University of Southern California Dissertations and Theses

Transcriptomic maturation of developing human cone precursors in fetal and 3D hESC-derived tissues

(USC Thesis Other)

Transcriptomic maturation of developing human cone precursors in fetal and 3D hESC-derived tissues

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

Transcriptomic Maturation of Developing Human Cone
Precursors in Fetal and 3D hESC-Derived Tissues
By
Dominic William Helsdon Shayler

A Dissertation Presented to the
Faculty of the USC Graduate School
University of Southern California
In Partial Fulfillment of the Requirements for the Degree
Doctor of Philosophy
Development, Stem Cells and Regenerative Medicine

August 2021

ii

Dedication
For my wonderful family who makes me who I am;
My father who inspires my journey with a smile,
My mother for driving me on that road with patience,
My sister and brother for blazing their own brilliant trails with humor and love,
And for Alina, who sees the man I aim to be.

iii

Acknowledgements
My first acknowledgement must go to my mentor, David Cobrinik, for all his constant
support over the years. His passion for the work, even when I felt discouraged or uncertain
about my results, helped me keep driving forward to complete this thesis. He provided me
opportunities to attend conferences as early as my first year in the lab to begin to learn the ins
and outs of the research community and presenting my data. I similarly want thank Jennifer
Aparicio, who established the retinal organoid work at CHLA. She taught me a huge amount
about working with the system and was always willing to answer my questions or discuss an
issue at length.
I also must thank my thesis committee members, Neil Segil (chair), Cheryl Craft, and
Justin Ichida. You consistently provided me with honest feedback, starting with my initial
proposal for a frankly too-large thesis project, and at each annual assessment but always with
fair criticism and a smile. All your time and wisdom has been immensely appreciated.
The next thanks must go to the members of the Cobrinik laboratory, the Stem Cell Core
team, and the Nagiel lab, past and present. Firstly, my training would have been poorer without
all of the fantastic post-docs. Particular thanks must go to Hardeep Singh (now assistant
professor), who has been an incredibly patient with me over the last 7 years as he taught me
laboratory basics, began the single cell RNA-sequencing work for our group, and put up with my
unending questions and requests for advice on all manner of issues. Another massive thank you
to Sunhye Lee, who streamlined our RNA-sequencing processes and directly helped with
sample collections, saving me many sleepless hours. Finally thank you to Zhengke Li, Donglai
Qi, and Yeha Kim for being mentors and wonderful lab mates.
Thank you, members of the CHLA Stem Cell Core team, Narine Harutyunyan, Andrew
Salas, and Kayla Stepanian, for teaching me to make organoids and juggling my (often poorly
planned) requests to begin a new round of preps. Thank you Narine for being ever patient and

iv

on top of every issue, while thank you Andrew and Kayla for being friends I could nerd out to
about video games and life in general. I’ll miss our group trips for thai food!
My fellow graduate students have both been phenomenal friends and labmates through
my time at USC. Kevin Stachelek joined the group when I did and almost single-handedly
became the bioinformatics expert in the lab. I cannot properly describe how much computational
work he did on this thesis work and I am eternally grateful. Thank you for being such a focused
co-worker and friend. Sijia Wang defended her work the same week as I did, and was a
sounding board for comparing our manuscripts, presentations, and indeed for all our work over
the last six years. Thank you for everything. Jinlun Bai is our newest grad student and has not
only done his own work but found time to help generate constructs for my work. Jay thank you
so much!
Thank you as well to former lab members Mitali Singh and Hann Hopp. Thank you Mitali
for all your work for the laboratory bioinformatics as well as deftly managing the lab. Hann thank
you for all organoid work you did as part of the Stem Cell team as well as for being a great
friend and work out buddy.
I also want to thank Matt Thornton and Brendon Grubbs for tissue collections that made
my fetal tissue work, as well as the members of the CHLA SC2 and FACS Cores for all the
hours on collecting single cells with me. Particularly, Jackie Lin for many hours of sample
collection and Annie Luong for additional collections and becoming a dear book buddy I could
rant about recent novels with. Thanks as well to the CHLA Vision Center and Thomas Lee for all
their support.
Lastly, thank you to my family and friends for all your comfort and encouragement on
this long road. I cannot properly convey everything my parents Lee and Sandra, or my siblings
Alex and Miles, mean to me. However, a special thank you goes to my partner, Alina Lieber.

v

Without you I would be a less confident and a much more disorganized man. Thank you for
loving and supporting me as I worked late hours in lab and on this dissertation, keeping me
sane and fed.

vi

Table of Contents
Dedication ................................................................................................................................... ii
Acknowledgements .................................................................................................................... iii
List of Figures .......................................................................................................................... viii
List of Tables ............................................................................................................................. xi
Abbreviations ............................................................................................................................ xii
Abstract.................................................................................................................................... xiii
Chapter 1 : Introduction .............................................................................................................. 1
The vertebrate eye ................................................................................................................. 1
Development of the eye field and optic cup ............................................................................ 4
Differentiation of retinal progenitor cells to terminal retinal cell types ...................................... 5
Structure of mature photoreceptors ........................................................................................ 6
Photoreceptor fate determination............................................................................................ 9
Order of photoreceptor marker gene and protein expression during development .................10
Development of 3D retinal organoid cultures .........................................................................11
Organoids as a model of human retina development and disease .........................................13
Thesis goals ..........................................................................................................................14
Chapter 2 : High resolution single cell transcriptomics reveals novel photoreceptor trajectories
and a cancer-predisposed developmental state ........................................................................16
Introduction ...........................................................................................................................16
Results ..................................................................................................................................17
FACS-enriched full-length scRNA-seq of human fetal photoreceptors ...............................17
NRL and RXRG isoform usage differs between early rod and cone populations ................33
High-resolution clustering identified two post-mitotic photoreceptor trajectories and a
common rod/cone precursor state ......................................................................................34
Features of early maturing L/M cone maturation trajectory .................................................53
Human cone cells express SYK, which impacts proliferative response to pRB loss ...........54
Discussion .............................................................................................................................63
Materials and Methods ..........................................................................................................68
Supplemental Data ................................................................................................................79
Chapter 3 : Similarities and Differences in Maturing Cone Photoreceptor Development in Fetal
Retina and hESC-Derived 3D Retinal Organoid Models .......................................................... 105
Introduction ......................................................................................................................... 105
Results ................................................................................................................................ 108
3a. Inconsistent expression of cone proliferation-related program associated proteins in
retinal organoids .............................................................................................................. 108
3b. scRNA-seq of maturing organoid-derived cone cells captured similarities and
differences between preparation methods and compared to human fetal cones .............. 123
Isolation of FACS-sorted single cells for RNA-sequencing ........................................... 123
Identification of cell types, age-related trajectories and preparation-specific cell
populations ................................................................................................................... 128
Identification and removal of aberrant organoid-derived cone populations .................... 129

vii

Establishing a trajectory with RNA Velocity and photoreceptor gene expression .......... 130
NRL and RXRG isoform usage in organoid photoreceptors .......................................... 135
S cones do not form a unique cell population in retinal organoids................................. 136
Marker Genes of Organoid Photoreceptor Maturation .................................................. 137
Cell cycle features divide cRPCs and RPC/MG populations ..................................... 145
Known rod genes are main markers of rod clusters .................................................. 145
Early organoid cones mimic the expression of the markers of early fetal cones. ....... 146
Late maturing organoid cones show coordinated upregulation of marker genes
indicating a state change .......................................................................................... 152
Identification of a transition point in transcription factor activity in maturing cones ........ 158
3C. Aberrant retinal organoid cone clusters defined by hypoxic markers and
metallothioneins ............................................................................................................... 165
Discussion ........................................................................................................................... 178
Materials and Methods ........................................................................................................ 185
Supplemental Data .............................................................................................................. 192
Discussion .............................................................................................................................. 203
References ............................................................................................................................. 209

viii

List of Figures
Figure 1.1: Structure of the human eye. ..................................................................................... 2
Figure 1.2: Structure and cell types of the human retina. ........................................................... 3
Figure 1.3: Structure of cone and rod photoreceptors ................................................................ 8
Figure 2.1: Overview of cell Isolation and single cell RNA-sequencing. .....................................17
Figure 2.2: FACS gating and sample summary .........................................................................18
Figure 2.3: UMAP plots of scRNA-seq of FACS-enriched human photoreceptor populations ....20
Figure 2.4: Aggregate scRNA-seq from PLAE and current scRNA-seq analyses ......................21
Figure 2.5: Individual low resolution cluster plots ......................................................................22
Figure 2.6: Identification of primary cell types ...........................................................................23
Figure 2.7: Expression plots for known RPC, MG, PR, rod, and cone genes ............................26
Figure 2.8: Differential expression between early and late maturing rod clusters ......................28
Figure 2.9: Differential expression between S and LM cone clusters .........................................30
Figure 2.10: Differential expression between late maturing LM cone group and LM cluster.......32
Figure 2.11: Differential NRL isoform use in rod and cone cell clusters .....................................36
Figure 2.12: Differential RXRG isoform use in rod and cone cell clusters ..................................38
Figure 2.13: Differential THRB isoform use in rod and cone cell clusters ..................................40
Figure 2.14: High resolution clustering and marker genes .........................................................41
Figure 2.15: RNA Velocity from retinal progenitor cells through maturing photoreceptors .........46
Figure 2.16: Cell cycle features distinguish high resolution cycling RPC (cRPC) and RPC/MG
clusters .....................................................................................................................................47
Figure 2.17: RPC-localized iPRP and TR cells upregulate cone and rod features .....................49
Figure 2.18: Bridge population has features of a common photoreceptor precursor state
adjacent to early cone ONECUT1/THRB pathway expression ..................................................51
Figure 2.19: High resolution L/M cluster marker genes and regulons ........................................56
Figure 2.20: Velocity identified two trajectories through L/M cones biased by age ....................58
Figure 2.21: L/M cone pseudotime reveals progression of lncRNA gene expression with
maturation .................................................................................................................................59
Figure 2.22: MYC targets and associated genes upregulated in LM cones over rods ...............61
Figure 2.23: SYK is expressed in fetal cones and inhibition decreases cell cycle entry after RB1
knockdown ................................................................................................................................62
Figure S2.1: SCENIC Regulon Specificity Scores (RSS) ..........................................................79
Figure 3.1: Organoid production overview ............................................................................... 110
Figure 3.2: Organoid appearance during development ............................................................ 111

ix

Figure 3.3: Variable pRB expression in retinal organoid cone precursors................................ 115
Figure 3.4: MDM2 expression is delayed compared to ARR3 in all organoid methods ............ 117
Figure 3.5: Organoid cones expressed intermittent nuclear MYCN that declined with age ...... 119
Figure 3.6: p27 signal remains nuclear across all organoid methods and ages ....................... 121
Figure 3.7: Overview of organoid collection and scRNA-sequencing ....................................... 125
Figure 3.8: FACS plots for example sorts from both methods of organoid production ............. 126
Figure 3.9: Summary of retinal organoid scRNA-seq data ....................................................... 127
Figure 3.10: Single-cell RNA-sequencing of retinal organoids and cell type identification ....... 131
Figure 3.11: Marker genes identify 2 cone clusters with unique marker expression that form
spatially distinct side populations ............................................................................................ 133
Figure 3.12: Visualization of scRNA-seq patterns after side population removal ..................... 134
Figure 3.13: RNA Velocity and photoreceptor marker genes define cone and rod trajectories 138
Figure 3.14: Differential NRL isoform use in rod and cone cell clusters ................................... 140
Figure 3.15: RXRG isoform use differences between rod and cone cell clusters ..................... 142
Figure 3.16: Organoid OPN1SW+ cells do not form a distinct population and have reduced S-
cone marker gene expression ................................................................................................. 143
Figure 3.17: Marker genes for reduced organoid dataset ........................................................ 149
Figure 3.18: Expression of retinal organoid rod cluster marker genes identified in Figure 3.17 in
organoid and fetal retina transcriptomes ................................................................................. 150
Figure 3.19: Marker genes of early organoid cone clusters show two patterns of downregulation
and similar behavior in fetal cones .......................................................................................... 151
Figure 3.20: Marker genes of early fetal photoreceptor precursors are expressed in immature
organoid cones ....................................................................................................................... 154
Figure 3.21: Late organoid cone cluster marker genes are not detected in fetal cones ........... 155
Figure 3.22: Transcription factor regulon upregulation in late maturing cones ......................... 157
Figure 3.23: Combined fetal and organoid scRNA-seq datasets show most overlap in RPCs and
opsin-expressing photoreceptors ............................................................................................ 161
Figure 3.24: Differential gene expression in equivalent cone precursor clusters in retinal
organoids and fetal retina ........................................................................................................ 162
Figure 3.25: Aberrant side population C-SG has unique marker genes and enriched for hypoxia-
related genes .......................................................................................................................... 169
Figure 3.26: Aberrant side population C3* shows increased expression of oxidative stress
upregulating genes ................................................................................................................. 172
Figure 3.27: Ages of cells in C-SG and C3* distinct populations ............................................. 173

x

Figure 3.28: Increased BHLHE40 protein expression in internal organoid cones, apical
Kuwahara cones, and foveal fetal cones. ................................................................................ 174
Figure 3.29: Increased ENPP2 protein expression in internal and border organoid cones, apical
Kuwahara cones, and in foveal fetal cones ............................................................................. 176
Figure S3.1: SCENIC Regulon Specificity Scores (RSS) ........................................................ 192
Figure S3.2: Gene ontologies for differential expression of organoid vs fetal rods ................... 193

xi

List of Tables
Table S2.1: Differentially Expressed Genes Between Rod Clusters ER and LR ........................80
Table S2.2: Differentially Expressed Genes Between Cone Clusters S and LM ........................83
Table S2.3: Differentially Expressed Genes Between L/M Opsin Quintet and Other LM cluster 85
Table S2.4: Genes Correlated with L/M Cone Pseudotime and Module Grouping .....................88
Table S2.5: Differentially Expressed Genes Between Low Resolution ER and LM ....................94
Table 3.1: Gene list for select LM4 ontologies from fetal LM4 and organoid Cone-3 differential
expression .............................................................................................................................. 164
Table 3.2: Gene list for select Cone-3 ontologies from fetal LM4 and organoid Cone-3
differential expression ............................................................................................................. 164
Table S3.1: Differentially Expressed Genes Between Fetal LM4 and Organoid Cone 3 .......... 194
Table S3.2: Differentially Expressed Genes Between Large Cone Side Group and Cone 3
Excluding Small Side Group .................................................................................................... 198
Table S3.3: Differentially Expressed Genes Between Small Cone Side Group and All Other
Cone 3 .................................................................................................................................... 202

xii

Abbreviations
Abbreviation Name
AC Amacrine Cell
BB Basal Body
BC Bipolar Cell
CC Connecting Cilium
cRPC Cycling Retinal Progenitor Cell
ESC Embryonic Stem Cells
FW Fetal Week
FPM Fragments Per Million
GC Ganglion Cell
GCL Ganglion Cell Layer
HC Horizontal Cell
hESC Human Embryonic Stem Cell
hPSC Human Pluripotent Stem Cells
iPRP Immature Photoreceptor Precursors
iPSC Induced Pluripotent Stem Cell
INL Inner Nuclear Layer
IPL Inner Plexiform Layer
IS Inner Segment
LM L/M Cone
MG Müller Glia
NR Neural Retina
OLM Outer Limiting Membrane
ONL Outer Nuclear Layer
OPL Outer Plexiform Layer
OS Outer Segments
PR Photoreceptor
PSC Pluripotent Stem Cell
RSS Regulon Specificity Scores
RGC Retinal Ganglion Cells
RMM Retinal Maturation Media
RPE Retinal Pigment Epithelium
RPC Retinal Progenitor Cell
pRB Retinoblastoma Protein
RT Room Temperature
SCENIC Single-Cell Regulatory Network Inference and Clustering
SLM Smart Local Moving
UMAP Uniform Manifold Approximation and Projection

xiii

Abstract

The human retina is a delicate tissue that provides the light-sensing capabilities that give us
vision. However, various congenital and age-related conditions can cause this system to
weaken and fail, leading to vision impairment or total blindness. Human cone photoreceptors in
particular have the unique capability to form retinoblastomas, indicating the presence of unique
human developmental processes that are not well elucidated. This thesis aimed to further
dissect cone maturation through full-length scRNA-seq of enriched fetal cone photoreceptors.
Through examination of known photoreceptor marker genes, I identified differential isoform
usage of canonical rod and cone marker genes NRL, RXRG, and THRB. Further parsing of
trajectories from retinal progenitor cells (RPCs) through photoreceptors suggested that there are
two potential routes of differentiation to rods, one of which was through a post-mitotic cone/rod
photoreceptor precursor state. I identified new gene expression defining this population, as well
as showed that cone-directed precursors expressed ONECUT1 and had increased OLIG2
regulon activity, elements previously only described in RPCs that were fated to become cones.
As these cones continued maturing, I found a progression of lncRNA genes that gained or lost
expression with development. Differential expression between developing rods and cones, to
explore cone-specific features relevant to the pRB-deficient cone precursor proliferative
response, identified SYK, which we then showed had expression in normal cone maturation and
was needed for the pRB-depleted cone entry into the cell cycle.
Modeling of human retina through 3D hESC-derived retinal organoids provides a more
available tissue for studying development and disease, however current organoid models are
imperfect, and the maturation of individual cell types has not been closely compared against
human retinal cells. To compare cone cell development between tissues, I first examined the
expression of cone proliferation-related proteins and found differences in timing and localization
in relation to expression of cone arrestin (ARR3). An scRNA-seq dataset was generated from

xiv

organoids of two production methods as done for fetal retina to directly compare cone
transcriptomes. Each production method generated cone populations poorly represented from
the other, with one more rapidly producing late maturation cone cells. However, overall organoid
cones recapitulated fetal features like fetal NRL and RXRG isoform patterns and photoreceptor
precursor gene expression. Late maturation cone states were also better isolated from
organoids than from fetal retina. Direct comparison of fetal and organoid cone cells
demonstrated major differences in the expression of ribosome and glycolytic genes which
suggest that culture conditions distinctly impact organoid gene expression. The scRNA-seq data
also identified aberrant organoid cone populations with hypoxic and oxidative stress signatures,
whose marker proteins BHLHE40 and ENPP2 were observed in atypical internal and border
region cone cells. Unexpected expression in fetal foveal cones indicated both proteins were
expressed during normal as well as abnormal cone development.
These dissertation results provide full-length photoreceptor transcriptomic datasets that
will be useful for future focused analyses. These observations provide insights into cone cell
differentiation and maturation in fetal retina, identify cone-cell intrinsic gene expression that has
a role in the proliferative response to pRB loss, and outline key similarities and differences in
retinal organoid-derived cone cells that will help researchers improve on organoid culture
methods and analysis.

: Introduction

Among our senses, vision is one of the most dominant for the average individual. Common
aspects of life, from basic navigation and communication to activities like watching TV and
reading, are all heavily dependent on sight. However, visual impairment or total blindness
impacted approximately 3.22 million people in the United States of America in 2015, as reported
by the Center for Disease Control, was among the top 10 most common disabilities, and has a
measurable effect on quality of life (Vu et al. 2005). The causes of vision loss vary, and can
occur due to congenital effects, later in life due to degenerative diseases or conditions like
diabetes and age-related macular degeneration, or due to direct injury to the eye. Within the
eye, the retina is particularly vulnerable to these various genetic and environmental effects and
is often the reason impairment occurred. To treat or repair the retina, we need a clear
understanding of how it forms and functions, and how disease or injury states modify or
manipulate those normal processes. This introduction will serve as an overview of the structure
and development of the vertebrate retina, as well as approaches for modeling retina using
human embryonic stem cells. It will conclude with an overview of photoreceptor development
and the current gaps in knowledge which drive this thesis work.

The vertebrate eye
The structure of the vertebrate eye is remarkably conserved across species (Fadool and
Dowling, 2008). The anterior portion of the eye consists of the cornea, the transparent outer
layer that partially focuses light into the front of the eye, the aqueous humor, a fluid filled
chamber that sits behind the cornea, the iris, a muscle system that controls the pupil aperture
for light to pass through, and the lens, a clear structure responsible for focusing light into the
back of the eye (Figure 1.1). The posterior of the eye is primarily a large chamber filled by the
vitreous humor, a more viscous liquid than the aqueous. The wall of this chamber is composed

2

of three layers which from the outside in are: the sclera, which provides the outer protective and
structural membrane, the choroid, which is made of the blood vessels supplying the eye, and
the retina, a neuroepithelium which is responsible for detecting light and transmitting sensory
signals to the brain.

Figure 1.1: Structure of the human eye
AH: Aqueous Humor. VH: Vitreous Humor. Created in BioRender.

A vertebrate retina consists of seven primary cell types: retinal ganglion cells, amacrine
cells, horizontal cells, bipolar cells, Müller glia, and the two photoreceptors, rod cells and cone
cells (Figure 1.2). These cell types form a structure with three nuclear layers, with cell synapses
in intervening plexiform layers (the inner and outer plexiform layer (IPL/OPL), a layout which is
highly conserved among vertebrate species. RGCs form the most interior nuclear layer adjacent
to the vitreous humor, where their axons create an interconnected nerve fiber layer that exits the
eye as the optic nerve (Thoreson 2016). The second layer, the inner nuclear layer (INL),
consists of horizontal cells, bipolar cells, and amacrine cells which form synapses with each
other as well as with photoreceptors, which make up the outer nuclear layer (ONL). Just
posterior to the retina, before the capillary beds and outer wall of the eye, sits the retinal

3

pigment epithelium (RPE); it consists of a single-cell layer of pigmented cells that surrounds the
outer segments (OS) of those light sensing cells. The RPE maintains the blood/retina barrier
and performs key recycling functions for components of the phototransduction pathway in
photoreceptors (Strauss 2005).

Figure 1.2: Structure and cell types of the human retina
Illustration of the layered structure of the mature human retina. Cell nuclei fall into three layers, outer
nuclear layer (ONL), inner nuclear layer (INL) and ganglion cell layer (GCL). Between are the synaptic
layers, inner plexiform layer (IPL) and outer plexiform layer (OPL). Retinal pigment epithelium (RPE) sits
apical to the photoreceptors as a separate layer and adjacent to vasculature. The seven major cell types
of the retina are labeled and colored: Rod cells in blue, cone cells in red, horizontal cells in purple, bipolar
cells in brown, amacrine cells in salmon, retinal ganglia in orange, and Müller glia in green. Created in
BioRender.

Evolutionary pressures for different species resulted in shifts in cell type distribution to
generate regions called acute zones, with higher visual acuity and greater retinal ganglion cell
Apical
Basal

4

density. In some birds and primates, including humans, a specific type of acute zone forms
called a fovea (Baden et al. 2020). The defining feature of a fovea is that INL cells and retinal
ganglion cells are shifted peripherally so that the only impediment to light reaching the
photoreceptors are their own axons, raising light detection and visual resolution (Hendrickson
and Yuodelis 1984). In humans, this is a single pit-like structure in the central vision (fovea
centralis) that is completely devoid of rod cells and has the highest density of cones. Peripheral
to this region, cone numbers fall off steadily and the greater portion of the retina is dominated by
rod cells (Mustafi et al. 2009). The only region absent of photoreceptors is the optic nerve head,
representing a blind spot in the visual field.

Development of the eye field and optic cup
The formation of the eye begins after gastrulation of the neural tube when a region of the medial
anterior neural plate is distinguished as the eye field, patterned by the combined action of
several eye field transcription factors, including Otx2, Lhx2, Rax, Six3, Pax6, and Six6 (Miesfield
and Brown 2019). This region gives rise to not only retina, but iris and RPE as well. Early
research on the eye field clarified that a single eye field formed that subsequently divided to
form bilateral eyes, rather than two separate eye fields developing independently. Removal of
portions of the anterior neural plate in developing salamanders showed singular eye formation
when a large percentage of the region was removed (Adelmann 1929a,b). Similarly, when these
same neural plate sections were transplanted to other locations in the developing animal, a
small, underdeveloped eye was sometimes observed. Co-transplanting with larger portions of
the underlying mesoderm produced two eyes, supporting its involvement in correctly generating
two organs.
The division of the eye field occurs along the midline of the developing embryo, specified
by the actions of Six3 and Shh (Geng et al. 2008, Graw 2010). This was first observed by
lineage tracing in the anterior neural plate of zebrafish, which identified posterior presumptive

5

diencephalon mar+ cells that migrated forward to divide the presumptive RPC population into
two (Varga et al. 1999). These two regions are referred to as optic sulci, or pits, and are
observed at around 22 days in fetal human development (Mϋller and O’Rahilly 1985, Quinn and
Wijnholds 2019, Verdijik and Herwig-Carl 2020).
The optic pits will subsequently evaginate from the diencephalon, retaining a connection
via the optic stalk, and contact the neighboring surface ectoderm, forming optic vesicles with the
ectoderm becoming the presumptive lens placode. The vesicle then begins to invaginate to form
the optic cup. Through a series of transcription factors and external signals, the vesicle begins
to pattern itself into presumptive retina, RPE and optic stalk regions (Heavner and Pevny 2012).
During invagination, the neuroepithelium folds into two layers, one surrounding the other with
the outermost of the two fated to become RPE and the inner the future retina.

Differentiation of retinal progenitor cells to terminal retinal cell types
The retinal layer of the new optic cup initially consists of retinal progenitor cells (RPCs).
Attached to the apical and basal membranes, RPCs undergo interkinetic nuclear migration
during the cell cycle, with nuclei basally located during S-phase and apical for M phase,
migrating between the two locations during G1 and G2. RPCs divide both symmetrically to form
two RPCs or two post-mitotic cells and asymmetrically to form one of each (Baye and Link,
2007 and 2008). Symmetric division serves to thicken the early retina by increasing the RPC
population before transitioning to producing differentiated cell types through asymmetric and
eventually symmetric division. RPCs generate all seven retinal cell types in all vertebrate
species (Turner and Cepko 1987, Wetts and Fraser 1988, Turner et al. 1990). However, these
cells are born in a biased order during tissue development. Earlier experiments utilizing [
3
H]
thymidine pulses in mouse and rat (Young 1985, Rapaport et al. 2004) demonstrated a temporal
pattern of S-phase incorporation: amacrine, ganglion, horizonal and cone cells predominately
were labelled after early age [
3
H] thymidine pulses, while rods, bipolar cells, and Müller glia all

6

were labelled in later weeks. This order of birth is relatively conserved across species and has
been more recently shown in human retina RNA-sequencing experiments (Aldiri et al. 2017,
Hoshino et al. 2017).
Along with this birth order, differentiation and maturation of all cell types is typically
graded along the central-to-peripheral axis of the retina (Prada et al. 1991, Malicki 2014). In
human retinal tissue, cells first differentiate and mature in the central fovea while more
peripheral regions proliferate and expand before undergoing these processes later. Macaques
show the same pattern, where cells in the fovea first incorporated [
3
H] thymidine labels (la Vail
et al. 1991). Mitotic cells also vanished first from the central human retina (Provis et al. 1985,
Hendrickson 2016). Other species manifest slightly different maturation patterns, such as
zebrafish neurogenesis starting in a ventral patch adjacent to the optic stalk and progressing
across the central retina to the dorsal side (Raymond et al. 1995, Hu and Easter Jr., 1999,
Stenkamp 2015). In human retina, one interesting exception to this patterning are the
photoreceptors of the human fovea, which are morphologically less mature than fovea-adjacent
photoreceptor populations from post-fertilization week 22 until birth, consistent with the pause in
central cone development prior to post-natal development of the foveal pit (Abramov et al. 1982,
Hendrickson and Drucker 1992). Infants have been demonstrated to use more peripheral vision
right after birth, also supporting that central vision is less well developed even though these
foveal cone cells were born from RPCs first.

Structure of mature photoreceptors
Photoreceptors are the key sensory cell type of the retina, performing the crucial function of
converting photons into an electrical signal to pass on to the brain. Both rods and cones share
the same basic polarized structure (Roepman and Wolfrum 2007), oriented parallel to the path
of incoming light. The basal end of the cell is closest to the INL and the interior of the eye; the
region between INL and ONL, the outer plexiform layer (OPL), is where the processes of the

7

bipolar and horizontal cells form a synapse with the terminals of photoreceptors (Figure 1.3).
Moving apically, the nuclear region is dominated by the photoreceptor nucleus and little else.
However further on, the photoreceptors project past the outer limiting membrane (OLM), which
serves to separate the cell body from the extended segment structures that give each cell type
their name. These rod and cone structures are modified cilia referred to as the outer segment
(OS), which are connected to a region of the cell body called the inner segment (IS) sitting just
outside the OLM (Roepman and Wolfrum 2007, Kennedy and Malicki 2009). The IS contains all
the standard organelles for a eukaryotic cell and commonly has two regions, the myoid area
which contains the Golgi apparatus and endoplasmic reticulum, and the ellipsoid area which is
densely packed with mitochondria. Photoreceptors have an uncommonly high number of
mitochondria to fuel the regular phototransduction processes occurring in the cell. The apical
end of the IS has a basal body that projects microtubules up to the OS by a connecting cilia
structure which then forms the core structural support for the OS. Finally, the OS is where the
light-sensing opsin proteins are housed and the phototransduction cascade begins. In rods, this
is in a stacked sequence of disc-shaped membranes internal to the main OS membrane
(Steinberg 1981). In cones however, the membrane itself becomes layered into lamellae and
serves as the location for opsins. These structures, both discs and lamellae, are regularly turned
over and replaced (Young 1967), which is part of the previously mentioned recycling role of the
RPE.
Rods and cones are uniquely specialized to prioritize different aspects of vision.
Scotopic vision is handled by the rods, which demonstrate sensitivity for even a single photon of
light via dramatic downstream amplification of signal from a single rhodopsin molecule (Baylor
et al. 1979, Thoreson 2017). In addition, multiple rod signals are collated to single retinal
ganglion cells to further amplify any signal detected across the entire group. By contrast, cones
are commonly present in a 1:1 ratio with ganglia, to allow greater levels of signal discrimination

8

per cell. The cones dominate in high light vision, producing both greater clarity and color
images. Cones subdivide into types based on the light spectra they are most sensitive to.

Figure 1.3: Structure of cone and rod photoreceptors
Structure of cone (red) and rod (blue) photoreceptors. The synaptic terminals (ST) of photoreceptors sit in
the OPL to interact with INL cells. The cell passes the outer limiting membrane (OLM) apical to the cell
body, where the inner segment (IS) contains organelles including a large number of mitochondria. The
basal body (BB) sits at the apical end of the inner segment, where it extends microtubules through the
connecting cilium (CC) into the outer segment (OS) where phototransduction begins in the lamellae or
discs. Created in BioRender.

Humans have three cone subtypes, S-cones (blue pigment, max absorption = 426nm), M-cones
(green pigment, max absorption = 530nm), and L-cones (red pigment, max absorption = 552-
557nm) (Merbs and Nathans, 1992). S-cones remain the most transcriptionally distinct of the
three, as L and M cones are the result of a single pigment gene duplication (Thoreson 2017).

9

Photoreceptor fate determination
Several proteins have been identified as key in the determination of RPCs to rod and cone cell
fates. The transcription factor OTX2 is critical for the birth of photoreceptors and ubiquitously
expressed in all rod and cone types. OTX2 deletion in mice prevents both photoreceptor and
bipolar cell development while overexpression increases photoreceptor cell numbers (Nishida et
al. 2003, Koike et al. 2007, Sato et al. 2007, Wang et al. 2014). Cone-Rod Homeobox (CRX) is
expressed downstream of OTX2 in all photoreceptors and is crucial for proper expression of
other photoreceptor genes, but it is not necessary for photoreceptor fate determination to take
place (Chen et al. 1997, Freund et al. 1997, Furukawa et al. 1997, Brzezinski and Reh, 2015).
Rod cells need expression of NRL and RORB, and without either the cells that would normally
become rods shift to S cones (Mears et al. 2001, Fu et al. 2014). Interestingly, no unique cone
transcription factor has been similarly identified. Thyroid Hormone Receptor Beta gene (THRB)
is key for L/M cone development, particularly the TRβ2 protein isoform, but its loss in mice
causes all M-opsin expression to vanish and S-opsin expression to rise (Roberts et al. 2006),
and a similar phenomenon was modeled with human embryonic stem cells (Eldred et al. 2018).
However, no factors have been shown to prevent any cone birth while promoting rod fate,
indicating cone fate is the default and that additional factors are required to produce rods.
It is also unclear whether photoreceptor fate is specified prior to terminal mitosis in
RPCs, or whether a post-mitotic, common rod and cone precursor state exists. The changing
bias in cell type birth over retinal development supports that some process restricts the cell
types that RPCs can produce at a specific time. A subpopulation of mouse RPCs expressing
Olig2-TVA were shown to terminally divide soon after gamma-retrovirus labeling as they only
generated 1-4 daughter cells. These daughters were cones and HCs at E13.5-14.5, while by P0
the same Olig2+ RPCs only generated rod and amacrine cells (Hafler et al. 2012, Cepko 2014).
These binary fate choices and the change in daughter cell bias were linked to RPC expression
of Onecut1 and Otx2; at earlier developmental stages these proteins bind the THRB promoter

10

and if cone fate is chosen Onecut1 drives up THRB expression and then downregulates
whereas in horizontal cell fate Otx2 levels decrease, THRB is not induced, and Onecut1
remains expressed (Emerson et al. 2013). Older age Olig2+ RPCs do not express Onecut1
when producing rod and bipolar cell fate, however elimination of Onecut1 in mice does not
prevent cone birth, only impairs maturation (Sapkota et al. 2014, Brzezinski and Reh 2014).
These findings support age-related cell fate restriction, and that cone/horizontal or rod/bipolar
fate is regulated by RPC expression of Onecut1. In contrast to this observation, non-RPC post-
mitotic cells co-expressing TRβ2 and Nrl have been observed in mouse retina, suggesting a
rod/cone hybrid state (Ng et al. 2011). Recent RNA-sequencing work also failed to distinguish
rod and cone directed post-mitotic PR transitional cell populations (Sridhar et al. 2020) possibly
consistent with the presence of common post-mitotic cone-rod precursors or reflecting a failure
of the low-read depth single cell RNA sequencing to resolve cone or rod fated precursor
populations where fate decisions have been made.

Order of photoreceptor marker gene and protein expression during development
Gene and protein expression changes continue in post-mitotic photoreceptors, suggesting that
while not dividing there is a developmental continuum that must occur in post-mitotic rods and
cones. Photoreceptors take time to form synapses (Hendrickson et al. 2008) as well as fully
construct their outer segments and express the phototransduction genes and proteins that
associate with it. Hoshino et al. demonstrated sequential upregulation of transcript expression in
bulk RNA-seq of human samples: OTX2 is expressed early, then CRX and NRL followed by
both photoreceptor transducin genes, GNAT1 and GNAT2 and the phosphodiesterase, PDE6H,
which catalyzes hydrolysis of cGMP during signaling from activated opsin proteins. OPN1SW
was first expressed just after GNAT2, but the other opsin genes (OPN1MW in cones, RHO in
rods) were some of the last detected (Hoshino et al. 2017). Protein expression matches with
some of these observations; OTX2 and CRX appear early in photoreceptors, rods express NRL

11

and NR2E3 before RHO (Xiao and Hendrickson 2000, Hendrickson et al. 2008) and S-opsin is
expressed before L/M opsin in cones. Together the well-established expression timing of these
photoreceptor components provide maturation signposts during analysis of photoreceptor
development.

Development of 3D retinal organoid cultures
The ability to directly interrogate human retina development is limited by the availability of pre-
natal tissues and the types of experiments allowed by that tissue. As such, much retinal
research is done in model organisms such as mouse or zebrafish. However, human pluripotent
stem cells (hPSCs) offer an alternative for human retina studies. Pluripotent stem cells are self-
renewing cell populations with the capacity to generate any differentiated cell type and are
either derived directly from embryos (ESC) or by transforming an adult differentiated cell
population back into a proliferating and multi-potent cell state (iPSC) (Thomson et al. 1998,
Takahashi et al. 2007). Embryonic stem cells (ESC) and, more recently, induced pluripotent
stem cells (iPSC) offer the capability to differentiate and grow both individual retinal cell types as
well as more complete tissues in vitro. In the last 20 years, the process of deriving retinal
tissues from stem cells has evolved rapidly, with the earliest work using two-dimensional
adherent cultures. Basic protocols demonstrated differentiation of neural progenitors which also
showed intermittent expression of photoreceptor markers in cultured cells (Zhao et al. 2002).
Procedures for deriving high percentage RPC-like populations from mouse (Ikeda et al. 2005)
and human (Lamba et al. 2006) ESCs followed shortly after. The laboratory of Yoshiki Sasai
was especially prolific and developed protocols for deriving photoreceptors from mouse, human,
and monkey ESCs (Ikeda et al. 2005, Osakada et al. 2008). Two groups built on these previous
works by first applying differentiation protocols to iPSCs and attempting to differentiate cells on
the expected fetal timeline (Hirami et al. 2009, Meyer et al. 2009). These two publications both
also utilized some of the earliest versions of retinal 3D aggregates.

12

The Sasai lab pioneered a ‘serum-free floating culture of embryoid body-like aggregates’
(SFEB) system, an early move away from adherent cultures (Watanabe et al. 2005, Eiraku and
Sasai, 2012). This involved dissociating and aggregating stem cells in a 96 well plate and
capitalized on the neural fate bias of ESCs to derive neural tissue structures without any
exogenous factor additions. A significant step was made when the Sasai group demonstrated
how to generate mouse cortical organoids with this method. Self-organizing structures had been
generated for other tissues previously (Sato et al. 2009); however, the Sasai cortical tissues
showed self-directed generation of multiple cell types in expected birth order and distinct cortical
layering (Eiraku et al. 2008). This was followed by the creation of the first mouse ESC-derived
optic cup-like structures in floating culture (Eiraku et al. 2011). Not only were the previous
structural advances retained, but the introduction of Matrigel, a secreted extracellular matrix
product, provided a rigid substrate that encouraged retinal fate and enabled the evagination and
invagination that occurs in the developing optic vesicle and cup.
The following year, the protocol was adapted to hESCs, producing only neural retina
cells until modified with a short Wnt antagonist treatment, which generated RPE cells and optic
cup-like structures (Nakano et al. 2012). However, this protocol still utilized Matrigel, which is an
inherently undefined factor that can vary greatly between lots, influencing organoid production.
In the paper of Kuwahara et al., the Sasai group overcame this issue and produced more
consistent results by replacing Matrigel with BMP4 treatment on day 6 in culture to induce eye
field expression (Kuwahara et al. 2015). Wnt activation and fibroblast growth factor (FGF)
inhibition would push the culture toward RPE, however transient treatment before return to
standard NR differentiation returns most cells to NR, a process referred to as ‘induction-reversal
culture’. This did not produce an optic cup-like structure, but rather a tapered epithelial edge to
the NR that contacts RPE-like cells. This structure is highly reminiscent of the ciliary margin
region at the outermost edge of the developing fetal retina (Kuwahara et al. 2015).

13

In parallel to the Sasai work, the Canto-Soler laboratory built upon Meyer et al. 2009,
where they generated disorganized 3D spheres with RPC and photoreceptor protein
expression, to develop a novel 3D-2D-3D culture model (Meyer et al. 2011, Phillips et al. 2012,
Zhong et al. 2014). Here stem cell colonies were lifted intact from culture and allowed to
aggregate in suspension for 7 days, initially with Wnt and BMP inhibitors to promote anterior
neuroectoderm fate. In earlier publications, these aggregates were plated onto a dish coated
with laminin but in Zhong et al. (2014) this was done instead with Matrigel. These aggregates
spread onto the scaffold and after 16 days arcs of neural retina can be excised and
reaggregated into 3D structures in culture medium. Finally, while earlier methods were
completely chemically defined, Zhong et al. added 10% fetal bovine serum (FBS) at d42 to
dramatically extend maturation (Aparicio et al. 2017b). This publication was also the first time an
organoid model demonstrated hESC-derived photoreceptor current changes in response to
light, a notable milestone.

Organoids as a model of human retina development and disease
The potential use of retinal organoids as a system to interrogate normal processes of
development and disease, as well as a potential platform for treatment testing, is enticing.
CRISPR/Cas9 tools allow for direct modifications to genetics that are impossible in human
tissue samples, while iPSCs in particular allow for tissue on a patient’s genetic background to
be generated, which has allowed for recapitulation of disease phenotypes in culture. Mutant
PRFPF31 lines from patients with retinitis pigmentosa recapitulated the disrupted cilia
morphology and degeneration seen in vivo (Buskin et al. 2018). Photoreceptor ciliopathies have
been more successfully modeled in organoids overall (Parfitt et al. 2016, Quinn et al. 2019).
Structural conditions like X-linked retinoschisis or CRB1 mutations have also been effectively
recreated (Huang et al. 2019, Quinn et al. 2019). CRISPR-edited cell lines have in some studies
even demonstrated rescued wild-type morphologies (Buskin et al. 2018).

14

However, these successful disease modeling studies are limited to the aspects of retinal
development that are sufficiently recreated in organoids. Organoid models do not sustain retinal
ganglion cell populations beyond a few months (Aparicio et al. 2017a, Sridhar et al. 2020), and
they develop without the surrounding tissues that would typically support the fetal retina, such
as vasculature or a contiguous layer of RPE (Bell et al. 2020). These are both aspects that
might affect the maturation and long-term survival of organoid photoreceptors, where the
formation of mature outer segments (Zhong et al. 2014, Hallam et al. 2018) as well as
detectable light response is limited. Even with current models sustaining organoids for several
hundred days in culture, all cell types show minimal transcriptional changes after around 30
weeks (Cowan et al. 2020). Retinoblastoma initiation and formation is one human cone-specific
disease process that has proven inconsistent and difficult to achieve; loss of RB1 has caused
high levels of apoptosis and no cone cell proliferative response in one study (Zheng et al. 2019),
while other groups have demonstrated tumor-like structures derived from RB1-mutant and
knockout ESCs (Liu et al. 2020). Overall, while retinal organoids are currently sufficient for
studying many aspects of human retina development, it is still unknown how closely the
developmental process in the in vitro retinal organoid cultures follow that which occurs in vivo.

Thesis goals
Due to the inherent variability and current flaws in retinal organoids, it becomes more important
to closely evaluate the developmental trajectory of each retinal cell type and compare it to the
standard of in vivo human retinal development for accuracy of timing, gene expression, and
state changes. Only recently have large scale RNA-sequencing projects begun to provide
insight into the transcriptomic changes underlying retinal cell maturation in human retina and
retinal organoids, but they typically focus on broad cell type differences or fate specification.
Cone cells have unique human-specific developmental features that we would want to replicate
in vitro; however, we do not have an understanding of transcriptional or protein expression

15

states that exist during maturation in fetal retina beyond known cell type marker genes, let alone
in organoids.
The goal of this thesis, then, is to further parse the developmental states and state
changes of human fetal cone cells as well as evaluate how accurately hESC-derived retinal
organoids mimic fetal cone maturation. To break down fetal cone photoreceptor development, in
Chapter 2 I used single-cell RNA-sequencing to investigate enriched photoreceptor
transcriptomes for details of photoreceptor fate specification, post-mitotic cone maturation
states, and disease-relevant gene expression differences between rods and cones Then in
Chapter 3, my aim was to compare retinal organoid cones from multiple production
methodologies against fetal cones, first by evaluating the expression timing of proteins
associated with fetal cone response to RB1 loss and then by generating single-cell
transcriptomes from enriched organoid photoreceptors to compare with our fetal photoreceptor
gene expression and cell state patterning data. These data will be valuable in deepening our
understanding of early human cone development, but also in identifying flaws in the organoid
model that may help refine future protocols for more accurate in vitro retinal development
models.

16

: High resolution single cell transcriptomics reveals novel
photoreceptor trajectories and a cancer-predisposed developmental
state
Contributions: Hardeep Singh: scRNA-seq methodology and samples. Sunhye Lee: scRNA-seq
methodology and samples, SYK inhibition. Kevin Stachelek: Bioinformatics. Mitali Singh: Bioinformatics.
Yeha Kim: SYK immunofluorescence. Matthew Thornton: Tissue collection. Brendan Grubbs: Tissue
Collection. Jackie Lin: CHLA FACS Core. Annie Luong: CHLA FACS Core. Michael Bonaguidi and Maxwell
Bay: Bioinformatics advice. Mark Reid: Statistics. David Cobrinik: PI

Introduction
Vision degradation and loss often relates to rod and cone photoreceptor developmental defects,
yet there is currently limited understanding of human photoreceptor differentiation and
maturation mechanisms. Whereas post-mitotic photoreceptor development stages have been
recognized based on morphologic features and phototransduction-related protein or gene
expression (Sehgal et al. 2006, Hendrickson 2012), it is unclear whether progression through
such stages can be subdivided into distinct cell states or subsumes a developmental continuum.
Defining the developmental states through which cells progress is important for understanding
how retinal disorders initiate. As one example, retinoblastomas arise following a proliferative
response to loss of the retinoblastoma protein (pRB) in maturing cone precursors, dependent on
the cone precursors’ intrinsically high MDM2 and MYCN (Xu et al. 2009 and 2014, Singh et al.
2018), yet the cell-intrinsic factors underlying the cancer-predisposed state have not been
defined. As no animal models recapitulate all human photoreceptor features, such as a rod-free
fovea, tri-color vision and the cone precursor proliferative response (Seabrook et al 2017, Singh
et al. 2018), analysis of human tissue is needed to define the human disease-relevant
photoreceptor features.
Single-cell RNA-sequencing (scRNA-seq) via 3’ end-counting has revealed gene
expression changes that accompany photoreceptor production in human as well as murine and
macaque retinae (Liang et al. 2019, Clark et al. 2019, Lu et al. 2019, Sridhar et al. 2020).
However, these analyses have not revealed isoform-level regulation or cell states that influence

17

photoreceptor disease. Here we used deep, full-length scRNA-seq of human retinal progenitor
cells (RPCs) and developing rod and cone photoreceptors, with enrichment of rare cone
precursor populations, in order to discriminate photoreceptor developmental states, identify
transcript isoform-specific roles in photoreceptor fate determination, elucidate developmental
trajectories, define a retinoblastoma origin-related cone developmental stage with intrinsically
high proto-oncogene MYCN and SYK expression, and provide a resource for future human
photoreceptor studies.

Cells were isolated from fetal retina of six different weeks between post-fertilization week 13 and 19,
either directly from FACS or collected in bulk and then isolated using the C1 microfluidic system.

Results
FACS-enriched full-length scRNA-seq of human fetal photoreceptors
To interrogate transcriptomic changes during human photoreceptor development, dissociated
fetal week 13–19 RPCs and photoreceptor precursors were FACS-enriched based on forward-
scatter, side-scatter, and CD133/CD44/CD49b staining, followed by microfluidic isolation of
FACS-enriched cells or sorting directly into microliter droplets, full-length cDNA synthesis,
paired-end sequencing, and alignment to ENSEMBL transcript isoforms (Figure 2.1). The FACS
enrichment was based on a prior cultured cone precursor isolation method (Xu et al. 2014), but
Figure 0.1: Overview of cell Isolation and single cell RNA-sequencing.

18

19

Figure 2.2: FACS gating and sample summary
A. Representative FACS plots for fetal week (FW) 13, 15, 16, 18, and 19 dissociated retinal cells stained
with CD133, CD44, and CD49b. Forward and side-scatter area gates were used to enrich for CD133+
photoreceptors, excluding the smaller SSC-low and FSC-low cells to remove more rods. B. Histogram of
total paired-end read counts for each cell sequenced, colored by sequencing run. Seq 1 cells were
isolated entirely by C1, Seq 2 cells were isolated by FACS or C1 as indicated, and in all other runs cells
were FACS-isolated only. Dotted line: 100,000 read cutoff for cell exclusion. C. Summary of retinae and
cell numbers at each age. D. Box plots of the read count, genes detected, and transcripts detected per
cell ordered by time of collection. Green bar: Average value across all samples (Rounded to nearest
whole number: read counts per cell = 3,750,417. Genes detected = 8,278. Isoforms detected = 20,344.)

20

Figure 2.3: UMAP plots of scRNA-seq of FACS-enriched human photoreceptor populations
A. UMAP plot with samples colored by retinal age and collection order. B. UMAP plot with samples
colored by method of single cell isolation, direct collection from FACS or C1 microfluidics system. C.
UMAP plot with samples colored by sequencing run; key indicates sample isolation method in each
run. D. UMAP plot colored by low resolution SLM cluster.

21

Figure 2.4: Aggregate scRNA-seq from PLAE and current scRNA-seq analyses
A. UMAP projection of available published 10x Genomics-derived fetal and adult retina datasets from the
Platform for Analysis of scEiad (PLAE) combined with our scRNA-seq data. PLAE data in purple, new
data in red. B. Our scRNA-seq data shown in isolation. C. Zoomed cone and rod regions from larger plot.
Black circles indicate most mature populations from current dataset (OPN1LW+ cone group and RHO+
late rod cluster)

22

Figure 2.5: Individual low resolution cluster plots
A. Low resolution clustering UMAP plot with cell type labels assigned in Figure 2.6. B. UMAP plot with
each individual cluster highlighted for clarity and labeled by cluster cell type.

23

Figure 2.6: Identification of primary cell types
A. Low resolution clustering UMAP plot with cell type labels. B. Expression of marker genes for RPC
(LHX2), Rod (NR2E3), S Cone (OPN1SW), and L/M Cone (THRB). Green boxes: Select regions shown
close up in C. which shows expression of later maturation photoreceptor genes. Top: Rod PDE6G and
RHO. Green arrow = late maturing RHO+ cell group. Bottom: Cone PDE6H and OPN1LW. Green arrow =
late maturing spatially distinct OPN1LW+ cell group. Green dotted box = Region of interest shown
enlarged in inset at bottom left.

24

with wider gating that also included rod precursors and RPCs (Figure 2.2A). Exclusion of cells
with <100,000 reads or expressing markers of retinal ganglion, amacrine, or horizontal cells
number of genes (POU4F1, POU4F2, POU4F3, TFAP1A, TFAP1B, ISL1) and without
photoreceptor lineage marker OTX2 yielded 794 single cells from 18 retinae, with averages of
3,750,417 uniquely aligned reads, 8,278 genes detected and 20,343 ENSEMBL transcripts
inferred per cell (Figure 2.2B-D). After batch normalization, single cell transcriptomes were
clustered using a smart local moving (SLM) algorithm and visualized in uniform manifold
approximation and projection (UMAP) plots that integrated cells across different retinae, single
cell isolation methods, and sequencing runs (Figure 2.3). When overlayed overlaid publicly
available fetal and adult retina scRNA-seq datasets generated by 3’ with end-counting
(https://plae.nei.nih.gov), the transcriptomes aligned with RPC, MG, rod precursor, and cone
precursor states with low representation of the most mature photoreceptor populations (Figure
2.4). The uneven distribution of late maturation cells likely indicates adult retina populations that
have no equivalent maturation state in my data.
Low resolution clustering generated six clusters segregated into distinct UMAP domains
(Figure 2.5). These clusters were assigned to cell types based on known photoreceptor marker
genes; one cluster was comprised of RPCs and early Mϋller glia, with specific expression of
LHX2, VSX2, SOX2, and SLC1A3, while five clusters were comprised of cells with
photoreceptor features, with expression of OTX2, CRX and diverse rod- and cone-specific
genes (Figure 2.5, 2.6, 2.7A,B). Two photoreceptor clusters expressing the rod determinant
NR2E3 (Figure 2.6B) were designated early maturing and late maturing rods, based on the
latter’s increased expression of rod phototransduction genes GNGT1, CNGB1, PDE6G, and
RHO (Figure 2.6C, 2.7C). Differential gene expression analysis confirmed that the late-maturing
rod cluster had significant upregulation of most of these rod genes as well as other
photoreceptor genes (NT5E, GNB1, SAMD7) (Figure 2.8B,C, Table S1), and the only enriched
gene ontologies were photoreceptor-related (not shown). Late rods also downregulated a small

25

number of genes, including CRABP2, DCT, and FABP7. The upregulation of numerous rod
phototransduction-related genes in the late maturing rod (LR) vs early maturing rod (ER)
clusters and the UMAP spatial separation imply that these clusters represent distinct rod
precursor states mainly defined by known rod photoreceptor genes. Cones segregated into
distinct S- and L/M-cone clusters, with differential expression of known cone subtype markers
(OPN1SW, THRB), as well as other genes recently identified in S-cones (CCDC136, UPB1)
(Lukowski et al. 2019, Peng et al. 2019, Kallman et al. 2020), and novel S-cone-specific genes
(MEGF10, NRXN3, ACKR3) (Figure 2.9, Table S2). Other less significantly upregulated features
of the S-cone cluster, such as SPINK4 and FAM19A4 were also found in rods or more mature
L/M cones. We also detected novel L/M cone-specific genes: FAAHP1, a lncRNA previously
implicated in a pain insensitivity phenotype (Habib et al. 2019), CNTN1, which has been
implicated in axon and synapse formation and cancer (Haenisch et al 2005, Nguyen-Ba-Charvet
and Chédotal 2014, Gu et al. 2020), and multiple genes encoding ribosomal proteins
represented by enriched ribosomal term ontologies (Figure 2.9B-D, Table S2).
In UMAP space, the THRB+ L/M-cone cluster separated into a large proximal population and a
small distal population distinguished by OPN1MW and OPN1LW expression (Figure 2.10A).
The latter were inferred to represent more mature L/M cones based on the increasing levels of
the cone maturation marker PDE6H (Figure 2.6C). Despite low cell numbers in the distal group
differential gene expression analysis revealed that the five-cell group had increased expression
of cone phototransduction gene GUCA1C, as well as PLA2G5, a member of a phospholipase
family thought to be involved in outer segment phagocytosis (Kolko et al. 2007), TTR which
codes for transthyretin, a retinol-binding protein, and MYL4, a myosin light chain gene
previously noted in retinal organoid L/M cones (Kallman et al. 2020) (Figures 2.10B,C, Table
S3), implying that it comprised a later maturation stage analogous to late maturing rods. Of
these four, only MYL4 had expression in other cell groups (2 rods and 3 S cones). The paucity
of OPN1MW/OPN1LW+ late maturing L/M cones is consistent with a prior analysis of similar-

26

27

Figure 2.7: Expression plots for known RPC, MG, PR, rod, and cone genes
A. Plots for genes shared between RPC and MG populations. B. Plots for genes expressed in all
photoreceptors, CRX and OTX2. C. Plots for genes associated with rod photoreceptors, ordered from
early to late maturation expression from left to right. D. Plots for genes associated with cone
photoreceptors, ordered from early to late maturation expression from left to right

28

29

Figure 2.8: Differential expression between early and late maturing rod clusters
A. UMAP plot with early maturing rod (ER) and late maturing rod (LR) clusters highlighted. Differential
expression between clusters was performed by Wilcoxon-rank-sum test and shown B. as a volcano plot.
Genes highlighted are most the significant and highest fold change as well as known photoreceptor
genes pAdj cutoff =0.05, log2FC. cutoff = |0.5|. C. Gene expression plots for rod marker genes
upregulated in late maturing rods (LR) and the top three upregulated features in early maturing rods (ER)
which are also expressed in S and L/M cones.

30

31

Figure 2.9: Differential expression between S and LM cone clusters
A. UMAP plot S Cone (S) and L/M Cone (LM) clusters highlighted. B. Volcano plot showing differential
expression between clusters as performed by Wilcoxon-rank-sum test. Genes highlighted have Padj <
10e-5 and log2FC > 1, except for MEGF10. Dashed lines indicate significant difference thresholds
Padj<0.05, log2FC. > |0.5|. C. Gene expression UMAP plots for S cluster enriched genes (CCDC136,
UPB1, MEGF10, NRXN3, ACKR3) and the two top LM cluster genes. D. LM cluster enriched
overrepresentation analysis (ORA) gene ontologies (genes used: Padj<0.05, log2FC. > |0.5|) reduced by
weighted set reduction to remove redundant terms.

32

Figure 2.10: Differential expression between late maturing LM cone group and LM cluster
A. Zoomed UMAP plot of L/M Cone (LM) cluster with OPN1LW+ late maturing population indicated by the
green arrowhead. B. Volcano plot of differential expression between the 5-cell late maturing population vs
the remaining LM cluster performed by Wilcoxon-rank-sum test. Genes highlighted are the highest pAdj
and Log2FC features. pAdj cutoff =0.05, log2FC cutoff = |0.5|. C. Gene expression UMAP plots for select
late maturing cone features.

33

age retinae (Lu et al. 2020) and with the paused cone maturation prior to post-natal foveal
morphogenesis (Hendrickson et al. 2012). The remaining cluster, here designated immature
photoreceptor precursors (iPRPs), extended from the RPC/MG group towards early cone
precursors and was comprised of cells expressing cone markers (THRB, GNAT2), rod markers
(NR2E3, GNAT1), or in some cases both (Figure 2.6, 2.7).
NRL and RXRG isoform usage differs between early rod and cone populations
Although cone and rod precursors were segregated, mRNAs encoding rod-determining factor
NRL and cone-determining factors RXR  and TR 2 were co-expressed in both rod and cone
populations (Figure 2.11A, 2.12A, 2.13A). However, interrogation of full-length scRNA-seq
isoform assignments and exon coverage plots revealed cell type-specific NRL, THRB, and
RXRG transcript isoform regulation.
For NRL, the transcript mapping program Stringtie assigned reads to previously defined
ENSEMBL transcripts and showed that the most highly assigned isoform (ENST00000397002),
which encoded full-length NRL, had a 7.0-fold higher average expression in early rod cluster
(ER) versus the LM-cone cluster (LM) (bootstrapped Welch’s t-test p<0.000001) (Figure 2.11A-
D). The second-most assigned isoform (ENST00000560550) used an alternative promoter and
first exon, encoded an N-terminally truncated NRL lacking the minimal transactivation domain
(Friedman et al. 2004), and had ~1.8-fold higher average expression in early rods (bootstrapped
Welch’s t-test p<0.0002) (Figure 2.11A-D). Moreover, the average ENST00000397002:
ENST00000560550 ratio was ~ 2.8:1 in early rods and ~ 0.72:1 in L/M cones indicative of a cell-
type-specific bias (Figure 2.11B). To assess the relative expression of these isoforms at the
single cell level without assigning reads to specific ENSEMBL isoforms, we compared reads
mapped to the unique first exons that define full-length vs truncated NRL protein expression
(Figure 11D). Specifically, the difference in read numbers mapping to each of the unique first
exons relative to the sum of reads mapping to both revealed a significant difference in isoform

34

use between L/M cones and both rod clusters. The truncated NRL first exon predominated in
developing L/M cones whereas the full-length NRL exon predominated in rods (Kruskal-Wallis
with post-hoc Dunn Test comparing ER-LM and LR-LM p < 0.0001) (Figure 2.11E).
For RXRG, ENST00000359842 was imputed as the most highly expressed isoform and
had 2.49-fold higher mean expression in L/M cones than in early rods (LM vs ER), whereas the
second-most abundant isoform ENST00000619224, which uses an alternative promoter and a
unique non-coding exon 3, had a 7.5-fold higher expression between the two groups (Figure
2.12A,B). These assignments reflected both overall higher levels of RXRG and increased use of
exon sequences that are unique to ENST00000619224 (Figure 2.12B-D, blue arrowheads).
Another L/M cone determining protein, TR 2, is encoded by THRB which was expressed
in a small set of rod cells in addition to broad cone signal (Figure 2.13A). The isoform containing
the TR 2-specific first exon, ENST00000280696, was imputed to be most highly expressed in
L/M cones as expected, and the TR 2-specific exon (Figure 2.13C, blue arrowhead) was
similarly used at far lower levels in ER and negligibly in LR. However, the neighboring (more 5’)
exon used in other THRB transcript isoforms that encode TR 1 (red arrowhead) was similarly
expressed in rod and L/M cone clusters which suggested low expression of this TR 1-encoding
isoform in late maturing rods as well (Figure 2.13B,C).

High-resolution clustering identified two post-mitotic photoreceptor trajectories and a
common rod/cone precursor state
To further define photoreceptor developmental states and trajectories, we subdivided the
initial six cell populations with higher resolution clustering, identified cluster-specific genes and
transcription factor regulons, and defined each cell’s rate and direction of transcriptomic change
using RNA Velocity. Regulons are produced by the program SCENIC and represent the overall
expression of a single transcription factor and its likely target genes (Van de Sande et al. 2020).

35

RNA Velocity is an algorithm that predicts a cell’s rate and direction of change towards a future
gene expression state based on gene-specific ratios of spliced versus unspliced scRNA-seq
reads (La Manno et al. 2018) (Figure 2.14, 2.15). Together these approaches allowed us to
predict a trajectory through each cell type identified based on the expression of known marker
genes and RNA Velocity, and to identify expressed genes and transcription factors that define
different cell populations of interest. Increased clustering resolution further divided the RPC, L/M
cone, early rod, and iPRP clusters (Figure 2.14A,B). Clustering resolution values are somewhat
arbitrary, but we chose the highest value that gave additional unique clusters that were not
disorganized or diffuse. The RPC/MG cluster became two groups with one displaying marker
genes PBK and E2F7, which are found in cycling neural progenitor cells (Dougherty et al. 2005,
Westendorp et al. 2012) (Figure 2.14C) and the other downregulated these features but shared
expression of SLC1A3 which is expressed in both RPCs and Muller Glia (Kanai et al. 2004). L/M
cones divided into four groups alongside the previous iPRP cluster which partially overlapped in
UMAP space and had overlapping marker gene expression. The final new cluster of cells,
derived from the low resolution iPRP and ER clusters, projected from the RPCs into the early
and late maturing rods, labeled transition rods (TR) (Figure 2.14A,B).
RNA Velocity generated individual vectors of predicted gene expression change for each
cell within UMAP space, which were then averaged within gridded areas to create a generalized
vector for all cells in that square (Figure 2.15A, B). We identified cell trajectories consistent with
RPC differentiation and PR maturation, where cells progressed out of the RPC region and
through both the rods and cones in a central to distal pattern (Figure 2.15B) similar to what was
predicted by observing photoreceptor marker genes (Figure 2.6B). Based on this clustering and
RNA Velocity, three regions of the UMAP plot were chosen for closer investigation: RPC/MG
and TR and iPRP cells which transition between RPCs and photoreceptors (green box), iPRP
cells positioned between early rods and cones (red box), and early maturing and late maturing
L/M cones (purple box).

36

37

Figure 2.11: Differential NRL isoform use in rod and cone cell clusters
A. Expression of NRL and the two most highly assigned NRL isoforms (ENST00000397002 and
ENST00000560550). B. The average NRL isoform assignments for low resolution clusters presented as
total transcript counts (top) and as a percentage of the total counts (bottom). C. Violin and box plots of
ENST00000397002 and ENST00000560550 assignments by cluster. D. Read coverage plot across
ENSEMBL-annotated NRL exons. Top, average read counts (FPM) for cells in each cluster. Bottom, exon
plot for all ENSEMBL transcripts of NRL, with coding sequences colored in dark blue. Below are the
predicted transcript structures for ENST00000397002 and ENST00000560550, numbered for the amino
acid positions of interest. Blue arrowhead = start of first exons for inferred isoform-specific promoters (P1
and P2). E. Relative difference plot of reads mapping to the first exons encoding truncated or full-length
NRL in each cell, according to cluster. Raw read counts were determined using DEXseq. The exon use
relative difference is the difference in reads mapping to the truncated (T) and full-length (F) NRL first
exons (T-F) divided by the sum of both (T+F). Values >0 indicate more reads assigned to the truncated
isoform first exon, while values < 0 indicate more reads assigned to the full-length isoform first exon.
Kruskal-Wallis Test p-value = 0.0001, post-hoc Dunn-Test with Benjamini-Hochberg correction: LM to ER
and LM to LR p-value <0.0001.

38

39

Figure 2.12: Differential RXRG isoform use in rod and cone cell clusters
A. Expression of RXRG and the two most highly assigned RXRG isoforms (ENST00000359842 and
ENST00000619224). B. The average RXRG isoform assignment for low resolution clustering presented
as (top) total transcript counts assigned to each isoform and (bottom) as a percentage of the total counts.
Fold change in average transcript counts between rod and cone clusters: ENST00000359842: 2.49-fold
higher in LR over ER. ENST00000619224: 7.5-fold higher in ER over LM. C. Violin and box plots of the
expression of ENST00000359842 and ENST00000619224 by cluster. D. Read coverage plot across
ENSEMBL-annotated RXRG exons. Top shows average read counts (FPM) for cells in each cluster.
Bottom shows the predicted transcript structures for ENST00000359842 and ENST00000619224 with
coding sequences in blue. Blue arrowhead = exon regions unique to the truncated protein transcript
ENST00000619224 indicated in ER and LM for comparison.

40

Figure 2.13: Differential THRB isoform use in rod and cone cell clusters
A. THRB expression plot. B. Violin plot of TRβ2 isoform ENST00000280696. C. Read coverage plot
across ENSEMBL-annotated THRB exons. Top shows average read counts (FPM) for cells in each
cluster. Bottom shows the predicted transcript structures for ENST00000280696 and a different isoform
utilizing different 5’ exons, with coding sequences in blue. Blue arrowhead = Unique ENST00000280696
TRβ2 first exon and coding sequence, indicated in LM and LR clusters for comparison. Red arrowhead:
THRB exon upstream used by TRβ2 (ENST00000280696) first exon indicated in LM and LR clusters.

41

42

Figure 2.14: High resolution clustering and marker genes
A. Side-by-side UMAP plots colored by low and high-resolution clustering. B. Each high-resolution cluster
is highlighted on individual plots, labeled by cell type or sub-type. L/M cones divide into 4 clusters
approximately numbered from less to more mature (1-4). C. Marker gene plot for high-resolution clusters.

43

Of the two RPC/MG clusters, one showed significantly increased expression of markers
cell cycle phases G1/S (CCNE2+) and G2/M (CCNA2/CCNB2+) along with the marker genes
PBK and E2F7 (Figure 2.14C, 16A,B), further supporting this as a cycling RPC (cRPC)
population. The second cluster was spatially segregated, lacked those cell cycle genes but
shared the G1 RPC and Mϋller cell characteristics CCND1, SLC1A3, and RLBP1 with the
first cluster and was designated RPC/MG. Differential expression (Figure 2.16C,D) also
revealed up-regulation of cell cycle-related genes like TOP2A and ontologies like G2/M
checkpoint and E2F targets in the RPC cluster, but there were no significantly upregulated
genes or ontologies in the RPC/MG population. This is consistent with early MGs resembling
quiescent RPCs that are distinguished by their lack of cell cycle-related gene expression
(Walcott & Provis 2003).
High resolution clustering also resolved subsets of iPRP and TR cells that were
positioned near the RPCs in the UMAP plot (2.17A). RNA Velocity analysis suggested that
these cell groups flowed towards the larger iPRP population, and directly towards early maturing
rods, respectively (Figure 2.17A). Compared to RPCs and MGs, RPC-localized iPRP and TR
cells had minimal expression of cell cycle-related cyclin RNAs, reduced expression of the RPC
markers PAX6, LHX2, and VSX2, and increased expression of the photoreceptor determinants
OTX2 and CRX (Figure 2.17B), consistent with their being immediately post-mitotic PR
precursors. The immediately post-mitotic iPRP cells more often expressed LM-cone
determinants ONECUT1 and THRB, the cone differentiation marker GNAT2, and lacked the rod
fate determinant NR2E3 and the early differentiation marker GNAT1, indicative of a cone-bias
(Figure 2.17C). In contrast, immediately post-mitotic TR cells more often expressed NRL,
NR2E3 and GNAT1, suggestive of rod bias (were any of these significant or have a suggestive
p value in t-test?).
SCENIC identified the top five most specific transcription factor regulons for each
cluster, and further clarified the identities of the four RPC-localized clusters (Figure 2.17D,

44

S2.1). RPC/MG regulons were best specified by PAX6 and EGR3, transcription factors known
to be associated with RPCs and MGs (Joly et al. 2011) and shared with the cRPC cluster.
Conversely, the E2F2 and E2F3 regulons, known cell cycle transcription factors, were
significantly expressed in cRPCs (Fig 2.17D). These RPC/MG and cRPC regulons were low or
absent in both the immediately post-mitotic iPRP and TR groups (Figure 2.17D). Both post-
mitotic groups showed THRB and NRL regulon signals; however, iPRP-clustered cells showed
greater THRB regulon signal than TR cells, while TR cells had a higher NRL regulon signal
(p<0.05 for both). In sum, high resolution clustering resolved cycling RPC and RPC/MG
populations distinguished mainly by cell cycle features, as well as two post-mitotic PR precursor
groups with higher cone-related gene expression and regulon activity in immediately post-
mitotic iPRPs and higher rod-related gene expression and regulon signal in TRs. The biased
regulon activity in immediately post-mitotic cells suggested that iPRP- versus TR-directed fate
decision were already made or made rapidly in the early post-mitotic period.
In addition to the immediately post-mitotic iPRPs located near RPCs, the iPRP cluster
included cells bridging the region between early maturing cones and early maturing rods (Figure
2.15B(red box), 16C). RNA Velocity indicated that the postmitotic iPRP trajectory flowed from
the RPC region towards this bridge region and then bifurcated towards rods or cones, with
individual cells expressing cone markers (GNAT2, THRB), rod markers (GNAT1, NR2E3), or
both (Figure 2.18A). Similarly, cells with NRL and THRB regulon signals intermix with a few
cells showing both regulon signals (Figure 2.18B, arrows), suggestive of common cone-rod
precursors (Ng et al. 2011). By examining expression patterns of iPRP marker genes (Figure
2.14B) we noted that CHRNA1 was mainly expressed in the bridge region whereas S100A6 was
expressed more widely, ONECUT1 was expressed in immediately post-mitotic PRPs, cone-
directed iPRPs in the bridge region, and early maturing L/M cones, and lncRNA CTC-378H22.2
was expressed in a tightly restricted zone of THRB+, GNAT2+ cone precursor cells (Figure
2.18C). ATOH7 and DLL3, which were previously proposed to define transitional PR precursors

45

and promote cone fate determination (Sridhar et al. 2020, Clark et al. 2019, Feng et al. 2010,
Zhang et al. 2018), were also detected in the bridge region, yet ATOH7 was largely restricted to
cone precursors similar to ONECUT1 and DLL3 had wider cone and rod precursor expression
similar to S100A6 (Figure 2.18D). Overall, CHRNA1 was identified as the most specific marker
of potential common cone and rod precursors or their immediate progeny, while other marker
genes were also detected in slightly more mature early rod and cone precursors.
Our detection of ONECUT1 in post-mitotic cone-directed iPRPs was consistent with evidence
that ONECUT1 promotes THRB expression during L/M cone fate determination and persists in
post-mitotic cone precursors, as previously shown in chick and mouse (Emerson et al. 2013).
Interestingly, we also detected OLIG2 and LHX9 regulon signals in these same ONECUT1+
early cone-directed iPRP cells immediately prior to broad THRB regulon upregulation in the
larger L/M cone population (Figure 2.18B,E). OLIG2 has been shown to mark an RPC
subpopulation that is preparing to differentiate and produce cones through the action of OTX2
and ONECUT1 on THRB (Hafler et al. 2012, Emerson et al. 2013), but evidence of its
downstream activity, via regulon signal, was not previously observed in post-mitotic cone-
directed cells. Our observations indicate that components of the OLIG2/ONECUT1/THRB axis
previously described to be active in RPCs are also present and active in human post-mitotic PR
precursors, coinciding with LHX9 regulon activity. The existence of the cone- and rod-directed
iPRP population and a distinct TR population suggest that the iPRP and TR trajectories
represent different routes to rods, with iPRP-derived PR precursors marked by CHRNA1+ and
initially characterized by co-expression and/or activity of the cone-related OLIG2, ONECUT1,
and THRB, and the rod-related NR2E3.

.

46

Figure 2.15: RNA Velocity from retinal progenitor cells through maturing photoreceptors
A. High resolution clustering plot and B. RNA Velocity clustering plot with per cell and grid averaged
velocity vectors. Regions of interest are indicated in A and B: green box = RPC and co-localized iPRP
and TR cells, red box = iPRP bridge population between rods and cones, purple box = early maturing L/M
cones.

47

48

Figure 2.16: Cell cycle features distinguish high resolution cycling RPC (cRPC) and RPC/MG
clusters
A. Left: zoomed view of RPC region of interest with cRPC and RPC/MG highlighted. Right: Gene
expression violin plots for cell cycle genes (G1/S = CCND1,CCNE2. G2/M = CCNA2,CCNB2) and
RPC/MG genes (SLC1A3, RLBP1). Significance of differential gene expression determined by Mann-
Whitney Test normal p-value approximation (* = p<0.005, *** = p<0.0005). The same genes are shown in
B. as zoomed UMAP plots C. Differential expression between RPC/MG and cRPC with labels for genes
p>10e-4 except for RPL34. pAdj cutoff =0.05, log2FC cutoff = |0.5|. D. Overrepresentation analysis for
genes upregulated in cRPC above p and log2FC cutoff.

49

50

Figure 2.17: RPC-localized iPRP and TR cells upregulate cone and rod features
A. Left: Zoomed plot of the RPC region of interest with iPRP and TR cluster cells highlighted. All cells
below black line are evaluated in this figure. Right: Same plot area showing RNA Velocity per cell with all
clusters colored. B. Violin plots for cRPC (CCNE2, CCNA2), RPC/MG (PAX6, LHX2, VSX2) and
photoreceptor (OTX2, CRX) genes in RPC region of interest (the iPRP and TR cells below line in A). C.
Violin plots for cone (THRB, GNAT2, ONECUT1) and rod (NR2E3, NRL, GNAT1) marker genes. D.
SCENIC regulon AUC violin and box plots for four RPC region of interest clusters, selected from top
regulon specificity scores for RPC/MG (PAX6 and EGR3) and cRPC (E2F2 and E2F3). The LM cone
regulon THRB and rod regulon NRL also displayed. Significant differences between clusters determined
by Kruskal-Wallis Test and if significant post-hoc Dunn test with Benjamini-Hochberg correction: * =
p<0.05, *** = p<0.0005.

51

52

Figure 2.18: Bridge population has features of a common photoreceptor precursor state adjacent
to early cone ONECUT1/THRB pathway expression
A. Zoomed bridge region plots for RNA Velocity and rod/cone marker gene expression. Red circles:
GNAT1/NR2E3/GNAT2+, Green circle: GNAT1/NR2E3/GNAT2/THRB+, Blue circle:
NR2E3/GNAT2/THRB+, Black circle: NR2E3/GNAT2+. B. Regulon AUC UMAP plots for NRL and THRB.
Green box around region of interest shown right; arrows indicate cells with signal for both regulons. C.
iPRP marker gene plots with boxed region of interest shown in inset. D. Transition state marker genes
ATOH7 and DLL3 (Sridhar et al. 2020) expression plots, boxed region shown in inset. E. UMAP plots of
top two regulon specificity score (RSS) iPRP regulons.

53

Features of early maturing L/M cone maturation trajectory
High resolution SLM clustering subdivided maturing L/M cones into four subpopulations, LM1,
LM2, LM3, and LM4 (Figure 2.14A,B, 2.19A). Gene expression differences between these
groups were subtle, with higher expression of the cone maturation marker PDE6H and novel
markers RTN1, ACOT7, OLAH in the most distal cluster LM4, and specific expression of NPFF
and OPN1LW/OPN1MW+ in the most distal LM4 late maturing cone subgroup (Figure 2.14C,
2.19B,C). Most of the transcription factor regulons identified by SCENIC also failed to
discriminate these cone subgroups, exemplified by the similar activity of the cone specific THRB
and ISL2 regulons, with only LM1 showing significant ISL2 differences to LM2 and 4 (Figure
2.19D). However, the LHX3 regulon showed increasing activity across LM1-4 and between early
and late maturing rods (Figure 2.19D), suggesting a possible role of this homeobox protein in
cone maturation.
RNA Velocity revealed diverging trajectories from photoreceptor precursors towards the
most mature cells in LM4 (Figure 2.15B, 2.20A), traversing the upper and lower edges of the
L/M cone precursor population, where some cells divert towards a central low velocity region.
These trajectories were distinguished by fetal age, with the upper route primarily comprised of
FW13 cells, the middle route mainly FW15-16 cells, and the lower route primarily FW17-19
(Figure 2.20A,B), implying that subtly different fetal cone maturation pathways were captured
depending on fetal age . When cell groups were defined by these upper and lower trajectories
and bounded by the low velocity region, differential expression analyses identified four
differentially expressed genes (Figure 2.20D-F) which were all upregulated in upper
trajectory/younger cells. These were F10, which encodes a coagulation factor, MFAP4, which
encodes a secreted fibrinogen-like protein associated with claudication and ischemia (Ea
Hemstra et al. 2018) and RP13-143G15.4, an antisense RNA within PDE7B. ACTB while
differentially expressed was too widespread and highly expressed to be informative.

54

To further evaluate gene expression changes across L/M cone maturation, L/M cones
were pseudotemporally ordered and pseudotime-correlated gene modules defined with Monocle
(Figure 2.21A, Table S4). The Monocle trajectory initiated in the immature PR precursors and
terminated at the OPN1LW+ cells, passing through the low velocity population. The analyses
revealed five co-regulated gene modules (Figure 2.21A) of which modules 1 and 2 had higher
expression in less mature L/M cones while modules 3, 4, and 5 increased in more mature cells.
However, the genes represented in individual modules had few significant ontologies; Module 2
had borderline significant stem cell proliferation ontology (FDR =0.057) (Figure 2.21B, Table
S4), while module 4 had significant phototransduction-related ontologies. This is supported by
the top 10 genes in module 4 best correlated with pseudotime, which were mostly known
photoreceptor/phototransduction genes (Table S4). Interestingly, across three modules (1,3,5)
the top correlated genes included multiple lncRNAs, several previously identified by prior
analyses: CTC-378H22.2 and HOTAIRM1 in module 1, RP13-143G15.4 in module 3, and CTD-
2034I21.2 in module 5 (Figure 2.21C,D). Together, these lncRNAs defined a sequence of
UMAP regions marked across early maturing cones. In contrast, we did not identify protein-
Coding RNAs that as clearly defined areas across the UMAP plot L/M cone region.

Human cone cells express SYK, which impacts proliferative response to pRB loss
Knowledge of human embryonic photoreceptor development is expected to provide insights into
the origins of developmental diseases. Retinoblastoma is a developmental cancer that
originates in pRB-deficient maturing (ARR3+) cone precursors based on the pRB-deficient cone
precursor’s unique abilities to proliferate and form retinoblastoma masses (Xu et al. 2014, Singh
et al. 2018). To identify early maturing cone precursor features that contribute to retinoblastoma
we compared gene expression between early maturing cones (which proliferate in response to
pRB loss) and early maturing rods, which fail to proliferate. This analysis identified 422 cone-
upregulated and 119 cone down-regulated genes (p<0.05, log2FC>|0.4|) (Figure 2.22A, Table

55

S5). Overrepresentation analysis of cone enriched genes revealed similar LM cone enrichment
of ribosome, translation, and protein targeting ontologies as between S and LM clusters (Figure
2.9D), but we also noted significant upregulation of MYC target genes (Figure 2.22B), consistent
with the upregulation of MYCN in this cone population (Figure 2.22C) (Xu et al. 2009) albeit with
relatively low fold change (log2FC = 0.54) (Figure 2.22C). Concordantly there was a
commensurate L/M-cone-specific activity of the MYCN regulon identified in SCENIC analyses
(Figure 2.22D). In examination of other low fold-change genes, we also noted cone-specific
upregulation of SYK, a non-receptor tyrosine kinase (Figure 2.22A,C, 2.23A, Table S5), that has
been shown to act upstream of MYC (Qu et al. 2018). Whereas SYK was previously implicated
in retinoblastoma genesis and proposed to be induced in response to pRB loss (Zhang et al.
2012), immunostaining (performed by Dr. Yeha Kim) revealed SYK protein in ARR3+ cones
(Figure 2.23B) (Zhang et al. 2001, Li et al. 2003). SYK was detected in peripheral, parafoveal,
and foveal regions at FW 16 but was excluded from the foveal cones at FW18, though still
expressed in an unknown bright INL cell type. Loss of cone SYK implied that expression
declines with terminal maturation and consistent with the lack of detectable SYK in the cones of
normal retina adjacent to a retinoblastoma tumor (Zhang et al. 2012). To determine whether
SYK contributes to retinoblastoma initiation Dr. Sunhye Lee transduced dissociated fetal retinal
cells with an RB1-directed shRNA (shRB1-733) or a scrambled control shRNA in order to
induce cone precursor proliferation (Xu et al. 2014) and examined effects of the specific SYK
inhibitor GS-9876 (GS, Gilead) (Blomgren et al. 2020) at 0.5 to 5uM. GS treatment suppressed
the proliferative response to pRB knockdown (Figure 2.23C) implying that intrinsic SYK
expression and activity is critical to the early maturing cone precursor proliferative response to
pRB loss. Thus, our single cell RNA-seq analyses identified an additional component of the
cone-specific pRB-related proliferation program which influences the L/M cone precursors’
ability to re-enter the cell cycle at the earliest stages after pRB loss.

56

57

Figure 2.19: High resolution L/M cluster marker genes and regulons
A. Zoomed plot of L/M cone region of interest. B. Violin plot of LM cluster marker genes and cone genes
PDE6H and OPN1LW. C. Gene expression plot of NPFF showing localization to late maturing cone group
(blue arrowhead). D. violin and box plots for highest three RSS regulons in all LM clusters. Significance
between clusters by Kruskal-Wallis Test and if significant post-hoc Dunn test with Benjamini-Hochberg
correction: * = p<0.05, *** = p<0.0005.

58

Figure 2.20: Velocity identified two trajectories through L/M cones biased by age
A. Grid-averaged RNA Velocity plot for L/M cone region of interest. B. Plots with samples colored by fetal
week of source retina, split into FW13, FW15 and 16, and FW17-19. Left line indicates where L/M cone
region of interest begins, right line is a shared axis across which samples shift from left to right with age.
C. Upper and lower RNA velocity trajectories bounded by blue and yellow, respectively. Both start in
iPRP, pass the edge of the low velocity region (circle) and then end in LM4 excluding the OPN1LW+
population. D. Differential expression between upper and lower regions identified four genes significantly
upregulated in upper cells, shown in E. as expression UMAP plots.

59

60

Figure 2.21: L/M cone pseudotime reveals progression of lncRNA gene expression with
maturation
A. Left: Pseudotime trajectory through L/M cone population derived with Monocle3, root cell is in late-
maturing cone cluster to define expected endpoint. Right: Heatmap of genes significantly correlated with
pseudotime divided into five shared expression pattern modules. Two modules show increased
expression in early maturation (1,2) while three show increased expression in later maturation samples
(3-5). B. Top gene ontology term with the most significant FDR and highest enrichment ratio for each
module (shared with heatmap). Italicized terms are nearly significant (FDR<0.05), bolded are significant.
C. Table of lncRNAs among 10 genes most highly correlated with pseudotime for modules 1, 3, 5. Red =
lncRNA, bold = photoreceptor/phototransduction gene. D. lncRNA expression plots for genes in C.

61

Figure 2.22: MYC targets and associated genes upregulated in LM cones over rods
A. Volcano plot of differential expression (Wilcoxon-rank sum test) between low resolution rod cluster ER
and cone LM. Labeled genes where pAdj <10e-32, except for GNAT1, PDE6H, SYK, MYCN. pAdj cutoff
=0.05, log2FC cutoff = |0.4|. B. Overrepresentation analysis for cone-enriched genes p-value <0.05,
log2FC ≥ |0.4|. C. Upregulation of MYCN, correlating with MYC target ontology, as well as SYK, which
interacts with MYC

62

Figure 2.23: SYK is expressed in fetal cones and inhibition decreases cell cycle entry after RB1
knockdown
A. Expression plot of SYK B. SYK immunostaining of FW18 and 16 retinae. Nuclear and cytoplasmic
signal in cones expressing ARR3 except in foveal region of older tissue where it diminishes. Green arrow:
ARR3+, SYK+. White arrow: ARR3+, SYK- Scale bar = 25um C. Quantitation of KI67+ cells among
GFP+/RXRγ+ cone cells in dissociated FW16.5 fetal retina after knockdown of RB1 with shRB1-733-GFP
construct and SYK inhibition with GS-9876. Samples were treated for 12 days in culture post-knockdown.
Two individual experimental replicates shown, total cells counted per experiment: Exp 1: Scramble control
n=237, shRB1 samples: DMSO n=366, 0.5uM n=107, 1uM n=159, 2.5uM n=255, 5uM n= 216. Exp 2:
Scramble control n=183. shRB1 samples: DMSO n=258, 0.5uM n=132, 1uM n=138, 2.5uM n=118, 5uM
n=158.

63

Discussion
The molecular mechanisms of photoreceptor differentiation and maturation are of considerable
interest, as to treat the plethora of retinal injuries, illnesses and genetic conditions which
negatively impact photoreceptors requires a strong understanding of the basic development.
scRNA-sequencing has recently expanded understanding of human retinal development,
generating large scale 3’ end counted datasets for all retinal cell populations at both fetal and
adult stages (Macosko et al. 2015, Lu et al. 2019, Lukowski et al. 2019, Liang et al. 2019, Voigt
et al. 2019, Sridhar et al. 2020). The most common focus among these and previous bulk RNA-
seq studies is on divining changes that mediate RPC differentiation, though some have
discussed the change in photoreceptor genes across varied ages and regions of the retina
(Hoshino et al. 2017). However, these data all lack full length transcript information, which
provides a greater read depth as well as insight into isoform usage, especially when working
with a focused cell population instead of the entire tissue.
In this chapter, our aim was to dissect cone photoreceptor development at a greater
level of transcript detail than previously presented to identify notable features of states and state
changes not previously described. We generated single cell transcriptomes from 18 retina for
794 fetal retina cells, enriching to capture a high percentage of cone photoreceptors, which are
one of rarer retinal cell populations, outnumbered by rods by approximately 20 times (Masland
2001). Our method of enrichment successfully captured not only cone cells but groups of rod
and RPCs. Initial comparison was done against the compiled fetal retina scRNA-seq sets in the
Platform for Analysis of single cell Eye in a Disk (PLAE), a recently developed tool from the NEI.
This demonstrated our samples were representative of major photoreceptor populations
captured in previous experiments (Figure 2.4), however the number of more mature cones was
very low, and our mature rods only covered one small part of a larger rod cluster present in
PLAE datasets. The small number of cones in our dataset that cluster near mature PLAE-
derived cones matched the L/M opsin positive late maturing cone cell quintet observed in our

64

data alone, suggesting these five are the best representative of that PLAE region. Due to the
delayed maturation of the fovea, L/M opsin is expressed most highly in a perifovea cell
population until ~1-2 years of age (Abramov et al. 1982, Hendrickson and Drucker 1991). Our
data suggests that the most opsin positive regions of the collected retinae were poorly enriched
in our FACS-sorted dissociated retinae samples, so most of our cone cells covered a pre-opsin
early to mid-maturation state. Due to the nature of FACS isolation via surface markers, it was
not possible to guarantee all states of cone or rod maturation were represented.
Among established photoreceptor markers, NRL and RXRγ are well described as unique
features of rod or cone cells, but not both. The expression of both genes across all
photoreceptors (Figure 2.11, 2.12) was unusual and inconsistent with typical protein patterns.
Rod cells preferentially used a unique exon that contained minimal transactivation domain
(MTD), which is required for NRL to perform transcription initiation functions. A previously
described truncated isoform, DD10, was shown to antagonize the function of full length NRL
protein (Rehemtulla et al. 1996). DD10 utilizes a small region of the MTD-coding exon that the
truncated transcript does not, and the truncated isoform also uses a unique non-coding first
exon; however, the behavior of DD10 suggests that the increased use of our different truncated
isoform in cones may serve to inhibit NRL function. This inhibition coupled with lower overall
NRL expression and potentially also translational regulation appears to be in lieu of completely
suppressing NRL transcription in developing cone cells. RXRG and THRB also demonstrated
differential use of predicted isoforms and exons. For RXRG, the isoform with a larger coding
region was used proportionally more highly in rods than cones, however it is unclear how this
extended coding region may influence canonical RXRG protein expression in rods. THRB was
minimally expressed in rod cells but these data indicated a lack of TRβ2 isoform and first exon
use in those cells, while other exons that are shared with TRβ1 are still expressed.
Most of the single cone transcriptomes gathered represented cone states from early
post-mitotic up to opsin gene expression, presenting as a continuous population until OPN1LW

65

was expressed. Of greatest interest to us were the gene expression features and changes that
occurred transitioning into and through the photoreceptor state. Recent publication by Sridhar et
al. beautifully demonstrated post-mitotic states in the transition from RPC to photoreceptor,
where ATOH7+ cells, which have previously been described as transiently expressed in post-
mitotic cells, formed the earliest born T1 transition population, and the last state T3 occurred
between bipolar cells and photoreceptors and expressed high FABP7, OTX2, and DLL3. In our
trajectory, cells transiting from the RPC region were directed towards rods or the iPRP bridge
region that appeared to be populated by common rod and cone photoreceptor precursors;
however, while cells from both groups (TR, iPRP) localizing near the RPCs expressed OTX2
(Figure 2.17B) only TR cells showed any ATOH7 or DLL3 expression (Figure 2.18D) until the
iPRP bridge region. There, focal ATOH7 and slightly broader expression of DLL3 were detected
in and around the bridge (Figure 2.18D), demonstrating the features of the T1 and T3 states
(Sridhar et al. 2020). Our data allowed us to identify a new feature of these shared PR precursor
cells, CHRNA1, which closely matched ATOH7 expression but was also found in additional
early rod-directed cells (Figure 2.18C). The function of a nicotinic acetylcholine receptor subunit
in early photoreceptor precursors is currently not clear, though other subunit proteins have been
detected across Rhesus monkey retina (Liu et al. 2009), so additional work is necessary to
clarify its tissue expression pattern and action.
Two models have been described for determining the timing of photoreceptor cell fate
determination, in RPCs or post-mitotically. States observed in mouse retina similar to those from
Sridhar et al. have been labeled as “neurogenic” progenitors (Clark et al. 2019). An OLIG2+
subset of RPCs with high ONECUT1 and OTX2 has been observed to rapidly differentiate and
only make cone and horizontal cells, with the final fate decision made by loss of ONECUT1 after
differentiation (Emerson et al. 2013). ONECUT1 was also shown to promote THRB expression
and thus push downstream L/M cone features even as it downregulates. In contrast to this
model of RPC-decided fate, cells with hybrid TRβ2 and NRL expression have been observed in

66

mouse retina, where L/M cone or rod fate is decided by which is retained while S-cone fate is
acquired if neither remains (Ng. et al. 2011). We observed gene and regulon expression for
THRB, ONECUT1, and OLIG2 in the bridge PR precursors directed (according to RNA Velocity
analyses) towards early cone cells but not in those directed towards early rods, showing the
RPC-specified fate determinants were retained as cones exited a shared post-mitotic cone and
rod precursor population. This may indicate that transitioning from a bipotential photoreceptor
precursor state to a committed early cone precursor requires continued involvement of the
ONECUT1/OTX2 activation of THRB. It is also possible, perhaps depending on the timing of cell
birth, that a direct photoreceptor progeny or a bipotential precursor may be formed. A direct cell
route to early rods appears in these data, similar but separate from rods born from the common
photoreceptor precursor cells. A direct rod trajectory from RPCs has been described in late
development when OLIG2+ RPCs no longer express ONECUT1, and it raises the possibility
that both cone and rod cell fates can be determined in the RPC stage as well as in a post-
mitotic shared precursor state, possibly changing depending on whether one or both cell types
are being born at a given point in development. For example, the earliest rod cells are born
alongside cone photoreceptors and – as rods were evolutionarily derived from cones (Lamb,
Collin, and Pugh Jr. 2007) – the initial switch to producing rods may be made by altering the
cone fate decision in a shared photoreceptor precursor. Then at later stages of retinogenesis
when cones are no longer being produced, rod fate may be directly determined in RPCs.
L/M cone cells formed one larger continuous population which we parsed using several
methods to identify subtle age-related changes. The regulon activity for ISL2 corroborates
previously observed ISL2 expression in human and chick cones (Lu et al. 2020); yet LHX3 has
only been described in bipolar cells (Balasubramanian et al. 2014). The most distinctive gene
features across cone maturation were the shifting expression patterns of lncRNAs. The
transition from PR precursors into the cone clusters was marked by focal expression of lncRNA
CTC-378H22.2 in cones. HOTAIRM1 has recently been described in human fetal cone cells but

67

not in mouse (Lu et al. 2020) and this was narrowed to earlier maturing cones in our data.
RP13-143G15.4 and CTD-2034I21.2 both upregulated in more mature L/M cones; interestingly
the neighboring gene CTD-2034I21.1 has been previously observed enriched in cone cells
(Welby et al. 2017). These non-coding genes described a general progression from early to
later maturation in L/M cones, but further analyses via in situ hybridization are needed to assess
whether they define distinct developmental stages in tissue sections.
Finally, we used the available cone and rod data to look for major gene expression
differences in early rod versus cone photoreceptors to identify additional factors that would
impact human cone cell sensitivity to RB1 loss (Xu et al. 2014). The most significant differences
in expression included well known markers of each cell type (NR2E3, NRL, THRB, GNAT2)
(Figure 2.22A). However, MYC-family caused significant but low-level upregulation of a group of
genes in cones. MYCN RNA was similarly weakly increased in cones versus rods at the RNA
level, however MYCN protein has dramatically higher expression in developing cones than in
other retinal cell types and is capable of influencing retinoblastoma (Xu et al. 2014) as well as
triggering tumor formation even without RB1 loss when amplified (Rushlow et al. 2013). This
suggests that the low fold change difference in gene expression may be enough to enable
proliferative response with pRB loss. Similarly, SYK, a gene shown to affect MYC expression
and thus its downstream pathways, showed low fold change increase in RNA expression but
distinct cone protein expression (Figure 2.23A,B). Previous work did not detect SYK expression
in healthy tumor associated retina (Zhang et al. 2012) but detected high expression in
retinoblastomas. We were able to demonstrate that ARR3+ cone photoreceptors express SYK
in early maturing cone precursors but not in the fovea after cones reached a certain point in
maturation. SYK inhibition also reduced the ability for cultured cone cells to re-enter the cell
cycle after RB knockdown, indicating that not only does SYK upregulate in retinoblastoma but
that its native expression in cones contributes to their ability to proliferate and form tumors,
possibly through activation of MYC proteins and their downstream targets.

68

Overall, this study provides a large repository of photoreceptor-enriched full-length
transcriptomes with deep levels of sequencing not achievable with high throughput, droplet-
based methods. These data show aspects of the early stages of rod and cone cell differentiation
with evidence that rods may develop directly from RPCs as well as from a shared post-mitotic
photoreceptor precursor state. Early cones exiting this shared state also express components
previously identified as cone fate-determining in RPCs, but not associated with exit from a post-
mitotic photoreceptor precursor state as they are currently modeled (Ng. et al. 2011). lncRNAs
served to mark both photoreceptor populations directly out of the shared precursor state and
describe maturation stages across otherwise apparently continuous early maturing L/M cones.
Finally, we identified SYK expression in developing human cone cells and showed its ability to
affect cone proliferative response to RB1 loss, supporting that it is not just upregulated in
retinoblastoma but important in sensitizing cone cells to the oncogenic effects of pRB loss.

Materials and Methods
Retinal tissue, dissociation, and single cell isolation: Following informed consent, fetal eyes
were obtained from authorized sources with approval by the University of Southern California
and Children’s Hospital Los Angeles Institutional Review Boards. Fetal age was determined
from fetal foot length and crown-rump length. Retinae were dissected and dissociated in ice-
cold PBS as previously described (Xu et al. 2014). Briefly, retina was removed and placed in
200ul Earle’s Balanced Salt Solution (EBSS) in a 6-well plate on ice. 2ml of 10u/ml Papain
(Worthington LK003176) solution was added and then incubated at 37°C for 10 minutes,
followed by pipetting with a 1000ul pipet tip to break up large pieces and then additional 5-
minute incubations until cells were dissociated to single cell level. Additional 1-2 ml Papain was
added after 20 minutes if dissociation was insufficient (~40 minutes total). Retina culture media
[Iscove’s Modified Dulbecco’s Medium (IMDM; Corning), 10% FBS, 0.28 U/mL insulin, 55uM β-
mercaptoethanol (Sigma-Aldrich), glutamine, penicillin, and streptomycin] was used to stop

69

enzyme activity and cells were centrifuged in 14ml round bottom tubes at 300g, 4°C for 10
minutes. Supernatants were centrifuged at 1100g for an additional 3 minutes. After
resuspension in HBSS, tubes were recombined, and cells counted.

FACS isolation of single cells and cDNA synthesis: Single cells were FACS-isolated as
described previously (Xu et al. 2014) with some differences. Briefly, after removing cells for an
unstained control, dissociated cells were centrifuged and resuspended in a 5% fetal bovine
serum in phosphate-buffered saline (FBS/PBS) with CD133-PE (Miltenyi Biotec 130-080-801),
CD44-FITC (BD Pharmingen 555478) and CD49b-FITC antibodies (BD Pharmingen 555498) to
a final concentration of 10,000 cells/ul. After incubation at room temperature (RT) for 1 hour,
samples were diluted in 5% FBS/PBS, centrifuged as above, and resuspended in 300-400ul
1%FBS/PBS containing 4′,6-diamidino-2-phenylindole (DAPI). Single cells were sorted on a BD
FACSAria I at 4°C using 100um nozzle in single-cell mode into 1.2ul lysis buffer droplets on
parafilm-covered glass slides (Lee et al. In preparation). Eight sample droplets were collected at
a time and then transferred on ice into individual low-retention PCR tubes (Bioplastics K69901,
B57801). Sample tubes were reverse transcribed and amplified using the SMART-Seq V4 Ultra
Low Input RNA kit (Takara Bio 634891) in 10x reduced volume reactions using a Mantis liquid
handling system (Lee et al. In preparation). Samples were stored at -20°C until quantitation and
library preparation. All RNA/cDNA volumes during FACS and processing were handled using
low retention siliconized tips (VWR 53503-800, 535093-794).

C1 isolation of single cells and cDNA synthesis:
Dissociated cells were stained and FACS sorted as above and collected into a single 1.5ml
tube. During collection, the microfluidics chip was primed per instructions for Clontech (now
Takara Bio) kit for the C1 system. Cells were centrifuged and resuspended at 400-800 cells per
microliter, combined with C1 suspension reagent and loaded onto the C1 chip. After, the chip

70

was imaged at each site to identify empty or multiple cell wells and then loaded for sample
preparation. First round of sequencing used SMARTer chemistry (SMARTer Ultra Low RNA Kit
for the Fluidigm C1 System, Clontech 634835/634935) (3 retina samples), all others used
SMART-Seq V4. Final cDNA was harvested in the morning into fresh low-retention tube strips
and stored at -20°C until quantitation and library preparation.

Quality control, library preparation and sequencing:
DNA quantitation was performed using Quant-iT PicoGreen dsDNA assay and a subset of
samples were checked on the Bioanalyzer platform for quality and to determine PicoGreen
thresholds for inclusion in libraries (Common threshold 0.05ng/ul). Library preparation was done
with the Nextera XT DNA Library Preparation Kit from Illumina (FC-131-1096). The first dataset
of three retina (2-FW17, 1-FW13) was sequenced on the Illumina NextSeq 500 (2x75) Other
experiments were sequenced on Illumina HiSeq 4000 (2x75) platform at the Center for
Personalized Medicine at Children’s Hospital Los Angeles.

Data processing and analysis:
FASTQ files had adapter sequences removed using the trimgalore wrapper for Cutadapt
(Martin 2011). Trimmed FASTQ files were used as input for HISAT2 (Kim et al. 2019) with a
non-canonical splice penalty of 20, maximum and minimum penalties for mismatch of 0 and
1, and maximum and minimum penalties for soft-clipping set to 3 and 1. Aligned bam files were
quantified with StringTie (Pertea et al. 2015), yielding cell-by-transcript and cell-by-gene count
matrices annotated according to Ensembl build 87. All analyses can be reproduced using a
custom snakemake workflow accessible at https://github.com/whtns/ARMOR adapted from
Orjuela et al 2019.
Dimensionality reduction and visualization of data was performed using the Seurat
toolset (Butler et al. 2018). To exclude technical effects between distinct tissue isolation and

71

sequencing batches, expression counts from all datasets were integrated using Seurat’s
standard integration workflow. Seven sample sequencing batches were integrated after default
normalization and scaling using the top 2000 most variable features in each set to identify
anchor features. Normalized read counts for gene expression are reported as raw feature read
counts divided by total cell read counts, then multiplied by a scaling factor of 10,000 and natural
log transformed.
These features of the integrated dataset were used to calculate principal component
analysis (PCA) and uniform manifold approximation and projection (UMAP) embeddings
(McInnes et al. 2018, Becht et al. 2018). A nearest-neighbors graph was constructed from the
PCA embedding and clusters were identified using smart local moving (SLM) algorithm at
resolutions 0.2 – 2.0 (Waltman & Van Eck, 2013).
To compare our range of cell states against other datasets, human retina scRNA-
sequencing experiments were pulled as an anndata object from the NEI-hosted Platform for the
Analysis of Single Cell Eye in a Disk (scEiaD) (https://plae.nei.nih.gov/). We utilized data from
the following SRA Accession numbers: SRP15023, SRP170761, SRP222001, SRP223254,
SRP238587, SRP255195, and E-MTAB-7316 (Lu et al. 2020, Menon et al. 2019, Sridhar et al.
2020, Yan et al. 2020, Lukowski et al. 2019). In experiments where retinal organoids were also
sequenced, those samples were excluded for clarity. scVI was applied using the model
developed for PLAE to project our data onto the downloaded combined dataset.
Differential expression was performed between groups of interest using the Seurat
FindMarkers function with the Wilcoxon-rank sum test method on log-normalized counts.
Adjusted p-values (Bonferroni correction on all features) of <0.05 were considered significant.
Results were displayed as volcano plots made with EnhancedVolcano.
Statistical comparison between more than two groups of cells for gene expression,
regulon signal or exon count proportions was done using the non-parametric Kruskal-Wallis
Test with post-hoc Dunn Tests in a pairwise fashion to determine differences. To compare mean

72

expression levels of NRL and RXRG isoforms across cell clusters, Welch's t-tests for groups
with unequal variance were used; p-values for these tests were estimated using bootstrapping
with 1,000,000 replications per comparison.
To identify marker features for each cluster, Wilcoxon-rank sum tests and specificity
scores were computed using the genesorteR R package. Only genes with an adjusted p-value
(pAdj) <0.05 were considered for further analyses and the top five markers for each cluster were
displayed where possible.
Overrepresentation and gene set enrichment analyses were performed with WebGestalt
(www.webgestalt.org) (Liao et al. 2019). Both analyses were run with the default settings (genes
per category 5-2000, Benjamini-Hochberg multiple-testing correction). All gene sets were
provided within WebGestalt: GO – Biological Process, KEGG and Hallmark50. To perform ORA
gene lists were compared to the “genome” reference set.
To generate coverage plots for genes of interest we utilized wiggleplotr to visualize
BigWig files with ENEMBL isoforms. Individual exon counts were identified using DEXseq,
which takes exons from available isoforms and bins them into intervals before assigning any
read counts that overlap a bin to that same bin. These counts were length normalized and used
to calculate proportion use of NRL truncated and full-length isoform first exons by calculating
(truncated-full)/(truncated+full).
RNA velocity analysis was performed with R package velocyto (La Manno et al. 2018).
Spliced and unspliced count matrices were assembled using the run_smartseq2 subcommand
with repeat-masked ENSEMBL build 87. RNA velocity was calculated using cell kNN pooling
with a gamma fit based on a 95% quantile and velocities were overlaid on integrated UMAP
visualizations using velocyto.
Transcription factor regulons were identified using pySCENIC, the python-based
implementation of Single-Cell Regulatory Network Inference and Clustering (SCENIC) (Van de
Sande et al. 2020, Aibar et al. 2017). The tool was run using the basic settings as shown

73

(https://pyscenic.readthedocs.io/en/latest/tutorial.html). Initial filtering required a gene to have a
minimum of 3 raw counts in 1% of cells to be considered for inclusion in a regulon.
Pseudotemporal trajectories were calculated using Monocle 3 (Cao and Spielmann et al.
2019). The SeuratWrappers package provided integration between Seurat’s final integrated
dimensional reduction and Monocle. The data was subset to only focus on one cell type at a
time (L/M cone, rod) for the following steps. A principal graph was fit across the UMAP plot and
a ‘root cell’ was chosen to represent the earliest point in the maturation of the chosen cell type.
Cells reachable from a given root node were then mapped onto the nearest point of the principal
graph and assigned a pseudotime value based on the distance from the root to that point.
Monocle also provides the ability to find genes that change as a function of this pseudotime, as
well as to group those genes into modules of co-regulation by performing UMAP and Louvain
community analysis on those genes. Genes with a correlation q-value of <0.05 were retained for
module identification.

Fixation and cryosectioning of fetal retina:
Retina were procured and then dissected in cold 1xPBS. The cornea and lens were removed
with a cut around the limbus of the iris leaving the front of the eye open. The tissue was
submerged in ~25ml of 4% paraformaldehyde (PFA) in 1xTBS and placed at 4°C on a rocker at
low speed ON. The tissue was washed 3 times with 1xPBS before incubation in 30% sucrose
ON at 4°C. A mold was cut from the top 5ml of a 50ml conical tube and placed on dry ice. A
solution of 1:2 OCT Compound:30%Sucrose was mixed and centrifuged to remove bubbles.
The mold was partially filled before adding the dissected retina and then covering fully.
Orientation of the open front was noted. 10um sections were cut on Cryostat at ~-24°C, 2-3
sections per slide with the front of the eye facing side on to the blade.

74

Immunostaining:
For SYK immunofluorescence, tissue sections were warmed at RT until dry and then washed in
Coplin staining jars two times with 1xPBS for 5 minutes at RT. Samples were permeabilized in
1xPBS-T for 5 minutes and washed again in 1xPBS twice for 5 minutes. Samples were blocked
for 1 hour in blocking solution (1xPBS, 0.1% Triton X-100, 5% normal donkey serum, 5% normal
horse serum) covered with parafilm to reduce evaporation. Primary antibodies, mouse anti-SYK
and rabbit anti-ARR3, were mixed in blocking solution and applied to respective samples
overnight (~16-18 hours) at 4°C. Slides were then washed with 1xPBS for 10 minutes three
times and secondary antibody solutions were mixed, again in blocking solution. These were
applied for 1 hour at RT, also covered as before, before two 1xPBS washes for 15 minutes
each. The third 1xPBS wash contained DAPI to stain nuclei and was done for 10 minutes before
mounting the slides with mowiol solution. Immunostaining of cells attached to coverslips was
performed as previously described (Singh et al. 2018).

shRB1 lentivirus production:
Four plasmids were each used in the production of three different lentiviral constructs (pMDL,
pVSVG, pREV and a lentiviral expression vector). 293T cells were cultured in 15cm dishes in
DMEM media with 10% FBS and split to the desired plate number the night before transfection
with no antibiotics. Plasmids were mixed at 10μg pMDL, 5μg pREV, 5μg pVSVG, and 20μg
Lenti expression vector in 3mL of fresh DMEM without serum for each plate. Separately, 160ul
of 0.6ug/ul polyethylenimine (PEI) and 3mL DMEM were mixed per plate and allowed to stand
for 5 minutes. Both mixtures were combined and allowed to incubate at RT for at least 30
minutes. All 293T plates were treated with trypsin (Corning 25-052-Cl) to single cell dissociation
and diluted in DMEM/10% FBS when collected. Once harvested, the cells were counted and the
volume was adjusted to approximately 30 million cells/plate. The cell mixture and PEI/plasmid
mix were combined and mixed for several minutes before being evenly distributed to the same

75

number of 20cm plates as started with. DMEM/10% FBS was added to bring the final volume to
25mL. After 16-24 hours at 37°C, media was exchanged for 20mL of UltraCULTURE medium
containing 1% HEPES, 1% GlutaMax and 1% PennStrep. After 60-64 hours of incubation,
media was harvested and then centrifuged, filtered through 0.45um PVDF flask filter,
concentrated via the Tangential Flow Filtration System (TFF) using a Midikros 20cm 500KD
0.5mm column (D02-5500-05-S), with both input and final supernatant containers kept in ice to
minimize virus loss. The concentrate was re-filtered using a 0.45um PVDF syringe filter
(Millipore) and stored in aliquots at -80°C. Final lentivirus was titered using a p24 ELISA kit
(ZeptoMetrix 801002).

SYK inhibitor treatment:
FW16.5 retina was partially dissociated with short papain treatment, cultured overnight in RB
media (Xu et al. 2014) at 37°C in a six-well plate, then frozen in 10%DMSO solution and stored
in liquid nitrogen. When needed, samples were removed and revived in RB media overnight (Xu
et al. 2014). Dissociated retina was divided into two groups and infected with lentivirus carrying
shRNAs targeting RB1 (shRB1-733 (Xu et al. 2014) or a scrambled control (shSCR) and diluted
β-mercaptoethanol, insulin, glutamine, penicillin, and streptomycin to the final concentrations in
the retina culture medium along with Polybrene, for 24 hours. After replacing 2/3 of media in
each well with fresh media, cells were treated with SYK inhibitor GS-9876 (Gilead) at one of four
concentrations (0.5uM, 1uM, 2.5uM, 5uM) or a DMSO control. After 12 days of the treatment,
cells were attached to poly-L-Lysine-coated coverslips for 3 hours, fixed in 4%
paraformaldehyde (PFA) for 10 min, washed in PBS three times, and stored at -20 °C until
immunostaining. To determine the role of SYK in cone precursors after RB loss, immunostaining
was performed on attached slides using rabbit anti-RXRγ, mouse anti-Ki67, and goat anti-GFP.

76

Antibody List:
Protein Target Species Source Cat.
Number
Dilution
SYK mouse Santa Cruz SC-1240 1:200
ARR3 rabbit Cheryl Craft
(Zhang et al. 2001,
Li et al. 2003)
NA 1:1000
RXRG rabbit Santa Cruz SC-555 1:800
Ki67 mouse BD Bioscience 550609 1:200
GFP goat Abcam ab6673 1:500

77

Software List:
TrimGalore: Available at: https://github.com/FelixKrueger/TrimGalore.
Cutadapt: Martin, M. (2011). Cutadapt removes adapter sequences from high-throughput
sequencing reads. EMBnet.Journal, 17(1), 10. https://doi.org/10.14806/ej.17.1.200
HISAT2: Kim, D., Paggi, J. M., Park, C., Bennett, C., & Salzberg, S. L. (2019). Graph-based
genome alignment and genotyping with HISAT2 and HISAT-genotype. Nature
Biotechnology, 37(8), 907–915. https://doi.org/10.1038/s41587-019-0201-4
StringTie: Pertea, M., Pertea, G. M., Antonescu, C. M., Chang, T. C., Mendell, J. T., & Salzberg,
S. L. (2015). StringTie enables improved reconstruction of a transcriptome from RNA-seq
reads. Nature Biotechnology, 33(3), 290–295. https://doi.org/10.1038/nbt.3122
Snakemake/ARMOR: Orjuela, S., Huang, R., Hembach, K. M., Robinson, M. D., & Soneson, C.
(2019). ARMOR: An automated reproducible modular workflow for preprocessing and
differential analysis of RNA-seq data. G3: Genes, Genomes, Genetics, 9(7), 2089–2096.
https://doi.org/10.1534/g3.119.400185. Custom workflow based on this work available at:
https://github.com/whtns/ARMOR
PLAE/scEiad: Available at: https://plae.nei.nih.gov/
WebGestalt: Liao, Y., Wang, J., Jaehnig, E. J., Shi, Z., & Zhang, B. (2019). WebGestalt 2019:
gene set analysis toolkit with revamped UIs and APIs. Nucleic Acids Research, 47(W1),
W199–W205. https://doi.org/10.1093/nar/gkz401
pySCENIC: Van de Sande, B., Flerin, C., Davie, K., De Waegeneer, M., Hulselmans, G., Aibar,
S., Seurinck, R., Saelens, W., Cannoodt, R., Rouchon, Q., Verbeiren, T., De Maeyer, D.,
Reumers, J., Saeys, Y., & Aerts, S. (2020). A scalable SCENIC workflow for single-cell
gene regulatory network analysis. Nature Protocols, 15(7), 2247–2276.
https://doi.org/10.1038/s41596-020-0336-2
tximport: Soneson, C., Love, M. I., & Robinson, M. D. (2015). Differential analyses for RNA-
seq: transcript-level estimates improve gene-level inferences. F1000Research, 4(2), 1521.
https://doi.org/10.12688/f1000research.7563.1.
Seurat: Stuart, T., Butler, A., Hoffman, P., Hafemeister, C., Papalexi, E., Mauck, W. M., Hao,
Y., Stoeckius, M., Smibert, P., & Satija, R. (2019). Comprehensive Integration of Single-Cell
Data. Cell, 177(7), 1888-1902.e21. https://doi.org/10.1016/j.cell.2019.05.031
EnhancedVolcano: Blighe, K., Rana, S., and Lewis, M. (2018). “EnhancedVolcano: Publication-
ready volcano plots with enhanced colouring and labeling.” R package version 1.10.0. Available
at: https://github.com/kevinblighe/EnhancedVolcano.
genesorteR: Mahmoud, IM. (2020). “genesorteR: Feature Ranking in Clustered Single Cell
Data.”R package version 0.4.3. http://github.com/mahmoudibrahim/genesorteR
wiggleplotr: Alasoo K. (2019). “wiggleplotr: Make read coverage plots from BigWig files.” R
package version 1.8.0. Available at: https://github.com/kauralasoo/wiggleplotr
DEXSeq: Anders, S., Reyes, A., & Huber, W. (2012). Detecting differential usage of exons from
RNA-seq data. Genome Research, 22(10), 2008–2017.
https://doi.org/10.1101/gr.133744.111

78

velocyto.R: La Manno, G., Soldatov, R., Zeisel, A., Braun, E., Hochgerner, H., Petukhov, V.,
Lidschreiber, K., Kastriti, M. E., Lönnerberg, P., Furlan, A., Fan, J., Borm, L. E., Liu, Z., van
Bruggen, D., Guo, J., He, X., Barker, R., Sundström, E., Castelo-Branco, G., …
Kharchenko, P. V. (2018). RNA velocity of single cells. Nature, 560(7719), 494–498.
https://doi.org/10.1038/s41586-018-0414-6
monocle3: Cao, J., Spielmann, M., Qiu, X., Huang, X., Ibrahim, D. M., Hill, A. J., Zhang, F.,
Mundlos, S., Christiansen, L., Steemers, F. J., Trapnell, C., & Shendure, J. (2019). The
single-cell transcriptional landscape of mammalian organogenesis. Nature, 566(7745),
496–502. https://doi.org/10.1038/s41586-019-0969-x.

79

Supplemental Data

Figure S2.1: SCENIC Regulon Specificity Scores (RSS)
Top five highest specificity scores per cluster are labeled.

80

Table S2.1: Differentially Expressed Genes Between Rod Clusters ER and LR
p-value>0.05, log2FC>|0.5|

Symbol log2FC P-Value Adj Symbol log2FC P-Value Adj
ER CRABP2 3.034048 5.36E-14 LR NT5E -2.1398 1.51E-29
DCT 1.463187 2.37E-07 AP000997.1 -1.98427 4.06E-26
FABP7 2.210345 1.73E-06 GNGT1 -3.84224 2.58E-23
GNG13 1.356835 8.07E-06 CLUL1 -2.43582 1.68E-22
LAYN -2.21892 2.59E-22
SAMD7 -1.83682 1.47E-21
YBX3 -1.89328 5.22E-21
F5 -1.3529 8.37E-20
GNB1 -2.05934 1.54E-19
CNGA1 -1.94968 5.44E-19
SVIP -1.43979 7.96E-19
UNC119 -1.71203 1.56E-18
FAR2P3 -0.58464 2.16E-18
RP11-23E19.1 -0.70934 3.99E-18
CNGB1 -1.80799 1.18E-17
RP11-96L14.7 -1.97954 4.43E-17
RP11-379K17.4 -0.73252 5.21E-17
PDE6G -1.84967 1.32E-16
GNAT1 -1.9401 2.27E-16
SAG -2.30363 6.30E-16
PPP2R2B -1.55112 3.65E-15
CADM1 -1.35728 5.46E-15
KCNV2 -1.62534 1.08E-14
RCVRN -0.67641 1.73E-13
TSPAN7 -1.719 1.81E-13
VAV3 -0.78122 2.08E-13
TRAJ49 -1.02236 2.52E-13
MET -0.73886 3.73E-13
CABP5 -1.57664 7.28E-13
TRAC -1.64642 1.69E-12
FAR2P2 -1.01868 2.07E-12
ARL4D -1.57777 7.51E-12
GPR160 -1.26392 1.13E-11
TLL1 -0.65884 2.14E-11
C11orf88 -1.6075 2.40E-11
RAX2 -1.26107 2.85E-11
REEP6 -1.53223 3.59E-11
PPEF2 -0.86656 5.74E-11

81

KCNIP2 -0.90055 6.16E-11
SPTBN1 -1.08298 7.19E-11
RGS9 -0.605 7.67E-11
GRIK2 -1.11384 8.61E-11
RASSF2 -0.76761 8.62E-11
DUX4L50 -0.65017 1.15E-10
LGALSL -0.93276 1.44E-10
RS1 -0.80411 1.75E-10
PLK2 -1.44176 2.41E-10
FAM138F -0.76215 3.09E-10
GABRR3 -1.04174 4.18E-10
CPLX4 -0.97184 4.94E-10
ANK3 -0.73473 5.81E-10
ELOVL4 -1.11071 6.81E-10
GNGT2 -1.11361 8.81E-10
FAM107A -1.0953 8.97E-10
FAM138C -0.57545 1.14E-09
ROM1 -1.07671 1.20E-09
FAM138B -0.65708 1.33E-09
AF131216.5 -1.12805 1.71E-09
HMGN2 -0.91442 1.93E-09
ZPBP -0.83096 2.05E-09
NRL -0.88795 2.88E-09
IMPG1 -1.55331 3.16E-09
FAM138A -0.81412 3.36E-09
ROBO3 -0.68586 3.59E-09
PLEKHB1 -1.18052 4.44E-09
SLIT2 -0.61683 6.27E-09
FAM138E -1.03399 7.50E-09
RTBDN -1.09369 1.66E-08
ARL6IP5 -1.0669 2.12E-08
RP1 -1.02106 2.49E-08
RABL3 -0.747 2.91E-08
MEGF11 -0.66191 2.91E-08
SERF1A -0.97652 3.54E-08
ENO2 -1.05026 3.90E-08
SERF1B -1.04461 4.19E-08
UBE2E3 -1.19869 4.53E-08
APLP1 -1.17394 4.64E-08
AC007349.7 -0.78455 5.32E-08
PSMG1 -0.74741 5.32E-08
TMCC2 -0.9121 5.79E-08

82

CPLX3 -1.45319 5.97E-08
CPE -0.84002 6.05E-08
SIL1 -0.83865 1.19E-07
HMGN1 -0.78195 1.19E-07
CKMT1B -0.80868 1.46E-07
MXRA7 -0.66102 1.81E-07
PLA2G4C-AS1 -1.24425 2.30E-07
KCTD7 -0.62636 2.61E-07
HPRT1 -1.28318 3.74E-07
DPP3 -0.71921 6.64E-07
ZNF593 -0.69832 6.64E-07
CTD-2193G5.1 -0.68527 6.75E-07
CKMT1A -0.67887 6.83E-07
PRCD -1.15978 7.21E-07
MIR124-2HG -0.87171 9.73E-07
SPCS1 -1.09433 1.03E-06
SVILP1 -0.59491 1.17E-06
TSPAN13 -0.57852 1.29E-06
GPC5 -0.7403 1.63E-06
AC004540.5 -1.09137 1.69E-06
AP000462.1 -1.14516 2.57E-06
PHYHD1 -1.00103 2.91E-06
C9orf16 -0.92339 3.03E-06
TLE4 -0.60614 3.93E-06
NR4A1 -0.52366 4.44E-06
PFKP -0.65272 4.54E-06
IFI27L2 -0.86943 4.56E-06
NDUFA4 -0.77013 4.57E-06
H2AFJ -1.04701 4.70E-06
PPARA -0.59503 6.09E-06
MFGE8 -0.71744 6.91E-06
VTN -0.63314 7.29E-06
SELENOK -0.79636 7.42E-06
TPD52 -0.88852 7.85E-06
CKB -0.66373 8.56E-06
GSKIP -0.69717 9.60E-06

83

Table S2.2: Differentially Expressed Genes Between Cone Clusters S and LM
p<0.05, log2FC>|0.5|

Symbol log2FC P-Value Adj
Symbol log2FC P-Value Adj
S Cone MEGF10 -0.70678 7.43E-35 LM Cone
THRB 1.830218 2.66E-08
CCDC136 -2.08752 1.19E-24
CNTN1 1.782047 2.64E-07
UPB1 -1.27343 5.85E-19
FAAHP1 4.909842 7.31E-07
NRXN3 -1.10998 4.34E-15
RPLP0 1.288001 2.44E-06
SEZ6L -0.9818 3.58E-14
RPS29 1.527113 2.82E-06
TMEM51 -0.7382 1.77E-12
ISL2 1.747289 2.64E-05
OPN1SW -5.78445 2.71E-12
RPL23A 1.151012 8.80E-05
ACKR3 -1.10078 3.40E-12
RPS27 1.187583 0.00011
CHRM3 -0.68764 7.49E-10
SLC40A1 2.208341 0.000116
TRIB1 -1.01327 3.24E-08
RXRG 1.321656 0.000348
ZPBP -1.40166 4.73E-08
LMO4 0.946545 0.000424
SPINK4 -1.83761 6.21E-08
HDDC2 1.816764 0.000441
RP11-99J16--A.2 -1.18557 1.31E-07
SPON2 2.028931 0.000643
THBS4 -2.30851 1.40E-07
MUC19 0.629442 0.000886
RP11-23E19.1 -0.59969 2.09E-07
EEF1A1P6 0.818115 0.001173
FAM19A4 -2.19053 3.39E-07
PRDX1 1.022738 0.002429
TBX2 -0.84242 4.13E-07
PCAT4 1.195536 0.002757
CHN2 -2.21211 1.23E-06 RP11-
761N21.2
0.531289 0.003328
CHN2 -2.21211 1.23E-06
GUK1 1.023578 0.004217
MAP2 -2.30951 5.46E-06
RPL23AP42 0.848005 0.004644
RS1 -1.27459 7.03E-06
EEF1A1 0.712331 0.006257
HHIP -0.40692 1.07E-05
NOL4 1.462102 0.006353
AGT -1.24608 1.24E-05
OLFM1 1.530466 0.006573
CNGB3 -2.01141 3.14E-05
SNHG6 1.358682 0.007488
PPP1CC -1.61107 7.33E-05
RBP4 1.691264 0.015019
CTSH -0.86164 9.48E-05
RPS3A 0.893996 0.015835
AC007349.7 -0.83143 0.000182802
RPS27A 0.856993 0.024055
NEUROG1 -1.26256 0.000231178
AIPL1 0.880937 0.029433
PDC -1.48574 0.00031019
RPS19 0.971124 0.036321
KCND3 -0.55781 0.000327672
CROT -1.33945 0.000637326
FZD10-AS1 -0.38226 0.001059615
UBAP1L -0.65046 0.001128335
TBX2-AS1 -0.70235 0.001615962
SCG2 -0.54944 0.002145943
HHIP-AS1 -1.18852 0.002602978
KCNH6 -1.0634 0.00271869
SFXN2 -0.82104 0.00307543

84

TMEM215 -0.65484 0.005720055
MYL4 -0.36342 0.008282631
SOCS2 -0.69817 0.008330643
GNGT2 -1.5769 0.010263919
KCNAB1 -0.52061 0.011004467
MFGE8 -0.4422 0.01233283
NPBWR1 -0.69416 0.013203532
NME1-NME2 -1.07747 0.015735492
PPARGC1A -0.49129 0.019793827
CPLX2 -0.89292 0.023433128
ESAM -0.75136 0.026280418
PCMTD1 -0.88785 0.036257609
TXNIP -0.68264 0.043098079
MCF2L2 -0.42059 0.048439748

85

Table S2.3: Differentially Expressed Genes Between L/M Opsin Quintet and Other LM cluster
p<0.05
Symbol log2FC P-Value Adj
RP11-907D1.2 0.775448 8.48E-61
IGSF5 0.329759 9.49E-33
MYL4 1.208249 8.22E-29
KRTDAP 0.535447 1.10E-28
LINC01549 0.348032 1.10E-28
RP11-464C19.2 0.402878 1.10E-28
RP13-349O20.2 0.42274 1.10E-28
CTC-458G6.2 0.693465 1.85E-28
RGS9 0.320837 3.57E-28
PLA2G5 0.984868 2.38E-25
TTR 1.905457 2.07E-22
AC007009.1 0.984427 3.79E-22
C6orf132 0.308001 1.50E-20
ANKRD29 0.574916 5.67E-18
RP11-143P4.2 0.507799 5.67E-18
RP11-279O9.4 0.644165 7.42E-18
CRISP2 1.196455 1.50E-14
RP11-44N21.1 0.58815 2.37E-14
MRVI1-AS1 1.772589 4.06E-14
PLK5 0.324502 4.97E-14
CTA-150C2.13 0.293832 1.09E-12
MGC45922 0.273564 1.09E-12
NPVF 0.297862 1.09E-12
OPN1MW2 1.772336 2.02E-12
OPN1MW3 1.234044 2.02E-12
FABP6 1.355433 3.27E-12
AC004854.4 0.284345 7.29E-12
PCP4 3.23864 3.00E-11
RP11-883G14.1 1.1929 3.78E-10
GLYATL1 0.827467 1.03E-09
RP11-38M8.1 0.969527 1.59E-09
RP11-240M16.1 1.427729 1.60E-09
OPN1LW 1.279345 2.32E-09
RP5-1115A15.1 0.32115 3.47E-09
AC063980.3 0.783461 7.86E-09
RP4-536B24.2 0.864879 7.86E-09
RP11-535M15.1 0.373377 3.51E-08
GUCA1C 4.825367 7.54E-08
CCDC136 0.831014 1.55E-07

86

TNNT3 0.254279 7.59E-07
TIMM8AP1 0.540896 8.81E-07
ATP6V0E2-AS1 0.270517 1.79E-06
TMEM35B 0.78914 3.48E-06
PSKH2 0.4123 5.06E-06
RSPH1 0.727776 6.87E-06
FRMD3 0.386741 9.01E-06
OPN1MW 2.027591 1.52E-05
C11orf88 0.476014 1.71E-05
RP11-6E9.5 0.403135 1.71E-05
RNU6-853P 0.443981 1.92E-05
CGREF1 0.804396 2.11E-05
SCG2 0.502788 2.15E-05
NAAA 0.505199 8.48E-05
ZNF30-AS1 0.269043 8.79E-05
CST6 0.646063 0.00015083
RNA5SP342 0.314306 0.00015083
RNU6-522P 0.782293 0.00015083
RP11-139K4.2 0.277793 0.00015083
IGKV1OR2-108 0.326115 0.000184695
DAW1 1.20198 0.00021192
ELOVL3 0.317691 0.000438797
CHAC2 0.420871 0.000600851
ICA1 0.811319 0.000973386
RP11-154D6.1 0.268269 0.001099128
IFI6 0.809267 0.001209457
NXNL1 0.586124 0.001231791
NPFF 1.071705 0.001343989
PPEF2 0.301135 0.001443943
RNF144B 1.098485 0.001580965
CDC37L1-AS1 0.961477 0.002376229
PPIC 0.656953 0.003310161
NPM2 0.606377 0.003362039
AC097724.3 0.297918 0.003536052
C3orf33 0.537517 0.005940128
RP11-441O15.3 0.324154 0.006277982
GYG1 0.807314 0.007699047
MATN2 0.959906 0.008848064
AF131216.5 0.561393 0.009153849
SPATA20 0.402286 0.009478499
KCNIP4 0.815266 0.009992235
CTC-543D15.8 0.438394 0.011674506

87

RP11-105N14.1 0.254907 0.012654934
RP11-169K17.3 1.694214 0.014936875
SPAG17 0.33113 0.018170194
TMEM63C 0.308237 0.020834363
DPP4 0.505317 0.023833774
DKK1 0.286252 0.029955452
RGR 0.853699 0.02995761
SLC37A1 0.26907 0.036742758
SCARNA20 0.384033 0.041654974
TDRG1 1.109417 0.043082099
ASMT 2.537289 0.044889533

88

Table S2.4: Genes Correlated with L/M Cone Pseudotime and Module Grouping

Symbol Module q value
CTC-378H22.2 1 2.96E-62
CHRNA1 1 3.79E-16
S100A6 1 2.71E-13
CD44 1 3.35E-10
MFAP4 1 1.81E-09
JMJD7 1 2.77E-09
HOTAIRM1-5 1 7.09E-09
HOTAIRM1-4 1 9.37E-09
ATOH7 1 2.79E-08
HOTAIRM1 1 4.18E-08
EDIL3 1 5.99E-08
STMN2 1 2.47E-07
NCCRP1 1 1.50E-06
DLL3 1 2.25E-06
DTNA 1 0.000162
SMIM20 1 0.000178
TOP1MT 1 0.000333
SLC20A1 1 0.000464
PCDH20 1 0.000543
TMEM17 1 0.000607
RBFOX2 1 0.000611
CEP83-AS1 1 0.000691
UQCRH 1 0.000827
ENO3 1 0.001075
FABP7 1 0.001489
BHMT 1 0.001674
MCM2 1 0.003697
ZNF239 1 0.005395
DYNLT1 1 0.005412
TSPAN33 1 0.006272
LIPT2 1 0.006583
ORC5 1 0.008963
FLVCR1-AS1 1 0.009868
RN7SL743P 1 0.009909
SURF1 1 0.012058
YEATS4 1 0.013239
CCKBR 1 0.014567
CSRP2 1 0.016304
MED9 1 0.018173

89

RND3 1 0.022292
RP11-1017G21.5 1 0.024709
GGH 1 0.02757
SLC7A3 1 0.02757
SPESP1 1 0.028463
RP11-159H10.3 1 0.034375
HIST1H4K 1 0.036924
COA4 1 0.037554
RP11-408A13.2 1 0.037554
ANKRD44 1 0.039073
AP001258.4 1 0.039868
MTHFD1 1 0.039868
CCNG1 1 0.040163
RP11-539G18.3 1 0.042186
HIST1H4J 1 0.043167
LRRN3 1 0.043884
RP11-282O18.3 1 0.047703
GPM6B 1 0.047732
NNAT 1 0.047987
TAGLN3 1 0.04847
AC062017.1 1 0.0494
LHX9 2 1.37E-23
PIK3IP1 2 1.91E-15
IFITM2 2 1.10E-10
HES4 2 1.04E-08
CPXM1 2 1.12E-06
PTPRK 2 5.26E-06
NECAB2 2 5.35E-06
MDK 2 2.11E-05
CRABP1 2 2.12E-05
SPARC 2 0.000183
ASCL1 2 0.000251
COL23A1 2 0.00201
MKNK1 2 0.002368
UBE2L6 2 0.002717
SFRP2 2 0.004232
DERL3 2 0.004336
PAX6 2 0.004577
MGARP 2 0.005831
IGFBP2 2 0.006732
IFITM3 2 0.009502
TMSB15A 2 0.009868

90

GATB 2 0.01084
C1orf61 2 0.013645
DAPL1 2 0.013645
CNN3 2 0.014881
CLU 2 0.015737
FBLN1 2 0.016445
RN7SL452P 2 0.024416
PIPOX 2 0.024709
TMEM98 2 0.02757
HES5 2 0.029867
DKK3 2 0.035469
MIR210HG 2 0.036924
RMST-10 2 0.036924
FAAHP1 3 5.43E-62
PDE6H 3 4.34E-42
MPP4 3 2.34E-33
OLFM1 3 6.18E-18
GRAMD1C 3 1.89E-17
HDDC2 3 2.11E-12
RP13-143G15.4 3 2.23E-12
CRABP2 3 4.00E-11
LOXL1-AS1 3 2.25E-10
DUSP2 3 3.56E-08
SMTN 3 1.21E-07
MT1X 3 1.09E-06
RP1-52J10.9 3 1.09E-06
SPON2 3 2.06E-06
PYY 3 3.72E-06
RXRG 3 4.28E-06
EMB 3 6.62E-06
MT1F 3 6.62E-06
RP11-354E11.2 3 1.15E-05
F10 3 1.49E-05
RTN1 3 2.14E-05
OLAH 3 3.80E-05
STN1 3 8.03E-05
LIPE-AS1 3 0.000146
PHF19 3 0.000273
PPA1 3 0.000276
ZNRD1 3 0.000324
TMEM101 3 0.000326
NPFF 3 0.000424

91

IFI16 3 0.000607
KCNIP4 3 0.000717
ZFR2 3 0.001075
AP000459.7 3 0.001361
RIBC1 3 0.001523
TMEM54 3 0.001697
SYT17 3 0.001977
MMS19 3 0.002601
RP11-395G23.3 3 0.0034
RP4-665J23.1 3 0.0034
ALG3 3 0.004415
HIGD2A 3 0.004474
MT3 3 0.004474
NME7 3 0.004474
CPLX3 3 0.00479
LINC01508 3 0.005485
LHFP 3 0.005759
DYNLT3 3 0.006047
DTNBP1 3 0.006818
MESP2 3 0.00708
RNA5-8S5 3 0.007102
ISOC2 3 0.010189
SH3GL2 3 0.010603
TAL2 3 0.01093
RRAD 3 0.010963
FUCA2 3 0.012191
RP11-545E17.3 3 0.018974
MPC1 3 0.020834
MPHOSPH6 3 0.020982
MED11 3 0.021745
ATP6V1B2 3 0.02665
CDC123 3 0.02757
OMA1 3 0.02757
TUBA1B 3 0.02757
RNF183 3 0.028463
TTC25 3 0.028861
DUSP14 3 0.03318
TMEM60 3 0.034927
FAM118B 3 0.037554
SLC39A4 3 0.039012
RAD51B 3 0.039372
SNHG8 3 0.039617

92

SNHG17 3 0.039868
SNORA63 3 0.041402
GRM6 3 0.043316
RCVRN 4 1.21E-88
GNB3 4 8.06E-43
PDC 4 5.73E-21
IMPG1 4 6.87E-20
PCP4 4 5.97E-18
IMPDH1 4 5.36E-16
UNC119 4 5.04E-15
CNGB3 4 1.59E-10
ACKR3 4 2.68E-10
KCNV2 4 1.22E-09
FAM19A4 4 1.62E-08
PPP2R2B 4 4.75E-08
PHYHD1 4 2.39E-07
LINC00575 4 5.69E-07
CLTB 4 1.03E-06
LINC01053 4 3.95E-06
ACOT7 4 5.71E-06
GNGT2 4 6.18E-06
RP11-165I9.9 4 6.84E-05
TDRD9 4 0.000178
RP11-3P22.2 4 0.000472
MRLN 4 0.002055
TMEM35A 4 0.002255
AC007349.7 4 0.002634
PLA2G4C-AS1 4 0.004577
CLUL1 4 0.007423
RNLS 4 0.008963
TRAJ49 4 0.008963
CCDC136 4 0.010189
GYG1 4 0.010768
ENO2 4 0.012654
PTS 4 0.012654
GPR160 4 0.013616
PDE6G 4 0.014567
MSR1 4 0.016604
CA2 4 0.017107
MAP2 4 0.022292
PPEF2 4 0.028463
VPS29 4 0.030209

93

COX14 4 0.03112
RP11-386J22.3 4 0.036924
MYL4 5 6.29E-21
CTD-2034I21.2 5 2.47E-13
GUCA1C 5 1.03E-12
VTN 5 3.69E-12
PLTP 5 6.80E-07
DAW1 5 0.000141
RP11-883G14.1 5 0.000165
POP5 5 0.00146
AMN1 5 0.001913
DNASE2 5 0.003523
HPRT1 5 0.003831
TP53TG1 5 0.004577
HIST1H2AC 5 0.004717
EPHA5-AS1 5 0.005861
CDC37L1-AS1 5 0.008537
C11orf71 5 0.010758
ALDOC 5 0.011298
RNU2-63P 5 0.014567
PCGF5 5 0.01486
LIN7B 5 0.019578
IFT172 5 0.019937
UBA3 5 0.02146
METTL5 5 0.022292
UBE2T 5 0.024396
RACGAP1 5 0.028463
CTH 5 0.035469
SNRNP25 5 0.039441
RP11-140K17.3 5 0.039973
ZNF702P 5 0.043577
SFXN4 5 0.045808

94

Table S2.5: Differentially Expressed Genes Between Low Resolution ER and LM
p<0.05, log 2FC>|0.4|

Symbol log2FC P-Value Adj Symbol log2FC P-Value Adj
ER NR2E3 -3.786789611 1.20E-58 LM RXRG 1.8979 3.09E-50
CTD-2524L6.3 -2.14343188 3.40E-39 CNTN1 1.543137 5.12E-50
ROM1 -2.599539183 4.89E-39 FAAHP1 4.064664 1.04E-49
PTPRZ1 -1.206860815 1.18E-33 THRB 1.583505 3.89E-47
MRLN -2.425383181 1.96E-33 OLFM1 2.065492 1.34E-46
NRL -1.86142902 3.18E-32 EEF1A1 1.152131 1.53E-45
PDC -1.496227446 1.26E-31 EEF1A1P6 1.178155 1.92E-44
SLC6A17 -0.619859856 4.15E-30 DHRS7 2.216071 3.79E-44
GNAT1 -2.707315148 1.95E-25 HDDC2 1.871092 9.58E-44
EPB41L2 -1.136884125 7.86E-25 C8orf46 1.502046 1.71E-42
C2orf71 -0.47341582 7.66E-24 EEF1A1P5 1.320542 5.20E-42
NFIB -1.173635098 7.74E-24 DCT 1.590486 2.88E-40
SOX4 -1.192701852 3.45E-23 LMO4 0.913629 6.63E-38
RCVRN -0.691220934 1.30E-21 SPON2 2.075747 3.09E-36
MALAT1 -0.843060463 1.52E-21 EEF1A1P11 0.774895 5.29E-36
SEMA3A -0.869227031 3.12E-21 EEF1A1P13 0.857507 5.60E-36
RP11-408A13.2 -0.830688233 4.59E-20 GNAT2 1.864293 1.57E-35
EZR -1.294990107 2.20E-19 EEF1A1P19 0.468462 4.53E-35
PLA2G4C -1.283565141 2.35E-19 RPLP0 1.101959 1.27E-34
FABP7 -1.850969755 2.84E-19 EEF1A1P9 0.635274 5.71E-34
SYTL4 -1.307912533 3.67E-19 RPLP0P6 1.036152 8.48E-34
RP5-1028L10.2 -1.483033149 5.79E-19 SALL3 1.138236 2.28E-33
TMEM91 -1.461925025 3.77E-18 CCDC141 1.381527 2.39E-33
MYO9A -0.406625347 1.28E-17 IGSF21 0.732484 1.03E-31
CNGB1 -1.557948113 1.95E-15 PCAT4 1.061869 3.77E-31
EPHA2 -0.645742007 5.32E-15 RBP4 1.719163 2.00E-30
MAP2 -1.159651844 8.52E-15 SEMA6D 1.120748 2.73E-30
BAZ2B -0.859126558 1.00E-14 MUC19 0.570649 3.80E-30
AL035610.1 -0.42038374 1.15E-13 TLE1 1.166286 1.69E-29
DUSP1 -1.322811459 1.27E-13 PPA1 1.680386 1.81E-29
CASC15 -0.811839386 1.49E-13 GUK1 1.093117 2.56E-29
SIX3 -0.763227563 2.50E-13 ISL2 1.396082 5.75E-29
SRSF5 -0.724720729 3.10E-13 TTN 0.64065 1.08E-28
HIST1H1C -1.556466782 3.12E-13 GSG1L 0.655857 2.88E-28
SRRM2 -0.912780151 8.44E-13 GRIP1 0.539493 8.34E-26
RASD1 -1.564889824 1.03E-12 RP11-849I19.1 0.815151 1.10E-25
EML4 -1.315025541 1.09E-12 RPL3P2 0.484148 1.27E-25
H3F3B -0.493842487 1.67E-12 SLC44A1 0.675239 3.08E-25

95

PRUNE2 -1.023345994 2.68E-12 FMN2 0.567055 3.32E-25
CABP5 -2.011133015 6.26E-12 RPL3 0.814749 6.08E-25
TDRD7 -0.840538836 1.62E-11 RTN1 1.175171 6.18E-25
JUND -0.749943398 1.71E-11 APMAP 1.195858 8.98E-24
NTM -1.762411176 2.54E-11 MCC 0.92208 4.04E-23
WI2-1896O14.1 -0.879338266 4.74E-11 SLC8A1 0.642183 4.12E-23
GPR160 -1.237788355 1.16E-10 PHACTR2 0.749511 1.10E-22
EPS8 -1.084686219 7.21E-10 CLTB 1.26419 1.62E-22
ABCA4 -0.82056882 2.32E-09 SH3GL2 1.489959 2.84E-22
KIF19 -0.894395063 2.88E-09 RRAD 1.64967 5.48E-22
FRMPD1 -0.613996336 6.62E-09 HOTAIRM1-4 2.08235 7.33E-22
CKAP4 -1.051284897 1.47E-08 CRABP2 1.281449 7.91E-22
TPI1 -0.617509015 2.00E-08 CCDC88C 0.519383 3.73E-21
PDE4DIP -0.891413501 3.56E-08 AKAP2 0.724998 1.67E-20
ZNF385A -0.562831747 5.75E-08 HOTAIRM1 1.005295 2.79E-20
SAMD11 -0.97306053 7.55E-08 STMN1 0.745033 3.44E-20
MMD -0.826710246 8.67E-08 CTB-63M22.1 0.848308 4.71E-20
DAND5 -0.730443836 1.09E-07 SYT4 0.99134 1.17E-19
BTBD8 -0.860515457 1.23E-07 ENAH 0.857399 1.96E-19
ATP1B2 -0.803368125 1.25E-07 NRN1L 1.324709 3.28E-19
TLE3 -0.538418137 1.63E-06 SIX6 1.015293 4.67E-19
RP11-87N24.3 -0.668754133 1.90E-06 SUSD2 0.794949 4.77E-19
GADD45G -0.745525482 2.03E-06 HNRNPA1P7 0.43733 9.34E-19
ANXA4 -0.661654773 2.82E-06 PYGB 0.902825 1.67E-18
STC1 -0.490348535 3.72E-06 RP11-379B18.5 0.879959 2.18E-18
ABHD15 -0.547784509 4.03E-06 RD3 0.828716 2.35E-18
PTP4A3 -0.898219181 5.92E-06 B3GAT2 0.491388 3.25E-18
PAG1 -0.565631542 6.87E-06 HNRNPA1P10 0.411405 4.27E-18
CCNG2 -0.784694604 7.71E-06 RPL3P4 0.568413 5.05E-18
CRYBG3 -0.502550887 7.75E-06 PLEKHB1 1.096999 5.61E-18
ABCA1 -0.702895518 7.78E-06 RPL18AP3 0.576166 6.97E-18
PLA2G4C-AS1 -0.596047623 8.64E-06 WWC1 0.635207 1.08E-17
MAP1LC3B2 -0.554079745 1.09E-05 PCBD1 1.176642 1.41E-17
SYNE2 -0.551464832 1.99E-05 RPL23AP42 0.569792 1.61E-17
MAP1LC3B -0.936302124 2.52E-05 RPL23A 0.607591 2.14E-17
ATP6V1B1 -0.770510744 2.62E-05 FAM174B 0.531579 3.12E-17
WNT10A -0.656780386 2.99E-05 NPTX1 0.542698 3.24E-17
LGI4 -0.444450371 6.44E-05 TMEM261 0.48002 5.00E-17
SELENOP -0.892208151 8.03E-05 HNRNPA1 0.71541 5.15E-17
MAP1LC3A -0.516986865 8.30E-05 SLC24A2 0.402598 8.75E-17
PHYHIPL -0.765716183 9.95E-05 RPL7P9 0.517254 1.83E-16
HIST1H2AC -1.002085047 0.000130264 KCTD17 0.729888 2.52E-16

96

PRPH2 -1.458862557 0.000139371 RPL7P1 0.513913 3.50E-16
ZBTB20 -0.663441248 0.000169661 DST 1.001243 3.97E-16
FBXO31 -0.659179489 0.000210133 PCSK1N 0.815998 4.09E-16
ADHFE1 -0.505626308 0.000235794 SLC38A1 0.537398 4.22E-16
CFAP36 -1.141318484 0.000297392 CABP2 1.056701 4.93E-16
DDX17 -0.529268866 0.000357983 MSR1 1.002866 6.08E-16
SNORA11E -0.489250211 0.000375717 RPL7 0.578882 6.42E-16
SNORA11D -0.551656515 0.000501225 DPP10 0.562694 1.01E-15
RP11-217O12.1 -0.769412117 0.000683977 RPL21P16 0.438508 1.02E-15
GLCCI1 -0.53147781 0.000694485 MPP4 1.092516 1.18E-15
MGLL -0.724191345 0.000729535 SMTN 0.997113 1.35E-15
FOXO3 -0.474660402 0.001229976 ATP5A1 0.623396 1.64E-15
CDKN1C -0.669829541 0.001301328 SLC4A8 0.701846 2.66E-15
PBRM1 -0.781530377 0.001442606 SLC25A6 0.842289 3.31E-15
PPM1D -0.477439126 0.002173822 SEZ6 0.405382 3.47E-15
ITSN1 -0.666167838 0.00225787 RPL18A 0.611612 3.51E-15
KDM4A-AS1 -0.797600006 0.002271973 TMEM38B 0.622452 5.62E-15
EIF1 -0.403182533 0.002344337 RPSA 0.944427 6.14E-15
AP003025.2 -0.448431328 0.002674953 HOTAIRM1-5 1.958519 6.69E-15
DDIT4 -0.997146574 0.003287451 ACOT7 0.942484 6.92E-15
BTG1 -0.602045372 0.003406322 LRRC75A-AS1 0.808237 8.57E-15
CSRNP3 -0.561557824 0.003680611 GRIK2 0.767981 1.13E-14
PTPRK -0.777788442 0.004092509 SLAIN1 0.993289 1.43E-14
PNRC1 -0.725417335 0.004219953 INA 0.885218 1.64E-14
ATF4 -0.771449951 0.004500293 MIR7-3HG 1.062917 1.64E-14
TOX -0.51127658 0.004663312 RPS2P5 0.645043 1.72E-14
FAM210A -0.566882781 0.006163425 AC079250.1 0.568052 1.75E-14
NDRG1 -0.638084135 0.006287224 OLAH 1.309173 1.83E-14
EPHA8 -0.448601713 0.006338847 PDE6H 3.075969 2.12E-14
ERI1 -0.536931255 0.008679133 AIG1 1.005389 2.14E-14
IMPDH1 -0.794768962 0.009506932 NEIL2 0.70327 2.77E-14
IER2 -0.789516783 0.01110937 LDHB 0.621527 2.84E-14
RAF1 -0.73671971 0.01193611 SYNE1 0.400993 4.35E-14
EPB41 -0.701157235 0.017155291 CPLX3 1.261198 5.16E-14
CEP112 -0.602132199 0.018153914 LGALSL 0.691609 6.41E-14
CYP27A1 -0.533758232 0.022196817 RPS3 0.576569 8.12E-14
FBLN1 -0.577626155 0.024207227 TPD52L1 0.679357 1.48E-13
SSX2IP -0.606596593 0.027763811 GSTP1 0.669723 1.54E-13
RP1 -0.500067689 0.036778493 SH3BGRL 1.359979 1.86E-13
BBS2 0.980711 2.11E-13
RPL4 0.686718 2.47E-13
CITED4 0.765575 3.26E-13

97

RPS19 0.69398 5.01E-13
MAP6 0.667138 5.37E-13
SPATA6 0.451523 5.43E-13
RPS5 0.776394 6.04E-13
CNTFR 0.556199 7.07E-13
ACAT1 0.944629 7.78E-13
TIAM2 0.496598 9.81E-13
CRMP1 0.860067 1.02E-12
RABGAP1L 0.554518 1.30E-12
LINGO3 0.602039 1.37E-12
PGM3 0.591356 1.40E-12
RP11-632C17--A.1 0.415243 1.80E-12
RPL7P23 0.423882 1.80E-12
CNGA3 0.456222 2.14E-12
RPS2P46 0.496856 2.35E-12
SLC40A1 1.017733 2.57E-12
ATP2B1 0.683428 3.01E-12
CHGA 0.629709 3.24E-12
EGFEM1P 1.204666 3.33E-12
ORAI2 0.497434 3.55E-12
GEM 0.777519 3.74E-12
ATOX1 0.884166 3.91E-12
RPL13A 0.580949 5.01E-12
RAX 0.746305 5.53E-12
MAP1B 0.800157 5.78E-12
PPP1R14A 0.601878 6.49E-12
RAMP1 1.083503 7.71E-12
EMB 0.568518 8.98E-12
AL162151.3 0.735606 1.20E-11
LMO3 0.748743 1.30E-11
RPL21 0.494489 1.32E-11
PACSIN1 0.532549 1.35E-11
ATP5G2 0.547955 1.97E-11
LINC00632 1.111929 1.97E-11
CCDC82 0.756303 2.21E-11
TJP3 0.719553 2.80E-11
RPL36A 0.659785 2.93E-11
BEX1 0.744949 3.27E-11
CPLX4 0.493139 3.61E-11
RPL19 0.562808 4.51E-11
PBX1 0.610864 5.72E-11
ATP5L 0.518338 6.45E-11

98

ASUN 0.813383 9.15E-11
SPECC1 0.547915 9.29E-11
RHOBTB1 0.542386 9.34E-11
APP 0.561267 1.15E-10
RPS2 0.656593 1.26E-10
IGFBPL1 0.455399 1.40E-10
HN1 0.737143 1.51E-10
IMPG2 0.531819 1.56E-10
RPS25 0.464144 1.78E-10
ITGA4 0.518396 1.85E-10
GNAS 0.614368 2.05E-10
COX6A1P2 0.435934 2.26E-10
RPS6 0.528619 2.83E-10
RP11-475C16.1 0.407 2.87E-10
CXADR 0.557374 2.99E-10
GAS5 0.500055 3.13E-10
WDR1 0.783464 3.49E-10
GPC5 0.640411 5.22E-10
NME2 0.777159 5.25E-10
NAP1L1 0.733415 5.80E-10
MIF-AS1 0.472586 7.69E-10
SPINT2 0.761449 7.86E-10
IFI27L2 0.539641 8.10E-10
SLC29A3 0.689713 9.88E-10
SLC8A1-AS1 0.654176 1.04E-09
PRDX1 0.448962 1.13E-09
TUBA1A 0.744835 1.74E-09
HSP90AB1 0.539229 1.75E-09
AKAP13 0.634635 2.16E-09
SPTSSA 0.728912 2.16E-09
FBXO10 0.400575 2.69E-09
TMSB4XP8 0.438835 2.99E-09
SYK 0.437902 3.15E-09
EFNA5 0.404089 3.17E-09
ARL6IP5 0.576601 3.30E-09
TAL2 0.740695 3.30E-09
MAP4 0.582072 3.39E-09
RP13-143G15.4 0.899738 3.94E-09
FAM213A 0.673013 4.12E-09
BEX3 0.697916 4.16E-09
TMEM14A 0.790913 4.40E-09
NDUFS5 0.617048 6.98E-09

99

LBH 0.508407 7.25E-09
IFI16 0.914381 7.73E-09
RPL5 0.60832 7.76E-09
GRAMD1C 0.781888 8.25E-09
SNAP91 0.636983 1.27E-08
AC018738.2 0.5354 1.28E-08
PAFAH1B1 0.4046 1.44E-08
PLPPR4 0.420522 1.51E-08
RPL15P3 0.440793 1.85E-08
ABHD12 0.766181 2.29E-08
ANAPC13 0.645435 2.56E-08
DZIP3 0.531212 2.64E-08
MYCN 0.543346 3.24E-08
PRDX3 0.545519 3.54E-08
RPS16 0.72209 3.56E-08
GCSH 0.470727 3.75E-08
YWHAZ 0.634643 3.99E-08
TMSB4X 0.650195 4.53E-08
RPL15 0.452391 5.04E-08
SSR4 0.692721 5.51E-08
LSM4 0.618481 7.55E-08
TBCB 0.58452 7.60E-08
NCBP2 0.751355 7.92E-08
ZNF480 0.552491 8.15E-08
TIGAR 0.404786 1.26E-07
RHBDD2 0.613301 1.27E-07
IMPG1 0.766255 1.44E-07
EEF1G 0.606621 1.66E-07
ATP5G1 0.64416 1.94E-07
KLHDC8B 0.794943 1.99E-07
CD200 0.854265 2.40E-07
RPS4X 0.578471 3.23E-07
NPDC1 0.456653 3.96E-07
TSPAN3 0.71966 3.98E-07
COX6A1 0.576381 5.11E-07
CANX 0.622569 5.62E-07
RP11-115N4.1 0.590511 5.67E-07
NDNF 0.448046 5.70E-07
NHP2 0.584834 6.10E-07
RP11-395G23.3 0.764401 6.51E-07
CTB-178M22.2 0.54928 6.68E-07
NR4A1 0.603576 7.36E-07

100

NDUFV2 0.615653 8.26E-07
HIST1H4K 0.577822 1.02E-06
ARMC9 0.546498 1.14E-06
SLC17A5 0.641331 1.22E-06
HOXA1 0.569724 1.48E-06
EGR1 0.631974 1.51E-06
APLP1 0.755471 1.73E-06
RPL23 0.683225 1.78E-06
RP11-363E7.4 0.504981 1.82E-06
GABRG2 0.684164 1.87E-06
RPS11 0.48485 1.94E-06
RPS3A 0.4467 2.02E-06
HIGD2A 0.488983 2.24E-06
SLC1A6 0.643798 2.48E-06
NUCB2 0.748575 2.51E-06
ATP5I 0.5883 2.71E-06
SSBP4 0.519433 2.87E-06
EEF2 0.464208 4.42E-06
CKMT1B 0.650632 4.49E-06
RPS12 0.468524 5.11E-06
AMPD2 0.439941 5.30E-06
MLLT11 0.533661 5.78E-06
RP11-386G11.10 0.491913 5.93E-06
IFT57 0.517263 6.03E-06
APOO 0.694408 6.29E-06
RPS13 0.456466 6.61E-06
TOMM5 0.567613 6.99E-06
ERGIC3 0.569071 7.55E-06
HIST1H4J 0.668381 7.86E-06
GOLPH3 0.560648 8.51E-06
GNG10 0.447806 8.58E-06
EGFLAM 0.613736 9.62E-06
PRKAR2B 0.509909 9.75E-06
STRA6 0.424679 1.02E-05
TCTN1 0.644149 1.05E-05
RPS14 0.43697 1.27E-05
PLTP 0.518984 1.31E-05
UCHL1 0.708344 1.32E-05
C1QBP 0.620317 1.37E-05
FAM107A 0.425348 1.54E-05
DPP10-AS1 0.581694 1.77E-05
ZFR2 0.6302 1.91E-05

101

ATP6V1F 0.709208 1.92E-05
NCALD 0.923671 2.31E-05
RPS28P7 0.476123 2.45E-05
RP11-666A20.4 0.620328 2.49E-05
PTPRN 0.580774 2.63E-05
GRM6 0.510235 2.72E-05
PPIB 0.496918 2.88E-05
ZFAS1 0.446298 3.00E-05
PDZRN3 0.416486 3.25E-05
PRDX4 0.650495 3.47E-05
SEC11C 0.670231 3.61E-05
WDR54 0.74521 4.05E-05
STRADB 0.608596 4.60E-05
TUSC3 0.421298 5.04E-05
RPS27A 0.412929 5.60E-05
EFCAB1 0.467116 6.48E-05
PPIA 0.496261 6.52E-05
HIST3H2A 0.571975 6.87E-05
RPL30 0.486076 7.21E-05
CKMT1A 0.533921 7.38E-05
NDUFB2 0.511132 8.01E-05
MSANTD4 0.593553 8.50E-05
DNAJA1 0.533838 8.55E-05
COTL1 0.661926 9.91E-05
CYCS 0.413065 0.000129375
NDUFB9 0.601572 0.00013195
TAGLN3 0.438094 0.000133222
PRELID2 0.497405 0.000144794
BASP1 0.42538 0.000177222
KCNV2 0.484136 0.000208218
TRIM2 0.42738 0.000227682
APRT 0.505947 0.00023009
CFL1 0.404405 0.000261603
NME5 0.577298 0.00026237
SLC6A15 0.574076 0.000266937
SERPINF1 0.572832 0.000274822
NME1 0.564104 0.000282882
PDZD11 0.766863 0.000304737
SERPINE2 0.566831 0.000307699
RPS28 0.42775 0.000313721
LIMK2 0.466815 0.00032077
EIF3L 0.484574 0.000362818

102

FAM162A 0.402638 0.000369607
IER3IP1 0.406308 0.000383319
RP3-375P9.2 0.40454 0.000391555
HMBS 0.509781 0.000470567
MIEN1 0.421868 0.000520462
PHPT1 0.449971 0.000545246
WSB1 0.513209 0.000546725
DNAJB11 0.44103 0.000563809
RPL12 0.435033 0.000575085
RABEPK 0.553022 0.000587976
RPL7A 0.469505 0.000601031
RSPO1 0.564997 0.000674441
TPM1 0.494259 0.000715792
ARL2 0.426298 0.000729388
PRAF2 0.425927 0.000741898
COX7C 0.432058 0.000827558
SPAG16 0.499793 0.000864755
MTHFD1L 0.476951 0.000880144
PEPD 0.628101 0.000885235
SULT1C4 0.471962 0.001008564
BEX4 0.552885 0.001018664
TIMP1 0.572023 0.001138648
SRM 0.611339 0.001153176
UQCR11 0.425131 0.001159975
NDUFS8 0.420849 0.001161145
DLL3 0.410394 0.001192734
RP11-408J6.1 0.430771 0.001237952
SERP2 0.403198 0.001238155
PRDM1 0.463729 0.001264535
PSMB7 0.561512 0.001338977
PHTF1 0.48449 0.001356922
HSPD1 0.443728 0.00137687
PKIG 0.439574 0.001786501
DNAJC12 0.605676 0.001860791
TSTD1 0.476768 0.001878982
CRB1 0.566876 0.002011175
BAG1 0.510793 0.002074501
TCEAL8 0.513146 0.002116072
RN7SKP139 0.545889 0.002142352
CDKN2AIPNL 0.46536 0.002196537
GOLM1 0.50615 0.002300951
YAF2 0.463217 0.002363664

103

UBE2V1 0.421877 0.002454254
NDUFA1 0.430269 0.002628108
ISYNA1 0.468801 0.002738044
FAM69B 0.420426 0.003085136
CD63 0.458088 0.003402919
MDH2 0.465449 0.003498488
SLC25A5 0.528999 0.003604609
CA2 0.807469 0.003677674
B4GAT1 0.535449 0.003699322
PGRMC1 0.528499 0.004272906
ATP6V0D1 0.443622 0.004925196
HPRT1 0.472108 0.005153294
TTC8 0.464092 0.005893763
ATP6V1B2 0.487026 0.00644999
DDOST 0.478998 0.006573144
SET 0.466435 0.006962252
C1orf53 0.566837 0.008570731
RNU6-713P 1.072589 0.009784711
DCTD 0.499305 0.010173975
CCT3 0.450673 0.010640563
MYDGF 0.447614 0.011129341
MZT2B 0.407295 0.011898438
NOB1 0.401926 0.012594413
RP11-620J15.3 0.425554 0.013427121
ECI1 0.430698 0.013580285
BSCL2 0.537377 0.013844202
TSPAN13 0.425198 0.014908849
RPL18 0.419178 0.014931266
MORN2 0.569112 0.015060047
CERKL 0.653653 0.015417121
SLCO3A1 0.4749 0.017450958
AC092675.3 0.638722 0.019226335
ROMO1 0.406184 0.019481047
GSKIP 0.430972 0.020996983
CCDC60 0.479622 0.021397552
RPL10P16 0.417817 0.022625114
NPY1R 0.474897 0.022717127
TUBA1B 0.507797 0.023205819
TMEM54 0.506287 0.025161801
COPS9 0.443224 0.028447006
PNMA1 0.624232 0.028587298
REEP5 0.501302 0.031762741

104

LYPLA1 0.425945 0.031937582
DAP3 0.43835 0.034354369
VTN 0.671811 0.035013097
LOXL1-AS1 0.579104 0.036503349
THRA1/BTR 0.554578 0.044147597
HINT2 0.414814 0.049410651

105

: Similarities and Differences in Maturing Cone Photoreceptor
Development in Fetal Retina and hESC-Derived 3D Retinal Organoid
Models
Contributions: Sunhye Lee: scRNA-seq methodology. Kevin Stachelek: Bioinformatics. Mitali Singh:
Bioinformatics. Jennifer Aparicio: Organoid production, embedding, and sectioning. Narine
Harutyunyan: Organoid production, embedding, and sectioning. Andrew Salas: Stem cell culture. Kayla
Stepanian: Organoid production, embedding, and sectioning. Yeha Kim: Organoid production,
embedding, and sectioning. Jackie Lin: CHLA FACS Core. Mark Reid: Statistics. Michael Bonaguidi and
Maxwell Bay: Bioinformatics advice. David Cobrinik: PI

Introduction
The human retina shares many morphologic and molecular features with the retina of other
vertebrates, such as the same major cell types, layered structure, and key transcription factors
defining individual cell types (Fadool and Dowling 2008, Thoreson 2016, Baden et al. 2020).
However, common models such as mouse and zebrafish do not recapitulate several
photoreceptor-related features such as the human rod-free fovea centralis, the three-color cone
subtypes (Seabrook et al. 2017), and the cone precursor proliferative response to pRB loss
(Singh et al. 2018). Studies with human tissues are inherently constrained by access to
available samples and by limits on the type of modification and perturbation experiments that
can be performed to examine development or disease processes. A potential solution to these
issues is using human embryonic stem cells (hESC) to derive retinal tissues, providing a more
accessible human tissue for study.
The earliest protocols to derive retinal cell populations from stem cells generated retinal
progenitor cells or limited post-mitotic cell types such as photoreceptors in two-dimensional
monolayer culture (Ikeda et al. 2005, Lamba et al. 2006, Osakada et al. 2008). However, in the
last decade protocols have progressed beyond production of simple monolayers. The Sasai
group demonstrated the formation of three-dimensional (3D) structures capable of replicating
optic cup formation (Eiraku et al. 2011, Nakano et al. 2012), ciliary margin like structures
(Kuwahara et al. 2015), and production of cells expressing markers of all major retinal cell types

106

born in the expected order (reviewed in Aparicio et al. 2017). Other retinal organoid approaches
also showed this consistent birth order while producing other more advanced retinal features
such as photoreceptor light response (Zhong et al. 2014, Cowan et al. 2020) and central to
peripheral maturation (Mao et al. 2019). Beyond producing expected retinal features, it has
been possible to alter the proportions of photoreceptor types and maturation rates (Kim et al.
2019, Kaya et al. 2019).
However, these retinal organoid tissues are imperfect retinal development models.
Retinal cell maturation has been observed to plateau at around 30 weeks in culture (Cowan et
al. 2020) and in multiple studies organoid transcriptomes better correlated with peripheral adult
retina over central retina, even though organoid rod:cone ratios often resemble the more central
perifovea (Gonzalez-Cordero et al. 2017, Capowski et al. 2019, O’Hara-Wright and Gonzalez-
Cordero 2020). One recent study produced cone-rich organoid photoreceptors whose
transcriptomes highly correlated with adult human macula and better correlated to macaque
foveal over peripheral photoreceptors (Kim et al. 2019); however, even these photoreceptors did
not achieve mature cilia morphology. Retinal organoids experience degradation of INL
lamination at ~ 100 days in culture (Capowski et al. 2019, Sridhar et al. 2020), and it has been
repeatedly shown that the RGC layer disappears early in culture, potentially due to lack of
synapse of retinal ganglion cell axons (Zhong et al. 2014, Aparicio et al. 2017a, Sridhar et al.
2020, O’Hara-Wright and Gonzalez-Cordero 2020). Culture of human fetal retina spheroids
demonstrated that culture conditions aren’t exclusively the cause of organoid RGC loss or layer
disorganization (Sridhar et al. 2020), indicating that additional organoid-specific factors must
contribute to these problems.
Parsing the developing protein and gene expression of retinal organoids is key to
understanding how effectively they serve as models of human retinal development. Recent bulk
and single-cell transcriptomic analyses have provided a broad array of information on
developing fetal and hESC-derived tissues, such as closely parsing the differentiation timing of

107

retinal cell types (Hoshino et al. 2017, Collin et al. 2019, Hu et al. 2019, Sridhar et al. 2020), and
differential gene expression between foveal and peripheral adult retina (Voight et al. 2019), as
well as temporal comparison of developing organoid cell types and transcriptomes against fetal
retina (Sridhar et al. 2020, Cowan et al. 2020, Kallman et al. 2020). However, few have
evaluated the post-mitotic development of a single cell type in depth across both tissues.
In the current chapter, I focus on deeper analysis of a single-cell type, cone
photoreceptors, to evaluate how organoid-grown cone cells mature and what processes or
states diverge from those observed in fetal cones. Our laboratory has developed organoid
cultures based on three prominent protocols for modeling cone development: Nakano et al.
2012, Zhong et al. 2014, and Kuwahara et al. 2015. These methods have distinct protocol
differences that can yield changes in organoid morphology and photoreceptor number, such as
the use of Matrigel for structure in the Nakano and Zhong methods, the addition of BMP4 in the
early Kuwahara method, or the sequential 3D-2D-3D culturing steps in the Zhong method.
These differences may produce greater or lesser approximation to fetal maturation, which we
want to evaluate.
Previously, it was demonstrated that retinoblastoma cells had properties of cone
photoreceptor precursors and that healthy cone precursors expressed proteins, such as MDM2
and MYCN, that could contribute to tumorigenesis (Xu 2009 and 2014, Singh et al. 2018). As
this human cone-specific program has disease relevance to retinoblastoma, Section 3a
examines the timing of this protein expression program in organoid cone cells and reveals
dysregulation or localization differences between organoid and fetal cones. I also examined
transcriptomic changes associated with cone maturation via full length single-cell RNA
sequencing on enriched organoid cone precursor populations in Section 3b. By comparing
transcriptomes of maturing cone precursors in organoids with those of human fetal cones
(Chapter 2), we demonstrated that organoid cones mature rapidly and undergo one major state
change, marked by upregulation of several photoreceptor genes and a shift in transcription

108

factor activity. However, the studies also revealed that maturing organoid cones did not
sequentially express lncRNAs that were identified in early maturing fetal cones and differentially
expressed genes composing glycolysis and ribosome ontologies in comparison to fetal cells.
Finally, the trajectory analyses revealed that organoid cones produced two cell populations for
which we detected no fetal retina equivalent. Section 3c indicates that these populations appear
to represent aberrant hypoxic internalized cones and cones from disordered tissue regions that
form an interface between ordered neural retina lobes solely in Zhong method organoid
preparations. Overall, this work found differences in retinal organoid versus fetal retina cone
protein and gene expression timing, differences in such timing between organoid protocols,
more mature captured organoid transcriptomic states than fetal of a similar age, as well as
organoid photoreceptor states with no fetal retina counterparts.

Results
3a. Inconsistent expression of cone proliferation-related program associated proteins in
retinal organoids
Prior work from our laboratory demonstrated the tumor-forming ability of pRB-depleted
developing human cone precursors in explanted fetal retina (Lee et al. 2006, Xu et al. 2009 and
2014, Singh et al. 2018). Specifically, cone cells reentered the cell cycle and expressed the
proliferation marker Ki67 after lentivirus mediated RB1 knockdown. However, attempts to
generate retinoblastoma-like structures in hESC-derived retinal organoids with lentiviral RB1
knockdown yielded no proliferating cone cells (data not shown) and had poor cone infection
compared to fetal tissues (data not shown). We wanted to know whether the developmental
expression patterns of proliferation-related proteins in retinal organoids matched those seen in
fetal tissues or might differ and explain why infected cones were not proliferating. To accomplish
this, the protein expression patterns of pRB, MDM2, MYCN and p27 were observed in retinal
organoids and compared against their patterns in human fetal retina.

109

For this purpose, retinal organoids were generated from the H9 human embryonic stem cell
(hESC) line through three different published methodologies (Nakano et al. 2012, Zhong et al.
2014, Kuwahara et al. 2015) with minor modifications (Methods) (Figure 3.1). Some key
differences between the protocols are indicated in Figure 3.1B and include the following: H9
hESCs were grown on a feeder cell layer for Nakano and Kuwahara protocols and grown under
feeder-free conditions for the Zhong protocol. The Nakano and Zhong methods utilize Matrigel
to form embryoid bodies and provide external structure and typically generate multi-lobed
organoids with occasional RPE regions (Figure 3.2, arrows), while the modified Kuwahara
protocol uses a factor combination ISL (Wnt inhibitor IWR1 and BMP/TGF-β inhibitors
SB431542 and LDN193189) and later BMP4 instead of Matrigel, which consistently generated
spherical organoids with little size increase after day 18 of differentiation (Figure 3.1, 3.2).
Individual organoids were kept for experimental use if they had a clearly defined bright
neuroepithelial layer, as indicated by the red and green lines in Figure 3.2.
In fetal tissue, cone arrestin (ARR3) is a cone specific protein that interacts with cone
opsin (Zhang et al. 2001, Li et al, 2003, Zhu et al. 2003) and whose upregulation serves as a
maturation milestone. Transcripts are first detected in fetal tissue at approximately 12 weeks
post conception (Welby et al. 2017), and protein expression marks cones at a development
stage that can proliferate with pRB loss (Singh et al. 2018). Initial studies in our and other labs
showed that organoid cone cells typically begin to express ARR3 between 60 and 70 days in
culture (Kallman et al. 2020). Thus, we used ARR3 as a marker of maturing cones in which to
evaluate the expression of proliferation-related proteins. Organoids were collected at three
different ages in multiple preparations, from approximately d68-125, to capture the early stages
of ARR3 expression. Due to the inherent variability both between and within organoids in a
single preparation, we specifically examined organized regions where cone cells were apically
positioned and showed appropriate apical-basal polarity.

110

Figure 3.1: Organoid production overview
A. Organoid production timelines for Nakano et al. 2012, Kuwahara et al. 2015, and Zhong et al. 2014.
The Kuwahara method has been modified to include BMP/TGF-β inhibitors SB431542 and LDN193189
(SB and LDN). B. Comparison of major features of the three production methodologies including
differences in stem cell culture, starting cell number, inhibitor and Matrigel use, media changes, and
addition of FBS and retinoic acid (RA).

111

Figure 3.2: Organoid appearance during development
Representative images of desired experimental organoids for each production method. Organoids
prepared by the Nakano and Zhong methods developed multi-lobed structures as well as intermittent
RPE (blue arrows). Organoids were chosen for experimental use if they presented a clearly bounded
yellow-white neural epithelium, shown between red and green lines in first column of images.

112

We first compared pRB expression in fetal and organoid cones. In fetal retina, pRB is expressed
peripherally in retinal progenitor cells as well as in the nuclei of apically localized cone
precursors (Lee et al. 2006). However, closer to the fovea, where ARR3 is expressed, the
highest pRB signal is found in the ARR3+ cones in the outer nuclear layer (ONL) with weaker
signal in Müller glia (Figure 3.3A). In all three organoid methods, younger (d68-70) samples had
widespread pRB expression that was not restricted to ARR3+ cells. As organoids aged, pRB
was found in fewer cells but retained variable expression in ARR3+ cones (Figure 3.3B). In all
methods across all ages a population of ARR3- cells showed the strongest pRB signal, usually
internal to the cone layer. In organoids prepared using the Nakano method, cones maintained a
similar level of pRB signal across ages with some showing high expression while most had
similar expression to non-ARR3+ cells at all ages. Organoids prepared with the Kuwahara
method had more ARR3+ cones at d68-70 and showed stable nuclear pRB levels across ages.
By d125, ARR3+ cells had brighter pRB+ nuclei than other cells (Figure 3.3B). Organoids
prepared using the Zhong method showed similar levels as non ARR3+ cells at d68-70, but by
d97 and 125 background levels dimmed along the apical edge and some cones with higher pRB
could be observed (Figure 3.3B). Thus, ARR3+ cones of all organoid methods expressed
nuclear pRB at all ages, with some cones heterogeneously expressing higher pRB at older ages
like what is observed in central retina, though unlike fetal retina they were not the strongest pRB
expressing cells.
MDM2 is an E3 ubiquitin ligase that mediates p53 degradation and is highly expressed
in foveal cone precursors (Xu et al. 2009). We observed similar MDM2 patterning in fetal retina,
with expression in the ONL tightly correlated to ARR3 expression and little or no MDM2
detected in immature ARR3-negative cones (Figure 3.4A), as previously reported (Singh et al.
2018). However, organoids produced with each of the three methods consistently produced
cones that demonstrated delayed MDM2 expression with respect to ARR3. For organoids
prepared by the Nakano method, some cones expressed a weaker cytoplasmic MDM2 signal at

113

d68 and d97 (Figure 3.4B, white arrows) and in older d125 samples MDM2 became detectable
in more ARR3+ cells and expressed at higher levels. The greater levels of early ARR3 in
organoids prepared by the Kuwahara method were not associated with increased MDM2 levels
as only very weak cytoplasmic protein signal was detectable at d69. However, by d97 MDM2
signal was consistently detected in ARR3+ cells which rose again at d125. Organoids prepared
by the Zhong method also showed the greatest levels of MDM2 in d125 cones but at all ages
cones with weak or undetectable MDM2 were present. Across methods, ARR3 signal did not
serve as a reliable indicator of MDM2, as cells with similar ARR3 could have noticeably different
MDM2 levels, particularly in d125 Nakano and Zhong organoids.
MYCN is a well-known oncogene which when amplified causes a small subset of
retinoblastomas without the need for the more common biallelic inactivation of RB1 (Dimaras et
al. 2015). Nuclear MYCN appeared in the more peripheral, less mature ARR3+ cones also have
a strong nuclear signal (Xu et al. 2009). That signal decreased towards the fovea and foveal
cones expressed weak MYCN in the cytoplasm and nucleus (Figure 3.5A). Heterogeneous
levels of MYCN signal were observed in organoids, with nuclear MYCN detected in only a
subset of ARR3+ cells in all three methods. In organoids prepared by the Nakano and
Kuwahara methods, both MYCN+;ARR3+ and MYCN-;ARR3+ cells were detected at d69-70
and d98-106, but by d125 little to no nuclear MYCN signals were visible (Figure 3.5B). MYCN+
nuclei were found up to d125 in Zhong organoids, though they were rare. Thus, in all methods,
cone MYCN+ signal was detected at d69-70, and was less common in the older tissues, and
was not notably higher in cones than in other retinal cell types.
p27 (encoded by CDKN1B) is involved in G1 arrest when present in the nucleus and its
SKP2-mediated degradation is often deregulated in cancers (Liang et al. 2002). p27 is highly
expressed in the nuclei of peripheral ARR3+ cells (Lee et al. 2006); however, in and around the
fovea nuclear p27 decreases and cytoplasmic levels increase in ARR3+ cells (Figure 3.6A). In

114

all organoid methods p27 was expressed in ARR3+ and ARR3- nuclei and not detected in the
cytoplasm. p27 expression in Nakano organoids did not greatly rise or fall with age, while
Kuwahara organoids showed increased p27 levels with age and Zhong organoids decreased
levels (Figure 3.6B).
These results together showcase the heterogeneity of organoids, with varied cone
expression patterns of different proliferation-related proteins even within one organoid.
However, apart from the heterogeneity, some general themes also emerged. The delay in
MDM2 expression relative to ARR3 cone precursors suggests decoupling of the expression
timing of these proteins relative to fetal retina. MYCN was either expressed in the nucleus or not
present in cones across these time points, with only a small number of cells copying the
maturation-associated cytoplasmic and nuclear expression seen in fetal cones. Similarly, p27
remained nuclear and did not achieve the cytoplasmic signal seen in foveal cones. Together
MYCN and p27 expression suggested that at these ages organoid cones do not reach a fovea-
like state and more closely mimic peripheral ARR3+ cone precursors, even though they are
morphologically advancing past what we see in available fetal tissue by developing outer
segments.

115

116

Figure 3.3: Variable pRB expression in retinal organoid cone precursors
A. Elevated pRB nuclear expression in ARR3+ cone cells of FW16.5 retina. Left: parafoveal region.
Middle: foveal region. Right: Extended image showing two layers of pRB+ cells in the fovea, cones at top
and Müller glia at bottom. B. pRB/ARR3 immunofluorescence on d68-70, d97-100, and d125 retinal
organoids from three production methods. For each method top image is ARR3/pRB, bottom is pRB only.
pRB in ARR3+ cones at d68-70 is a similar level to surrounding cells. At later ages some ARR3+ cells
show higher pRB expression. Arrows: ARR3+ RB+ cells. Scale bar: 25um. Number of organoids
evaluated by timepoint: Nakano (n=8,5,5), Kuwahara (n=6,8,10), Zhong (n=11,8,10).

117

118

Figure 3.4: MDM2 expression is delayed compared to ARR3 in all organoid methods
A. MDM2 cytoplasmic expression in ARR3+ cone cells of FW16.5 retina. Left: mid-peripheral region with
weak ARR3 and MDM2. Right: foveal region with high ARR3 and cytoplasmic MDM2 expression B.
MDM2/ARR3 immunofluorescence on d69-70, d98-106, and d125 retinal organoids from three production
methods. For each method top image is ARR3/MDM2, bottom is MDM2 only. D69-70 ARR3+ cells show
low or no MDM2 expression. The number of MDM2+ cells increases with age, though ARR3+, MDM2
low

cells are still found at d125 for Nakano and Zhong methods. White arrows: ARR3+ MDM2+, blue arrow:
ARR3+ MDM2
low/-
. Scale bar: 25um. Number of organoids evaluated by timepoint: Nakano (n=10,6,7),
Kuwahara (n=5,9,10), Zhong (n=9,8,10)

119

120

Figure 3.5: Organoid cones expressed intermittent nuclear MYCN that declined with age
A. MYCN expression in ARR3+ cone cells of FW16.5 retina. Left: peripheral region with weak ARR3 and
high nuclear MYCN (arrow). Right: foveal region with high ARR3 and low cytoplasmic and nuclear MYCN
(yellow arrow) B. MDM2/ARR3 immunofluorescence on d69-72, d97-106, and d125 retinal organoids
from three production methods. For each method top image is ARR3/MYCN, bottom is MYCN only. D68-
72 Some ARR3+ cells show nuclear MYCN. At 97-106 MYCN+ nuclei are detectable in Nakano and
Zhong method organoids, while at d125 only Zhong organoids show any MYCN+ ARR3+ cells. White
arrows: ARR3+ MYCN+, blue arrow: ARR3+ MDM2
low/-
. Scale bar: 25um. Number of organoids examined
at each time point: Nakano (n=7,4,6), Kuwahara (n=10,9,13), Zhong (n=14,8,10).

121

122

Figure 3.6: p27 signal remains nuclear across all organoid methods and ages
A. MYCN expression in ARR3+ cone cells of FW16.5 retina. Left: parafoveal region with weak ARR3 and
high nuclear p27 (white arrow). Right: foveal region with high ARR3 and low nuclear, increased
cytoplasmic p27 (green arrow) B. p27/ARR3 immunofluorescence on d68-70, d97-100, and d125 retinal
organoids from three production methods. For each method top image is ARR3/p27, bottom is p27 only.
ARR3+ cells show -27+ nuclear signal in all organoids at all ages. Nakano and Zhong organoids show
decreases in p27 with age while Kuwahara organoids show increased p27 from d97-100 on. ARR3- cells
along the apical border also express similar levels of p27, forming a layer of increased p27 compared to
more internal cells. Scale bar: 25um. Number of organoids examined at each time point: Nakano
(n=9,6,8), Kuwahara (n=9,11,12), Zhong (n=12,8,10).

123

3b. scRNA-seq of maturing organoid-derived cone cells captured similarities and
differences between preparation methods and compared to human fetal cones

Isolation of FACS-sorted single cells for RNA-sequencing
The above evaluation of the RB-related cone precursor proliferation-related proteins
demonstrated inherent developmental differences present in retinal organoid cultures as
compared to human fetal retina. To determine whether the differences in this single program
reflect broader changes at the transcriptomic level, I performed scRNA-sequencing on retinal
organoids. The scRNA-seq data for fetal retina discussed in Chapter 2 provided a deep analysis
of developing photoreceptors in fetal retina, and by gathering a similar dataset from retinal
organoids we can directly compare the developmental gene expression across both tissues.
The results of section 3a as well as current retinal organoid literature has demonstrated
how organoids produced, via different protocols, vary in a range of aspects, from size and cell
type composition to protein expression timing. This makes it worthwhile to investigate whether
cone precursor transcriptomes differ according to retinal organoid production method in addition
to differing from fetal cone precursors. The Nakano methodology proved more inconsistent than
the other methods for generating retinal organoids, with the highest percentage of failed or
disorganized tissues. Organoids made using the Kuwahara method were highly consistent in
shape and generated a larger number of ARR3+ cells at earlier time points than other methods.
Zhong method organoids were also more consistent than Nakano organoids, utilized a unique
2D to 3D culture that the other two methods did not, and typically produced fewer cone cells
than Kuwahara organoids. Based on the consistency and potential differences in developing
cones, we chose to collect single cells from two replicates each of Kuwahara and Zhong method
organoids for transcriptomic analyses
Using the FACS isolation methods previously described in Chapter 2, single cells were
collected at seven timepoints approximately every two weeks starting at ~d55 and ending at ~

124

d140 (±1-2 days). At each timepoint, scRNA-seq was performed on two different organoid
preparations of each method with the exception of only one Zhong organoid preparation at days
76 and 98 (Figure 3.7). In addition, cells from one set of organoids made using the Zhong
method were gathered at d225. For each collection timepoint, cells were isolated directly with
FACS from a pool of three dissociated organoids. Across all collections, we consistently used
wide FSC/SSC gating and were able to identify distinct CD133
hi
/CD44,49b
low
cell populations
(Figure 3.8), unlike in fetal samples where more restricted size gating was needed to identify the
target population (Figure 2.2). All rounds of sample isolation displayed distinct CD133
hi
clusters
which were larger and more distinct in older organoid samples.
. cDNA was prepared and amplified for each experiment using the SMART-seq V4
chemistry before being stored at -20°C. All amplified cDNA samples were quantified as
described in Chapter 2 to remove those with low cDNA amounts (common threshold for
inclusion = 0.05ng/ul). Once all timepoints for both organoid preparation methods had been
collected, 938 barcoded libraries were made for all samples and placed in a single sequencing
run, which eliminated batch variation following the cDNA synthesis step. Nextera XT index
adapters allowed for 384 samples on a single sequencing lane, divided across 8 lanes to
accommodate all samples in one sequencing run. After alignment of the raw sequencing reads,
undesired single cell transcriptomes with under 100,000 read counts or expressing gene
markers of non-photoreceptor or RPC cell types were removed as described in Chapter 2
(Figure 3.9A). This yielded a final count of 867 single cell transcriptomes, most of which derived
from two independent collection of cells with each of two methods (Figure 3.9B). Per cell, there
was an average of 2,899,688 uniquely aligned reads, 8,321 genes detected, and 23,468
transcripts inferred (Figure 3.9C).

125

Figure 3.7: Overview of organoid collection and scRNA-sequencing
Samples were collected from two replicate preparations of each method at each timepoint with the
organoid days in culture shown on the timeline. Exceptions where only one preparation was collected
were Zhong d76, d98, and d225. The grey line to d225 indicates one sample collected from Zhong
method organoids only. Three organoids were collected and dissociated for each experiment to reduce
effects of organoid heterogeneity. Samples were FACS-enriched for photoreceptors as they were
isolated, then cDNA was made for each sample. Samples went through DNA quantitation and any that
passed had barcoded libraries made and then sequenced. The final data was aligned and cells
expressing INL cell markers (POU4F1, POU4F2, POU4F3, TFAP1A, TFAP1B, ISL1) were removed from
further analysis.

126

Figure 3.8: FACS plots for example sorts from both methods of organoid production
Example FACS plots for three timepoints of organoid collection, d56, d98 and d140-141 for A. Zhong
method-produced organoids and B. Kuwahara method-produced organoids. For each pair of plots the left
shows the forward and side-scatter gating used to exclude the smallest cells before sorting by antibody
signal on the right. Cells are collected from a CD133-high population as CD133 marks photoreceptors.
CD44 and CD49b are used to remove inner nuclear layer cell types to assist in purifying the
photoreceptor population.

127

Figure 3.9: Summary of retinal organoid scRNA-seq data
A. Histogram of total read counts for each cell sequenced colored by read count. Red dotted line:100,000
read cutoff, samples below line are excluded from final dataset. B. Summary of final dataset, with number
of cells captured at each timepoint for each method of organoid production. (*=1 collection). C. Box plots
of the read count, genes detected, and transcripts detected per cell sorted by production method
(Kuwahara (Ku) or Zhong (Zh)) and age of organoid collection. Green bar: Average value across all
samples (Rounded to nearest whole number: read counts per cell = 2,868,684. Genes detected = 8,256.
Isoforms detected = 23,279.).

128

Identification of cell types, age-related trajectories, and preparation-specific cell
populations
UMAP visualization resolved three major cell groups from the final scRNA-seq dataset
(Figure 3.10A). Cell type markers identified RPCs (LHX2), cones (THRB), and rod
photoreceptors (NR2E3) (Figure 3.10B). Increasing ARR3 expression across the cone
population trajecting away from the RPC cluster suggested a direction of cone maturation
(Figure 3.10B, blue arrow). This was supported by cone opsin gene expression (OPN1LW,
OPN1SW), which upregulates after ARR3, in small distal cell populations. Rod cells showed a
similar direction of maturation with rod opsin (RHO) expressed in the cells most distal to RPCs
(Figure 3.10B, red arrow).
High resolution SLM clustering divided transcriptomes into six cone clusters, three rod clusters,
and one RPC cluster (Figure 3.10A). Both photoreceptor groups showed a progression of
clusters organized from least mature to most mature samples based on their increasing
expression of ARR3 and opsin genes, with cone cells forming two spatially distinct side groups,
one larger cluster (Cone-SG) and one smaller which was grouped in Cone-3 (C3*). The
predicted maturation trajectory is also supported by sample ages, with the earliest ages
positioned proximal to the RPC cluster and the oldest samples most distal (Figure 3.10C). Since
transcriptionally similar cells are placed adjacent to each other in UMAP space, any regions
where cells are primarily from one production method indicated unique states for that method.
Cells from both organoid methods intermixed in RPCs, early cones and rods, indicating that
both organoid methods produce similar transcriptome states in these regions. However, four
regions display preparation method bias: late maturing cones (Cone-4-5) are dominated by
Kuwahara organoid-derived cells (Figure 3.10D, red arrowhead), the small spatially distinct
Cone-C3* and part of Cone-1 (C1*) are formed from only Zhong cells (blue arrowheads), and
most mature rods are specifically d225 Zhong (green arrowhead). Generally, Zhong method
tissues produced a unique early cone state (C1*), unique side population (C3*) as well as more

129

mature rods at d225, while late maturation cones were derived from Kuwahara method
organoids.

Identification and removal of aberrant organoid-derived cone populations
To characterize the gene expression differences across clusters and look for any unusual
organoid populations, we first computationally identified cluster-specific marker genes using the
program genesortR (Figure 3.11A, B). Cone-specific marker genes (SLC24A2, GRK7, CNGB3,
PDE6C, GUCA1C, ARR3) were identified in the later Cone-4-5 and rod-specific marker genes
(RHO, PDE6A, SAG, GNGT1, GNAT1, NR2E3) were identified in all rod clusters. However, we
found two groups with unique gene expression markers. These two groups are Cone-SG (C-
SG), which formed most of a large side group alongside an otherwise linearly arranged cone
population (Figure 3.10D, 3.11A blue arrow and circle), and Cone-3, which included the
previously noted C3* group (green arrow and circle) opposite C-SG. Both populations had a
unique set of marker genes that were not shared by other cell clusters, including MFSD2A,
SCD, PFKB3, BHLHE40, and RP11-798M19.6 for C-SG and THY1 and KCNK15 for C3
Focused around C3* (Figure 3.11B). Further analyses revealed that these spatially distinct side
groups represented aberrant cone populations whose gene expression signatures had no
counterparts in fetal cone development, as described below in Section 3C. So, we chose to
exclude these aberrant cell groups, comprising 86 cells, in order to further evaluate the main
trajectory of organoid cone development and compare it to fetal cone development. After
removing the two side populations, a newly generated UMAP plot and clustering yielded distinct
RPC, cone, and rod populations retaining the linear age-related sequence of photoreceptor
subclusters (proximal to distal from RPCs: Cone-1 to Cone-6, Rod-1 to Rod-3) as well as
populations that were specific to the organoid preparation methods (Figure 3.12), now including
a late rod region dominated by Kuwahara method samples (magenta arrowhead). By counting
the number of cells for a given age group in each cluster, I found that for the early timepoints the

130

majority of Zhong cells for a specific age group were found one cluster proximal to RPCs than
Kuwahara cells (Figure 3.12D). Most D55-57 cells from Zhong organoids were found in the
cluster Cone-1, while most D55-57 cells from Kuwahara-derived organoids were found in Cone-
2. This pattern persisted until d98-99 where the number of Zhong organoid cones decreased.

Establishing a trajectory with RNA Velocity and photoreceptor gene expression
To further validate the cone and rod maturation trajectories indicated by sample age,
RNA Velocity was applied to the revised dataset. This identified a predicted direction of
transcriptome change for each individual cell (Figure 3.13A) or the region-averaged direction of
change for the cell populations (Figure 3.13B). Both the cell-specific and region-averaged
trajectories demonstrated that both rod and cone photoreceptor transcriptomes were generally
trajecting away from the RPC population, consistent with the observed increase in age. One
point where this was different was in early cones. The earliest cone cells co-clustered with
RPC/MG similar to the iPRP population observed in Chapter 2, and these cells and those
directly adjacent to them presented velocities towards and through Cone-1. Cone-2 cells show
smaller velocity vectors but from Cone-2 to Cone-3 the grid average trajectories shift to point
toward the more distal cone populations. Unexpectedly, late maturing rod and cone trajectories
also do not converge exactly at the highest opsin-expressing populations but at a point along
the edge.

131

132

Figure 3.10: Single-cell RNA-sequencing of retinal organoids and cell type identification
A. UMAP plot of final scRNA-seq organoid dataset. Three main cell type groups labeled in bold, with high
resolution cell clusters labeled (6 cone clusters, 3 rod clusters, 1 RPC cluster). Two spatially distinct cone
populations were identified (C-SG and C3*) B. Expression plots of RPC/MG (LHX2), Rod (NR2E3), and
cone (THRB) markers, as well as later maturation cone gene ARR3 and photoreceptor opsin genes (Rod,
RHO) (Cone, OPN1LW, OPN1SW). C. UMAP plot of cells colored according to sample age (days in
culture). Arrows indicate direction of increasing sample age for cones (blue) and rods (red) D. UMAP plot
of samples colored by method of organoid preparation. Red arrowhead: Kuwahara method-dominant late
maturing cone region. Blue arrowheads: Zhong method only side population (C3*) and early cone
population (C1*). Green arrowhead: d225 Zhong late maturing rod region.

133

Figure 3.11: Marker genes identify 2 cone clusters with unique marker expression that form
spatially distinct side populations
A. UMAP clustering plot with the two spatially distinct populations indicated. Cone-SG (C-SG) indicated
by blue circle. C3* is included in Cone-3 and indicated by green arrowhead and orange circle. B. Marker
gene plot with top 5 most specific markers for each cluster. Marker genes for C-SG (blue box) and Cone-
3 (olive green box) were uniquely expressed in only that cluster, though only 2 markers were specific to
Cone-3. Based on the features of C-SG and C3*, and additional work in Section 3c, both spatially
separated groups were removed from following analysis.

134

Figure 3.12: Visualization of scRNA-seq patterns after side population removal
A. Recalculated UMAP plot with high resolution clustering after removal of C-SG and C3*. RPC/MG form
two clusters (cRPC and RPC/MG), Cone cells form 6 clusters (C1 to C6), and rod cells form three clusters
(R1 to R3). B. Samples colored by collection age group show young to old samples progress distally from
the RPC/MG population. C. Samples colored by preparation method. Method-specific regions: red
arrowhead = Kuwahara high late maturing cones, blue arrowhead = Zhong high early maturation
population, green arrowhead = Zhong only late maturing rods captured from d225 organoids, magenta
arrowhead = Kuwahara high late rod group. D. Plot of number of cells for a given collection age in each
cone cluster. Top: Zhong method cells, bottom: Kuwahara method cells.

135

Similar patterns of photoreceptor gene expression were observed in this new UMAP
space as in the original UMAP plot shown in Figure 3.10. The cone population showed early
expression of THRB, while PDE6H and ARR3 both upregulated with maturation but were first
detected in the youngest cone clusters (Figure 3.13C). OPN1LW and OPN1SW were only
detected in the distal cone populations, with OPN1LW forming a small group at the farthest
edge. All rods expressed NR2E3, and the later maturation genes SAG and RHO upregulate in
the distal rod clusters. Together RNA Velocity and photoreceptor gene expression indicated that
photoreceptor maturation occurs in a proximal to distal fashion relative to the RPC/MG
population, with an exception for the earliest cones directed to enter the Cone-1 population.
Along with the previous observations about the number of samples of a specific age group in
each cluster, this also indicated that Kuwahara-derived cones reach later maturation states than
Zhong-derived cones at the same age.

NRL and RXRG isoform usage in organoid photoreceptors
Our fetal retina scRNA-seq analyses revealed cell type specific expression of transcript isoforms
related to the canonical rod and cone marker genes NRL and RXRG. To assess whether this
cell type-specific isoform expression is retained in retinal organoids we examined isoform and
exon usage for each gene. Irrespective of isoform, the two genes did not uniquely mark the
corresponding rod and cone cell clusters but were expressed across both photoreceptor
populations as in fetal retina, with higher NRL expression in rods and higher RXRG expression
in cones (Figure 3.14A, 3.15A, 2.11, 2.12). In addition, both genes replicated the same isoform
dynamics as observed in fetal photoreceptors. Full-length NRL isoform ENST00000397002 and
truncated isoform ENST00000560550 were most highly expressed, with ENST00000397002
dominant in rod clusters and ENST00000560550 most expressed in cone clusters (Figure
3.14B). To support this observation, we compared read counts mapping to the unique first
exons of each transcript as described in Chapter 2. Mean read coverage for the two regions

136

shows higher counts for full-length first exon than the truncated first exon (Figure 3.14C). The
relative difference for first exon reads (defined as the difference of normalized first exon read
counts divided by the sum of first exon read counts) supported that observation as there was
greater use of the truncated NRL-encoding isoform exon 1 (Ratio>0) in cone clusters 2-6 and
greater use of full-length isoform exon 1 (Ratio<0) in rod clusters 1-3 (Figure 3.14D, post-hoc
Dunn Test to compare medians: p<0.005 for any pair of cone C2-C6 versus rod R1-R3
clusters). Cone-1 did not show significant relative exon difference to rods, but this cluster had
the lowest overall NRL expression potentially making it hard to compare to other groups (Figure
3.14B). RXRG isoform ENST0000359842 made up the biggest percentage of total average
RXRG expression in all clusters besides Cone-3-5 where it was closely matched by
ENST00000619224 (Figure 3.15A,B). Rod groups showed decreased use of
ENST00000619224, as observed in developing fetal rods (Figure 2.12). This demonstrates that
isoform use patterns in developing photoreceptors are in line with developing fetal cells for
these two genes.

S-cones do not form a unique cell population in retinal organoids
We next examined whether S-cones formed a distinct population, as in our fetal retina scRNA-
seq analyses. No spatially distinct S-cone population was found and furthermore, OPN1SW+

cells were found near OPN1LW+

cells in the large cone population (Figure 3.13C, 3.16A,B). To
determine whether these retinal organoid OPN1SW+ cells that failed to segregate from LM
cones resembled the well-defined fetal S-cones, I looked at the expression of other S-cone
markers identified in Chapter 2, CCDC136, UPB1, and MEGF10 (Lukowski et al. 2019, Peng et
al. 2019, Kallman et al. 2020). CCDC136 was detected across organoid cones, with UPB1 and
MEGF10 sparsely expressed at lower levels (Figure 3.16B). Visual assessment of high
OPN1SW-expressing cells revealed varying levels of each marker, with most expressing
CCDC136 while only one cell expressed UPB1 (Figure 3.16B, circles), and it was aberrantly

137

located near rods (green circle). To test whether the co-expression of OPN1SW+ with each of
the other genes was significantly different than observed in fetal scRNA-seq analyses, I
performed a Fischer’s exact test on the number of cells with OPN1SW>0.5 that had counts
greater than zero or equal to zero across tissue types (Figure 3.16C). CCDC136 co-expression
was not significantly different between tissues (84.6% organoid, 65.6% fetal, p-value = 0.136),
however co-expression of UPB1 and MEGF10 was lower in organoid cells (1.02e-4 and 0.0350
respectively).
While not significantly different in OPN1SW+ cells, organoid cones also showed broad
expression of CCDC136 in OPN1SW-cells (Figure 3.16A,D) while all three genes showed more
restricted S-cone cluster expression in fetal samples (Figure 3.16D). These data suggested two
possibilities: that S and L/M cones were too similar in these organoid tissues to separate into
distinct clusters, possibly due to low S-cone capture or downregulation of other S-cone genes,
or that L/M cones expressed OPN1SW and/or other S-cone-enriched RNAs, particularly
CCDC136, and no S cones were either captured or produced. This issue is re-visited in an
analysis of the combined fetal retina and retina organoid scRNA-seq datasets.

Marker Genes of Organoid Photoreceptor Maturation
To further parse the major gene expression changes between clusters, I used marker gene
analysis to identify the top 5 most cluster-specific gene expression features for the retinal
organoid scRNA-seq dataset (Figure 3.17). The expression of these features was then
compared between the organoid and fetal datasets.

138

139

Figure 3.13: RNA Velocity and photoreceptor marker genes define cone and rod trajectories
A-B. RNA Velocity vectors displayed on high resolution clustering and displaying individual cell velocity
vectors (A) or grid averaged velocity vectors (B). Blue arrows indicate direction of velocity through cones
and red through rods. C. Expression of cone-associated genes. THRB is expressed in the youngest
cones, while PDE6H and ARR3 show more scattered early expression and upregulate through the most
distal cones. Cone opsin genes OPN1LW and OPN1SW are detected in distal cell groups. D. Expression
of rod-associated genes. NR2E3 is observed in all rods while SAG and the rod opsin RHO upregulate in
more distal cells.

140

141

Figure 3.14: Differential NRL isoform use in rod and cone cell clusters
A. Expression of NRL B. The average predicted NRL isoform expression for high resolution clustering
presented as (left) total transcript counts assigned to each isoform and (right) as a percentage of the total
counts. C. Read coverage plot across ENSEMBL-annotated NRL exons. Top shows average read counts
(FPM) for cells in each cluster. Bottom shows the predicted transcript structures for ENST00000397002
and ENST00000560550. Coding sequences colored in dark blue. Blue arrowhead = first exons for
truncated (T) and full-length (FL) isoform sequences in C5 and R2. D. Box plot comparing first exon read
counts in each cell and segregated by cluster. Raw read counts for exon of interest were determined
using DEXseq. The exon use proportion is calculated by taking the difference T – FL and dividing by the
sum of both T+FL for each cell. Horizontal lines indicate median. Values >0 indicate more reads assigned
to the truncated isoform first exon, while values < 0 indicate more reads assigned to the full-length
isoform first exon. Kruskal-Wallis Test p-value < 0.0001, post-hoc Dunn-Test with Benjamini-Hochberg
correction: All cone groups besides Cone-1 (Cone-2-6) are significantly different than every Rod cluster
(Rod-1-3) p-value <0.01.

142

Figure 3.15: RXRG isoform use differences between rod and cone cell clusters
A. Expression of RXRG. B. The average predicted RXRG isoform expression for low resolution clustering
presented as (left) total transcript counts assigned to each isoform and (right) as a percentage of the total
counts. C. Read coverage plot across the RXRG coding region. Top shows average read counts (FPM)
for cells in each cluster group. Bottom shows the predicted transcript structures for ENST00000359842
and ENST00000619224 with coding exons in blue. Blue arrowhead = exon regions unique to the
truncated protein transcript ENST00000619224 marked in C2 and R2 for comparison.

143

144

Figure 3.16: Organoid OPN1SW+ cells do not form a distinct population and have reduced S-cone
marker gene expression
A-B. Expression of OPN1LW and OPN1SW upregulated in distal cones (A), and S-cone marker genes
CCDC136, UPB1, MEGF10 (B). Colored circles: red = CCDC136+/OPN1SW+, blue
=MEGF10/CCDC136/OPN1SW+, green = CCDC136/UPB1/OPN1SW+. C. Number of S-cone gene-
expressing cells (n>0) in OPN1SW+ (>0.5) cells for organoid and fetal scRNA-seq. 2-tailed Fisher’s exact
test comparing cells with counts >0 vs cells with counts = 0 across tissue types. D. Violin plot of cone
opsin genes and S-cone marker genes in retinal organoid and fetal scRNA seq datasets.

145

Cell cycle features divide cRPCs and RPC/MG populations
Markers of RPC/MG cells divided clearly into two populations similar to those observed in fetal
data (Figure 2.16), one with markers of the cycling RPCs like NDC80 and FAM64A (Matson and
Stukenberg 2012, Hashimoto et al. 2017) (cRPC), the other with shared RPC/MG markers like
HES1 and SOX9 (Poché et al. 2008, Furukawa et al. 2000). However, the only marker shared
between organoid and fetal cRPC and RPC/MG clusters (Figure 2.14) was HES1, present in
both clusters.

Known rod genes are main markers of rod clusters
In the three rod cell clusters, well known rod-specific genes were the most common marker
gene features. We observed upregulation of NR2E3, GNGT1, and GNAT1 in all rod clusters
versus those of other cell types (Figure 3.17, 3.18A) with GNGT1 and GNAT1 rising further in
the more mature clusters. A lncRNA CTD-2524L6.3 also showed more scattered expression
across all rod clusters and had low or no expression in some of the least mature cells along with
GNGT1. The least mature rod cluster, Rod-1, showed upregulation of two marker genes that
decreased in more mature clusters: PAG1, a transmembrane adaptor protein that inhibits Src
Family Kinases (Agarwal et al. 2016) and lncRNA RP11-64C1.1 (Figure 3.18B). Rod-specific
genes RHO, SAG, and PDE6A (Figure 3.18C) upregulated in Rod-2 and increased into Rod-3,
along with another lncRNA CTD-2193G5.1). These data indicate that the retinal organoid Rod-2
cluster represents a transition population where marker genes for Rod-1 downregulate while
late features such as RHO and SAG begin to upregulate, with no distinctly high marker gene
expression of its own.
In our fetal retina RNA-seq dataset described in Chapter 2, we captured two distinct rod
populations, early maturing rods (ER) and late maturing rods (LR) (Figure 2.8,3.18A). The
widely expressed organoid rod marker genes NR2E3, GNAT1, and CTD-2524L6.3 were
prominent in organoid clusters 1-3 as well as in both the fetal LR and ER clusters. However,

146

GNGT1 was only weakly expressed in ER cells, suggesting the low expression in the least
mature organoid rods may reflect this ER state. The early organoid rod marker PAG1 was
expressed more widely in fetal cones as well as in both the ER and LR groups with no clear
downregulation as was observed in organoids (Figure 3.18B). Interestingly, the early lncRNA
RP11-64C1.1 was barely detectable at low levels in any fetal rods in comparison to the far
stronger expression of early organoid rods. All late organoid rod marker genes RHO, SAG,
PDE6A, and CTD-2193G5.1 were primarily found in LR fetal rods, though besides SAG only a
small number of cells had detectable expression (Figure 3.18C). Fetal rods capture similar early
and late patterns of most organoid rod marker genes; however, the most mature rods may be
underrepresented in the fetal retina scRNA-seq data accounting for the inability to detect PAG1
downregulation and the low numbers of cells expressing late rod markers RHO, PDE6A, and
CTD-2193G5.1, which appears most similar to expression in Rod-2. Only one marker gene
(lncRNA RP11-64C1.1) was absent from fetal data when strongly expressed in organoid rods.

Early organoid cones mimic the expression of the markers of early fetal cones.
Retinal organoid cone cluster markers segregated into several categories: early marker genes
which were most highly expressed in Cone-1 and notably downregulated in subsequent
clusters, a second group which downregulated gradually across all clone clusters, a third group
which upregulated in Cone-4-6, and a fourth group that upregulated in Cone-5/6 (Figure 3.16).
Among the early expressed markers that rapidly down-regulated, we observed the lncRNA
CTC-378H22.2, a feature of the early cone population from fetal scRNA-seq data (Figure
2.16C), as well as SNCG, which encodes γ-Synuclein and has been observed in RGCs
(Surgucheva et al. 2008), SSTR2, which produces a somatostatin receptor, and nc-RNA PCAT4
(Figure 3.16, 3.19A). SNCG was detected in a focal cell group in Cone 1, overlapping CTC-
378H22.2, while SSTR2 and PCAT4 were expressed in a broader range of early organoid
cones with PCAT4 appearing in a small subset of more mature cells, (Figure 3.19A).

147

A second group of marker genes downregulated slowly with maturation (Figure 3.19B),
including a lncRNA MIR7-3HG, RBP4, a retinol-binding protein typically found in the blood for
transport (Steinhoff et al. 2021), and C8orf46, which codes the protein Vexin which is required
for primary neurogenesis and retinal cell fate specification (Moore et al. 2018). All three showed
low or no expression in late Cone-6.
In the fetal retina scRNA-seq analyses we captured an immature photoreceptor
precursor (iPRP) population that bifurcated into larger rod and cone populations, with the
adjacent L/M cones designated an early maturing cone population and a segregated group of
five cells representing late maturing cones (Figure 2.10). On examination of the early organoid
cone markers in the fetal scRNA-seq UMAP plots, CTC-378H22.2 tightly localized to a small
population of photoreceptor precursor/early cone cells, as previously described (Figure 2.16C),
and, though only detected in a few cells, so did SNCG (Figure 3.19A). This was similar to the
highly focal CTC-378H22.2 and SNCG expression observed in the organoid Cone-1 cluster.
SSTR2 and PCAT4 showed broader fetal cone expression, with SSTR2 expression mainly in
younger age L/M cones and PCAT4 expressed across all available fetal cone groups, while in
organoids this signal was limited to Cone-1/2. All three “slow downregulation genes” (MIR7-
3HG, RBP4, C8orf46) were found across all L/M cones in the fetal scRNA-seq, (Figure 3.19B).
As organoids demonstrated focal expression of the previously identified marker of the early fetal
iPRP cluster, CTC-378H22.2, I also examined the expression of other fetal iPRP cluster
markers identified in Chapter 2 (Figure 3.20). CHRNA1 was expressed across the shared
rod/cone photoreceptor precursor bridge in cells that co-clustered with RPCs, and in organoids
expression was detected in the small group of cone cells co-clustering with RPC/MG (Figure
3.20, arrowhead) and some neighboring Cone 1 and 2 cells. S100A6 was intermittently
expressed in the early organoid Cone-1 and 2 as well as in early maturing fetal cones but
decreasing with maturation. ONECUT1 was mainly expressed in organoid Cone-1-2, where the
strongest expression was overlapped with CTC-378H22.2+ cells. CTC-378H22.2 was found

148

adjacent to CHRNA1 in fetal and organoid cones (Figure 3.19A), and organoid RNA Velocity
indicated that CHRNA1+ cells likely transition into the CTC-378H22.2+ cells before further
maturation, as in the fetal retina trajectory (Figure 3.13A, 3.20C, Figure 2.15B). Sridhar et al.
also previously described features of transition populations from RPCs to photoreceptors,
ATOH7 in T1 and DLL3 in T3. ATOH7 was expressed in the bridge and early cones in the fetal
dataset overlapping the expression of CHRNA1, ONECUT1, and S100A6, while DLL3 was most
highly expressed in iPRP cells but remained expressed in later cone clusters (Figure 3.20B). In
the organoid data, ATOH7 was highly expressed in the RPC/MG co-clustered cone cells that
expressed CHRNA1 (arrowhead) similar to in fetal, while DLL3 appeared in the marker gene
plot and was highly expressed in early cones and slowly downregulated with maturation, also
similar to fetal cones (Figure 3.16).
Overall, early organoid cone marker genes (CTC-378H22.2, SNCG, SSTR2, PCAT4)
and early fetal cone marker genes (CHRNA1, ONECUT1, S100A6) were similarly found in early
maturing fetal cones, with RNAs only present in clusters Cone-1 and 2 detected in fetal cones,
indicating that these early organoid clusters best replicate early maturing fetal cone features.
Similarly, slow downregulating organoid genes (MIR7-3HG, RBP4, C8orf46) do not
downregulate in fetal data, suggesting a later maturation state that was not isolated from fetal
retina.

149

Figure 3.17: Marker genes for reduced organoid dataset
Up to five of the most specific markers are shown for each high-resolution cell cluster

150

Figure 3.18: Expression of retinal organoid rod cluster marker genes identified in Figure 3.17 in
organoid and fetal retina transcriptomes
Expression plots of marker genes that are most prominent in A. entire rod population, B. early maturation
rods, C. late maturation rods. Shown in organoid and fetal scRNA-seq datasets. Fetal rods divided into
two main clusters, labeled in A as ER (early rod) and LR (late rod).

151

Figure 3.19: Marker genes of early organoid cone clusters show two patterns of downregulation
and similar behavior in fetal cones
A. Expression plots of marker genes for early organoid cone clusters showing A. rapid downregulation
with maturation or B. slow downregulation with maturation. Shown in organoid and fetal scRNA-seq
datasets.

152

Notably, on examination of these gene expression patterns and RNA Velocity, there was
no indication of a CHRNA1+ rod population or a CHRNA1+ common cone-rod precursor state
as observed in developing fetal retina (Figure 2.16, 3.20A). A possible explanation is that a
CHRNA1+ common photoreceptor precursor state was present but not sampled. It is also
possible that this population may exist in younger organoid samples than those included in this
dataset, however CHRNA1+ cones were isolated from d55-57 and d83-84 samples and as rods
differentiate after cones it would be anticipated to find CHRNA1+ rods at the same or older
timepoints.

Late maturing organoid cones show coordinated upregulation of marker genes indicating
a state change
The third group of markers defining the mature cone clusters Cone-4 through 6 included known
cone features (GUCA1C, ARR3, CNGB3, MYL4, SLC24A2 (Figure 3.16, 3.21). Among these,
ARR3 was initially expressed in the least mature cone clusters and steadily upregulated to
Cone-6 (Figure 3.21A), the only example of slow upregulation among markers. GUCA1C,
CNGB3, and SLC24A2 were in all three clusters and increased with maturation (Figure 3.21B),
as was MCF2, which encodes a guanine nucleotide exchange factor (Komai et al. 2002), MYL4,
and a lncRNA CTD-2050N2.1 which was previously observed in cone cells (Welby et al. 2017).
MYL4 and GUCA1C were also found in the small Kuwahara method-enriched subgroup of late
maturing rods.
The fourth and final group of markers upregulated in the most mature Cone-5/6 clusters
and included PLA2G5, which we previously observed in late maturing fetal cones (Chapter 2),
GRK7, and a lncRNA TDRG1 (Figure 3.21C). These late cone markers upregulated preceding
OPN1LW, which was not identified by the cluster marker program but nonetheless was specific
to the most mature population of L/M cones collected (Figure 3.20D) The large group of genes,
including phototransduction genes like GUCA1C and CNGB3, undergoing a concerted

153

upregulation between Cone-3 and 4 showed that Cone-4 marks a distinct maturation-related
state change, with a smaller gene set upregulating in Cone 5 and opsin gene OPN1LW
increasing in Cone-6.
In fetal scRNA-seq analyses, ARR3 was sparsely detected across the early L/M cones (Figure
3.21A). Of the Cone-4-6 markers, MCF2 was weakly expressed in a few cells while CNGB3
showed sparse but increased expression in more mature L/M cones and S-cones (Figure
3.21B). SLC24A2 and CTD-2050N2.1 were found in more cells somewhat biased to the center
of the L/M cone population but did not show clear upregulation in the direction of the late
maturing cones, as observed in organoids. On the other hand, MYL4 and GUCA1C were tightly
expressed in the late maturing OPN1LW+ cells (Figure 3.21B,D), though in organoids both
genes upregulated before OPN1LW was detected. Finally, the Cone-5/6 genes were all
detected in very few cells, mainly in the late maturing cone group or adjacent to it, though
TDRG1 was more randomly found across the fetal early L/M cone population
Overall, the latest organoid cone maturation features PLA2G5 and GRK7 were
expressed in or adjacent to late-maturing OPN1LW+ cells in the fetal retina, while earlier Cone-
4-6 genes like MYL4 and GUCA1C are also closely correlated to OPN1LW expression and
supported the notion that the five most-distally positioned L/M cones comprise the late maturing
fetal cone state discussed in Chapter 2.

154

Figure 3.20: Marker genes of early fetal photoreceptor precursors are expressed in immature
organoid cones
A. Expression plots of marker genes for fetal iPRP (immature photoreceptor precursor) cluster, shown in
fetal and organoid scRNA-seq datasets. Markers of interest include CHRNA1 (expressed in the bridge
between fetal rods and cones), S100A6 and ONECUT1 (expressed in more mature early fetal cones and
similarly in more mature organoid cones). B. Expression plots for previously identified photoreceptor
precursor markers ATOH7 and DLL3. Green arrowheads: region of highest ATOH7 and CHRNA1
expression. Box: bridge region between fetal rods and cones where CHRNA1 is expressed in cone and
rod precursors.

155

156

Figure 3.21: Late organoid cone cluster marker genes are not detected in fetal cones
A. ARR3 detected early in organoid cones and upregulates into the most mature cells, while fetal
expression was only sparsely detected across L/M cones. B. Six markers showing upregulation in
organoid cluster Cone-4, with scattered expression in less mature cones. Fetal cones show GUCA1C and
MYL4 tightly localize to the small, late maturing cone group (green arrowhead) C. A smaller group of
markers upregulate in more mature clusters Cone-5 and 6 but in fetal cells genes are minimally
expressed in few cells. All genes in B and C upregulate before expression of D. OPN1LW.

157

Figure 3.22: Transcription factor regulon upregulation in late maturing cones
Violin plots for positive SCENIC regulons with highest cone cluster regulon specificity scores (RSS, see
Figure S3.1). Area-under the curve (AUC) scores are a measure of the combined activity of all genes in a
regulon, including the TF that drives it. Regulons are grouped by highest activities among A. all
photoreceptor clusters, B. early cone clusters, C. middle cone clusters, and D. late cone clusters.

158

Identification of a transition point in transcription factor activity in maturing cones
The patterns of cone feature expression observed between fetal and organoid retinal scRNA-
seq datasets suggest that the early organoid clusters are most comparable to fetal samples.
Specifically, the later clusters also provided an opportunity to examine features of late maturing
cone states we were unable to observe in fetal retina scRNA-seq data owing to the lack of late
maturing cells captured. We utilized SCENIC to identify transcription factor regulons in these
late states and evaluated whether early organoid regulons mimicked those from fetal data.
Regulon specificity scores were calculated for each cluster and all regulons that were in the top
five highest scores for at least one cluster were examined further (Figure S3.1). Broad regulon
signals for NEUROD1 and OTX2 were present in all cone and rod clusters and were also
observed in fetal data as well (Figure 3.19A). However, unlike in fetal regulons cone
transcription factors THRB and ISL2 did appear, though NRL appeared for rods (Figure S2.1,
S3.1). RAX was the only regulon with strong signal bias to early cones, thought it also showed
lesser AUC scores in early rods as well, while PRDM1 had similarly upregulated signal in early
cone and rod clusters (Figure 3.19B).
More unique signals were distinguishable for the late maturation cone clusters (Cone-4-
6). NR2F6 was the only regulon to plateau Cone-4, all others increased in signal through Cone-
6. GSC2 has been observed in late cones in previous organoid publications (Welby et al. 2017),
and MEF2C has been suggested as a regulator of rod homeostasis (Hao et al. 2011) though
there was low rod signal for this regulon (Figure 3.19C). MEF2D has been previously reported
to be influential in all photoreceptor maturation and health but had greater AUC levels in late
cones than rods. TLX2 and ESRRG regulons had similarly high Cone-5-6 expression with
consistent but lower rod cluster signals. These results show a large shift in transcription factor
regulon signals between Cone-3 and Cone-4, with far more TFs becoming active with
maturation rather than downregulating.

159

Differential expression of glycolytic and ribosomal protein gene expression between fetal
and organoid cone cells
Examination of the retinal organoid-derived transcriptomes indicated that the earliest organoid
clusters C1-3 most closely match our early maturing fetal cone clusters, with later organoid
clusters showing changes in gene and regulon expression as they expressed genes with only
minimal expression in late maturing fetal cones. To directly compare the developing cone
transcriptomes from both datasets, still excluding the aberrant organoid cone cell groups C-SG
and C3*, the two datasets were merged into a joint UMAP space (Figure 3.23A). Importantly,
organoid samples were originally sequenced alongside a subset of the fetal samples, and all
were normalized as a single sequencing batch to mitigate Seurat normalization of organoid-
specific features. We discerned the locations of RPCs, cones, and rods using markers as before
(Figure 3.23A,B (dotted boxes)).Likewise, the maturation direction was inferred by the
expression of cone and rod opsins, which were expressed most distal to the RPCs (Figure
3.23C).
In the combined UMAP space, the RPCs of both tissues overlapped, implying that they
had similar transcriptomic profiles. However, the fetal retina and retinal organoid cone and rod
cells segregated (Figure 3.23). Coloring the cells according to their original high-resolution SLM
clusters (Figure 2.14, 3.12) revealed that the relationships between fetal retina cells and retinal
organoid cells were largely retained (Figure 3.23D). Fetal photoreceptor cell clusters maintained
a similar structure as discussed in Chapter 2. The bridge population between rods and cones,
comprised of fetal iPRP cluster cells, was still detectable, fetal rods still segregated into two
distinct early rod (ER) and late rod (LR) groups, and S cones form a separate spatial group
(Figure 3.23A,D). The L/M cone clusters LM1 – LM4 were less clearly segregated likely
reflecting their similar transcriptomic profiles. The organoid PR groups retained the linear
relationships of the Rod-1-3 and Cone-1-6 clusters but remained distinctly separated and ran
parallel to their fetal photoreceptor counterparts. Notably, organoid rods localized more closely

160

with the late maturing fetal rods (LR). Besides RPCs, the region with most overlap between
tissues were the late maturing opsin-expressing cell groups for both tissues. RHO+ late
maturing fetal rods tightly bordered RHO+ organoid rods and a small number of RHO- fetal cells
overlapped the RHO+ organoid region (Figure 3.23A,C (red arrowheads)). The five late
maturing fetal cones, two of which were OPN1LW+, were also present within the late maturing
organoid cone population adjacent to OPN1LW+ organoid cells (Figure 3.23A,C (red
arrowheads)). Fetal S-cones still formed a spatially separate group, and some OPN1SW+ and
OPN1SW- organoid cells were placed near it (Figure 3.23A,C (blue arrowheads)), suggesting
they were closer to S-cones, although several other OPN1SW+ cells remained within the larger
mostly L/M cone cluster.
The patterns of photoreceptor marker genes, and regulons observed through cones from
both tissues indicated that the midpoint of the organoid cone trajectory, Cone-3-4, was similar to
the most mature fetal cluster LM4. To identify differentially expressed genes and processes that
distinguish these most related cones from each tissue, differential expression analysis was
performed between the cells of Cone-3 and LM4, excluding the five late-maturing LM4 cells
(Table S1, Figure 3.23D). Ontology analysis of significantly differentially expressed genes (p-
value < 0.05) showed significant LM4 over-representation (FDR < 0.05) of terms related to
translational targeting to the endoplasmic reticulum and RNA catabolic processes, which include
a large number of ribosomal proteins (Table 3.1), as well as cholinergic and glutamatergic
synapses (Figure 3.24A). Upregulated Cone-3 terms included glycolysis/gluconeogenesis and
nucleoside phosphate biosynthetic process, both caused by a similar gene set (Table 3.2).
Organoid terms also included GABAergic synapse, different from the two synapse terms for the
fetal cones and perhaps indicative of different synapse formation events occurring in fetal and
organoid cells at this time Both groups also shared the sensory perception of light stimulus
ontology (Table 3.1, 3.2).

161

Figure 3.23: Combined fetal and organoid scRNA-seq datasets show most overlap in RPCs and
opsin-expressing photoreceptors
A. UMAP plot of combined fetal (Chapter 2) and organoid scRNA-seq. Dotted line boxes: Black:
Combined RPC/MG population. Red: tissue separated cone populations. Green: tissue separated rod
populations. Solid line boxes: late maturing organoid cones and S cones (red) and late maturing rods
(green) shown C. Fetal S-cone population shown by arrowhead and label B. Known genes for RPC/MG
(VSX2), L/M cones (THRB), and rods (NR2E3). C. Zoomed regions of interest from A marked by green
and red boxes. Green: RHO+ cells from organoid and fetal samples. Fetal cells present in the organoid
population (red arrowheads). Red: late maturation organoid cone and fetal S-cone population, OPN1LW
on left and OPN1SW on right. OPN1LW+ late maturing cone cells (red arrowheads). OPN1SW+ organoid
cells (blue arrowheads) near the fetal S-cone population. D. SLM clusters for individual fetal and organoid
datasets overlaid on combined UMAP plot. Red box: iPRP rod/cone bridge. Dotted lines: Fetal cluster
LM4 (excluding late maturing cones) and organoid cluster C3 used in differential expression.

162

163

Figure 3.24: Differential gene expression in equivalent cone precursor clusters in retinal
organoids and fetal retina
A. Ontologies for significantly differentially expressed genes (p-value<0.05, log2FC>|0.5|) between fetal
LM4 and organoid Cone-3 (See Figure 3.23). Gene list for ontologies of interest for LM4 (Protein
localization to endoplasmic reticulum, sensory perception of light stimulus) and Cone-3 (Glycolysis and
gluconeogenesis, sensory perception of light stimulus) showing in Tables 3.1 and 3.2. B. Top 20 most
significantly upregulated genes for LM4 and Cone-3. C. Expression of select genes from B. D. Structural
protein genes differentially expressed between fetal and organoid photoreceptors.

164

LM4 Ontology Gene Lists
Cone-3 Ontology Gene Lists
Table 0.1: Gene list for select LM4 ontologies from fetal LM4 and organoid Cone-3 differential
expression
Table 0.2: Gene list for select Cone-3 ontologies from fetal LM4 and organoid Cone-3 differential
expression

165

Close examination of the top 20 most significant gene features (log2FC>|0.5|)
upregulated in each cluster (Figure 3.24B) showed that many were not unique to cone cells of
each tissue and were upregulated across all organoid or all fetal samples. Four which stood out
as different were CCDC141 and HDDC2 upregulated in fetal cones with low fetal rod
expression, MXNIP which was upregulated mainly in organoid cones, and FAM161 which was
downregulated in organoid cones compared to organoid rods and fetal photoreceptors (Figure
3.21C). Finally, two of the genes that were biased by only tissue type were of interest,
COL16A1, coding the alpha chain of XVI collagen and a component of extracellular matrix
(Grässel and Bauer 2012), was mainly expressed in fetal photoreceptors and TUBA4A, an
alpha tubulin subunit, was expressed primarily in organoid photoreceptors. Overall, many of the
most significant differences between these organoid cluster Cone-3 and fetal cluster LM4 were
tissue-wide and did not capture unique differences between only cone cells. Glycolytic terms are
increased in organoid samples and ribosome-related terms in fetal samples even when other
fetal/organoid comparisons are done such as rods (Figure S3.2). However, while we did not
identify features specific to the most mature fetal cluster or its organoid counterpart, we were
able to identify several genes that were biased to cones from one tissue.

3C. Aberrant retinal organoid cone clusters defined by hypoxic markers and
metallothioneins
The primary trajectory across retinal organoid cone transcriptomes replicated several
aspects of human fetal cone maturation, in particular expression of genes that are characteristic
of photoreceptor precursors and earliest cones. However, the initial organoid dataset also
included two spatially distinct populations (Figure 3.11A) whose clusters expressed marker
gene features only sparsely expressed in other cone clusters and had no parallel group in fetal
data. One of the side groups formed a distinct SLM cluster, termed C-SG and was comprised of
cells from organoids prepared with both the Kuwahara and Zhong methods. The other side

166

group, termed C3*, was a subset of the Cone-3 cluster and comprised solely of organoids
prepared using the Zhong method. It was necessary to remove the cells comprising these to
evaluate the major cone maturation trajectory, however we also needed to know what made
these cells unusual in order to recognize them in vivo and in transcriptomic data, and ultimately
to prevent their formation in future organoid culture.
As noted in the Chapter 3b marker gene analyses (Figure 3.11) the Cone-SG marker
genes were MFSD2A, BHLHE40, SCD, PFKFB3, and a non-coding RNA RP11-798M19.6
(Figure 3.25A). MFSDA had sparse expression in all cell groups outside the Cone-SG cluster,
and in fetal retinae only three early cones show any signal. This gene encodes a
lysophosphatidylcholine symporter that has been described in RPE controlling uptake of
docosahexaenoic acid for use in PR outer segments (Eser Ocak et al. 2018), but not in PRs.
BHLHE40 had a similar organoid gene expression pattern but with weaker expression in the
smaller side group (C3*) and the mature cones (in clusters C4-6), but only a few cells in the fetal
data had any expression. The protein is a transcriptional repressor that has been observed in
iPSC derived RPE as well as induced by hypoxia in mouse cell lines (Chuang et al. 2018, Choi
et al. 2008). SCD had much broader expression across the younger organoid cones and RPCs
which downregulated in the most mature cones. There was similar expression in fetal RPCs and
scattered expression across the fetal cones. The Stearoyl-CoA desaturase protein synthesizes
monounsaturated fatty acids also described in RPE (Samuel et al. 2001). PFKFB3 and RP11-
798M19.6 both had increased expression in a group of mature organoid cones, however RP11-
798M19.6 was also found in early cones as well as more rods. Both genes had similar scattered
expression all fetal cell groups. PFKFB3 produces a glycolytic enzyme and regulator 6-
phosphofructo-2-kinase/fructose-2,6-biphosphatase 3 (Xu et al. 2014) that helps drive
neovascularization in mouse retina. While several of these features were described in RPE, no
canonical RPE markers were expressed in Cone-SG (data not shown). Differential expression
of the cells around the spatially segregated portion of Cone-SG and the other cells versus

167

neighboring cluster Cone-3 revealed 101 genes up and 145 genes downregulated in the side
population (Figure 3.25B,C, Table S2) (pAdj<0.05, log2FC>|0.5|). Overrepresentation analysis
revealed that the upregulated side group genes comprised significantly over-represented
ontologies (Figure 3.22D) including protein localization to endoplasmic reticulum, cholesterol
homeostasis, ribosome, hypoxia response and mTORC signaling. mTORC is critical in aerobic
glycolysis and metabolism and along with hypoxia features suggest this side population was
hypoxic. The most significantly differentially expressed gene, FTL, encodes ferritin light chain
which controls iron sequestration. It also has native expression in retina (Hahn et al. 2004) and
upregulates in hypoxic conditions (Liu et al. 2020). Ontologies downregulated in the Cone-SG
cluster included cytoskeletal intracellular transport, oxidative phosphorylation, apoptosis and
sensory perception of light stimulus. Unusually, the top ontology was E. coli infection, however
the genes recognized in this set are mainly cytoskeletal structural genes (ACTB, ACTG1,
TUBA1A/B/C, TUBB3, YWHAQ) suggesting a reduction in structural protein expression in
Cone-SG cells. The downregulated oxidative phosphorylation genes further supported that this
population is hypoxic. Cone photoreceptors are often mis-localized internally in organoid
development, which could be a possible hypoxic photoreceptor population source for this cluster
in our sequencing data.
Marker genes for the Cone-3* cluster showed two genes mainly expressed in this
cluster, THY1 and KCNK15 (Figure 3.11). In the UMAP plot, both genes were most highly
expressed in and adjacent to the organoid Cone-3* group, while only KCNK15 had appreciable
expression in fetal cones (Figure 3.26A). THY1(CD90), a cell surface protein with predicted cell-
cell interaction functions (Rege and Hagood 2006), is known to be highly and specifically
expressed in developing fetal RGCs but more broadly expressed in retinal organoids (Aparicio
et al. 2017a). KCNK15 is part of a superfamily of potassium channels but little has been
published about it. These genes did not differentiate the Zhong-only C3* side population from
the rest of the cluster, so additional comparison was needed.

168

Differential expression analysis was again used to identify genes more specific to the small C3*
population (Figure 3.26B,C, Table S3). There were only 11 genes whose differential expression
was below the significance threshold (pAdj<0.05, log2FC>|0.5|), with 10 upregulated in the C3*
side population. All significantly differentially represented gene ontologies were related to metal
ion response processes (Figure 3.23D) driven by increased metallothionein (MT) gene
expression such as MT1M, MT1E, MT1F, and MT1G (Figure 3.23C,D, Table S3). MTs in the
retina have broad functions in zinc homeostasis and defense against oxidative damage
(Álvarez-Barrios et al. 2021). Besides metallothioneins, the gene with the highest fold change
and significance was ENPP2 (Figure 3.23 C,D, Table S3), which produces a secreted protein
member of the ectonucleotide pyrophosphatase/phosphodiesterase family also known as
autotaxin. ENPP2 had sparse expression across cone clusters, including the larger Cone-SG,
as well as weakly in rods, but showed greatly increased and focal expression in the C3* side
population. ENPP2 has been shown to upregulate in response to oxidative stress (Cholia et al.
2018), similarly to metallothioneins, which suggested that this population experienced increased
oxidative stress.

169

Figure 3.25: Aberrant side population C-SG has unique marker genes and enriched for hypoxia-
related genes
A. Marker genes for cluster Cone-SG. B. Original organoid scRNA-seq clustering plot with the C-SG
labeled and the two populations used for differential expression circled. C. Volcano plot of genes
upregulated in Cone-3 or C-SG cells. Labeled genes have p-value<10e-11 or are marker genes from A.
pAdj cutoff =0.05, log2FC. cutoff = |0.5|. D. Ontologies for upregulated genes in each group.

170

ENPP2 and BHLHE40 are upregulated in internal and border region organoid cone cells
and in normal fetal photoreceptor maturation
The hypoxia related features of the large cone side group suggested that these cells may be
internalized inside the organoid where they have poor access to oxygen, while the Zhong-
specific C3* side group showed features of oxidative stress. To determine whether internalized
cone cells could represent one or both side populations, immunofluorescence for ENPP2 and
BHLHE40 was performed. A representative age for each tissue was chosen for evaluation
based on side group sample ages in the sequencing data. The large C-SG cluster included cells
from Zhong-method organoids of all ages except d55-56, with most being d76 and d111-112 (10
and 14 cells), and from Kuwahara method organoids from d70 to d112 (Figure 3.27). The
smaller C3* side group included cells that were prepared from d76 to d112 and only from Zhong
organoids. Based on this, d70 Zhong and d100 Kuwahara organoid samples were chosen for
immunofluorescence analyses.
Three different tissue regions were assessed: neural retina (NR) is the bright organoid
border described in Figure 3.2, where cell nuclei form a clear layer that contains all cell types in
a layered fashion similar to the fetal retina; internal regions that are basal to the NR, including
the mostly hollow space within an organoid; and border regions that sit between the lobes of
multilobulated organoids characteristic of the Zhong organoid method. Both organoid production
methods produce NR and internal region (Figure 3.25A-C, (gold and magenta boxes)), while
only Zhong tissues develop border regions (white box).
In Zhong organoids, BHLHE40 showed minimal expression in cones populating the NR
but had significantly higher nuclear BHLHE40 expression in internal and border cones (Figure
3.28B,D) though border cells were poorly represented overall (n=10 cells from 1 organoid). In
Kuwahara organoids, the apical cone layer in the NR formed a well-ordered array of cells co-
expressing ARR3 and BHLHE40 (Figure 3.28C,D). Kuwahara organoids had fewer internal

171

ARR3+ cells but their BHLHE40 protein expression levels were not significantly different from
those in the NR, unlike the Zhong organoids. This similarity suggested that BHLHE40 was
broadly expressing in Kuwahara cones from all regions, and so we also looked at the protein
expression in human fetal retina. We observed that in the central regions of the retina,
BHLHE40 signal upregulated and in the presumptive fovea, where the ONL is only cone
photoreceptors, all ONL cells expressed BHLHE40. This increased protein expression in more
mature fetal cones suggested that d70 Kuwahara-derived cones in the NR achieved a similar
maturation level to express BHLHE40 protein. d100 Zhong cones did not upregulate BHLHE40
in NR but did internally, indicating that cell position correlated with early upregulation of the
protein.
ENPP2 was detected as a cytoplasmic signal in organoids produced with both methods
(Figure 3.26A, B, (insets)). Zhong organoids had low expression in NR cones compared to
internal and border cones while Kuwahara organoids had higher overall levels than Zhong
tissues with similar signal in cones from both NR and internal zones (Figure 3.29A-C).
Interestingly, the internal zone in Zhong organoids showed broad background expression of
ENPP2 (Figure 3.29A, (magenta region)), possibly suggesting high amounts of protein
secretion, while border cells had generally higher levels of cytoplasmic ENPP2 overall. In fetal
tissue cytoplasmic expression of ENPP2 in apical cone cells started to express the protein as
the nuclear layers began to separate and the signal increased into the presumptive foveal ONL
(Figure 3.29D). Kuwahara cone cells again showed high expression of ENPP2 at d70 across
organoid regions, suggesting that they mimicked the developing expression of the fetal cones.
Both internal and border cones from Zhong organoids had higher protein expression, which
indicates either population could be the source of the small side group cells. However, the
border region is Zhong method unique, making it the most likely location for a non-hypoxic,
ENPP2 high cone population.

172

Figure 3.26: Aberrant side population C3* shows increased expression of oxidative stress
upregulating genes
A. Marker genes for cluster Cone-3 identified in Figure 3.11. B. Original dataset clustering plot with the
C3* labeled and the two populations used for differential expression circled. C. Volcano plot of genes
upregulated in Cone-3 or C3* cells. pAdj cutoff =0.05, log2FC. cutoff = |0.5|. D. Expression plots for two
top upregulated C3* genes MT1M and ENPP2 in organoid and fetal datasets. E. Ontologies for
upregulated genes C3* are all metal ion related. No weighted set cover for ontology reduction used due to
small gene list.

173

Figure 3.27: Ages of cells in C-SG and C3* distinct populations
A. Retinal organoid Cone population with spatially distinct side populations C-SG and C3* circled and
labeled. B. Top: Zhong and Kuwahara method samples in C-SG by age of sample at collection. Bottom:
Zhong samples in C3* by age of sample at collection.

174

175

Figure 3.28: Increased BHLHE40 protein expression in internal organoid cones, apical Kuwahara
cones, and foveal fetal cones.
A. DAPI staining of d100 Zhong method organoids. Gold line: neural retina (NR). Magenta line: internal
(I). White line: border (B). B and C. Representative BHLHE40 (red) and ARR3 (green)
immunofluorescence in Zhong d100 (B) and Kuwahara d70 organoids (C). Regions are labeled and
colored as in A. Scale bar: 50um White arrows: ARR3+ nuclear BHLHE40+ cones. Blue arrows: ARR3+
nuclear BHLHE40 low/negative cone. Yellow arrows: ARR3+ nuclear and cytoplasmic BHLHE40+. D.
Quantitation of nuclear BHLHE40 mean grey value for cone cells in all three regions for each preparation
method. Cell numbers counted: Zhong: NR (59 cells in 4 organoids). Internal (66 cells in 4 organoids).
Border (10 cells in one organoid). Kuwahara: NR (132 cells in 5 organoids). Internal (21 cells in 5
organoids). Significance between regions by Kruskal-Wallis Test and if significant post-hoc Dunn test with
Benjamini-Hochberg correction: * = p<0.05, ** = p<0.005. Error bars are 1 standard deviation. E.
BHLHE40 expression in midperiphery and central FW16.6 retina.

176

177

Figure 3.29: Increased ENPP2 protein expression in internal and border organoid cones, apical
Kuwahara cones, and in foveal fetal cones
A and B. Representative ENPP2 (red) and ARR3 (green) immunofluorescence in Zhong d100 (A) and
Kuwahara d70 organoids (B). Regions: Gold line: neural retina (NR). Magenta line: internal (I). White line:
border (B). Scale bar: 50um. Green arrow: ARR3+ cytoplasmic ENPP2+ cone. Magenta arrow: ARR3+
ENPP2 low/negative cone. C. Quantitation of ENPP2 mean grey value for cone cells in all three regions
for each preparation method. Cell numbers counted: Zhong: NR (n=84 cells, N= 5). Internal (n= 49, N=5).
Border (n=14, N=3). Kuwahara: NR (137 cells in 4 organoids). Internal (31 cells in 4 organoids). Internal
(21 cells in 5 organoids). Significance between regions by Kruskal-Wallis Test and if significant post-hoc
Dunn test with Benjamini-Hochberg correction: * = p<0.05, ** = p<0.005*, *** = p<0.0005. Error bars are
1 standard deviation. D. ENPP2 expression in midperiphery and central FW16.6 retina.

178

Discussion
The accuracy of retinal organoid models is crucial for using them as replacement tissues or to
study human retinal development and disease. However, the continued observations of inter-
and intra-organoid heterozygosity (Browne et al. 2017, Phillips et al. 2018, Capowski et al.
2019) already demonstrates inconsistency in maturation timing between cells of the same type,
if not failure to express the same gene expression programs entirely. Recent scRNA-seq
comparisons between organoid and fetal retina, both focused on photoreceptor development or
on all cell types, consistently showed a birth order and progression of known cell type features
closely consistent with what is observed in fetal retina (Hoshino et al. 2017, Welby et al. 2017,
Kim et al. 2019, Sridhar et al. 2020). Differences do still occur though, with the death of RGCs
and lack of supporting RPE/vasculature, but also in gene expression levels between similarly
aged organoids and retina (Sridhar et al. 2020). These subtle gene expression differences have
not been reported for photoreceptors and may well be obscured in low read-count, large cell
number retinal scRNA-seq datasets.
In this chapter, the aim was to examine the cone cell maturation in H9-derived retinal
organoids produced with different methodologies and evaluate similarities and differences with
developing fetal cones. To that end, we looked at protein expression for a disease pathway of
interest before examining gene expression via scRNA-seq. We first looked at four protein
members of the pRB-loss proliferation program (Xu et al. 2014), pRB, MDM2, MYCN, and p27,
in ARR3+ cone cells. The three culture methods produced different number of ARR3+ cones
and had different timing of nascent OS formation (apical circular structures), with Kuwahara
method organoids producing more cones at d68-72 with some even showing small OS regions
(Figure 3.4B). By d98+ most cones in the apical photoreceptor layer of Kuwahara organoids had
a clear OS, while the other two methods still only intermittently produced them by d125, which
suggested an overall more rapid maturation of Kuwahara organoids.

179

pRB was detected in the nuclei of cones from all methods, with levels only decreasing
slightly in older age tissues. In fetal tissue at the time of pRB expression in ARR3+ cones, only
MG express a low level of pRB, while in organoids a large number of cells express the protein at
the same time (Figure 3.3B), which suggests either other cell types are aberrantly expressing
pRB or a large number of RPCs and/or MG persist in close proximity to these cones. Cones
also do not generally express higher pRB than surrounding cells either, indicating other cell
types are more highly expressing pRB than in fetal compared to cones, or cone expression is
lower.
MDM2 expression in fetal cones is concurrent with ARR3 and MDM2 levels are robust in
central cones/ even prior to detectable OS formation is detectable (due to tissue age limitations)
(Figure 3.4A). Organoid cones consistently showed low levels of MDM2 at young ages even
with strong ARR3 expression and OS formation, while even in the oldest tissues cells with
similar morphology and ARR3 intensity show notably different MDM2 signals. Nuclear MYCN
signal was also heterogeneous in cones across all methodologies, and while d69-70 Kuwahara
organoids showed some indication of cytoplasmic and nuclear expression similar to that
observed in foveal fetal cones, older organoids had no detectable MYCN in most cones. As fetal
cones do not complete OS formation in our fetal tissue, it is possible that fetal samples may lose
all MYCN expression at tissue ages we cannot obtain, but in organoids the low MYCN
expression state similar to available foveal cones was almost never observed in any method.
p27 expression was always nuclear in all ages of all methods, never showing the
cytoplasmic downregulation observed in foveal fetal cones. This may be a feature of foveal
cones that our culture is unable to recapitulate. However, interestingly, p27 sometimes showed
upregulation (Kuwahara) or downregulation (Zhong) with age. Morphologically ARR3+ cones in
organoid models achieve a state of OS formation rarely seen in available fetal tissue samples,
and yet they did not capture foveal expression patterns (MYCN, p27) or protein timing that in
fetal retinae correlated with ARR3 (RB, MDM2).

180

The scRNA-seq data gathered in Chapter 2 provided a large human photoreceptor
lineage transcriptome dataset against which organoid-derived cone cells could be compared in
terms of gene expression timing and maturation states. Two of our chosen culture methods,
based on the Kuwahara et al and Zhong et al methods (Kuwahara et al. 2015, Zhong et al.
2014), proved more reliably reproducible while having noticeable differences in cone number
and maturation rate, so we chose to use organoids from these two methods to FACS isolate
867 cells from seven different timepoints. These timepoints were chosen to cover ages from
around the earliest ARR3 expression up to a timepoint similar to that of the oldest available fetal
sample (FW19 = 133 days), with one older timepoint at 225 days. We isolated primarily cones,
rods, and an RPC/MG population, similar to what was captured for fetal retina.
Examining these data for any production method differences, we found that Kuwahara-
derived cone cells overwhelming made up the most mature cell clusters, while Zhong-derived
cones made up an early cone cell cluster as well as a small side population termed C3* (Figure
3.10A). Along with the observations of rapid ARR3 expression and OS formation in Kuwahara
method, as well as the Kuwahara cells occupying more mature clusters than Zhong samples of
the same age (Figure 3.12D), these findings supported that the Kuwahara method generated
more cones and caused more rapid expression of mature cone features than other methods.
Mainly rod transcriptomes were isolated from d225 Zhong organoids, which also indicated this
methodology produced a greater number of rods than cones.
There are protocol differences which likely cause these variations in photoreceptor
development between protocols, however there is no concrete evidence for one specific factor
being responsible. Our modified Kuwahara protocol utilized specific factors to drive eye field and
retinal cell fate (BMP4/SB/LDN) compared to using Matrigel as in the Zhong protocol, which
may cause more rapid retinal cell fate specification and thus cell type differentiation; however,
there is no published evidence for this. This is also a major difference with the Nakano method
organoids, which also utilized Matrigel and had fewer ARR3+ cone cells compared to Kuwahara

181

method organoids (Figure 3.3-3.6). The Kuwahara protocol also used an added lipid
concentrate in the early culture media; however, while lipids can function as energy sources for
photoreceptors there is no indication this might accelerate photoreceptor maturation (Fu et al.
2020). Zhong organoids are first exposed to retinoic acid almost a month after Kuwahara
organoids, and while retinoic acid has been shown to induce cone features in retinoblastoma
cell lines (Li et al. 2002), retinoic acid has been particularly shown to foster rod cell fate while in
some cases inhibiting cone cell maturation in cultured cells, rodents, and zebrafish (Kelley et al.
1994, Hyatt et al. 1996, Zhong et al. 2014, Brzezinski and Reh 2015, Kim et al. 2019).
Kuwahara organoids receive earlier and longer retinoic acid exposure which may explain more
rapid cone cell maturation; however, based on literature showing increased rod production with
added retinoic acid it doesn’t explain the reduced number of rod cells relative to cones that this
method produced. FBS is also added later in the Zhong protocol, which is intended to preserve
long-term tissue structure (Zhong et al. 2014) but there is no indication it may alter maturation
rates of cells in any tissue. Finally, the mechanical changes due to the Zhong protocol’s 3D-2D-
3D organoid culture steps could cause delayed cell type maturation as the tissue has to
restructure twice which might impact cell-cell interactions and signaling gradients involved in
normal retinal cell development.
Two unusual side populations appeared in our cone data, which were initially removed
to evaluate the features of the main cone population. By looking at marker genes for the
remaining dataset, we found that well-known photoreceptor-development genes were the most
specific markers of the cell clusters during progression through both rods and the later
maturation cones (Figure 3.17). The most distal cone cluster was also where OPN1LW was
detected, which was similar to the small population expressing the gene in the fetal data.
However unlike in fetal retina, no S-cone specific population was resolved (Figure 3.16) and
fewer OPN1SW+ cells expressed S-cone markers other than CCDC136. Eldred et al. 2018
demonstrated the capability to produce and manipulate S-cones in culture, suggesting either

182

that our chosen production methods or FACS isolation did not provide enough captured S-
cones for sequencing.
Organoids successfully recapitulated several aspects of fetal cone maturation. The
organoid photoreceptors demonstrated the same NRL and RXRG isoform use patterns between
rods and cones as seen in fetal samples (Figure 3.17). Marker gene expression features of
early organoid clusters (SNCG, SSTR2, PCAT4) were also found to have clear expression in
our fetal scRNA-seq analyses; SSTR2 and PCAT4 were found in most of the fetal L/M cone
cells, indicating these earlier organoid cells had gene expression similar to that of most of the
sequenced fetal cone precursors. Early organoid cone clusters also expressed features of
immature photoreceptor precursors identified in Chapter 2: CHRNA1, CTC-378H22.2, S100A6,
and ONECUT1. Most of the cells expressing these genes were captured from d55-56 samples,
showing consistent expression of these earliest cone features. Notably, CTC-378H22.2+
samples were most captured from Zhong organoids and formed cluster Cone-1. This is
consistent with Zhong cells being less mature, so I was able to capture this early focal
expression (Figure 3.20).
Several late maturation cone marker genes (GUCA1C, MYL4, CNGB3, SLC24A2)
showed a coordinated upregulation of gene expression in the middle cone clusters (Cone-3-4)
suggestive of a state change that occurred before upregulation of OPN1LW. Indeed, in our fetal
retina scRNAseq analyses MYL4 and GUCA1C appeared in the small population of OPN1LW+
cones that formed a spatially distinct UMAP group, hinting that they comprise a new state.
SCENIC regulon signals further support the late maturation state changes. Late appearing
regulon signals, GSC2, MEF2C, MEF2D, TLX2, and ESRRG were similar in Cone-1 and Cone-
2, but from Cone-3 on the activities of these regulons rose (Figure 3.22C). MEF2C was recently
noted in organoid and human retina data to be enriched in cones (Kallman et al. 2020), but our
data connects increased MEF2C regulon activity to increased activity of the other four TFs in
Cone-3 and later, aligning well with the marker gene expression change from Cone-3 to Cone-4.

183

This state change also included two lncRNAs not detected in the Chapter 2 fetal scRNA-seq
analysis, CTD-2050N2.1, which was previously observed in cones by Welby et al., and TDRG1.
Both tissues have demonstrated changing lncRNA expression patterns with photoreceptor
maturation, showing early CTC-378H22.2 expression in developing cones for example, which
indicates this phenomenon is consistent across tissue types. This raises an interesting broader
question about whether similar maturation-related lncRNA behavior occurs in other tissues
during development.
For a direct comparison between organoid and fetal cone transcriptomes, we combined
the two into a single dataset. The state change that occurred in organoid Cone-3 and Cone-4
included upregulation of MYL4 and GUCA1C, which was only detected in the small late
maturing fetal cone group. We chose to compare these two clusters, the most mature captured
fetal cones and the midpoint of our organoid cones, where this upregulation occurred.
Ontologies identified an increased activity in ribosome and endoplasmic reticulum protein
targeting genes in fetal cones, while organoid cones had increased glycolysis/gluconeogenesis
and nucleoside phosphate synthetic processes. The altered metabolic features of organoid
cones could be a culture-related artifact (Bhaduri et al. 2020), as culture stress may exacerbate
the need for energy or provide excess glucose that must be metabolized; however, it is not clear
why fetal samples have increased ribosomal gene expression. One possibility is that the fetal
cone precursors have increased MYCN expression, as MYCN is a major driver of ribosomal
protein gene expression. Two specific genes of interest also stood out between the organoid
versus fetal photoreceptor gene expression, COL16A1 and TUBA4A (Figure 3.21). Fetal
photoreceptors express greater levels of the ECM component COL16A1, potentially indicative of
the more organized structure achieved in native tissue over organoid culture. Tubulin is
important the cilia structure for photoreceptors (Woodford and Blanks 1989, Arikawa and
Williams 1992); thus, the increased levels of tubulin in organoid cells may be related to the
rapidly forming segment structures observed in organoid cones.

184

Finally, we more closely examined the aberrant cone populations identified in the initial
retinal organoid dataset. Marker gene analysis found unique markers for both groups as well as
increased hypoxia or metal ion/oxidative stress terms from differential expression against a
neighboring cell cluster. We found that internalized ARR3+ cells expressed higher levels of C-
SG marker BHLHE40 in Zhong organoids, consistent with these cells lacking access to oxygen.
Internal and border region cones also had increased C3* marker ENPP2 in Zhong, and though
the sample number for the border cells is too low to make a statistical evaluation it is suggestive
that this population may be the source of the Zhong-specific side group undergoing oxidative
stress (Figure 3.25, 3.26). No similar internal/external expression difference was noted in
Kuwahara organoids, however we found that more mature fetal photoreceptors in the central
ONL had high endogenous expression of both BHLHE40 and ENPP2 proteins, indicating that
while the less mature Zhong cones only expressed these proteins in the aberrant cell
populations, Kuwahara cones had reached a point in maturation where they were also
expressing the two proteins normally.
The work in this chapter broadly examined the features and timing of photoreceptor, and
particularly cone, protein, and gene expression. We observed abnormal timing and
heterogeneous expression of proliferation related proteins in three different methods of organoid
production. Gene expression data showed that the Kuwahara methodology produces a greater
number of rapidly maturing cone cells, but cones from both organoid production methods
successfully capture some features of photoreceptor maturation, including a distinct late
maturation state switch distinguished by joint transcription factor and cone gene expression
changes. Finally, while not typically addressed in publications, organoids produce abnormal
cone populations outside organized neural retina regions with distinct stress-related
transcriptomic profiles.

185

Materials and Methods
Stem cell culture:
H9 line human embryonic stem cells were grown and maintained by the Children’s Hospital Los
Angeles Stem Cell core. Cells were grown on 35mm dishes.
Feeder-based: To passage, 100,000 irradiated mouse embryonic fibroblasts were plated
on a 0.1% gelatin (SIGMA G1890) coated dish. 1mg/ml collagenase (Gibco 17104-019) was
made with DMEM/F12 (Gibco 11330-032). Stem cell plates were washed with 2mL of 1xPBS
and then covered with 1mL of collagenase solution while being incubated at 37C for 2 minutes.
Plates were evaluated under a dissection microscope and a p200 pipettor tip was used to
manually scrape up pieces of optimal appearance colonies. 40-50 pieces were seeded onto the
new MEF-covered plate and then returned to the incubator to attach and grow.
Feeder free: 35mm dishes were coated in a solution of 40ul Vitronectin-XF (Stem Cell
Technologies 07180) per 1mL of CellAdhere dilution buffer (Stem Cell Technologies 07183) and
left for at least 1 hour at room temperature (RT) before use or stored at 4°C. Plates were
washed with 1mL of fresh CellAdhere buffer and then 1mL of TeSR-E8 before use. To release
cultured cells, the dish was washed with 1xPBS before 1mL of ReLeSR (Stem Cell
Technologies 05872) was added and then the majority removed after 1 minute leaving a low
covering over the cells. The plate was incubated for 7-9 minutes before 1mL of TeSR-E8 was
added and colonies were detached by tapping the sides of the plate. The cells were removed
with a 5mL glass pipette to a 15mL conical tube and triturated to break aggregates down to
smaller sizes. A newly coated plate was then seeded with the desired density for a new
passage every 4-5 days.

Retinal organoid differentiation:
The three production methods for retinal organoid differentiation were implemented as
published (Nakano et al. 2012, Kuwahara et al. 2015, Zhong et al. 2014), with some alterations.

186

Prior to use colonies are cleared of any differentiated or poorly growing cell groups via
mechanical removal. To briefly describe each:
Nakano: H9 ESCs were treated with 10uM Rock inhibitor (RI) (Tocris #1254) for at least
an hour before the cells were treated with 750ul RT Accutase (Life Technologies A1110501) for
approximately 3 minutes at 37°C to lift the supporting fibroblasts. After removal, an additional 8–
10-minute Accutase treatment followed by gentle pipetting with a 1mL serological pipet
dissociated the colonies fully. The volume was collected in DMEM/F12 (5mL per dish of H9) and
centrifuged at 300g for 5 minutes at RT. The supernatant was removed and the cell pellet gently
resuspended in Aggrewell media (Stem Cell Technologies 05893) containing RI (~500ul/dish)
before being counted to determine concentration. 5000 cells in 30ul were placed into Lipidure-
coated u-bottom 96 well plates and spun at 400g for 5 minutes before overnight incubation. On
day 1, media was replaced with 50 ul retinal differentiation medium (RDM). On day 2, 130ul of
RDM with 3.6uM of the WNT inhibitor IWR1 was added to each well, then 20ul of cold 5%
Matrigel in DMEM/F12 was added to the center of each well to fall over the organoid. After four
days of incubation, media is replaced on day 6 with 200ul of RDM containing 3uM IWR1, and
this change is repeated on day 9. Days 12 and 15 the media is replaced with 200ul RDM2
(RDM in which 10% FBS replaces the KSR) containing 100nM SAG and on day 18, organoids
are transferred to HEMA-coated 10cm dishes and cultured long term in retinal maturation media
(RMM)containing 0.5uM retinoic acid and 0.1mM taurine. A maximum of 20 organoids are
placed in each dish with 100ul of media per organoid per day.
Kuwahara: ESCs are handled as described for Nakano, except 20uM of RI is used in all
preparation media and the Aggrewell media contains 3uM IWR1, 10uM SB431542 (SB)
(Cayman Chemical #13031), and 100nM LDN193189 (LDN) (Sigma SML0559) both BMP/TGF-
β inhibitors. The combination is collectively referred to as ISL and was previously described as a
modification on the Kuwahara protocol (Browne et al. 2017). 12,000 cells in 76ul of Aggrewell
media suspension are dispensed into each u-bottom well to form aggregates. On day 1

187

Aggrewell media is replaced with 100ul of gfCDM + ISL, while on Day 6 all previous media is
removed and 200ul gfCDM + 1.5nM BMP4 (R&D Systems 314-BP) is added. On days 9, 12,
and 15 half of the media is removed and replaced with 100ul fresh gfCDM to dilute the initial
BMP4 concentration each time. Day 18 organoids are moved to coated 6cm plates with RMM +
RA/Taurine as Nakano method.
Zhong: on the first day, feeder-free ESCs were incubated for 1 hour with 4mL mTeSR
media containing 10uM blebbistatin before being washed with 1x PBS and incubated with 1mL
accutase for 5-10 minutes at 37°C until the colonies dislodged. Colonies were dissociated to
smaller clumps via gentle trituration, then placed into a 15mL tube containing 3mL
mTeSR+blebbistatin and centrifuged at 100g for 1 minute. The supernatant was aspirated, and
cells were resuspended in 6mL mTeSR+Blebbistatin and placed onto a 60mm, uncoated dish. If
aggregates are larger than approximately 200um, gently pipette until small enough. On day 2
the new aggregates were collected in a 15ml tube, the plate was washed with 2mL of mTeSR
and transferred to the same tube, then the volume was centrifuged at 100g for 1 minute, the
media is aspirated and replaced with 5mL of mTeSR:NIM (3:1) media mix before being returned
to the culture dish. The media was similarly changed on day 2 but replaced with mTeSR:NIM
(1:1). On day 6 a plate is coated with Matrigel so that on day 7 the aggregates can be
transferred in NIM media (~250 aggregates/60mm dish) and allowed to attach. NIM media is
replaced on day 9 and 13, then on day 16 it is replaced with Zhong 1 media and changed every
day until day 27. Under a dissection microscope, regions of neural retina are identified and cut
with tungsten needles, then transferred to a new low-adhesion dish. The dish is checked for
several days for new regions to dissect. Media is changed to Zhong 2 on day 41, retinoic acid is
added to the media on day 69 (1:10,000), then halved in concentration at day 97 onward
(1:20,000)

188

In all three methods, organoids in long term culture were periodically separated with
forceps if two or more began to adhere to each other and media changes were done every 2-3
days.

Cryosectioning of Retinal Organoids:
Retinal organoids were collected for fixation based after light microscopy evaluation of
morphology and structure. The desired number of tissues were collected in a 1.5mL tube in
culture media and allowed to sink to the bottom of the tube. The media was removed by pipet
and then the tissues were washed in 1x PBS. The PBS was removed and 1mL of 4% PFA in 1x
PBS was added. The samples sat at RT for 12 minutes and the tube was mixed every 2-3
minutes to ensure even exposure of the organoids. After, the PFA was removed the tissue was
washed 3 times for five minutes with 1x PBS before being incubated in cold 30% sucrose ON at
4°C. Samples were then mixed with a trypan blue/OCT mixture before being placed in molds
with additional OCT; this allows us to identify the region of the mold containing the small tissues
clearly. Molds were frozen on dry ice before storing at -80°C. Samples were removed from mold
forms for sectioning on a Cryostat at ~-14°C. Samples were 10um thick and approximately 4-6
were placed on a single slide before storage at -80°C

Immunostaining:
To perform immunofluorescence for ENPP2 and BHLHE40, stored retina and organoid tissue
sections were warmed on a 37°C slide warmer for 2 minutes, before being removed and left at
RT for 10 minutes to dry. Slides were washed with 1xPBS for 5 minutes to remove excess OCT,
a border was drawn with wax pen around all sections on each slide and allowed to dry. Samples
were simultaneously permeabilized and blocked in super block solution (2.5% horse serum,
2.5% donkey serum, 2.5% human serum, 1% BSA, 0.1% Triton-X-100 and 0.05% Tween-20 in
PBS; filtered with 0.22mm filter) for one hour at RT. Primary antibodies were diluted in super

189

block and then placed on tissue sections and stored overnight at 4°C in a humid chamber, The
next day, the liquid was aspirated and 1xPBS was used to wash each section three times, one
rapidly removed and the following washes for 10 and 5 minutes, respectively. Secondary
antibodies were diluted in super block, then applied to the tissue sections for 30 minutes at RT.
Samples were then washed twice more for 5 and 10 minutes, then a third time with 1xPBS
containing DAPI (1:1000). Slides were aspirated and then mowiol mounting media was applied
before covering with a cover slide. Cone Arrestin antibody was provided by Cheryl Craft (Zhang
et al. 2001, Li et al, 2003).

Organoid tissue dissociation, single cell isolation, and processing for scRNA-seq:
To collect samples for scRNA-seq, samples were gathered from organoids generated with the
Kuwahara-ISL and Zhong methodologies across 8 timepoints. Two separate preparations for
each method were gathered for each timepoint unless otherwise indicated. Kuwahara
timepoints: d56-57, d70-71, d84-85, d98-99, d112-113, d126-127, d140-141. Zhong timepoints:
d55-56, d76 (single replicate), d83-84, d98 (single replicate), d111-112, d125-126, d139-140,
d225 (single replicate). At each collection point, three retinal organoids were selected based on
morphology under light microscopy. These were imaged and placed into a 12 well plate with 1ml
of 10u/ml Papain (Worthington LK003176) solution before being incubated at 37°C for 10
minutes. Dissociation proceeded as described for fetal tissue from this step; organoid tissues
proved more difficult to break apart and required additional mechanical dissociation through
pipettor tips at both 1000ul and 200ul sizes. As previously described, the reaction was halted
and samples were centrifuged and resuspended for counting.
Dissociated organoids were FACS-enriched for cone cells as described in chapter 1 and
single cells were directly collected from the FACS. Samples were similarly processed in low
volume to produce cDNA before being stored at -20°C. All DNAS quantitation and library
preparation were done as previously described before being sequenced on the Illumina HiSeq

190

4000 (2x75) platform at the Center for Personalized Medicine at Children’s Hospital Los
Angeles.

Data processing and analysis:
The software packages and processes used for all steps of data alignment, quality control,
visualization, and analysis are the same as described in chapter 2, with the exception that no
Seurat sequencing batch normalization was needed. The combined fetal retina and retinal
organoid dataset was generated using the same batch normalization as for the fetal retina in
chapter 2 with the organoid samples included with the fetal samples they were sequenced with.
The organoid samples were not identified as a unique batch so that differences between tissue
types would be better preserved. Differential expression analyses in the combined dataset were
performed as described for the individual datasets.

191

Antibody List:
Protein Target Species Source Cat. Number Dilution
ARR3 rabbit Cheryl Craft
(Zhang et al. 2001, Li et al. 2003)
NA 1:5000
p27 mouse BD Bioscience 610241 1:600
MDM2 mouse Santa Cruz SC-965 1:100
MYCN mouse Santa Cruz SC-555 1:100
RB mouse Cell Signaling Technologies 9309 1:200
ENPP2/Autotaxin rabbit Proteintech 14243-1-AP 1:200
BHLHE40/DEC1 rabbit Novus Biologicals NB100-1800 1:1000

192

Supplemental Data

Figure S3.1: SCENIC Regulon Specificity Scores (RSS)
Top five highest specificity scores per cluster are labeled.

193

Figure S3.2: Gene ontologies for differential expression of organoid vs fetal rods
Reduced by weighted set cover. Increased glycolytic/metabolic terms in organoid rods captured under
terms mTORC1 signaling, metabolic pathways. Increased ribosomal terms in fetal rods by protein
localization to endoplasmic reticulum and RNA catabolic process.

194

Table S3.1: Differentially Expressed Genes Between Fetal LM4 and Organoid Cone 3

p-value>0.05, log 2FC>|0.5|
Symbol log2FC P-Value Adj Symbol log2FC P-Value Adj
Cone 3 GUCA1A -4.530821936 1.20E-31 LM4 COL16A1 2.546062347 1.80E-26
AL365357.1 -2.87163571 4.65E-24 SLC40A1 2.763117505 2.24E-24
AL591846.1 -2.645091403 1.69E-23 HDDC2 1.956538759 1.01E-22
VAX2 -2.402623634 4.04E-21 FAAHP1 5.800563017 4.67E-22
TXNIP -2.174747076 1.22E-20 FAM161A 1.534452987 3.36E-21
MAP2 -2.076985345 1.04E-19 EEF1A1P5 0.946256402 1.10E-19
ITM2C -2.054362682 1.63E-18 LOXL1 1.74922239 1.77E-19
IMPDH1 -2.028630823 2.02E-18 RCVRN 1.73000342 1.56E-18
LMOD1 -1.875972357 4.17E-18 DUSP1 2.483253833 5.15E-18
SEPT4 -1.742406466 1.95E-17 LMO1 1.813514447 8.31E-16
HRASLS -1.717328989 1.73E-16 DUSP26 1.617994339 9.03E-16
KIF2A -1.71677853 1.24E-15 EEF1A1 0.899590094 2.17E-15
ENO2 -1.693184938 1.25E-14 TMX1 1.423614934 3.96E-15
GDPD5 -1.660664327 1.57E-14 SPON2 1.783309499 4.72E-15
RS1 -1.654272048 2.00E-14 MAR1 1.338646742 5.65E-15
MLXIP -1.596824986 4.13E-14 EEF1A1P6 0.79350766 9.38E-15
TUBA4A -1.552209293 1.77E-13 APMAP 1.285857936 3.05E-14
GUCA1B -1.53140332 1.04E-12 CCDC141 1.833126274 7.93E-14
PFKP -1.497190488 1.09E-12 GRAMD1C 1.469254549 9.57E-14
ALDOA -1.489808526 3.35E-12 CCDC60 1.670397942 1.48E-13
ARR3 -1.416704607 4.17E-12 CABP2 1.784214716 2.46E-13
NANOS1 -1.375136895 2.23E-11 IMPG1 1.590681211 4.21E-13
CHN2 -1.356047532 2.94E-11 SLC6A15 0.97989967 3.56E-12
MAP1LC3A -1.331073861 3.64E-11 FOS 1.649099075 1.21E-11
DPYSL3 -1.274514003 4.14E-11 LOXL1-AS1 2.431277303 1.54E-11
MLLT11 -1.262183119 1.53E-10 RPL23A 0.841096362 1.75E-11
SHISA5 -1.244203322 1.63E-10 AC011603.2 0.895790277 5.18E-11
MVP -1.221492775 3.01E-10 RAX2 1.052037143 5.21E-11
UNC119 -1.21364919 1.43E-09 AC026366.1 0.95130693 7.06E-11
TMEM35A -1.177373 1.55E-09 RPL23AP42 0.663977262 9.52E-11
DPYSL5 -1.153368431 3.41E-09 TUBA1C 0.706835949 1.43E-10
AC245297.1 -1.151327455 5.22E-09 RPL7 0.81049403 3.02E-10
ARHGAP42 -1.151023201 8.49E-09 AIPL1 0.855908719 1.05E-09
HLA-A -1.140135993 9.90E-09 ID1 2.085700505 1.12E-09
PCDHB2 -1.135626576 1.68E-08 RXRG 0.732925168 1.69E-09
FJX1 -1.131226397 1.82E-08 MFNG 1.380155791 1.97E-09
NEUROG1 -1.128427743 2.12E-08 RPL7P1 0.541656476 4.43E-09

195

NME1 -1.123423206 2.74E-08 DCT 1.201680574 4.67E-09
CAMK1D -1.118037149 2.98E-08 NIPAL1 1.038839631 1.46E-08
AMPD2 -1.117978606 3.20E-08 CPLX3 1.368336938 1.84E-08
NREP -1.114430922 4.96E-08 LHX3 0.707601593 2.16E-08
SLC38A3 -1.104327451 6.43E-08 GRM6 1.13856057 2.72E-08
GNGT2 -1.101504275 6.80E-08 SLC25A6 0.94078391 3.20E-08
KCTD6 -1.100571487 9.51E-08 LINC01896 1.116853926 3.59E-08
CNNM4 -1.084047861 1.12E-07 ZNF667-AS1 1.479402633 4.96E-08
THY1 -1.077526208 2.63E-07 EPB41L2 0.839728538 7.98E-08
NXNL2 -1.076313021 6.42E-07 MAGEH1 1.347903614 1.00E-07
Vax2os1-3 -1.06944834 1.02151E-06 RPL18A 0.79996385 1.14E-07
USP11 -1.068460586 1.07002E-06 NEUROD1 0.799667313 1.48E-07
APOE -1.06270972 1.07885E-06 CAMK2B 1.115068634 1.54E-07
PKM -1.060657764 1.11064E-06 JUND 0.816132519 1.93E-07
TSPAN7 -1.058657556 1.4599E-06 DUSP2 1.434973401 3.76E-07
CNTNAP2 -1.054355659 1.83621E-06 OTX2 0.829014608 3.78E-07
PIK3R1 -1.050898584 2.00297E-06 CTSF 1.640135589 4.74E-07
ALDOC -1.04862725 2.14181E-06 SCN1B 0.842611754 5.23E-07
CCDC144NL-AS1 -1.035724363 2.37811E-06 TMEM244 1.058984091 1.03688E-06
EDIL3 -1.021745891 3.9083E-06 GEM 1.440572442 1.16385E-06
SMIM4 -1.021454807 6.23411E-06 RP1 0.961711328 1.27761E-06
HLA-B -1.008722411 8.47866E-06 RPL7P9 0.627758179 1.92499E-06
PDC -0.987606333 8.72337E-06 ANO2 0.914175248 2.01921E-06
GSTM3 -0.977585912 1.10225E-05 CBWD1 0.708234322 2.56161E-06
UBAP1L -0.966755649 1.43733E-05 TUBA1B 1.154305948 3.33691E-06
CA2 -0.966236314 1.51166E-05 LINC00632 1.373782919 5.60821E-06
NEDD4L -0.947813418 1.55724E-05 DOCK8 0.886600582 7.49505E-06
BLOC1S2 -0.940398277 1.64422E-05 GNB1 0.94491155 1.19632E-05
PDE4DIP -0.936462929 2.00032E-05 RPL19 0.752558235 1.27889E-05
CALCOCO2 -0.930772597 2.24411E-05 GPC5 1.102352684 1.45896E-05
WFDC1 -0.929611218 2.29092E-05 RPL10P16 0.547771384 1.75544E-05
PGAM1 -0.923172067 2.66971E-05 AC079760.2 0.92591018 1.87869E-05
CD24 -0.920697559 3.9931E-05 CD151 0.523443519 1.89002E-05
FAM135B -0.919205469 4.28123E-05 OLAH 1.425474599 2.76362E-05
GPI -0.915939645 4.42447E-05 GADD45G 1.055354351 4.00941E-05
AKAP9 -0.915892051 5.29547E-05 HNRNPA1 0.631933843 6.44619E-05
AC012085.1 -0.906935228 6.62238E-05 NR4A1 1.148793313 6.83505E-05
FAM162A -0.906606215 8.64933E-05 GRIK2 0.835837518 6.87203E-05
PPP4R4 -0.904879492 9.65137E-05 MSR1 1.165833315 6.93736E-05
SEMA4B -0.901028912 0.000123429 NRN1L 1.383340044 7.8638E-05
CRABP2 -0.899294333 0.00013225 CAMK2N1 0.801013746 0.000104593
AC093752.1 -0.895920573 0.000149405 RPLP0 0.670931873 0.000110487

196

AC087783.2 -0.890112491 0.000154496 MAP1B 0.912759037 0.000118602
KLF10 -0.889867037 0.000157601 ARL4A 1.066136604 0.000142645
LRRN2 -0.877082911 0.000215713 ME3 1.060019337 0.00016606
CORO1C -0.870051006 0.000225528 INSM1 0.740209669 0.000179981
UBC -0.865109824 0.00023186 ZFP36L2 0.771962444 0.000197643
P4HA2 -0.856993758 0.000235787 CKAP4 0.768692143 0.000201373
AC024610.2 -0.854175111 0.00024214 AC004448.1 0.521575619 0.000229422
ATP5I -0.850833397 0.000246468 BTG2 1.076899775 0.000245222
SCG3 -0.847414161 0.000254635 PCAT4 0.819730561 0.000283021
TMEM237 -0.84435132 0.000270272 GNG5 0.712966066 0.000357121
GABRR2 -0.838128037 0.000273496 ENC1 0.565914057 0.000445571
C4orf48 -0.823464238 0.000274206 HNRNPA3 0.546194072 0.000512217
NR2F1-AS1 -0.822331577 0.000298346 GNG13 0.707218268 0.000553749
ARRDC4 -0.822078127 0.000345859 GALNT13 0.726429628 0.000598946
TROVE2 -0.81407486 0.000380811 FXYD6 0.861611938 0.000599659
FAAH -0.811852087 0.000409018 GRIK3 0.848903208 0.000617402
TUBB2B -0.801328637 0.000415987 LINC01563 1.036170869 0.00064328
NRN1 -0.79131975 0.000481962 ISL2 0.80692745 0.000691722
PRPH2 -0.791183528 0.000494617 MATK 0.717125355 0.00072148
CDADC1 -0.78440249 0.000744107 AMER2 0.544217317 0.000741641
EPB41 -0.776838925 0.00075753 PODXL2 0.72096337 0.000742065
CDK16 -0.774923584 0.000793651 SUSD2 1.028182394 0.000764562
DPY19L3 -0.771141299 0.000885322 PTMA 0.798184795 0.000834757
IQGAP2 -0.76451484 0.001027929 PALMD 0.978323746 0.000915468
AP1S2 -0.758646702 0.001127111 RPS3A 0.557517248 0.000981153
PARP1 -0.756824002 0.001614942 RPL7A 0.761297411 0.001025342
CCDC136 -0.754772167 0.001638991 ANKRD53 0.791990766 0.001153614
PKP2 -0.752806479 0.001750592 RAMP1 1.317426962 0.001209221
MYO3B -0.751362885 0.00178297 NEIL2 0.714772631 0.001333622
ADCY1 -0.746286093 0.002121936 TTC8 0.88803947 0.00133705
UCKL1 -0.742583718 0.002316217 KLF6 0.832019903 0.001416302
SMARCA1 -0.739937737 0.002464079 RPL13A 0.625603871 0.001622393
TPI1 -0.738397012 0.00340866 CNTFR 0.600200762 0.00228622
MTMR6 -0.731857746 0.003450247 NOP53 0.553186416 0.002303112
NME1-NME2 -0.730675774 0.00358279 AKAP2 0.756165738 0.002496674
THBS4 -0.729066 0.003600774 WDR34 0.883166159 0.002785072
FTSJ1 -0.721718342 0.003751448 CENPV 0.757212975 0.002936103
HES6 -0.721557802 0.004240743 RPL10A 0.758669037 0.003110939
UBE2Q2 -0.720348131 0.004365536 LINGO3 0.618819489 0.003495891
AC091849.1 -0.716491498 0.004974724 QTRT1 0.75894465 0.003554292
PDE6H -0.713108239 0.005148264 SHANK2-AS2 0.691840832 0.003627283
PTP4A3 -0.704852061 0.005156031 MYCN 0.96736033 0.003634435

197

KCNG4 -0.701619472 0.0057669 PDZRN3 0.8597634 0.003798212
ARHGEF9 -0.691095677 0.006109404 MAOA 0.789659737 0.003870136
GNG3 -0.687557532 0.006271828 RASA4 0.555346591 0.004350969
AC009404.1 -0.67543836 0.006317913 FAM174B 0.664174712 0.004825488
HSF2 -0.661794237 0.006449933 TMSB10 0.692540634 0.004840785
RABAC1 -0.661208249 0.006461303 RPL13 0.503210264 0.005437556
LYPLA1 -0.660626028 0.007336669 RPS6 0.688035353 0.005718269
PLXDC1 -0.659284296 0.008778438 EIF3E 0.75039861 0.005787795
A1BG -0.65789708 0.008969474 POLR2L 0.725992842 0.006240183
CERS4 -0.650912694 0.009024544 FABP5 0.902707854 0.006728666
HK1 -0.649890226 0.009627377 RPS4Y1 1.182554912 0.0067362
MIR7-3HG -0.642413036 0.009848591 TMX4 0.604053393 0.006969003
HOPX -0.642249374 0.010213759 SVIL-AS1 0.788678509 0.0071443
CCDC28B -0.628608984 0.010846612 AC092902.2 0.856388362 0.007219274
SEC61G -0.623032827 0.011914292 METTL26 0.621361769 0.007624001
SNCB -0.62209535 0.016137072 RPL10P6 0.670470591 0.007868267
TSPAN13 -0.621779559 0.016334644 RPS2 0.568635202 0.008628962
PPP2R2B -0.619726384 0.016338137 HGNC:24955 0.645972941 0.009516911
DYM -0.612989098 0.016377744 TIPARP 0.678819714 0.010215875
PSIP1 -0.604322091 0.016420664 MFAP4 1.376275367 0.01054491
GLYATL1 -0.601839139 0.016711487 SLAIN1 0.92028955 0.013792643
HIGD1A -0.599825543 0.01733231 RPS27A 0.617302798 0.014266305
TMEM9 -0.599715322 0.017355571 RPL10 0.606608857 0.014962845
B2M -0.583168131 0.017591791 ITGA4 0.858972858 0.015579042
RGS9 -0.559388191 0.018422142 RPL10P9 1.031187629 0.015668401
BCO2 -0.548393122 0.020925931 ENAH 0.635839682 0.015796793
TNNI3 -0.538703034 0.021328777 CHRNA3 0.63749045 0.020315291
SPCS1 -0.530941708 0.021413494 LINC01451 0.518322242 0.021779168
ADD1 -0.528981118 0.02203435 AL138828.1 1.446000862 0.021970401
P4HA1 -0.524977904 0.02224471 PRDM1 0.790230805 0.022209393
NPM2 -0.521124712 0.024050092 HNRNPK 0.605304455 0.02519262
PFKL -0.517441584 0.025698191 SERPINE2 0.896680346 0.028195569
ENO1 -0.517309734 0.031883242 FAM89A 0.745782394 0.029441618
PGK1 -0.507949031 0.042848397 KCNQ2 0.754695269 0.03429971
SULT1A1 -0.504870198 0.043918803 HNRNPA2B1 0.607819865 0.035223309
CCNC -0.504015692 0.044221562 ATP1B2 0.535200345 0.036180881
B3GLCT 0.688262588 0.036430023
PSPHP1 0.662802813 0.038281353
INAFM1 0.608820737 0.045824633
NAV2 0.538411313 0.048196817

198

Table S3.2: Differentially Expressed Genes Between Large Cone Side Group and Cone 3
Excluding Small Side Group
p-value>0.05, log 2FC>|0.5|

Symbol log2FC P-Value Adj Symbol log2FC P-Value Adj
Cone-3 CRABP2 -2.53087299 8.63E-19 Cone-SG FTLP3 0.572117637 5.62E-16
CKB -1.548929491 4.96E-16 FTL 1.524876489 4.84E-15
MTCO3P12 -1.158560178 3.16E-15 RP11-798M19.6 1.711162579 1.25E-08
MTCO2P12 -0.871609671 9.69E-15 SRP54 1.309510077 2.00E-08
GSTP1 -1.297127838 3.73E-14 HBP1 1.352806713 2.33E-08
MTATP6P1 -2.036835656 6.39E-13 PGK1 1.033163037 4.59E-08
PLEKHB1 -1.120835725 9.68E-13 IGFBP2 1.628788444 1.54E-07
TUBA1B -1.601586806 1.21E-12 HMGB2 1.32330216 1.73E-07
PRDX1 -1.185394469 1.81E-12 BHLHE40 1.473394584 2.03E-07
VAX2 -1.138573381 1.40E-11 ALKBH5 0.950092706 2.23E-07
NME1 -1.282637603 1.42E-11 SEC31A 1.268429485 3.02E-07
MTCO1P12 -1.368985676 1.97E-11 MORF4L2 0.919435197 4.11E-07
RRAD -1.805637837 2.58E-11 SEC61G 1.076046655 4.20E-07
MAP1B -1.409493819 6.30E-11 GOLT1B 1.191795811 6.15E-07
FAM107A -1.171826796 1.07E-10 GNL3 0.772279748 1.07E-06
ROGDI -1.032958945 2.13E-10 MFSD2A 0.990592438 1.08E-06
UBB -0.813872977 2.26E-10 HLA-B 1.156865775 1.09E-06
STMN1 -1.091367146 2.44E-10 BNIP3 1.414580043 2.09E-06
H3F3AP6 -0.541530939 3.43E-10 RNMT 0.894632965 2.72E-06
COTL1 -1.294028688 3.71E-10 UFM1 1.074775059 3.65E-06
OLFM1 -1.273722136 6.92E-10 INSIG2 1.157644805 3.71E-06
H3F3AP4 -0.717922147 7.68E-10 SCD 1.343515597 4.04E-06
EGR1 -1.635838314 8.49E-10 PFKFB3 0.694673165 4.62E-06
MLLT11 -0.907121809 2.03E-09 PJA2 1.019162517 5.90E-06
GNGT2 -1.302590878 3.45E-09 RPL21 0.92073682 5.93E-06
RP11-386G11.10 -0.858925235 3.77E-09 RPS25 0.757134517 6.20E-06
NDUFA11 -1.053943222 5.03E-09 CHCHD7 1.123364194 7.34E-06
IMPDH1 -1.12320783 1.09E-08 ARF4 1.338345743 8.18E-06
ACTG1 -0.89584689 1.17E-08 RPL21P16 0.695147468 9.35E-06
MTND2P28 -1.01486613 1.63E-08 TRAM1 0.981550932 2.92E-05
H3F3A -0.755576956 2.06E-08 INSIG1 1.174061429 4.05E-05
TUBA1A -1.165617832 2.22E-08 RPL21P75 0.555009852 5.87E-05
ACTB -0.937465215 2.51E-08 ARF1 0.965502863 6.25E-05
GNB3 -0.910007996 4.93E-08 RBPJ 0.883912309 6.69E-05
TUBA1C -0.592579694 5.70E-08 SRPRA 1.541887589 7.77E-05
NDUFB2 -1.012691001 5.96E-08 ALDOA 0.674215914 8.68E-05
MTND4P12 -0.677589438 8.14E-08 RPS3A 0.703897604 8.68E-05

199

PDE6H -1.014487299 2.00E-07 KDELR2 1.105527074 0.000102951
ACOT7 -0.894667023 2.61E-07 CCNI 0.976394392 0.000143621
SLC38A3 -0.986731668 2.98E-07 HILPDA 1.324561479 0.000144003
CHRNA3 -1.160991423 3.41E-07 MALAT1 0.809770522 0.000176342
SLC38A5 -1.153155532 4.27E-07 PAM 0.714571497 0.000180739
SEPT4 -0.845558891 4.61E-07 ANKRD37 1.245932077 0.000182865
RAX2 -0.851900742 4.77E-07 PDK1 0.899581468 0.000188567
ATP5A1 -0.584883972 6.07E-07 AC090498.1 0.530072658 0.000247696
FOS -2.237612904 6.81E-07 ZFPL1 0.808803553 0.000263064
HES6 -0.969747968 7.13E-07 IDI1 1.363095776 0.000331607
LMOD1 -0.541175795 1.55E-06 SAP30 0.787239474 0.000482473
NANOS1 -1.13394145 1.72E-06 RPL34 0.786755934 0.000565947
CFL1 -0.563260768 1.80E-06 HNRNPC 0.566710114 0.00066101
CCDC28B -0.914227204 1.86E-06 SIKE1 0.890918311 0.000763006
MYL6 -0.61359129 2.25E-06 CTD-2287O16.1 0.506635371 0.000990413
THY1 -0.772787539 2.51E-06 ECE2 0.505271962 0.001149713
DYNLL1 -0.775301302 3.28E-06 SELENOM 0.672529434 0.00117667
UQCC2 -0.712112352 4.27E-06 NDRG1 0.560203221 0.001193583
MTND1P23 -0.676353387 4.49E-06 COPB1 0.934623532 0.001300601
TMEM14B -0.822733248 6.61E-06 ADM 0.666773322 0.00137494
SNAP25 -0.893535633 7.21E-06 RPL18AP3 0.594119756 0.001422168
MPPED2 -1.269721171 7.98E-06 AK4 0.682964457 0.001497799
BLOC1S2 -1.48971222 9.04E-06 RPL13A 0.588735087 0.001654307
EWSR1 -0.764610072 1.17E-05 STARD4 0.660096501 0.001751703
HIGD1A -0.813511831 1.72E-05 FDFT1 1.239685237 0.001934265
RXRG -0.771793734 2.50E-05 MRPL17 0.773991547 0.002061659
HMGB1P5 -0.514606961 3.35E-05 COPB2 0.887015229 0.002192648
CRMP1 -0.820900903 3.42E-05 FUT11 0.881314793 0.002405417
CHRNB4 -0.912228661 3.64E-05 TYMS 0.569646241 0.002917759
NDUFC2 -0.592016169 3.64E-05 FGF11 0.892219623 0.003885404
NREP -0.804847853 3.96E-05 SQLE 1.0168311 0.00393274
CLTB -0.899282886 4.50E-05 PGM3 0.757762012 0.004166461
NDUFB1 -0.745809684 4.77E-05 RPS27 0.8247256 0.004268907
YWHAB -0.722570805 4.95E-05 POLRMT 0.61298493 0.004967557
WDR1 -0.792423217 5.29E-05 MED10 1.164747425 0.006104111
ATP6V0E2 -0.718061414 6.58E-05 PFKFB4 0.775309756 0.006584574
MARCKSL1 -0.682962712 6.92E-05 GNL2 1.157801253 0.006646835
SELENOH -0.723195086 7.57E-05 EIF2S3 0.962649684 0.006970963
R3HCC1 -0.80255001 7.89E-05 HMGCS1 1.276026335 0.00877514
GAL3ST3 -0.672568363 9.21E-05 RPL18 0.567880324 0.008807684
UNC119 -0.843777398 0.000125158 EEF1D 0.659779113 0.009978002
HNRNPA2B1 -0.674268273 0.000159478 BNIP3L 0.726350877 0.00998032

200

MOK -0.746189727 0.000160387 TCP1 0.89955696 0.010341674
PRDX6 -0.808702204 0.000177551 CKS2 1.430264404 0.011133133
ABHD14A -0.820950417 0.000201853 GLTSCR2 0.533286294 0.012606957
ARR3 -1.597098625 0.0002078 GAS5 0.802073974 0.013690579
NDN -0.690739128 0.000219781 RPL18A 0.741267774 0.015261151
SAMD7 -0.998192052 0.000237693 RPS18 0.813035205 0.017558164
LAPTM4B -0.635343827 0.000346856 PRPF3 1.022366784 0.018178871
IFT22 -0.78224703 0.000372821 ANXA5 0.962963029 0.018497541
AK1 -0.505422593 0.00053212 CBFA2T2 0.878698693 0.023131203
PLXDC1 -0.738437658 0.000668303 GOLGA5 0.795094177 0.023404941
DPYSL5 -0.685115722 0.000703055 BTG1 0.502450391 0.024350044
PPP1CA -0.750899501 0.00087283 P4HB 0.730086253 0.02547079
BZW2 -0.790828535 0.000912519 RPLP0 0.579070061 0.025728864
HMGN3 -0.763968935 0.000916931 RNF24 0.552780538 0.025867759
HMGN2 -0.702109719 0.000953189 TM9SF2 1.13469776 0.031751848
RS1 -0.741659039 0.001004402 RPL7P9 0.608021358 0.033862122
DCTN1 -0.576079996 0.001238899 HK2 1.005584412 0.036701739
YWHAQ -0.589121344 0.00134354 NANS 0.855297257 0.041350393
PODXL2 -0.617173395 0.001432437 BLVRB 0.600101201 0.041741903
CKMT1B -0.659515212 0.001451744 RPS27L 0.52089377 0.044069725
BEX3 -0.605871792 0.001462067 RPL7P1 0.503216225 0.046757872
MTERF2 -0.545258082 0.001559532 BLZF1 0.787637674 0.047837669
NDUFV2 -0.677881647 0.001852132
POLR2L -0.562390331 0.00191651
TSPAN7 -0.740198178 0.002034349
STXBP1 -0.602443351 0.002452378
GUK1 -0.521653654 0.00269151
VPS28 -0.504694932 0.002846043
TMEM53 -0.593512461 0.003072841
JUNB -1.070171351 0.003087091
UCHL1 -0.592092612 0.003487552
CKMT1A -0.59804932 0.003647285
TCN2 -0.75790422 0.003739617
COX6C -0.556904562 0.006034496
ARL4D -0.739920897 0.006364454
PAFAH1B3 -0.616424843 0.006621825
TSPAN15 -0.696839814 0.006889071
SLC1A7 -0.609110187 0.007097487
TUBB3 -0.517209629 0.008123397
TSC22D1 -0.593177632 0.008651261
SVOP -0.513153275 0.008983063
PPP4R4 -0.964356532 0.009128746

201

LOXL1 -0.755669132 0.009420726
SYP -0.556120669 0.012552152
RBP4 -0.674542867 0.012863432
SIX6 -0.683655052 0.015644903
ATP5I -0.672934245 0.016954447
TPM1 -0.71522549 0.019096463
PAQR4 -0.547436646 0.01959684
NDUFA13 -0.688324222 0.020540052
SLC25A11 -0.611308314 0.023888227
HSPA8 -0.630339387 0.028043076
KCNV2 -0.637175648 0.029023892
MPP4 -0.896819813 0.029498326
PDE6D -0.7731433 0.032158098
HMGB1 -0.530089254 0.032168872
GADD45G -0.882474648 0.034349314
APOO -0.639763 0.036031973
TPD52 -0.632413341 0.039473194
POLR2G -0.715609529 0.041452066
MFSD11 -0.713560778 0.041510918
ATP5O -0.619543661 0.041528241
NDRG3 -0.647224346 0.043325187
CD200 -0.741683185 0.045991197
CISD1 -0.597892228 0.049258207
ETNK2 -0.699763352 0.049733274

202

Table S3.3: Differentially Expressed Genes Between Small Cone Side Group and All Other Cone 3
p-value>0.05

Symbol log2FC P-Value Adj
MT1M 1.433441764 1.12E-05
RWDD3 1.454274113 3.72E-05
ENPP2 3.772354139 4.46E-05
MT1F 2.697899765 0.000526803
SMIM24 0.80978278 0.000942219
CYP26B1 0.7806377 0.001453908
MT1G 4.160263837 0.001587387
MT1E 2.809127417 0.002003899
MT2A 3.173891193 0.002213517
NREP 1.15850056 0.005249399
RSPO2 0.680341396 0.008375886
PDE6H -1.974162333 0.008929818
RP11-193M21.1 0.409202094 0.011831242
MT1H 1.474271367 0.036817188

203

Discussion

Understanding the processes that define differentiation and maturation of cell types in the retina
is key to being able to understand how diseases alter those processes and how we can best
recreate them for retinal regeneration and in vitro modeling. A great deal of previous work in the
retina focused on the mechanisms of RPC differentiation into each cell type (Young et al. 1985,
Wetts and Fraser 1988, Bassett and Wallace 2012), and even though recent large scRNA-seq
experiments provide a wealth of information for both fetal and adult retina (Macosko et al. 2015,
Mellough et al. 2019, Lukowski et al. 2019, Hu et al. 2019, Sridhar et al. 2020, Cowan et al.
2020, Yan et al. 2020), as well as retinal organoids (Kaya et al. 2019, Kim et al. 2019, Kallman
et al. 2020, Sridhar et al. 2020, Cowan et al. 2020, Cui et al. 2020), analysis of each cell type
usually focuses on evaluating already-identified marker gene expression. Organoid scRNA-seq
datasets are typically compared against fetal through correlation of known markers between
ages, which has shown a great deal of similarity but also gene expression differences between
tissues. Our group’s interest in cones, stemming from their unique capability to form
retinoblastomas, required a greater understanding of their differentiation and cell state changes
during development.
The full-length fetal photoreceptor transcriptomes generated in this thesis have
provided a number of novel insights through analysis of the known and unknown photoreceptor
gene expression and transcription factor activity changes that occur in photoreceptor fate
specification and cone maturation. This scRNA-seq study were the first to describe differential
isoform use of canonical rod and cone genes NRL and RXRG between these cell types. In
particular, increased cone usage of the truncated NRL ENST00000560550, which lacks the
transactivation domain critical for (Friedman et al 2004) NRL transcription factor function, may
dimerize with full length NRL and inhibit its activity. Future experiments would aim to evaluate

204

the potential inhibitory functionality of truncated NRL in cone cells and how this affects final
protein expression.
While other publications have recently described transcriptomic transition states
between RPCs and photoreceptors (Collin et al. 2019, Sridhar et al. 2020), our data resolved
two potential trajectories to rods, one directly from an RPC/MG population to early maturing
rods, and the other from a shared immature photoreceptor precursor state which then bifurcated
into rods or cones (Chapter 2, Figure 2.17, 2.18). These observations are consistent with two
existing models of photoreceptor fate determination (Ng et al. 2011, Emerson et al. 2013),
where fate is determined in a post mitotic photoreceptor precursor state or in the RPC stage,
respectively. Our data also provided evidence that this shared photoreceptor precursor
population expressed elements of both models of cone fate specification. In the larger iPRP
cluster, a small post-mitotic population linked to rod and cone clusters showed bifurcating RNA
Velocity towards both photoreceptor types with shared expression of CHRNA1 and known
transition state gene ATOH7 (Sridhar et al. 2020). Cone directed cells leaving this shared
precursor population expressed ONECUT1, THRB, and showed OLIG2 regulon activity, which
are all elements described in RPCs fated to produce cones or horizontal cells in early retina
development (Emerson et al. 2013). Our data captures cones emerging from a post-mitotic
precursor population expressing genes previously tied to photoreceptor fate specification in the
RPC state. Photoreceptor fate could be defined both pre- and post-mitotically at different times
in development, however ONECUT1 and THRB signaling may be key in both cases to define
cones. These observations suggest that both fate specification methods exist, and that the
existence of a shared precursor state needs to be validated through in situ hybridization and
immunostaining for post-mitotic cells co-expressing shared state markers (such as CHRNA1)
with early cone markers (such as lncRNA CTC-378H22.2 or RXRγ protein) and with early rod
marker proteins (such as NRL or NR2E3). One potential explanation for the existence of fate

205

decision points both in RPCs and in common cone-rod precursors may relate to timing of cell
birth; at developmental timepoints where only cones or only rods are born, a direct fate
specification may be utilized, but when both types are being produced an alternate shared
photoreceptor precursor pathway could be used instead. Rods appeared evolutionarily as an
offshoot of cones, so this could be the remains of early processes converting between the two
cell types (Lamb, Collin, and Pugh Jr. 2007).
My data primarily captured three cell types, RPC/MG, rods, and cones, with clear
distinction of cycling RPCs, RPC/MG, a shared immature photoreceptor precursor population,
early and late maturing rods, and L/M and S-cones. A large number of early maturing L/M cones
were sequenced, but we were unable to capture more than a few representative cells of the
most mature cone cell states (OPN1LW+ macula cones). However, this small mature cone
population showed a spatial and gene expression separation from the remaining L/M cones,
with increased expression of features like TTR, MYL4 and GUCA1C. This separation suggests
these few cells represent a distinct state change during maturation. The bulk of our cones,
comprising an early maturing population, did not form spatially distinct groups in UMAP but were
patterned by lncRNA genes. Specifically, lncRNAs like CTC-378H22.2 HOTAIRM1 were
expressed in cells closer to the iPRP region, while others like RP13-143G15.4 upregulated
nearer to the late maturing group. Several lncRNAs have been tangentially observed in cone
populations previously (Welby et al. 2017). This kind of subtle lncRNA patterning of a visually
continuous population suggests a role in the post-mitotic development of cone cells which may
well extrapolate to other retinal and non-retinal cell types. Additional localization and functional
evaluation experiments are needed to better visualize the patterns of lncRNAs and determine
whether they influence the birth or development of cone cells or simply serve as maturation
signposts.

206

Lastly, we were also able to identify new elements comprising the cone sensitivity to
RB1 loss. Specifically, we found increased expression of MYC target genes and increased
MYCN expression in cones, which matches the nuclear protein expression present in more
peripheral cells (Figure 3.5A) (Xu et al. 2014), along with strongly expressed SYK which can
regulate MYC activity (Puissant et al. 2014). SYK protein was detected in all but the most
mature fetal cones, and SYK inhibition in cultured retina directly decreased the proliferative
response to pRB loss. Unlike previous work which only identified SYK in retinoblastoma, we
show that SYK expression is an intrinsic feature of human fetal cones, the cell of origin, and not
acquired after pRB loss.
In Chapter 3, I examined retinal organoid cone cell maturation through expression of
cone proliferation-program related proteins and scRNA-sequencing. These experiments
identified cone maturation differences between organoid production methods as where organoid
cone maturation was similar or diverged from that of fetal cones. The initial comparison of retinal
organoid to fetal cone maturation through the expression of RB1-loss proliferative program
proteins demonstrated the inherent heterogeneity of a cell type within a single organoid,
reflected in ARR3+ cones positively and negatively expressing pRB, MDM2 and MYCN in the
same tissue. But organoids also showcased general timing shifts, with MDM2 expression
delayed relative to cone arrestin expression, as well as a complete lack of foveal cone protein
expression patterns such as cytoplasmic p27 and cytoplasmic/nuclear MYCN. While these
protein expression patterns were similar to those in fetal cones, it was enough to suggest subtle
changes during maturation that could be further illuminated using scRNA-seq.
Through generation of a cone-enriched transcriptomic dataset similar to the fetal dataset
in Chapter 2, we were able to capture cone maturation differences between retinal organoid
production methods as well as more mature cone states that we were unable to isolate from
fetal retina. Differences in cone cell number and maturation rate between production methods
may be caused by several differences between methodologies; addition of defined factors such

207

as BMP4 to the early Kuwahara protocol encourages retinal cell fate from hESCs, which may
lead to earlier RPC differentiation than without them, while the 3D-2D-3D system used in the
Zhong protocol along with later addition of retinoic acid may similarly delay photoreceptor
maturation (O’Hara-Wright and Gonzalez-Cordero 2017). These late maturing cones had a
distinctive upregulation of both cone marker genes (MYL4, GUCA1C, GRK7) and transcription
factor activity SCENIC regulons for GSC2, MEF2C, ESRRG), suggesting a state change similar
to the small late maturing cone population detected in fetal retina scRNA-seq analyses. Early
iPRP and cone features that were first detected in fetal cone precursors were similarly shown in
organoids, indicating another area of consistency between tissues. Direct comparison between
fetal and organoid cones, however, revealed that the broadest differences are tissue-wide, not
only between cones, with notable differences in ribosomal and glycolytic as well as structural
genes in cones and rods alike. As others have shown, retinal organoids are similar to fetal retina
in many ways (Zhong et al. 2014, Kuwahara et al. 2015, Aparicio et al. 2017b, Kim et al. 2019,
Sridhar et al. 2020, O’Hara-Wright and Gonzalez-Cordero 2020) and our results show capture
of mature photoreceptor states is easier from organoids without more involved manual
dissection of retina. However, the process of differentiation and culture clearly imparts broader
metabolic differences on organoid photoreceptors, along with other individual gene expression
differences such as the structural genes COL16A1 and TUBA4A (Figure 3.24). Future work
would look closely at culture conditions that exacerbate glycolysis and metabolism in organoids,
while immunohistochemical evaluation of collagen and tubulin may provide insight into fetal and
organoid tissue structure and photoreceptor morphology.
Lastly, we investigated the features of aberrant cone cell populations identified in our
sequencing data. We demonstrated that the internalized photoreceptors observed this and other
works (Cowan et al. 2020) were the likely source of a hypoxic aberrant cone cluster in our data,
while the Zhong methodology specifically produced a small population expressing oxidative
stress-related metallothionine and stress-induced ENPP2 gene in the borders between neural

208

retina regions. The idea that organoids are hypoxic internally at a certain size has existed
previously, but no one has addressed this or its influence on cone cells in larger sequencing
datasets. However, the proteins BHLHE40 and ENPP2 used to mark these populations were
unexpectedly upregulated in foveal cone cells, indicating they have normal roles in cone
maturation as well and that post-transcriptional regulation can provide protein expression
patterns not predicted by scRNA-seq data.
Our work comparing of retinal organoid and fetal retina scRNA-seq profiles upholds
previous organoid work showing strong similarities between fetal and organoid tissues, as well
as that organoids may be better suited to isolate examples of mature, non-foveal cones that are
rarely obtained from fetal retina samples. Current retinal organoids may also be suitable to
interrogate early iPRP and cone genes, such as the lncRNA CTC-378H22.2, for their role in cell
fate determination and early cone maturation. However, future work would use our findings to
identify aberrant internalized or border region cones from sequencing datasets, as well as to
interrogate aspects of retinal organoid culture that increase the expression of glycolysis genes
in organoids. Overall, organoid cones show many similarities to fetal cones during development
which allow us to discover new elements of fetal development; however, we must be cautious of
the systemic differences in maturation and gene expression caused by organoid culture and
differences in production method, as well as aberrantly localized cell populations when
considering whole organoid transcriptomes.

209

References
Abramov, I., Gordon, J., Hendrickson, A., Hainline, L., Dobson, V., & Labossiere, E. (1982). The retina of
the newborn human infant. Science, 217(4556), 265–267. https://doi.org/10.1126/science.6178160
Adelmann, H. B. (1929). Experimental studies on the development of the eye. I. The effect of the removal
of median and lateral areas of the anterior end of the urodelan neural plate on the development of
the eyes (Triton teniatus and Amblystoma punctatum). Journal of Experimental Zoology, 54(2), 249–
290. https://doi.org/10.1002/jez.1400540204
Adelmann, H. B. (1929). Experimental studies on the development of the eye. II. The eye ‐forming
potencies of the median portions of the urodelan neural plate (Triton teniatus and Amblystoma
punctatum). Journal of Experimental Zoology, 54(2), 291–317.
https://doi.org/10.1002/jez.1400540205
Agarwal, S., Ghosh, R., Chen, Z., Lakoma, A., Gunaratne, P. H., Kim, E. S., & Shohet, J. M. (2016).
Transmembrane adaptor protein PAG1 is a novel tumor suppressor in neuroblastoma. Oncotarget,
7(17), 24018–24026. https://doi.org/10.18632/oncotarget.8116
Aibar, S., González-Blas, C. B., Moerman, T., Huynh-Thu, V. A., Imrichova, H., Hulselmans, G., Rambow,
F., Marine, J. C., Geurts, P., Aerts, J., Van Den Oord, J., Atak, Z. K., Wouters, J., & Aerts, S. (2017).
SCENIC: Single-cell regulatory network inference and clustering. Nature Methods, 14(11), 1083–
1086. https://doi.org/10.1038/nmeth.4463
Aldiri, I., Xu, B., Wang, L., Chen, X., Hiler, D., Griffiths, L., Valentine, M., Shirinifard, A., Thiagarajan, S.,
Sablauer, A., Barabas, M. E., Zhang, J., Johnson, D., Frase, S., Zhou, X., Easton, J., Zhang, J.,
Mardis, E. R., Wilson, R. K., … Dyer, M. A. (2017). The Dynamic Epigenetic Landscape of the
Retina During Development, Reprogramming, and Tumorigenesis. Neuron, 94(3), 550-568.e10.
https://doi.org/10.1016/j.neuron.2017.04.022
Álvarez-Barrios, A., Álvarez, L., García, M., Artime, E., Pereiro, R., & González-Iglesias, H. (2021).
Antioxidant defenses in the human eye: A focus on metallothioneins. Antioxidants, 10(1), 1–33.
https://doi.org/10.3390/antiox10010089
Anders, S., Reyes, A., & Huber, W. (2012). Detecting differential usage of exons from RNA-seq data.
Genome Research, 22(10), 2008–2017. https://doi.org/10.1101/gr.133744.111
Aparicio, J. G., Hopp, H., Choi, A., Mandayam Comar, J., Liao, V. C., Harutyunyan, N., & Lee, T. C.
(2017a). Temporal expression of CD184(CXCR4) and CD171(L1CAM) identifies distinct early
developmental stages of human retinal ganglion cells in embryonic stem cell derived retina.
Experimental Eye Research, 154(8), 177–189. https://doi.org/10.1016/j.exer.2016.11.013
Aparicio, J. G., Shayler, D. W. H., & Cobrinik, D. (2017b). Retinal organoids: An emerging technology for
retinal disease research and therapy. In S. D. Schwartz, A. Nagiel, & R. Lanza (Eds.), Cellular
Therapies for Retinal Disease: A Strategic Approach (pp. 117–138). Springer International
Publishing. https://doi.org/10.1007/978-3-319-49479-1_10

210

Arikawa, K., & Williams, D. S. (1993). Acetylated α-tubulin in the connecting cilium of developing rat
photoreceptors. Investigative Ophthalmology and Visual Science, 34(6), 2145–2149.
Baden, T., Euler, T., & Berens, P. (2020). Understanding the retinal basis of vision across species. Nature
Reviews Neuroscience, 21(1), 5–20. https://doi.org/10.1038/s41583-019-0242-1
Balasubramanian, R., Bui, A., Ding, Q., & Gan, L. (2014). Expression of LIM-homeodomain transcription
factors in the developing and mature mouse retina. Gene Expression Patterns, 14(1), 1–8.
https://doi.org/10.1016/j.gep.2013.12.001
Bassett, E. A., & Wallace, V. A. (2012). Cell fate determination in the vertebrate retina. Trends in
Neurosciences, 35(9), 565–573. https://doi.org/10.1016/j.tins.2012.05.004
Baylor, D. A., Lamb, T. D., & Yau, K. W. (1979). Responses of retinal rods to single photons. The Journal
of Physiology, 288(1), 613–634. https://doi.org/10.1113/jphysiol.1979.sp012716
Becht, E., McInnes, L., Healy, J., Dutertre, C. A., Kwok, I. W. H., Ng, L. G., Ginhoux, F., & Newell, E. W.
(2019). Dimensionality reduction for visualizing single-cell data using UMAP. Nature Biotechnology,
37(1), 38–47. https://doi.org/10.1038/nbt.4314
Bhaduri, A., Andrews, M. G., Mancia Leon, W., Jung, D., Shin, D., Allen, D., Jung, D., Schmunk, G.,
Haeussler, M., Salma, J., Pollen, A. A., Nowakowski, T. J., & Kriegstein, A. R. (2020). Cell stress in
cortical organoids impairs molecular subtype specification. Nature, 578(7793), 142–148.
https://doi.org/10.1038/s41586-020-1962-0
Blomgren, P., Chandrasekhar, J., Di Paolo, J. A., Fung, W., Geng, G., Ip, C., Jones, R., Kropf, J. E.,
Lansdon, E. B., Lee, S., Lo, J. R., Mitchell, S. A., Murray, B., Pohlmeyer, C., Schmitt, A., Suekawa-
Pirrone, K., Wise, S., Xiong, J. M., Xu, J., … Currie, K. S. (2020). Discovery of Lanraplenib (GS-
9876): A Once-Daily Spleen Tyrosine Kinase Inhibitor for Autoimmune Diseases. ACS Medicinal
Chemistry Letters, 11(4), 506–513. https://doi.org/10.1021/acsmedchemlett.9b00621
Browne, A. W., Arnesano, C., Harutyunyan, N., Khuu, T., Martinez, J. C., Pollack, H. A., Koos, D. S., Lee,
T. C., Fraser, S. E., Moats, R. A., Aparicio, J. G., & Cobrinik, D. (2017). Structural and functional
characterization of human stem-cell-derived retinal organoids by live imaging. Investigative
Ophthalmology and Visual Science, 58(9), 3311–3318. https://doi.org/10.1167/iovs.16-20796
Brzezinski, J. A., & Reh, T. A. (2015). Photoreceptor cell fate specification in vertebrates. Development
(Cambridge), 142(19), 3263–3273. https://doi.org/10.1242/dev.127043
Buenaventura, D. F., Corseri, A., & Emerson, M. M. (2019). Identification of genes with enriched
expression in early developing mouse cone photoreceptors. Investigative Ophthalmology and Visual
Science, 60(8), 2787–2799. https://doi.org/10.1167/iovs.19-26951
Buskin, A., Zhu, L., Chichagova, V., Basu, B., Mozaffari-Jovin, S., Dolan, D., Droop, A., Collin, J.,
Bronstein, R., Mehrotra, S., Farkas, M., Hilgen, G., White, K., Pan, K. T., Treumann, A., Hallam, D.,
Bialas, K., Chung, G., Mellough, C., … Lako, M. (2018). Disrupted alternative splicing for genes
implicated in splicing and ciliogenesis causes PRPF31 retinitis pigmentosa. Nature
Communications, 9(1). https://doi.org/10.1038/s41467-018-06448-y

211

Cao, J., Spielmann, M., Qiu, X., Huang, X., Ibrahim, D. M., Hill, A. J., Zhang, F., Mundlos, S.,
Christiansen, L., Steemers, F. J., Trapnell, C., & Shendure, J. (2019). The single-cell transcriptional
landscape of mammalian organogenesis. Nature, 566(7745), 496–502.
https://doi.org/10.1038/s41586-019-0969-x
Capowski, E. E., Samimi, K., Mayerl, S. J., Phillips, M. J., Pinilla, I., Howden, S. E., Saha, J., Jansen, A.
D., Edwards, K. L., Jager, L. D., Barlow, K., Valiauga, R., Erlichman, Z., Hagstrom, A., Sinha, D.,
Sluch, V. M., Chamling, X., Zack, D. J., Skala, M. C., & Gamm, D. M. (2019). Reproducibility and
staging of 3D human retinal organoids across multiple pluripotent stem cell lines. Development
(Cambridge), 146(1), dev.171686. https://doi.org/10.1242/dev.171686
Cepko, C. (2014). Intrinsically different retinal progenitor cells produce specific types of progeny. Nature
Reviews Neuroscience, 15(9), 615–627. https://doi.org/10.1038/nrn3767
Choi, S. M., Cho, H. J., Cho, H., Kim, K. H., Kim, J. B., & Park, H. (2008). Stra13/DEC1 and DEC2 inhibit
sterol regulatory element binding protein-1c in a hypoxia-inducible factor-dependent mechanism.
Nucleic Acids Research, 36(20), 6372–6385. https://doi.org/10.1093/nar/gkn620
Cholia, R. P., Dhiman, M., Kumar, R., & Mantha, A. K. (2018). Oxidative stress stimulates invasive
potential in rat C6 and human U-87 MG glioblastoma cells via activation and cross-talk between
PKM2, ENPP2 and APE1 enzymes. Metabolic Brain Disease, 33(4), 1307–1326.
https://doi.org/10.1007/s11011-018-0233-3
Chuang, J. H., Yarmishyn, A. A., Hwang, D. K., Hsu, C. C., Wang, M. L., Yang, Y. P., Chien, K. H., Chiou,
S. H., Peng, C. H., & Chen, S. J. (2018). Expression profiling of cell-intrinsic regulators in the
process of differentiation of human iPSCs into retinal lineages. Stem Cell Research and Therapy,
9(1), 1–17. https://doi.org/10.1186/s13287-018-0848-7
Clark, B. S., Stein-O’Brien, G. L., Shiau, F., Cannon, G. H., Davis-Marcisak, E., Sherman, T., Santiago, C.
P., Hoang, T. V., Rajaii, F., James-Esposito, R. E., Gronostajski, R. M., Fertig, E. J., Goff, L. A., &
Blackshaw, S. (2019). Single-Cell RNA-Seq Analysis of Retinal Development Identifies NFI Factors
as Regulating Mitotic Exit and Late-Born Cell Specification. Neuron, 102(6), 1111-1126.e5.
https://doi.org/10.1016/j.neuron.2019.04.010
Cowan, C. S., Renner, M., De Gennaro, M., Gross-Scherf, B., Goldblum, D., Hou, Y., Munz, M.,
Rodrigues, T. M., Krol, J., Szikra, T., Cuttat, R., Waldt, A., Papasaikas, P., Diggelmann, R., Patino-
Alvarez, C. P., Galliker, P., Spirig, S. E., Pavlinic, D., Gerber-Hollbach, N., … Roska, B. (2020). Cell
Types of the Human Retina and Its Organoids at Single-Cell Resolution. Cell, 182(6), 1623-
1640.e34. https://doi.org/10.1016/j.cell.2020.08.013
Dimaras, H., Corson, T. W., Cobrinik, D., White, A., Zhao, J., Munier, F. L., Abramson, D. H., Shields, C.
L., Chantada, G. L., Njuguna, F., & Gallie, B. L. (2015). Retinoblastoma. Nature Reviews Disease
Primers, 1(1), 15021. https://doi.org/10.1038/nrdp.2015.21
Dougherty, J. D., Garcia, A. D. R., Nakano, I., Livingstone, M., Norris, B., Polakiewicz, R., Wexler, E. M.,
Sofroniew, M. V., Kornblum, H. I., & Geschwind, D. H. (2005). PBK/TOPK, a proliferating neural
progenitor-specific mitogen-activated protein kinase kinase. Journal of Neuroscience, 25(46),
10773–10785. https://doi.org/10.1523/JNEUROSCI.3207-05.2005

212

Ea Hemstra, L., Schlosser, A., Lindholt, J. S., & Sorensen, G. L. (2018). Microfibrillar-associated protein 4
variation in symptomatic peripheral artery disease. Journal of Translational Medicine, 16(1), 1–7.
https://doi.org/10.1186/s12967-018-1523-6
Eiraku, M., & Sasai, Y. (2012). Self-formation of layered neural structures in three-dimensional culture of
ES cells. Current Opinion in Neurobiology, 22(5), 768–777.
https://doi.org/10.1016/j.conb.2012.02.005
Eiraku, M., Takata, N., Ishibashi, H., Kawada, M., Sakakura, E., Okuda, S., Sekiguchi, K., Adachi, T., &
Sasai, Y. (2011). Self-organizing optic-cup morphogenesis in three-dimensional culture. Nature,
472(7341), 51–58. https://doi.org/10.1038/nature09941
Eiraku, M., Watanabe, K., Matsuo-Takasaki, M., Kawada, M., Yonemura, S., Matsumura, M., Wataya, T.,
Nishiyama, A., Muguruma, K., & Sasai, Y. (2008). Self-Organized Formation of Polarized Cortical
Tissues from ESCs and Its Active Manipulation by Extrinsic Signals. Cell Stem Cell, 3(5), 519–532.
https://doi.org/10.1016/j.stem.2008.09.002
Eldred, K. C., Hadyniak, S. E., Hussey, K. A., Brenerman, B., Zhang, P. W., Chamling, X., Sluch, V. M.,
Welsbie, D. S., Hattar, S., Taylor, J., Wahlin, K., Zack, D. J., & Johnston, R. J. (2018). Thyroid
hormone signaling specifies cone subtypes in human retinal organoids. Science, 362(6411).
https://doi.org/10.1126/science.aau6348
Emerson, M. M., Surzenko, N., Goetz, J. J., Trimarchi, J., & Cepko, C. L. (2013). Otx2 and Onecut1
promote the fates of cone photoreceptors and horizontal cells and repress rod photoreceptors.
Developmental Cell, 26(1), 59–72. https://doi.org/10.1016/j.devcel.2013.06.005
Eser Ocak, P., Ocak, U., Sherchan, P., Zhang, J. H., & Tang, J. (2020). Insights into major facilitator
superfamily domain-containing protein-2a (Mfsd2a) in physiology and pathophysiology. What do we
know so far? Journal of Neuroscience Research, 98(1), 29–41. https://doi.org/10.1002/jnr.24327
Fadool, J. M., & Dowling, J. E. (2008). Zebrafish: a model system for the study of eye genetics. Progress
in Retinal and Eye Research, 27(1), 89–110. https://doi.org/10.1016/j.preteyeres.2007.08.002
Feng, L., Xie, Z. H., Ding, Q., Xie, X., Libby, R. T., & Gan, L. (2010). MATH5 controls the acquisition of
multiple retinal cell fates. Molecular Brain, 3(1), 1–16. https://doi.org/10.1186/1756-6606-3-36
Friedman, J. S., Khanna, H., Swain, P. K., DeNicola, R., Cheng, H., Mitton, K. P., Weber, C. H., Hicks, D.,
& Swaroop, A. (2004). The minimal transactivation domain of the basic motif-leucine zipper
transcription factor NRL interacts with TATA-binding protein. Journal of Biological Chemistry,
279(45), 47233–47241. https://doi.org/10.1074/jbc.M408298200
Fu, Z., Kern, T. S., Hellström, A., & Smith, L. E. H. (2021). Fatty acid oxidation and photoreceptor
metabolic needs. Journal of Lipid Research, 62(26), 100035.
https://doi.org/10.1194/JLR.TR120000618
Furukawa, T., Mukherjee, S., Bao, Z. Z., Morrow, E. M., & Cepko, C. L. (2000). rax, Hes1, and notch1
promote the formation of Muller glia by postnatal retinal progenitor cells. Neuron, 26(2), 383–394.
https://doi.org/10.1016/S0896-6273(00)81171-X

213

Geng, X., Speirs, C., Lagutin, O., Inbal, A., Liu, W., Solnica-Krezel, L., Jeong, Y., Epstein, D. J., & Oliver,
G. (2008). Haploinsufficiency of Six3 Fails to Activate Sonic hedgehog Expression in the Ventral
Forebrain and Causes Holoprosencephaly. Developmental Cell, 15(2), 236–247.
https://doi.org/10.1016/j.devcel.2008.07.003
Gonzalez-Cordero, A., Kruczek, K., Naeem, A., Fernando, M., Kloc, M., Ribeiro, J., Goh, D., Duran, Y.,
Blackford, S. J. I., Abelleira-Hervas, L., Sampson, R. D., Shum, I. O., Branch, M. J., Gardner, P. J.,
Sowden, J. C., Bainbridge, J. W. B., Smith, A. J., West, E. L., Pearson, R. A., & Ali, R. R. (2017).
Recapitulation of Human Retinal Development from Human Pluripotent Stem Cells Generates
Transplantable Populations of Cone Photoreceptors. Stem Cell Reports, 9(3), 820–837.
https://doi.org/10.1016/j.stemcr.2017.07.022
Grässel, S., & Bauer, R. J. (2013). Collagen XVI in health and disease. Matrix Biology, 32(2), 64–73.
https://doi.org/10.1016/j.matbio.2012.11.001
Graw, J. (2010). Eye development. Current Topics in Developmental Biology, 90(C), 343–386.
https://doi.org/10.1016/S0070-2153(10)90010-0
Gu, Y., Li, T., Kapoor, A., Major, P., & Tang, D. (2020). Contactin 1: An important and emerging
oncogenic protein promoting cancer progression and metastasis. Genes, 11(8), 1–22.
https://doi.org/10.3390/genes11080874
Habib, A. M., Okorokov, A. L., Hill, M. N., Bras, J. T., Lee, M. C., Li, S., Gossage, S. J., van Drimmelen,
M., Morena, M., Houlden, H., Ramirez, J. D., Bennett, D. L. H., Srivastava, D., & Cox, J. J. (2019).
Microdeletion in a FAAH pseudogene identified in a patient with high anandamide concentrations
and pain insensitivity. British Journal of Anaesthesia, 123(2), e249–e253.
https://doi.org/10.1016/j.bja.2019.02.019
Haenisch, C., Diekmann, H., Klinger, M., Gennarini, G., Kuwada, J. Y., & Stuermer, C. A. O. (2005). The
neuronal growth and regeneration associated Cntn1 (F3/F11/Contactin) gene is duplicated in fish:
Expression during development and retinal axon regeneration. Molecular and Cellular
Neuroscience, 28(2), 361–374. https://doi.org/10.1016/j.mcn.2004.04.013
Hafler, B. P., Surzenko, N., Beier, K. T., Punzo, C., Trimarchi, J. M., Kong, J. H., & Cepko, C. L. (2012).
Transcription factor Olig2 defines subpopulations of retinal progenitor cells biased toward specific
cell fates. Proceedings of the National Academy of Sciences of the United States of America,
109(20), 7882–7887. https://doi.org/10.1073/pnas.1203138109
Hahn, P., Dentchev, T., Qian, Y., Rouault, T., Harris, Z. L., & Dunaief, J. L. (2004). Immunolocalization
and regulation of iron handling proteins ferritin and ferroportin in the retina. Molecular Vision,
10(August), 598–607.
Hao, H., Tummala, P., Guzman, E., Mali, R. S., Gregorski, J., Swaroop, A., & Mitton, K. P. (2011). The
transcription factor Neural Retina Leucine Zipper (NRL) controls photoreceptor-specific expression
of myocyte enhancer factor Mef2c from an alternative promoter. Journal of Biological Chemistry,
286(40), 34893–34902. https://doi.org/10.1074/jbc.M111.271072

214

Hashimoto, K., Kodama, A., Honda, T., Hanashima, A., Ujihara, Y., Murayama, T., Nishimatsu, S. I., &
Mohri, S. (2017). Fam64a is a novel cell cycle promoter of hypoxic fetal cardiomyocytes in mice.
Scientific Reports, 7(1), 1–17. https://doi.org/10.1038/s41598-017-04823-1
Heavner, W., & Pevny, L. (2012). Eye development and retinogenesis. Cold Spring Harbor Perspectives
in Biology, 4(12). https://doi.org/10.1101/cshperspect.a008391
Hendrickson, A. (2016). Development of Retinal Layers in Prenatal Human Retina. American Journal of
Ophthalmology, 161, 29-35.e1. https://doi.org/10.1016/j.ajo.2015.09.023
Hendrickson, A. E., & Yuodelis, C. (1984). The Morphological Development of the Human Fovea.
Ophthalmology, 91(6), 603–612. https://doi.org/10.1016/S0161-6420(84)34247-6
Hendrickson, A., Bumsted-O’Brien, K., Natoli, R., Ramamurthy, V., Possin, D., & Provis, J. (2008). Rod
photoreceptor differentiation in fetal and infant human retina. Experimental Eye Research, 87(5),
415–426. https://doi.org/10.1016/j.exer.2008.07.016
Hendrickson, A., & Drucker, D. (1992). The development of parafoveal and mid-peripheral human retina.
Behavioural Brain Research, 49(1), 21–31. https://doi.org/10.1016/S0166-4328(05)80191-3
Hendrickson, A., Possin, D., Vajzovic, L., & Toth, C. A. (2012). Histologic Development of the Human
Fovea From Midgestation to Maturity. American Journal of Ophthalmology, 154(5), 767-778.e2.
https://doi.org/10.1016/j.ajo.2012.05.007
Hirami, Y., Osakada, F., Takahashi, K., Okita, K., Yamanaka, S., Ikeda, H., Yoshimura, N., & Takahashi,
M. (2009). Generation of retinal cells from mouse and human induced pluripotent stem cells.
Neuroscience Letters, 458(3), 126–131. https://doi.org/10.1016/j.neulet.2009.04.035
Hoshino, A., Ratnapriya, R., Brooks, M. J., Chaitankar, V., Wilken, M. S., Zhang, C., Starostik, M. R.,
Gieser, L., La Torre, A., Nishio, M., Bates, O., Walton, A., Bermingham-McDonogh, O., Glass, I. A.,
Wong, R. O. L., Swaroop, A., & Reh, T. A. (2017). Molecular Anatomy of the Developing Human
Retina. Developmental Cell, 43(6), 763-779.e4. https://doi.org/10.1016/j.devcel.2017.10.029
Hu, M., & Easter, J. (1999). Retinal neurogenesis: The formation of the initial central patch of postmitotic
cells. Developmental Biology, 207(2), 309–321. https://doi.org/10.1006/dbio.1998.9031
Huang, K. C., Wang, M. L., Chen, S. J., Kuo, J. C., Wang, W. J., Nhi Nguyen, P. N., Wahlin, K. J., Lu, J.
F., Tran, A. A., Shi, M., Chien, Y., Yarmishyn, A. A., Tsai, P. H., Yang, T. C., Jane, W. N., Chang, C.
C., Peng, C. H., Schlaeger, T. M., & Chiou, S. H. (2019). Morphological and Molecular Defects in
Human Three-Dimensional Retinal Organoid Model of X-Linked Juvenile Retinoschisis. Stem Cell
Reports, 13(5), 906–923. https://doi.org/10.1016/j.stemcr.2019.09.010
Hyatt, G. A., Schmitt, E. A., Fadool, J. M., & Dowling, J. E. (1996). Retinoic acid alters photoreceptor
development in vivo. Proceedings of the National Academy of Sciences of the United States of
America, 93(23), 13298–13303. https://doi.org/10.1073/pnas.93.23.13298

215

Ikeda, H., Osakada, F., Watanabe, K., Mizuseki, K., Haraguchi, T., Miyoshi, H., Kamiya, D., Honda, Y.,
Sasai, N., Yoshimura, N., Takahashi, M., & Sasai, Y. (2005). Generation of Rx+/Pax6+ neural retinal
precursors from embryonic stem cells. Proceedings of the National Academy of Sciences of the
United States of America, 102(32), 11331–11336. https://doi.org/10.1073/pnas.0500010102
Joly, S., Pernet, V., Samardzija, M., & Grimm, C. (2011). PAX6-positive müller glia cells express cell cycle
markers but do not proliferate after photoreceptor injury in the mouse retina. Glia, 59(7), 1033–1046.
https://doi.org/10.1002/glia.21174
Kallman, A., Capowski, E. E., Wang, J., Kaushik, A. M., Jansen, A. D., Edwards, K. L., Chen, L.,
Berlinicke, C. A., Joseph Phillips, M., Pierce, E. A., Qian, J., Wang, T. H., Gamm, D. M., & Zack, D.
J. (2020). Investigating cone photoreceptor development using patient-derived NRL null retinal
organoids. Communications Biology, 3(1), 82. https://doi.org/10.1038/s42003-020-0808-5
Kanai, Y., & Hediger, M. A. (2004). The glutamate/neutral amino acid transporter family SLC1: Molecular,
physiological and pharmacological aspects. Pflugers Archiv European Journal of Physiology, 447(5),
469–479. https://doi.org/10.1007/s00424-003-1146-4
Kaya, K. D., Chen, H. Y., Brooks, M. J., Kelley, R. A., Shimada, H., Nagashima, K., de Val, N., Drinnan,
C. T., Gieser, L., Kruczek, K., Erceg, S., Li, T., Lukovic, D., Adlakha, Y. K., Welby, E., & Swaroop, A.
(2019). Transcriptome-based molecular staging of human stem cell-derived retinal organoids
uncovers accelerated photoreceptor differentiation by 9-cis retinal. Molecular Vision, 25(November),
663–678. https://doi.org/10.1101/733071
Kelley, M. W., Turner, J. K., & Reh, T. A. (1994). Retinoic acid promotes differentiation of photoreceptors
in vitro. Development, 120(8), 2091–2102. https://doi.org/10.1242/dev.120.8.2091
Kennedy, B., & Malicki, J. (2009). What drives cell morphogenesis: A look inside the vertebrate
photoreceptor. Developmental Dynamics, 238(9), 2115–2138. https://doi.org/10.1002/dvdy.22010
Kim, D., Paggi, J. M., Park, C., Bennett, C., & Salzberg, S. L. (2019). Graph-based genome alignment
and genotyping with HISAT2 and HISAT-genotype. Nature Biotechnology, 37(8), 907–915.
https://doi.org/10.1038/s41587-019-0201-4
Kim, S., Lowe, A., Dharmat, R., Lee, S., Owen, L. A., Wang, J., Shakoor, A., Li, Y., Morgan, D. J., Hejazi,
A. A., Cvekl, A., DeAngelis, M. M., Jimmy Zhou, Z., Chen, R., & Liu, W. (2019). Generation,
transcriptome profiling, and functional validation of cone-rich human retinal organoids. Proceedings
of the National Academy of Sciences of the United States of America, 166(22), 10824–10833.
https://doi.org/10.1073/pnas.1901572116
Kolko, M., Wang, J., Zhan, C., Poulsen, K. A., Prause, J. U., Nissen, M. H., Heegaard, S., & Bazan, N. G.
(2007). Identification of intracellular phospholipases A2 in the human eye: Involvement in
phagocytosis of photoreceptor outer segments. Investigative Ophthalmology and Visual Science,
48(3), 1401–1409. https://doi.org/10.1167/iovs.06-0865
Komai, K., Okayama, R., Kitagawa, M., Yagi, H., Chihara, K., & Shiozawa, S. (2002). Alternative splicing
variants of the human DBL (MCF-2) proto-oncogene. Biochemical and Biophysical Research
Communications, 299(3), 455–458. https://doi.org/10.1016/S0006-291X(02)02645-1

216

Kuwahara, A., Ozone, C., Nakano, T., Saito, K., Eiraku, M., & Sasai, Y. (2015). Generation of a ciliary
margin-like stem cell niche from self-organizing human retinal tissue. Nature Communications, 6(1),
6286. https://doi.org/10.1038/ncomms7286
La Manno, G., Soldatov, R., Zeisel, A., Braun, E., Hochgerner, H., Petukhov, V., Lidschreiber, K., Kastriti,
M. E., Lönnerberg, P., Furlan, A., Fan, J., Borm, L. E., Liu, Z., van Bruggen, D., Guo, J., He, X.,
Barker, R., Sundström, E., Castelo-Branco, G., … Kharchenko, P. V. (2018). RNA velocity of single
cells. Nature, 560(7719), 494–498. https://doi.org/10.1038/s41586-018-0414-6
Lamb, T. D., Collin, S. P., & Pugh, E. N. (2007). Evolution of the vertebrate eye: opsins, photoreceptors,
retina and eye cup. Nature Reviews Neuroscience, 8(12), 960–976. https://doi.org/10.1038/nrn2283
Lamba, D. A., Karl, M. O., Ware, C. B., & Reh, T. A. (2006). Efficient generation of retinal progenitor cells
from human embryonic stem cells. Proceedings of the National Academy of Sciences of the United
States of America, 103(34), 12769–12774. https://doi.org/10.1073/pnas.0601990103
Lee, T. C., Almeida, D., Claros, N., Abramson, D. H., & Cobrinik, D. (2006). Cell cycle-specific and cell
type-specific expression of Rb in the developing human retina. Investigative Ophthalmology and
Visual Science, 47(12), 5590–5598. https://doi.org/10.1167/iovs.06-0063
Li, A., Zhu, X., Brown, B., & Craft, C. M. (2003). Gene Expression Networks Underlying Retinoic Acid–
Induced Differentiation of Human Retinoblastoma Cells. Investigative Opthalmology & Visual
Science, 44(3), 996. https://doi.org/10.1167/iovs.02-0434
Li, A., Zhu, X., & Craft, C. M. (2002). Retinoic acid upregulates cone arrestin expression in retinoblastoma
cells through a cis element in the distal promoter region. Investigative Ophthalmology and Visual
Science, 43(5), 1375–1383.
Liang, J., Zubovitz, J., Petrocelli, T., Kotchetkov, R., Connor, M. K., Han, K., Lee, J. hwa, Ciarallo, S.,
Catzavelos, C., Beniston, R., Franssen, E., & Slingerland, J. M. (2002). PKB/Akt phosphorylates
p27, impairs nuclear import of p27 and opposes p27-mediated G1 arrest. Nature Medicine, 8(10),
1153–1160. https://doi.org/10.1038/nm761
Liang, Q., Dharmat, R., Owen, L., Shakoor, A., Li, Y., Kim, S., Vitale, A., Kim, I., Morgan, D., Liang, S.,
Wu, N., Chen, K., DeAngelis, M. M., & Chen, R. (2019). Single-nuclei RNA-seq on human retinal
tissue provides improved transcriptome profiling. Nature Communications, 10(1), 1–12.
https://doi.org/10.1038/s41467-019-12917-9
Liao, Y., Wang, J., Jaehnig, E. J., Shi, Z., & Zhang, B. (2019). WebGestalt 2019: gene set analysis toolkit
with revamped UIs and APIs. Nucleic Acids Research, 47(W1), W199–W205.
https://doi.org/10.1093/nar/gkz401
Liu, H., Zhang, Y., Zhang, Y.-Y., Li, Y.-P., Hua, Z.-Q., Zhang, C.-J., Wu, K.-C., Yu, F., Zhang, Y., Su, J., &
Jin, Z.-B. (2020). Human embryonic stem cell-derived organoid retinoblastoma reveals a cancerous
origin. Proceedings of the National Academy of Sciences, 117(52), 33628–33638.
https://doi.org/10.1073/pnas.2011780117

217

Liu, J., McGlinn, A. M., Fernandes, A., Milam, A. H., Strang, C. E., Andison, M. E., Lindstrom, J. M.,
Keyser, K. T., & Stone, R. A. (2009). Nicotinic acetylcholine receptor subunits in rhesus monkey
retina. Investigative Ophthalmology and Visual Science, 50(3), 1408–1415.
https://doi.org/10.1167/iovs.08-2398
Liu, J., Gao, L., Zhan, N., Xu, P., Yang, J., Yuan, F., Xu, Y., Cai, Q., Geng, R., & Chen, Q. (2020).
Hypoxia induced ferritin light chain (FTL) promoted epithelia mesenchymal transition and
chemoresistance of glioma. Journal of Experimental and Clinical Cancer Research, 39(1), 1–17.
https://doi.org/10.1186/s13046-020-01641-8
Lu, Y., Shiau, F., Yi, W., Lu, S., Wu, Q., Pearson, J. D., Kallman, A., Zhong, S., Hoang, T., Zuo, Z., Zhao,
F., Zhang, M., Tsai, N., Zhuo, Y., He, S., Zhang, J., Stein-O’Brien, G. L., Sherman, T. D., Duan, X.,
… Clark, B. S. (2020). Single-Cell Analysis of Human Retina Identifies Evolutionarily Conserved and
Species-Specific Mechanisms Controlling Development. Developmental Cell, 53(4), 473-491.e9.
https://doi.org/10.1016/j.devcel.2020.04.009
Lukowski, S. W., Lo, C. Y., Sharov, A. A., Nguyen, Q., Fang, L., Hung, S. S., Zhu, L., Zhang, T., Grünert,
U., Nguyen, T., Senabouth, A., Jabbari, J. S., Welby, E., Sowden, J. C., Waugh, H. S., Mackey, A.,
Pollock, G., Lamb, T. D., Wang, P., … Wong, R. C. (2019). A single ‐cell transcriptome atlas of the
adult human retina. The EMBO Journal, 38(18), 1–15. https://doi.org/10.15252/embj.2018100811
Lukowski, S., Lo, C., Sharov, A., Nguyen, Q., Fang, L., Hung, S., Zhu, L., Zhang, T., Nguyen, T.,
Senabouth, A., Jabbari, J., Welby, E., Sowden, J., Waugh, H., Mackey, A., Pollock, G., Lamb, T.,
Wang, P.-Y., Hewitt, A., … Wong, R. (2018). Generation of human neural retina transcriptome atlas
by single cell RNA sequencing. BioRxiv, 114(10), 425223. https://doi.org/10.1101/425223
Macosko, E. Z., Basu, A., Satija, R., Nemesh, J., Shekhar, K., Goldman, M., Tirosh, I., Bialas, A. R.,
Kamitaki, N., Martersteck, E. M., Trombetta, J. J., Weitz, D. A., Sanes, J. R., Shalek, A. K., Regev,
A., & McCarroll, S. A. (2015). Highly parallel genome-wide expression profiling of individual cells
using nanoliter droplets. Cell, 161(5), 1202–1214. https://doi.org/10.1016/j.cell.2015.05.002
Malicki, J. (2004). Cell fate decisions and patterning in the vertebrate retina: the importance of timing,
asymmetry, polarity and waves. Current Opinion in Neurobiology, 14(1), 15–21.
https://doi.org/10.1016/j.conb.2004.01.015
Martin, M. (2011). Cutadapt removes adapter sequences from high-throughput sequencing reads.
EMBnet.Journal, 17(1), 10. https://doi.org/10.14806/ej.17.1.200
Masland, R. H. (2001). The fundamental plan of the retina. Nature Neuroscience, 4(9), 877–886.
https://doi.org/10.1038/nn0901-877
Matson, D. R., & Todd Stukenberg, P. (2012). Cdt1 throws kinetochore-microtubule attachments for a
loop. Nature Cell Biology, 14(6), 561–563. https://doi.org/10.1038/ncb2513
Menon, M., Mohammadi, S., Davila-Velderrain, J., Goods, B. A., Cadwell, T. D., Xing, Y., Stemmer-
Rachamimov, A., Shalek, A. K., Love, J. C., Kellis, M., & Hafler, B. P. (2019). Single-cell
transcriptomic atlas of the human retina identifies cell types associated with age-related macular
degeneration. Nature Communications, 10(1). https://doi.org/10.1038/s41467-019-12780-8

218

Merbs, S. L., & Nathans, J. (1992). Absorption spectra of human cone pigments. Nature, 356(6368), 433–
435. https://doi.org/10.1038/356433a0
Meyer, J. S., Howden, S. E., Wallace, K. A., Verhoeven, A. D., Wright, L. S., Capowski, E. E., Pinilla, I.,
Martin, J. M., Tian, S., Stewart, R., Pattnaik, B., Thomson, J. A., & Gamm, D. M. (2011). Optic
vesicle-like structures derived from human pluripotent stem cells facilitate a customized approach to
retinal disease treatment. Stem Cells, 29(8), 1206–1218. https://doi.org/10.1002/stem.674
Meyer, J. S., Shearer, R. L., Capowski, E. E., Wright, L. S., Wallace, K. A., McMillan, E. L., Zhang, S. C.,
& Gamm, D. M. (2009). Modeling early retinal development with human embryonic and induced
pluripotent stem cells. Proceedings of the National Academy of Sciences of the United States of
America, 106(39), 16698–16703. https://doi.org/10.1073/pnas.0905245106
Miesfeld, J. B., & Brown, N. L. (2019). Eye organogenesis: A hierarchical view of ocular development. In
Current Topics in Developmental Biology (1st ed., Vol. 132). Elsevier Inc.
https://doi.org/10.1016/bs.ctdb.2018.12.008
Moore, K. B., Logan, M. A., Aldiri, I., Roberts, J. M., Steele, M., & Vetter, M. L. (2018). C8orf46 homolog
encodes a novel protein Vexin that is required for neurogenesis in Xenopus laevis. Developmental
Biology, 437(1), 27–40. https://doi.org/10.1016/j.ydbio.2018.03.003
Müller, F., & O’Rahilly, R. (1985). The first appearance of the neural tube and optic primordium in the
human embryo at stage 10. Anatomy and Embryology, 172(2), 157–169.
https://doi.org/10.1007/BF00319598
Mustafi, D., Engel, A. H., & Palczewski, K. (2009). Structure of cone photoreceptors. Progress in Retinal
and Eye Research, 28(4), 289–302. https://doi.org/10.1016/j.preteyeres.2009.05.003
Nakano, T., Ando, S., Takata, N., Kawada, M., Muguruma, K., Sekiguchi, K., Saito, K., Yonemura, S.,
Eiraku, M., & Sasai, Y. (2012). Self-formation of optic cups and storable stratified neural retina from
human ESCs. Cell Stem Cell, 10(6), 771–785. https://doi.org/10.1016/j.stem.2012.05.009
Ng, L., Lu, A., Swaroop, A., Sharlin, D. S., Swaroop, A., & Forrest, D. (2011). Two transcription factors
can direct three photoreceptor outcomes from rod precursor cells in mouse retinal development.
Journal of Neuroscience, 31(31), 11118–11125. https://doi.org/10.1523/JNEUROSCI.1709-11.2011
Nguyen-Ba-Charvet, K. T., & Chédotal, A. (2014). Development of retinal layers. Comptes Rendus -
Biologies, 337(3), 153–159. https://doi.org/10.1016/j.crvi.2013.11.010
O’Hara-Wright, M., & Gonzalez-Cordero, A. (2020). Retinal organoids: A window into human retinal
development. Development (Cambridge), 147(24), dev189746. https://doi.org/10.1242/dev.189746
Orjuela, S., Huang, R., Hembach, K. M., Robinson, M. D., & Soneson, C. (2019). ARMOR: An automated
reproducible modular workflow for preprocessing and differential analysis of RNA-seq data. G3:
Genes, Genomes, Genetics, 9(7), 2089–2096. https://doi.org/10.1534/g3.119.400185

219

Osakada, F., Ikeda, H., Mandai, M., Wataya, T., Watanabe, K., Yoshimura, N., Akaike, A., Sasai, Y., &
Takahashi, M. (2008). Toward the generation of rod and cone photoreceptors from mouse, monkey
and human embryonic stem cells. Nature Biotechnology, 26(2), 215–224.
https://doi.org/10.1038/nbt1384
Peng, Y. R., Shekhar, K., Yan, W., Herrmann, D., Sappington, A., Bryman, G. S., van Zyl, T., Do, M. T.
H., Regev, A., & Sanes, J. R. (2019). Molecular Classification and Comparative Taxonomics of
Foveal and Peripheral Cells in Primate Retina. Cell, 176(5), 1222-1237.e22.
https://doi.org/10.1016/j.cell.2019.01.004
Pertea, M., Kim, D., Pertea, G. M., Leek, J. T., & Salzberg, S. L. (2016). Transcript-level expression
analysis of RNA-seq experiments with HISAT, StringTie and Ballgown. Nature Protocols, 11(9),
1650–1667. https://doi.org/10.1038/nprot.2016.095
Pertea, M., Pertea, G. M., Antonescu, C. M., Chang, T. C., Mendell, J. T., & Salzberg, S. L. (2015).
StringTie enables improved reconstruction of a transcriptome from RNA-seq reads. Nature
Biotechnology, 33(3), 290–295. https://doi.org/10.1038/nbt.3122
Phillips, M. J., Jiang, P., Howden, S., Barney, P., Min, J., York, N. W., Chu, L. F., Capowski, E. E., Cash,
A., Jain, S., Barlow, K., Tabassum, T., Stewart, R., Pattnaik, B. R., Thomson, J. A., & Gamm, D. M.
(2018). A Novel Approach to Single Cell RNA-Sequence Analysis Facilitates In Silico Gene
Reporting of Human Pluripotent Stem Cell-Derived Retinal Cell Types. Stem Cells, 36(3), 313–324.
https://doi.org/10.1002/stem.2755
Phillips, M. J., Wallace, K. A., Dickerson, S. J., Miller, M. J., Verhoeven, A. D., Martin, J. M., Wright, L. S.,
Shen, W., Capowski, E. E., Percin, E. F., Perez, E. T., Zhong, X., Canto-Soler, M. V., & Gamm, D.
M. (2012). Blood-derived human iPS cells generate optic vesicle-like structures with the capacity to
form retinal laminae and develop synapses. Investigative Ophthalmology & Visual Science, 53(4),
2007–2019. https://doi.org/10.1167/iovs.11-9313
Poché, R. A., Furuta, Y., Chaboissier, M. C., Schedl, A., & Behringer, R. R. (2008). Sox9 is expressed in
mouse multipotent retinal progenitor cells and functions in Müller Glial cell development. Journal of
Comparative Neurology, 510(3), 237–250. https://doi.org/10.1002/cne.21746
Provis, J. M., Van Driel, D., Billson, F. A., & Russell, P. (1985). Development of the human retina:
Patterns of cell distribution and redistribution in the ganglion cell layer. Journal of Comparative
Neurology, 233(4), 429–451. https://doi.org/10.1002/cne.902330403
Puissant, A., Fenouille, N., Alexe, G., Pikman, Y., Bassil, C. F., Mehta, S., Du, J., Kazi, J. U., Luciano, F.,
Rönnstrand, L., Kung, A. L., Aster, J. C., Galinsky, I., Stone, R. M., DeAngelo, D. J., Hemann, M. T.,
& Stegmaier, K. (2014). SYK Is a Critical Regulator of FLT3 in Acute Myeloid Leukemia. Cancer
Cell, 25(2), 226–242. https://doi.org/10.1016/j.ccr.2014.01.022
Qu, C., Zheng, D., Li, S., Liu, Y., Lidofsky, A., Holmes, J. A., Chen, J., He, L., Wei, L., Liao, Y., Yuan, H.,
Jin, Q., Lin, Z., Hu, Q., Jiang, Y., Tu, M., Chen, X., Li, W., Lin, W., … Hong, J. (2018). Tyrosine
kinase SYK is a potential therapeutic target for liver fibrosis. Hepatology, 68(3), 1125–1139.
https://doi.org/10.1002/hep.29881

220

Quinn, P. M., Buck, T. M., Mulder, A. A., Ohonin, C., Alves, C. H., Vos, R. M., Bialecka, M., van
Herwaarden, T., van Dijk, E. H. C., Talib, M., Freund, C., Mikkers, H. M. M., Hoeben, R. C.,
Goumans, M.-J., Boon, C. J. F., Koster, A. J., Chuva de Sousa Lopes, S. M., Jost, C. R., &
Wijnholds, J. (2019). Human iPSC-Derived Retinas Recapitulate the Fetal CRB1 CRB2 Complex
Formation and Demonstrate that Photoreceptors and Müller Glia Are Targets of AAV5. Stem Cell
Reports, 12(5), 906–919. https://doi.org/10.1016/j.stemcr.2019.03.002
Quinn, P. M. J., & Wijnholds, J. (2019). Retinogenesis of the human fetal retina: An apical polarity
perspective. Genes, 10(12), 987. https://doi.org/10.3390/genes10120987
Rege, T. A., & Hagood, J. S. (2006). Thy ‐1 as a regulator of cell ‐cell and cell ‐matrix interactions in axon
regeneration, apoptosis, adhesion, migration, cancer, and fibrosis. The FASEB Journal, 20(8),
1045–1054. https://doi.org/10.1096/fj.05-5460rev
Rehemtulla, A., Warwar, R., Kumar, R., Ji, X., Zack, D. J., & Swaroop, A. (1996). The basic motif-leucine
zipper transcription factor Nrl can positively regulate rhodopsin gene expression. Proceedings of the
National Academy of Sciences of the United States of America, 93(1), 191–195.
https://doi.org/10.1073/pnas.93.1.191
Roepman, R., & Wolfrum, U. (2007). Protein networks and complexes in photoreceptor cilia. In
Subcellular Biochemistry (Vol. 43, pp. 209–235). Springer Netherlands. https://doi.org/10.1007/978-
1-4020-5943-8_10
Rushlow, D. E., Mol, B. M., Kennett, J. Y., Yee, S., Pajovic, S., Thériault, B. L., Prigoda-Lee, N. L.,
Spencer, C., Dimaras, H., Corson, T. W., Pang, R., Massey, C., Godbout, R., Jiang, Z.,
Zacksenhaus, E., Paton, K., Moll, A. C., Houdayer, C., Raizis, A., … Gallie, B. L. (2013).
Characterisation of retinoblastomas without RB1 mutations: Genomic, gene expression, and clinical
studies. The Lancet Oncology, 14(4), 327–334. https://doi.org/10.1016/S1470-2045(13)70045-7
Samuel, W., Kutty, R. K., Nagineni, S., Gordon, J. S., Prouty, S. M., Chandraratna, R. A. S., & Wiggert, B.
(2001). Regulation of Stearoyl Coenzyme A Desaturase Expression in Human Retinal Pigment
Epithelial Cells by Retinoic Acid. Journal of Biological Chemistry, 276(31), 28744–28750.
https://doi.org/10.1074/jbc.M103587200
Sapkota, D., Chintala, H., Wu, F., Fliesler, S. J., Hu, Z., & Mu, X. (2014). Onecut1 and Onecut2
redundantly regulate early retinal cell fates during development. Proceedings of the National
Academy of Sciences of the United States of America, 111(39), E4086–E4095.
https://doi.org/10.1073/pnas.1405354111
Sato, T., Vries, R. G., Snippert, H. J., Van De Wetering, M., Barker, N., Stange, D. E., Van Es, J. H., Abo,
A., Kujala, P., Peters, P. J., & Clevers, H. (2009). Single Lgr5 stem cells build crypt-villus structures
in vitro without a mesenchymal niche. Nature, 459(7244), 262–265.
https://doi.org/10.1038/nature07935
Seabrook, T. A., Burbridge, T. J., Crair, M. C., & Huberman, A. D. (2017). Architecture, Function, and
Assembly of the Mouse Visual System. Annual Review of Neuroscience, 40, 499–538.
https://doi.org/10.1146/annurev-neuro-071714-033842

221

Sehgal, R., Andres, D. J., Adler, R., & Belecky-Adams, T. L. (2006). Bone morphogenetic protein 7
increases chick photoreceptor outer segment initiation. Investigative Ophthalmology and Visual
Science, 47(8), 3625–3634. https://doi.org/10.1167/iovs.06-0079
Singh, H. P., Wang, S., Stachelek, K., Lee, S., Reid, M. W., Thornton, M. E., Craft, C. M., Grubbs, B. H.,
& Cobrinik, D. (2018). Developmental stage-specific proliferation and retinoblastoma genesis in RB-
deficient human but not mouse cone precursors. Proceedings of the National Academy of Sciences,
115(40), E9391–E9400. https://doi.org/10.1073/pnas.1808903115
Soneson, C., Love, M. I., & Robinson, M. D. (2015). Differential analyses for RNA-seq: transcript-level
estimates improve gene-level inferences. F1000Research, 4(2), 1521.
https://doi.org/10.12688/f1000research.7563.1
Sridhar, A., Hoshino, A., Finkbeiner, C. R., Chitsazan, A., Dai, L., Haugan, A. K., Eschenbacher, K. M.,
Jackson, D. L., Trapnell, C., Bermingham-McDonogh, O., Glass, I., & Reh, T. A. (2020). Single-Cell
Transcriptomic Comparison of Human Fetal Retina, hPSC-Derived Retinal Organoids, and Long-
Term Retinal Cultures. Cell Reports, 30(5), 1644-1659.e4.
https://doi.org/10.1016/j.celrep.2020.01.007
Steinberg, R. H., Fisher, S. K., & Anderson, D. H. (1981). Disc morphogenesis in vertebrate
photoreceptors. Vision Research, 21(11), 1725. https://doi.org/10.1016/0042-6989(81)90067-5
Steinhoff, J. S., Lass, A., & Schupp, M. (2021). Biological Functions of RBP4 and Its Relevance for
Human Diseases. Frontiers in Physiology, 12(March), 1–15.
https://doi.org/10.3389/fphys.2021.659977
Stenkamp, D. L. (2015). Development of the Vertebrate Eye and Retina. In Progress in Molecular Biology
and Translational Science (1st ed., Vol. 134). Elsevier Inc.
https://doi.org/10.1016/bs.pmbts.2015.06.006
Strauss, O. (2005). The retinal pigment epithelium in visual function. Physiological Reviews, 85(3), 845–
881. https://doi.org/10.1152/physrev.00021.2004
Stuart, T., Butler, A., Hoffman, P., Hafemeister, C., Papalexi, E., Mauck, W. M., Hao, Y., Stoeckius, M.,
Smibert, P., & Satija, R. (2019). Comprehensive Integration of Single-Cell Data. Cell, 177(7), 1888-
1902.e21. https://doi.org/10.1016/j.cell.2019.05.031
Surgucheva, I., Weisman, A. D., Goldberg, J. L., Shnyra, A., & Surguchov, A. (2008). γ-Synuclein as a
marker of retinal ganglion cells. Molecular Vision, 14(April), 1540–1548.
Takahashi, K., Tanabe, K., Ohnuki, M., Narita, M., Ichisaka, T., Tomoda, K., & Yamanaka, S. (2007).
Induction of Pluripotent Stem Cells from Adult Human Fibroblasts by Defined Factors. Cell, 131(5),
861–872. https://doi.org/10.1016/j.cell.2007.11.019
Thomson, J. A. (1998). Embryonic stem cell lines derived from human blastocysts. Science, 282(5391),
1145–1147. https://doi.org/10.1126/science.282.5391.1145

222

Thoreson, W. B. (2017). The Vertebrate Retina. In T. Ikezu & H. E. Gendelman (Eds.), Neuroimmune
Pharmacology (pp. 55–68). Springer International Publishing. https://doi.org/10.1007/978-3-319-
44022-4_5
Trapnell, C., Cacchiarelli, D., Grimsby, J., Pokharel, P., Li, S., Morse, M., Lennon, N. J., Livak, K. J.,
Mikkelsen, T. S., & Rinn, J. L. (2014). The dynamics and regulators of cell fate decisions are
revealed by pseudotemporal ordering of single cells. Nature Biotechnology, 32(4), 381–386.
https://doi.org/10.1038/nbt.2859
Turner, D. L., & Cepko, C. L. (1987). A common progenitor for neurons and glia persists in rat retina late
in development. Nature, 328(6126), 131–136. https://doi.org/10.1038/328131a0
Turner, D. L., Snyder, E. Y., & Cepko, C. L. (1990). Lineage-independent determination of cell type in the
embryonic mouse retina. Neuron, 4(6), 833–845. https://doi.org/10.1016/0896-6273(90)90136-4
Van de Sande, B., Flerin, C., Davie, K., De Waegeneer, M., Hulselmans, G., Aibar, S., Seurinck, R.,
Saelens, W., Cannoodt, R., Rouchon, Q., Verbeiren, T., De Maeyer, D., Reumers, J., Saeys, Y., &
Aerts, S. (2020). A scalable SCENIC workflow for single-cell gene regulatory network analysis.
Nature Protocols, 15(7), 2247–2276. https://doi.org/10.1038/s41596-020-0336-2
Varga, Z. M., Wegner, J., & Westerfield, M. (1999). Anterior movement of ventral diencephalic precursors
separates the primordial eye field in the neural plate and requires cyclops. Development, 126(24),
5533–5546. https://doi.org/10.1242/dev.126.24.5533
Verdijk, R. M., & Herwig-Carl, M. C. (2020). Development of the Human Eye. In Fetal and Neonatal Eye
Pathology (pp. 1–16). Springer International Publishing. https://doi.org/10.1007/978-3-030-36079-
5_1
Voigt, A. P., Whitmore, S. S., Flamme-Wiese, M. J., Riker, M. J., Wiley, L. A., Tucker, B. A., Stone, E. M.,
Mullins, R. F., & Scheetz, T. E. (2019). Molecular characterization of foveal versus peripheral human
retina by single-cell RNA sequencing. Experimental Eye Research, 184(April), 234–242.
https://doi.org/10.1016/j.exer.2019.05.001
Vu, H. T. V., Keeffe, J. E., McCarty, C. A., & Taylor, H. R. (2005). Impact of unilateral and bilateral vision
loss on quality of life. British Journal of Ophthalmology, 89(3), 360–363.
https://doi.org/10.1136/bjo.2004.047498
Walcott, J. C., & Provis, J. M. (2003). Müller cells express the neuronal progenitor cell marker nestin in
both differentiated and undifferentiated human foetal retina. Clinical and Experimental
Ophthalmology, 31(3), 246–249. https://doi.org/10.1046/j.1442-9071.2003.00638.x
Watanabe, K., Kamiya, D., Nishiyama, A., Katayama, T., Nozaki, S., Kawasaki, H., Watanabe, Y.,
Mizuseki, K., & Sasai, Y. (2005). Directed differentiation of telencephalic precursors from embryonic
stem cells. Nature Neuroscience, 8(3), 288–296. https://doi.org/10.1038/nn1402

223

Welby, E., Lakowski, J., Di Foggia, V., Budinger, D., Gonzalez-Cordero, A., Lun, A. T. L., Epstein, M.,
Patel, A., Cuevas, E., Kruczek, K., Naeem, A., Minneci, F., Hubank, M., Jones, D. T., Marioni, J. C.,
Ali, R. R., & Sowden, J. C. (2017). Isolation and Comparative Transcriptome Analysis of Human
Fetal and iPSC-Derived Cone Photoreceptor Cells. Stem Cell Reports, 9(6), 1898–1915.
https://doi.org/10.1016/j.stemcr.2017.10.018
Westendorp, B., Mokry, M., Groot Koerkamp, M. J. A., Holstege, F. C. P., Cuppen, E., & De Bruin, A.
(2012). E2F7 represses a network of oscillating cell cycle genes to control S-phase progression.
Nucleic Acids Research, 40(8), 3511–3523. https://doi.org/10.1093/nar/gkr1203
Wetts, R., & Fraser, S. E. (1988). Multipotent precursors can give rise to all major cell types of the frog
retina. Science, 239(4844), 1142–1145. https://doi.org/10.1126/science.2449732
Woodford, B. J., & Blanks, J. C. (1989). Localization of actin and tubulin in developing and adult
mammalian photoreceptors. Cell and Tissue Research, 256(3), 495–505.
https://doi.org/10.1007/BF00225597
Xiao, M., & Hendrickson, A. (2000). Spatial and temporal expression of short, long/medium, or both
opsins in human fetal cones. Journal of Comparative Neurology, 425(4), 545–559.
https://doi.org/10.1002/1096-9861(20001002)425:4<545::AID-CNE6>3.0.CO;2-3
Xu, X. L., Fang, Y., Lee, T. C., Forrest, D., Gregory-Evans, C., Almeida, D., Liu, A., Jhanwar, S. C.,
Abramson, D. H., & Cobrinik, D. (2009). Retinoblastoma Has Properties of a Cone Precursor Tumor
and Depends Upon Cone-Specific MDM2 Signaling. Cell, 137(6), 1018–1031.
https://doi.org/10.1016/j.cell.2009.03.051
Xu, X. L., Singh, H. P., Wang, L., Qi, D. L., Poulos, B. K., Abramson, D. H., Jhanwar, S. C., & Cobrinik, D.
(2014). Rb suppresses human cone-precursor-derived retinoblastoma tumours. Nature, 514(7522),
385–388. https://doi.org/10.1038/nature13813
Xu, Y., An, X., Guo, X., Habtetsion, T. G., Wang, Y., Xu, X., Kandala, S., Li, Q., Li, H., Zhang, C.,
Caldwell, R. B., Fulton, D. J., Su, Y., Hoda, M. N., Zhou, G., Wu, C., & Huo, Y. (2014). Endothelial
PFKFB3 plays a critical role in angiogenesis. Arteriosclerosis, Thrombosis, and Vascular Biology,
34(6), 1231–1239. https://doi.org/10.1161/ATVBAHA.113.303041
Yan, W., Peng, Y. R., van Zyl, T., Regev, A., Shekhar, K., Juric, D., & Sanes, J. R. (2020). Cell Atlas of
The Human Fovea and Peripheral Retina. Scientific Reports, 10(1), 9802.
https://doi.org/10.1038/s41598-020-66092-9
Young, R. W. (1967). The renewal of photoreceptor cell outer segments. The Journal of Cell Biology,
33(1), 61–72. https://doi.org/10.1083/jcb.33.1.61
Young, R. W. (1985). Cell differentiation in the retina of the mouse. The Anatomical Record, 212(2), 199–
205. https://doi.org/10.1002/ar.1092120215

224

Zhang, J., Benavente, C. A., McEvoy, J., Flores-Otero, J., Ding, L., Chen, X., Ulyanov, A., Wu, G.,
Wilson, M., Wang, J., Brennan, R., Rusch, M., Manning, A. L., Ma, J., Easton, J., Shurtleff, S.,
Mullighan, C., Pounds, S., Mukatira, S., … Dyer, M. A. (2012). A novel retinoblastoma therapy from
genomic and epigenetic analyses. Nature, 481(7381), 329–334. https://doi.org/10.1038/nature10733
Zhang, X. M., Hashimoto, T., Tang, R., & Yang, X. J. (2018). Elevated expression of human bHLH factor
ATOH7 accelerates cell cycle progression of progenitors and enhances production of avian retinal
ganglion cells. Scientific Reports, 8(1), 1–13. https://doi.org/10.1038/s41598-018-25188-z
Zhang, Y., Li, A., Zhu, X., Wong, C. H., Brown, B., & Craft, C. M. (2001). Cone Arrestin Expression and
Induction in Retinoblastoma Cells. In R. E. Anderson, M. M. LaVail, & J. G. Hollyfield (Eds.), New
Insights Into Retinal Degenerative Diseases (pp. 309–317). Springer US.
https://doi.org/10.1007/978-1-4615-1355-1_33
Zheng, C., Schneider, J. W., & Hsieh, J. (2020). Role of RB1 in human embryonic stem cell-derived
retinal organoids. Developmental Biology, 462(2), 197–207.
https://doi.org/10.1016/j.ydbio.2020.03.011
Zhong, X., Gutierrez, C., Xue, T., Hampton, C., Vergara, M. N., Cao, L. H., Peters, A., Park, T. S.,
Zambidis, E. T., Meyer, J. S., Gamm, D. M., Yau, K. W., & Canto-Soler, M. V. (2014). Generation of
three-dimensional retinal tissue with functional photoreceptors from human iPSCs. Nature
Communications, 5(1), 4047. https://doi.org/10.1038/ncomms5047
Zhu, X., Brown, B., Li, A., Mears, A. J., Swaroop, A., & Craft, C. M. (2003). GRK1-dependent
phosphorylation of S and M opsins and their binding to cone arrestin during cone phototransduction
in the mouse retina. Journal of Neuroscience, 23(14), 6152–6160.
https://doi.org/10.1523/jneurosci.23-14-06152.2003

Abstract (if available)

Abstract The human retina is a delicate tissue that provides the light-sensing capabilities that give us vision. However, various congenital and age-related conditions can cause this system to weaken and fail, leading to vision impairment or total blindness. Human cone photoreceptors in particular have the unique capability to form retinoblastomas, indicating the presence of unique human developmental processes that are not well elucidated. This thesis aimed to further dissect cone maturation through full-length scRNA-seq of enriched fetal cone photoreceptors. Through examination of known photoreceptor marker genes, I identified differential isoform usage of canonical rod and cone marker genes NRL, RXRG, and THRB. Further parsing of trajectories from retinal progenitor cells (RPCs) through photoreceptors suggested that there are two potential routes of differentiation to rods, one of which was through a post-mitotic cone/rod photoreceptor precursor state. I identified new gene expression defining this population, as well as showed that cone-directed precursors expressed ONECUT1 and had increased OLIG2 regulon activity, elements previously only described in RPCs that were fated to become cones. As these cones continued maturing, I found a progression of lncRNA genes that gained or lost expression with development. Differential expression between developing rods and cones, to explore cone-specific features relevant to the pRB-deficient cone precursor proliferative response, identified SYK, which we then showed had expression in normal cone maturation and was needed for the pRB-depleted cone entry into the cell cycle. ? Modeling of human retina through 3D hESC-derived retinal organoids provides a more available tissue for studying development and disease, however current organoid models are imperfect, and the maturation of individual cell types has not been closely compared against human retinal cells. To compare cone cell development between tissues, I first examined the expression of cone proliferation-related proteins and found differences in timing and localization in relation to expression of cone arrestin (ARR3). An scRNA-seq dataset was generated from organoids of two production methods as done for fetal retina to directly compare cone transcriptomes. Each production method generated cone populations poorly represented from the other, with one more rapidly producing late maturation cone cells. However, overall organoid cones recapitulated fetal features like fetal NRL and RXRG isoform patterns and photoreceptor precursor gene expression. Late maturation cone states were also better isolated from organoids than from fetal retina. Direct comparison of fetal and organoid cone cells demonstrated major differences in the expression of ribosome and glycolytic genes which suggest that culture conditions distinctly impact organoid gene expression. The scRNA-seq data also identified aberrant organoid cone populations with hypoxic and oxidative stress signatures, whose marker proteins BHLHE40 and ENPP2 were observed in atypical internal and border region cone cells. Unexpected expression in fetal foveal cones indicated both proteins were expressed during normal as well as abnormal cone development. ? These dissertation results provide full-length photoreceptor transcriptomic datasets that will be useful for future focused analyses. These observations provide insights into cone cell differentiation and maturation in fetal retina, identify cone-cell intrinsic gene expression that has a role in the proliferative response to pRB loss, and outline key similarities and differences in retinal organoid-derived cone cells that will help researchers improve on organoid culture methods and analysis.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Signaling cascades: a functional characterization of cone arrestin and a differential gene expression analysis of developing retinal ganglion cells

PDF

Understanding human nephrogenesis and scaling synthesis of organoids facilitate modeling of kidney development and disease

PDF

The autism-associated gene SYNGAP1 regulates human cortical neurogenesis

PDF

Genetics and the environment: evaluating the role of noncoding RNA in autism spectrum disorder

PDF

Cross-species comparison of non-canonical Wnt signaling in the developing retina

PDF

Investigating molecular roadblocks to enhance direct cellular reprogramming

PDF

Dissecting the heterogeneity of mouse hematopoietic stem cells in vivo

PDF

Generation and long-term culture of human cerebellar organoids to study development and disease.

PDF

Fibroblastic connective tissue cells: the blastema stem cells and source of large-scale chondrogenesis in the regenerating lizard tail

PDF

pRB structures and functions mediating suppression of retinoblastoma initiation

PDF

Elements of photoreceptor homeostasis: investigating phenotypic manifestations and susceptibility to photoreceptor degeneration in genetic knockout models for retinal disease

PDF

Transcriptional and epigenetic mechanisms underlying sensory hair cell differentiation and regeneration

PDF

The organ of Corti sensory progenitor cell competence is controlled by SoxC transcription factors

PDF

Identification of therapeutic targets in human cerebral brain organoid models of neurodegeneration

PDF

Single cell gene expression analysis of KSHV infection in three-dimensional oral epithelial tissue cultures

PDF

Characterization of human embryonic stem cell derived retinal pigment epithelial cells for age-related macular degeneration

PDF

RPE secretome for the treatment of retinal degeneration in the RCS rat

PDF

Novel roles for Maf1 in embryonic stem cell differentiation and adipogenesis

PDF

Cone arrestin 4 contributes to vision, cone health, and desensitization of the dopamine receptor D4

PDF

Skeletal cell fate plasticity in zebrafish bone development and regeneration

Asset Metadata

Creator Shayler, Dominic William Helsdon (author)

Core Title Transcriptomic maturation of developing human cone precursors in fetal and 3D hESC-derived tissues

School Keck School of Medicine

Degree Doctor of Philosophy

Degree Program Development, Stem Cells and Regenerative Medicine

Degree Conferral Date 2021-08

Publication Date 07/22/2022

Defense Date 06/10/2021

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag cone cell,eye,human,maturation,OAI-PMH Harvest,organoid,photoreceptor,retina,single-cell RNA sequencing

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Segil, Neil (committee chair), Cobrinik, David (committee member), Craft, Cheryl (committee member), Ichida, Justin (committee member)

Creator Email dominicshayler@comcast.net,shayler@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-oUC15622359

Unique identifier UC15622359

Legacy Identifier etd-ShaylerDom-9879

Document Type Dissertation

Format application/pdf (imt)

Rights Shayler, Dominic William Helsdon

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright. The original signature page accompanying the original submission of the work to the USC Libraries is retained by the USC Libraries and a copy of it may be obtained by authorized requesters contacting the repository e-mail address given.

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA

Repository Email cisadmin@lib.usc.edu