University of Southern California Dissertations and Theses

Quantitative analysis of mouse medial prefrontal cortex whole brain structural connectivity

(USC Thesis Other)

Quantitative analysis of mouse medial prefrontal cortex whole brain structural connectivity

PDF

Download

Open document

Flip pages

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content University of Southern California
Quantitative Analysis of
Mouse Medial Prefrontal Cortex
Whole Brain Structural Connectivity
by
Muye Zhu
A dissertation presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In partial fulllment of the
Requirements for the degree
Doctor of Philosophy
In Neuroscience
December 2018
Advisory Committee:
Hongwei Dong, MD, PhD
Larry Swanson, PhD
Li Zhang, PhD
Samuel Andrew Hires, PhD
Daniel Philipp Holschneider, PhD
Acknowledgements
I would like to thank Dr. Hongwei Dong, whose mentorship, encouragement and scientic
insights have made these few years the most intellectually enriching time of my life. I arrived
full of doubts, now at this rite of passage I know I shall overcome. Thank you to the members
of my qualications and dissertation committees: Dr. Larry Swanson, Dr. Li Zhang, Dr.
Samuel Andrew Hires and Dr. Daniel Philipp Holschneider, whose guidance expanded my
visions and helped me to evaluate and plan experiments critically. Many thanks to my
colleagues and friends at the Dong lab, whose talent, dedication and support have been a
constant source of inspiration.
I thank my husband Dmitriy for being my best friend, my safe harbor, my trusted companion.
I am grateful for your steadfast support and earnestness. I feel both excited about and
content with our future. I know you do too.
I thank my parents for their personal sacrices that aorded me greater opportunities
throughout the decades. To my mother, I can't wait to share more moments of excite-
ment with you. To my father, we wish you were here but we know you would be proud. We
will miss you always.
1
Contents
Abbreviations 4
1 Chapter 1: Introduction 10
2 Chapter 2: Quantication of Whole Brain Connectivity 12
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2 Methods and Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2.1 Overview of Image Processing Pipeline . . . . . . . . . . . . . . . . . 13
2.2.2 Image Registration . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2.3 Injection Site Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.2.4 Tracer Segmentation . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.2.5 Statistical Model of Brain Connectivity . . . . . . . . . . . . . . . . . 37
2.3 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3 Chapter 3: Whole Brain Connectivity of Mouse mPFC 45
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.2 Material and Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.2.1 Tracers and Stereotaxic Surgeries . . . . . . . . . . . . . . . . . . . . 45
3.2.2 Tissue Preparation and Immunohistochemistry . . . . . . . . . . . . . 46
3.2.3 Image Acquisition and Processing . . . . . . . . . . . . . . . . . . . . 47
3.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.3.1 Denition of DP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.3.2 Stereiotaxic Tracer Injections in mPFC . . . . . . . . . . . . . . . . . 53
3.3.3 Overview: Connectivity of mPFC Cortices . . . . . . . . . . . . . . . 55
3.3.4 mPFC Connectivity with Cortex . . . . . . . . . . . . . . . . . . . . 61
3.3.5 mPFC Connectivity with Thalamus . . . . . . . . . . . . . . . . . . . 67
3.3.6 mPFC Connectivity with Striatum . . . . . . . . . . . . . . . . . . . 75
2
3.3.7 mPFC Connectivity with Hippocampal Formation, Midbrain and Hind-
brain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
3.3.8 mPFC Connectivity with Hypothalamus and Amygdalar Complex . . 84
3.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
4 Bibliography 87
5 Appendix 99
5.1 Prefrontal Cortex Region Name Label . . . . . . . . . . . . . . . . . . . . . . 99
5.2 Thalamic Region Name Label . . . . . . . . . . . . . . . . . . . . . . . . . . 100
5.3 Table of Mouse Brain Structure Names and their Hierarchy Groups . . . . . 101
3
Abbreviations
ACAd Anterior Cingulate Area, Dorsal Part.
ACAv Anterior Cingulate Area, Ventral Part.
ACB Nucleus Accumbens.
AI Agranular Insular Area.
AId Agranular Insular Area, Dorsal Part.
AIp Agranular Insular Area, Posterial Part.
AIv Agranular Insular Area, Ventral Part.
AMd Anteromedial Nucleus, Dorsal Part.
AMv Anteromedial Nucleus, Ventral Part.
AMY Amygdalar Complex.
AON Anterior Olfactory Nucleus.
AONd Anterior Olfactory Nucleus, Dorsal Part.
AONl Anterior Olfactory Nucleus, Lateral Part.
ARA Allen Reference Atlas.
AUD Auditory Area.
AUDd Dorsal Auditory Area.
AUDp Primary Auditory Area.
AUDv Ventral Auditory Area.
BLAa Basolateral Amygdalar Nucleus, Anterior Part.
4
BST Bed Nuclei of Stria Terminalis.
BSTam Bed Nuclei of Stria Terminalis, Anterior Division, Anteromedial Area.
BSTif Bed Nuclei of Stria Terminalis, Posterior Division, Interfascicular Nucleus.
BSTpr Bed Nuclei of Stria Terminalis, Posterior Division, Principal Nucleus.
CA1 Field CA1.
CL Central Lateral Nucleus of the Thalamus.
CLA Claustrum.
COA Cortical Amygdalar Area.
COApl Cortical Amygdalar Area, Posterial Part, Lateral Zone.
COApm Cortical Amygdalar Area, Posterial Part, Medial Zone.
CP Caudate Putamen.
CS Superior Central Nucleus Raphe.
CSl Superior Central Nucleus Raphe, Lateral Part.
CSm Superior Central Nucleus Raphe, Medial Part.
CTX Cortex.
DP Dorsal Pedunclar Area.
DPd Dorsal Pedunclar Area, deep layers.
DPs Dorsal Pedunclar Area, supercial layers.
DR Dorsal Nucleus Raphe.
ECT Ectorhinal Area.
5
ENTl Entorhinal Area, lateral part.
EPd Endopiriform Nucleus.
FS Fundus of Striatum.
GP Globus Pallidus.
GPe Globus Pallidus, External Segment.
GPi Globus Pallidus, Internal Segment.
GU Gustatory Area.
HB Hindbrain.
HPF Hippocampal Formation.
HY Hypothalamus.
ILA Infralimbic Area.
LD Lateral Dorsal Nucleus of the Thalamus.
LHA Lateral Hypothalamus.
LP Lateral Posterior Nucleus of the Thalamus.
LSc Lateral Septum, Caudal Part.
LSr Lateral Septum, Rostral Part.
LSv Lateral Septum, Ventral Part.
MA Magnocellular Nucleus.
MB Midbrain.
6
MD Mediodorsal Nucleus of the Thalamus.
MDc Mediodorsal Nucleus of the Thalamus, Central Part.
MDl Mediodorsal Nucleus of the Thalamus, Lateral Part.
MDm Mediodorsal Nucleus of the Thalamus, Medial Part.
MOp Primary Motor Area.
MOs Secondary Motor Area.
mPFC Medial Prefrontal Cortex.
MS Medial Septal Nucleus.
NDB Diagonal Band Nucleus.
ORB Orbital Area.
ORBl Orbital Area, Lareral Part.
ORBm Orbital Area, Medial Part.
ORBvl Orbital Area, Ventrolateral Part.
OT Olfactory Tubercle.
PAG Periaqueductal Gray.
PCG Pontine Central Gray.
PERI Perirhinal Area.
PF Parafascicular Nucleus.
PIR Piriform Cortex.
PL Prelimbic Area.
7
PRN Pontine Reticular Nucleus.
PRNc Pontine Reticular Nucleus, Caudal Part.
PRNr Pontine Reticular Nucleus, Rostral Part.
PT Parataenial Nucleus.
PTLp Posterior Parietal Association Areas.
PVT Paraventricular Nucleus of the Thalamus.
RE Nucleus Reunions.
RSP Retrospenial Area.
RSPagl Retrospenial Area, Lateral Granular Part.
RSPd Retrospenial Area, Dorsal Part.
RSPv Retrospenial Area, Ventral Part.
RT Reticular Nucleus of the Thalamus.
SC Superior Colliculus.
SCm Superior Colliculus, Motor Part.
SF Septombrial Nucleus.
SH Septohippocampal Nucleus.
SI Substantia Innominata.
SNr Substantia Niagra, Reticular Part.
STR Striatum.
SUB Subiculum.
8
SUBd Subiculum, Dorsal Part.
SUBv Subiculum, Ventral Part.
TEa Temporal Association Areas.
TH Thalamus.
TRS Triangular Nucleus of Septum.
TTd Taenia Tecta dorsal.
TTv Taenia Tecta Ventral.
VAL Ventral Anterior-Lateral Complex of the Thalamus.
VIS Visual Areas.
VISam Anteromedial Visual Area.
VISpm Posteromedial Visual Area.
VM Ventral Medial Nucleus of the Thalamus.
VTA Ventral Tegmental Area.
9
Chapter 1: Introduction
It has been well established that the medial prefrontal cortex (mPFC) in primates (in-
cluding human) plays an essential role in regulating cognition, emotion, and motivation [1].
Consequently, disruption of the mPFC has been implicated in numerous neurological and
psychiatric disorders, including PTSD, frontotemporal dementia, autism, OCD, ADHD, de-
pression, anxiety, schizophrenia, and drug addiction [2]. Existing evidence reveals that a
broad range of these symptoms are associated with dysfunction of the mPFC with neural
circuitry or cell type specicities. Therefore, the systemic mapping of neural circuitries of
the mPFC is of signicant research and clinical interest.
In primates, the prefrontal cortex (PFC) consists of three major divisions: the orbital,
medial and lateral parts [3]. The orbital and medial divisions serve well-recognized roles in
emotional behavior and the dorsolateral prefrontal cortex (or lateral) in executive functions
of the PFC [3, 4]. Neural circuits of these prefrontal areas were characterized by a series of
neural tract tracing studies in macaque monkeys [4, 5]. Accordingly, the primate mPFC is
generally characterized by its outputs to visceral control areas in the hypothalamus (both
medial and lateral) and periaqueductal gray (PAG) [1, 6].
In rats, the mPFC consists of four main divisions which from dorsal to ventral are (1)
the medial agranular area (AGm), which includes the secondary motor cortex (MOs) that
specically controls eye movement; (2) the anterior cingulate (ACA, dorsal and ventral
divisions); (3) the prelimic (PL) [7, 8] and nally (4) the infralimbic area (ILA) [4, 9{11].
Neuronal connectivity of these individual cortical areas was examined using classic neural
tract tracing methods. In general, it was suggested that the dorsal regions of mPFC (AGm
and ACAd) are homologous to the frontal eye elds in primates associated with various
motor behaviors; the infralimbic area of rats is primarily involved in visceromotor functions,
homologous to the orbitomedial cortex of primates, and the PL and ventral anterior cingulate
ACAv is involved in limbic/cognitive activity, homologous to the lateral/dorsolateral cortex
of primates.
The cytoarchitectural delineations of the mouse cerebral cortex mostly follow that in
10
the rat brain [12{14]. Consequently, homological mPFC areas, namely, the ILA, PL, ACA
and adjacent MOs frontal area (corresponding to the AGm), are identied in several stan-
dard mouse brain atlases (i.e., [14]). Previous work of the Mouse Connectome Project
(www.MouseConnectome.org) conducted in our laboratory generated a cortical connectivity
atlas, based on over 600 labeled neural pathways from tracer injections placed across the
entire neocortex [15]. Detailed analysis of the intracortical connectivity data showed that
projections from all dierent cortical subnetworks are able to topographically converge onto
discrete regions of the mPFC; while the mPFC provides an interface for integrating infor-
mation from the medial cortical subnetworks that process external stimuli (such as visual,
auditory, somatic sensory) and the lateral cortical subnetworks that process internal stimuli
(such as visceral and gustatory information).
Together with numerous other studies reported in rats and primates, the mPFC appears
to serve as a critical hub that provides an interface for interactions among the neocortex,
amygdala, and hippocampus, thus, bridging the communication throughout the entire cere-
brum. In turn, the MPF generates wide range projections to the thalamus, striatum, and
hypothalamus to regulate motivated behavior. Despite its scientic and clinical importance,
a whole brain wide examination and comparison of neuronal connectivity of the mouse mPFC
cortices has not yet been conducted. Additionally, the organization of neural circuits associ-
ated with individual mPFC areas into global neural networks remains undetermined. In this
thesis work, we quantitatively characterize the global structural connectome of mPFC from
terabyte range microscopic imaging data. We develop a full image processing and statistical
analysis pipeline over a massively parallel computing infrastructure to enable connectivity
mapping at large scale. An accompanying interactive software has also been built to sup-
port data visualization and quality assurance. We expect our systematic and quantitative
analysis of detailed connectivity of the mPFC will provide a framework for exploring mPFC
connectional pathologies of neurological/neuropsychiatric diseases.
11
Chapter 2: Quantication of Whole Brain Connectivity
2.1 Introduction
Analysis of rodent neuronal connectivity and other neuroanaomical data has been heavily
reliant on manual data annotation. To do this, a human expert or neuroanatomist examines
microscopic images of the brain, and manually notes the impression of strength of histological
labeling (i.e., tracer labeled axons or neurons) in dierent brain regions. A typical notation
system uses a set of symbols +++, ++, +, - to semi-quantitatively indicate strengths of
labeling as strong, moderate, weak and no connectivity, respectively. Other notation systems
are also used, for example the symbol set from +++++ to - to oer even ner granularities.
This traditional manual annotation oers much insight into structural organization of the
brain in the last several decades. Nevertheless, this approach has several drawbacks.
Firstly, manual annotation is very time consuming and labor intensive, therefore, clearly
restricts the researchers productivity. It is challenging to maintain intra-person annotation
consistency during the lengthy process. Additionally, inter-person interpretation of connec-
tivity strength may dier. All of these factors may contribute to a large degree of variability
and inconsistency of neuroanatomical data (i.e., tract tracing, histological/immunohistological
staining, in situ hybridization mRNA expression) reported in literatures.
Secondly, manual annotation of anatomical data usually is reported in a highly schematic
tabulate form using a limited set of symbols to indicate labeling strength in dierent anatom-
ical regions without specic spatial information (i.e., specic stererotaxic coordinates or spa-
tial correlations with adjacent brain structures). However, many classically delineated brain
structures in current standard brain atlases [1] arguably can be further parceled or subdivided
into sub-domains, where each sub-domain is distinguishable by its own innervation patterns.
A notable example is the classically dened caudate putamen (CP, or dorsal striatum), which
was recently subdivided into 29 domains based on specicities of axonal projections from in-
dividual cortical areas of the entire mouse brain using newly developed computational tools
[16]. Such connectivity-based parcellation is otherwise very dicult to uncover using classic
12
manual annotation approaches. In recent years, neural connectivity patterns (i.e., axonal
projection pathways or retrogradely labeled neurons) are frequently presented or reported in
schematic drawings or maps manually produced either using camera luscida or with commer-
cial software (i.e., Adobe Illustrator or Photoshop) These drawings are capable of providing
insights to spatial patterns of neural connectivity, but are extremely laborious. Moreover,
manual drawings can be subjective with inconsistency of data interpretation (i.e., drawing
of anatomical boundaries) and may not necessarily constitute a high quality transformation
into the atlas space. Additionally, it is practically impossible to manually draw and map
connectivity and other anatomical data for every single section, therefore, large amount of
data from those sections are typically omitted.
Last but not least, neither an annotation table nor a manual representative drawing of
neuroanatomical data captures the range of connectivity strengths observed in target regions
from multiple brain samples with comparable injection region, which can be considerably
wide [17]. These factors complicate eorts to make statistical estimations and comparisons
from multiple brain samples, and to provide quantitative analysis of connectivity patterns.
To address these limitations of the traditional manual annotation and to systematically
handle the large datasets generated in experiments, we have developed an image processing
and data analysis pipeline, deployed over a high performance computing cluster, to enable
a quantitative and statistical understanding of whole brain structural connectivity.
2.2 Methods and Results
2.2.1 Overview of Image Processing Pipeline
For a typical experiment in the Mouse Connectome Project (www.MouseConnectome.org),
one animal receives injections of up to 4 anterograde and/or retrograde tracers in dierent
brain structures. Post surgery survival times were 2-3 weeks, after which animals were sac-
riced with 4% PFA transcardiac perfusion. Brains were extracted and sectioned on coronal
planes at the thickness of 50 um. 1 in 4 series of sections ( 40-50 sections with 200 um
13
apart) along the rostrocaudal axis were processed to reveal tracer labeling with
uorescent
dyes (see chapter 3 for detailed tissue processing methods), resulting in a series of 40 50
sampled sections for processing and analysis. Microscopic scanning of a 4 tracer section at
10x magnication generates a 5 channel (including an additional reference Nissl channel)
16-bit image with approximate xy-dimensions 14k by 11k, corresponding to 1.54 GB of im-
age content. Therefore a 4 tracer section series generates between 60 and 80 GB of image
data. Studies that simultaneously examines 15 20 section series, such as those presented
in this thesis, involve over 1 TB of raw microscopic images. In order to gain insight into
complex brain connectivity patterns from such large data collections, it is essential to make
use of powerful image processing pipelines and computing infrastructures, as well as develop
strategies to organize, navigate and visualize the raw and processed images.
For this purpose, we have developed a pipeline that achieves high quality image regis-
tration and tracer signal quantication. Image processing algorithms are often CPU and
memory intensive. Some modules of our pipeline require memory usage a few times the
size of an input image, and several minutes of computing time for each section channel.
Since computed output from each pipeline module is saved to disk, 150 200 GB of storage
space is also required for a section series. Due to the high amount of computing and storage
resources needed, we deploy our pipeline on the high performance computing cluster and dis-
tributed le system operated by the University of Southern California Laboratory of Neuro
Imaging (http://www.loni.usc.edu/about loni/resources/ComputingResources.php)
[18, 19]. We also developed our in house software, Outspector, to allow management, pro-
cessing, analysis and visualization of large image datasets on a typical personal computer.
Importantly, Outspector provides a simple user interface to initiate our pipeline modules on
the computing cluster, and allows seamless integration of user interactions and ne tuning
at every pipeline stage.
14
2.2.2 Image Registration
Image registration is a process where a given image is spatially transformed to align
with a reference image or common coordinate system. In the eld of medical registration,
frequently the central task is to establish anatomical region correspondence between the
imaging data and a reference atlas. Several well known mouse brain atlases are currently
available to the research community: the Allen Reference Atlas (ARA, available at http:
//mouse.brain-map.org/static/atlas) [14], the Waxholm Space (available at https://
scalablebrainatlas.incf.org/mouse/WHS12)[20], the Paxinos mouse brain in stereotaxic
coordinates (available at https://www.elsevier.com/books/paxinos-and-franklins-t
he-mouse-brain-in-stereotaxic-coordinates/paxinos/978-0-12-391057-8) [12] and
the Allen Common Coordinate Framework (CCF, downloadable at http://help.brain-m
ap.org/display/mousebrain/API) [21]. By default, our pipeline uses the Allen Reference
Atlas. Both its accompanying high resolution nissl stained section images and interactive,
scalable region boundary drawings are easily viewable and downloadable online. Most of
our connectivity data are collected from coronal sections and registered to the coronal ARA.
However, any 2D atlas, including ones of dierent species and sectioning planes, can be easily
incorporated into our work
ow.
Several automatic approaches to register mouse brain coronal section series to the refer-
ence atlas have been proposed. The Allen Institute registration process rst constructs serial
sections into a 3D volume, then uses tissue auto
uorescence to register the aligned volume
to a 3D template volume generated by averaging a large number of brain specimens [22, 23].
This registration pipeline has several limitations. Firstly, the Allen Institute meso-scale
connectome datasets were collected using serial two-photon (STP) tomography [24]. STP
tomography is known to yield inherently pre-aligned serial images with little tissue distor-
tion, allowing the sections to be easily constructed into a 3D volume. For many widely used
imaging systems, the section pre-alignment assumption is not applicable. Aligning serial 2D
sections can be especially challenging due to large tissue sampling gap, tissue distortion and
damage, variability in section orientation during mounting and section position in the imag-
15
ing eld. The diculty in achieving robust 3D volume reconstruction across a large number
of specimens is a signicant barrier to a 3D to 3D registration framework. Secondly, the
need for a dedicated auto
uorescence channel eectively reduces the number of wavelengths
available for tracers and other informative biological markers. Finally, the lack of accurate
annotated 3D reference volume negatively aects correct connectivity quantications. CCF,
the leading eort in 3D mouse brain atlas annotation, is generated from aligning and inter-
polating annotated ARA 2D sections. Though the project is of great signicance, the level of
agreement between its annotation outcome and the ground truth 2D annotations is not yet
satisfactory. Other whole brain meso-scale connectome projects have focused on 2D to 2D
registration methods. Zhang et al use geometric transformation to minimize the dierence
between manually selected correspondence points on specimen sections and reference sec-
tions. Dierent scaling factors along x and y axis are allowed to model tissue distortion [17].
However since the transformation is still linear, it can only achieve a coarse alignment. F urth
et al extract a tissue mask, and apply thin plate spline transformation with correspondence
points automatically selected along the tissue contour with respect to principal axes of the
tissue [25]. We found that tissue mask point cloud principal axes often do not align well with
the expected dorsal-ventral and medial-lateral axis of a section. Additionally, although thin
plate spline methods model non linear deformations well [26, 27], transformations based on
tissue contours alone are insucient for ne registration of many anatomical regions.
We employ a semi-automatic, dieomorphic 2D to 2D registration process, where op-
tional user interaction can be used to ne tune the registration when needed. The steps of
registration are described below.
(1) Assign an ARA level for each section in the series. This is done by a researcher
comparing and matching the section Nissl channel to the ARA Nissl images.
We are interested in automating section level assignments since it can take an experi-
enced research an hour to complete the task for one brain sample. We tested a near
automatic level assignment approach, where the section corresponding to level 53 is
rst identied, then all remaining sections receive a level estimation by assuming all
16
neighboring section pairs are 2 levels apart (see tissue preparation in chapter 3). 591
sections were each given a manually assigned level l and an estimated level
^
l. Average
magnitude of error

^
ll

in the level estimation is 5.47 (95%CI[5:04; 5:90]). We found
section planes further away from the pivot plane of level 53 are more prone to error
propagation and accumulate larger errors (Figure 2.1). Expressing the level estimation
error as a linear regression model:
error =
^
ll =c
1
(ll
0
) +c
0
where c
1
and c
0
are the model coecients, and l
0
= 53, we apply a correction to the
level estimator
^
l. This gives signicantly better estimated levels with an average error
magnitude of 3.39 (95%CI[3:10; 3:67],p = 4:2410
54
in two-sided dependent t-test).
Figure 2.1: Error in ARA coronal level estimation increases as sectioning plane moves further
away from level 53. A linear regression model explains 71.16% of error variability.
A linear regression model of estimation error assumes independence between sections.
Leveraging the sequential nature of the data may provide a better estimator
d
l
i+1
=
b
l
i
+ l
i!i+1
=
b
l
i
+f(S(i + 1;j(n)))
17
where the level dierential between theith and (i+1)th sections is a function of then-
dimensional vectorS(i + 1;j(n)), representing geometric similarity measures between
the contours of the (i+1)th section and itsn neighbor sections, indexed byj(n), with
n being an input to the algorithm. This is a direction that we are actively pursuing to
further improve the eciency of our pipeline.
(2) Perform a coarse alignment with geometric transformations. We rst extract
a binary tissue mask from the Nissl channel with a gaussian mixture model (GMM). A
set of geometric transformation parameters is analytically derived using binary masks
of the tissue and reference image. The transformation dened by these parameters are
then applied to the sections to initialize an alignment.
Image registration is frequently formulated as an optimization problem, where a target
function expressing a chosen registration quality metric is numerically minimized. The
target functions have complex landscape and are not globally convex, even in the case
of ane registration with small degrees of freedom (DoF) [28]. Non-linear registration
algorithms have much higher DoF, and benet greatly from an appropriate coarse
alignment that help prevent instances where the optimization is trapped at a bad local
minimum [29, 30].
Histogram of a typical Nissl stain section is shown in Figure 2.2A. Due to variability in
the tissue preparation, immunostaining and image processing, the threshold value that
separates tissue foreground intensity distribution and background intensity distribution
needs to be found for each individual section. Under GMM, the probability of observing
each pixel intensity x is given as following:
p(x) =w
f
N (xj
f
;
2
f
) +w
b
N (xj
b
;
2
b
)
N (xj;
2
) =
1
p
2
2
e

(x)
2
2
2
w
f
+w
b
= 1
18
the model parametersw
f
,
f
,
f
,w
b
,
b
,
b
are found with Expectation Maximization
(EM) algorithm. The model gives probabilities of each pixel belonging to background
or foreground according to Bayesian rule:
p( cjx
c2ff;bg
) =
w
c
N (xj
c
;
2
c
)
p(x)
The GMM classication results include background pixels immediately adjacent to the
tissue, where the intensity values are contributed to by the point spread function of
the imaging system. Therefore we use a harsher intensity threshold valuex
t
=
b
+6
b
instead. A typical tissue mask segmentation result is shown in Figure 2.2A. Dust
particles and other small debris are removed with simple connected component analysis.
In histology section image registration, ane transformation (rotation, translation,
scaling and shearing) models mechanical tissue handling introduced deviation from
atlas images, while the non-linear transformations account for deformations caused by
xation, temperature changes as well as inter-brain variations [31]. In practice, we do
not always observe shearing as a global force over an entire section. For the initial lin-
ear transformation alignment, we instead use geometric transformation, composed of a
rotation matrix, a translation vector and a scaling factor. We derive each component
individually and analytically, since we had limited success with common registration
algorithms that optimize for these parameters. To obtain the rotation matrix, the
smallest rectangle Rec
section
that fully contains the binary tissue mask is found by
rotating calipers method [32]. A similar rectangle Rec
atlas
is found for the atlas im-
age. The angle between Rec
section
and Rec
atlas
determines the rotation matrix. The
translation vector and scaling factor are dened as dierences between binary mask
centers of mass, and the ratio of mask areas respectively. Representative results of the
geometric transformation are shown in Figure 2.2B.
19
Figure 2.2: Tissue Segmentation and Initialization of Tissue Alignment by Geometric
Transform. (A) Typical result of Nissl stained tissue segmentation. The intensity value of 33.56
found by the algorithm is shown at the red line over the image intensity histogram. Result from a
range of other values are shown for comparison. (B) Geometric transformation provides an initial
alignment for Nissl stained sections. The tightest tting bounding box and dierence between
desired and actual tissue angle is shown with the extracted tissue binary mask. A reference atlas
image and the transformed tissue is shown in the right two gures.
(3) Minimize image energy by applying dieomorphic non-parametric demon
algorithm.
(4) Modify correspondence points and repeat step (3) if necessary. Following au-
tomatic registration, Outspector allows the user to visually inspect registration quality,
by overlaying a wireframe of the atlas on top of the Nissl channel tissue image. The
correspondence points in the tissue image and atlas image can both be modied to
improve the current registration.
20
Some brain regions are prone to registration errors. The thin neocorcial layer 6b,
the small and fast varying (along z axis) thalamic nuclei, regions showing only subtle
cyto-architectural dierences with their neighbors (e.g. claustrum) are all "hotspots"
for misalignment. Human ne-tuning of registration is helpful in studies when precise
registration is desired for even the error prone regions.
Availability of high quality Nissl stain image is important for a researcher to establish
correspondence points eciently. The Nissl image is often degraded from several fac-
tors, including blurring introduced by out-of-focus imaging and image downsampling,
and stripe artifacts introduced by image tiling. Stripe artifacts also adversely aect
registration algorithms by manufacturing false local features. We perform deconvolu-
tion and destriping to counter these issues (Figure 2.3).
Deconvolution: In an imaging system, the point spread function (psf) describes the
system's response to a point source. For an image with true intensity I(x;y), the
observed image O at spatial location (i;j) has intensity values equal to the additions
of response to all true image pixels at (i;j):
O(i;j) =
Z Z
u;v2DxDy
I(u;v) psf(iu;jv)dudv
Setting psf as a Gaussian function g, and allowing a maximum of 1000 iterations, the
estimated true image
^
I is iteratively updated according to Richardson-Lucy algorithm
[33]:
^
I
t+1
=
^
I
t

(O (
^
I
t
g)) ~ g

~ g(x;y) =g(supfD
x
gx; supfD
y
gy)
Destripe: Tile scan is a common imaging technique when the specimen exceeds the
eld of view (FoV) in size. The FoV periphery receives reduced excitation light, cre-
ating artifacts in the form of darker stripes at tile boundaries. More recent stitching
algorithms have improved their eciency in stripe pattern correction [34, 35]. How-
21
ever stripe artifacts are still commonly seen in our datasets, due to the data collection
process spanning a few years with evolving stitching methods within the image acqui-
sition software bundle, as well as dicult cases where imaging tiles exhibit uneven base
illumination levels.
A horizontal stripe in an image f(x;y) can be considered as addition of a bias term b
to an entire width e row of pixels:
f
s
(x;y;b;e;y
0
) =
8
>
>
<
>
>
:
f(x;y) +b; if y
0
yy
0
+e:
f(x;y); otherwise:
Vertical stripes are similarly dened.
Multi-scale wavelet decomposition is particularly eective at encoding image stripes.
In the wavelet decomposition, a lter bank of high and low pass lters are successively
applied to the input data, giving coecients for a sequence of frequency bands at
increasingly coarse scales. For a 2D image f(x;y), its discrete wavelet decomposition
with scales l2f1;:::;l
m
g is as following:
f(x;y) =
X
i
X
j
C
LL
lm;ij

LL
lm;ij
(x;y) +
X
l
X
i
X
j
C
HL
l;ij

HL
l;ij
(x;y)+
X
l
X
i
X
j
C
LH
l;ij

LH
l;ij
(x;y) +
X
l
X
i
X
j
C
HH
l;ij

HH
l;ij
(x;y)
where
LL
is the scaling function, and
HL
;
LH
;
HH
are the wavelet functions. At
each level l, C
LL
l
is decomposed to C
LL
l+1
, C
HL
l+1
, C
LH
l+1
and C
HH
l+1;ij
. Each higher level
coecients array contain 1=4 the number of elements of C
LL
l
.
LH
applies low pass
lter along x axis and high pass lter along y axis, condensing horizontal details of
the image into the coecients C
LH
. Similarly, C
HL
and C
HH
carry vertical and diag-
onal details respectively. The image is encoded by low pass coecients and vertical,
22
horizontal and diagonal detail coecients at dierent scales as
f(x;y) () fC
LL
lm
;C
HL
l
;C
LH
l
;C
HH
l
:l2f1;:::;l
m
gg
For an image with ideal horizontal stripes as dened by f
s
(x;y;b;e;y
0
), all stripe
information is contained in C
LH
l
;l2f1;:::;l
m
g. Applying 1D Fourier transform to
rows of C
LH
l
further condenses stripe information to F (0). Multiplication by a zero
mean gaussian function in the Fourier domain dampens F (0), subsequently erasing
the stripes from image spatial domain. Destriped image is reconstructed from reverse
fourier transform and reverse wavelet transform [36].
Figure 2.3: Nissl channel image correction. (A) Stripe artifacts are diminished by combined
wavelet and fourier ltering. (B) Two examples of applying deconvolution to sharpen blurry Nissl
staining.
23
2.2.3 Injection Site Analysis
The pattern of synaptic connections between regions of the brain denes the structural
connectome. Correct annotation of the spatial location of injection sites is critical in inter-
preting neuroanatomical data. In anterograde tracing experiments, with appropriate imag-
ing exposure time settings (avoiding overexposure), the starter cell population, where axonal
bers across the brain originate from at the injection site, are labeled. In retrograde trac-
ing experiments, the volume of tracer deposit denes axon terminal locations of labeled cell
populations. Additionally, pixel intensities inside and adjacent to injection sites are excep-
tionally high and appear as positive in segmentations. However, these high intensity regions
yield very little useful connectivity information and tend to skew the overall connectivity
quantication results. We combine several techniques, including multi-scale wavelet decom-
position, non-linear adaptive intensity adjustment and maximally stable extremal regions
(MSER) detection to annotate tracer injection sites, as well as exclude injection site area
from connectivity quantication.
Starter cells at anterograde tracer injection sites are surrounded by highly irregular and
intense background. To successfully segment them, we begin by dampening signals from
unwanted frequencies. From a section image I(x;y), a rectangular subregion that contains
the injection site is selected by a user. The resulting smaller image I
inj
(x;y) is gaussian
smoothened and wavelet decomposed into 5 levels L =f1; 2; 3; 4; 5g:
I
inj
(x;y) () fC
LL
5
;C
HL
l
;C
LH
l
;C
HH
l
:l2Lg
Detail coecients at level 1, 2, 3 are all dampened to suppress noise. Cell bodies are blobs
within a certain range of physical dimensions, and their image signals are expected to concen-
trate in C
HH
4
andC
HH
5
. At these two levels, C
HH
values large in magnitude are augmented
while small C
HL
and C
LH
magnitudes are suppressed. The low frequency approximation
coecientsC
LL
5
are dampened and convolved with a gaussian kernel to smoothen image back-
24
ground. Specically, for somek
c"
l
2 (1;1),k
c#
l
2 [0; 1),
2 (1;1),c2fLL;HL;LH;HHg:
]
C
HH
l;ij
=
8
>
>
>
>
>
>
<
>
>
>
>
>
>
:
C
HH
l;ij
k
HH"
l
; if

C
HH
l;ij

>

C
HL
l;ij

^ C
HH
l;ij
C
HL
l;ij
> 0
C
HH
l;ij
k
HH"
l
; if

C
HH
l;ij

>

C
LH
l;ij

^ C
HH
l;ij
C
LH
l;ij
> 0
C
HH
l;ij
; otherwise:
g
C
HL
l;ij
=
8
>
>
>
>
>
>
<
>
>
>
>
>
>
:
C
HL
l;ij
k
HL"
l
; if

C
HL
l;ij

<
jC
HH
l;ij
j

^ C
HL
l;ij
C
HH
l;ij
> 0
C
HL
l;ij
k
HH"
l
; if

C
HL
l;ij

<
jC
LH
l;ij
j

^ C
HH
l;ij
C
LH
l;ij
> 0
C
HL
l;ij
; otherwise:
g
C
HL
l;ij
=
8
>
>
>
>
>
>
<
>
>
>
>
>
>
:
C
LH
l;ij
k
LH"
l
; if

C
LH
l;ij

<
jC
HH
l;ij
j

^ C
LH
l;ij
C
HH
l;ij
> 0
C
LH
l;ij
k
LH"
l
; if

C
LH
l;ij

<
jC
HL
l;ij
j

^ C
LH
l;ij
C
HL
l;ij
> 0
C
LH
l;ij
; otherwise:
g
C
LL
5
= (C
LL
5
k
LL#
5
)g(x;y)
The updated coecients are used to reconstruct the injection site image
g
I
inj
(x;y) =
X
i
X
j
g
C
LL
5;ij

LL
lm;ij
(x;y) +
5
X
l=1
X
i
X
j
g
C
HL
l;ij

HL
l;ij
(x;y)+
5
X
l=1
X
i
X
j
g
C
LH
l;ij

LH
l;ij
(x;y) +
5
X
l=1
X
i
X
j
]
C
HH
l;ij

HH
l;ij
(x;y)
Given an imagef(x;y), the linear transformationT (;) =f(x;y)+ achieves a global
intensity adjustment, with controlling gain (contrast) and controlling bias (brightness).
We want to transform the pixel intensities of the wavelet ltered
g
I
inj
(x;y) such that the
foreground and background intensities are further separated. This can not be achieved by
linear mapping of pixel intensities. We use a non-linear local adaptive contrast enhance
method [37], where the initial intensity mapping for a normalized input f(x;y) is dened as
25
[38]:
f(x;y)7!sin
2

2
f(x;y)
p

; f(x;y)2 [0; 1]
By normalizing the Taylor expansion approximation of sin(u), the transformation function
becomes
T (f(x;y);c) =f(x;y)
c
; c =c
1

f(x;y)g(x;y) +
(1f(x;y)g(x;y)) +
+c
2
whereg(x;y) is a gaussian kernel,c
1
,c
2
are user supplied constants. Denotingf(x;y)g(x;y)
as I
g
(x;y), the mapped intensity is adjusted for local contrast enhancement:
E(x;y) =
f(x;y)
f
g
(x;y)
T (f(x;y);c) +
f(x;y)
f
g
(x;y)
@T (f(x;y);c)
@c
(f(x;y)f
g
(x;y))
With the now wavelet ltered and intensity adjusted image E
inj
(x;y), we use MSER al-
gorithm to extract injection site cell bodies [39], which applies a sequence of threshold values
T =ft
1
;t
2
;:::t
i
;t
i+1
;:::g to an imagef(x;y). A collection of nested connected components is
produced, each element in the collectionfR
1
;R
2
;:::R
i
;R
i+1
;:::g satisfying R
i
R
i+1
. With
algorithm parameter42 (0;jTj), R
i
is maximally stable when its variation, dened as
Var(i) =

R
i4
nR
i+4

=jR
i
j
has a local minimum at i. 4 controls the number of threshold levels that region needs
to remain stable to be considered maximally stable. Two additional parameters s
min
and
v
max
are used to prune regions: jR
i
j > s
min
, Var(i) < v
max
. To break components
with similar intensity values and close spatial proximity, we iteratively apply the algorithm
MSER(4;s
min
;v
max
) with increasingly stringent v
max
values. At the beginning of each it-
eration t, the algorithm has a maximal variation value v
t
max
. If some maximally stable
component has a size greater than a gating value s
max
, these components are subject to
another application of MSER with parameters (4;s
min
;v
t+1
max
). v
t+1
max
is obtained with some
> 0: v
t+1
max
= v
t
max
. The algorithm terminates when no component is greater than
s
max
or whenv
t
max
becomes negative. Figure 2.4A shows the results from anterograde tracer
26
injection site start cell detection.
Starter cell detection benets from appropriate imaging protocols. Importantly, pro-
longed exposure to excitation light diminishes contrast within the image, especially in gap
areas between densely packed starter cells. As the local intensity minimum becomes satu-
rated, two or more cells join into a thick line segment like structure. Since we design our
wavelet decomposition to lter away line segments from dendritic spines and axons will, at
some coarse scale such joined cells will be eroded. High resolution scans are also naturally
helpful. The increased pixel dimension of objects make them more robust to poisson and
gaussian noises common in microscopy.
For retrograde tracer injection sites, single cell extraction is not meaningful. Rather, it is
important to determine the injection center, the volume where tracer concentration peaks.
We use a similar wavelet decomposition and MSER approach. Since no single cell resolution
is required, stronger gaussian smoothing is applied toI
inj
(x;y), while the contrast amplifying
intensity adjustment step is dropped. The wavelet ltering sets to zero all detail coecients
from levels 1 through 3 and left the approximation coecients unmodied. Harsher stability
parameters are used in MSER to favor large components that maintain their sizes across
a wider range of threshold levels. Identied injection centers are shown with pink shade
in Figure 2.4B left column. Injection center pixels have intensity ranges well above signals
generated by tracer labeling or auto
uorescence (Figure 2.4B), sometimes saturating the
entire imaging bit width. A set of pixels adjacent to the injection center also accumulate
substantial intensity generated by the center due to the point spread function, as described
in chapter 2.2.2. Often very little connectivity information can be derived from this adjacent
pixel set, since its intensity is mainly explained by vicinity to injection center. We dene
this set as the injection periphery, and extract it with the identical method and parameters
for identical injection center, but using the entire image I(x;y) as input. Intuitively, we can
consider the injection center and periphery as a contiguous regions of exceptional brightness
with respect to I
inj
(x;y) and I(x;y) respectively. The injection periphery forms a volume
that envelopes the injection center (Figure 2.4B). The image partitions (injection site center,
27
periphery and exterior) have rather dierent intensity distributions. Their histograms and
their mean, median, mode values are shown in Figure 2.4B. Because our method of tracer
label segmentation (described in next section) uses image statistics to construct probability
distributions, the highly irregular intensity proles at the injection center and periphery
are detrimental to algorithm performance. We nd masking of extracted injection sites,
including both its center and periphery, greatly improves segmentation outcome.
28
Figure 2.4: Injection site analysis. (A) Injection site starter cell detection. Original image
and detected cells (green dots) are shown in the left gures. Intermediate results after wavelet
ltering and intensity transform are shown in right gures (B). Extracted injection site center and
periphery. The center is shown as pink, and periphery as yellow over the original image. Top row
charts show the histogram of the center, periphery and exterior respectively. Left most chart at
bottom row shows the mean, median and mode statistic of the three image partitions. The center
chart at the bottom row plots the average intensities of pixels at given distance into and away from
the border of injection center. Border of injection periphery is plotted as a blue dot. The right
chart at the bottom row plots the average intensities of pixels with given distance into and away
from the border of injection periphery. Border of injection center is plotted as a blue dot.
29
2.2.4 Tracer Segmentation
Segmentation of axonal bers from whole brain sections needs to take into consideration
high variance in true signal
uorescence intensity, saliency of image features at varying scales
and frequent occurrence of strong tissue auto
uorescence. To address these challenges, we
generate a belief map from both pixel intensity and local features in the linear gaussian scale
space, and a texture map by cross referencing the Nissl channel. Fiber pixels are segmented
by applying dual thresholds in both the belief map and the texture map.
In a microscopic scan of the mouse coronal plane, axonal bers appear as curvilinear
structures. In digital images, lines are characterized as rapid changes in the derivatives of
image intensity. Therefore the image second derivatives are widely used in line detection
and enhancement lters.
In the simplest 1D case, a line I is modeled as a gaussian prole:
f(x;
2
I
) =e

x
2
2
I

2
L
is the variance of the prole function and represents the physical scale of the 1D line.
The response R
I
of I to a line lter is the second derivative of I:
R
I
(x;
2
I
) =
d
2
f(x;
2
I
)
dx
2
=f(x;
2
I
)h
xx
whereh
xx
is the second derivative convolution kernel. Since second derivatives are sensitive to
noise, often a gaussian convolution smoothing operationg(x;
2
g
) is applied to the image prior
to dierentiation. By associativity of convolution, responses to line lter can be obtained as
R
f
(x;
2
f
;
2
g
) =f(x;
2
f
)g(x;
2
g
)h
xx
=
d
2
g(x;
2
g
)
dx
2
f(x;
2
f
)
Mathematically, the maximum line lter response occurs at x = 0, with
2
f
=
2
g
=2. By
applying a sequence of gaussian convolutions [g
0
(x;
2
g
0
);g
1
(x;
2
g
1
);:::g
i
(x;
2
g
i
):::] with ap-
propriate scale parameters, one can achieve multi-scaled responses at line structures [40]
30
[41][42].
The linear gaussian scale space is a framework to represent images at multiple scales. For
a 2D imagef(x;y) and a setT of non-negative scale parameters, its scale space representation
is a family of derived images dened by the convolution betweenf(x;y) and a gaussian kernel
G[43]:
fI(x;y;t);t2Tg =ff(x;y)
1
2t
e

x
2
+y
2
2t
;t2Tg
At t = 0, I(x;y; 0) is dened as f(x;y) itself. To reduce CPU and memory usage during
computation, we use the gaussian pyramids to approximate the scale space of f(x;y). Level
0 of the pyramid, I
0
(x;y), is the original input f(x;y). At pyramid level l> 0:
I
l+1
(x;y;
2
g
) =Subsample

I
l
(x;y;
2
g
)g(x;y;
2
g
)

We found both rate of changes in intensity and magnitude of intensity to be important
in ber segmentation. To re
ect these two aspects, at a given levell, we apply the Laplacian
lter to I
l
as a line lter:
r
2
(I
l
(x;y;
2
g
)) =
@
2
I
l
(x;y;
2
g
)
@x
2
+
@
2
I
l
(x;y;
2
g
)
@y
2
An image feature map F
l
is calculated as element wise multiplication of the image and
Laplacian of the image:
F
l
(x;y;
2
g
) =I
l
(x;y;
2
g
)r
2
(I
l
(x;y;
2
g
))
Let L be the set of all levels used in the scale space representation. We useA
L
to denote a
sequence of matrices (vectors) related by the pyramid levelsL, andA
l
to denote the matrix
(vector) at level l: A
L
= [A
0
;A
1
;:::;A
i
;:::];i2 L. By the construction process described
above we obtain a feature pyramidF
L
for an input image f(x;y).
We now dene some probabilistic events used to generate belief map for axonal bers:
(i;j)
+
jl: pixel (i;j) belong to a axonal ber in the level l image representation
31
(i;j)
+
: pixel (i;j) belong to a axonal ber in some level of the image representation
!
ij
(r)
+
jl:9(m;n)2!
ij
(r) = [i
r
2
;i +
r
2
] [j
r
2
;j +
r
2
] s:t: (m;n)
+
jl
We are ultimately interested in the belief mapB, withB
ij
approximating the probability of
the event (i;j)
+
. By law of total probability, we dene B as a linear combination of B
L
,
the beliefs at scales represented byL. Let
L
be the frequencies of axonal bers occuring
at the scale represented by the image pyramid levelsL:
P

(i;j)
+

=
X
l2L

l
ij
P

(i;j)
+
jl

=
X
l2L

l
ij
B
l
ij
=
X
l2L

l
ij
P

!
ij
(r)
+
jl

P

(i;j)
+

!
ij
(r)
+
;l

P

!
ij
(r)
+

(i;j)
+
;l

At each level l, by denition of the events, P

!
ij
(r)
+

(i;j)
+
;l

is equal to 1 and can
simply be dropped from the equation. P

!
ij
(r)
+

l

is estimated as the probability of
drawing a pixel from the image with feature value greater than the average feature value of
!
i;j
(r):
P

!
ij
(r)
+

l

=
Z
u
!
ij
(r)
1
D(t;
F
l;
2
F
l
)dt
u
!
ij
(r)
=
P
(m;n)2!
ij
(r)
F
l
m;n
r
2
whereD is a probability density function, whose parameters are estimated from
F
l and
2
F
l
.
P

(i;j)
+

!
ij
(r)
+
;l

is similarly estimated with feature values within !
ij
(r):
P

(i;j)
+

!
ij
(r)
+
;l

=
Z
F
l
ij
1
D(t;
!
ij
(r)
;
2
!
ij
(r)
)dt
Probabilistic estimations at levell yields a belief matrixB
l
#
with reduced dimensions, deter-
mined by the subsample ratio. Successive upsampling is applied to B
l
#
a total of l times to
generate a map B
l
with the same matrix dimensions as B
0
:
B
l
= 4g(x;y;
2
g
)Upsample(B
l
#
)
| {z }
l times recursively
32
AlthoughB
L
can be calculated, the values of
L
are dicult to know a priori. Various
factors in
uence
L
, including tracer type, anatomical locations of tracer injection and pro-
jection, imaging settings, etc. At each pixel location (i, j),
L
ij
is ajLj-dimensional Dirichlet
distributionD
ij
(). Because assuming a
atD
ij
() causes the segmentation results to lose
too much detail, we seek to yield a D
ij
() concentrated at a single level that suciently
represents the ber signal, with ner resolution outcomes favored over coarse resolutions.
For l 0 and some chosen > 1, we nalize B as
B
ij
=max
l2L
B
l
ij

1(B
l
ij
>B
0
ij
) + 1(l = 0)

Segmentation of cell bodies is a task of blob detection, subject to size and morphology
constraints. We generate a pixel belief map B similarly as described for axonal bers, but
without using multi-scale representation:
B
ij
=P

(i;j)
+

=
P

!
ij
(r)
+

P

(i;j)
+

!
ij
(r)
+

P

!
ij
(r)
+

(i;j)
+

while the feature map is generated with:
F (x;y;
2
G
) =

f(x;y)g(x;y;
2
g
)

2
To re
ect the spatial clustering of foreground pixels in blobs, B is then convolved with an
averaging kernel h(r
c
) with radius r
c
:
B Bh(r
c
)
Within a tissue section, both the extra-cellular matrix and cellular contents emit auto
u-
orescence when excited by suitable wavelengths [44]. In our datasets, auto
uorescence can
be considered as image texture and contributes to noise during segmentation. To quantify
33
the amount of variance in an image explained by auto
uorescence, we compute a texture
map T , dened by the Pearson correlation coecients between pixel neighborhoods of two
images f(x;y) and g(x;y):
T
ij
=
cov(f
w
ij
(r)
;g
w
ij
(r)
)

f
w
ij
(r)

g
w
ij
(r)
=
E[f
w
ij
(r)
g
w
ij
(r)
]E[f
w
ij
(r)
]E[g
w
ij
(r)
]

f
w
ij
(r)

g
w
ij
(r)
f
w
ij
(r)
=ff(m;n) : (m;n)2w
ij
(r)g; g
w
ij
(r)
=fg(m;n) : (m;n)2w
ij
(r)g
In the anterograde tracer channels, we use input tracer image asf(x;y) and the Nissl channel
image of the same section as g(x;y). For ideal images f(x;y) and g(x;y) with no auto
uo-
rescence, since the
uorescence arise from two dierent markers with no intrinsic tendency of
co-localization, we expect to see T
ij
with small magnitudes. In the presence of
uorescence,
regions of f(x;y) and g(x;y) where signicant pixel intensities are generated by auto
uo-
rescence, magnitudes T
ij
will be closer to 1. For a retrograde tracer image I(x;y), we found
the following f(x;y) and g(x;y) generate good texture map:
f(x;y) =I(x;y)g(x;y;
2
g
); g(x;y) =f(x;y)h(r)
where h(r) is an averaging kernel with radius r.
With two chosen threshold valuest
B
andt
T
, a tracer image segmentationS is dened as:
S
ij
= 1(B
ij
t
B
) 1(T
ij
t
T
)
The retrograde segmentations are post-processed to impose morphological constraints: (1)
lling holes and convex decits in segmented objects (2) separating touching objects with
watershed algorithm and (3) removing objects that are small or eccentric. It is possible to
use the segmented axonal ber as input to a tract tracing algorithm. However, our image
data are collected with 0:495 m/pixel in the x, y axis, and have sampling gaps of 200 m
along thez axis. These resolution parameters are inappropriate for meaningful tract tracing.
Therefore we base our quantication on segmented pixels instead.
34
Figure 2.5: Tracer Segmentation. (A) Anterograde tracer segmentation result. The quality
of segmentation rapidly improves in a narrow threshold range (between 45 and 50 in this example),
corresponding to the sharp peak in SNR. (B) Auto
uorescence removal by texture map. (C)
Retrograde tracer segmentation result. (D) Signal to noise ratio, mean foreground intensity and
number of foreground pixels are plotted as functions of belief map threshold value, averaged over
98 images. (E) Multi-scale segmentation captures bers with a wide range of physical dimensions.
35
Signal to noise ratio (SNR) of the segmentation was quantied in 98 anterograde tracer
images. We use the SNR denition given by NEMA (The National Electrical Manufac-
turers Association)[45, 46], calculated as SNR
fb
=

foreground

background
. An alternative calculation
with SNR
ff
=

foreground

foreground
is simultaneously looked at. Three probability distribution, the
gaussian distributions, the exponential distribution and the Rayleigh distribution is plugged
into the algorithm as the presumed distribution followed by the feature map for comparison.
Parameters for each distribution are estimated from the mean (
tissue
) and variance (
2
tissue
)
of the feature map within the segmented tissue mask (see chapter 2.2.2).
The probability density function (PDF) of gaussian distribution probability is
f(x;;
2
) =
1
p
2
2
e

(x)
2
2
2
, with ^ =
tissue
,
b

2
=
2
tissue
PDF of exponential distribution is
f(x;;b) =e
(xb)
, with
^
=
1

,
^
b =
tissue

1
^

PDF of Rayleigh distribution is
f(x;;b) =
x

2
e

(xb)
2
2
2
, with
b

2
=
r
2
4

2
tissue
; ,
^
b =
tissue

r

2
^
Both average SNR
fb
and SNR
fb
values show a sharp peak for all three distributions as
the threshold value t to the belief map increases (in range 1 to 100), and remains steady
over a wide range. The average intensities of segmented foreground tissue and number of
foreground pixels also show non-linear rapid change at SNR peak. The eect on an image is
shown in Figure 2.5A. Segmentation results rapidly improved within a narrow t value range
[45; 50], and remained representative of original data as t value further increases (result of
t = 80 is shown). Figure 2.5B shows typical results from texture map threshold, useful
for images with low signal intensity and high level of auto
uorescence. Figure 2.5C shows
detected cells from a retrograde tracer image. We also show the capability of our multi-scale
36
segmentation method to capture dense ber bundles (Figure 2.5E).
Tracer segmentation is employed as an component of Outspector. A set of default belief
and texture map threshold values is provided for the user, which empirically give good results
on most images. Numeric values of the belief and texture maps are computed in parallel on
the high performance computing cluster, and stored on the distributed le system. Research
can access these maps, and perform interactive parameter adjustment with side-to-side view
of original data and segmentation result from Outspector when necessary.
2.2.5 Statistical Model of Brain Connectivity
Whole brain neuroanatomical tracing data production, collection and quantication is a
long process, where completion time of each experiment spans 4 6 weeks. Numerous wet
lab and informatics processes involved are expected to introduce random noise to measured
outcomes, on top of the intrinsic biological variability among experiment animals. To name
a few, number of start cells in anterograde injections and volume of tracer in retrograde
injections, delivery time, concentration and uptake eciency of the tracer, quality of the
transcardial perfusion and immuno-staining, integrity of the brain sections and their agree-
ment with the atlas, sampling frequency and location of the coronal plane and microscopic
imaging settings are all factors that in
uence the amount and strength of signals eventually
observed. In order to meaningfully compare data from dierent individuals, it is necessary
to normalize at least some of these non-biological variations. A few factors can be accounted
for explicitly. For example, the injection site can be quantied (chapter 2.2.3) and serve as
a normalizing constant [47]. Staining quality may be expressed as a function of peak signal
to noise ratio during segmentation. However, the exact eects on quantication outcome
from many aspects mentioned above remain unclear. Noting that sections within a specimen
brain is subject to consistent experiment conditions, we use a self-normalization approach
to obtain whole brain connectivity fractions within each brain specimen [48].
The ARA denes 688 gray matter regions. To analyze the ipsilateral and contralateral
brains separately, the number of distinct regions is doubled to 1376. Let us refer to this
37
number as n
ARA
. Given a specimen brain with sections sampled at a set of ARA levels L,
we generate an initialjLj 1376 matrixC of connectivity counts data. The matrix entryC
lr
is the quantied amount of connectivity (pixel and cell count for anterograde and retrograde
tracers respectively) in anatomical region r at level l. A length 1376 count vectorn can be
generated by pooling connectivity within the same region across levels
n
r
=
X
l2L
C
lr
n is an un-normalized raw data vector. The self-normalized length n
ARA
fraction vectorf
is calculated as
f =
n
knk
1
where f
r
, the r
th
entry in the vector, characterizes the proportion of connectivity that falls
into a target/source region r out of all axons originating from/projecting into the injection
area.
We additionally dene a connectivity descriptor vectord that approximates the projec-
tion density into target regions. This is done by tracking a volume vectorv for target regions
sampled in the specimen
v
r
=
Z
z
s
r
(z)dz'
X
l2Lr
s
r
(l)
where L
r
L is the sampled levels where region r is present, s
r
(l) is the area of region
r's cross section with the x-y plane at level l, and the constant is section thickness (for
convenience we use = 1). Element-wise division off andv givesd
d =f (v +)
This density calculation has a implicit total intensity normalization across all brain spec-
imens, removing the eects of some experiment condition variations such as injection site
size, tissue staining and imaging quality. It should be interpreted as the projection density
of each specimen, assuming all specimens contain the same amount of positive segmented
38
pixels. Density measure provides an intuitive read into the connectivity data, since pro-
jection densities have been widely considered in neuroanatomical literature, and are easily
appreciable by human experts. The count, fraction and density vectors from 18 anterograde
injections in mPFC are composed into three 181376 data matrices (151376 for retrograde
injections), the count matrix N, the fraction matrix F and density matrix D. In N, the
entryN
i;j
corresponds ton
j
of thei
th
brain specimen. In other words, rows and columns of
F represent brain specimens and gray matter regions as dened by ARA respectively (see
chapter 3.3.1 experiment evidence based modication to ARA). F and D's layout is identi-
cal to N. Although based on the same underlying collection of experiments, there is subtle
dierence in the information content of F and D. The fraction matrix describes how the
injection site region interacts with the brain. The density matrix carries more information
about regions connected with the injection site. Viewed on its columns, density matrix in
anterograde tracing experiments reveals the dierent degrees of in
uence of mPFC cortices
on their target regions, while in retrograde tracing experiments it implies how important
each mPFC cortices are as projection targets for their upstream regions.
From a probabilistic point of view, the whole brain structural projectome at a given
spatial locationx is dened by a non-negative probability vectorp
x
, such that
P
r
p
x;r
= 1,
and the expected frequency of observing projection from x in region r is equal to p
r
. For
each neural tracing experiment with injection atx (x is a vector of spatial coordinates), its
associated count vector n
x
is then a multinomial random variable whose probability mass
function is
f (n
x
;S;p
x
) =S!
Y
r
p
nr
x;r
n
x;r
!
where S =
P
r
n
x;r
. Conducting tracing experiments at x in k animals gives a collection
of multinomial random variables N
x;k
=fn
(1)
x
;n
(2)
x
;:::;n
(k)
x
g, with the unknown p
x
to be
estimated. When k = 1, the f vector we constructed above is already equal to p
x
. For
39
k 2, the log likelihood function forp
x
is
L(p
x
jN
x;k
) =
X
k
logf (n
(k)
x
;S
(k)
;p
x
) =
X
k
logS
(k)
! +
X
k
X
r
n
(k)
x;r
logp
x;r

X
k
X
r
logn
(k)
x;r
Posing the constraints
P
r
p
x;r
= 1 with Lagrange multiplier
L(p
x
;) =L(p
x
jN
x
) +(1
X
r
p
x;r
) =
X
k
X
r
n
(k)
x;r
logp
x;r
+(1
X
r
p
x;r
) +c
The maximum likelihood estimation (MLE) ofc p
x
=argmax
px2
L(p
x
jN
x
) is obtained by
@L(p
x
;)
@p
x;r
= 0 =)d p
x;r
=
P
k
n
(k)
x;r

@L(p
x
;)
@
= 0 =) =
X
r
X
k
n
(k)
x;r
The simple closed form ofc p
x
oers an easy solution to infer connectivity atx from mul-
tiple animal experiments. Segmented signals are added for all experiments, and normalized
by a grand sum. For multi-experiment datasets,N,F andD matrices from a total ofk brain
specimens are rst constructed. Denote the injection regions associated with each matrix
row as ak-dimensional vector!, and the set of unique elements of! as
. Iterating through
contents of
, for

i
, thei
th
element encountered, thei
th
row of thej
jn
ARA
sample count
matrix
b
N and sample fraction matrix
b
F are calculated as
b
N
i;
=
X
j:!
j
=

i
N
j;
This step poolsk brain specimens with common tracer injection location

i
and observations
N

i
;k
=fn
(1)

i
;n
(2)

i
;:::;n
(k)

i
g with numbers of trialsfS
(1)

i
;S
(2)

i
;:::;S
(k)

i
g into the aggregate
specimen
b
N
i;
with number of trial
P
k
S
(k)

i
. The aggregate fraction matrix, whosei
th
row is
40
the MLE c p

i
, is given by
b
F
i;
=
b
N
i;

b
N
i;

1
Connectivity dierences between two brain regions

i
and

j
is re
ected by the dierences
of
b
F
i;
and
b
F
j;
.
To calculate sample density matrix
b
D, target region volumes in dierent brains are used
as well. Denote V as the volume matrix, composed from volume vectorsv
b
D
i;
=
P
j:!
j
=

i
F
j;
P
j:!
j
=

i
V
j;
Row vectors in these sample matrices are denoted similarly as b n,
b
f and
b
d. Due to the
mathematical property of the fraction vectors n and b n , we mostly present, describe and
interpret our mPFC connectivity data with from the fraction matrix. We do derive biological
intuitions and insights from density matrix in a qualitative (occasionally quantitative as well)
manner.
Another operation of interest is to combine matrix entries according to some anatomical
region grouping rules (we described two possible rules in the subsection immediately above).
This is done by rst dening a mapping from the original ARA regions to a set of region
groups:
M :r7!g; r2R;g2G
where R is a set of anatomical regions and G is a set of region groups. For example, when
we want to combine all regions according to our 8 major hierarchical groups, we would have
for ipsilateral ILA layer 5: M(ILA 5 i) = CTX i. The count and volume vector,n
M
and
v
M
of lengthjGj over the sampled levels set L are calculated as
n
M;g
=
X
r:M(r)=g
n
r
=
X
r:M(r)=g
X
l2L
C
lr
41
v
M;g
=
X
r:M(r)=g
v
r
'
X
r:M(r)=g
X
l2Lr
s
r
(l)
Under region grouping ruleM, the matrices N
M
, F
M
, D
M
, N
M
, F
M
, D
M
can then be
calculated exactly as when no grouping rule is present.
2.3 Conclusion
To address the myriad of challenges in quantitative analysis of structural connectivity
data of the brain, we develop a comprehensive image processing and data analytical pipeline
and its accompanying software suite Outspector. Our goal is to leverage large parallel
computing infrastructures for the kind of high throughput processing required for massive
scale projects such as the Mouse Connectome Project. To assure high quality quantication,
we also incorporate user interaction as one of the design principles of Outspector, and provide
researchers the capability to ne tune outcomes from automatic algorithms. This combined
high-throughput and ne-tuning approach generates annotated connectivity data in a reliable
and quantitative manner from terabyte range raw microscopic imaging data.
In order to normalize data over experimental condition variations, we quantify the connec-
tivity patterns in each brain as self-normalized quantities: fractions of connectivity observed
in particular brain regions over the total amount of connectivity seen in the brain. Results
from multiple brains are combined with maximum likelihood estimation. The traditional
notion of connectivity densities are also quantied through normalizing fraction data over
sampled brain region volumes. A natural consideration that arises with using fractions as a
connectivity descriptor is that the "strength" of connectivity is not encoded: for anterograde
tracing data for example, we can not know if a region sends massive or only moderate amount
projections across the brain by looking at its fraction vector. However, with a collection of
experiments covering the brain, we can leverage information that lies in the relationship of
connectivity between brain regions, and the value of projection strength. Brain connectivity
can be seen as a graph [49][50]. In our connectome data, each brain regionv is a node of the
graph G(V;E), where weighted edges e in and out of the node are inputs and projections
42
of the region. Let us use the anterograde tracing dataset to frame the nature of graph G
probabilistically. Suppose we have the projection and input fraction matrix F
out
and F
in
,
both containing a row for each of then nodes of the graphG: V =fv
1
;v
2
;:::;v
n
g. In the pro-
jection data, the value F
out
ij
is the conditional probability of observing projection from v
i
to
v
j
, given projection fromv
i
to any node inV has occurred: F
out
ij
=P
out
(v
i
!v
j
jv
i
). We are
interested in the joined probabilityP
out
((v
i
!v
j
)^v
i
) (can be simplied toP
out
(v
i
!v
j
)),
which provides us the strength of the projection from v
i
to v
j
in the context of the entirety
of V . However F
out
does not give us P
out
(v
i
). Instead, we can use the fact that any input
connection received by v
j
must be provided by projections from a subset of V, and write
down a relation linkingF
out
,w
out
= [P
out
(v
1
);P
out
(v
2
);:::;P
out
(v
n
)] and the input probabil-
ities to nodesw
in
= [P
in
(v
1
);P
in
(v
2
);:::;P
in
(v
n
)]. To begin with, we establish P
in
(v
j
) for
any j2f1; 2;:::;ng:
P
in
(v
j
) =
n
X
i=1
P
out
(v
i
!v
j
) =
n
X
i=1
P
out
(v
i
!v
j
jv
i
)P
out
(v
i
)
In matrix form, this is equivalent to
(F
out
)
T
w
out
=w
in
The system of equation has 2n unknowns inw
out
,w
in
andn equations. The input fraction
matrix F
in
provides an additional n equations by identical reasoning:
(F
in
)
T
w
in
=w
out
The coecient matrix

F
out
F
in

has rank 2n2< 2n due to each row ofF
out
andF
in
summing
to 1. We supply two additional constraints:

w
out

1
= 1 and

w
in

1
= 1. Then the least
43
squares solutions to
0
B
B
B
B
B
B
B
@
(F
out
)
T
I
I (F
in
)
T
1
T
n
0
T
n
0
T
n
1
T
n
1
C
C
C
C
C
C
C
A
0
B
@
w
out
w
in
1
C
A
=
0
B
B
B
B
@
0
n
1
1
1
C
C
C
C
A
provides the strength of input and output at each node ofV , and with it speciesP
out
(v
i
!
v
j
) P
in
(v
i
v
j
) for all pairs of v
i
;v
j
2V (0
n
and 1
n
denote length n all 0 and all 1 vector
respectively). This provides the strength of every single structural connectivity pathway in
the brain.
We note that with complete F
out
and F
in
, covering all brain regions of interest, we are
able to fully specify all connections in the brain together with their magnitude, without
the necessity to normalize for injection start cell number or tracer spread volume. This
is a large scale yet incredibly exciting endeavor that will quantitatively layout the entire
wiring diagram of the brain. The Mouse Connectome Project is particularly well positioned
for the task, beneting from its large data collection and high quality data processing and
analytics. We expect our nal results to greatly facilitate the understanding of complex
neural circuitries, provide biological foundations for computational models of the brain, and
drive testable hypotheses on the functions of connectivity pathways.
44
Chapter 3: Whole Brain Connectivity of Mouse mPFC
3.1 Introduction
3.2 Material and Methods
3.2.1 Tracers and Stereotaxic Surgeries
Surgical and microscopic imaging procedures have been described previously [15, 51].
All animals used in this study are 8-week-old male C57BL/6J mice (Jackson Laboratories).
Animals were housed in a temperature (21 - 22

C), humidity (51%) and light (12 hour light
and 12 hour dark cycle, with lights on at 6:00 am and o at 6:00 pm) controlled room. Food
and water were given ad libitum. Upon arrival at the facility, the animals were allowed a
minimum of 1 week to adapt to the housing environment before surgeries were performed.
All experiments were conducted according to the standards set by the National Institutes of
Health Guide for the Care and Use of Laboratory Animals and the institutional guidelines
of the University of Southern California (USC).
Tracers were injected into the following medial prefrontal areas: ILA, PL, ACAd, ACAv,
DP supercial layers, DP deep layers. Stereotaxic coordinates of targeted injection cen-
ters were determined through the Allen Reference Atlas and empirically adjusted when
needed. Anterograde tracers included phaseolus vulgaris leucoagglutinin (PHAL; 2.5%; Vec-
tor Laboratories) and adeno-associated viruses encoding enhanced green
uorescent pro-
tein (AAV-GFP; AAV2/1.hSynapsin.EGFP.WPRE.bGH; Penn Vector Core) and tdTomato
(AAV1.CAG.tdtomato.WPRE.SV40; Penn Vector Core) and biotinylated dextran amine
(BDA; FluoroRuby, 5%; Invitrogen). Retrograde tracers included cholera toxin subunit b
(CTB AlexaFluor 647 conjugate, 0.25%; Invitrogen) and Fluorogold (FG; 1%; Fluorochrome,
LLC). At each targeted coordinate, an anterograde tracer and a retrograde tracer are co-
injected to simultaneously reveal the input and output pathways [52]. Tracer combinations
included PHAL/CTB, BDA/FG and AAV/FG.
Surgeries for tracer injections were performed under iso
urane anesthesia (Hospira, Inc.).
45
Mice were initially anesthetized in an induction chamber primed with iso
urane and were
subsequently mounted to the stereotaxic apparatus where they were maintained under anes-
thetic state via a vaporizer (Datex-Ohmeda). The iso
urane was vaporized and mixed with
oxygen (0.5 L/min) and nitrogen (1 L/min). The percent of iso
urane in the gas mixture was
maintained between 2 and 2.5 throughout the surgery. Tracers were delivered iontophoret-
ically using glass micropipettes whose outside tip diameters measured approximately 1030
m. A positive 5Amp, 7-second alternating injection current was delivered for 10 minutes
(Stoelting Co.). The analgesic buprenorphine was administered the day of and day following
surgeries (0.05 mg/kg).
3.2.2 Tissue Preparation and Immunohistochemistry
Animals were sacriced with an overdose injection of sodium pentobarbital (6 mg/kg) 7
days or 14 days (when AAV was injected) following surgeries. Each animal was transcardially
perfused with approximately 50 ml of 0.9% NaCl followed by 50 ml of 4% paraformaldehyde
solution (PFA; pH 9.5). The brains were post-xed in 4% PFA for 24 48 hours at 4

C
after which they were embedded in 3% Type I-B agarose (Sigma-Aldrich) prior to sectioning.
Four series of coronal sections were sliced at 50 m thickness with a Compresstome (VF-
700, Precisionary Instruments, Greenville, NC) and prepared for processing. One of the four
section series was immunostained for imaging. We selected the series that contained coronal
level 53 of the Allen Reference Atlas to maintain similar section level distributions across
experiments.
Sections were stained for PHAL in animals where PHAL was injected using the free-

oating method. Brie
y, sections were transferred to a blocking solution containing normal
donkey serum (Vector Laboratories) and Triton X-100 (VWR) for 1 hour. Following three
5-minute rinses, sections were incubated in a KPBS solution comprised of donkey serum,
Triton, and a 1:1000 concentration of rabbit anti-PHAL antibody for 4872 hours at 4

C
(Vector Laboratories, #AS-2300). Sections were rinsed three times in KPBS and then soaked
for 3 hours in the secondary antibody solution, which contained donkey serum, Triton, and
46
a 1:500 concentration of anti-rabbit IgG conjugated with AlexaFluor 488 or 647 (Invitrogen,
488: #A-21206, 647: #A-31573). Sections were again rinsed with KPBS three times. All
selected section series were counterstained with a
uorescent Nissl stain, NeuroTrace 435/455
(NT; 1:500; Invitrogen, #N21479). The sections were then mounted and coverslipped using
65% glycerol.
3.2.3 Image Acquisition and Processing
All sections were scanned as high-resolution virtual slide image (VSI) les using an Olym-
pus VS110 high-throughput microscope. Post-acquisition processing methods, including im-
age registration, tracer segmentation and quantication, were described in detail in chapter
2 of this thesis. Brie
y, all images were
ipped when needed to have the injection side of
brain appear in the right half of the image in VS-Desktop software accompanying the mi-
croscope. Images were then imported into Outspector as a project and processed on a high
performance computing cluster. A rst pass fully automatic registration was executed, its
results inspected and ne-tuned when needed. Tracer segmentation was performed on reg-
istered images with default parameters. Segmentation results were also inspected, followed
by interactive parameter adjustment if needed. Segmented pixel counts and cell counts in
each brain region at all specimen section levels were generated and used for analysis.
3.3 Results
3.3.1 Denition of DP
In the Allen Reference Atlas, several mPFC areas have been well delineated, including,
from the dorsal to ventral, the ACAd, ACAv, PL, and ILA. In addition, one small cortical
area located ventral to ILA and dorsal to TTd (dorsal TT) were dened as the dorsal
pedundular area (DP). Nevertheless, it still remains controversial whether the DP should be
dened as a distinct cortical area or part of the ILA[14, 53], and very little is known about
connectivity of DP [54{56]. Cyto-architecturally, DP has several interesting characteristics.
47
Its cortical layer 1 is substantially thicker than other mPFC cortices and TTd. Also of
intrigue is the unique juxtaposition of DP between the classically dened neocortex (in six
layers) and paleocortex (in three layers). This unique anatomical position suggests the DP
serves as an interface to mediate interactions between these two functional systems. These
observations motivated us to systematically examine the input/output connectivity of the
DP and compare it with that of the rest mPFC areas.
Taking advantage of the large data collection of the Mouse Connectome Project (ht
tps://www.mouseconnectome.org), we were able to identify multiple brain regions whose
connections with DP discerns DP from both mPFC and TTd. For each of those regions, we
were also able to verify their connections through brain samples with tracer injections in DP
(Figure 3.1).
48
Figure 3.1: Unique Connectivity Pathways of DP. (A) Top: Anterograde tracer injection
at AUDv labels layer 1 of DP. Bottom: Retrograde tracer injection at DP labels layer 5 of AUDv.
(B) Top: Anterograde tracer injection at PIR labels layer 1 of DP. Bottom: Retrograde tracer
injection at DP labels PIR. (C) Top: Retrograde tracer injection at ACA labels supercial layers
of DP. Bottom: Anterograde tracer injection at DP labels supercial and deep layers of ACAv,
ACAd and MOs. (D) Top: Anterograde tracer injection at ENTl labels layer 1 of DP. Bottom:
Retrograde tracer injection at DP labels ENTl. (E) Top: Anterograde tracer injection at RE labels
layer 1 of DP. Bottom: Retrograde tracer injection at DP labels RE. (F) Top: Retrograde tracer
injection at LHA labels DP. Bottom: Anterograde tracer injection at DP labels LHA.
We report that DP receives signicant input from the ventral auditory area (AUDv) as
well as the piriform cortex. To the best of our knowledge, DP is the only cortical region
to be heavily innervated by both the auditory cortex and the olfactory cortex. These two
connectivity pathways suggest the participation of DP in sensory information integration.
Nucleus reunions (RE) of the thalamus is another major source of aerent inputs to the
DP. Classic neuroanatomical studies have established that RE is important in information
49
transfer from mPFC to the hippocampal formation (HPF) [57, 58][59, 60]. The substantial
RE to DP projection reported here, together with the known extensive limbic and limbic-
associated aerents to RE [61] greatly expand the complexity of neural circuitry between
mPFC, RE and HPF. A further source of rich DP innervation is the entorhinal area lateral
part (ENTl). The ENTl projects robustly to the ILA as well. However, the DP receives
inputs predominately from ENTl layer 2, while projections to the ILA predominantly arise
from neurons in the ENTl layer 5. These layer specic connectivity suggest that the DP plays
a dierent role from the ILA in cognitive functions such as spatial memory and navigation
[62{64].
We also report that DP displays distinct projection patterns into other mPFC regions.
The caudal ventral DP, potentially a sub-domain within DP, heavily innervates layer 1 and
the deep layers of ACAv, ACAd, caudal ILA, caudal PL and MOs, allowing DP to exert
broad in
uence over the mPFC. Additionally, DP generates prominent projections to the
lateral hypothalamus, a region involved in feeding and homeostasis.
Based on these newly identied connectivity pathways, we consider DP as a distinct
brain region. In addition, we generated updated levels (37 41) of the ARA atlas, with
boundaries of DP, ILA and TTd redrawn based on anatomical tracing data described above
(Figure 3.2A). With the updated boundaries, we calculated cortical layer 1 thickness for
mPFC cortices from ARA levels 37 41. On each level l, denote the curve intersections
of cortical layer 1 pial and inner surfaces with the x-y plane C
l
pial
(r
l
(u)) and C
l
inner
(r
l
(v)),
whereu;v2 [0; 1]. The cortical thickness is calculated as distance between the surfacesS
pial
and S
inner
:
50
(S
pial
;S
inner
)
X
l2L
L(C
l
pial
)(C
l
pial
;C
l
inner
)
X
l2L
L(C
l
pial
)
=
X
l2L
Z
1
0
argmin
v2[0;1]

r
l
(u) r
l
(v)

2

r
0
l
(u)

du
X
l2L
Z
1
0

r
0
l
(u)

du
We obtain cortical layer 1 thickness for ACAd, PL, ILA, DP and TTd as 146.89 m,
160.64 m, 132.98 m, 209.15 m, 107.56 m respectively (ACAv is not present at ARA
levels 37 41). Kruskal-Wallis test on cortical thickness at points along C
inner
showed
highly signicant dierences among the 5 anatomical regions (H(4) = 8137:11, p = 0:00).
Figure 3.2B shows the thickness distribution and bootstrapped condence intervals for me-
dian thickness. These statistical ndings support our DP delineation with cyto-architecture
evidence.
Additionally, we observe dierences in input/output patterns between the supercial and
deep layers of DP. Among others, DP-thalamic projections mainly originate from its deep
layers, while DP-mPFC projections mainly arise from its supercial layers. In the ensuing
sections, we discuss the inputs and outputs to the following mPFC regions: ILA, PL, ACAd,
ACAv, DP supercial (DPs), DP deep (DPd).
51
Figure 3.2: Updated Boundaries of Dorsal Peduncle. (A) ARA levels 33 41 where DP
is present are shown. Supercial and deep layers of DP are colored red and green respectively. DP
boundaries are updated on ARA levels 37 41 (second and third row). ARA original delineation of
DP is shown as a dark shade (ARA level 41 did not delineate DP). (B) Boxplot of mPFC cortices
layer 1 thickness along its inner surface. Within each group, the orange line and notch in the box
represent data median and bootstrapped 95% condence interval for the median. The box extends
from lower to upper quantile of the data. The whisker extends from Q1 1:5IQR toQ3 + 1:5IQR.
52
3.3.2 Stereiotaxic Tracer Injections in mPFC
In total, section series from 18 anterograde and 15 retrotrade injections in ILA, DPs,
DPd, PL, ACAv, ACAd were processed through our pipeline and quantitatively analyzed
for this study. Anterograde and retrograde experiments generated a total of 752 and 604
coronal sections respectively (Figure 3.3B). To our knowledge, this is the most comprehensive
medial prefrontal cortex anatomical connectivity dataset in mice to date. The amount of
connectivity observed across all atlas levels are plotted for each of the mPFC cortices (Figure
3.4). All parts of mPFC display wide ranging connections with the rostral-caudal extent of
the brain. Proximity to the injection site alone is not informative on amount of connection
observed, suggesting functional signicance of the connectivity level distributions. Some
aspects of connectivity pattern dierences between mPFC regions are also re
ected. For
instance, despite the fact that it is often considered part of ILA, both the projection and
input distributions of DPs are noticeably dierent from those of ILA. These results are an
example of the greater richness of information under our quantitative and systematic analysis
framework over traditional symbol annotation based frameworks.
For each of the 33 mPFC injections, their surgical case ID (used to uniquely identify an
animal), anatomical locations and tracer type are shown in Figure 3.3A. Atlas levels 38
41 have updated boundaries for ILA, DP and TTd, as discussed in Chapter 3.3.1. Figure
3.3B displays a coloring scheme for the 6 mPFC cortices, which we will be consistently using
through out this thesis.
53
Figure 3.3: Stereotaxic Injections in mPFC. (A) Drawings of injection center locations for
tracing experiments used in this study over the Allen Reference Atlas. The atlas wire frame for
levels 38, 39, 40, 41 have updated DP boundaries. (B) Color representation of the 6 mPFC cortices
used in (A). Number of brain specimens and sections available per mPFC area is summarized in
the table.
54
Figure 3.4: Connectivity Distributions across ARA Levels. (A) and (B) Distributions of
mPFC whole brain projections and inputs. Vertical lines indicate the ARA levels of injection site
centers
3.3.3 Overview: Connectivity of mPFC Cortices
Anatomical region groupings and notations: We organize the 1376 ARA grey matter re-
gions into 8 major groups according to their anatomical hierarchies: cortex (CTX), thala-
mus (TH), hypothalamus (HY), striatum (STR), amygdalar complex (AMY), hippocampal
formation (HPF), midbrain (MB) and hindbrain (HB), and discuss connectivities between
55
mPFC cortices and these major groups in separate sections. To present data in an intuitive
and digestible manner, we also group cortical layers as supercial and deep parts. For 6
layered neocortical structures, layers 1 through 3 are grouped as supercial, and layers 4
through 6 as deep. For 3 layered paleocortical structures, layers 1 and 2 are grouped as
supercial, and layer 3 as deep. A table listing anatomical regions and their groupings is
provided in the appendix (Table 5.3). For brevity, we use a couple of notations: (1) The let-
ters "i" and "c" are used to denote ipsilateral and contralateral side of the brain respectively.
They appear appended to an anatomical region name. For example, "CP i" and "ILA 5 c"
refer to ipsilateral CP and contralateral ILA layer 5. (2) The letters "d" and "s", when
appearing after a cortical region, are used to denote deep and supercial layers of the cortex
respectively. The "d/s" notation precedes the "i/c" notation. For example, "ILA s i" refers
to ipsilateral ILA deep layers.
Figure 3.5 shows a high level view of mPFC whole brain projection patterns to 8 hierar-
chical groups (Table 5.3). Between 70 80 % of outputs from ILA, DPd, PL, ACAv and
ACAd innervate the ipsilateral brain. DPs output has a higher ipsilateral concentration at
92% (Figure 3.5 and Figure 3.5B). All mPFC areas have substantial axonal ber distribu-
tions to the cortex, while the hippocampal formation receive very little input from all 18
mPFC anterograde injections studied. Other hierarchical groups receive varying levels of
projections from mPFC. These results are consistent with earlier ndings in other species,
including rats and macaque monkeys [6, 65]. When total volume of all 8 groups are ac-
counted for, the cortex is relatively sparsely innervated, while the hypothalamus, striatum,
midbrain, amygdalar complex and thalamus receive denser mPFC input (Figure 3.5D). In
all 6 mPFC areas, input distributions are heavily concentrated in the cortex. STR, HY, MB
and HB generate little input to mPFC. Inputs from HPF are substantially greater in ILA
and DPd than in other regions. TH and AMY project to mPFC areas at varying amounts
(Figure 3.6A) from dense populations of cells.
We perform a hierarchical 2D clustering with the mPFC anterograde whole brain pro-
56
jection data F
global
and D
global
to observe if injections into the same anatomical location
yield similar whole brain projections. Hierarchical clustering along both axes of F (or D)
re-organizes its rows and columns, such that brain specimens with similar whole brain pro-
jection patterns, and target brain regions with similar input patterns from mPFC are shued
close to each other. Our data is in a 1376-dimensional space, making the cosine distance
a suitable similarity measure. The cosine distance is frequently used in text mining and
information retrieval, where data are distributed in high dimensional space with semantic
embeddings. In the context of our connectivity matrix F (or D), its semantics lies in the
brain wiring patterns that it describes, which form the foundations of complex physiological
and cognitive functions of an organism. The cosine distance between two vectorsu andu
0
is dened as

cos
(u;u
0
) = 1
uu
0
kuk
2
ku
0
k
2
= 1
P
i
u
i
u
0
i
p
P
i
u
2
i
p
P
i
u
02
i

cos
(u;u
0
) measures angular dierences, and carries no information on the magnitude of
vector norms. However the row and column norms ofF also contain meaningful information
regarding strengths of connectivity. Therefore we weigh the cosine distance with the ratio
in total connectivity betweenu andu
0
(u;u
0
) =
max (kuk
1
;ku
0
k
1
)
min (kuk
1
;ku
0
k
1
) +

cos
(u;u
0
) =
max (
P
i
ju
i
j;
P
i
ju
0
i
j)
min (
P
i
ju
i
j;
P
i

u
0
i

) +

1
P
i
u
i
u
0
i
p
P
i
u
2
i
p
P
i
u
02
i

On the global anterograde fraction matrix,
cos
(u;u
0
) and (u;u
0
) have no in
uence on
distance measures between brain specimens, represented by F
i;
(row vectors of F ). By
denition,

F
i;

1
= 1. For grouping target anatomical regions which receive similar input
patterns from across mPFC, (u;u
0
) gives more meaningful results. In a simple 2 dimen-
sional example, suppose region A receives 20% of ILA's total projections and 20% of PL's
total projections, while region B receives 1% and 1% of total projections from ILA and PL
respectively. The cosine distance between A and B is zero, clearly not a reasonable inter-
pretation. Distances between row and column vectors in D are calculated using
cos
(u;u
0
),
since the division between fraction and volume removes interpretable additivity from vector
57
elements.
Clustering by F
global
and D
global
both show higher intra- than inter-group projection
similarities (Figure 3.5C and Figure 3.5D). With 3 injections each, ILA, PL, ACAd and
ACAv form triplet leaves in the hierarchy tree. All 4 DPs injections cluster together as well.
The 2 DPd injections are exceptions, with one DPd case producing more similar pattern to
ILA, and the other more similar to DPs. Inter-group comparison results diverge between
F
global
and D
global
at the second level of the hierarchical tree. Fraction matrix groups DPs,
DPd, ILA into one branch, and PL, ACAd, ACAv into another. Density matrix groups DPs
into one branch, the other regions into another, and DPd split between two branches. D
global
gives much higher consideration to midline thalamus than F
global
, where many small nuclei
receive dense innervation. While mPFC is classically characterized by its midline thalamus
connections, DPs projection to thalamus is very weak (Figure 3.5A and Figure 3.5B). This is
an example of how interpretation of connectivity pattern may dier, based on the selection
of data matrix: in general, the density matrix gives higher weights to connectivity observed
in small nuclei than the fraction matrix.
Results of clustering on retrograde experiment data show one ACAd injection clustered
with PL injections and vice versa. One plausible reason is the existence of subdomains within
either PL or ACAd, which have dierential input patterns with other subdomains. Another
possibility is that a subset of the PL/ACAd inputs are laminar specic, leading to large
variability between injections placed at dierent cortical layers. Nevertheless, the majority
of our retrograde cases are clustered as expected (Figure 3.6B).
58
59
Figure 3.5: mPFC Cortices Whole Brain Projections. (A) Left panel: Scatter plots of
mPFC fractional projections to CTX, STR, MB, HY, TH, HB, AMT and HPF. Results from the
two hemispheres are displayed separately. The top left panel shows combined projections to the
entire ipsilateral and contralateral brain. Right panel: MLE projection fractions calculated from
data points in left panel. (B) 2D hierachical clustering of mPFC projection fraction matrix. Given
the diculty to show the entire 1376 matrix columns, only regions with high projection fractions in
at least 1 brain sample are displayed. (C) 2D hierachical clustering of mPFC anterograde tracing
density matrix. Similar to (B), only high density regions are displayed
Figure 3.6: mPFC Cortices Whole Brain Inputs. (A) Scatter plots of mPFC input fractions
from CTX, HPF, TH, AMY, STR, HY, MB and HB. MLE fractions of input from the 8 groups are
given in the table in the right panel. (B) 2D hierachical clustering of mPFC input fraction matrix.
Only regions with high input fractions are displayed.
60
3.3.4 mPFC Connectivity with Cortex
Among the 8 hierarchy groups we examine, the cerebral cortex is the largest projection
target of mPFC, as well as providing the most inputs to mPFC (Figure 3.5A and Figure
3.6A). Cortical regions receiving mPFC projections are seen to form several clusters (Figure
3.7A).
The cluster receiving most mPFC input consists of prefontal cortex (PFC) regions, in-
cluding mPFC itself, as well as three orbitofrontal regions, orbital area medial (ORBm),
ventral lateral(ORBvl) and lateral (ORBl) parts, and the agranular frontal regions, pri-
mary(MOp) and secondary (MOs) motor areas. The PFC cluster accounts for a signicant
amount of cortical projections from mPFC. The ratios of
b
f
PFC
b
f
CTX
is 61.21%, 62.07%, 53.43%,
51.17%, 62.84% and 81.31% for ILA, DPs, DPd, PL, ACAv and ACAd respectively. PFC
cluster regions are also a signicant cortical source of input to mPFC, with ratios of
b
f
PFC
b
f
CTX
equal 34.50%, 10.41%, 21.43%, 67.48%, 57.40% and 68.92% for ILA, DPs, DPd, PL, ACAv
and ACAd respectively. These values are likely underestimations, since in each brain sample
signals from some mPFC volumes are excluded from quantication due to high intensity
from tracer injection (see chapter 2.2.3). Amount of mPFC projection to and input from
PFC are shown as heat maps in the right panel of Figure 3.7A. Color coded patterns of
mPFC-PFC connections are shown in Figure 3.7D.
61
62
Figure 3.7: mPFC Connectivity with Cortex. (A) Left panel: 2D hierarchical clustering
of mPFC-Cortex projection fraction matrix reveal several clusters of cortical projection targets of
mPFC. PFC areas are in the columns to the left of the gray line. Other clusters are highlighted
with colored box. Right panel: mPFC projection to (top) and input from (bottom) PFC are
shown as heatmaps. (B) 2D hierarchical clustering of mPFC-Cortex input fraction matrix reveal
show an additional cluster not revealed by the projection data. (C) Reciprocity of connectivity
between mPFC and cortex. (D) Reconstruction of mPFC-PFC connecitivty. Segmented tracer
from collective experiments shown on top of a ARA wire frame. Considering the density of the
display, we do not use text label of region names in this gure, but provide such a label in the
appendix (Appendix 5.1). (E) Reconstruction of mPFC connectivity with ENTl, PERI, ECT and
TEa. (F) Reconstruction of mPFC connectivity with RSP, VIS and PTLp. (G) Reconstruction of
mPFC connectivity with CLA, AI, GU and VISC. (H) Reconstruction of mPFC input from PIR
and AUD.
Interestingly, both DPs and DPd show projections to PFC that are several folds greater
than their input from PFC. Apart from its projection to neighboring ILA, DPs projection
to PFC is concentrated in two ber pathways to layer 1 and deep layer 5 of ACAv, ACAd
and MOs, observed posterior to ARA level 42 (Figure 3.7D). These two pathways and the
ILA projection constitute 51.78% of total cortical projections from DPs. Meanwhile, regions
satisfying
b
f
Region
b
f
CTX
0:01 for DPs are supercial layers of piriform cortex (PIR, 26.27%) and
entorhinal area, lateral part (ENTl, 10.52%), perirhinal area (PERI, 1.81%), ectorhinal area
(ECT, 1.20%), deep layers of PIR (8.18%) and ENTl (1.24%), ventral auditory area (AUDv,
1.33%), temporal association areas (TEa, 1.22%) and the endopiriform nucleus (EPd, 5.94%),
none of which belong to PFC. The fractions of input that DPs receive from PIR, AUDv
and ENTl is greater than any other mPFC areas (Figure 3.7H, ARA level 53 and 64).
Additionally, ENTl cell populations innervating DPs show dierence in laminar distribution
than those innervating other mPFC area (Figure 3.7E). DPs has a
b
f
Entls
b
f
ENTl
d
value equal to
6.84. This value ranges between 0.5 and 0.75 for others, indicating segregate information
pathways from ENTl to mPFC. In contrast to the notable amount of input DPs receives
from PIR, AUDv and ENTl, its output to them are of much lesser magnitude at 7.68%,
0.2% and negligible. Viewing the evidence together, DPs relays a unique set of pathways,
with a convergence of olfactory and auditory information, to ILA and specic layers of
ACAv, ACAd and MOs, largely with weak feedback mechanisms. DPs and DPd are also
63
associated with two functionally related cluster groups, consisting of the olfactory system
cortices. The group of PIR and TTd receives stronger input than the group of TTv, OT and
two AON nuclei AONd and AONl. Note that although we discussed the uneven amount of
input versus output between DPs and PIR, DPs projection to PIR is still stronger than other
mPFC areas. The olfactory areas PIR, TT, AON and cortical amygdalar (COA) combined
recive 34.11%, 32.54%, 15.62%, 9.22%, 0.97% and 0.69% of total cortical projections of DPs,
DPd, ILA, PL, ACAv and ACAd respectively.
The ILA projection pattern is associated with a cluster group including deep layers of
ENTl, ECT and TEA, CLA (claustrum) and EPd (Figure 3.7E and Figure 3.7G). Its bilateral
projection to layer 6 of ENTl is unique among the mPFC areas. Back projection from ENTl
to ILA mainly arise from ENTl layer 5. ILA sends the highest fraction of projections to ENTl,
PERI, ECT, and TEA compared to other mPFC areas (Figure 3.7E). CLA is a notable region
with reciprocal dense connections with mPFC. The density of ILA to CLA projection is 3
rd
among all of ILA's cortical projections, and PL to ILA projection ranks 7
th
. Both in ILA
and PL, all regions with denser input are part of PFC. mPFC projection fractions to the
claustrum (CLA) are such that
b
f
ILA
CLA
>
b
f
PL
CLA
>
b
f
ACAv
CLA
>
b
f
ACAd
CLA
>
b
f
DPs
CLA
>
b
f
DPd
CLA
' 0.
With the exception of DP, another interesting pattern is a proportion contralateral versus
ipsilateral clustrum projection such that
b
f
ILA
CLAc
b
f
ILA
CLA
i
<
b
f
PL
CLAc
b
f
PL
CLA
i
<
b
f
ACAv
CLAc
b
f
ACAv
CLA
i
<
b
f
ACAd
CLAc
b
f
ACAd
CLA
i
. mPFC inputs
from CLA are also dense (except DPs which has no input from CLA), being the 3rd, 1
st
, 3
th
,
6
th
and 11
th
densiest out of all cortical areas for ILA, PL, ACAv, ACAd and DPd respectively,
and predominantly arise from ipsilateral CLA. mPFC-CLA inputs and outputs are shown in
Figure 3.7G. Finally, ILA has the highest amount of projection to EPd, followed by DPd and
PL. Inputs from EPd is strong in DPs and DPd (7.18% and 7.22% of total cortical input),
and weaker in other mPFC areas, following the same ranking order of mPFC to olfactory
area projections. EPd is in close spatial proximity with CLA. Given the large amount of
bers in CLA, we want to be certain that observations at EPd are not false positives from
mis-registering CLA. We test for a null hypothesis that amount of projections in EPd has no
linear correlation with that of CLA. This hypothesis is rejected with p 0:03, although the
64
correlation coecient is low (r = 0:50). We do know that for ILA, PL, ACAv and ACAd, the
order of projection fractions into olfactorcy areas and CLA exactly coincide. The correlation
found between EPd and CLA projections can very well be biologically factual, rather than
a byproduct of registration error. Additionally, using the retrograde dataset, fractions of
mPFC input from EPd and CLA have a negative correlation coecient (r =0:65). This
strongly argues against CLA signal spilling into EPd, as that would give a positive coecient,
and demonstrate the reliability of our quantication system.
The PL projection pattern is associated with a cluster group including gustatory areas
(GU) and agranular insular areas dorsal and ventral part (AId and AIv), with PL generating
higest projection fraction to all three regions among mPFC areas. Its projection to AI is
nearly evenly divided on the two hemisphere. mPFC inputs from AId and AIv are of more
similar amounts for all areas, and more ipsilaterally conned. GU provides few input to
mPFC (Figure 3.7G).
The nal projection pattern cluster group is associated with ACAv and consists of retro-
splenial areas lateral agranular part, dorsal part and ventral part(RSPagl, RSPd and RSPv)
and anteromedial and posteromedial visual area (VISam and VISpm). ACAv projections to
the three RSP areas combined and to VISam are over 10 times greater than other mPFC
regions. RSP and VIS inputs to mPFC are also at least 3 times higher in ACAv. Though
left out of major cluster groups, the posterior parietal association areas (PTLp) have a fair
amount of bilateral connections with mPFC areas. ACAv and ACAd project the most to
PTLp, while ACAv and PL receive the most input from PTLp. Projections to PTLp are
concentrated in its deep layers, with
b
f
PTLp
d
b
f
PTLps
equals to 1.27, 2.28, 2.30, 3.62 and 2.35 for ILA,
DPs, PL, ACAv and ACAd respectively. Inputs from PTLp have dierent laminar distri-
butions for mPFC areas. Using binomial test, there is a signicant dierence of proportion
of input cells arising from deep as opposed to supercial layers of PTLp in DPs (p 0:01),
DPd (p 4:01e 5), ILA(p 3:14e 4) and PL (p 3:15e 3), with DPs and DPd in-
puts concentrated in deep PTLp, and ILA and PL inputs in supercial PTLp. PTLp input
to ACAv and ACAd shows no signicant laminar distribution dierences (p 0:058 and
65
p 0:37). These data suggest segregated input pathways from PTLp to mPFC.
The above projection cluster groups generally capture signicant cortical input regions
to mPFC. DPd has a set of strong inputs not covered above from cortical amygdalar area,
posterial part, lateral and medial zone (COApl and COApm), accounting for 13.81% of
its total cortical input (Figure 3.7B). Cortical amygdalar are shown to drive appetitive or
aversive behavior according to innate ordor preference [66]. The mPFC, through heavy
projection to DPd from the COA, may factor the odor aect encoding into behaviors such
as feeding and
ight.
We are also interested in systemically identifying regions with strong reciprocal connec-
tions with mPFC. To do this, we dene a measure of connection reciprocity between a given
mPFC area and a region r,
r
as

r
=
min(
b
f
projection
r
;
b
f
input
r
)
max(
b
f
projection
r
;
b
f
input
r
)

(
b
f
projection
r
+
b
f
input
r
)
2

r
penalizes regions that show signicant dierence in the amount of their mPFC input and
output, as well as regions that have low amount of connectivity with mPFC to suppress
spuriously large division ratios with no biological signicance. The result of hierarchical
clustering on is shown in Figure 3.7C and consistent with our discussion above. Within
cortical regions, reciprocal connectivity is found beteen ILA and ENTl, ECT, CLA, PFC,
weakly between DPs and PFC, between DPd and PFC, between PL and AI, CLA and PFC,
between ACAv and RSP, VIS, PFC, and nally strongly between ACAd and PFC. We nd
is an eective measure to gain insight into connectivity datasets and derive evidence based
hypotheses.
66
Figure 3.8: mPFC Projection Densities are Enriched in Thalamic and Midbrain Re-
gions. Boxplots of mPFC whole brain projection densities. Each region are plotted individually
and arranged into their own hierarchical groups. Meaning of graph elements is identical as Figure
3.2B.
3.3.5 mPFC Connectivity with Thalamus
With the exception of DPs, mPFC cortices send more than 10% of their total projec-
tions to ipsilateral thalamic nuclei, and to a lesser extent to contralateral thalamus as well
(Figure 3.5A and Figure 3.5B). mPFC is a very in
uential source of input for the thala-
mus (Figure 3.8). Using the density data matrix D, two-tailed Mann-Whitney U test with
Holm-Bonferroni correction shows signicant dierence between mPFC projection densi-
ties in regions of thalamus versus regions of CTX (U = 254090.0, p 2.39e-10), STR (U =
483919.5, p 2.87e-3), HY(U = 1194642.0, p = 0.0), AMY(U = 329722.0, p = 0.0), HPF (U
= 76004.0, p 7.60e-5) and HB (U = 6099700.0, p = 0.0). The only hierarchical group with
a comparable range of mPFC projection density as the thalamus is the midbrain (MB) (U =
1088609.5, p 0.49). All mPFC areas receive nearly 10% or more of their inputs from the
thalamus (Figure 3.6). Many thalamic nuclei, especially the midline thalamus, have strong
reciprocal connections with mPFC (Figure 3.11C). Quantied amounts of mPFC-thalamic
connections are shown as scatter plots in Figure 3.11A. All mPFC-thalamic connectivity
67
from our entire mPFC anterograde and retrograde tracing experiment dataset are plotted
on top of the ARA wire frame, and shown below in Figure 3.9 and Figure 3.10. We provide
thalamus region name labels in appendix in order not to crowd the gure (Appendix 5.2).
68
Figure 3.9: Reconstruction of mPFC-Thalamic Projections. Composite mPFC-thalamic
projections reconstructed from 18 mPFC anterograde tracing experiments. The circles under each
image indicate if the level is present in any of the experiment with injection into the region repre-
sented by the circle. Level 78 is omitted since no data is available for it.
69
Figure 3.10: Reconstruction of Thalamic-mPFC Projections. Composite mPFC-thalamic
projections reconstructed from 15 mPFC anterograde tracing experiments. Symbols have identical
meaning as Figure 3.9. ARA levels 84 and later are omitted because no cells are found in any
experiment.
70
This aggregated view of large, complex dataset exposes us to great amounts of informa-
tion within one page space, and allow us to develop insights and hypotheses in a data driven
manner. We observe three groups of thalamic nuclei as dierentially innervated by parts
of mPFC (Figure 3.11). These dierential patterns are visible from hierarchical clustering
(shown in colored boxes in Figure 3.11B) with both the individual fraction matrix F and
the sample fraction matrix
b
F . The individual fraction matrix F shows a clustering pattern
over biological variation of individual specimens that is in general agreement with clustering
results from sample fraction matrix
b
F .
b
F gives more robust results with increased sample
sizes, due to its statistical property (discussed in chapter 2.2.5). Patterns of thalamic pro-
jections from ILA and DPd are highly similar, while DPs has very little thalamic eerents.
This is consistent with the fact that cortico-thalamic projecting neurons reside in a deeper
cortical layer (layer 5). PL shows similarity in its thalamic projections with both ILA and
ACAd, sharing a few common nuclei with ILA and ACAd as target for strong projections.
ACAv on the other hand, has several unique projection targets in the thalamus. The three
thalamic nuclei groups are described below.
(1) Mediodorsal nucleus of the thalamus, medial part (MDm), paraventricular nucleus of
the thalamus (PVT), parataenial nucleus (PT) and nucleus reunions (RE). These four mid-
line thalamic nuclei are major targets of ILA and DPd. Together they account for 60.46% and
70.82% of total thalamic projections of ILA and DPd respectively. MDm also receives strong
input from PL. PVT and PT are postulated to relay information of arousal. It receives input
from the brain stem, hypothalamus and the mPFC, and projects heavily to limbic regions,
including the nucleus accumbens (ACB) [67, 68], the amygdalar complex, BST and mPFC
[69]. More recent studies also show that PVT is innervated by axonal terminals releasing two
feeding related peptides, orexin and CART (cocaine and amphetamine related transcript)
[70]. Projections to PT from ILA and DPd are distinctively dense (Figure 3.9 ARA level
58). PT and PVT together account for 28.90% and 28.99% of total thalamic projections
for ILA and DPd. RE serves the important function of relaying mPFC information to the
hippocampus, due to a lack of direct mPFC-hippocampal projection [57{60], and is critical
71
in many aspects of memory [71{73]. Apart from DPs, all mPFC areas have signicant RE
projections, accounting for 23.82%, 14.30%, 14.34%, 11.91% and 13.35% of total thalamic
projections for DPd, ILA, PL, ACAv and ACAd respectively. Notably, DPs receives enor-
mous RE input (Figure 3.9 ARA levels 58 74 and Figure 3.1E), that amounts to 76.79%
of its total thalamic input. Fractions of RE input as a ratio of total thalamic input for other
mPFC areas are 28.57%, 34.81%, 5.20%, 4.47% and 2.19% for DPd, ILA, PL, ACAv and
ACAd respectively.
Figure 3.11: mPFC-Thalamic Connectivity Quantication and Hierarchical Cluster-
ing. (A) Top panel: Scatter plots of mPFC-thalamic projection fraction data. Bottom panel:
Scatter plots of mPFC-thalamic input fraction data.
(2) Mediodorsal nucleus of the thalamus, lateral and caudal part (MDl and MDc), retic-
ular nucleus of the thalamus (RT), ventral medial nucleus of the thalamus (VM), ventral
anterior-lateral complex of the thalamus (VAL), parafascicular nucleus (PF) comprise the
72
second group of thalamic nuclei group revealed by hierarchical clustering. These areas are
predominantly innervated by ACAd and PL, with ACAv providing additional strong inputs
to MDl and VM. Mediodorsal dorsal thalamus (MD) plays important roles in cognitive func-
tions including working memory and goal-oriented behavior [74{76]. Schizophrenia patients
show decreased mPFC-MD connectivity and impaired working memory [77]. Optogenetic
inhibition of the mPFC-MD pathway in mice show the requirement of reciprocal mPFC-
MD connections in spatial working memory [78]. Neuroanatomically, ILA, PL and ACA
are known to have a medial lateral shift in their connections with MD [79, 80]. We show
this classic nding quantitatively (also see Figure 3.9 and Figure 3.10, ARA levels 63 74
for graphical representation). For mPFC-thalamic projections,
b
f
MDm
b
f
MD
l
is equal to 8.26, 3.54,
1.72, 0.16 and 0.35 for DPd, ILA, PL, ACAv and ACAd respectively. For mPFC inputs from
thalamus,
b
f
MDm
b
f
MD
l
is equal to 4.44, 2.58, 0.40, 0.10 and 0.14 for above regions in the same
order. This reveals an interesting case for PL, where its outputs are preferentially to MDm
and more of its inputs are from MDl. VM and VAL are shown to aect motor behavior,
including posture, locomotion [81] and sensorimotor learning [82]. Optogenetic activation
and inhibition of calbindin1-positive matrix cells in VM are responsible for arousal through
broad cortical activation [83]. By our reciprocity measure , VM has reciprocal connections
with ILA, PL, ACAv and ACAd, with being greatest in ACAd, weaker in PL and ACAV
and further weaker in ILA. DP receives input from VM in DPs, but does not have signicant
projection to it Figure 3.9 and Figure 3.10 ARA levels 63 and 64. RT is also implicated
in attention and arousal [84{86]. mPFC projects to RT unilaterally, with PL giving the
most projection. RT projection accounts for 8.14%, 5.42%, 2.57%, 1.87% of total thalamic
projections for PL, ACAd, ACAv and ILA respectively, while DPd projects little to RT. PF
gives dense projections to basal ganglia and regulates motor action execution [87]. It is also
thought to control balance relaying input from vestibular nucleus to the striatum [88, 89].
mPFC projection to PF is of greater magnitude than its input from PF, with PL and ACAd
having the strongest projection to PF.
73
Figure 3.11: (continued) mPFC-Thalamic Connectivity Quantication and Hierarchi-
cal Clustering. (B) 2D hierarchical clustering of mPFC-thalamic projection fraction matrix
reveal several clusters of thalamic projection targets of mPFC. Top panel: Clustering with all
individual experiment cases. Bottom panel: Clustering with MLE projections.
(3) Anteromedial nucleus dorsal and ventral part (AMd and AMv), CL, central latera,
lateral dorsal and lateral posterior nucleus of the thalamus (CL, LD and LP) comprise a third
group of thalamic nuclei. These nuclei mainly receive input from ACAv. AM is suggested in
attention and predatory defense [90, 91]. ACAv has the highest amount of projections to as
74
well as inputs from AMd and AMv, followed by ACAd and then PL, though AMv input to
ACAv and ACAd is near identical. The functions of CL, LD and LP are poorly understood.
ACAv, ACAd and PL have reciprocal connections with CL, again in decreasing order of
magnitude. The connection between ACAv and LD, LP are more specic to ACAv (over 10
fold greater than other mPFC areas). In general, little is known about this thalamic group
with preferentially reciprocal connections with ACAd. We hope our ndings will help the
development and testing of hypotheses regarding their roles in the corcico-thalamic circuitry.
Figure 3.11: (continued) mPFC-Thalamic Connectivity Quantication and Hierarchi-
cal Clustering. (C) Reciprocity of mPFC-thalamus connections.
3.3.6 mPFC Connectivity with Striatum
The STR hierarchical group contains regions of the dorsal and ventral striatum-pallidum
(STR
d
and STR
v
) and bed nuclei of stria terminalis (STR
BST
). STR
d
consists of caudate
putamen (CP) and globus pallidus external and internal segments (GPe and GPi). STR
v
in-
cludes nucleus accumbens (ACB), substantia innominata (SI), lateral septum rostral, caudal
and ventral parts (LSr, LSc, LSv), NDB, MA, FS, SH, MS, SF, TRS. STR
BST
is a collec-
tion of small nuclei, including BST am, BST pr, BST if (please see Table 5.3 for a detailed
list of these nuclei). Full region names omitted for brevity can be found in the preamble
75
Abbreviation section.
Figure 3.12: mPFC Projections to Dorsal Striatum-Pallidum, Ventral Striatum-
Pallidum and BST Nuclei. (A) Top panel: Scatter plots of mPFC projection fractions in
STR
d
, STR
v
and STR
BST
. Bottom panel: MLE mPFC projection fractions in STR
d
, STR
v
and
STR
BST
. (B) Scatter plots of mPFC projection fractions in individual regions of the STR hierar-
chical group. (C) Amount of CP projection along the anterior-posterior axis from ILA, PL, ACAv
and ACAd. Location of injection site is shown in color representing the region. Projection center
of mass on the anterior-posterior axis is shown as a gray line.
76
Figure 3.12: (continued) mPFC Projections to Dorsal Striatum-Pallidum, Ventral
Striatum-Pallidum and BST Nuclei. (D) Left panel: Subdomains of CP dened by their
cortical projection terminal elds, image is modied from [16]. Right panel: Reconstruction of
mPFC-STR projection at ARA levels 43, 53 and 61, corresponding to rostral, intermediate and
caudal CP.
77
STR regions combined receive large portions of mPFC input (Figure 3.5B), second only
to the CTX regions. Chi-square test of independence show signicant dierences in amount
of projections into STR
d
, STR
v
and STR
BST
(
2
(10) = 1:21e8, p = 0:0) from mPFC re-
gions. DPs, DPd and ILA innervates STR
v
more than STR
d
, with DPs and DPd showing
more extreme ratios of projection fractions in STR
v
versus STR
d

b
f
STRv
b
f
STR
d
= 27:59, 20.98 and
2.44 for DPs, DPd and ILA respectively

. ACAv, ACAd and PL preferentially innervate
STR
d
, with ACAv showing the strongest bias toward STR
d

b
f
STR
d
b
f
STRv
= 18:81, 8.16 and 2.93
for ACAd, ACAv and PL respectively

(Figure 3.12A). It should be noted that despite a
STR
d
leaning connectivity prole, PL gives comparable levels of projections into the nucleus
accumbens as ILA and DPd. ILA and DPd have highest fractions of projections into BST
(2.9% and 1.8% respectively), while projections to BST from other mPFC areas are several
fold to several magnitude weaker (Figure 3.12A). In addition to their heavier projections to
STR
v
compared to STR
d
, two-tailed Mann-Whitney U test shows that ILA, DPs and DPd
brain specimens have projections more heavily concentrated in the ipsilateral STR, compared
to PL, ACAv and ACAd specimens (U = 5, p 0.0027). During goal oriented behavior,
the ventral striatum-pallidum is involved in limbic neural circuits of reward and motivation,
while its dorsal counterpart processes more sensorimotor and spatial location related infor-
mation [92, 93]. BST receives strong input from amygdala and regulates activities of the
hypothalamic-pituitary-adrenal(HPA) axis [94, 95]. The quantied connectivity proportion
gradient from mPFC areas to STR
d
, STR
v
and STR
BST
suggest a shift of sensorimotor
and limbic involvement in mPFC dorsal-ventrally, with increasing limbic involvement in the
ventral areas.
mPFC Projections to STR
d
are predominantly accounted for by CP (Figure 3.12B bot-
tom panel). Fiber terminations in CP are concentrated in the medial portions of rostral and
intermediate CP, and in dorsomedial portion of caudal CP. These conclusions are consistent
with our previous ndings on cortico-striatal protectome [16] (Figure 3.12D). Among ILA,
PL, ACAv and ACAd, we are interested to see if these regions dier in the CP projec-
tion densities along the rostral-caudal axis. We nd no evidence supporting this: one-way
78
ANOVA testing of projection center of mass shows no signicant dierence (F(3, 8) = 2.03,
p 0:23). Average ARA levels of center of mass are 48.04, 50.72, 49.11, 51.52 for ILA, PL,
ACAv and ACAd respectively. Within each cortices, rostral-caudal position of the injection
site also does not correlate with that of center of mass, suggesting a lack of tight topography
rostral-caudally (Figure 3.12C). All these 4 mPFC regions preferentially project to more
rostral parts of CP (prior to ARA level 56, beginning level of GPe). Binomial proportions
tests were signicant for all 4 regions, when volume dierence between CP levels 40 55
and levels 56 78 are accounted for (
V
4055
V
4055
+V
5678
= 0:57).
In the ventral striatum-pallidum, ACB, SI and LSr together account for 92.67% 99.48%
of mPFC to STR
v
projections. In the ACB, ber spatial distribution from mPFC cortices
follow some patterns. One-way Anova shows signicant dierences in location of projection
center of mass along ARA levels (F(6, 12) = 10.50, p 4.69e-4). Fibers from ACAv and
ACAd are more rostrally distributed (center of mass at ARA level 38.58 and 39.58 on average
respectively), while ACB projections from ILA, DPs, DPd and PL center more caudally than
ARA level 41 (41.0, 43.50, 41.98, 43.06 respectively). ACAv and ACAd also projects only
to a small, conned area of the dorsal shell in rostral ACB. ILA and PL projects to two
larger, mostly non overlapping oblique stripes in ACB, similar to the pattern described in
[92]. DP projection does not follow the stripe pattern, and spans ventral ACB along the
core. The LSr projection is a pronounced distinction for DP, especially DPs (Figure 3.12B
and Figure 3.12D top panel). LSr regulates emotion and reward through its connection with
the limbic system [96, 97], as well as modulates olfaction cued learning such as pair bonding
[98, 99]. DP receives strong inputs from the piriform cortex, also known as the olfactory
cortex, the auditory cortex (Figure 3.3.1 and Figure 3.7H), and from the olfactory bulb [51],
bringing convergence of sensory information into the aect regulating circuitry of the limbic
system. SI is considered as the ventral pallidum. mPFC projections into SI is dorsally and
medially concentrated in SI. SI's input and output pathways are not well documented in
mice. Our own unpublished data suggest a medio-lateral distinction in the neuroanatomical
connections of SI. The medial SI shows strong reciprocal connections between SI and laterla
79
ACB, basolateral amygdalar nucleus, anterior part (BLAa), and prominent input from the
olfactory tubercle (OT) as well as output to ventral tegmental area (VTA) and substantia
nigra, reticular part (SNr). The lateral SI has a more olfactory area leaning connectivity
prole. The higher projection fractions in SI from DPs and ILA (Figure 3.12B, Figure 3.12D
central and bottom panel) suggest their involvement in limbic functions including fear and
reward seeking.
3.3.7 mPFC Connectivity with Hippocampal Formation, Midbrain and Hind-
brain
mPFC connections with the hippocampal formation, midbrain and hindbrain are largely
unilateral (Figure 3.5B and Figure 3.6A). Hippocampal input to mPFC mainly arise from
CA1 and subiculum ventral and dorsal parts (SUBv and SUBd), with ILA and DPd receiving
the most hippocampal input (17.78% and 15.42% of their global input) (Figure 3.13B ARA
level 89). We nd a signicantly higher proportion of SUBv input to DPd than to ILA with
binomial test (p 8:06e-20, Figure 3.13B ARA levels 87 and 88). This dierence allows a
more complex hippocampal in
uence on mPFC.
mPFC projections to the midbrain are dense (Figure 3.8) and mostly concentrated in a
few regions, including periaqueductal gray (PAG), superior colliculus, motor related (SCm),
midbrain reticular nucleus (MRN) and ventral tegmental area (VTA) (Figure 3.13A bottom
panel). The cortices dier in the amount of their midbrain projections. DPs, DPd and PL
send 8.39%, 8,12% and 9.71% of their total projections to the midbrain, while ILA, ACAv
and ACAd have greater midbrain projections at 20.31%, 21.08% and 19.04%.
PAG is the largest midbrain target of mPFC, accounting for over 30% of MB projections
across all mPFC areas (64.20%, 30.48%, 48.66%, 33.58%, 41.16% and 34.22% for DPs, DPd,
ILA, PL, ACAv and ACAd). For ILA, ACAv and ACAd, their projection to PAG accounts
for 9.88%, 8.68% and 6.52% of their global projections. Considering the long physical dis-
tance between midbrain and mPFC, these numbers are highly signicant. Four subdivisions
have been delineated for PAG: dorsomedial, dorsolateral, lateral, and ventrolateral, with
80
each subdivision having its own descending and ascending connections [100, 101]. PAG is
important in fear and defense, with the four subdivisions mediating dierent types of com-
plex fear and defensive behavior response. mPFC projections to lateral and ventrolateral
PAG regulate context fear discrimination [102], while the dorsolateral PAG produces un-
conditioned freezing behavior [103]. The lateral and ventrolateral PAG are also thought
to control
ight versus freezing respectively [104]. We found dierences in mPFC cortices
projections to PAG subdivions. The dorsomedial and dorsolateral divisions are most concen-
trated in ACAv projections (Figure 3.13A ARA level 97). Most ILA and ACAd projections
are in lateral and ventrolateral PAG, while PL projections are more limited to ventrolateral
PAG (Figure 3.13A ARA levels 98 and 99). In caudal PAG (later than ARA level 105),
ILA projection in PAG seems to shift dorsally. These patterns of mPFC-PAG subdivision
projection are highly suggestive of dierential fear and defense responses. Currently we do
not have quantitative data for the mPFC-PAG division specic projections, because ARA
does not have the four PAG subdivisions delineated. Our future work will examine the
division specic mPFC projections quantitatively. Dorsal nucleus raphe (DR), a midline
necleus neighboring ventrolateral PAG also receives mPFC input, with ILA, DPd, PL and
ACAd having relatively high projection fractions (Figure 3.13A ARA levels 97, 98 and 99)
and ACAv to a lesser extent (DPs has no projections to DR). DR serotonergic neuron trig-
gers behavior changes under uncontrollable stressors [105], while ILA and PL suppress DR
serotonergic neuron activities when a stress is infact controllable [106]. The cell type of DR
neurons innervated by ACA and their role in stress responses remain to be investigated.
81
Figure 3.13: mPFC Connections with the Hippocampal Formation , Midbrain and
Hindbrain. (A) Top panel: Reconstruction of mPFC-midbrain projection at representative ARA
levels. Bottom panel: Scatter plots of mPFC-midbrain projection fractions. Only region with
top amount of mPFC inputs are shown. (B) Reconstruction of mPFC-hippocampal input at
representative ARA levels.
82
SCm is the second largest midbrain projection target of mPFC. ACAv and ACAd have the
highest projection fractions to SCm at 29.57% and 16.00% of their total midbrain projection.
Their ber distributions in SCm show spatial segregation ( Figure 3.13A ARA levels 97 and
99), with ACAv preferentially projecting to medial SCm and ACAd to lateral SCm. SC
directs visual and auditory spatial attention either by producing eye movement or a shifting
attention covertly [107{111]. Arm movements toward novel sensory stimuli are also encoded
by SC [112, 113]. The mPFC projection pattern to SCm suggests ACAv and ACAd control
sensory stimuli guided attention and movements through distinct SCm neural substrates.
The function and connectivity of MRN is poorly understood. However the mPFC-MRN
projection does constitute a signicant portion of all mPFC-midbrain projections (Figure
3.13A top panel). On the other hand, VTA is well established to signal reward and aversion
through its dopaminergic neurons [114, 115]. In the global context, ILA gives the most
projection to VTA, over 2 folds greater than other mPFC cortices.
mPFC projections to the hindbrain are mostly found in pontinine reticular nucleus rostral
and caudal parts (PRNr and PRNc), pontine central gray (PCG), superior central nucleus
raphe medial and lateral parts (CSm and CSl) and tegmental reticular nucleus (TRN) (Fig-
ure 3.13C). PRN and TFN induces REM sleep [116{118], and receives the most projections
from ACAv and ACAd. PRN receives 2.36% and 3.31% of the global projections of ACAv
and ACAd, while ACA projections to TRN is at least twice the amount of other mPFC ar-
eas. Considering ACAv and ACAd's connections with the thalamic arousal and wakefulness
regulating nuclei VM and RT (Figure 3.11), and their connections with SCm, the anterior
cingulate cortex appear more involved in guiding attention and evoking arousal than ILA, PL
and DP. Serotonergic neurons in CS is known to mediate fear memory [119, 120]. However
the CS neuronal cell types innervated by mPFC may not be serotonergic [121]. It is possible
that mPFC-CS pathways regulate aspects of stress and fear related behavior dierent than
those regulated by serotonergic neurons.
83
Figure 3.13: (continued) mPFC Connections with the Hippocampal Formation , Mid-
brain and Hindbrain. (C) Top panel: Reconstruction of mPFC-hindbrain projection at repre-
sentative ARA levels. Bottom panel: Scatter plots of mPFC-hindbrain projection fractions. Only
region with top amount of mPFC inputs are shown.
3.3.8 mPFC Connectivity with Hypothalamus and Amygdalar Complex
Hypothalamus and amygdalar are situated at the ventral most brain. We notice these
two hierarchy groups often do not have the same best matching ARA level as more dorsally
located regions in a coronal brain section. Although their borders with other hierarchy groups
can be properly aligned, precise registration of nuclei within hypothalamus and amygdalar
can be dicult to achieve. We are currently modifying our pipeline software to allow the
registration of hypothalamus and amygdalar to a dierent ARA level than the rest of the
section and unications of the resulting quantications. Therefore we leave the detailed
discussion of mPFC-hypothalamus and mPFC-amygdalar connections to future work. We
84
do note that ILA, DPs and DPd distribute more of their projections to both hypothalamus
and amygdalar complex than PL, ACAv and ACAd. DP projections to the LHA is especially
strong.
3.4 Conclusion
We present the quantitative connectome of the medial prefrontal cortex. Our report
is signicant in several aspects. First of all, our result is the most comprehensive dataset
on mice mPFC structural connectivity to date. Secondly, we provide fully quantitative
connectivity measures with a combination of careful interactive quality control and parallel
computing architecture enabled image processing and statistical analysis pipelines. The
comprehensiveness and quantitative nature of the data together allow an unprecedented
level of insight into the wiring diagrams of mPFC. Last but not least, we present cyto-
architectural and neuroanatomical evidence that dorsal peduncle is a distinct region from its
neighboring ILA. The connectivity of DP relays convergent information from olfactory and
auditory systems as well as the hippocampal formation to mPFC, and expands the in
uence
of mPFC in the ventral striatum-pallidum and hypothalamus, suggesting greater complexity
of mPFC function than previously known. Our future work aim to dissect the cell type
specic neural circuitries of mPFC and their functionality with genetically targeted neural
tracing and optogenetic activation/inhibition of neuronal activities.
Additionally, it is important to address the limitations in the state-of-art quantication
and analysis framework for neuroanatomical data and invest our further eort accordingly.
Firstly, our results on mPFC projectome relies on approximating the number of synapses
by the amount of axon bers seen in coronal sections. This approximation is an over es-
timation in some brain regions that contain bers of passage, instead of axonal terminals.
Incorporation of pre- and post-synaptic markers in the data collection and image processing
work
ow, as well as referencing available manual annotations from classic neuroanatomical
studies are both valuable avenues to reduce the bias in synapse number approximation. Sec-
ondly, due to the current lack of well annotated 3D mouse brain atlas, we consider a 2D-2D
85
registration framework the most reliable approach for structural connectivity studies. How-
ever, we acknowledge that a satisfactory high resolution 3D atlas will address several dicult
issues, such as asymmetries introduced by oblique tissue section planes. If the image data
is collected as 3D volumes as well, bers of passage can be discerned from ber terminals
by tract tracing within the imaged volume, addressing the over estimation concern. Finally,
it is meaningful to carefully evaluate the number of injections required for the statistical
estimations of connectivity to reach stable values. Simultaneous reporting of the variabil-
ity and condence interval of the estimated fractions will also be informative. Despite the
limitations discussed above, our quantied mPFC structural connectivity dataset is a mile-
stone accomplishment in the comprehensive understanding of the complex and important
functions of the mPFC. We expect the results reported here to be valuable assets for the
research community to further investigate the involvement of mPFC neural pathways in goal
oriented behavior and neuropsychiatric diseases.
86
Bibliography
[1] Joseph L. Price and Wayne C. Drevets. Neural circuits underlying the pathophysiology
of mood disorders. Trends in Cognitive Sciences, 16(1):61{71, Jan 2012.
[2] Joshua Buckholtz and Andreas Meyer-Lindenberg. Psychopathology and the human
connectome: Toward a transdiagnostic model of risk for mental illness. Neuron,
74(6):990{1004, Jun 2012.
[3] Joaquin M. Fuster. The prefrontal cortex{an update: time is of the essence. Neuron,
30(2):319{333, May 2001.
[4] D Ongr and Joseph Price. The organization of networks within the orbital and medial
prefrontal cortex of rats, monkeys and humans. Cerebral Cortex, 10:206{19, 04 2000.
[5] Kadharbatcha S. Saleem, Hideki Kondo, and Joseph L. Price. Complementary cir-
cuits connecting the orbital and medial prefrontal networks with the temporal, insu-
lar, and opercular cortex in the macaque monkey. Journal of Comparative Neurology,
506(4):659{693, 2007.
[6] S.T. Carmichael and J.L. Price. Connectional networks within the orbital and medial
prefrontal cortex of macaque monkeys. Journal of Comparative Neurology, 371(2):179{
207, 1996.
[7] Harry B.M. Uylings, Henk J. Groenewegen, and Bryan Kolb. Do rats have a prefrontal
cortex? Behavioural Brain Research, 146(1):3 { 17, 2003. The Rodent Prefrontal
Cortex.
[8] Sylvie Granon and Bruno Poucet. Involvement of the rat prefrontal cortex in cognitive
functions: A central role for the prelimbic area. Psychobiology, 28(2):229{237, Jun
2000.
[9] H.W. Berendse and H.J. Groenewegen. Restricted cortical termination elds of the
midline and intralaminar thalamic nuclei in the rat. Neuroscience, 42(1):73 { 102,
1991.
[10] James P. Ray and Joseph L. Price. The organization of the thalamocortical connections
of the mediodorsal thalamic nucleus in the rat, related to the ventral forebrainprefrontal
cortex topography. Journal of Comparative Neurology, 323(2):167{197, 1992.
[11] Christian A Heidbreder and Henk J Groenewegen. The medial prefrontal cortex in
the rat: evidence for a dorso-ventral distinction based upon functional and anatomical
characteristics. Neuroscience & Biobehavioral Reviews, 27(6):555 { 579, 2003.
87
[12] George Paxinos and Keith B. J. Franklin. Paxinos and Franklin's the Mouse Brain in
Stereotaxic Coordinates, Fourth Edition. Academic Press, 4 edition, November 2012.
[13] H. J. J. M. Van De Werd, G. Rajkowska, P. Evers, and Harry B. M. Uylings. Cytoar-
chitectonic and chemoarchitectonic characterization of the prefrontal cortical areas in
the mouse. Brain Struct Funct, 214(4):339{353, May 2010.
[14] Hong Wei Dong. The Allen reference atlas: A digital color brain atlas of the C57Bl/6J
male mouse. John Wiley & Sons Inc., Hoboken, NJ, US, 2008.
[15] Brian Zingg, Houri Hintiryan, Lin Gou, MonicaY. Song, Maxwell Bay, MichaelS. Bi-
enkowski, NicholasN. Foster, Seita Yamashita, Ian Bowman, ArthurW. Toga, and
Hong-Wei Dong. Neural networks of the mouse neocortex. Cell, 156(5):1096 { 1111,
2014.
[16] Houri Hintiryan, Nicholas N. Foster, Ian Bowman, Maxwell Bay, Monica Y. Song, Lin
Gou, Seita Yamashita, Michael S. Bienkowski, Brian Zingg, Muye Zhu, X. William
Yang, Jean C. Shih, Arthur W. Toga, and Hong-Wei Dong. The mouse cortico-striatal
projectome. Nat Neurosci, 19(8):1100{1114, Aug 2016. 27322419[pmid].
[17] Siyu Zhang, Min Xu, Wei-Cheng Chang, Chenyan Ma, Johnny Phong Hoang Do,
Daniel Jeong, Tiany Lei, Jiang Lan Fan, and Yang Dan. Organization of long-
range inputs and outputs of frontal cortex for top-down control. Nature neuroscience,
19(12):1733, 2016.
[18] Ivo Dinov, Kamen Lozev, Petros Petrosyan, Zhizhong Liu, Paul Eggert, Jonathan
Pierce, Alen Zamanyan, Shruthi Chakrapani, John Van Horn, D. Stott Parker, Rico
Magsipoc, Kelvin Leung, Boris Gutman, Roger Woods, and Arthur Toga. Neuroimag-
ing study designs, computational analyses and data provenance using the loni pipeline.
PLOS ONE, 5(9):1{15, 09 2010.
[19] Ivo D. Dinov, Federica Torri, Fabio Macciardi, Petros Petrosyan, Zhizhong Liu, Alen
Zamanyan, Paul Eggert, Jonathan Pierce, Alex Genco, James A. Knowles, Andrew P.
Clark, John D. Van Horn, Joseph Ames, Carl Kesselman, and Arthur W. Toga. Appli-
cations of the pipeline environment for visual informatics and genomics computations.
BMC Bioinformatics, 12(1):304, Jul 2011.
[20] G. Allan Johnson, Alexandra Badea, Jerey Brandenburg, Gary Cofer, Boma Fubara,
Song Liu, and Jonathan Nissanov. Waxholm space: An image-based reference for coor-
dinating mouse brain research. Neuroimage, 53(2):365{372, Nov 2010. 20600960[pmid].
[21] Allen Mouse Common Coordinate Framework. Technical white paper: Allen mouse
common coordinate framework. 2015.
88
[22] Leonard Kuan, Yang Li, Chris Lau, David Feng, Amy Bernard, Susan M. Sunkin,
Hongkui Zeng, Chinh Dang, Michael Hawrylycz, and Lydia Ng. Neuroinformatics of
the allen mouse brain connectivity atlas. Methods, 73:4 { 17, 2015. Spatial mapping
of multi-modal data in neuroscience.
[23] Ed S. Lein, Michael J. Hawrylycz, Nancy Ao, Mikael Ayres, Amy Bensinger, Amy
Bernard, Andrew F. Boe, Mark S. Boguski, Kevin S. Brockway, Emi J. Byrnes, Lin
Chen, Li Chen, Tsuey-Ming Chen, Mei Chi Chin, Jimmy Chong, Brian E. Crook,
Aneta Czaplinska, Chinh N. Dang, Suvro Datta, Nick R. Dee, Aimee L. Desaki,
Tsega Desta, Ellen Diep, Tim A. Dolbeare, Matthew J. Donelan, Hong-Wei Dong,
Jennifer G. Dougherty, Ben J. Duncan, Amanda J. Ebbert, Gregor Eichele, Lili K.
Estin, Casey Faber, Benjamin A. Facer, Rick Fields, Shanna R. Fischer, Tim P. Fliss,
Cli Frensley, Sabrina N. Gates, Katie J. Glattfelder, Kevin R. Halverson, Matthew R.
Hart, John G. Hohmann, Maureen P. Howell, Darren P. Jeung, Rebecca A. Johnson,
Patrick T. Karr, Reena Kawal, Jolene M. Kidney, Rachel H. Knapik, Chihchau L.
Kuan, James H. Lake, Annabel R. Laramee, Kirk D. Larsen, Christopher Lau, Tracy A.
Lemon, Agnes J. Liang, Ying Liu, Lon T. Luong, Jesse Michaels, Judith J. Morgan,
Rebecca J. Morgan, Marty T. Mortrud, Nerick F. Mosqueda, Lydia L. Ng, Randy
Ng, Geralyn J. Orta, Caroline C. Overly, Tu H. Pak, Sheana E. Parry, Sayan D.
Pathak, Owen C. Pearson, Ralph B. Puchalski, Zackery L. Riley, Hannah R. Rockett,
Stephen A. Rowland, Joshua J. Royall, Marcos J. Ruiz, Nadia R. Sarno, Katherine
Schanit, Nadiya V. Shapovalova, Taz Sivisay, Cliord R. Slaughterbeck, Simon C.
Smith, Kimberly A. Smith, Bryan I. Smith, Andy J. Sodt, Nick N. Stewart, Kenda-
Ruth Stumpf, Susan M. Sunkin, Madhavi Sutram, Angelene Tam, Carey D. Teemer,
Christina Thaller, Carol L. Thompson, Lee R. Varnam, Axel Visel, Ray M. Whitlock,
Paul E. Wohnoutka, Crissa K. Wolkey, Victoria Y. Wong, Matthew Wood, Murat B.
Yaylaoglu, Rob C. Young, Brian L. Youngstrom, Xu Feng Yuan, Bin Zhang, Theresa A.
Zwingman, and Allan R. Jones. Genome-wide atlas of gene expression in the adult
mouse brain. Nature, 445:168, Dec 2006. Article.
[24] Stephen P. Amato, Feng Pan, Joel Schwartz, and Timothy M. Ragan. Whole brain
imaging with serial two-photon tomography. Front Neuroanat, 10:31, Mar 2016.
27047350[pmid].
[25] Daniel F urth, Thomas Vaissi ere, Ourania Tzortzi, Yang Xuan, Antje M artin, Iakovos
Lazaridis, Giada Spigolon, Gilberto Fisone, Raju Tomer, Karl Deisseroth, Marie
Carl en, Courtney A. Miller, Gavin Rumbaugh, and Konstantinos Meletis. An interac-
tive framework for whole-brain maps at cellular resolution. Nat Neurosci, 21(1):139{
149, Jan 2018. 29203898[pmid].
[26] R. Sprengel, K. Rohr, and H. S. Stiehl. Thin-plate spline approximation for image
registration. In Proceedings of 18th Annual International Conference of the IEEE
Engineering in Medicine and Biology Society, volume 3, pages 1190{1191 vol.3, Oct
1996.
89
[27] Shihui Ying, Yuanwei Wang, Zhijie Wen, and Yuping Lin. Nonlinear 2d shape regis-
tration via thin-plate spline and lie group representation. Neurocomputing, 195:129 {
136, 2016. Learning for Medical Imaging.
[28] Mark Jenkinson and Stephen Smith. A global optimisation method for robust ane
registration of brain images. Medical Image Analysis, 5(2):143 { 156, 2001.
[29] B. B. Avants, C. L. Epstein, M. Grossman, and J. C. Gee. Symmetric dieomorphic
image registration with cross-correlation: Evaluating automated labeling of elderly and
neurodegenerative brain. Med Image Anal, 12(1):26{41, Feb 2008. 17659998[pmid].
[30] B. B. Avants and G. Tustison, N.and Song. Advanced normalization tools (ants).
Insight J, 2:1{35, 2009.
[31] Julien Dauguet, Thierry Delzescaux, Franoise Cond, Jean-Franois Mangin, Nicholas
Ayache, Philippe Hantraye, and Vincent Frouin. Three-dimensional reconstruction of
stained histological slices and 3d non-linear registration with in-vivo mri for whole
baboon brain. Journal of Neuroscience Methods, 164(1):191 { 204, 2007.
[32] Godfried T. Toussaint. Applications of the rotating calipers to geometric problems in
two and three dimensions. International Journal of Digital Information and Wireless
Communications, 4:106, Jul 2014.
[33] D. A. Fish, A. M. Brinicombe, E. R. Pike, and J. G. Walker. Blind deconvolution by
means of the richardson{lucy algorithm. J. Opt. Soc. Am. A, 12(1):58{65, Jan 1995.
[34] Stephan Preibisch, Stephan Saalfeld, and Pavel Tomancak. Globally optimal stitching
of tiled 3d microscopic image acquisitions. Bioinformatics, 25(11):1463{1465, 2009.
[35] LEGESSE F.B., CHERNAVSKAIA O., HEUKE S., BOCKLITZ T., MEYER T.,
POPP J., and HEINTZMANN R. Seamless stitching of tile scan microscope images.
Journal of Microscopy, 258(3):223{232, 2015.
[36] Beat M unch, Pavel Trtik, Federica Marone, and Marco Stampanoni. Stripe and ring
artifact removal with combined wavelet | fourier ltering. Opt. Express, 17(10):8567{
8591, May 2009.
[37] Zhigang Zhou, Nong Sang, and Xinrong Hu. A parallel nonlinear adaptive enhancement
algorithm for low- or high-intensity color images. EURASIP Journal on Advances in
Signal Processing, 2014(1):70, May 2014.
[38] Rupal Patel and Vijayan K. Asari. A neighborhood dependent nonlinear technique
for color image enhancement. In Aur elio Campilho and Mohamed Kamel, editors,
Image Analysis and Recognition, pages 23{34, Berlin, Heidelberg, 2010. Springer Berlin
90
Heidelberg.
[39] M. Donoser and H. Bischof. Ecient maximally stable extremal region (mser) track-
ing. In 2006 IEEE Computer Society Conference on Computer Vision and Pattern
Recognition (CVPR'06)(CVPR), volume 01, pages 553{560, 06 2006.
[40] T. M. Koller, G. Gerig, G. Szekely, and D. Dettwiler. Multiscale detection of curvilinear
structures in 2-d and 3-d image data. In Proceedings of IEEE International Conference
on Computer Vision, pages 864{869, Jun 1995.
[41] Yoshinobu Sato, Shin Nakajima, Nobuyuki Shiraga, Hideki Atsumi, Shigeyuki Yoshida,
Thomas Koller, Guido Gerig, and Ron Kikinis. Three-dimensional multi-scale line lter
for segmentation and visualization of curvilinear structures in medical images. Medical
Image Analysis, 2(2):143{168, Jun 1998.
[42] Alejandro F. Frangi, Wiro J. Niessen, Koen L. Vincken, and Max A. Viergever. Mul-
tiscale vessel enhancement ltering. In William M. Wells, Alan Colchester, and Scott
Delp, editors, Medical Image Computing and Computer-Assisted Intervention | MIC-
CAI'98, pages 130{137, Berlin, Heidelberg, 1998. Springer Berlin Heidelberg.
[43] Luc M.J Florack, Bart M ter Haar Romeny, Jan J Koenderink, and Max A Viergever.
Scale and the dierential structure of images. Image and Vision Computing, 10(6):376
{ 388, 1992. Information Processing in Medical Imaging.
[44] Monica Monici. Cell and tissue auto
uorescence research and diagnostic applications.
volume 11 of Biotechnology Annual Review, pages 227 { 256. Elsevier, 2005.
[45] National Electrical Manufacturers Association (NEMA). Determination of signal-to-
noise ratio (SNR) in diagnostic magnetic resonance imaging. Rosslyn: National Elec-
trical Manufacturers Association, 2001.
[46] Dietrich Olaf, Raya Jos G., Reeder Scott B., Reiser Maximilian F., and Schoenberg Ste-
fan O. Measurement of signal-to-noise ratios in mr images: In
uence of multichan-
nel coils, parallel imaging, and reconstruction lters. Journal of Magnetic Resonance
Imaging, 26(2):375{385, 2007.
[47] Rolf J. F. Ypma and Edward T. Bullmore. Statistical analysis of tract-tracing exper-
iments demonstrates a dense, complex cortical network in the mouse. PLOS Compu-
tational Biology, 12(9):1{22, 09 2016.
[48] N. T. Markov, P. Misery, A. Falchier, C. Lamy, J. Vezoli, R. Quilodran, M. A.
Gariel, P. Giroud, M. Ercsey-Ravasz, L. J. Pilaz, C. Huissoud, P. Barone, C. Dehay,
Z. Toroczkai, D. C. Van Essen, H. Kennedy, and K. Knoblauch. Weight consistency
species regularities of macaque cortical networks. Cerebral Cortex, 21(6):1254{1272,
91
2011.
[49] Ed Bullmore and Olaf Sporns. Complex brain networks: graph theoretical analysis
of structural and functional systems. Nature Reviews Neuroscience, 10:186 EP {, Feb
2009. Review Article.
[50] Olaf Sporns. Networks of the brain. MIT Press, Cambridge, MA, US, 2011.
[51] Houri Hintiryan, Lin Gou, Brian Zingg, Seita Yamashita, Hannah Lyden, Monica Song,
Arleen Grewal, Xinhai Zhang, Arthur Toga, and Hong-Wei Dong. Comprehensive con-
nectivity of the mouse main olfactory bulb: analysis and online digital atlas. Frontiers
in Neuroanatomy, 6:30, 2012.
[52] Richard H. Thompson and Larry W. Swanson. Hypothesis-driven structural connectiv-
ity analysis supports network over hierarchical model of brain architecture. Proceedings
of the National Academy of Sciences, 107(34):15235{15239, 2010.
[53] Swanson Larry W. Brain maps 4.0 structure of the rat brain: An open access atlas with
global nervous system nomenclature ontology and
atmaps. Journal of Comparative
Neurology, 526(6):935{943, 2018.
[54] Haberly Lewis B. and Price Joseph L. Association and commissural ber systems of the
olfactory cortex of the rat. i. systems originating in the piriform cortex and adjacent
areas. Journal of Comparative Neurology, 178(4):711{740, 1978.
[55] Haberly Lewis B. and Price Joseph L. Association and commissural ber systems of
the olfactory cortex of the rat ii. systems originating in the olfactory peduncle. Journal
of Comparative Neurology, 181(4):781{807, 1978.
[56] Luskin Marla B. and Price Joseph L. The topographic organization of associational
bers of the olfactory system in the rat, including centrifugal bers to the olfactory
bulb. Journal of Comparative Neurology, 216(3):264{291, 1983.
[57] Robert P. Vertes, Walter B. Hoover, Klara Szigeti-Buck, and Csaba Leranth. Nucleus
reuniens of the midline thalamus: Link between the medial prefrontal cortex and the
hippocampus. Brain Research Bulletin, 71(6):601 { 609, 2007.
[58] Wouterlood Floris G., Saldana Enrique, and Witter Menno P. Projection from the
nucleus reuniens thalami to the hippocampal region: Light and electron microscopic
tracing study in the rat with the anterograde tracer phaseolus vulgaris-leucoagglutinin.
Journal of Comparative Neurology, 296(2):179{203, 1990.
[59] Vertes Robert P. Analysis of projections from the medial prefrontal cortex to the thala-
mus in the rat, with emphasis on nucleus reuniens. Journal of Comparative Neurology,
92
442(2):163{187, 2001.
[60] Hiroshi T. Ito, Sheng-Jia Zhang, Menno P. Witter, Edvard I. Moser, and May-Britt
Moser. A prefrontal-thalamo-hippocampal circuit for goal-directed spatial navigation.
Nature, 522:50 EP {, May 2015. Article.
[61] McKenna James Timothy and Vertes Robert P. Aerent projections to nucleus re-
uniens of the thalamus. Journal of Comparative Neurology, 480(2):115{142, 2004.
[62] Menno P. Witter, Thanh P. Doan, Bente Jacobsen, Eirik S. Nilssen, and Shinya Ohara.
Architecture of the entorhinal cortex a review of entorhinal anatomy in rodents with
some comparative notes. Frontiers in Systems Neuroscience, 11:46, 2017.
[63] Wilson David I.G., Watanabe Sakurako, Milner Helen, and Ainge James A. Lateral en-
torhinal cortex is necessary for associative but not nonassociative recognition memory.
Hippocampus, 23(12):1280{1290, 2013.
[64] Albert Tsao, May-Britt Moser, and EdvardI. Moser. Traces of experience in the lateral
entorhinal cortex. Current Biology, 23(5):399 { 405, 2013.
[65] Susan R. Sesack, Ariel Y. Deutch, Robert H. Roth, and Benjamin S. Bunney. To-
pographical organization of the eerent projections of the medial prefrontal cortex in
the rat: An anterograde tract-tracing study with phaseolus vulgaris leucoagglutinin.
Journal of Comparative Neurology, 290(2):213{242, 1989.
[66] Cory M. Root, Christine A. Denny, Ren e Hen, and Richard Axel. The participation
of cortical amygdala in innate, odor-driven behavior. Nature, 515(7526):269{273, Nov
2014. 25383519[pmid].
[67] Michael Bubser and Ariel Y. Deutch. Thalamic paraventricular nucleus neurons col-
lateralize to innervate the prefrontal cortex and nucleus accumbens. Brain Research,
787(2):304 { 310, 1998.
[68] Sa Li and Gilbert J. Kirouac. Projections from the paraventricular nucleus of the
thalamus to the forebrain, with special emphasis on the extended amygdala. Journal
of Comparative Neurology, 506(2):263{287, 2008.
[69] Robert P. Vertes and Walter B. Hoover. Projections of the paraventricular and parate-
nial nuclei of the dorsal midline thalamus in the rat. Journal of Comparative Neurology,
508(2):212{237, 2008.
[70] Matthew P. Parsons, Sa Li, and Gilbert J. Kirouac. The paraventricular nucleus of
the thalamus as an interface between the orexin and cart peptides and the shell of the
nucleus accumbens. Synapse, 59(8):480{490, 2006.
93
[71] Tatiana D. Viena, Stephanie B. Linley, and Robert P. Vertes. Inactivation of nu-
cleus reuniens impairs spatial working memory and behavioral
exibility in the rat.
Hippocampus, 28(4):297{311, 2018.
[72] Karthik R. Ramanathan, Jingji Jin, Reed L. Ressler, and Stephen Maren. Nucleus
reuniens is required for encoding and retrieving precise contextual fear memories in
rats. bioRxiv, 2018.
[73] G.R.I. Barker and E.C. Warburton. A critical role for the nucleus reuniens in long-term,
but not short-term associative recognition memory formation. JournalofNeuroscience,
2018.
[74] Yuri B. Saalmann and Sabine Kastner. The cognitive thalamus. Frontiers in Systems
Neuroscience, 9:39, 2015.
[75] Anna Mitchell and Subhojit Chakraborty. What does the mediodorsal thalamus do?
Frontiers in Systems Neuroscience, 7:37, 2013.
[76] Anna S. Mitchell. The mediodorsal thalamus as a higher order thalamic relay nucleus
important for learning and decision-making. Neuroscience & Biobehavioral Reviews,
54:76 { 88, 2015. The Cognitive Thalamus.
[77] Sbastien Parnaudeau, Scott S. Bolkan, and Christoph Kellendonk. The mediodor-
sal thalamus: An essential partner of the prefrontal cortex for cognition. Biological
Psychiatry, 83(8):648 { 656, 2018. Circuits and Symptoms in Schizophrenia.
[78] Scott S. Bolkan, Joseph M. Stujenske, Sebastien Parnaudeau, Timothy J. Spellman,
Caroline Rauenbart, Atheir I. Abbas, Alexander Z. Harris, Joshua A. Gordon, and
Christoph Kellendonk. Thalamic projections sustain prefrontal activity during working
memory maintenance. Nature Neuroscience, 20:987 EP {, May 2017. Article.
[79] H.J. Groenewegen. Organization of the aerent connections of the mediodorsal thala-
mic nucleus in the rat, related to the mediodorsal-prefrontal topography. Neuroscience,
24(2):379 { 431, 1988.
[80] Fabien Alcaraz, Alain R. Marchand, Gilles Courtand, Etienne Coutureau, and Mathieu
Wol. Parallel inputs from the mediodorsal thalamus to the prefrontal cortex in the
rat. European Journal of Neuroscience, 44(3):1972{1986, 2016.
[81] M.S. Starr and M. Summerhayes. Role of the ventromedial nucleus of the thalamus in
motor behaviouri. eects of focal injections of drugs. Neuroscience, 10(4):1157 { 1169,
1983.
[82] M Jeljeli, C Strazielle, J Caston, and R Lalonde. Eects of ventrolateral-ventromedial
94
thalamic lesions on motor coordination and spatial orientation in rats. Neuroscience
Research, 47(3):309 { 316, 2003.
[83] Sakiko Honjoh, Shuntaro Sasai, Shannon S. Schiereck, Hirotaka Nagai, Giulio Tononi,
and Chiara Cirelli. Regulation of cortical activity and arousal by the matrix cells of
the ventromedial thalamic nucleus. Nature Communications, 9(1):2100, 2018.
[84] G. Daniel Weese, Janice M. Phillips, and Verity J. Brown. Attentional orienting is
impaired by unilateral lesions of the thalamic reticular nucleus in the rat. J Neurosci,
19(22):10135{10139, 1999.
[85] Allison Young and Ralf D. Wimmer. Implications for the thalamic reticular nucleus
in impaired attention and sleep in schizophrenia. Schizophrenia Research, 180:44 { 47,
2017. Pathologies of the Thalamus in Schizophrenia.
[86] Laura D Lewis, Jakob Voigts, Francisco J Flores, L Ian Schmitt, Matthew A Wilson,
Michael M Halassa, and Emery N Brown. Thalamic reticular nucleus induces fast and
local modulation of arousal state. eLife, 4:e08760, oct 2015.
[87] Gil Mandelbaum, Julian Taranda, Trevor Haynes, Daniel Hochbaum, Kee Wui Huang,
Minsuk Hyun, Kannan Umadevi Venkataraju, Christoph Straub, Wengang Wang,
Keiramarie Robertson, Pavel Osten, and Bernardo Sabatini. Distinct cortical-thalamic-
striatal circuits through the parafascicular nucleus. bioRxiv, 2018.
[88] Lucy Stiles and Paul F. Smith. The vestibularbasal ganglia connection: Balancing
motor control. Brain Research, 1597:180 { 188, 2015.
[89] Hong Lai, Toshiko Tsumori, Takashi Shiroyama, Shigefumi Yokota, Katsuma Nakano,
and Yukihiko Yasui. Morphological evidence for a vestibulo-thalamo-striatal pathway
via the parafascicular nucleus in the rat. Brain Research, 872(1):208 { 214, 2000.
[90] Eduardo F. Carvalho-Netto, Raquel C.R. Martinez, Marcus Vinicius C. Baldo, and
Newton Sabino Canteras. Evidence for the thalamic targets of the medial hypothalamic
defensive system mediating emotional memory to predatory threats. Neurobiology of
Learning and Memory, 93(4):479 { 486, 2010.
[91] Paul Rinne, Mursyida Hassan, Despina Goniotakis, Kiran Chohan, Pankaj Sharma,
Dawn Langdon, David Soto, and Paul Bentley. Triple dissociation of attention networks
in stroke according to lesion location. Neurology, 81(9):812{820, 2013.
[92] Pieter Voorn, Louk J.M.J Vanderschuren, Henk J Groenewegen, Trevor W Robbins,
and Cyriel M.A Pennartz. Putting a spin on the dorsalventral divide of the striatum.
Trends in Neurosciences, 27(8):468 { 474, 2004.
95
[93] Matthijs A.A. van der Meer, Adam Johnson, Neil C. Schmitzer-Torbert, and A. David
Redish. Triple dissociation of information processing in dorsal striatum, ventral stria-
tum, and hippocampus on a learned spatial decision task. Neuron, 67(1):25 { 32,
2010.
[94] Hong-Wei Dong, Gorica D Petrovich, and Larry W Swanson. Topography of projec-
tions from amygdala to bed nuclei of the stria terminalis. Brain Research Reviews,
38(1):192 { 246, 2001.
[95] Dennis C. Choi, Amy R. Furay, Nathan K. Evanson, Michelle M. Ostrander, Yvonne M.
Ulrich-Lai, and James P. Herman. Bed nucleus of the stria terminalis subregions
dierentially regulate hypothalamic{pituitary{adrenal axis activity: Implications for
the integration of limbic inputs. Journal of Neuroscience, 27(8):2025{2034, 2007.
[96] Teige P Sheehan, R.Andrew Chambers, and David S Russell. Regulation of aect by
the lateral septum: implications for neuropsychiatry. Brain Research Reviews, 46(1):71
{ 117, 2004.
[97] Georg M. Singewald, Alesja Rjabokon, Nicolas Singewald, and Karl Ebner. The mod-
ulatory role of the lateral septum on neuroendocrine and behavioral stress responses.
Neuropsychopharmacology, 36:793 EP {, Dec 2010. Original Article.
[98] Yan Liu, J. Thomas Curtis, and Zuoxin Wang. The prefrontal cortex—an
update auty - jour pmid - 11508730. Behavioral Neuroscience, 115(4):910{919, 2001.
[99] Larry J. Young and Zuoxin Wang. The neurobiology of pair bonding. Nature Neuro-
science, 7:1048 EP {, Sep 2004. Review Article.
[100] Eduardo E. Benarroch. Periaqueductal gray: An interface for behavioral control. Neu-
rology, 78(3):210{217, 2012.
[101] D.M.L. Vianna and M.L. Brand
~
Apoundso. Anatomical connections of the periaque-
ductal gray: specic neural substrates for dierent kinds of fear. Brazilian Journal of
Medical and Biological Research, 36:557 { 566, 05 2003.
[102] Robert R. Rozeske, Daniel Jercog, Nikolaos Karalis, Fabrice Chaudun, Suzana Khoder,
Delphine Girard, N^ anci Winke, and Cyril Herry. Prefrontal-periaqueductal gray-
projecting neurons mediate context fear discrimination. Neuron, 97(4):898{910.e6,
Feb 2018.
[103] D.M.L Vianna, J Landeira-Fernandez, and M.L Brand~ ao. Dorsolateral and ventral
regions of the periaqueductal gray matter are involved in distinct types of fear. Neu-
roscience & Biobehavioral Reviews, 25(7):711 { 719, 2001.
96
[104] Neda Assareh, Mahsa Sarrami, Pascal Carrive, and Gavan P. McNally. The organiza-
tion of defensive behavior elicited by optogenetic excitation of rat lateral or ventrolat-
eral periaqueductal gray. Behavioral Neuroscience, 130(4):406{414, 2016.
[105] Steven F. Maier and Linda R. Watkins. Stressor controllability and learned help-
lessness: The roles of the dorsal raphe nucleus, serotonin, and corticotropin-releasing
factor. Neuroscience & Biobehavioral Reviews, 29(4):829 { 841, 2005. Animal Models
of Depression and Antidepressant Activity.
[106] J. Amat, M. V. Baratta, E. Paul, S. T. Bland, L. R. Watkins, and S. F. Maier. Medial
prefrontal cortex determines how stressor controllability aects behavior and dorsal
raphe nucleus. Nature Neuroscience, 8:365 EP {, Feb 2005. Article.
[107] Alla Ignashchenkova, Peter W. Dicke, Thomas Haarmeier, and Peter Thier. Neuron-
specic contribution of the superior colliculus to overt and covert shifts of attention.
Nature Neuroscience, 7:56 EP {, Dec 2003. Article.
[108] Husam A. Gandhi, Neeraj J.; Katnani. Motor functions of the superior colliculus.
Annual Review of Neuroscience, 34(1):205{231, 2011.
[109] Eliana M. Klier, Hongying Wang, and J. Douglas Crawford. The superior colliculus
encodes gaze commands in retinal coordinates. Nature Neuroscience, 4:627 EP {, Jun
2001. Article.
[110] Richard J. Krauzlis, Lee P. Lovejoy, and Alexandre Znon. Superior colliculus and
visual spatial attention. Annual Review of Neuroscience, 36(1):165{182, 2013. PMID:
23682659.
[111] Gidon Felsen and Zachary F. Mainen. Neural substrates of sensory-guided locomotor
decisions in the rat superior colliculus. Neuron, 60(1):137 { 148, 2008.
[112] Veit Stuphorn, Erhard Bauswein, and Klaus-Peter Homann. Neurons in the primate
superior colliculus coding for arm movements in gaze-related coordinates. Journal of
Neurophysiology, 83(3):1283{1299, 2000. PMID: 10712456.
[113] Attila Nagy, Wolfgang Kruse, Silke Rottmann, Sabine Dannenberg, and Klaus-Peter
Homann. Somatosensory-motor neuronal activity in the superior colliculus of the
primate. Neuron, 52(3):525 { 534, 2006.
[114] Stephan Lammel, Byung Kook Lim, Chen Ran, Kee Wui Huang, Michael J. Betley,
Kay M. Tye, Karl Deisseroth, and Robert C. Malenka. Input-specic control of reward
and aversion in the ventral tegmental area. Nature, 491:212 EP {, Oct 2012. Article.
[115] Jeremiah Y. Cohen, Sebastian Haesler, Linh Vong, Bradford B. Lowell, and Naoshige
97
Uchida. Neuron-type-specic signals for reward and punishment in the ventral tegmen-
tal area. Nature, 482:85 EP {, Jan 2012.
[116] YY Lai and JM Siegel. Muscle tone suppression and stepping produced by stimu-
lation of midbrain and rostral pontine reticular formation. Journal of Neuroscience,
10(8):2727{2734, 1990.
[117] Margarita L. Rodrigo-Angulo, Susana Heredero, Elisia Rodrguez-Veiga, and Fernando
Reinoso-Surez. Gabaergic and non-gabaergic thalamic, hypothalamic and basal fore-
brain projections to the ventral oral pontine reticular nucleus: Their implication in
rem sleep modulation. Brain Research, 1210:116 { 125, 2008.
[118] Kazue Semba. Aminergic and cholinergic aerents to rem sleep induction regions of the
pontine reticular formation in the rat. Journal of Comparative Neurology, 330(4):543{
556, 1993.
[119] Karina Genaro Borelli, Ana Carolina Grgaro, Jlia Maria dos Santos, and Marcus Lira
Brando. Eects of inactivation of serotonergic neurons of the median raphe nucleus
on learning and performance of contextual fear conditioning. Neuroscience Letters,
387(2):105 { 110, 2005.
[120] Yu Ohmura, Takeshi Izumi, Taku Yamaguchi, Iku Tsutsui-Kimura, Takayuki Yoshida,
and Mitsuhiro Yoshioka. The serotonergic projection from the median raphe nucleus
to the ventral hippocampus is involved in the retrieval of fear memory through the
corticotropin-releasing factor type 2 receptor. Neuropsychopharmacology, 35:1271 EP
{, Jan 2010. Original Article.
[121] Iskra PollakDorocic, Daniel Frth, Yang Xuan, Yvonne Johansson, Laura Pozzi, Gilad
Silberberg, Marie Carln, and Konstantinos Meletis. A whole-brain atlas of inputs to
serotonergic neurons of the dorsal and median raphe nuclei. Neuron, 83(3):663 { 678,
2014.
98
Appendix
5.1 Prefrontal Cortex Region Name Label
99
5.2 Thalamic Region Name Label
100
5.3 Table of Mouse Brain Structure Names and their Hierarchy
Groups
ARA Name Hierarchy Group Cortical Layers
1 A13 HY NA
2 AAA AMY NA
3 ACAd1 CTX Supercial
4 ACAd2/3 CTX Supercial
5 ACAd5 CTX Deep
6 ACAd6 CTX Deep
7 ACAd6a CTX Deep
8 ACAd6b CTX Deep
9 ACAv1 CTX Supercial
10 ACAv2/3 CTX Supercial
11 ACAv5 CTX Deep
12 ACAv6a CTX Deep
13 ACAv6b CTX Deep
14 ACB STR NA
15 ACVII HB NA
16 AD TH NA
17 ADHp HY NA
18 ADP HY NA
19 AHNa HY NA
20 AHNc HY NA
21 AHNp HY NA
22 AId1 CTX Supercial
23 AId2/3 CTX Supercial
24 AId5 CTX Deep
25 AId6 CTX Deep
26 AId6a CTX Deep
27 AIp1 CTX Supercial
28 AIp2/3 CTX Supercial
29 AIp5 CTX Deep
30 AIp6a CTX Deep
31 AIp6b CTX Deep
32 AIv1 CTX Supercial
33 AIv2/3 CTX Supercial
34 AIv5 CTX Deep
35 AIv6a CTX Deep
36 AIv6b CTX Deep
37 AMB HB NA
38 AMBd HB NA
101
ARA Name Hierarchy Group Cortical Layers
39 AMBv HB NA
40 AMd TH NA
41 AMv TH NA
42 ANCr2mo CB NA
43 ANcr1gr CB NA
44 ANcr1mo CB NA
45 ANcr2gr CB NA
46 AOB CTX Deep
47 AOBgl CTX Deep
48 AOBgr CTX Deep
49 AOBmi CTX Deep
50 AON CTX Deep
51 AON1 CTX Deep
52 AONd CTX Deep
53 AONe CTX Deep
54 AONl CTX Deep
55 AONm CTX Deep
56 AONpv CTX Deep
57 AP HB NA
58 APN MB NA
59 AQ VENTRICLE NA
60 ARH HY NA
61 AT MB NA
62 AUDPo1 CTX Supercial
63 AUDPo2/3 CTX Supercial
64 AUDPo4 CTX Deep
65 AUDPo5 CTX Deep
66 AUDPo6a CTX Deep
67 AUDd1 CTX Supercial
68 AUDd2/3 CTX Supercial
69 AUDd4 CTX Deep
70 AUDd5 CTX Deep
71 AUDd6a CTX Deep
72 AUDd6b CTX Deep
73 AUDp1 CTX Supercial
74 AUDp2/3 CTX Supercial
75 AUDp4 CTX Deep
76 AUDp5 CTX Deep
77 AUDp6a CTX Deep
78 AUDp6b CTX Deep
79 AUDpo6b CTX Deep
102
ARA Name Hierarchy Group Cortical Layers
80 AUDv1 CTX Supercial
81 AUDv2/3 CTX Supercial
82 AUDv4 CTX Deep
83 AUDv5 CTX Deep
84 AUDv6a CTX Deep
85 AUDv6b CTX Deep
86 AV TH NA
87 AVP HY NA
88 AVPV HY NA
89 Ald6b CTX Deep
90 B HB NA
91 BA AMY NA
92 BLAa AMY NA
93 BLAp AMY NA
94 BLAv AMY NA
95 BMAa AMY NA
96 BMAp AMY NA
97 BSTAL STR NA
98 BSTAM STR NA
99 BSTBAC STR NA
100 BSTOV STR NA
101 BSTd STR NA
102 BSTdm STR NA
103 BSTfu STR NA
104 BSTif STR NA
105 BSTju STR NA
106 BSTmg STR NA
107 BSTpr STR NA
108 BSTrh STR NA
109 BSTse STR NA
110 BSTtr STR NA
111 BSTv STR NA
112 CA1slm HPF NA
113 CA1so HPF NA
114 CA1sp HPF NA
115 CA1sr HPF NA
116 CA2slm HPF NA
117 CA2so HPF NA
118 CA2sp HPF NA
119 CA2sr HPF NA
120 CA3slm HPF NA
103
ARA Name Hierarchy Group Cortical Layers
121 CA3slu HPF NA
122 CA3so HPF NA
123 CA3sp HPF NA
124 CA3sr HPF NA
125 CEAc AMY NA
126 CEAl AMY NA
127 CEAm AMY NA
128 CENT2gr CB NA
129 CENT2mo CB NA
130 CENT3gr CB NA
131 CENT3mo CB NA
132 CL TH NA
133 CLA CTX Deep
134 CLI MB NA
135 CM TH NA
136 CNlam HB NA
137 CNspg HB NA
138 COAa1 CTX Supercial
139 COAa2 CTX Supercial
140 COApI1 CTX Supercial
141 COApI2 CTX Supercial
142 COApl3 CTX Deep
143 COApm1 CTX Supercial
144 COApm2 CTX Supercial
145 COApm3 CTX Deep
146 COPYgr CB NA
147 COPYmo CB NA
148 CP STR NA
149 CSl HB NA
150 CSm HB NA
151 CU HB NA
152 CUL4/5gr CB NA
153 CUL4/5mo CB NA
154 CULmo CB NA
155 CUN MB NA
156 DCO HB NA
157 DECgr CB NA
158 DECmo CB NA
159 DGmo HPF NA
160 DGpo HPF NA
161 DGsg HPF NA
104
ARA Name Hierarchy Group Cortical Layers
162 DMHa HY NA
163 DMHp HY NA
164 DMHv HY NA
165 DMX HB NA
166 DN CB NA
167 DP1 CTX Supercial
168 DP2/3 CTX Supercial
169 DP5 CTX Deep
170 DP6a CTX Deep
171 DR MB NA
172 DTN HB NA
173 ECT1 CTX Supercial
174 ECT2/3 CTX Supercial
175 ECT5 CTX Deep
176 ECT6a CTX Deep
177 ECT6b CTX Deep
178 ECU HB NA
179 ENTl1 CTX Supercial
180 ENTl2 CTX Supercial
181 ENTl2/3 CTX Supercial
182 ENTl2a CTX Supercial
183 ENTl2b CTX Supercial
184 ENTl3 CTX Supercial
185 ENTl4 CTX Deep
186 ENTl4/5 CTX Deep
187 ENTl5 CTX Deep
188 ENTl6 CTX Deep
189 ENTl6a CTX Deep
190 ENTl6b CTX Deep
191 ENTm1 CTX Supercial
192 ENTm2 CTX Supercial
193 ENTm2a CTX Supercial
194 ENTm2b CTX Supercial
195 ENTm3 CTX Supercial
196 ENTm4 CTX Deep
197 ENTm5 CTX Deep
198 ENTm6 CTX Deep
199 ENTmv1 CTX Supercial
200 ENTmv2 CTX Supercial
201 ENTmv3 CTX Supercial
202 ENTmv5/6 CTX Deep
105
ARA Name Hierarchy Group Cortical Layers
203 EPd CTX Deep
204 EW MB NA
205 Epv CTX Deep
206 FC HPF NA
207 FLgr CB NA
208 FLmo CB NA
209 FN CB NA
210 FOTUgr CB NA
211 FOTUmo CB NA
212 FRP1 CTX Supercial
213 FRP2/3 CTX Supercial
214 FS STR NA
215 GR HB NA
216 GRN HB NA
217 GU1 CTX Supercial
218 GU2/3 CTX Supercial
219 GU4 CTX Deep
220 GU5 CTX Deep
221 GU6a CTX Deep
222 GU6b CTX Deep
223 Gpe STR NA
224 Gpi STR NA
225 HY HY NA
226 IA AMY NA
227 IAD TH NA
228 IAM TH NA
229 ICB HB NA
230 ICc MB NA
231 ICd MB NA
232 ICe MB NA
233 IF MB NA
234 IG HPF NA
235 IGL TH NA
236 III MB NA
237 ILA1 CTX Supercial
238 ILA2 CTX Supercial
239 ILA2/3 CTX Supercial
240 ILA3 CTX Supercial
241 ILA5 CTX Deep
242 ILA6a CTX Deep
243 ILA6b CTX Deep
106
ARA Name Hierarchy Group Cortical Layers
244 IMD TH NA
245 INC MB NA
246 IO HB NA
247 IP CB NA
248 IPN MB NA
249 IRN HB NA
250 ISN HB NA
251 IV MB NA
252 KF HB NA
253 LA AMY NA
254 LAV HB NA
255 LC HB NA
256 LD TH NA
257 LDT HB NA
258 LGd TH NA
259 LGv TH NA
260 LH TH NA
261 LHA HY NA
262 LIN HB NA
263 LINGgr CB NA
264 LINGmo CB NA
265 LM HY NA
266 LP TH NA
267 LPO HY NA
268 LRNm HB NA
269 LRNp HB NA
270 LSc STR NA
271 LSr STR NA
272 LSv STR NA
273 LT MB NA
274 MA STR NA
275 MARN HB NA
276 MB MB NA
277 MDI TH NA
278 MDRNd HB NA
279 MDRNv HB NA
280 MDc TH NA
281 MDm TH NA
282 ME HY NA
283 MEAad AMY NA
284 MEAav AMY NA
107
ARA Name Hierarchy Group Cortical Layers
285 MEApda AMY NA
286 MEApdb AMY NA
287 MEApdc AMY NA
288 MEApv AMY NA
289 MEPO HY NA
290 MEV MB NA
291 MGd TH NA
292 MGm TH NA
293 MGv TH NA
294 MH TH NA
295 MM HY NA
296 MMme HY NA
297 MOB CTX Deep
298 MOBgl CTX Deep
299 MOBgr CTX Deep
300 MOBipl CTX Deep
301 MOBmi CTX Deep
302 MOBopl CTX Deep
303 MOp1 CTX Supercial
304 MOp2/3 CTX Supercial
305 MOp5 CTX Deep
306 MOp6 CTX Deep
307 MOp6a CTX Deep
308 MOp6b CTX Deep
309 MOs1 CTX Supercial
310 MOs2/3 CTX Supercial
311 MOs5 CTX Deep
312 MOs6 CTX Deep
313 MOs6a CTX Deep
314 MOs6b CTX Deep
315 MPN HY NA
316 MPNc HY NA
317 MPNl HY NA
318 MPNm HY NA
319 MPO HY NA
320 MPT MB NA
321 MRN MB NA
322 MS STR NA
323 MV HB NA
324 MY HB NA
325 NB MB NA
108
ARA Name Hierarchy Group Cortical Layers
326 NC HY NA
327 ND STR NA
328 NDB STR NA
329 NI HB NA
330 NLLd HB NA
331 NLLh HB NA
332 NLLv HB NA
333 NLOT1 CTX Supercial
334 NLOT2 CTX Supercial
335 NLOT3 CTX Deep
336 NODgr CB NA
337 NODmo CB NA
338 NOT MB NA
339 NPC MB NA
340 NR HB NA
341 NTB HB NA
342 NTSce HB NA
343 NTSco HB NA
344 NTSge HB NA
345 NTSl HB NA
346 NTSm HB NA
347 OP MB NA
348 ORBl1 CTX Supercial
349 ORBl2/3 CTX Supercial
350 ORBl5 CTX Deep
351 ORBl6 CTX Deep
352 ORBl6a CTX Deep
353 ORBl6b CTX Deep
354 ORBm1 CTX Supercial
355 ORBm2 CTX Supercial
356 ORBm2/3 CTX Supercial
357 ORBm3 CTX Supercial
358 ORBm5 CTX Deep
359 ORBm6 CTX Deep
360 ORBm6a CTX Deep
361 ORBvl1 CTX Supercial
362 ORBvl2/3 CTX Supercial
363 ORBvl5 CTX Deep
364 ORBvl6 CTX Deep
365 ORBvl6a CTX Deep
366 ORBvl6b CTX Deep
109
ARA Name Hierarchy Group Cortical Layers
367 OT1 CTX Supercial
368 OT2 CTX Supercial
369 OT3 CTX Deep
370 OV HY NA
371 PA AMY NA
372 PAA1 CTX Supercial
373 PAA2 CTX Supercial
374 PAA3 CTX Deep
375 PAG MB NA
376 PAR1 HPF NA
377 PAR2 HPF NA
378 PAR3 HPF NA
379 PARN HB NA
380 PAS HB NA
381 PBG MB NA
382 PBllc HB NA
383 PBlld HB NA
384 PBlle HB NA
385 PBlls HB NA
386 PBllv HB NA
387 PBmm HB NA
388 PBmme HB NA
389 PCG HB NA
390 PCN TH NA
391 PERI1 CTX Supercial
392 PERI2/3 CTX Supercial
393 PERI5 CTX Deep
394 PERI6a CTX Deep
395 PERI6b CTX Deep
396 PF TH NA
397 PFLgr CB NA
398 PFLmo CB NA
399 PG HB NA
400 PGRNd HB NA
401 PGRNl HB NA
402 PH HY NA
403 PIR1 CTX Supercial
404 PIR2 CTX Supercial
405 PIR3 CTX Deep
406 PL1 CTX Supercial
407 PL2 CTX Supercial
110
ARA Name Hierarchy Group Cortical Layers
408 PL2/3 CTX Supercial
409 PL3 CTX Supercial
410 PL5 CTX Deep
411 PL6 CTX Deep
412 PL6a CTX Deep
413 PL6b CTX Deep
414 PMd HY NA
415 PMv HY NA
416 PO TH NA
417 POL TH NA
418 PONS HB NA
419 POR HB NA
420 POST1 HPF NA
421 POST2 HPF NA
422 POST3 HPF NA
423 PP TH NA
424 PPN MB NA
425 PPT MB NA
426 PPY HB NA
427 PR TH NA
428 PRC TH NA
429 PRE1 HPF NA
430 PRE2 HPF NA
431 PRE3 HPF NA
432 PRMgr CB NA
433 PRMmo CB NA
434 PRNc HB NA
435 PRNr HB NA
436 PRP HB NA
437 PS HY NA
438 PST HY NA
439 PSTN HY NA
440 PSV HB NA
441 PT TH NA
442 PTLp1 CTX Supercial
443 PTLp2/3 CTX Supercial
444 PTLp4 CTX Deep
445 PTLp5 CTX Deep
446 PTLp6a CTX Deep
447 PTLp6b CTX Deep
448 PVHa HY NA
111
ARA Name Hierarchy Group Cortical Layers
449 PVHap HY NA
450 PVHdp HY NA
451 PVHf HY NA
452 PVHlp HY NA
453 PVHmm HY NA
454 PVHmpd HY NA
455 PVHmpv HY NA
456 PVHpml HY NA
457 PVHpmm HY NA
458 PVHpv HY NA
459 PVRpd HY NA
460 PVT TH NA
461 PVa HY NA
462 PVi HY NA
463 PVp HY NA
464 PVpo HY NA
465 PYRgr CB NA
466 PYRmo CB NA
467 RCH HY NA
468 RE TH NA
469 RH TH NA
470 RL MB NA
471 RM HB NA
472 RN MB NA
473 RO HB NA
474 RPA HB NA
475 RPO HB NA
476 RR MB NA
477 RSPagl1 CTX Supercial
478 RSPagl2/3 CTX Supercial
479 RSPagl5 CTX Deep
480 RSPagl6a CTX Deep
481 RSPagl6b CTX Deep
482 RSPd1 CTX Supercial
483 RSPd2/3 CTX Supercial
484 RSPd5 CTX Deep
485 RSPd6a CTX Deep
486 RSPd6b CTX Deep
487 RSPv1 CTX Supercial
488 RSPv2 CTX Supercial
489 RSPv2/3 CTX Supercial
112
ARA Name Hierarchy Group Cortical Layers
490 RSPv5 CTX Deep
491 RSPv6a CTX Deep
492 RSPv6b CTX Deep
493 RT TH NA
494 SAG MB NA
495 SBPV HY NA
496 SCH HY NA
497 SCMig-a MB NA
498 SCMig-b MB NA
499 SCMig-c MB NA
500 SCmdg MB NA
501 SCmdw MB NA
502 SCmig MB NA
503 SCmiw MB NA
504 SCsop MB NA
505 SCssg MB NA
506 SCszo MB NA
507 SF STR NA
508 SFO HY NA
509 SFO/V3 VENTRICLE NA
510 SG HB NA
511 SGN TH NA
512 SH STR NA
513 SI STR NA
514 SIMgr CB NA
515 SIMmo CB NA
516 SLC HB NA
517 SLD HB NA
518 SMT TH NA
519 SNc MB NA
520 SNr MB NA
521 SO HY NA
522 SOCl HB NA
523 SOCm HB NA
524 SPA TH NA
525 SPFm TH NA
526 SPFp TH NA
527 SPIV HB NA
528 SPVOcdm HB NA
529 SPVOmdmd HB NA
530 SPVOmdmv HB NA
113
ARA Name Hierarchy Group Cortical Layers
531 SPVOrdm HB NA
532 SPVOvl HB NA
533 SPVc HB NA
534 SPVi HB NA
535 SSp-un1 CTX Supercial
536 SSp-un2/3 CTX Supercial
537 SSp-un4 CTX Deep
538 SSp-un5 CTX Deep
539 SSp-un6a CTX Deep
540 SSp-un6b CTX Deep
541 SSp1 CTX Supercial
542 SSp2/3 CTX Supercial
543 SSp4 CTX Deep
544 SSp5 CTX Deep
545 SSp6a CTX Deep
546 SSp6b CTX Deep
547 SSpbfd1 CTX Supercial
548 SSpbfd2/3 CTX Supercial
549 SSpbfd4 CTX Deep
550 SSpbfd5 CTX Deep
551 SSpbfd6a CTX Deep
552 SSpbfd6b CTX Deep
553 SSpll1 CTX Supercial
554 SSpll2/3 CTX Supercial
555 SSpll4 CTX Deep
556 SSpll5 CTX Deep
557 SSpll6a CTX Deep
558 SSpll6b CTX Deep
559 SSpm1 CTX Supercial
560 SSpm2/3 CTX Supercial
561 SSpm4 CTX Deep
562 SSpm5 CTX Deep
563 SSpm6a CTX Deep
564 SSpm6b CTX Deep
565 SSpn1 CTX Supercial
566 SSpn2/3 CTX Supercial
567 SSpn4 CTX Deep
568 SSpn5 CTX Deep
569 SSpn6a CTX Deep
570 SSpn6b CTX Deep
571 SSptr1 CTX Supercial
114
ARA Name Hierarchy Group Cortical Layers
572 SSptr2/3 CTX Supercial
573 SSptr4 CTX Deep
574 SSptr5 CTX Deep
575 SSptr6a CTX Deep
576 SSptr6b CTX Deep
577 SSpul1 CTX Supercial
578 SSpul2/3 CTX Supercial
579 SSpul4 CTX Deep
580 SSpul5 CTX Deep
581 SSpul6a CTX Deep
582 SSpul6b CTX Deep
583 SSs1 CTX Supercial
584 SSs2/3 CTX Supercial
585 SSs4 CTX Deep
586 SSs5 CTX Deep
587 SSs6a CTX Deep
588 SSs6b CTX Deep
589 STN HY NA
590 SUBdm HPF NA
591 SUBdsp HPF NA
592 SUBdsr HPF NA
593 SUBvm HPF NA
594 SUBvsp HPF NA
595 SUBvsr HPF NA
596 SUMl HY NA
597 SUMm HY NA
598 SUT HB NA
599 SUV HB NA
600 Scigb MB NA
601 SubG TH NA
602 TEa1 CTX Supercial
603 TEa2/3 CTX Supercial
604 TEa4 CTX Deep
605 TEa5 CTX Deep
606 TEa6a CTX Deep
607 TEa6b CTX Deep
608 TMd HY NA
609 TMv HY NA
610 TR1 CTX Supercial
611 TR2 CTX Supercial
612 TR3 CTX Deep
115
ARA Name Hierarchy Group Cortical Layers
613 TRN HB NA
614 TRS STR NA
615 TTd1 CTX Supercial
616 TTd2 CTX Supercial
617 TTd3 CTX Deep
618 TTd4 CTX Deep
619 TTv1 CTX Supercial
620 TTv2 CTX Supercial
621 TTv3 CTX Deep
622 TU HY NA
623 UVUgr CB NA
624 UVUmo CB NA
625 V HB NA
626 V3 VENTRICLE NA
627 V4 VENTRICLE NA
628 V4r VENTRICLE NA
629 VAL TH NA
630 VCO HB NA
631 VI HB NA
632 VII HB NA
633 VIIIn WHITE MATTER NA
634 VISC1 CTX Supercial
635 VISC2/3 CTX Supercial
636 VISC4 CTX Deep
637 VISC5 CTX Deep
638 VISC6a CTX Deep
639 VISC6b CTX Deep
640 VISal1 CTX Supercial
641 VISal2/3 CTX Supercial
642 VISal4 CTX Deep
643 VISal5 CTX Deep
644 VISal6a CTX Deep
645 VISal6b CTX Deep
646 VISam1 CTX Supercial
647 VISam2/3 CTX Supercial
648 VISam4 CTX Deep
649 VISam5 CTX Deep
650 VISam6a CTX Deep
651 VISam6b CTX Deep
652 VISl1 CTX Supercial
653 VISl2/3 CTX Supercial
116
ARA Name Hierarchy Group Cortical Layers
654 VISl4 CTX Deep
655 VISl5 CTX Deep
656 VISl6a CTX Deep
657 VISl6b CTX Deep
658 VISp1 CTX Supercial
659 VISp2/3 CTX Supercial
660 VISp4 CTX Deep
661 VISp5 CTX Deep
662 VISp6a CTX Deep
663 VISp6b CTX Deep
664 VISpl1 CTX Supercial
665 VISpl2/3 CTX Supercial
666 VISpl4 CTX Deep
667 VISpl5 CTX Deep
668 VISpl6a CTX Deep
669 VISpl6b CTX Deep
670 VISpm1 CTX Supercial
671 VISpm2/3 CTX Supercial
672 VISpm4 CTX Deep
673 VISpm5 CTX Deep
674 VISpm6a CTX Deep
675 VISpm6b CTX Deep
676 VL VENTRICLE NA
677 VLPO HY NA
678 VM TH NA
679 VMHa HY NA
680 VMHc HY NA
681 VMHdm HY NA
682 VMHvl HY NA
683 VPL TH NA
684 VPLpc TH NA
685 VPM TH NA
686 VPMpc TH NA
687 VTA MB NA
688 VTN MB NA
689 Vlln WHITE MATTER NA
690 X HB NA
691 XII HB NA
692 Y HB NA
693 ZI HY NA
694 ZIA13 HY NA
117
ARA Name Hierarchy Group Cortical Layers
695 ZIFF HY NA
696 aco WHITE MATTER NA
697 act WHITE MATTER NA
698 alv WHITE MATTER NA
699 amc WHITE MATTER NA
700 arb WHITE MATTER NA
701 bic WHITE MATTER NA
702 bsc WHITE MATTER NA
703 centralcanal VENTRICLE NA
704 cic WHITE MATTER NA
705 cpd WHITE MATTER NA
706 csc WHITE MATTER NA
707 df WHITE MATTER NA
708 dhc WHITE MATTER NA
709 dscp WHITE MATTER NA
710 ect WHITE MATTER NA
711 em WHITE MATTER NA
712 fa WHITE MATTER NA
713 WHITE MATTER NA
714 fr WHITE MATTER NA
715 fx WHITE MATTER NA
716 gVLLn WHITE MATTER NA
717 hbc WHITE MATTER NA
718 hsc WHITE MATTER NA
719 icp WHITE MATTER NA
720 im WHITE MATTER NA
721 int WHITE MATTER NA
722 isl WHITE MATTER NA
723 islm WHITE MATTER NA
724 ll WHITE MATTER NA
725 lot WHITE MATTER NA
726 mcp WHITE MATTER NA
727 ml WHITE MATTER NA
728 mlf WHITE MATTER NA
729 moV WHITE MATTER NA
730 mp WHITE MATTER NA
731 mtg WHITE MATTER NA
732 mtt WHITE MATTER NA
733 mtv WHITE MATTER NA
734 och WHITE MATTER NA
735 onl WHITE MATTER NA
118
ARA Name Hierarchy Group Cortical Layers
736 opt WHITE MATTER NA
737 pc WHITE MATTER NA
738 pm WHITE MATTER NA
739 py WHITE MATTER NA
740 rust WHITE MATTER NA
741 sAMY AMY NA
742 scp WHITE MATTER NA
743 sctv WHITE MATTER NA
744 sm WHITE MATTER NA
745 smd WHITE MATTER NA
746 st WHITE MATTER NA
747 ts WHITE MATTER NA
748 tsp WHITE MATTER NA
749 vhc WHITE MATTER NA
750 von WHITE MATTER NA
751 vtd WHITE MATTER NA
119

Abstract (if available)

Abstract In numerous studies reported in rodents and primates, the mPFC (medial prefrontal cortex) appears to serve as a critical hub that provides an interface for interactions among the neocortex, amygdala, and hippocampus, thus, bridging the communication throughout the entire cerebrum. In turn, the mPFC generates wide range projections to the thalamus, striatum, and hypothalamus to regulate motivated behavior. Despite its scientific and clinical importance, a whole brain wide examination and comparison of neuronal connectivity of the mouse mPFC cortices has not yet been conducted. Additionally, the organization of neural circuits associated with individual mPFC areas into global neural networks remains undetermined. In this thesis work, we quantitatively characterize the global structural connectome of mPFC from terabyte range microscopic imaging data. We develop a full image processing and statistical analysis pipeline over a massively parallel computing infrastructure to enable connectivity mapping at large scale. An accompanying interactive software has also been built to support data visualization and quality assurance. We expect our systematic and quantitative analysis of detailed connectivity of the mPFC to provide a framework for exploring mPFC connectional functions and its pathologies in neurological/neuropsychiatric diseases.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Mapping multi-scale connectivity of the mouse posterior parietal cortex

PDF

Molecular diversity of neurons in developing and mature brain circuits

PDF

Subnetwork organization of the superior colliculus and visual system in the mouse brain

PDF

Development of a toolbox for global functional brain imaging of wake and sleep states in zebrafish

PDF

Functional connectivity analysis and network identification in the human brain

PDF

Neural circuits underlying the modulation and impact of defensive behaviors

PDF

Neural circuits control and modulate innate defensive behaviors

PDF

Applications of graph theory to brain connectivity analysis

PDF

Measuring functional connectivity of the brain

PDF

Contextual modulation of sensory processing via the pulvinar nucleus

PDF

Shortcomings of the genetic risk score in the analysis of disease-related quantitative traits

PDF

Computational investigation of cholinergic modulation of the hippocampus and directional encoding of touch in somatosensory cortex

PDF

Fathers’ social-cognitive network connectivity: associations with prenatal father-infant attachment and postpartum parenting behavior

PDF

Balancing security and performance of network request-response protocols

PDF

Detection, classification and functional annotation of mouse L1 retrotransposon promoters

PDF

Diffusion MRI of the human brain: signal modeling and quantitative analysis

PDF

Exploiting novel transport properties of adeno-associated virus for circuit mapping and manipulation

PDF

Functional models of fMRI BOLD signal in the visual cortex

PDF

A million-plus neuron model of the hippocampal dentate gyrus: role of topography, inhibitory interneurons, and excitatory associational circuitry in determining spatio-temporal dynamics of granul...

PDF

Computational methods for translation regulation analysis from Ribo-seq data

Asset Metadata

Creator Zhu, Muye (author)

Core Title Quantitative analysis of mouse medial prefrontal cortex whole brain structural connectivity

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Neuroscience

Publication Date 10/15/2018

Defense Date 08/29/2018

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

Tag medial prefrontal cortex,mouse,mPFC,OAI-PMH Harvest,quantitative,whole brain connectivity

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Zhang, Li (committee chair), Dong, Hongwei (committee member), Hires, Samuel Andrew (committee member), Holschneider, Daniel (committee member), Swanson, Larry (committee member)

Creator Email muyezhu@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c89-88877

Unique identifier UC11675752

Identifier etd-ZhuMuye-6826.pdf (filename),usctheses-c89-88877 (legacy record id)

Legacy Identifier etd-ZhuMuye-6826.pdf

Dmrecord 88877

Document Type Dissertation

Format application/pdf (imt)

Rights Zhu, Muye

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 a...

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