University of Southern California Dissertations and Theses

Value-based decision-making in complex choice: brain regions involved and implications of age

(USC Thesis Other)

Value-based decision-making in complex choice: brain regions involved and implications of age

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content VALUE-BASED DECISION-MAKING IN COMPLEX CHOICE: BRAIN REGIONS INVOLVED AND IMPLICATIONS OF AGE by Jekaterina Zyuzin A Dissertation Presented to the FACULTY OF THE GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA in Partial Fulllment of the Requirements for the Degree of Doctor of Philosophy Neuroscience May 2020 Acknowledgments I believe that I am not an exception, and that everyone who has gone through the graduate school and reached a culmination point had to, at least at that point, try to comprehend the amount of work that went into this process. It is overwhelming for me to look back and try to see the path as a linear progression towards a goal, as it was anything but a straight line, and it was paved with uncertainty and countless surprises. In my case, everything that lead to this moment was a big roller coaster ride which included the life that lead to the PhD and the life during the PhD. In this process, I was constantly presented with numerous challenges, but what is really remarkable, is that those challenges brought opportunities, as well as, second chances that I could not have dreamed of earlier and that changed my life for the best. Along the way I have met and became friends with an incredible amount of wonderful bright inspiring and amazing people from all over the world. I am not able to put everyone's name onto this manuscript, as it will take more space than the dissertation itself, but believe me, all of you mattered, and because of you I was able to succeed, and ii you are eternally engraved in my heart, and I will always be grateful for you. Importantly, none of this work would have been possible without the unconditional love and support from my family, my son Nikita, David F. and my friends, who picked me up when I fell and who brightened this path on some days that were gloomy. I am forever grateful to Isabelle Brocas (PI), John Monterosso, Giorgio Coricelli and Katie Page, as I was incredibly lucky to have such an outstanding committee of most supportive intelligent and fascinat- ing people, whose time and dedication made it possible to complete this project despite the odds and inevitable pressure of time. To make each contribution to this project clear, I need to outline several important steps. First, the study was conducted with the University of Southern California IRB approval number UP-13-00235. We acknowledge the nancial support of award 1R21AG046917-01A1 from the National Institute on Aging. Second, main contributors to this research were Jekaterina Zyuzin, Isabelle Brocas (PI) and John Monterosso (Co-PI). Third, for Chapter 1 of this study Dalton Combs and Juan Carrillo provided considerable contribution by designing the task and providing guidance and feedback throughout the data col- iii lection and analysis. Moreover, Dalton designed the nipype pipeline for analysis and conducted a preliminary study (Combs, 2016, Neu- roEconomic Mechanisms for Valuing Complex Options) on which the rest of the experiments were constructed. Fourth, in addition, for Chapter 2 Mara Mather, helped with the subjects' pool and pro- vided guidance regarding OA strategies for analysis. Fifth, we are also grateful to members of the Los Angeles Behavioral Economics Laboratory (LABEL), Giorgio Coricelli and Katie Page for their in- sights and comments in the various phases of the project. We also thank especially Xochitl Cordova, Niree Kodaverdian, Calvin Leather and James Melrose for excellent research assistance on specic aspects of the project, as well as, Eustace Hsu, Natalie Poppa, Nina Christie, Xiaobei Zhang, Milad Kassaie, Max Ibrahimzade, Sara Doyle, Madi- son Burger, Jared Gilbert, Cesar Sul and many others for help and encouragement throughout these work. iv Contents 1 The neural correlates of complex value 1 1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . 1 1.2 Materials and methods . . . . . . . . . . . . . . . . . . 5 1.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 1.3.1 Behavioral measures . . . . . . . . . . . . . . . 23 1.3.2 Regions tracking subjective value . . . . . . . . 27 1.3.3 Regions responding dierentially to conditions 43 1.3.4 Regions tracking diculty/salience . . . . . . . 50 1.3.5 Connectivity analysis . . . . . . . . . . . . . . . 53 1.3.6 Correlates of behavioral markers . . . . . . . . . 57 1.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . 63 1.5 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . 72 2 The impact of aging on value representation 94 v 2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . 94 2.2 Materials and methods . . . . . . . . . . . . . . . . . . 99 2.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 114 2.3.1 Behavioral measures . . . . . . . . . . . . . . . 114 2.3.2 Overall dierences between OA and YA . . . . 119 2.3.3 Evaluation of YA core task-related regions . . . 122 2.3.4 Connectivity analysis . . . . . . . . . . . . . . . 128 2.3.5 Neural correlates of behavior . . . . . . . . . . . 132 2.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . 137 2.5 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . 141 Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . 143 Chapter 1 The neural correlates of complex value 1.1 Introduction Evidence from many lesion and fMRI studies converges in identifying the medial orbito-frontal cortex (MOFC) or sometimes more narrowly, the ventromedial prefrontal cortex (VMPFC) as a critical region in valuation when deciding between alternatives (Rangel et al., 2008; Henri-Bhargava et al., 2012; Fellows and Farah, 2007) or how much to pay for a good or item (Chib et al., 2009; Hare et al., 2008; Plassmann et al., 2007). This nding has been consistently reported in studies involving food items, trinkets and money (see Clithero and Rangel 1 (2013b) for a meta-analysis). Most studies however have focused on choices involving single items, as opposed to complex options made of several single items, or bundles. Among the few studies involving bundles, the VMPFC has been associated with the ability to make con- sistent choices between bundles (Camille et al., 2011) and the MOFC has been shown to re ect the dierence in subjective value between monetary options and bundled options (FitzGerald et al., 2009). In other forms of complex options, such as multi-attribute options, ac- tivity in the VMPFC re ected also the value of the combined items (Kahnt et al., 2011). It seems intuitive that complex options are dicult to evaluate. Hence, the ability to make more complex value comparisons is likely to also involve the working memory system, responsible for the short- term mental maintenance and manipulation of information. In the case of value-based decision-making, working memory has been re- ported to be associated with consistent choices involving complex bundles in older adults (Brocas et al., 2019a). Also, during tasks that tax executive function, activation is evoked in the dorsolateral prefrontal cortex (DLPFC) as demonstrated by many studies (Gold- 2 berg et al., 1998; Osherson et al., 1998; Goel et al., 1997; Baker et al., 1996; Berman et al., 1995; Nichelli et al., 1994; Petrides, 1994). Ac- tivation studies have shown that dorsal frontal regions are activated during tasks that are experienced as dicult (Braver et al., 1997; Co- hen et al., 1994, 1997; Monterosso et al., 2007; Luo et al., 2012), during task switching (Dove et al., 2000), and the DLPFC is dierentially re- cruited as tasks become more complex (Carlson et al., 1998; Braver et al., 1997; Cohen et al., 1997; Baker et al., 1996; Demb et al., 1995; Christo et al., 2001). This relationship extends to tasks requiring the explicit representation and manipulation of knowledge, where the ability to reason relationally is essential (Kroger et al., 2002). The role of DLPFC in value-based decision making has not been clearly established. It is sometimes reported to be activated and, whenever it is reported, its involvement is interpreted in the context of the question of interest. For instance, the DLPFC has been found to encode the variability of multi-attribute objects (Kahnt et al., 2011) and to be more activated when trade-os between attributes are re- quired (McFadden et al., 2015). In food choices, the DLPFC has been reported to modulate value (Camus et al., 2009; Hare et al., 2011; 3 Sokol-Hessner et al., 2012) and craving (Fregni et al., 2008), and to be involved in self regulation and self control (Hutcherson et al., 2012; Harris et al., 2013). The DLPFC has also been found to be function- ally connected with the value coding regions in self-control paradigms (Hare et al., 2009) and in multi-attribute paradigms (Rudorf and Hare, 2014). Last, the DLPFC was found signicantly more activated in studies in which options involved a con ict to be resolved (Baumgart- ner et al., 2011; de Wit et al., 2009). The DLPFC is however not usually reported to be active in studies involving single uni-attribute items. Taken together, these ndings indicate that the potential role of DLPFC is to support value calculation (perhaps in various ways) when those are complex. Here we report the results of an fMRI study in which participants were asked to choose between real food options involving single item options and bundled items options. Bundles varied in complexity and were either composed of the same two single items or of two dierent single items. Despite our primary interest in the role of the DLPFC and related structure in subjective value computation, our design al- lows us to address the following broader questions: (1) Is there a com- 4 mon value tracking region when options are simple and complex? (2) What are the neural mechanisms used to compute value as a function of complexity? (3) Does complexity involve brain networks implicated in attention and working memory? 1.2 Materials and methods Subjects Sixty eight healthy young adults (mean age 22 years old, 36 female and 32 male, all right-handed) were recruited from the Los Angeles Behavioral Economics Laboratory's subject pool at the University of Southern California. The Institutional Review Board of USC approved the study. Subjects could participate if they satised the standard el- igibility criteria for fMRI studies (no cognitive disorder or psychiatric condition, no medication aecting cognition, no history of seizure, no metal implants). We excluded subjects who reported to have food allergies, food restrictions or to be picky eaters. All participants re- ceived a $50 show-up fee for participating. They were also rewarded with one of their choices, selected randomly at the end of the session. Eight participants were excluded because of incomplete data collection 5 or excessive head movement during scanning. Procedure Participants were instructed to not eat for at least 4 hours before the experimental session. They were also instructed that they would have to stay after the session to consume what they had obtained and that they could not take any of the food items with them when they leave. This was implemented to make sure participants were hungry and thinking carefully about their choices during the session. The procedure was explained beforehand so that each participant knew that their reward would be based on their choices, and they should make their best decision in every trial. Each participant was asked to rank 30 single item options by order of preference. Each option was a small food serving. All were cali- brated to represent between 20 and 50 calories and to look visually similar. This ranking was used to create 40 bundles, 20 combinations of 2 same single items, and 20 combinations of 2 dierent single items. The participant was then asked to include those bundles in their pre- vious ranking. We then selected (see below) 11 single item options, 10 combinations of 2 same single items and 10 combinations of 2 dierent 6 single items to include in the experimental task, where each partici- pant made binary choices in the scanner. One of the 11 single item option was a reference option, denoted hereafter by REF. Choices were divided into three conditions (see Fig.1.1(A)): CONTROL, SCALING and BUNDLING. In each of the CONTROL trials, the participant had to choose between REF and one of the 10 remaining single items. In each of the SCALING trials, the participant had to choose between REF and one of the 10 combinations of 2 same single items. In each of the BUNDLING trials, the participant had to choose between REF and one of the 10 combinations of 2 dierent single items. SCAL- ING trials were included to control for quantity. In all cases, REF was o-screen, it was the same for each trial and it was shown to the participant at the beginning of the experiment. The other option was on-screen and it was displayed at the beginning of each trial. (see Fig.1.1(B)). Each individual trial was repeated 9 times for a total of 90 trials in each condition. The circles at the bottom of the screen told the participant what button selected which option, the solid cir- cle always representing REF. The button mappings were randomly assigned for each trial. 1 When the participant responded, the circle 1 Subjects had button boxes in each hand when they were in the scanner. They were instructed to 7 representing the chosen option was framed in a square to let the par- ticipant know that the their answer was recorded. The screen then advanced to a xation cross for the remainder of the trial. The fMRI task was optimized for detecting neural responses. We used Optsec2, a tool that automatically schedules events for rapid-presentation event- related fMRI experiments. Trials order and inter-stimulus intervals were optimized for task regressor estimation eciency (Dale, 1999) and organized into 5 runs. We chose the options in order to ensure that each of the three conditions CONTROL, SCALING and BUNDLING had symmetrical sets of low, medium and high on-screen value options centered around REF. To separate value specic activity from task specic activity, we also made sure that the distribution of value was similar across con- ditions. Specically, for each participant, we used an adapted genetic algorithm that produced three "populations" of options (for CON- TROL, SCALING and BUNDLING) satisfying criteria ensuring simi- larity across populations in terms of spread of values and distribution make choices by pressing a button in the hand corresponding to the option, as represented by the circle, they wanted. For example, if they wanted the reference option and the solid circle was on the right side of the screen in that trial, they could select it by pressing a button in their right hand. If they wanted the on-screen option instead, they could select it by pressing a button corresponding to the hollow circle, which in that case would be a button in their left hand. 8 around REF (see Fig.1.1(C)). CONTROL SCALING BUNDLING A SCANNER TASK Off-Screen REF B C Figure 1.1: Experimental Design. A. Each trial was a choice between the reference item (REF) and a food option in either of 3 conditions: CONTROL (one single item), SCALING (two same single items) or BUNDLING (two dierent single items). B. Only the latter food option was presented on screen. All trials were self-paced. C. We designed the task to best approximate a distribution of options centered around a REF item (orange) in each task CONTROL (red), SCALING (dark blue) and BUNDLING (light blue). MRI data acquisition Neuroimaging data was collected using the 3T Siemens MAGNETOM 9 Tim/Trio scanner at the Dana and David Dornsife Cognitive Neuro- science Imaging Center at USC with a 32-channel head-coil. Partic- ipants were laid supine on a scanner bed, viewing stimuli through a mirror mounted on head coil. Blood oxygen level-dependent (BOLD) response were measured by echo planar imaging (EPI) sequence with PACE (prospective acquisition correction) (TR = 2 s; TE = 25 ms; ip angle= 90; resolution = 3 mm isotropic; 64 x 64 matrix in FOV = 192 mm). A total of 41 axial slices, each 3 mm in thickness were acquired in an ascending interleaved fashion with no interslice gap to cover the whole brain. The slices were tilted on a subject by subject basis { typically 30 from the anterior commissure posterior commis- sure plane { to minimize signal dropout in the orbitofrontal cortex (Deichmann et al., 2003). Anatomical images were collected using a T1-weighted three-dimensional magnetization prepared rapid gradi- ent echo (MP-RAGE with TI = 900 ms; TR=1.95 s; TE: 2260 ms; ip angle=9; resolution = 1 mm isotropic; 256 256 matrix in FOV = 256-mm) primarily for localization and normalization of functional data. These scans were co-registered with the participant's mean EPI images. These images were averaged together to permit anatomical 10 localization of the functional activations at the group level. MRI data preprocessing Image analysis was performed using FSL algorithms organized in a nipype pipeline. Computation for the work described in this paper was supported by the University of Southern California's Center for High- Performance Computing (hpcc.usc.edu). The structural images were skull-striped then aligned and spatially normalized to the standard Montreal Neurological Institute (MNI) EPI template. The functional images were motion and time corrected. They were spatially smoothed using a Gaussian Kernel with a full width at half-maximum of 5mm. We also applied a high-pass temporal ltering using a lter width of 120s. Behavioral analysis Participants were asked to make choices between two snack options. The rst option was always the same o-screen reference option de- noted by REF while the second option was an on-screen variable op- tionVAR j (j =f1;:::;Ng). We constructed a Random Utility Model (McFadden et al., 1973; Train, 2009; Clithero and Rangel, 2013a) in 11 which we assumed that the utility derived by option VAR j depended on the value of the food snack and a stochastic unobserved error com- ponent j . Formally, u(REF ) = v 0 + 0 and u(VAR j ) = v j + j . In this model, the probability of choosing option VAR j is therefore P j = Pr[ 0 j < v j v 0 ]. Assuming that the error terms are in- dependent and identically distributed and follow an extreme value distribution with cumulate density function F ( k ) = exp(e k ) for allk = 0;j, the probability that the participant chooses optionVAR j is the logistic function P j = 1 1 +e v j v 0 We then constructed a likelihood function and we used Maximum Likelihood Estimation techniques to retrieve parameters v j given the observed choices. This procedure was implemented in Matlab with standard algorithms. For each individual, we also assigned an implicit ranking of all options based on these retrieved values. This ranking was used as a value-tracking regressor, referred to as VT. We also as- signed implicit rankings in the CONTROL condition (CV), SCALING condition (ScV) and BUNDLING condition (BV). Note that estimated 12 value and implicit ranking contain the same information. In principle, if a subject's choices are well represented by the Ran- dom Utility Model, we should observe that most choices are consistent with estimates. For each individual, we generated the choices they should have made in all trials if they were always choosing accord- ing to the value estimates and we compared with their actual choices. More precisely, we counted all choices that were not consistent with the value estimates (and henceforth with the implicit ranking) and we computed the percentage of these inconsistent choices. This exercise equipped us with a behavioral measure, a score we will refer to as Con- sistency Rate, that captured logical / consistent value representation. We computed these rates across all conditions and within condition. These measures were used to study individual dierences and neural correlates of behavior. Analysis of reaction times We recorded the onset of the stimulus and the time at which a choice was made in each trial. We looked at whether trials deemed to be more dicult, as measured by a smaller distance between the estimated value of the on-screen and o-screen options, were also taking longer. 13 We also looked for systematic dierences across conditions and across the type of choices (on screen vs. o-screen) ended up to be made. For each participant, we computed the mean Reaction Time it took them to deliberate in each of the three conditions. These measures were designed to analyze individual dierences across conditions. MRI data analysis We estimated several general linear models (GLMs) of BOLD re- sponses. Each aspect of the task was encoded in a regressor for the GLM. To identify what signal was associated with a particular condi- tion, we constructed indicator regressors that take value 1 whenever the participant is performing a trial within a condition and 0 other- wise. To identify the neural activity associated with the subjective value of the on-screen option, we created a parametric regressor that is equal to the value proxy (details below) of the on screen option and changes every time the on-screen option changes. When there is nothing on the screen, both regressors are 0. The models also include motion parameters (regressors for translation and rotation as well as artifact regressors controlling for quick jerking movements) and regres- sors for each run as nuisance regressors. All regressors were convolved 14 with the canonical form of the hemodynamic response. The values in the regressors were applied from the onset of the stimulus until a choice was made (average duration, 1.47s). All of our GLMs took the general form: BOLD i = [H 1 (R a )] a i +R b b i +e i WhereBOLD i is the time-series of BOLD signal at each voxeli,H 1 is the hemodynamic response function (HDF) used by FSL (Jenkinson et al., 2012) applied to the primary regressor matrix R a (each column is a primary regressor),R b are regressors of no interest ande i is a gaus- sian noise. The GLM solves for a i and b i to minimize the errore i . To analyze the in uence of an indicator regressor the coecients a i are contrasted against each other. These-contrasts are used to generate interpretable statistics. Every GLM was estimated in several steps. First, we estimated the model separately for each participant. After each GLM was t to the image time-series, the-contrasts were com- bined at the subject level using a Fixed Eects Model, then combined in a Mixed Eects Model to create group level voxel-wise t-statistics converted into z-statistics. All images were thresholded at z = 3:1 15 (p=0.001). The resulting image was rened further using cluster cor- rection and a signicance level of p 6 0:05 adjusted for family wise error. Clusters were reported if they passed that threshold unless otherwise noted. We used FSL Harvard-Oxford Subcortical and Cortical Structural Atlas and Talairach Daemon Labels (https://fsl.fmrib.ox.ac.uk/fsl/ fslwiki/Atlases) to list every gray matter region identied within each cluster. Functional regions were added where it was appropriate, using some of the notations from Dixon et al. (2017). For the regions for which we formed a priori hypothesis, namely the VMPFC, the MOFC and the DLPFC, and given the ambiguity around their description in the literature, we set an ex ante rule regarding how we would report our evidence (Figure 1.2). Identifying the value tracking regions across all conditions We used GLM 1 to identify the value tracking region across all con- ditions. This GLM consisted of 4 regressors of interest: C, S and B were indicator functions denoting CONTROL (C), SCALING (S) and BUNDLING (B) trials, while VT was the value regressor represent- ing the subjective value of the on-screen option across all conditions. 16 Y: +28 LEFT RIGHT X: +4 Z: -20 Harvard-Oxford Cortical Structural Atlas - ACC +10 +98 +10 +96 - FP +10 +87 - PaCG +10 +98 - SCA +10 +97 - FMC - FOC +10 +96 A B Y: +14 X: 0 Z: +64 +10 +79 - SFG +10 +83 - MFG +10 +99 - PCC +10 +86 - SMA +10 +84 - PrG +10 +81 - IFG VMPFC MOFC DLPFC Figure 1.2: Location of VMPFC, MOFC and DLPFC with regard to Harvard- Oxford Cortical Structural Atlas. A. Posterior part of VMPFC was dened by the anterior part of Subcallosal Cortex (at Y=+28), while the anterior part of VMPFC was dened by the posterior part of Frontal Pole (FP), and it overlayed with dorsal part of Frontal Medial Cortex (FMC) and ventral part of Paracingulate Gyrus/Anterior Cingulate Cortex. MOFC was dened by ventral part of FMC/FP and surrounded by medial parts of Frontal Orbital Cortex. B. DLPFC was dened according to Dixon et al., 2017 paper (BA = 9, 46 and 8), by the presence of Middle Frontal Gyrus (as well as junctions with Superior Frontal Gyrus and Inferior Frontal Gyrus), with the most posterior border right before Supplementary Motor Area (Y=+14). We labeled VT the set of regions that tracked the value regressor VT across all conditions. Given the decision-making literature on subjec- tive value has consistently reported certain regions to be signicantly associated with value tracking, we were interested in identifying which regions in VT overlapped with regions reported elsewhere. For that 17 purpose, we used the Meta Value (hereafter MV) from Clithero and Rangel (2013b) that comprises VMPFC and DLPFC among others. Tests for dierences in value tracking across conditions To study dierences in value tracking between conditions, we used GLM 2 consisting of the three parametric regressors representing the subjective value of the on-screen option in the CONTROL condition (CV), SCALING condition (ScV) and BUNDLING condition (BV) and the 3 indicator functions C, S and B capturing conditions. We labeled CV, ScV and BV the sets of regions that tracked the value re- gressor CV, ScV and BV. 2 We analyzed dierences between conditions within the value tracking region identied byGLM 1 that collapsed all conditions. Identifying a common core value tracking system To identify voxels that were tracking value in all conditions, we ran a conjunction analysis within GLM 2 and retained voxels that were tracking CV, ScV and BV at z> 1:645 (p6 0:05). Testing for dierences in responses to conditions To study dierences in direct responses to conditions, we used GLM 3 2 Technically, each of the value regressor is the interaction between VT and the condition regressor. 18 and we computed contrasts of the indicator regressors in SCALING vs. CONTROL (SC), BUNDLING vs. CONTROL (BC) and BUNDLING vs. SCALING (BS). We were particularly interested in identifying regions that were dierentially activated in each condi- tion compared to the two others. We therefore ran several conjunction analyses and retained voxels that responded signicantly more (less) in CONTROL than SCALING and BUNDLING, signicantly more (less) in SCALING than CONTROL and BUNDLING and signi- cantly more (less) in BUNDLING than CONTROL and SCALING. Testing for neural correlates of diculty/salience To identify regions that were tracking the value dierence between the on-screen option and the o-screen option, we used GLM 1 consisting of the three parametric regressors representing the demeaned absolute dierence between the estimated subjective values of the on-screen and o-screen options in the CONTROL condition (CD), SCALING condition (ScD) and BUNDLING condition (BD) and the 3 indicator functions C, S and B capturing conditions. Regressors CD, ScD and BD were interpreted as measure of diculty: items with a distant (respectively close) value from REF had a large (respectively small 19 positive or negative) score and were assumed to be easier (respectively more dicult) to compare with the o-screen option. Alternatively, the measure captured the degree to which the participant was indif- ferent between the on-screen and the o-screen options, or how salient the o-screen option was. These three interpretations were equivalent (see Fig.1.3). We will use both the dicult and salience interpreta- tions interchangeably. Figure 1.3: Diculty/Salience measure. The diculty/salience measure was con- structed from the estimated values. It measured the distance between the values of the on-screen and (xed) o-screen options. Region of interest (ROI) analysis Given earlier research is pointing to the signicant role of the MOFC and the VMPFC in value representation, we had a strong a priori 20 interest in those regions (Plassmann et al., 2007; Hare et al., 2009; Sokol-Hessner et al., 2012; Kable and Glimcher, 2007). The ROI for the VMPFC, hereafter VMPFC, was dened by 10 voxel sphere with the center at [0,46,-6] in MNI152 space. It encompasses VMPFC ac- tivity reported in Kahnt et al. (2011), Chib et al. (2009), McClure et al. (2004), O'Doherty et al. (2006), Kim et al. (2010), Lim et al. (2011) and Levy and Glimcher (2011). The MOFC ROI, hereafter MOFC, was dened by 7 voxel sphere with the center at [-8,44,-20] in MNI152 space. This corresponds to the area where the \value track- ing" activity was reported by Arana et al. (2003). We were also interested in analyzing clusters that were tracking value in our study and we dened ROI based on the value-tracking regions identied byGLM 1 andGLM 2 . Last, given one objective was to pinpoint regions that responded dierentially to condition (and independently of value), we also dened ROI based on the regions as- sociated to specic conditions in GLM 1 . All ROIs were performed on the second level cope images and we extracted each subject's contrast estimates averaged across all of the voxels in the mask. The ROI analysis was designed to address several questions. First, 21 we looked at similarities and dierences across parameter weights to investigate the specic roles of regions in value representation. Second, we studied the correlations between parameter weights to assess which ROIs were contributing to similar functions. Last, we looked for asso- ciations between parameter weights within each ROI and behavioral measures. Testing for functional connectivity. To test for dierences in condition-dependent functional connectivity, we used PPI, a general psychophysiological interaction (gPPI) model (McLaren et al., 2012; De Martino et al., 2013b; O'Reilly et al., 2012; Clewett et al., 2014b). We dened a region of interest (ROI) which we used as the seed for our analysis. The model created a new GLM in which the deconvolved activity of the seed region was assigned to the regressors modeling the condition and reconvolved with the hemody- namic response function. The gPPI model searched for how and when other regions connected to that seed region during a specic condi- tion, but not in any other condition. The seed region in PPI was our a priori ROI VMPFC. We preferred to use this independent seed rather than our peak activity value region to avoid inference problems 22 due to circular analysis (Kriegeskorte et al., 2009). 1.3 Results 1.3.1 Behavioral measures Distribution of options in the pairwise fMRI choice task We targeted an ideal distribution of options (see Fig. 1.1(C)) but bundled options were often more attractive to participants who were hungry. We ended up with a distribution of options favoring on-screen options in both SCALING and BUNDLING. More precisely, on aver- age, options were evenly distributed around REF in CONTROL, there were however 1.93 (respectively 1.73) more options to the left of REF in SCALING and BUNDLING. These preferences for bundled options were compensated by choosing REF slightly to the right of the median rank (see Fig. 1.4). Distributions in SCALING and BUNDLING were otherwise comparable and not excessively dierent from the distribu- tion of options in CONTROL. As a direct consequence, we observed that 50.7% of trials resulted in choices in favor of the on-screen option in CONTROL, against 60% in SCALING and 61% in BUNDLING. 3 3 We did not nd any relation between individual asymmetries across these distributions and behavioral measures or patterns of brain activity. 23 Figure 1.4: Realized distribution of on-screen options. Options were evenly dis- tributed around REF (orange) in CONTROL (red) but slightly biased towards on-screen options in SCALING (dark blue) and BUNDLING (light blue). Consistency rates We counted very few missed trials resulting in no choice (1.64% of the trials) indicating that participants were attentive and had enough time to select their preferred option. We also found that, on average, 90% of the choices of a subject were consistent with their implicit rankings (90% in the CONTROL condition, 91% in the SCALING condition and 89% in the BUNDLING condition), with signicant dierences only in the comparison between SCALING and BUNDLING (t = 2:76, p = 0:008). These results indicate that value estimates and implicit rankings were good proxies of the underlying subjective values (see Fig. 1.5(A)). We observed some variance within consistency rates (sd =0.07 in CONTROL, SCALING and BUNDLING) but these rates were corre- lated across conditions (Pearson's r =0.74, p <0.001 between CON- 24 TROL and SCALING, Pearson's r =0.69, p <0.001 between CON- TROL and BUNDLING and Pearson's r =0.80, p <0.001 between SCALING and BUNDLING). Therefore, participants were similarly consistent across all conditions. We also found that the proportions of trials that con icted with the best estimate of value (and implicit rank) was higher when im- plicit ranks were close (see Fig.1.5(B)). Note that there is no way of knowing whether a specic trial is \inconsistent" or not because we do not observe true subjective values. However, we can still infer from this result that there were more reversals or con icting choices when options were best tted as similar. This indicates that such options were more dicult to compare. Reaction times We found that it took on average longer to make decisions in BUNDLING (mean=1.62 s) compared to CONTROL (mean=1.50 s) and SCAL- ING (mean=1.42 s). A series of paired T tests and Wilcoxon signed rank tests conrmed that reaction times were signicantly longer in BUNDLING compared to CONTROL and SCALING and signicantly lower in SCALING compared to CONTROL and BUNDLING (in all 25 Figure 1.5: Consistency Rates. A. Consistency rates were high and similar across conditions. B. The proportion of choices con icting with the best value estimate was higher when choices were similar. cases p<0.001). This suggests that, compared to CONTROL, sub- jective valuation was a more complex task in BUNDLING but it was easier in SCALING (see Fig. 1.6(A)). Despite some variance within conditions (sd =0.32 in CONTROL, sd =0.32 in SCALING and sd =0.37 in BUNDLING), reaction times were correlated across conditions (Pearson's r =0.91, p <0.001 be- tween CONTROL and SCALING, Pearson's r =0.90, p <0.001 be- tween CONTROL and BUNDLING and Pearson's r =0.894, p<0.001 between SCALING and BUNDLING). Therefore, participants ranked similarly in terms of speed across conditions. 26 Reaction times were also longer in trials showing an on-screen op- tion ranked close to REF (see Fig. 1.6(B)). This indicates that trials involving items closer in value were more dicult and required more deliberation. The same was observed by splitting data by condition. Reaction times were also shorter when participants ended up choosing the on-screen option (see Fig. 1.6(C)). Figure 1.6: Reaction Times. A. It took more time to choose in BUNDLING and less time in SCALING. B. Mean reaction times as a function of option closeness (diculty). C. Mean reaction times as a function of nal choice (on-screen or o-screen). 1.3.2 Regions tracking subjective value Value tracking regions across conditions We identied candidates for regions associated with the computation of subjective value by estimating GLM 1 and by looking at signicant BOLD responses to the VT regressor. We found that activity in MV correlated signicantly with VT (see Fig. 1.7). We also found signif- 27 icant activity in regions not formerly associated with value-tracking (e.g. Fusiform Gyrus and Lateral Occipital Cortex) and usually re- ported in complex visual processing. A signal within these clusters was also found in the Left Cerebellum. We report in Table 1.1 the neural correlates of value during the evaluation period in all trials within MV and outside MV respectively. X: -20 Y: +23 Z: +46 X: -4 Y: +42 Z: -18 VALUE TRACKING META VALUE LEFT RIGHT left DLPFC (MFG, SFG, FP) VMPFC, MOFC, ACC (FMC, FOC, PaCG, SCA) +3.1 +6.95 +3.1 +5.04 Figure 1.7: Value tracking regions. Value tracking VT regions (in red) correlate sig- nicantly with MV (in green). Considerable overlap is present in Left VMPFC, Anterior Cingulate Cortex, MOFC and left DLPFC. 28 Region k z-score x y z (A) Meta Value (MV) regions Left Superior Frontal Gyrusz 341 4.44 -22 20 42 Left Subcallosal Cortexy 285 4.41 -6 28 -18 (B) Regions outside MV Right Supramarginal Gyrus 1381 4.9 48 -38 50 Left Occipital Fusiform Gyrus 943 4.95 -32 -66 -18 Right Temporal Occipital Fusiform Cortex 855 4.99 28 -48 -16 Left Lateral Occipital Cortex 307 3.89 -30 -78 40 Region is identied as peak activity. Clusters are reported at p < 0.05 y Overlap with VMPFC/MOFC;z Overlap with DLPFC Table 1.1: Neural correlates of value during the evaluation period across all conditions Dierences in value tracking correlates between conditions We used GLM 2 to identify dierences in value tracking across con- ditions. We focused primarily on these dierences within the regions VT that were identied by GLM 1 (these regions responded to the regressor VT) and we looked for signicant BOLD responses to re- gressors CV, ScV and BV. There were signicant dierences in value tracking across conditions (Table 1.2). Within MV, only regions within the VMPFC, the MOFC and the Anterior Cingulate Cortex (ACC) were signicantly activated in the CONTROL condition (Fig. 1.8(A)). Surprisingly, DLPFC was not signicantly active during any of the three conditions (Fig. 1.8(B)). We noticed a lack of signicant activity in regions usually associ- 29 ated with value tracking in the complex conditions (SCALING and BUNDLING). More generally, there was no signicant value tracking activity in the SCALING condition. Signicant value tracking activity in CON- TROL and BUNDLING was located outside MV. Also, regions iden- tied to respond to VT in visual areas (Fig. 1.8(C)) were dierentially activated in response to CV (left Lateral Occipital Cortex) and BV (Fusiform Gyrus). We will collectively refer to value tracking regions located in visual areas as Visual Value. Last, a cluster located in the Left Cerebellum was identied to respond to BV (Fig. 1.8(D)). Dierences between conditions within VMPFC and DLPFC We retained VT clusters that overlapped with MV as regions of in- terest and we studied parameter weights of value regressors VT, CV, ScV and BV. Namely, we retained the cluster with peak activity in the Left Subcallosal Cortex that overlapped with VMPFC, referred to as VMPFC-1, and the cluster with peak activity in the Left Superior Frontal Gyrus that overlapped with DLPFC, referred to as DLPFC-1 (see Table 1.1 for details). We found that parameter weights in VMPFC-1 were on average 30 A Meta Value Control Value Bundling Value +3.1 +5.04 +5.11 +3.1 +3.1 +5.46 Value Tracking X: -2 B Y: +19 L R C Z: -16 Meta Value Control Value Bundling Value +3.1 +5.04 +5.11 +3.1 +3.1 +5.46 Value Tracking Meta Value Control Value Bundling Value +3.1 +5.04 +5.11 +3.1 +3.1 +5.46 Value Tracking D X: -21 Meta Value Control Value Bundling Value +3.1 +5.04 +5.11 +3.1 +3.1 +5.46 Value Tracking L R Figure 1.8: Core value tracking activity A. The VMPFC is dierentially activated in response to value across conditions (orange) and in CONTROL (red); B. The left DLPFC is dierentially activated in response to value across conditions only; C. Regions involved in visual processing are tracking value across conditions and specically in CONTROL and BUNDLING; D. clusters in the Left cerebellum are tracking value across condition and in BUNDLING (blue). (MV regions (green) represented for reference) 31 Region k z-score x y z (A) CV: Regions responding signicantly to CV Right Temporal Occipital Fusiform Cortex ] 782 5.06 48 -56 -24 Left Lateral Occipital Cortex ] 773 5.00 -50 -74 -10 Left Subcallosal Cortexy 760 4.52 -6 30 -18 Left Lateral Occipital Cortex 341 4.32 -30 -76 40 Right Posterior Cingulate Cortex 302 4.02 6 -52 20 (B) ScV: Regions responding signicantly to ScV none (C) BV: Regions responding signicantly to BV Right Precentral Gyrus 712 4.29 36 -10 62 Left Temporal Occipital Fusiform Cortex ], [ 373 5.41 -28 -44 -22 Right Occipital Fusiform Gyrus ] 207 4.25 40 -64 -14 Region is identied as peak activity. Clusters are reported at p < 0.05. Overlap with MV. y Overlap with VMPFC/MOFC ] Overlap with Visual Value; [ Overlap with Left Cerebellum Table 1.2: Neural correlates of value during the evaluation period within conditions signicantly dierent from 0 in response to VT (t = 2:49, p=0.016) and to CV (t = 3:37, p-value =0.001). They were however not dierent from 0 in response to value in SCALING and BUNDLING, indicating that the value tracking signature identied in GLM 1 in the VMPFC was mostly due to activity in the CONTROL condition. We also found that parameter weights in VMPFC-1 were signicantly dierent in response to CV compared to BV (t=2.48, p=0.016). Parameter weights in DLPFC-1 were signicantly dierent from 0 in response to VT (t =2.30, p=0.03) and only marginally signicantly dierent from 32 0 in response to ScV (t = 1.68, p = 0.09). They were otherwise on average similar across conditions (Fig. 1.9). Meta Value VMPFC-1 X: -4 A B Y: +21 L R Meta Value DLPFC-1 Figure 1.9: Value tracking activity in VT clusters overlapping with MV. A. Distribution of mean parameter weights in VMPFC-1 ; B. Distribution of mean parameter weights in DLPFC-1. We next investigated the relationship between parameter weights in VMPFC-1 and DLPFC-1 across and within conditions. We found that they were overall signicantly correlated (Pearson's r =0.20, p- value =0.008). They were strongly correlated in CONTROL (Pear- son's r =0.47, p-value <0.001) and BUNDLING (Pearson's r =0.34, p-value<0.007) but no signicant association was found in SCALING 33 (Fig. 1.10(A)). This provided indication that the two regions were part of the same network in CONTROL and BUNDLING. Figure 1.10: Relationship between mean activity in VMPFC and DLPFC. Parameter weights are correlated in A. clusters that respond to value inGLM 1 and in B. clusters that respond to value in all conditions in GLM 2 . The analysis conrmed that, while regions in the VMPFC and DLPFC were signicantly associated with value tracking in the ex- perimental paradigm, (i) VMPFC was mostly activated in response to value in CONTROL and (ii) DLPFC was not dierentially activated across conditions. Common value tracking regions within the MV map across conditions The previous results show that conditions recruit dierent regions, or 34 involve specic regions dierentially. A natural question is whether these conditions also recruit a common set of regions. Said dierently, does there exist a common core of regions involved in value tracking? To address this question, we restricted our analysis to regions for which we had a strong a priori hypothesis that they would be involved in value tracking and we retained clusters that were responding to CV, ScV and BV. Specically, we focuses on the interaction map of CV, ScV and BV that overlapped with MV (Table 1.3). These clusters were located primarily in the left DLPFC, the left VMPFC and the ACC. Region k x y z Left Dorsolateral Prefrontal Cortex 287 -24 12 42 Left Ventromedial Prefrontal Cortex 49 -6 30 -16 Right Superior Frontal Gyrus 44 6 38 30 Right Frontal Pole 20 12 60 18 Right Anterior Cingulate Cortex 8 6 30 10 7 12 40 -2 2 6 2 26 Anterior Cingulate Cortex 3 4 -2 30 Left Anterior Cingulate Cortex 1 -6 40 4 Images thresholded at p< 0:05, uncorrected. Table 1.3: Neural correlates of value common to all conditions within MV. Clusters in the interaction map of CV, ScV and BV that overlap with MV. Within VMPFC and DLPFC, the clusters identied in the common core were located close to the clusters VMPFC-1 and DLPFC-1 that 35 were tracking value across conditions. We conducted an ROI analysis of theses clusters, hereafter VMPFC-2 and DLPFC-2, to investigate similarities and dierences between activity in the interaction map and activity in VMPFC-1 and DLPFC-1. Parameter weights in VMPFC-2 were on average signicantly dif- ferent from 0 in response to CV (t = 3:24, p-value =0.002) and ScV (t=2.28, p=0.03), but there was no statistically signicant dierence across conditions. We also found that parameter weights in VMPFC-2 were not signicantly correlated with parameter weights in VMPFC-1. Parameter weights in DLPFC-2 were on average similar across con- ditions in response to value regressors and they were signicantly dif- ferent from 0 (t=3.95, p<0.001 in CONTROL; t=2.56, p=0.013 in SCALING; t=2.27, p=0.027 in BUNDLING). We found no signicant dierence in DLPFC-2 across conditions. We also found that param- eter weights in DLPFC-2 were correlated with parameter weights in VMPFC-2 in all conditions (Pearson's r =0.60, p <0.001 in CON- TROL; Pearson's r = 0.33, p=0.01 in SCALING and Pearson's r = 0.38, p=0.003 in BUNDLING). Last, we found that parameter weights in DLPFC-2 were not signicantly correlated with parameter weights 36 in DLPFC-1. Meta Value VMPFC-2 X: -4 A B Y: +21 L R Meta Value DLPFC-2 Figure 1.11: Value tracking activity in core clusters overlapping with MV. A. Distribution of parameter weights in VMPFC-2 across conditions; B. Distribution of parameter weights in DLPFC-2 across conditions. Overall, the analysis conrmed that both VMPFC and DLPFC were part of the value tracking system irrespective of contextual rep- resentations of choices. Even though VMPFC was more involved in value representation in the task analogous to tasks reported in the lit- erature, some clusters within that region were equally recruited by our new tasks. The analysis also conrmed the existence of a relationship 37 between value tracking activity in VMPFC and DLPFC irrespective of task related contexts. Value-tracking activity in a priori regions of interest Independent analyses based on a priori ROIs are critical to conrm ef- fects. We performed an analysis of our two independent ROI, VMPFC and MOFC (Fig. 1.12). We found that parameter weights in the VMPFC in response CV were higher compared to ScV (t = 1:78, p = 0:081) and BV (t = 1:98,p = 0:052). We also found that parame- ter weights in the MOFC in response to value were strongly correlated with parameter weights in VMPFC (Pearson's r =0.62 in CONTROL, Pearson's r =0.64 in scaling, Pearson's r =0.65 in BUNDLING, all p<0.001) indicating that the two ROI were part of the same value- tracking network. Finally, parameter weights in VMPFC were highly correlated with parameter weights in VMPFC-2 (Pearson's r =0.92, p<0.001 in CONTROL; Pearson's r =0.86, p<0.001 in SCALING; Pearson's r =0.89, p<0.001 in BUNDLING) but not with parameter weights in VMPFC-1. Overall, the common core VMPFC region as- sociated with value tracking across all conditions was strongly related (and overlapping) with the VMPFC region already identied in the 38 literature. Meta Value ROI [0, 46, -6] X: 0 A B X: -8 Meta Value ROI [-8, 44, -20] Figure 1.12: Value tracking activity in a priori regions of interest. A. Distri- bution of parameter weights in VMPFC across conditions; B. Distribution of parameter weights in MOFC across conditions. Unexpected value tracking patterns Clusters at the junction of the Temporal Occipital Fusiform Cortex, the Temporal Fusiform Cortex, the Occipital Fusiform Gyrus and the Lateral Occipital Cortex were systematically identied as value track- ing regions on both sides of the brain and in both GLM 1 and GLM 2 . 39 More specically, activity in clusters within these areas were found to be neural correlates of VT, CV and BV. Similar clusters were also found within the interaction map describing voxels that were activated in response CV, ScV and BV. The Fusiform Gyrus and the lateral Oc- cipital cortex are implicated in complex visual processing. The recur- rence of ndings in our dierent conditions suggest that these regions were generally supporting the computation of subjective value and were more heavily taxed to track CV and BV. In addition, we found that voxels in the Left Cerebellum were associated to value tracking. We found in particular a signal common to all tasks in a specic clus- ter. This cluster overlapped left cerebellum clusters responding to both VT and BV (Fig. 1.8(C)). Table 1.4 summarizes the clusters of relevance within the core value region. We refer to the visual value cluster located on the left (re- spectively right) as L-Visual (respectively R-Visual) and the cluster located in the left Cerebellum as L-Cb. ROI analysis of these clus- ters showed that the means of mean parameters in visual regions were all signicantly dierent from 0 (t.tests, p > 0.02 in L-Visual and R-Visual) but not dierent from each other. The mean of mean pa- 40 rameters in L-Cb was dierent from 0 in BUNDLING only (t-test, p = 0:01245). Figure 1.13 summarizes the distribution of mean pa- rameters in those clusters. Region k x y z Right Temporal Occipital Fusiform Gyrus ] 82 40 -48 -24 Left Lateral Occipital Cortex ] 60 -26 -54 -24 Left Cerebellum [ 6 -22 -54 -26 Images thresholded at p< 0:05, uncorrected. ] Overlap with Visual Value; [ Overlap with Left Cerebellum Table 1.4: Neural correlates of value in visual regions and the Left Cerebellum. Clusters in the interaction map of CV, ScV and BV that overlap with the non MV regions of VT. MV L-Visual Z: -20 A B R-Visual C X: -20 L-Cb L R Z: -20 L R VT MV VT MV VT Figure 1.13: Value tracking activity in core visual value and left cerebellum. A. Distribution of parameter weights in L-Visual across conditions; B. Distribution of parameter weights in R-Visual across conditions; C. Distribution of parameter weights in L-Cb across conditions. Last, we investigated the correlations between mean parameters 41 across clusters of relevance in the core value tracking region. Mean activity in the common core regions was dierentially correlated across conditions (Fig. 1.14). The strongest correlations were observed between the DLPFC and VMPFC in CONTROL. They were how- ever strongest between the two visual regions in both SCALING and BUNDLING. Figure 1.14: Associations between mean parameters in core value regions: Core value tracking regions were dierentially correlated across conditions, except for VMPFC- 2 and DLPFC-2, the two visual regions L-Visual and R-Visual and L-Visual and Left Cb. (insignicant correlations left blank) Value tracking: summary Taken together, these results show dierences in activation patterns across conditions. They also show that a core set of regions com- prising clusters in (i) the Left VMPFC and the Left DLPFC, (ii) the visual cortex and (iii) the Left Cerebellum are tracking value. Clus- ters located in the Left VMPFC and Left DLPFC operate under all 42 conditions. Clusters in the visual cortex are required in all conditions but are more specically active in CONTROL and BUNDLING. Last, clusters in the left cerebellum are required in all conditions but are more specically active in BUNDLING. 1.3.3 Regions responding dierentially to conditions Dierences in responses to conditions We found several dierences among regions that responded to condi- tions. First, clusters within the DLPFC were recruited dierentially. Regions in the left DLPFC including clusters at the junction with the left VLPFC as well as regions in the right DLPFC including clusters at the junction with the right VLPFC were signicantly more active in CONTROL or BUNDLING compared to SCALING (Fig. 1.15(A)). Second, clusters located in the Left/Right cerebellum were more ac- tive in BUNDLING compared to each of the other two conditions (Fig. 1.15(B)). To identify the unique patterns associated with a given condition, we concentrated on regions that were systematically more or less ac- tive in that condition compared to the two others. No region was more active in SCALING, indicating that the task may have been easier to 43 complete. This was consistent with the fact that reaction times were signicantly shorter. Neural correlates revealed that CONTROL and BUNDLING were resulting in more complex activity patterns com- pared to SCALING. Notably, activity in the DLPFC, the VLPFC, the Middle Frontal Gyrus (MFG) and the Superior Frontal Gyrus (SFG) was larger in BUNDLING compared to the two other conditions while none of these regions were more active in BUNDLING. Conversely, activity in L/R cerebellum regions was larger in BUNDLING com- pared to the two others while none of these regions were more active in CONTROL. Due to these unique patterns, we focused on these regions exclusively. DLPFC and VLPFC activity Several clusters involving the DLPFC and the VLPFC were uniquely associated with conditions. In particular, clusters in the left DLPFC and the left VLPFC were more active in CONTROL compared to the two other conditions. We also found that regions in the left and right DLPFC, at the junction of the VLPFC were signicantly less active in SCALING (Table 1.5). Overall, regions in the left and right DLPFC were recruited dif- 44 A Y: +13 L R C > B B > S +3.1 +7.34 +7.02 +3.1 +3.1 +7.93 C > S +3.1 +7.34 B > C Y: +13 L R Y: +13 L R Y: +13 L R B C > B B > S +3.1 +7.34 +7.02 +3.1 +3.1 +7.93 C > S +3.1 +7.34 B > C X: -5 Figure 1.15: Dierences in responses to condition. A. Several clusters in the DLPFC and at the junction of the VLPFC responded dierentially to conditions; B. Several clusters in the Left/Right Cerebellum were dierentially active in BUNDLING compared to CONTROL and SCALING. (Heat-maps represent z-values) 45 Region k x y z CONTROL> SCALING & BUNDLING Left Dorsolateral Prefrontal Cortex 391 -24 16 36 Left Dorso- and Ventro-lateral Prefrontal Cortex 60 -54 2 28 SCALING<CONTROL & BUNDLING Right Dorso- and Ventro-lateral Prefrontal Cortex 554 44 16 14 Left Dorso- and Ventro-lateral Prefrontal Cortex 245 -36 8 18 Images thresholded at p< 0:05, uncorrected. Table 1.5: Clusters in DLPFC/VLPFC regions uniquely associated to a condition. ferentially. The results indicate that a distinction was made between single items and bundles, resulting in higher activity in regions of the left DLPFC in CONTROL. At the same time, a distinction was made between perceived simple (SCALING) vs. perceived complex (CONTROL and BUNDLING) computations resulting in lower ac- tivity in the right DLPFC in SCALING. To investigate these regions further, we conducted a ROI analysis of each cluster reported in Ta- ble 1.5, hereafter referred to as DLPFC-C, L-VLPFC-C, L-VLPFC-Sc, R-VLPFC-Sc. Mean parameters in DLPFC-C and L-VLPFC-C were larger in CONTROL compared to the two other conditions (t-tests, all p< 0.001 for DLPFC-C ; p< 0.02 for L-VLPFC-C ; Fig. 1.16 (A,B)) and mean parameters in L-VLPFC-Sc and R-VLPFC-Sc were larger in CONTROL compared to SCALING (t-test, p<0.001 for both ROI, 46 Fig. 1.16 (C,D)). C > Sc Y: +16 A B X: -54 X: -42 C D Y: +44 C > B C > Sc&B L R C > Sc C > B C > Sc&B Sc < C Sc < B Sc < C&B Sc < C Sc < B Sc < C&B DLPFC-C L-VLPFC-C L-VLPFC-Sc R-VLPFC-Sc Figure 1.16: Responses to condition in VLPFC/DLPFC regions. Distribution of parameter weights in A. DLPFC-C, B. L-VLPFC-C, C. L-VLPFC-Sc and D. R-VLPFC- Sc across conditions. 47 Unexpected responses to conditions We did not nd any unique pattern of activity in the DLPFC in BUNDLING. However, we found that a subset of voxels in the Left/Right Cerebellum were systematically more active in BUNDLING. Also, mean parameters were larger in BUNDLING compared to the other two conditions in that ROI, hereafter referred to as L/R Cb (t-tests; p<0.0025; Figure 1.17). B > C Y: +16 B > Sc B > C&Sc L/R Cb Figure 1.17: Responses to condition in the Cerebellum. Distribution of parameter weights in L/R Cb. 48 Network of regions We asked whether activity was related in the condition-tracking re- gions. We found that mean parameters in the 5 clusters were highly correlated (Fig. 1.18) in all conditions. Figure 1.18: Associations between mean parameters in regions representing conditions: Mean parameters in all regions involved in condition responses were strongly correlated. Condition tracking: summary Clusters located in the DLPFC/VLPFC were uniquely related to con- ditions: specic patterns in those regions seem to support discrimi- nation between conditions and provide contextual information about the kind of trial a participant faces. We also found that regions in the R/L Cerebellum were specically associated to BUNDLING. 49 1.3.4 Regions tracking diculty/salience UsingGLM 2 , we investigated the correlates of diculty/salience. We did not nd signicant involvement of prefrontal regions. Large clus- ters were found in areas involved in the integration sensory informa- tion (Superior Parietal Lobule, Postcentral Gyrus). We also noticed dierent patterns of activity across conditions. Even though visual re- gions were active in all three conditions (Temporal Occipital Fusiform Gyrus, Lateral Occipital Cortex), distinct activation patterns were ob- served in the Right Insula to track CD and in the Left Insula to track ScD. Results are reported in Table 1.6 and illustrated in Figure 1.19. We also found that regions posterior to the DLPFC, more precisely the Middle and Superior Frontal Gyri were asociated with diculty in both SCALING and BUNDLING. We investigated whether there was a common region tracking dif- culty/salience across conditions by looking at the interaction map between CD, ScD and BD and we identied clusters in the Visual, SFG/MFG and Insula areas. Because these regions were playing a role in tracking diculty in the task, we retained those clusters to perform ROI analysis. Table 1.7 summarizes the clusters of relevance 50 Region k z-score x y z (A) CD: Regions responding signicantly to CD Right Superior Parietal Lobule 4024 5.07 26 -50 52 Left Temporal Occipital Fusiform Gyrus 1603 4.65 -22 -52 -18 Right Postcentral Gyrus 375 4.17 52 -16 30 (B) ScD: Regions responding signicantly to ScD Left Precuneus cortex ? 4026 5.35 -10 -64 52 Right Lateral Occipital Cortex 298 4.4 46 -82 -8 Left Central Operculum 242 4.79 -50 2 2 Right Lateral Occipital Cortex 200 4.22 38 -84 14 (C) BD: Regions responding signicantly to BD Left Lateral Occipital Cortex ? 11070 5.58 -28 -88 18 Left Precentral Gyrus ? 357 4.17 -28 -8 44 Region is identied as peak activity. Clusters are reported at p < 0.05. Overlap with Insula; ? Overlap with MFG/SFG. Table 1.6: Neural correlates of diculty during the evaluation period within conditions. within the core diculty regions. We refer to the clusters in the visual cortex located on the right (respectively left) as R-Visual-Di (respec- tively L-Visual-Di ), the clusters located on the right (respectively left) SFG/MFG as R-SFG (respectively R-SFG), and the clusters lo- cated in the neighborhood of the Insula on the right (respectively left) as R Ins (respectively L Ins). Figure 1.20 summarizes the distribu- tion of mean parameters in those clusters. Mean parameters in the 6 clusters were highly correlated (Fig. 1.21) in all conditions but the relationships were stronger in SCALING. 51 A Scaling Difficult Bundling Difficult +5.40 +3.1 +3.1 +5.64 Y: -52 Control Difficult +5.12 +3.1 B Scaling Difficult Bundling Difficult +5.40 +3.1 +3.1 +5.64 Control Difficult +5.12 +3.1 L/R-SPL L R L R Z: -16 L/R-TOFC C Scaling Difficult Bundling Difficult +5.40 +3.1 +3.1 +5.64 Y: -14 Control Difficult +5.12 +3.1 R-PoG L R D Scaling Difficult Bundling Difficult +5.40 +3.1 +3.1 +5.64 Y: +2 Control Difficult +5.12 +3.1 L-CO L R Figure 1.19: Dierences in responses to diculty A. Considerable activation was found in the Superior Parietal Lobule in all conditions; B. Activation was found in both CONTROL and BUNDLING in the Temporal Occipital Fusiform Gyrus and the C. Post- central Gyrus; D. Clusters in the Central Operculum were activated in SCALING only. (Heat-maps represent z-values) 52 Region k x y z Right Temporal Occipital Fusiform Gyrus 595 26 -64 -26 Left Temporal Occipital Fusiform Gyrus 398 -30 -48 -30 Right Superior/Middle Frontal Gyrus 99 28 -10 48 Left Superior/Middle Frontal Gyrus 75 -22 -12 46 Right Postcentral Gyrus, Parietal Operculum 61 62 -22 22 Left Insula 46 -36 -2 8 Images thresholded at p< 0:05, uncorrected. Table 1.7: Neural correlates of diculty common to all conditions. Main clusters (k>40 voxels) in the interaction map of CD, ScD and BD in the Visual, SFG/MFG and Insular regions. Diculty: Summary The analysis suggests that a common core of clusters tracks di- culty/salience in visual regions, in regions of the SMG/MFG, posterior to the DLPFC, and in insular regions. They are however dierentially recruited across conditions: Right insular regions are more heavily taxed in CONTROL while Left insular regions are more heavily taxed in SCALING. Both SCALING and BUNDLING require the involve- ment of SMG/SFG regions. 1.3.5 Connectivity analysis Using PPI analysis, we identied regions that were functionally asso- ciated with regions in the VMPFC and we asked whether these associ- ations were modulated by conditions. Our independent vmPFC ROI, 53 Z: -16 L R R-Visual-Diff Z: -16 A B L-Visual-Diff C Y: -6 R-SFG L R MV VT MV VT MV VT L-SFG D E R-Ins F X: -38 L-Ins MV VT Y: -8 MV VT L R L R Y: -22 MV VT L R Figure 1.20: Diculty tracking activity. Distribution of parameter weights in A. R- Visual-Di, B. L-Visual-Di, C. R-SFG, D. L-SFG, E. R-Ins, F. L-Ins across conditions. which was overlapping signicantly with the core VMPFC-2 region in our study, was chosen as the seed of this exercise. We identied one cluster in the Right SFG that was signicantly connected to vmPFC in SCALING and three clusters, in the Right and Left Occipital Pole and the left Interior Frontal Gyrus, that were more strongly connected 54 Figure 1.21: Associations between mean parameters in regions tracking di- culty/salience. Mean parameters in regions tracking diculty were strongly correlated across regions, most signicantly in SCALING (insignicant correlations left blank). to vmPFC in SCALING compared to CONTROL (Table 1.8). Region k x y z SCALING connectivity Right Superior Frontal Gyrus 655 22 8 48 SCALING connectivity> CONTROL connectivity Left Inferior Frontal Gyrusz 486 -52 20 24 Right Occipital Pole ] 485 16 -100 -10 Left Occipital Pole ] 433 -30 -94 -18 Images thresholded at p< 0:05. z Overlap with DLPFC; ] Overlap with Visual Value. Table 1.8: PPI analysis of VMPFC (seed) connectivity. The cluster signicantly connected to the VMPFC in SCALING appeared to overlap with clusters that responded more strongly to CONTROL than SCALING and/or BUNDLING. This may indicate that voxels in that region were signaling condition dierences to the 55 VMPFC. A signicantly lower activity combined with a signicant coupling suggests that the mechanism allowing information transmis- sion from one region to the other was direct. The region showing a more signicant coupling with VMPFC in SCALING compared to CONTROL in the Left IFG was overlapping with clusters previously identied to play a role in condition response (namely L-VLPFC-C and L-VLPFC-Sc). ROI analysis of mean pa- rameters in these clusters showed that mean parameters of conditions regressors were correlated in all conditions (Pearson's R>0.31 and all p>0.14) suggesting that voxels in these regions were responding in a similar fashion. The two clusters in the Right and Left Occipital Pole were located in the visual system and they were overlapping with clusters involved in value tracking (more precisely R-Visual and L-Visual). More in- terestingly, mean parameters of value regressors in R-Visual and the cluster located in the Right Occipital Pole were strongly correlated in SCALING and BUNDLING (Pearson r =0.48, p< 0:001 in SCALING and Pearson r =0.68, p< 0:001 in BUNDLING). Similarly, mean pa- rameters of value regressors in L-Visual and the cluster located in the 56 Left Occipital Pole were correlated in all conditions but to a smaller extent (Pearson r =0.28, p= 0:03 in CONTROL; Pearson r =0.24, p= 0:07 in SCALING and Pearson r =0.22, p< 0:09 in BUNDLING). The absence of fundamental coupling with the VMPFC in CON- TROL and BUNDLING suggests that the VMPFC was more strongly connected to other regions only in the simpler and quicker SCALING condition. SCALING CONNECTIVITY X: 0 Z: -6 LEFT RIGHT R-SFG SCALING CONNECTIVITY > CONTROL CONNECTIVITY X: 0 Z: -6 LEFT RIGHT L-IFG L-OP R-OP Figure 1.22: VMPFC Connectivity: A. cluster in the Right SFG was signicantly connected to the VMPFC in CONTROL. B. regions in the Left and Right Occipital Pole and the Left Inferior Frontal Gyrus showed signicantly more coupling with the VMPFC in SCALING compared to BUNDLING. 1.3.6 Correlates of behavioral markers Value-tracking and behavior We related the behavioral measures with value tracking activity within specic ROI through regression analysis. We concentrated on activity in the clusters we identied as part of the common value system: 57 DLPFC-2, VMPFC-2, L-Visual, R-Visual, and L-Cb. The dierences in reaction times across conditions and across par- ticipants were expected to relate to dierences in patterns of activity across value-tracking regions. We found that lower reaction times were associated with higher activity in VMPFC-2, L-Visual and R-Visual. The remarkably similar results obtained for L-Visual and R-Visual were consistent with the fact that parameter weights in these clus- ters were strongly correlated (Pearson's r =0.50 in CONTROL, Pear- son's r =0.70 in SCALING and Pearson's r =0.66 in BUNDLING, all p< 0:001). There was no relationship between parameters weights in DLPFC-2 and reaction times (Table 1.9, rst and last two column). As noted earlier, consistency scores were quite high in all condi- tions, indicating that consistent value representation was not harder to achieve in any specic condition. There were still individual dier- ences to exploit. We found that higher parameter weights in VMPFC- 2 and in DLPFC-2 were associated with higher consistency scores. There was no signicant relationship between parameter weights in L-Visual or R-Visual and consistency scores (Table 1.9, second and third columns). 58 Even though parameter weights in L-Cb were strongly correlated with parameter weights in L-Visual in all conditions (Pearson's r =0.67 in CONTROL, Pearson's r =0.35 in SCALING and Pearson's r =0.53 in BUNDLING, all p< 0:006) and in R-Visual in CONTROL and BUNDLING (Pearson's r =0.35 in CONTROL, and Pearson's r =0.58 in BUNDLING, all p< 0:005), we did not nd any relationship between L-Cb and measures associated with L-Visual and R-Visual. VMPFC-2 VMPFC-2 DLPFC-2 L-Visual R-Visual Mean Reaction Time -34.93 - - -25.59 -24.13 (18.87) - - (14.24) (13.03) Consistency rate - 91.52 68.56 - - - (53.07) (32.45) - - SCALING -15.22 -13.27 -9.05 -6.24 -1.40 (11.65) (11.62) (6.07) (9.41) (10.12) BUNDLING -12.43 -15.79 -7.37 -11.52 10.42 (14.22) (13.18) (7.24) (10.23) (10.26) Constant 81.95 -52.91 -41.64 58.42 50.40 (30.36) (47.06) (28.49) (24.11) (23.64) Observations 60 60 60 60 60 Adj. R 2 0.042 0.022 0.028 0.031 0.012 p< 0:1, p< 0:05, p< 0:01, p< 0:001 clustered standard errors at the subject level Table 1.9: Relationship between ROI parameter weights and behavioral mark- ers in regions tracking value. Parameter weights in VMPFC-2 were negatively asso- ciated with mean reaction times and positively associated with consistency rates (rst and second columns); parameter weights in DLPFC-2 were positively associated with consistency rates (third column); parameter weights in both L-Visual and R-Visual were negatively associated with mean reaction times (last two columns) 59 Condition-tracking and behavior We performed a similar analysis by correlating behavioral measures with activity in regions that were identied to respond to condition, DLPFC-C, L-VLPFC-C, L-VLPFC-Sc, R-VLPFC-Sc, and L/R Cb. Table 1.10 shows that mean activity in the vast majority of regions were associated with smaller reaction times. L/R Cb was the only one associated with consistency. These results taken together suggested that active voxels in response to conditions were supporting ecient and consistent value representations. DLPFC-C L-VLPFC-C L-VLPFC-Sc R-VLPFC-Sc L/R Cb Mean Reaction Time -35.93 -95.102 -36.38 -59.60 - (12.71) (33.30) (18.20) (19.80) - Consistency rate - - - - 102.15 - - - - (55.89) SCALING -10.06 -10.18 -12.70 -12.18 -2.30 (1.95) (3.08) (2.47) (2.43) (2.94) BUNDLING -8.47 -7.91 0.66 -1.71 7.96 (1.75) (3.34) (2.30) (2.78) (2.54) Constant 108.86 253.52 118.56 174.31 -37.73 (21.62) (54.67) (29.50) (33.45) (50.84) Observations 60 60 60 60 60 Adj. R 2 0.126 0.163 0.062 0.105 0.033 p< 0:1, p< 0:05, p< 0:01, p< 0:001 clustered standard errors at the subject level Table 1.10: Relationship between ROI mean activity and behavioral markers in regions tracking conditions. Mean activity in DLPFC-C, L-VLPFC-C, L-VLPFC-Sc and R-VLPFC-Sc was negatively associated with mean reaction times (rst four columns); Mean activity in L/R Cb was positively associated with consistency rates (last column). 60 Diculty-tracking and behavior Regions involved in diculty/salience tracking were also associated with behavioral measures (Table 1.11). In particular, mean parame- ters in visual regions R-Visual-Di and L-Visual-Di and in R Ins were negatively associated with reaction times. Also, mean parameters in L-SFG were negatively associated with consistency rates: partici- pants exhibiting a higher response to salience were less consistent. R-Visual-Di L-Visual-Di L-SFG R-Ins Mean Reaction Time -1.00 -0.65 - -0.65 (0.33) (0.30) - (0.29) Consistency rate - - -2.81 - - - (1.50) - SCALING -0.38 -0.37 -0.18 -0.08 (0.23) (0.22) (0.22) (0.25) BUNDLING -0.14 0.005 -0.33 -0.04 (0.23) (0.25) (0.23) (0.27) Constant 2.45 1.83 3.25 1.56 (0.60) (0.55) (1.38) (0.58) Observations 60 60 60 60 Adj. R 2 0.054 0.022 0.010 0.084 p< 0:1, p< 0:05, p< 0:01, p< 0:001 clustered standard errors at the subject level Table 1.11: Relationship between ROI mean activity and behavioral markers in regions tracking diculty/salience. Mean activity in visual regions R-Visual-Di and L-Visual-Di was negatively associated with reaction times (rst two columns); Mean activity in L-SFG was negatively associated with consistency rates (third column); Mean activity in R Ins was negatively associated with reaction times (last column). 61 Left IFG R-Visual-PPI R-Visual PPI L-Visual PPI Mean Reaction Time -54.54 -38.37 - -55.20 (16.46) (15.36) - (20.85) Consistency score - - 89.13 - - - (48.90) - SCALING -7.36 -5.63 -11.44 -3.25 (3.04) (2.75) (9.96) (2.58) BUNDLING 7.68 9.91 -6.76 -1.94 (2.60) (2.96) (10.38) (2.26) Constant 193.42 151.32 -68.33 148.83 (28.31) (24.64) (45.80) (34.67) Observations 60 60 60 60 Adj. R 2 0.119 0.073 0.002 0.107 p< 0:1, p< 0:05, p< 0:01, p< 0:001 clustered standard errors at the subject level Table 1.12: Relationship between ROI mean activity and behavioral markers in regions connected to VMPFC. Mean activity in regions that were more connected to VMPFC in SCALING than in CONTROL was associated with behavioral markers. Higher mean parameters of condition regressors in Left IFG, R-Visual-PPI and L-Visual PPI were associated with lower reaction times; higher mean parameters of value regressors in R-Visual PPI were associated with higher Consistency rates. 62 A note on missed trials We also found that higher parameter weights in value tracking regions vmPFC-2, dlPFC-2, in condition tracking regions L-VLPFC-Sc, R- VLPFC-Sc and L/R Cb and in diculty/salience tracking regions L- Ins, R-Ins, L-SFG and L-Visual-Di were associated with lower rates of missed trials. The results are partly due to the fact that higher mean parameters in those regions were often associated with lower reaction times. 1.4 Discussion We hypothesized that value representation was modulated by con- text eects related to the number and type of options presented to a decision-maker. We expected that choices between single items would be simple while choices involving bundles would be complex. In line with the hypothesis, we expected to see dierent neural pat- terns within known value-tracking regions, such as the VMPFC and the MOFC. We also expected to see that complexity would require the involvement of cognitive processes, including working memory. Therefore, we hypothesized that clusters in the DLPFC would be dif- 63 ferentially activated in complex contexts. To test that hypothesis, we asked three questions: (1) is there a common value tracking region when options are simple and complex? (2) Which brain regions compute value as a function of complexity? (3) Does complexity involve brain networks implicated in attention and working memory? Our ndings suggest that there is a common value region comprising the VMPFC and the DLPFC, however dier- ent additional regions may be recruited to represent contextual ele- ments of choices and to help computation. We have identied clusters in the left and right DLPFC at the junction of the VLPFC to be in- volved in response to conditions. We have also observed interesting patterns of activation in clusters located in the Cerebellum associated with both value tracking (Left Cerebellum) and complex representa- tion (Right/Left Cerebellum). Last, we have found that value was tracked by regions involved in high-order visual processing. We made a series of observations that we discuss below. Involvement of VMPFC in value-tracking across contexts We found dierences in value-tracking patterns in the VMPFC, a region traditionally associated to subjective value representation. 64 More precisely, our study suggests that the computation of subjective value depends critically on the context in which choices are presented. We found strong support for the involvement of the VMPFC primar- ily in the condition closest to earlier experimental settings (Konovalov and Krajbich, 2019) but less for its involvement in the valuation of bundles, indicating that dierent kinds of calculations were imple- mented. In particular, there was no signicant activity during the evaluation period in SCALING and BUNDLING after cluster correc- tion. This result is reminiscent of recent studies that did not observe value-related activity in the VMPFC (de Berker et al., 2019; Hunt et al., 2013; Jocham et al., 2014). Still, using a more liberal threshold and restricting to voxels that were active in response to value in all tasks, we found that a cluster located in the VMPFC belonged to this map. Overall, VMPFC belonged to a set of regions that tracked value across conditions. The role of DLPFC in value-tracking We found that clusters in the left DLPFC were responding to value. While the eect was small and not always detected after appropriate whole-brain corrections, a cluster in the DLPFC was also present in 65 the interaction map representing voxels active in all tasks. The result indicates that DLPFC belongs to the common set of regions that track value, together with VMPFC. This is consistent with Domenech et al. (2017) that suggests that the primary function of the VMPFC is to aggregate information into a value signal while selection occurs in regions involving the DLPFC that integrate the VMPFC signal. In our case, we have demonstrated that activity in these two regions was correlated. Value tracking neural patterns and quality of choice Even though behavioral measures were very similar across individ- uals, which is to be expected in the context of value-based decisions in a population of undergraduate students (Brocas et al., 2019a), we found that activity patterns in both the VMPFC and the DLPFC was explaining individual behavioral heterogeneity. In particular, higher value-tracking activity in the VMPFC and the DLPFC was associated with higher consistency rates. Said dierently, VMPFC and DLPFC were working jointly to provide an accurate representation of prefer- ences in each trial, resulting in consistent choices across trials. More- over, higher value-tracking activity in the VMPFC was correlated with 66 smaller reaction times, suggesting that the more ecient or diligent participants were those for which the VMPFC responded more to value dierences. Value tracking and Cerebelum activity One of the most salient unexpected result of our study is the involve- ment of clusters located around the cerebellum in the computation of value in BUNDLING. Even though the connection between value- based decision-making and cerebellar functions remains unclear, the nding is reminiscent of other studies (in other domains) that revealed a role for the cerebellum in decision-making. Cerebellar circuitry has been associated to computations that support accurate performance in perceptual decision-making task (Deverett et al., 2018) and high-level functions (Rosenbloom et al., 2012; Cardoso et al., 2014). Lesion stud- ies have shown that patients with cerebellar damage perform worse than control groups in tasks such as the Stroop Test (Gottwald et al., 2004), the Wisconsin Card Sorting Task (Karatekin et al., 2000) and in instruments that assess cognitive exibility (Manes et al., 2009). The cerebellum has also been linked more directly to the maintenance of working memory (Deverett et al., 2019; Grimaldi and Manto, 2012). 67 These associations are likely promoted by reciprocal connections be- tween the cerebellum and the prefrontal cortex in humans and mon- keys (Rosenbloom et al., 2012). Moreover, it has also been shown that the right cerebellar hemisphere is associated with logical reason- ing while the left cerebellum mediates attentional and visuo-spatial skills (Baillieux et al., 2010). These latter skills might be more rele- vant in the BUNDLING context. Complexity in the brain We unexpectedly found that SCALING was perceived as a simple task. Even though the on-screen option was a bundle a priori more complex to evaluate than a single item, the task took signicantly less time and recruited fewer regions. We also found that the functional coupling between the VMPFC and regions identied to support value computation was signicantly stronger in SCALING. This suggests that information pathways across these regions were more connected in that condition, hence enabling a quicker and more ecient processing. Contextual representations We found that regions in the DLPFC and VLPFC were associated with the representation of conditions. Our evidence suggests a later- 68 alization of the task representation within DLPFC/VLPFC, the left DLPFC/VLPFC being associated with the representation of on-screen bundled items and the right DLPFC/VLPFC being recruited in con- junction to perceived complex tasks (CONTROL and BUNDLING). The VLPFC is typically associated with response inhibition and goal appropriate response selection (Aron et al., 2004). In our setting, activity in the VLPFC may help select responses to incorporate the unique features of the on-screen option. In particular, the Left VLPFC has also been found to play a central role in integrating the size of choice sets in the representation of subjective value (Fujiwara et al., 2018). Even though our choice set is xed, the size of the choice it- self varies across conditions. In this context, and consistent with our observation, the Left VLPFC may be signaling the kind of trial (one item or two items on screen) the participant is facing. Experimental contexts The dierences observed between our study and earlier ones may re ect dierences in experimental contexts. In cases where partici- pants are exposed to a single condition, the brain does not need to encode unique features of trials. In our case, disentangling the three 69 conditions may be part of the valuation process. For instance, activity in the Cerebellum observed during CONTROL may result from the existence of BUNDLING trials for which Cerebellum is required. Visual value The result that value-tracking recruited regions generally involved in visual processing (fusiform gyrus, lateral occipital cortex) in our task is consistent with the evidence that higher value targets recruit greater visual activation (Serences, 2008; Serences and Saproo, 2010). It has also been shown that activity in the fusiform gyrus corre- lated with the aesthetic of visual attributes and exhibited functional connectivity with VMPFC area involved in value computation (Lim et al., 2013). This is reminiscent of the observed coupling between the VMPFC and the lateral occipital cortex in our setting. Insula The involvement of the insula and related regions (such as Opercu- lum) in the processing of diculty in our task may be consistent with the known role of insula in attention and salience processing (Tregel- las et al., 2006; Eckert et al., 2009). One can argue that an easy trial consisting of a choice between a very valuable item and a signicantly 70 less valuable item may be sorted out by detecting and focusing on the salient valuable item. In our task, the Insula may be involved in identifying such unique salient features, which is consistent with re- cent research reporting the role of Insula in novelty detection (Uddin, 2015; Uddin et al., 2017). 71 1.5 Appendix Appendix 0: Abbreviations are reported in Table 1.13. Abbreviation Region Abbreviation Region ACC Anterior Cingulate Cortex OFG Occipital Fusiform Gyrus Amg Amygdala OL Occipital Lobe Ang Angular Gyrus OP Occipital Pole Cb Cerebellum OrbG Orbital Gyrus ccs Calcarine Sulcus pa Pallidum Cd Caudate PaCG Paracingulate Gyrus CO Central Operculum paHG Parahipppocampal gyrus Cun Cuneus PCC Posterior Cingulate Cortex DLPFC Dorso Lateral Prefrontal Cortex PCun Precuneous cortex DMPFC Dorso Medial Prefrontal Cortex PO Parietal Operculum FG Fusiform Gyrus PoG Postcentral Gyrus FMC Frontal Medial Cortex PPo Planum Polare FO Frontal Operculum PrG Precentral Gyrus FOC Frontal Orbital Cortex PTe Planum Temporale FP Frontal Pole Pu Putamen GRe Gyrus Rectus R Red Nucleus HschGy Heschl'sGyrus RLPFC Rostro Lateral Prefrontal Cortex Hi Hippocampus SCA Subcallosal Cortex IFG Inferior Frontal Gyrus SFG Superior Frontal Gyrus Ins Insula SMG Supramarginal gyrus IntCalC Intracalcarine Cortex SN Substantia Nigra IPL Inferior Parietal Lobule SPL Superior Parietal Lobule ITG Inferior Temporal Gyrus STG Superior Temporal Gyrus SMA Supplementary Motor Area SupCalcC Supracalcarine Cortex LingG Lingual Gyrus TFC Temporal Fusiform Cortex LOC Lateral Occipital Cortex th Thalamus LOFC Lateral Orbito-Frontal Cortex TOFC Temporal Occipital Fusiform Cortex MFG Middle Frontal Gyrus TP Temporal Pole MOFC Medial Orbito-Frontal Cortex VLPFC Ventro Lateral Prefrontal Cortex MOG Middle Occipital Gyrus VMPFC Ventro Medial Prefrontal Cortex MTG Middle Temporal Gyrus WM White Matter NA Nucleus Accumbens Table 1.13: Brain Regions and Abbreviations 72 Appendix 1. We report results related to Value tracking in GLM 1 . Table 1.14 describes the clusters reported brie y in the text as track- ing VT. The uncorrected results thresholded at z=3.1 are reported in Table 1.15. The tables are organized to separate regions of the Meta Value MV reported in Clithero and Rangel (2014) from regions specically identied in the present study. Regions k z-score x y z (A) Meta Value (MV) Regions L SFG, MFG, DLPFC, FP 341 4.44 -22 20 42 RL PaCG, ACC, FMC, VMPFC, FOC, MOFC, SCA 285 4.41 -6 28 -18 (B) Regions outside MV R SMG, PoG, PrG, PO, SPL, Ang, LOC 1381 4.9 48 -38 50 L OFG, TOFC, TFC, LOC, LingG, Cb, MTG, ITG 943 4.95 -32 -66 -18 R OFG, TOFC, TFC, LOC, LingG, Cb, ITG, MTG 855 4.99 28 -48 -16 L LOC 307 3.89 -30 -78 40 Clusters are reported at p< 0:05. Table 1.14: Neural correlates of value in GLM 1 73 Regions k z-score x y z Meta Value L DLPFC, MFG, SFG, FP 341 4.44 -22 20 42 RL VMPFC, L ACC, RL FMC, RL mOFC, RL PaCG, RL SCA 285 4.41 -6 28 -18 R ACC, PaCG 63 3.79 12 44 6 R dmPFC, FP, PaCG, SFG 48 3.65 10 42 42 R FOC, lOFC 39 3.86 28 28 -14 R FP , SFG 19 3.57 12 56 22 L ACC 17 4.17 -2 30 8 R ACC 15 3.54 8 28 12 L DMPFC, FP 8 3.64 -12 46 36 L vmPFC, FMC, FP, PaCG 6 3.3 -2 56 -2 L ACC, PaCG 5 3.39 -6 40 4 Outside Meta Value R POG, SMG, SPL, LOC, Ang, PO, PCun, PrG 1381 4.9 48 -38 50 L LingG, LOC, TFC, TOFC, OFG, Cb 950 4.95 -32 -66 -18 R ITG, MTG, LingG, LOC, TOFC, Cb, OFG 855 4.99 28 -48 -16 L LOC 307 3.89 -30 -78 40 R ITG, MTG 105 4.2 62 -18 -28 L LOC 78 3.78 -16 -72 54 L ITG 76 4.25 -56 -40 -14 R IFG, PrG, vVLPFC 62 4.5 48 4 22 L SMG, SPL 59 3.91 -36 -48 46 R FP 39 3.93 48 36 8 L FOC, LOFC 33 3.94 -24 20 -26 L ITG, MTG 33 4.31 -56 -12 -28 L SMG 28 3.55 -52 -40 54 L Brain Stem, Hi, th 27 4.28 -16 -24 -10 L paHG 26 3.73 -32 4 -18 L MFG, Amg, TP 20 3.39 -34 10 40 L POG, SMG 16 3.63 -56 -26 42 R FP, LOFC 13 3.91 46 50 -18 R MTG 13 3.49 64 -50 -4 R LOC 13 3.42 38 -78 4 L MFG, PrG 12 3.37 -50 2 40 R FP 10 3.7 6 66 16 R FOC, Ins 8 3.46 28 12 -14 R Ins , CO 7 3.36 38 -2 12 R PrG 7 3.71 36 -10 62 R IFG, PrG 6 3.37 60 14 24 L FOC, MOFC 5 3.38 -10 6 -18 R POG, PrG 5 3.36 36 -24 50 z < 3:1, uncorrected, minimum extent k= 4 voxels. Table 1.15: Neural correlates of value in GLM 1 (z = 3:1, uncorrected results). 74 Appendix 2. We report results related to Value tracking in GLM-2. Tables 1.16 and 1.17 describe the clusters involved in value-tracking inGLM 2 in CONTROL and BUNDLING. No signicant cluster were associated with value tracking in SCALING. Tables 1.18, 1.19 and 1.20 report the uncorrected results. Figure 1.23 summarizes these results. Regions k z-score x y z R TOFC, TFC, ITG, paHG, LOC, OFG, Cb, LingG 782 5.06 48 -56 -24 L LOC, OFG, TOFC, TFC, paHG, Cb, ITG, MTG 773 5.00 -50 -74 -10 RL SCA, FMC, MOFC, PaCG, ACC, VMPFC, FP 760 4.52 -6 30 -18 L LOC, OP 341 4.32 -30 -76 40 RL PCun, PCC, IntCalC; R SupCalcC, R Cun 302 4.02 6 -52 20 Clusters are reported at p< 0:05. overlaps with MV, bold overlaps with VT at z=3.1. Table 1.16: Neural correlates of CV in GLM 2 . Regions k z-score x y z R PrG, PoG, SMG, SPL, Ang, MFG, SFG 712 4.29 36 -10 62 L TOFC, TFC, Cb, LingG, OFG, LOC 373 5.41 -28 -44 -22 R OFG, LOC, TOFC, LingG, ITG, Cb 207 4.25 40 -64 -14 Clusters are reported at p< 0:05. overlaps with MV, bold overlaps with VT at z=3.1. Table 1.17: Neural correlates of BV in GLM 2 . 75 Regions k z-score x y z L TFC, TOFC, LOC, MTG, OFG 836 5 -50 -74 -10 R TFC, TOFC, LOC, MTG, OFG 787 5.06 48 -56 -24 L/R FP, PaCG, FMC, SCA, MOFC, VMPFC 765 4.52 -6 30 -18 L LOC 341 4.32 -30 -76 40 L/R PCC, L IntCalC, L SupCalcC, PCun 302 4.02 6 -52 20 L FOC, FP, MOFC, LOFC 177 4.67 -22 22 -18 L LOC 153 4.13 32 -70 36 L SFG, MFG, DLPFC 107 4.16 -22 18 42 R SFG, MFG, FP, DMPFC 104 3.92 18 36 44 L PCC, PCun, LingG 91 4.44 -4 -50 4 L Brain-Stem, Hi, paHG 87 4.84 -16 -24 -14 R SMG, SPL 83 4.79 48 -40 48 L SFG, MFG, FP, DMPFC 52 4.47 -10 40 42 R FOC, mOFC, lOFC 39 4.4 28 30 -14 L PCC 37 3.69 -8 -42 -42 R FP, rlPFC 34 3.49 14 64 2 L MTG, STG 32 4.17 -60 -4 -14 L Amg, paHG 30 3.88 -26 0 -20 R PCun, SupCalcC 27 3.6 4 -70 28 R FP, IFG, VLPFC 21 3.74 48 38 10 R FP 18 3.46 -14 62 12 L FOC, MOFC 17 3.71 -12 6 -20 L ITG, MTG 17 3.55 -58 -40 -16 R SPL, Ang 16 3.6 38 -52 50 R Cb 15 3.4 24 -52 -20 L ITG, MTG 15 3.5 -56 -50 -14 R LOC, PCun 13 3.51 18 -68 40 L LOC 13 3.73 -18 -70 52 R FOC, LOFC, TP 11 3.47 28 18 -28 L/R PCC, PCun 10 3.29 -4 -48 20 R PaCG, SFG 9 3.63 8 42 28 R MTG, ITG 9 3.72 58 -20 -24 L ACC, PaCG 8 3.27 -4 42 6 L FP 8 3.53 -40 44 -10 L PoG, SMG 7 3.27 -52 -26 44 L FP, RLPFC, FMC,PaCG 7 3.47 -8 56 2 R TFC 6 3.46 34 -30 -30 R Ins, TP 6 3.35 36 10 -16 R LOC 6 3.31 32 -60 42 L TFG, ITG 6 3.34 -40 -16 -26 L LOFC, FP 5 3.4 -46 54 -14 R PrG, IFG, VLPFC 5 3.48 46 4 20 L paHG 5 3.4 -24 -20 -26 z< 3:1, uncorrected, minimum extent k= 4 voxels. overlaps with MV, bold overlaps with VT. Table 1.18: Neural correlates of CV in GLM 2 (z = 3:1, uncorrected). 76 Regions k z-score x y z R SMG, PoG 51 3.71 62 -24 44 L ACC, PaCG 39 4.1 14 42 12 L SMG, SPL 25 4.57 -50 -42 56 R NA, Cd, SCA 19 3.61 6 12 -4 L ITG, MTG 18 3.68 -56 -14 -28 L th, Midbrain 16 3.5 -2 -16 -4 L SFG, MFG, DLPFC 15 3.51 -22 16 52 L ACC 12 4.16 -2 32 8 R FP, RLPFC 12 3.6 12 64 16 R LOC, MTG 7 3.46 62 -66 20 R SMG 6 3.43 54 -30 54 L FP, FMC, VMPFC 6 3.46 -4 56 -18 L SMG 6 3.61 -56 -26 50 R FP, SFG 6 3.32 14 54 20 z< 3:1, uncorrected, minimum extent k= 4 voxels. overlaps with MV, bold overlaps with VT Table 1.19: Neural correlates of ScV in GLM 2 (z = 3:1, uncorrected). Regions k z-score x y z R PrG, PoG,SPL,SMG 712 4.29 36 -10 62 L TOFC, TFC,OFG, LingG, LOC 373 5.41 -28 -44 -22 R ITG, MTG, LOC, TOFC, LingG 207 4.25 40 -64 -14 L/R LingG, IntCalC, SupCalcC 125 4.19 -4 -72 2 L LingG, paHG 104 4.07 -12 -48 -10 R Ins 54 4.18 38 -6 14 L MFG, PrG, RLPFC 33 3.53 -50 6 42 L Brain-Stem, Midbrain, Hi, th 27 4.02 -16 -26 -10 R Cb 24 3.61 10 -70 -26 L SPL, SMG 23 3.53 -30 -50 44 L LOC 15 3.63 -50 -76 2 R LOC, PCun 14 3.59 18 -68 52 R SPL, PoG 11 3.93 24 -50 70 R CO, R Po 11 3.65 46 -16 16 L PoG, PrG 10 3.69 -54 -8 42 L MFG, PrG, DLPFC 9 3.49 -34 6 38 L OFG, LingG 8 3.32 -18 -74 -12 R SMG 8 3.48 56 -36 42 L ITG, MTG 8 3.47 -60 -42 -14 R FOC, FP, LOFC 8 3.24 30 32 -8 R FP, IFG, VLPFC 8 3.56 54 36 10 L SFG, MFG, DLPFC 7 3.5 -22 12 60 R Hi, 7 3.49 16 -26 -12 R Brain-Stem, th, Midbrain 6 3.34 12 -24 -6 L LOC 6 3.23 -36 -86 10 L LOC 6 3.36 -26 -78 20 R PoG, SMG, SPL 6 3.4 36 -36 42 L SMG, Ang 5 3.31 -50 -46 48 R LOC 5 3.3 38 -82 12 z < 3:1, uncorrected, minimum extent k= 4 voxels. overlaps with MV, bold overlaps with VT Table 1.20: Neural correlates of BV in GLM 2 (z = 3:1, uncorrected). 77 LEFT RIGHT BUNDLING CONTROL +3.1 +5.11 +3.1 +5.46 OVERLAP Figure 1.23: Value accros conditions CV, ScV and BV. Appendix 3. We report value tracking dierences across conditions in GLM 2 (value contrasts). Table 1.21 presents value contrasts infor- mation forGLM 2 . Table 1.22 reports the uncorrected value contrasts information. Figures 1.24 provides a visual representation of these contrasts. Regions k z-score x y z CV > BV LR FP, FMC, MOFC, PaCG 108 3.78 2 58 -16 Clusters are reported at p< 0:05. overlaps with MV, bold overlaps with VT Table 1.21: Dierences in value tracking response across conditions. 78 Regions k z-score x y z CV< ScV R SMG, Ang 16 4.22 66 -40 44 R SMG, Ang 9 3.54 66 -46 34 CV < BV R Cb 10 3.67 8 -68 -24 R Pu 5 3.52 30 -4 18 ScV < BV R LingG, Cb, TOFC 38 3.48 14 -62 -14 R Cun, L/R IntCalC, L/R SupCalcC, L/R LingG 28 3.38 6 -78 14 R LingG, R IntCalC 25 3.72 16 -62 4 R Hi, paHG, th 19 3.52 24 -34 -6 R CO, PO 18 3.74 44 -16 18 L/R Cb 16 3.46 0 -82 -24 R Hi, Brain stem, paHG 16 3.31 20 -26 -16 R Pu, Ins 15 3.52 28 -10 16 L Cb 14 3.77 -20 -60 -20 R LingG 12 3.9 8 -48 -6 R STG, MTG 11 3.53 56 -4 -12 L LingG, paHG 9 3.5 -14 -42 -12 R PoG 8 3.65 30 -30 40 R IntCalC, SupCalcC, LingG 7 3.28 8 -82 4 L TP 5 3.24 -34 8 -28 R th 5 3.42 10 -14 22 R Ins, CO 5 3.24 40 -8 16 L IntCalC 5 3.38 -24 -70 4 R SPL, Pcun, PoG 5 3.22 10 -50 70 CV > ScV L LOC, OFG, ITG 44 3.95 -52 -70 -14 L Brain-Stem, Hi, paHG 43 3.96 -16 -24 -14 R paHG, LingG 31 3.81 12 -36 -12 L STG, MTG 29 4.36 -64 -32 6 L PCC, LingG, PCun 29 4.1 -14 -52 0 L Hi, paHG 25 3.59 26 -30 -14 L Hi, paHG 24 3.69 20 62 0 L STG, MTG 23 3.86 58 -36 4 R PoG, SMG 23 3.66 32 -28 38 R Cb, LinG 22 3.64 18 -62 -16 R Hi 20 3.6 26 -36 -4 L LOC 16 3.77 -56 -74 -2 R PCun, SupCalcC 14 3.43 14 -56 18 L STG, TP 10 3.54 -38 12 -28 R PCun, LOC 10 3.37 10 -82 50 L FP, FOC, LOFC, MOFC 8 3.57 -20 36 -18 R Cun, PCun 8 3.29 6 -70 24 R TOFC, ITG 6 3.44 46 -54 -22 L PCun 6 3.31 -14 -52 24 R PrG 6 3.56 42 -8 32 L LOFC, FP, FOC 5 3.34 -30 32 -12 L Pte, STG 5 3.46 -66 -20 10 CV > BV L/R FP, VMPFC, MOFC, FOC 109 3.78 2 58 -16 L TP, STG, MTG 58 3.97 -42 12 -24 R PCun, PCC 24 3.82 14 -44 34 R FP, LOFC, FOC 16 3.75 -22 38 -18 L PoG 13 3.32 -44 -26 52 L MTG, STG 12 3.44 -60 -4 -14 L PCun, SupCalcC 12 3.44 -18 -58 20 R paHG, TFC, TOFC 11 3.39 26 -36 -20 R FP 10 3.54 20 64 2 L MTG 9 3.73 -68 -10 -24 L PoG 9 3.73 -48 -20 58 R VMPFC, ACC, PaCG 6 3.28 6 36 -14 L MTG 6 3.52 -48 -18 -10 L PrG 5 3.39 -34 -18 58 R SFG, MFG 5 3.37 22 26 36 R PCun, SupCalcC 5 3.42 14 -56 18 ScV > BV L FP, VMPFC, MOFC, FMC 36 4.05 -4 56 -18 L/R PaCG, SFG, FP 14 3.45 -2 54 18 R FP, VMPFC,MOFC 8 3.53 8 60 -16 R LOC 6 3.6 60 -68 20 z < 3:1, uncorrected, minimum extent k= 4 voxels. overlaps with MV, bold overlaps with VT Table 1.22: Dierences in value tracking response across conditions (z = 3:1, uncor- rected). 79 LEFT RIGHT CONTROL > BUNDLING +3.1 +3.82 Figure 1.24: CV bigger. Appendix 4. We report here common value tracking regions across all conditions. Table 1.23 details the clusters that track value across all three conditions. These are represented in Figure 1.25. The results reported here are uncorrected. 80 Regions k x y z R LOC, SPL, Ang, SMG, PoG 415 24 -64 32 * L MFG, SFG, FP, DLFPC 287 -24 12 42 L LOC 108 -26 -78 16 R LOC 82 32 -70 24 R TOFC, TFC, Cb 82 40 -48 -24 R PoG, IFG 81 60 8 18 L TOFC, TFC, Cb 60 -26 -54 -24 L ITG, MTG 55 -62 -46 -18 * L FMC, SCA, VMFPC, ACC, paCG 49 -6 30 -16 * R PaCG, SFG, FP 44 6 38 30 R OFG, LOC, TOFC 37 36 -72 -18 R PoG, SMG 35 58 -18 30 R ITG, MTG 30 64 -22 -28 L LOC, OFG 27 -40 -72 -10 R LOC, OP, ITG 21 32 -84 6 *R FP 20 12 60 18 L SPL, SMG, Ang 18 -34 -52 48 L OP, LOC 15 -32 -90 12 L OFG, LingG, TOFC 15 -26 -70 -14 R ITG, MTG 12 58 -42 -18 L FOC, LOFC 9 -20 16 -26 R FP, IFG, RLPFC 9 50 36 6 *R ACC 8 6 30 10 R PoG 8 40 -34 52 R ITG, MTG 7 54 -54 -8 *R ACC, PaCG 7 12 40 2 R MFG 7 38 16 46 L OFG, LOC 7 -30 -78 -14 L SMG 6 -50 -38 48 L Cb 6 -22 -54 -26 L MFG, DLPFC 5 -34 10 38 R FP 5 6 66 14 R ITG, MTG 4 46 -56 -6 R FP, RLPFC 4 54 40 10 L SPL, SMG 3 -34 -46 42 N/A outside LOC 3 -56 -74 -16 * ACC 3 4 -2 30 R ins, FOC 2 28 12 -14 R FP, LOFC 2 48 54 -16 L ITG, MTG 2 -54 -52 -16 R ITG 2 62 -46 -18 R FOC, LOFC 2 22 22 -20 R MOFC 2 22 22 -22 *R ACC 2 6 2 26 R ITG, MTG 2 -56 -12 -30 R PaCG, SFG 1 10 46 28 R PrG, IFG 1 58 14 28 R LOC, OP 1 32 -88 22 R IFG 1 58 22 14 R LOC 1 38 -84 4 R MTG 1 62 -48 0 L SFG 1 -4 44 40 L Midbrain 1 -12 -14 -6 R ITG, LOC 1 42 -58 -6 L SMG, SPL, Ang 1 -36 -50 44 L LOC 1 -32 -88 -8 R FP, LOFC 1 38 50 -18 R ITG 1 62 -52 -18 L SMG, SPL, PoG 1 -46 -42 54 L SFG 1 -20 6 54 L SMG, SPL 1 -44 -46 56 *L ACC 1 -6 40 4 z < 3:1, uncorrected, minimum extent k= 1 voxel. overlaps with MV Table 1.23: Regions tracking value in all conditions. Interaction mask of CV, ScV and BV. 81 LEFT RIGHT CONJUNCTION MASK VALUE TRACKING META VALUE +3.1 +6.95 +3.1 +5.04 Figure 1.25: Overlap between CV, ScV and BV (z = 1.645, uncorrected). Appendix 5. This appendix addresses responses to conditions in GLM-1. Table 1.24 presents value contrasts information for condi- tions inGLM 1 . Figures 1.26, 1.27, 1.28, 1.29, 1.30 and 1.31 represent these contrasts. Table 1.25 present the uncorrected results. Table 1.26 details the clusters that respond to a condition more than to the two others. Table 1.27 details the clusters that respond to a condition less than to the other two combined. (Figure 1.32, Figure 1.33) 82 Regions k z-score x y z CONTROL > SCALING RL PCC, PCun, Cun, IntCalC, SupCalcC, LingG, LOC, OP, Ang, SMG, MTG, ITG, STG, SPL, L Cb 13591 6.61 42 -86 24 R MFG, SFG, DLPFC, PrG, PoG, IFG, VLPFC 2173 5.69 30 6 58 L MFG, SFG, DLPFC, PrG, PoG, SMG, SPL, IFG, VLPFC 1937 5.08 -28 2 56 R PoG, SMG, PTe, PO, PrG 853 4.67 54 -18 46 RL FMC, MOFC, VMPFC, SCA, ACC, PaCG, NA, FOC, L LOFC, R FP 835 5.19 6 32 -18 CONTROL > BUNDLING L PoG, SMG, MTG, STG, ITG, Ang, LOC, OP, SPL, PO, CO, PTe, HschGy, PrG, SFG, MFG, DLPFC 6733 6.6 -50 -30 46 R LOC, Ang, OP, MTG, ITG, STG, SMG, PO, PTe, PoG 4857 6.95 44 -78 36 RL PCC, PCun, SupCalcC, IntCalC, Cun 1108 4.85 8 -40 34 R SFG, MFG, PrG 460 4.87 24 6 48 RL SMA, L ACC, L PaCG 384 4.49 -2 -2 48 L PrG, MTG, STG, CO, FO, TP, PPo, HschGy, Ins, MTG, ITG, STG 378 4.7 -58 -6 -16 L PrG, PoG, IFG, MFG 316 4.95 -56 6 34 R MTG, ITG, STG, TP 276 4.39 68 -16 -22 SCALING > CONTROL RL LingG, IntCalC, SupCalC 309 4.94 -4 -80 -10 L OP, Cun 291 6 -6 -102 16 SCALING > BUNDLING L SMG, PoG, SPL, PO, CO, PTe 258 4.17 -52 -32 40 BUNDLING > CONTROL RL OP, SupCalcC, IntCalC, Cun, LingG, Cb, L LOC, L OFG, L TOFC, L TFC, L ITG 4118 7.27 -4 -102 12 R OFG, LingG, LOC, OP, TOFC, ITG, Cb 1690 6.49 26 -74 -14 BUNDLING > SCALING RL TOFC, TFC, LingG, PrG, Cb, PCun, Cun, SupCalcC, PCC, LOC, SPL, Ang, SMG, OP, OFG, L ITG, L paHG 13703 7.86 -30 -56 -18 R PrG, PoG, IFG, VLPFC, MFG, DLPFC, FP, FOC, Ins, FO, CO, SPL, SMG, PCun, PCC 3278 5.74 46 2 30 L IFG, VLPFC, MFG, DLPFC, PrG, FO 633 5.09 -36 22 16 RL Cb 442 5.71 -6 -78 -22 Clusters are reported at p< 0:05. Table 1.24: Dierences in condition responses. 83 Regions k z-score x y z CONTROL > SCALING L/R LOC, PCun, IntCalC, SupCalcCCun, PCC, SPL, SMG, Ang, MTG, STG 14013 6.61 42 -86 24 R MFG, DLPFC, VLPFC, PrG, IFG, FOC, IFG 2174 5.69 30 6 58 R PoG, PrG, SMG, PO, SFG, MFG, DLPFC 1937 5.08 -28 2 56 R PoG, PrG, PTe, SMG, PO 867 4.67 54 -18 46 L/R ACC, SCA, NA, FOC, FMC, VMPFC, MOFC, PaCG 838 5.19 6 32 -18 R OP, OFG, LOC 303 4.8 30 -94 -10 L ITG, MTG, STG 289 4.98 -62 -20 -28 L/R ACC, PaCG, SMA 246 4.51 8 22 32 L LingG, TOFC, TFC, paHG, Cb 186 4.42 -26 -48 -8 L OP, OFG, LOC 144 4.53 -28 -94 -10 R PCC, PrG, PCun 95 4.11 8 -28 44 R TFC, paHG, TOFC, Hi, LingG 91 4.78 30 -38 -14 R MTG 81 4.15 58 2 -18 L pa, th, Cd 72 4.28 -12 -2 -2 R Ins, FOC 51 3.86 32 22 -12 R Cb 47 3.99 6 -60 -18 L Hi, paHG 46 3.8 -20 -18 -26 R Ins, PPo 41 3.81 38 -6 -12 CONTROL > BUNDLING L CO, PO, HschGy, LOC, Ang, ITG, MTG, STG, SMG, PTe, PoG 6815 6.6 -50 -30 46 R LOC, Ang, ITG, MTG, STG, SMG, PTe, PoG 5466 6.95 44 -78 36 L/R PCC, PCun, Cun, IntCalC, SupCalcC 1108 4.85 8 -40 34 R MFG, DLPFC, SFG, PrG 460 4.87 24 6 48 L/R SMA, ACC, PaCG 384 4.49 -2 -2 48 L Ins, MTG, STG, ITG, PPo, CO 382 4.7 -58 -6 -16 L PrG, MFG, VLPFC, IFG 322 4.95 -56 6 34 R MTG 290 4.39 68 -16 -22 L/R SCA, ACC, FMC, MOFC, VMPFC, PaCG 199 4.37 0 30 -24 L PCC 125 4.1 -6 -32 36 R LOC, PCun 109 4.49 16 -74 50 R PrG, IFG 97 4.62 60 8 20 R Cb 56 4.02 18 -58 -24 L FP, RLPFC 41 3.92 -22 60 0 SCALING > CONTROL L OP 358 6 -6 -102 16 L/R LingG, IntCalC, OFG, SupCalcC 309 4.94 -4 -80 -10 R OP, LOC 238 4.61 14 -102 14 L OFG, LingG 108 4.71 -22 -78 -8 R OFG, LingG 107 4.28 22 -74 -14 SCALING > BUNDLING L PO, CO, SPL, SMG, PoG 258 4.17 -52 -32 40 L Ins, CO 64 4.43 -40 -2 10 L PrG, IFG 57 4.27 -52 8 10 BUNDLING > CONTROL L/R OP, LOC, OFG, TOFC, LingG, Cb, IntCalC, SupCalcC, Cun 6636 7.27 -4 -102 12 R IFG, FP, MFG, DLPFC, VLPFC 84 3.95 44 38 12 R LOC, SPL 61 3.61 32 -60 48 L/R th 49 4.39 -2 -30 6 BUNDLING > SCALING L/R TOFC, Ang, OFG, ITG, LOC, OP, Cb, LingG, SMG, TFC, paHG 14993 7.86 -30 -56 -18 R PrG, PoG, MFG, DLPFC, SPL, CO 3278 5.74 46 2 30 L IFG, MFG, DLPFC, VLPFC, FO 633 5.09 -36 22 16 L/R Cb 457 5.71 -6 -78 -22 L/R PaCG, ACC, SFG, SMA 217 4.97 4 18 40 L/R PCC 71 3.67 4 -48 8 R th 58 4.01 18 -28 12 L/R OP 55 4.15 4 -98 12 R FOC, LOFC, MOFC 54 4.25 18 22 -24 L TFC, paHG, ITG 43 4.11 -36 -26 -30 z < 3:1, uncorrected, minimum extent k= 40 voxel. . Table 1.25: Dierences in condition responses (z = 3:1, uncorrected) 84 LEFT RIGHT +3.1 +6.67 CONTROL > SCALING CONTROL > BUNDLING +3.1 +7.02 CONTROL > SCALING & BUNDLING Figure 1.26: Condition contrasts CONTROL bigger. LEFT RIGHT +3.1 +6.06 SCALING > CONTROL SCALING > BUNDLING +3.1 +4.21 SCALING > CONTROL & BUNDLING Figure 1.27: Condition contrasts SCALING bigger. 85 LEFT RIGHT BUNDLING > CONTROL BUNDLING > SCALING +3.1 +7.34 +3.1 +7.93 BUNDLING > CONTROL & SCALING Figure 1.28: Condition contrasts BUNDLING bigger. LEFT RIGHT +3.1 +6.06 CONTROL < SCALING CONTROL < BUNDLING +3.1 +7.34 CONTROL < SCALING & BUNDLING Figure 1.29: Condition contrasts CONTROL less. LEFT RIGHT +3.1 +6.67 SCALING < CONTROL SCALING < BUNDLING +3.1 +7.93 SCALING < CONTROL & BUNDLING Figure 1.30: Condition contrasts SCALING less. 86 LEFT RIGHT BUNDLING < CONTROL BUNDLING < SCALING +3.1 +7.02 +3.1 +4.21 BUNDLING < CONTROL & SCALING Figure 1.31: Condition contrasts BUNDLING less. +3.1 +6.67 X: -9 Y: +13 Z: -20 CONTROL > SCALING CONTROL > BUNDLING +3.1 +7.02 +3.1 +6.06 SCALING > CONTROL SCALING > BUNDLING +3.1 +4.21 BUNDLING > CONTROL BUNDLING > SCALING +3.1 +7.34 +3.1 +7.93 LEFT RIGHT Z: +31 B A C CONTROL > SCALING & BUNDLING SCALING > CONTROL & BUNDLING X: 0 Y: -29 Z: +2 Z: +20 BUNDLING > CONTROL & SCALING X: -4 Y: +21 Z: +20 Y: +18 Figure 1.32: Dierences in responses to condition. A. Regions more active in CONTROL compared to SCALING (red), BUNDLING (blue) and both (green); B. Re- gions more active in the SCALING compared to CONTROL (red), BUNDLING (blue) and both (green); C. Regions more active in BUNDLING compared to CONTROL (red), SCALING (blue) and both (green). (Heat-maps represent z-values) 87 Regions k x y z CONTROL >SCALING & BUNDLING R ITG, MTG, LOC, SMG, Ang, STG, OP 2220 60 -62 -14 L MTG, LOC, OP, Ang, SMG, PTe 1686 -58 -54 -6 RL PCC, PCun 420 10 -50 22 L PCun, IntCalC, SupCalcC, Cun, PCC 405 -14 -62 10 L MFG, SFG, DLPFC, PrG 391 -24 16 36 R MFG, SFG, PrG 357 26 12 40 L PoG, PrG, SMG, SPL 317 -50 -20 28 R SMG, PO, STG, PoG 192 68 -28 22 L PrG, MFG, IFG, DLPFC, VLPFC 60 -54 2 28 L SPL, LOC 31 -32 -56 62 L SPL, SMG 16 -34 -48 44 L SFG 10 -16 -2 68 R PO, SMG 4 48 -26 28 R Ang, SMG 2 50 -46 26 R PO, SMG 2 46 -34 28 R MFG 2 38 16 50 R SMG 1 58 -40 22 BUNDLING > SCALING & CONTROL L Cb, LOC, OFG, LinG, TOFC, TFC, ITG 1478 -36 -84 -28 R Cb, LOC, OFG, LinG, TOFC, TFC 1245 40 -82 -26 L OP, LOC 253 -30 -96 -10 RL Cb (Posterior Lobe.Pyramis of Vermis) 35 4 -78 -30 R OP, LingG 16 14 -96 -16 R Cb (Posterior Lobe.Declive.) 10 2 -80 -20 L OP 4 -4 -96 -16 L OP 2 -14 -106 -4 L LingG 1 -4 -84 -20 SCALING > CONTROL & BUNDLING None Table 1.26: Regions responding signicantly more to one condition. 88 Regions k x y z CONTROL <SCALING & BUNDLING RL LingG, OFG, IntCalC, SupCalcC 285 -4 -84 -14 L OP 280 -8 -106 0 BUNDLING < SCALING & CONTROL L CO, PO, PoG, SMG, PTe, SPL 258 -56 -24 16 SCALING < CONTROL & BUNDLING RL LOC, OP, PCun, Cun, PCC, Ang, SPL, R SupCalcC 3967 40 -72 8 R IFG, PrG, MFG, DLPFC/VLPFC 554 44 16 14 R PrG, PoG, SMG, SPL 407 44 -22 42 L IFG, PrG, MFG, DLPFC/VLPFC 245 -36 8 18 R PrG, MFG 121 34 -8 40 R LOC, ITG 5 54 -66 -14 L LOC, SPL, Ang 4 -30 -60 42 R PrG, MFG 4 46 -2 56 L SPL, LOC, Ang 3 -28 -56 38 R PCun 3 -28 -56 38 R SFG, PrG 1 18 -8 72 Table 1.27: Regions responding signicantly less to one condition. 89 +3.1 +6.06 X: -7 Y: -75 Z: -20 CONTROL < SCALING CONTROL < BUNDLING +3.1 +7.34 +3.1 +6.67 SCALING < CONTROL SCALING < BUNDLING +3.1 +7.93 BUNDLING < CONTROL BUNDLING < SCALING +3.1 +7.02 +3.1 +4.21 LEFT RIGHT Y: -97 B A C CONTROL < SCALING & BUNDLING SCALING < CONTROL & BUNDLING X: -5 Y: +20 Z: +26 Y: +13 BUNDLING < CONTROL & SCALING X: -5 Y: +13 Z: +16 Y: -32 Figure 1.33: Dierences in responses to condition. A. Regions less active in CON- TROL compared to SCALING (red), BUNDLING (blue) and both (green); B. Regions less active in the SCALING compared to CONTROL (red), BUNDLING (blue) and both (green); C. Regions less active in BUNDLING compared to CONTROL (red), SCALING (blue) and both (green). (Heat-maps represent z-values) Appendix 7. This section reports results from GLM 3 . Table 1.28 reports the clusters involved in salience-tracking in GLM-3 in the three conditions. Table 1.29 reports clusters involved in common salience- tracking regions across all conditions. 90 Regions k z-score x y z CD R PoG, SPL, SMG, Ang, LOC, PCun, Cun, 4024 5.07 26 -50 52 OP, Cb, TOFC, TFC, OFG, ITG L TOFC, TFC, ITG, OFG, OP, LOC, paHG, LingG, Cb 1603 4.65 -22 -52 -18 R PoG, SMG, CO, PO, PTe, Ins, Pu 375 4.17 52 -16 30 ScD RL PCun, LOC, OP, PCC; L PoG, PrG, SPL, SMG, PO, 4026 5.35 -10 -64 52 PTe, CO, SFG, MFG R LOC, OP, MTG, Ang, 298 4.4 46 -82 -8 L CO, PPo, Ins, FO, IFG, PrG, TP 242 4.79 -50 2 2 R LOC, OP 200 4.22 38 -84 14 BD RL LOC, OP, TOFC, OFG, Cb, LingG, PoG, SMG; 11070 5.58 -28 -88 18 R SFG, MFG, PrG L SFG, MFG, PrG, paHG, Hi 357 4.17 -28 -8 44 Clusters are reported at p< 0:05. Table 1.28: Regions tracking salience during the deliberation period. Appendix 8. This section reports results from model PPI. Table 1.30 reports the clusters showing signicant connectivity with the seed region VMPFC. 91 Regions k x y z R OFG, LOC, OP, ITG, MTG, Cun, PCun, SPL, PoG, SMG, Ang 3138 42 -74 -14 L ITG, LOC, MTG, OP, Cun, PCun, SPL, PoG, SMG 1793 -50 -62 -10 R Cb, TOFC, TFC, LinG, OFG, LOC, paHG 595 26 -64 -26 L Cb, TOFC, TFC, LinG, OFG, LOC, ITG, paHG 398 -30 -48 -30 L CO, PrG, PoG 173 -52 2 4 R SMG, PoG 107 50 -24 32 R PrG, SFG, MFG 99 28 -10 48 L PrG, SFG, MFG 75 -22 -12 46 R SMG, PO, PoG, PTe 61 62 -22 22 R ACC, SMA 59 6 -2 44 L CO, PoG, PO 53 -52 -18 14 L Ins, CO, Pu 46 -36 -2 8 L PO, SMG, PTe 32 -52 -40 24 L PoG, SMG 29 -62 -20 32 R OFG, LOC 13 46 -66 -20 L OP, LOC, Cun 12 -14 -96 16 L LOC, OP, OFG 11 -32 -90 -12 R PrG 8 38 -16 60 L SMG, PoG 7 -50 -24 30 L ACC, SMA 4 -2 -6 48 L PO, CO, SMG 3 -50 -26 22 L LOC, OFG 2 -34 -86 -16 R PCun, PCC, LingG 2 18 -52 6 L ACC, SMA 2 -8 -2 50 R PoG, SMG 2 64 -14 28 L Cb 2 -24 -58 -30 L ACC, SMA 1 -6 -2 42 L SMG 1 -50 -24 26 L PO, PTe, SMG 1 -60 -36 24 R PCun 1 20 -52 10 L LOC, MTG 1 -52 -62 10 R MTG 1 46 -56 8 R LingG, PCC 1 20 -50 2 R SFG 1 20 -10 68 z< 2:3, uncorrected, minimum extent k = 1 voxel. Table 1.29: Regions tracking salience in all conditions. Interaction map of CD, ScD and BD. 92 Regions k z-score x y z SCALING Connectivity R MFG, SFG, PrG 655 3.5 22 8 48 SCALING Connectivity > CONTROL Connectivity L IFG, MFG, DLPFC, VLPFC 486 3.92 -52 20 24 R OP, LOC, OFG 485 3.82 16 -100 -10 L OP, LOC, OFG, MTG 433 3.66 -30 -94 -18 Table 1.30: Connectivity analysis of the VMPFC 93 Chapter 2 The impact of aging on value representation 2.1 Introduction Reasoning about value is an essential element of decision-making; most day-to-day decisions involve comparing items and making trade-os between them. Economic theories assume that individuals have sta- ble preferences. When faced with repeated choices between bundles of goods, individuals are expected to make choices that are internally consistent. The notion refers to the fact that choices should not con- tradict one another: if choosing item A over item B and item B over item C, an individual should also choose item A over item C. Such 94 \consistent" choice is guaranteed if the individual is bale to rank all alternatives and to choose systematically according to this rank. While behavioral studies indicate generally high choice consistency among adults (Battalio et al., 1973; Cox, 1997; Sippel, 1997; Andreoni and Miller, 2002; Choi et al., 2007a,b), recent studies demonstrates high sensitivity to cognitive development (Harbaugh et al., 2001; Bro- cas et al., 2019b) or impairment associated with normal aging (Fin- ucane et al., 2005, 2002). In particular, Brocas et al. (2019a) shows marked dierences in consistency between young and old adults when choices become complex. Given the growing complexity of economic products (insurance, pension and health plans, mortgages, telephone plans, etc.), older adults may have diculty making decisions that ac- curately re ect their underlying preferences. The study also reveals a correlation between inconsistency and working memory and proposes that the ability to make more complex value comparisons involves the working memory system, responsible for the short-term mental main- tenance and manipulation of information. Our study tests the prediction that dierences in consistency be- tween young and old adults stem from dierences in value representa- 95 tion in a network of brain regions that have been associated to value tracking. These include the medial orbitofrontal cortex (MOFC), or sometimes more narrowly the ventromedial prefrontal cortex (VMPFC), and the left dorsolateral prefrontal cortex (DLPFC). Evidence from lesion studies converges in identifying the MOFC/VMPFC as a critical region in valuation when deciding be- tween two single alternatives (Rangel et al., 2008; Henri-Bhargava et al., 2012; Fellows and Farah, 2007). Parallel ndings from fMRI studies are in agreement with those results. A participant's willing- ness to pay for a good or item is thought to represent their valuation of the item (Chib et al, 2009; Hare et al, 2008) and is correlated with activation in the VMPFC (Plassmann et al., 2007). There is also convergent evidence that the DLPFC is involved in value representation requiring higher order information, such as in self-regulation and self-control (Hutcherson et al., 2012; Harris et al., 2013). The DLPFC has also been found to be functionally connected with the value coding regions in self-control paradigms (Hare et al., 2009) and in multi-attribute paradigms (Rudorf and Hare, 2014). Last, the DLPFC was found signicantly more activated in studies in 96 which options involved a con ict to be resolved (Baumgartner et al., 2011; de Wit et al., 2009). This is consistent with activation studies that have shown that dorsal frontal regions are activated during tasks that are experienced as dicult (Braver et al., 1997; Cohen et al., 1994, 1997; Monterosso et al., 2007; Luo et al., 2012) and during task switch- ing (Dove et al., 2000). During tasks that tax executive function, activation is evoked in the DLPFC and the posterior parietal cortex (PCC) (Goldberg et al., 1998; Osherson et al., 1998; Goel et al., 1997; Baker et al., 1996; Berman et al., 1995; Nichelli et al., 1994; Petrides, 1994). In such contexts, the DLPFC is dierentially recruited as tasks become more complex (Carlson et al., 1998; Braver et al., 1997; Co- hen et al., 1997; Baker et al., 1996; Demb et al., 1995; Christo et al., 2001). This relationship extends to tasks requiring the explicit repre- sentation and manipulation of knowledge, where the ability to reason relationally is essential (Kroger et al., 2002). The VMPFC shows little structural decline with aging (Fjell et al., 2009) and functions involving VMPFC are usually well maintained (MacPherson et al., 2002; Mather, 2012). By contrast, in activation studies with younger and older adults, the DLPFC was dierentially 97 activated with age in executive function and working memory tasks (Grady et al., 2006; Rypma et al., 2001). Furthermore, longitudinal studies measuring brain changes have shown that the dorsal and lat- eral regions of the prefrontal cortex experience signicant atrophy with age (Raz et al., 2005; Resnick et al., 2003). These results taken to- gether suggest that the aging process should aect primarily complex value computations that require the DLPFC. At the same time, older adults should keep their ability to make simple value comparisons that primarily involve the VMPFC. The present study involves simple choice tasks directed at distin- guishing functional elements contributing to choice complexity that may deteriorate with aging. Participants were asked to choose between real food options involving single item options and bundled items op- tions. Bundles varied in complexity and were either composed of the same two single items or of two dierent single items. Participants performed these tasks during functional Magnetic Resonance Imaging (fMRI), allowing us to track the neural correlates of their decisions and to compare them between young and old adults. We hypothesize that simple value computations rely on the MOFC, which remains 98 intact among older adults, allowing them to make consistent choices in simple settings. However, complex value computations required to evaluate bundles of goods depend also on the network of brain regions recruited during executive function tasks. These include lateral sectors of the prefrontal cortex, the anterior cingulate cortex, and the intra- parietal sulcus, which we will collectively refer to as \frontoparietal Network (FPN)" (Seeley et al., 2007). As these regions (in particu- lar the lateral prefrontal regions) decline with age, value computation for complex goods is predicted to be impaired among older adults, resulting in inconsistencies. Thus we predict that the functional com- ponents critical for decision-making in tasks that require the DLPFC are more compromised in older adults (Peters et al., 2007). 2.2 Materials and methods Subjects Forty old adults (mean age 66 yo, 20 female and 20 male, all right handed), hereafter OA, and sixty eight healthy young adults (mean age 23 yo, 36 female and 32 male, all right-handed), hereafter YA, were recruited from the Los Angeles Behavioral Economics Laboratory's 99 subject pools and the Healthy Minds subject pool, all at the University of Southern California. Subjects could participate if they satised the standard eligibility criteria for fMRI studies. We excluded subjects who reported to have food allergies, food restrictions or to be picky eaters. All participants received a $50 show-up fee for participating. They were also rewarded with one of their choices, selected randomly at the end of the session. Eight hereafter YA and six hereafter OA participants were excluded because of incomplete data collection or excessive head movement during scanning. The Institutional Review Board of USC approved the study. Procedure Participants were instructed to not eat for at least 4 hours before the experimental session. They were also instructed that they would have to stay after the session to consume what they had obtained and that they could not take any of the food items with them when they leave. This was implemented to make sure participants were hungry and thinking carefully about their choices during the session. The procedure was explained beforehand so that each participant knew that choices were real, and they should make their best decision in 100 every trial. Each participant was asked to rank 30 single item options by order of preference. This ranking was used to create 40 bundles, 20 com- binations of 2 same single items, and 20 combinations of 2 dierent single items. The participant was then asked to include those bundles in their previous ranking. We then selected 11 single item options, 10 combinations of 2 same single items and 10 combinations of 2 dierent single items. In the main task, each participant made binary choices involving these 31 options in the scanner. One of the 11 single item option was a reference option, denoted hereafter by REF. Choices were divided into three conditions (see Fig. 1.1A): CONTROL, SCALING and BUNDLING. In each of the CONTROL trials, the participant had to choose between REF and one of the 10 remaining single item. In each of the SCALING trials, the participant had to choose between REF and one of the 10 combinations of 2 same single items. In each of the BUNDLING trials, the participant had to choose between REF and one of the 10 combinations of 2 dierent single items. In all cases, REF was o-screen, it was the same for each trial and was shown to the participant at the beginning of the experiment. The other option 101 was on-screen and it was displayed at the beginning of each trial. (see Fig. 1.1 (B)). Each individual trial was repeated 9 times for a total of 90 trials in each condition. The circles at the bottom of the screen told the participant what button selected which option, the solid cir- cle always representing REF. The button mappings were randomly assigned for each trial. 1 When the participant responded, the circle representing the chosen option was framed in a square to let the par- ticipant know that the their answer was recorded. The screen then advanced to a xation cross for the remainder of the trial. The fMRI task was optimized for detecting neural responses. Trials order and inter-stimulus intervals were optimized by Optsec2, a tool that auto- matically schedules events for rapid-presentation event-related fMRI experiments (Dale, 1999) and organized into 5 runs. We chose the options in order to ensure that each of the three conditions CONTROL, SCALING and BUNDLING had symmetrical sets or low, medium of high on-screen value options centered around REF and that the distribution of value was similar across conditions 1 Subjects had button boxes in each hand when they were in the scanner. They were instructed to make choices by pressing a button in the hand corresponding to the option, as represented by the circle, they wanted. For example, if they wanted the reference option and the solid circle was on the right side of the screen in that trial, they could select it by pressing a button in their right hand. If they wanted the on-screen option instead, they could select it by pressing a button corresponding to the hollow circle, which in that case would be a button in their left hand. 102 (see Fig. 2.1(C)). CONTROL SCALING BUNDLING A SCANNER TASK Off-Screen REF B C Figure 2.1: Experimental Design. A. Each trial was a choice between the reference item (REF) and a food option in either of 3 conditions: CONTROL (one single item), SCALING (two same single items) or BUNDLING (two dierent single items). B. Only the latter food option was presented on screen. All trials were self-paced. C. We designed the task to best approximate a distribution of options centered around the REF item (orange) in each task CONTROL (red), SCALING (dark blue) and BUNDLING (light blue). MRI data acquisition Neuroimaging data were collected using the 3T Siemens MAGNETOM Tim/Trio scanner at the Dana and David Dornsife Cognitive Neuro- 103 science Imaging Center at USC with a 32-channel head-coil. Partic- ipants were laid supine on a scanner bed, viewing stimuli through a mirror mounted on head coil. Blood oxygen level-dependent (BOLD) response were measured by echo planar imaging (EPI) sequence with PACE (prospective acquisition correction) (TR = 2 s; TE = 25 ms; ip angle= 90; resolution = 3 mm isotropic; 64 x 64 matrix in FOV = 192 mm). A total of 41 axial slices, each 3 mm in thickness were acquired in an ascending interleaved fashion with no interslice gap to cover the whole brain. The slices were tilted on a subject by subject basis { typically 30 from the anterior commissure posterior commis- sure plane { to minimize signal dropout in the orbitofrontal cortex (Deichmann et al., 2003). Anatomical images were collected using a T1-weighted three-dimensional magnetization prepared rapid gradient echo (MP-RAGE with TI = 900 ms; TR=1.95 s; TE: 2260 ms; ip angle = 9; resolution = 1 mm isotropic; 256 256 matrix in FOV = 256-mm) primarily for localization and normalization of functional data. These scans were co-registered with the participant's mean EPI images. These images were averaged together to permit anatomical localization of the functional activations at the group level. 104 MRI data preprocessing Image analysis was performed using FSL algorithms organized in a nipype pipeline. Computation for the work described in this paper was supported by the University of Southern California's Center for High- Performance Computing (hpcc.usc.edu).The structural images were skull-striped then aligned and spatially normalized to the standard Montreal Neurological Institute (MNI) EPI template. The functional images were motion and time corrected. They were spatially smoothed using a Gaussian Kernel with a full width at half-maximum of 5mm. We also applied a high-pass temporal ltering using a lter width of 120s. Behavioral analysis Participants were asked to make choices between two snack options. The rst option was always the same o-screen reference option de- noted by REF while the second option was an on-screen variable op- tionVAR j (j =f1;:::;Ng). We constructed a Random Utility Model (McFadden et al. (1973); Clithero and Rangel (2013a)) in which we as- sumed that the utility derived by optionVAR j depends on the value of the food snack and a stochastic unobserved error component j . For- 105 mally, u(REF ) =v 0 + 0 and u(VAR j ) =v j + j . The probability of choosing optionVAR j is thereforeP j =Pr[ 0 j <v j v 0 ]. Assum- ing that the error terms are independent and identically distributed and follow an extreme value distribution with cumulate density func- tionF ( k ) =exp(e k ) for allk = 0;j, then the probability that the participant chooses option VAR j is the logistic function P j = 1 1 +e v j v 0 We then constructed a likelihood function and we used Maximum Likelihood Estimation techniques to retrieve parameters v j given the observed choices. This procedure was implemented in Matlab with standard algorithms. For each individual, we assigned an implicit ranking of all options based on these retrieved values. We then as- signed implicit rankings in the CONTROL condition (CV), SCALING condition (ScV) and BUNDLING condition (BV). Checking for consistency across trials In principle, if a subject's choices are well represented by the Random Utility Model, we should observe that most choices are consistent with 106 estimates. For each individual, we generated the choices they should have made in all trials if they were choosing according to the value estimates and we compared these with their actual choices. More precisely, we counted all choices that were not consistent with the value estimates (and henceforth with the implicit ranking) and we computed the percentage of these inconsistent choices. We computed an overall Consistency Rate (CR) and one for each condition. Testing for matching between implicit and explicit ranking. Recall that the snacks that we included in the fMRI task were se- lected by asking participants to rank all options. The selection was made in such a way that the distributions of expected binary choices between the on-screen and the o-screen options in the main task would be balanced and comparable across conditions. This implicitly assumes that binary choices in the scanner should be consistent with the ranking of all options outside the scanner. To check that assump- tion, we counted all choices in the scanner that were consistent with the ranking elicited outside the scanner. This exercise amounted to compare the consistency between that ranking and the implicit rank- ing based on estimated values. We computed an overall Matching rate 107 (MR) and one for each condition. Analysis of reaction times We recorded reaction times between the onset of the stimulus and the time at which a choice was made in each trial. We analyzed indi- vidual and group dierences across conditions. For each individual, we computed a mean Reaction Time (RT) it took them to deliberate overall and in each condition. We also looked for systematic dier- ences across conditions and across the type of choices (on screen vs. o-screen) ended up to be made. For each participant, we computed the mean Reaction Time (RT) it took them to deliberate in each of the three conditions. These measures were designed to analyze individual dierences across conditions. MRI data analysis We estimated general linear models (GLMs) of BOLD responses. Each aspect of the task was encoded in a regressor for the GLM. To identify what signal was associated with a particular condition, we constructed indicator regressors that take value 1 whenever the participant is per- forming a trial within a condition and 0 otherwise. To identify the 108 neural activity associated with the subjective value of the on-screen option, we created a parametric regressor that is equal to the value proxy (details below) of the on screen option and changes every time the on-screen option changes. When there is nothing on the screen, both regressors are 0. The models also include motion parameters (regressors for translation and rotation as well as artifact regressors controlling for quick jerking movements) and regressors for each run as nuisance regressors. All regressors were convolved with the canon- ical form of the hemodynamic response. The values in the regressors were applied from the onset of the stimulus until a choice was made (average duration, 1.47s). The GLMs took the general form: BOLD i = [H 1 (R a )] a i +R b b i +e i WhereBOLD i is the time-series of BOLD signal at each voxeli,H 1 is the hemodynamic response function (HDF) used by FSL (Jenkinson et al. (2012)) applied to the primary regressor matrixR a (each column is a primary regressor) ande i is a gaussian noise. The GLM solves for a i and b i to minimize the error e i . To analyze the in uence of an in- dicator regressor the coecients a i are contrasted against each other. 109 These -contrasts are used to generate interpretable statistics. Every GLM was estimated in several steps. First, we estimated the model separately for each participant. After each GLM was t to the image time-series, the -contrasts are combined at the subject level using a Fixed Eects Model, then combined in a Mixed Eects Model to cre- ate group level voxel-wise t-statistics. We implemented the Randomise procedure is FSL (Winkler et al., 2014) that uses permutation meth- ods to produce a test statistic image using threshold-free cluster en- hancement (Smith et al. (2009)). All images are thresholded at p<0.05 FWE corrected unless otherwise noted. We used FSL Harvard-Oxford Subcortical and Cortical Structural Atlas and Talairach Daemon La- bels to list every gray matter region identied within each cluster. Functional regions were added where it was appropriate, using some of the notations from Dixon et al. (2017). For the regions for which we formed a priori hypothesis, namely the VMPFC, the MOFC and the DLPFC, and given the ambiguity around their description in the literature, we set an ex ante rule regarding how we would report our evidence (see Fig. 2.2). 110 Y: +28 LEFT RIGHT X: +4 Z: -20 Harvard-Oxford Cortical Structural Atlas - ACC +10 +98 +10 +96 - FP +10 +87 - PaCG +10 +98 - SCA +10 +97 - FMC - FOC +10 +96 A B Y: +14 X: 0 Z: +64 +10 +79 - SFG +10 +83 - MFG +10 +99 - PCC +10 +86 - SMA +10 +84 - PrG +10 +81 - IFG VMPFC MOFC DLPFC Figure 2.2: Location of VMPFC, MOFC and DLPFC with regard to Harvard- Oxford Cortical Structural Atlas. A. Posterior part of VMPFC was dened by the anterior part of Subcallosal Cortex (at Y=+28), while the anterior part of VMPFC was dened by the posterior part of Frontal Pole (FP), and it overlayed with dorsal part of Frontal Medial Cortex (FMC) and ventral part of Paracingulate Gyrus/Anterior Cingulate Cortex. MOFC was dened by ventral part of FMC/FP and surrounded by medial parts of Frontal Orbital Cortex. B. DLPFC was dened according to Dixon et al., 2017 paper (BA = 9,46 and 8), by the presence of Middle Frontal Gyrus (as well as junctions with Superior Frontal Gyrus and Inferior Frontal Gyrus), with the most posterior border right before Supplementary Motor Area (Y=+14). Testing for dierences across populations We used GLM* consisting of the three parametric regressors repre- 111 senting the subjective value of the on-screen option in the CONTROL condition (CV), SCALING condition (ScV) and BUNDLING condi- tion (BV) and the 3 indicator functions C, S and B capturing con- ditions. We labeled CV, ScV and BV the clusters of regions that tracked the value regressor CV, ScV and BV. Given the decision- making literature on subjective value has consistently reported certain regions to be signicantly associated with value tracking, we were in- terested in identifying whether activity overlapped with regions re- ported elsewhere. We used the Meta Value (hereafter MV) from Clithero and Rangel (2013b) that comprises VMPFC and DLPFC among others. GLM* allowed us to model (i) dierences in value tracking between conditions and (ii) non value-related dierences be- tween conditions. To test for dierences across populations, we con- trasted the results obtained in the two populations. Region of interest (ROI) analysis Given earlier research is pointing to the signicant role of the MOFC and the VMPFC in value representation, we had a strong a priori interest in those regions (Plassmann et al., 2007; Hare et al., 2009; Sokol-Hessner et al., 2012; Kable and Glimcher, 2007). The ROI for 112 the VMPFC, hereafter vmPFC, was dened by 10 voxel sphere with the center at [0,46,-6] in MNI152 space. It encompasses VMPFC ac- tivity reported in Kahnt et al. (2011), Chib et al. (2009), McClure et al. (2004), O'Doherty et al. (2006), Kim et al. (2010), Lim et al. (2011) and Levy and Glimcher (2011). The MOFC ROI, hereafter mOFC, was dened by 7 voxel sphere with the center at [-8,44,-20] in MNI152 space. This corresponds to the area where value tracking activity was reported by Arana et al. (2003). We also used the results of GLM* obtained on the YA population as a template and we conducted ROI analysis on regions of relevance identied to respond to value and condition. All ROIs were performed on the second level cope images and we extracted each subject's con- trast estimates averaged across all of the voxels in the mask. Testing for functional connectivity To test for dierences in condition-dependent functional connectivity, we used PPI*, a general psychophysiological interaction (gPPI) model (McLaren et al., 2012; De Martino et al., 2013a; O'Reilly et al., 2012; Clewett et al., 2014a). We dened a region of interest (ROI) which we used as the seed for our analysis. The model created a new GLM 113 in which the deconvolved activity of the seed region was assigned to the regressors modeling the condition and reconvolved with the HDF. The gPPI model searched for how and when other regions connected to that seed region during a specic condition, but not in any other condition. The seed region in PPI* was our a priori ROI vmPFC. 2.3 Results 2.3.1 Behavioral measures We counted very few missed trials resulting in no choice (1.64% of the trials among YA and 4.12% among OA) indicating that participants were attentive and had enough time to select their preferred option. We found dierences in Consistency Rates (CR) across populations. On average, 87% of the choices of a subject were consistent with their implicit rankings (87.5% in the CONTROL condition, 88% in the SCALING condition and 86.5% in the BUNDLING condition). Across populations, 90% of all choices were consistent among YA against 83% among OA, indicating a high level of consistency in both popula- tions. However, OA were signicantly less consistent in all conditions compared to YA (t =3:29, p = 0:002 in CONTROL, t =3:23, 114 p = 0:002 in SCALING and t =3:31, p = 0:001 in BUNDLING) suggesting the existence of dierences in the process through which values were represented. Within populations, the percentages of con- sistent choices were not signicantly dierent across conditions except for the YA group that was signicantly more consistent in SCALING compared to BUNDLING (t = 2:7637, p< 0:008). These ndings are summarized in Figure 2.3 (A). There were also dierences in Matching Rates (MR). On average 72% of the choices of YA in the fMRI task were consistent with the rankings elicited outside the scanner (71% in the CONTROL condi- tion, 76% in the SCALING condition and 68% in the BUNDLING condition). This number dropped to 59% for OA (59% in the CON- TROL condition, 60% in the SCALING condition and 59% in the BUNDLING condition). These dierences between populations were signicant (t =3:64,p< 0:001 in CONTROL,t =4:41,p< 0:001 in SCALING and t =2:91, p = 0:005 in BUNDLING) and sug- gested a diculty among OA to represent value similarly in dierent contexts such as ranking all options altogether vs. making pairwise choices (Fig. 2.3 (B)). 115 Consistency Rates were correlated across conditions among YA (Pearson = 0.74,p< 0:001 between CONTROL and SCALING; Pear- son = 0.69, p< 0:001 between CONTROLand BUNDLING, Pearson = 0.80, p< 0:0001 between SCALING and BUNDLING) and among OA (Pearson = 0.86, p< 0:001 between CONTROL and SCALING; Pearson = 0.77, p < 0:001 between CONTROL and BUNDLING, Pearson = 0.84, p < 0:0001 between SCALING and BUNDLING). Matching Rates were also signicantly correlated across conditions among YA (Pearson = 0.41, p = 0:001 between SINGLE and SCAL- ING; Pearson = 0.40, p = 0:002 between SINGLE and BUNDLING, Pearson = 0.72,p< 0:0001 between SCALING and BUNDLING) and OA (Pearson = 0.84, p< 0:001 between CONTROL and SCALING; Pearson = 0.73, p < 0:001 between CONTROL and BUNDLING, Pearson = 0.79, p < 0:0001 between SCALING and BUNDLING). Hence, participants who tended to be more consistent in their choices in one condition tended to be also more consistent in the two others. Also, participants who were more able to re ect the rankings they re- ported outside the scanner in one condition were more able to re ect them in the other two. 116 Consistency Rates and Matching Rates were not always signi- cantly correlated however. They were signicantly correlated among YA (Pearson = 0.28, p = 0:028 in CONTROL; Pearson = 0.49, p< 0:001 in SCALING and Pearson = 0.29,p = 0:024 in BUNDLING) but they were only correlated in BUNDLING among OA (Pearson = 0.20, p = 0:2 in CONTROL; Pearson = 0.22, p = 0:16 in SCALING and Pearson = 0.36, p = 0:02 in BUNDLING). Hence among YA, those who were more able to make consistent pairwise choices were also responding according to what they previously reported. This was consistent with a model in which participants would try to draw noisily from their preferences. More noise resulted in more reversals in pair- wise choices and more dierences with respect to stated preferences outside the scanner. This was not the case among OA: a signicant fraction of participants were making choices that did not re ect the preferences stated outside the scanner but were still relatively consis- tent with each other in the pairwise choice-task (Fig, 2.3(C)). Reaction times We found no signicant dierences in the time it took OA and YA to evaluate options in each of the conditions. We found however that 117 . Figure 2.3: Behavior and reaction times. A. OA were signicantly less consistent across trials than YA in all conditions. B. OA were also less consistent with their explicit rankings in all conditions. C Participants CR and MR where signicantly correlated among YA only. D. In both populations, it took more time to participants to choose in BUNDLING and less time in SCALING compared to CONTROL. it took participants signicantly more time to choose in BUNDLING (t =5:86, p< 0:001 for YA and t =3:46, p = 0:001 for OA) and 118 less time to choose in SCALING (t = 4:45, p < 0:001 for YA and t = 3:13, p = 0:003 for OA) compared to CONTROL (Fig. 2.3(D)). Behavior: summary YA and OA did not dier in terms of reaction times but they did in terms of quality of decision-making: CR were signicantly lower among the older participants indicating that their choices were lacking internal consistency. Further, MR were signicantly lower, suggesting that the context of the choices (binary in the scanner vs. full rankings outside the scanner) were playing a role. We however did not nd qualitative dierences in the way participants approached conditions. SCALING was quicker and yielded higher CR and MR in both popu- lations. BUNDLING was taking more time and yielded lower CR and MR in both populations. Furthermore, CR and MR were correlated among YA but not among OA. 2.3.2 Overall dierences between OA and YA We assessed dierences across OA and YA in BOLD responses to regressors modeled in GLM*. This exercise (Table 2.1) revealed that only few regions were signicantly more active in the YA population 119 compared to the OA population. In particular, there was no signif- icant dierence in value tracking neural correlates. Only regions in the left ACC were signicantly more active in response to conditions CONTROL and BUNDLING in YA. By contrast, several regions were signicantly more active in the OA population compared to the YA population. Clusters in the primary motor cortex (Precentral Gyrus) and in the PCC were signicantly more active in all conditions. Sig- nicant activity was found in the SCALING condition, in particular in regions involved in somatosensory processing (Supramarginal Gyrus, Postcentral Gyrus) and activity in the frontal gyrus (Middle and Su- perior Frontal Gyri) and the DLPFC. This is summarized in Figure 2.4. Overall dierences: summary There was no signicant dierences regarding value-tracking but condition- tracking was operating dierently between populations. Activity in the ACC was specic to the YA population while activity in the PCC was specic to the OA population. These dierences were also modu- lated by condition. Signicantly more activity was found in SCALING in the OA population. 120 Region k Max-score x y z OA > YA ScV: Left Lingual Gyrus 1 0.951 -14 -62 -8 C Right Precentral Gyrusy,z 6082 0.996 10 -20 46 Sc Left Postcentral Gyrusy 22389 1 -40 -18 46 Left Planum Temporale 119 0.962 -50 -42 18 Left Supramarginal Gyrus 110 0.965 -64 -32 22 Left Frontal Pole 84 0.97 -38 40 -8 Left Precentral Gyrus 60 0.961 -64 0 8 Left Precuneus Cortex 38 0.953 -16 -62 12 Right Pallidum 14 0.954 20 -12 -8 B Left Precentral Gyrusy 9864 1 -32 -24 62 YA > OA C Right/Left Anterior Cingulate Cortex 436 0.991 0 30 24 Left Frontal Orbital Cortex 4 0.953 -18 14 -16 B Left Anterior Cingulate Cortex 213 0.986 -2 32 22 Left Anterior Cingulate Cortex 25 0.968 0 26 6 Right Caudate 22 0.96 14 18 4 Region is identied as peak activity. y Overlap with PCC.z Overlap with DLPFC. Table 2.1: Dierences in neural correlates of regressors between OA and YA 121 OA > YA OA > YA Y: +24 Meta Value OA > YA A OA > YA X: 0 CONTROL YA > OA Meta Value Meta Value YA > OA Z: -18 X: 0 Meta Value B C A X: -4 SCALING Meta Value Meta Value Y: +74 B C L R L R L R A OA > YA Y: +18 BUNDLING YA > OA Meta Value YA > OA X: -4 Meta Value B C L R X: -4 Meta Value . Figure 2.4: Regions responding dierentially in the YA and OA populations. In CONTROL, more activity is observed in A. the ACC and B. Left MOFC among YA and in C. the PCC among OA. In SCALING, more activity is observed among OA in A. the left LOFC, the B. left DLPFC and C. the PCC. In BUNDLING, more activity is observed in A. the ACC and B. the Caudate among YA and in C. the PCC among OA. Green represents MV. 2.3.3 Evaluation of YA core task-related regions The previous analysis suggests that the two populations did not dier in terms of value-tracking. We put that result to the test by conducting ROI analysis on the core regions identied in the independent study of the YA population. The results were obtained by implementing GLM* on the YA data only. 122 Value Tracking in YA was associated with (i) regions in the left VMPFC and the left DLPFC identied as part of the value tracking system in earlier literature; (ii) regions usually involved in complex visual processing and (iii) regions in the left Cerebellum. Condi- tion responses were identied in regions involving the DLPFC and the VLPFC as well as the mid-Cerebellum. Table 2.2 summarizes the ROI we retained to evaluate dierences between populations. We re- port below the distribution of mean parameters in these ROIs (Fig. 2.5, Fig. 2.6 and Fig. 2.7). We used regression analysis controlling for age and contrasts to assess age-related dierences in BOLD responses in these regions (Table 2.3 and Table 2.4). The regressions inform us of the nature and signicance of the relationship between BOLD responses to regressors and the age of the participants after removing the eects of conditions. We found that BOLD responses were decreasing signicantly with age in the VMPFC and DLPFC. Visual Value regions and the Cere- bellum were more spared by aging. Also, only BOLD responses to conditions in the left VLPFC were decreasing with age. Other regions involved in condition tracking did not exhibit signicant age-related 123 Region Identier k x y z (A) V-Core YA Left Ventromedial Prefrontal Cortex VMPFC 49 -6 30 -84 Left Dorsolateral Prefrontal Cortex DLPFC 287 -24 12 -120 Left Temporal Occipital Fusiform Cortex L-Visual 60 -26 -54 -124 Right Temporal Occipital Fusiform Cortex R-Visual 82 40 -48 8 Left Cerebellum Left Cb 6 -22 -54 -116 (B) C-Core YA Left Dorsolateral Prefrontal Cortex L-DLPFC-C 391 -24 16 36 Left Dorso- and Ventro-lateral Prefrontal Cortex L-VLPFC-C 60 -54 2 28 Left Dorso- and Ventro-lateral Prefrontal Cortex L-VLPFC-Sc 245 -36 8 18 Right Dorso- and Ventro-lateral Prefrontal Cortex R-VLPFC-Sc 554 44 16 14 Left/Right Cerebellum R/L Cb 35 4 -78 -30 Table 2.2: Regions of interest in YA. Clusters active in response to value (A) and to specic conditions (B) in the YA population. VMPFC DLPFC L-Visual R-Visual Left Cb Age -0.42 -0.27 -0.32 -0.11 -0.22 (0.15) (0.12) (0.19) (0.17) (0.20) SCALING -6.55 -4.37 3.08 4.58 1.51 (8.52) (4.82) (7.79) (7.63) (8.28) BUNDLING -12.89 -0.61 8.16 4.40 1.51 (9.08) (6.24) (8.43) (7.68) (8.28) Constant 35.94 22.27 24.53 16.13 10.75 (9.65) (6.06) (9.68) (8.35) (10.42) Observations 282 282 282 282 282 Clusters 94 94 94 94 94 Adj. R 2 0.029 0.013 0.006 -0.007 -0.004 p< 0:1, p< 0:05, p< 0:01, p< 0:001 clustered standard errors at the subject level Table 2.3: BOLD responses in V-Core YA. Age-related changes in BOLD responses were found in value-tracking regions VMPFC, DLPFC and L-Visual. 124 Meta Value VMPFC X: -4 A B Y: +21 L R Meta Value L-DLPFC . Figure 2.5: Activity in VMPFC and DLPFC. Responses to Value Regressors were signicantly lower in the CONTROL condition in the OA population for both core value tracking regions A. VMPFC and B. DLPFC. dierences. These results taken together indicate that age-related dif- ferences existed within the YA core neural correlates of value and conditions. Last we also compared mean parameters in our two a pri- ori ROIs. We found that activity in our a priori vmPFC and mOFC were following the same patterns as activity in VMPFC. They both were impacted by age (Fig. 2.8). Evaluation of core task-related regions: summary Aging was associated with lower responses to regressors in the left 125 L-Visual Z: -20 A B R-Visual C X: -20 L-Cb L R Z: -20 L R Meta Value Meta Value Meta Value . Figure 2.6: Activity in Visual Value Regions and Left Cerebellum. Responses to Value regressors were marginally lower in the BUNDLING condition in A. L-Visual. No dierences are observed in B. R-Visual and C. L-Cb. DLPFC-C L-VLPFC-C L-VLPFC-Sc R-VLPFC-Sc L/R Cb Age 0.14 -0.60 -0.38 -0.05 -0.02 (0.13) (0.26) (0.27) (0.18) (0.20) SCALING -6.45 -6.93 -7.36 -9.00 1.42 (1.63) (2.35) (2.01) (1.94) (2.21) BUNDLING -5.83 -5.99 -1.58 0.12 9.16 (1.43) (2.53) (2.03) (1.71) (2.17) Constant 50.15 121.73 91.74 63.71 53.70 (6.67) (13.70) (12.23) (9.49) (8.24) Observations 282 282 282 282 282 Clusters 94 94 94 94 94 Adj. R 2 0.075 0.029 0.015 0.001 0 p< 0:1, p< 0:05, p< 0:01, p< 0:001 clustered standard errors at the subject level Table 2.4: BOLD responses in C-Core YA. Age-related dierences in BOLD re- sponses were found only in L-VLPFC-C. 126 C > Sc Y: +16 A B X: -54 X: -42 C D Y: +44 C > B C > Sc&B L R Sc < C Sc < B Sc < C&B DLPFC-C L-VLPFC-C L-VLPFC-Sc R-VLPFC-Sc C > Sc C > B C > Sc&B Sc < C Sc < B Sc < C&B E B > C Y: +16 B > Sc B > C&Sc L/R Cb . Figure 2.7: Regions responding dierentially to conditions. Mean activity was signicantly smaller in OA only in B. L-VLPFC-C and in D. R-VLPFC-Sc. There was no diference in A. DLPFC-C, C. L-VLPFC-Sc and E. L/R Cb. VMPFC and MOFC, the left DLPFC and the left VLPFC. Dierences were also found in the left Temporal Occipital Fusiform Gyrus (visual value regions). 127 . Figure 2.8: Regions responding dierentially to conditions. Mean activity was signicantly smaller in OA in both a priori ROIs. 2.3.4 Connectivity analysis Using PPI analysis, we identied regions that were functionally con- nected with regions in the VMPFC among the OA population and we asked whether these associations were modulated by condition. We identied a single large cluster (peak activity at [16,16,36]) showing signicant connectivity in CONTROL. This cluster spanned through the right/left Anterior Cingulate Cortex, Supramarginal Gyrus and Frontal Pole reaching the right Middle and Inferior Frontal Gyri, 128 DLPFC and VLPFC (Fig. 2.9). This cluster overlapped with ROI R-VLPFC-Sc involved in condition tracking and showing diminished activity in SCALING (hence higher in CONTROL). OA: CONTROL CONNECTIVITY LEFT RIGHT X: +42 Z: +18 X: 0 RL-PaCG/SFG/ACC R-FP R-DLPFC/VLPFC MFG/IFG . Figure 2.9: Functional coupling with VMPFC among OA. We retained this cluster, hereafter PPI-OA, to conduct ROI analy- sis and we compared the distribution of the mean parameters associ- ated with condition regressors across both YA and OA participants (Fig. 2.10 (A)). We found that mean parameters were lower among OA on average in CONTROL and BUNDLING (t.tests, t = 2.11, df =67.42, p-value = 0.04 in CONTROL; t = 2.28, df =85.54, p-value = 0.025 in BUNDLING). We also found that parameter weights in PPI-OA were generally strongly correlated with parameter weights in 129 VMPFC in both populations. This was the case for YA in all condi- tions (Pearson's r =0.39, p=0.002 in CONTROL; Pearson's r =0.45, p=0.0005 in SCALING; Pearson's r =0.346, p=0.0002 in BUNDLING, Fig. 2.10 (B)). This was also the case for OA in both CONTROL and SCALING (Pearson's r =0.46, p=0.006 in CONTROL; Pearson's r =0.41, p=0.017 in SCALING) but no relationship was present in BUNDLING (Fig. 2.10 (C)). These results were consistent with the functional coupling found in OA in CONTROL. They also indicated that activity in this network of regions was also related in SCALING for OA participants and that similar patterns were present among YA. Last, we found that parameter weights in PPI-OA were strongly correlated with parameter weights in R-VLPFC-Sc also in both pop- ulations. This was the case for YA in all conditions (Pearson's r =0.58, p<0.001 in CONTROL; Pearson's r =0.59, p<0.001 in SCAL- ING; Pearson's r =0.51, p<0.001 in BUNDLING, Fig. 2.11 (A)). This was also the case for OA (Pearson's r =0.46, p=0.006 in CONTROL; Pearson's r =0.62, p<0.001 in SCALING, Pearson's r =0.39, p=0.021 in BUNDLING; (Fig. 2.11 (B)). 130 . Figure 2.10: Activity in region signicantly connected to the VMPFC in CON- TROL. . Figure 2.11: Relationship betweenPPI-OA andR-VLPFC-Sc. Mean parameters in the cluster signicantly functionally connected to the VMPFC in CONTROL were strongly correlated with mean parameters in core condition tracking region located in the right VLPFC. Connectivity analysis: summary The main message from this analysis was that regions in the right VLPFC showed similar patterns of activity in both populations. Also in both populations activity in these regions was correlated with ac- 131 tivity in the VMPFC in both CONTROL and SCALING. Model PPI* suggested a signicant functional coupling in CONTROL among OA. 2.3.5 Neural correlates of behavior YA core task-related regions We investigated the relationship between value-tracking and condition- tracking neural patterns and behavioral markers through regression analysis (Table 2.5 and Table 2.6 ), controlling for age and condition. The regressions inform us of the nature and signicance of the rela- tionship between BOLD responses to value and behavioral measures after removing direct eects of age and conditions. We found that mean parameters for value tracking regressors were negatively associated with reaction times in VMPFC, positively as- sociated with Consistency Rates in DLPFC and positively associated with Matching Rates in Visual Value regions. We also found that mean parameters for condition regressors were negatively associated with reaction times in ROIs in the DLPFC and VLPFC regions while they were positively related to consistency rates in L/R Cb (Table 2.6). Overall, patterns of brain activity were signicantly related to dierences in behavioral measures. 132 VMPFC VMPFC DLPFC-2 L-Visual R-Visual Mean RT -36.35 - - - - (12.75) - - - - CR - 153.44 127.38 - - - (79.42) (48.77) - - MR - - - 44.23 53.70 - - - (21.98) (18.06) Age -0.42 2.23 1.58 -0.20 0.03 (0.15) (1.29) (0.87) (0.19) (0.17) SCALING -8.38 -6.92 -4.77 1.83 3.06 (8.61) (8.57) (4.86) (7.76) (7.62) BUNDLING -9.92 -12.43 -0.11 9.28 5.75 (9.55) (9.04) (6.24) (8.69) (7.88) Constant 75.36 -101.02 -92.35 -9.76 -25.50 (23.11) (70.24) (42.59) (20.65) (16.62) Observations 282 282 282 282 282 Clusters 94 94 94 94 94 Adj. R 2 0.036 0.022 0.022 0.016 0.015 p< 0:1, p< 0:05, p< 0:01, p< 0:001 clustered standard errors at the subject level Table 2.5: Neural activity and behavioral markers in core value-tracking re- gions. Parameter weights in VMPFC were negatively associated with mean Reaction Times and positively associated with Consistency Rates (rst and second columns); pa- rameter weights in DLPFC were positively associated with Consistency Rates (third col- umn); parameter weights in both L-Visual and R-Visual were positively associated with mean Matching Rates (last two columns). 133 DLPFC-C L-VLPFC-C L-VLPFC-Sc R-VLPFC-Sc L/R Cb Mean RT -27.76 -75.95 -29.83 -49.68 - (9.44) (23.95) (13.53) (16.81) - CR - - - - 78.45 - - - - (41.84) Age 0.14 -0.59 -0.05 -0.38 0.10 (0.13) (0.25) (0.17) (0.26) (0.21) SCALING -6.45 -6.93 -9.00 -7.36 1.42 (1.63) (2.35) (1.94) (2.02) (2.21) BUNDLING -5.83 -5.99 0.12 -1.58 9.16 (1.43) (2.54) (1.71) (2.03) (2.17) Constant 91.71 235.42 108.36 166.11 -19.61 (18.01) (42.90) (24.24) (30.15) (40.21) Observations 282 282 282 282 282 Clusters 94 94 94 94 94 Adj. R 2 0.085 0.156 0.048 0.091 0.028 p< 0:1, p< 0:05, p< 0:01, p< 0:001 clustered standard errors at the subject level Table 2.6: Neural activity and behavioral markers in core regions tracking conditions. Mean activity in DLPFC-C, L-VLPFC-C, L-VLPFC-Sc and R-VLPFC-Sc was negatively associated with mean Reaction Times (rst four columns); Mean activity in L/R Cb was positively associated with Consistency Rates (last column). 134 Predictors of behavior We ran multinomial regressions to explain the relationship between age and patterns of activity within our ROIs. The multinomial logistic regression estimates a separate binary logistic regression model for each category, here each cohort. Each model conveys the eect of predictors on the probability of success in that category, in comparison to a reference category. We chose OA as reference. We conducted a similar analysis to determine what patterns of activity were predicting whether a participant belonged to the OA or YA cohorts. We found that increased mean activity in DLPFC and L-VLPFC-C predicted that the participant was young (Table 2.7). YA DLPFC 0.009 (0.002) L-VLPFC-C 0.008 (0.001) Constant -0.224 (0.137) Observations 282 Clusters 94 Akaike 1052 p< 0:1, p< 0:05, p< 0:01, p< 0:001 clustered standard errors at the subject level. Table 2.7: Predictors of Age groups: Higher activity in the left DLPFC and the left VLPFC were predicting a participant belonged to the YA cohort. 135 Given CR and MR were impacted by age, we asked whether activity in the regions that predicted age could predict CR and MR. We found that higher mean parameters in DLPFC and L-VLPFC-C predicted higher CR and MR (Table 2.8, second and third columns). CR MR DLPFC 0.0002 -0.0005 (0.0001) (0.0002) L-DLPFC-C 0.0002 - (0.0001) - Constant 0.85 0.67 (0.017) (0.015) Observations 282 282 Clusters 94 94 Adj. R 2 0.030 0.048 p< 0:1, p< 0:05, p< 0:01, p< 0:001 clustered standard errors at the subject level. Table 2.8: Predictors of behavioral markers. Behavioral markers and neural correlates: summary We reported two main ndings. First, activity patterns in YA core task-related regions were associated with behavioral markers. Some of these patterns were also modulated by age. Notably, higher activity in the left VMPFC and in the left DLPFC (two regions exhibiting lower BOLD responses among OA) were associated with higher CR. Second, the age of the participant was predicted by activity patterns 136 in the left DLPFC and the left VLPFC. Consistent with that result, both MR and CR were predicted by these patterns as well. Observing lower mean parameters tracking value and conditions in those regions indicated that the participant was likely old and making inconsistent choices. Taking these results together, reduction in behavioral perfor- mance were associated with disrupted patterns of activity in a network of regions involving the left VMPFC, left DLPFC and left VLPFC. 2.4 Discussion We have reported a comparison of value-based decision-making be- tween young and old adults. We have tested the hypothesis that age- related changes in the DLPFC and more generally the FPN network would be related to decreased quality in decision-making in tasks that involve those regions. We have not found qualitative dierences in core regions involved in value-tracking in our task, indicating that value-based computations take place in the same areas in both populations. Nevertheless, we have found that BOLD responses were signicantly lower among OA in many of these regions such as the VMPFC and the DLPFC. Some 137 regions, such as the VLPFC, involved in contextual representation of specic choices (single items, or bundles) were also aected by age. This suggests that computations are less eciently performed in the aging brain even though the same regions are recruited. We have also found that age-related dierences in BOLD responses predicted age-related dierences in behavioral measures. In particular, age-related changes in the DLPFC and the VLPFC were associated with a signicant decrease in consistency rates among OA. These results are consistent with the known decline of the DLPFC during the aging process (Raz et al., 2005; Resnick et al., 2003). Similar decline has been reported for the VLPFC in the context of routine problem solving (Hampshire et al., 2008) and food choices (Chung et al., 2017). We also found that dierent mechanisms operated to represent con- textual information regarding the trial faced. Most signicantly, activ- ity in the ACC was specic to the YA population while activity in the PCC was specic to the OA population. The cingulate cortex is a site where information from dierent sources is integrated and it comprises areas that mediate attention and the integration of cognitive processes. 138 Two of these regions (ACC and PCC) were dierentially activated in our task. Activation of the ACC has been observed in cognitive tasks requiring con ict monitoring (Botvinick et al., 2004) as well as action selection (Rushworth et al., 2004) and it is believed to to play a role when the relationship between actions and their consequences changes. In particular some regions in the ACC are activated when it is optimal to switch between response rules (Liston et al., 2006). Activation of the PCC has also been observed in a wide range of cognitive tasks, including the encoding of the subjective value (Kable and Glimcher, 2007) and strategic planning (Leech et al., 2011). Animal studies have shown that neurons in the PCC encode reward sizes and variance (Mc- Coy and Platt, 2005) and track the history of consequences, suggesting that PCC is responsible for representing the consequences of choices within a changing environment (Pearson et al., 2011). The PCC is also subject to deactivation during tasks that require attention and it is a key node in the default mode network (Buckner et al., 2008). It is therefore involved in the regulation of the focus of attention (Leech and Sharp, 2013). Interestingly, the cingulate cortex has often shown vulnerability to gray matter loss (Resnick et al., 2003; Salat et al., 139 2009) most signicantly in the ACC (Bergeld et al., 2010) while a relative structural preservation occurs in the PCC (Kalpouzos et al., 2009; Smith et al., 2007). These elements taken together help explain the asymmetry in activation patterns of the ACC and the PCC across populations. Overall, our results indicate that the quality of decision-making de- creases with age because several critical regions involved in the compu- tation of subjective value and the representation of relevant contextual information are aected by aging. Indeed, even in the case of simple choices, cognitive functions and attention-based mechanisms need to be recruited. As these functions decline, the way older adults represent their underlying preferences is critically impaired. 140 2.5 Appendix Appendix 1. Dierences between OA and YA resulting from FSL Randomize. CONTROL YA>OA Regions k MAX-score region MAX x y z RL ACC, PaCG 436 0.991 RL ACC 0 30 24 L FOC, MOFC 4 0.953 L FOC -18 14 -16 CONTROL OA>YA Regions k MAX-score region MAX x y z RL SMA, ACC, PCC, PrG, PoG, PCun, SPL, LOC 6082 0.996 R PrG 10 -20 46 SCALING OA>YA Regions k MAX-score region MAX x y z RL PoG, PrG, CO, MFG, SFG, DLPFC, SPL, PCun, Cun, 22389 1 L PoG -40 -18 46 LingG, PCC, ACC, SMA, LOC, OP, SMG, th, brainstem/midbrain, Cb, Hi, PaHG, Ins, ; R PTe, PO, FO, HschGy, IFG, pa; L Pu, PPo, LOC, OP L PTe, SMG, PO 119 0.962 L PTe -50 -42 18 L SMG, PO, PTe 110 0.965 L SMG -64 -32 22 L FP, LOFC, FOC 84 0.97 L FP -38 40 -8 L PrG 60 0.961 L PrG -64 0 8 L PCun, SupCalcC, IntCalC 38 0.953 L PCun -16 -62 12 R pa 14 0.954 R pa 20 -12 -8 BUNDLING YA>OA Regions k MAX-score region MAX x y z L ACC, PaCG 213 0.986 L ACC -2 32 22 RL ACC 25 0.968 L ACC 0 26 6 R Cd 22 0.96 R Cd 14 18 4 BUNDLING OA>YA Regions k MAX-score region MAX x y z RL PrG, PoG, PCC, ACC, SMA, PCun, SPL 9864 1 L PrG, PoG -32 -24 62 R IFG, CO, PPo, HschGy, PO, PTe, SMG, Ang SCALING VALUE OA>YA Regions k MAX-score region MAX x y z L LingG 1 0.951 L LingG -14 -62 -8 Table 2.9: Dierence in neural correlates between OA and YA. 141 Appendix 2. PPI analysis of OA using a priori VMPFC as seed. VMPFC (CONTROL CONNECTIVITY) MNI 152 Regions k z-score x y z RL PaCG, ACC, SFG, FP; R MFG, IFG, DLPFC, VLPFC 1174 3.66 16 16 36 Table 2.10: PPI OA 142 Bibliography Andreoni, J. and Miller, J. (2002). Giving according to garp: An ex- perimental test of the consistency of preferences for altruism. Econo- metrica, 70(2):737{753. Arana, F. S., Parkinson, J. A., Hinton, E., Holland, A. J., Owen, A. M., and Roberts, A. C. (2003). Dissociable contributions of the human amygdala and orbitofrontal cortex to incentive motivation and goal selection. Journal of Neuroscience, 23(29):9632{9638. Aron, A. R., Robbins, T. W., and Poldrack, R. A. (2004). Inhibition and the right inferior frontal cortex. Trends in cognitive sciences, 8(4):170{177. Baillieux, H., De Smet, H. J., Dobbeleir, A., Paquier, P. F., De Deyn, P. P., and Mari en, P. (2010). Cognitive and aective disturbances following focal cerebellar damage in adults: a neuropsychological and spect study. Cortex, 46(7):869{879. Baker, S., Frith, C., Frackowiak, S., and Dolan, R. J. (1996). Active representation of shape and spatial location in man. Cerebral Cortex, 6(4):612{619. Battalio, R. C., Kagel, J. H., Winkler, R. C., Fisher Jr, E. B., Bas- mann, R. L., and Krasner, L. (1973). A test of consumer demand theory using observations of individual consumer purchases. Eco- nomic Inquiry, 11(4):411{428. 143 Baumgartner, T., Knoch, D., Hotz, P., Eisenegger, C., and Fehr, E. (2011). Dorsolateral and ventromedial prefrontal cortex orchestrate normative choice. Nature neuroscience, 14(11):1468. Bergeld, K. L., Hanson, K. D., Chen, K., Teipel, S. J., Hampel, H., Rapoport, S. I., Moeller, J. R., and Alexander, G. E. (2010). Age-related networks of regional covariance in mri gray matter: reproducible multivariate patterns in healthy aging. Neuroimage, 49(2):1750{1759. Berman, K. F., Ostrem, J. L., Randolph, C., Gold, J., Goldberg, T. E., Coppola, R., Carson, R. E., Herscovitch, P., and Weinberger, D. R. (1995). Physiological activation of a cortical network during performance of the wisconsin card sorting test: a positron emission tomography study. Neuropsychologia, 33(8):1027{1046. Botvinick, M. M., Cohen, J. D., and Carter, C. S. (2004). Con ict monitoring and anterior cingulate cortex: an update. Trends in cognitive sciences, 8(12):539{546. Braver, T. S., Cohen, J. D., Nystrom, L. E., Jonides, J., Smith, E. E., and Noll, D. C. (1997). A parametric study of prefrontal cortex involvement in human working memory. Neuroimage, 5(1):49{62. Brocas, I., Carrillo, J. D., Combs, T. D., and Kodaverdian, N. (2019a). Consistency in simple vs. complex choices by younger and older adults. Journal of Economic Behavior & Organization, 157:580{601. Brocas, I., Carrillo, J. D., Combs, T. D., and Kodaverdian, N. (2019b). The development of consistent decision-making across economic do- mains. Games and Economic Behavior, 116:217{240. Buckner, R. L., Andrews-Hanna, J. R., and Schacter, D. L. (2008). The brain's default network: anatomy, function, and relevance to disease. 144 Camille, N., Griths, C. A., Vo, K., Fellows, L. K., and Kable, J. W. (2011). Ventromedial frontal lobe damage disrupts value maximiza- tion in humans. Journal of Neuroscience, 31(20):7527{7532. Camus, M., Halelamien, N., Plassmann, H., Shimojo, S., O'Doherty, J., Camerer, C., and Rangel, A. (2009). Repetitive transcranial magnetic stimulation over the right dorsolateral prefrontal cortex decreases valuations during food choices. European Journal of Neu- roscience, 30(10):1980{1988. Cardoso, C. d. O., Branco, L. D., Cotrena, C., Kristensen, C. H., Schneider Bakos, D. D. G., and Fonseca, R. P. (2014). The impact of frontal and cerebellar lesions on decision making: evidence from the iowa gambling task. Frontiers in neuroscience, 8:61. Carlson, S., Martinkauppi, S., R am a, P., Salli, E., Korvenoja, A., and Aronen, H. J. (1998). Distribution of cortical activation during visu- ospatial n-back tasks as revealed by functional magnetic resonance imaging. Cerebral cortex (New York, NY: 1991), 8(8):743{752. Chib, V. S., Rangel, A., Shimojo, S., and O'Doherty, J. P. (2009). Ev- idence for a common representation of decision values for dissimilar goods in human ventromedial prefrontal cortex. Journal of Neuro- science, 29(39):12315{12320. Choi, S., Fisman, R., Gale, D., and Kariv, S. (2007a). Consistency and heterogeneity of individual behavior under uncertainty. American economic review, 97(5):1921{1938. Choi, S., Fisman, R., Gale, D. M., and Kariv, S. (2007b). Revealing preferences graphically: an old method gets a new tool kit. American Economic Review, 97(2):153{158. Christo, K., Prabhakaran, V., Dorfman, J., Zhao, Z., Kroger, J. K., Holyoak, K. J., and Gabrieli, J. D. (2001). Rostrolateral prefrontal 145 cortex involvement in relational integration during reasoning. Neu- roimage, 14(5):1136{1149. Chung, H.-K., Tymula, A., and Glimcher, P. (2017). The reduc- tion of ventrolateral prefrontal cortex gray matter volume correlates with loss of economic rationality in aging. Journal of Neuroscience, 37(49):12068{12077. Clewett, D., Luo, S., Hsu, E., Ainslie, G., Mather, M., and Mon- terosso, J. (2014a). Increased functional coupling between the left fronto-parietal network and anterior insula predicts steeper delay discounting in smokers. Human brain mapping, 35(8):3774{3787. Clewett, D., Schoeke, A., and Mather, M. (2014b). Locus coeruleus neuromodulation of memories encoded during negative or unex- pected action outcomes. Neurobiology of Learning and Memory, 111:65{70. Clithero, J. A. and Rangel, A. (2013a). Combining response times and choice data using a neuroeconomic model of the decision process improves out-of-sample predictions. unpublished, California Institute of Technology. Clithero, J. A. and Rangel, A. (2013b). Informatic parcellation of the network involved in the computation of subjective value. Social cognitive and aective neuroscience, 9(9):1289{1302. Cohen, J. D., Forman, S. D., Braver, T. S., Casey, B., Servan- Schreiber, D., and Noll, D. C. (1994). Activation of the prefrontal cortex in a nonspatial working memory task with functional mri. Human brain mapping, 1(4):293{304. Cohen, J. D., Perlstein, W. M., Braver, T. S., Nystrom, L. E., Noll, D. C., Jonides, J., and Smith, E. E. (1997). Temporal dynam- 146 ics of brain activation during a working memory task. Nature, 386(6625):604. Cox, J. C. (1997). On testing the utility hypothesis. The Economic Journal, 107(443):1054{1078. Dale, A. M. (1999). Optimal experimental design for event-related fmri. Human brain mapping, 8(2-3):109{114. de Berker, A. O., Kurth-Nelson, Z., Rutledge, R. B., Bestmann, S., and Dolan, R. J. (2019). Computing value from quality and quantity in human decision-making. Journal of Neuroscience, 39(1):163{176. De Martino, B., Fleming, S. M., Garrett, N., and Dolan, R. J. (2013a). Condence in value-based choice. Nature neuroscience, 16(1):105. De Martino, B., O'Doherty, J. P., Ray, D., Bossaerts, P., and Camerer, C. (2013b). In the mind of the market: theory of mind biases value computation during nancial bubbles. Neuron, 79(6):1222{1231. de Wit, S., Corlett, P. R., Aitken, M. R., Dickinson, A., and Fletcher, P. C. (2009). Dierential engagement of the ventromedial prefrontal cortex by goal-directed and habitual behavior toward food pictures in humans. Journal of Neuroscience, 29(36):11330{11338. Deichmann, R., Gottfried, J. A., Hutton, C., and Turner, R. (2003). Optimized epi for fmri studies of the orbitofrontal cortex. Neuroim- age, 19(2):430{441. Demb, J. B., Desmond, J. E., Wagner, A. D., Vaidya, C. J., Glover, G. H., and Gabrieli, J. (1995). Semantic encoding and retrieval in the left inferior prefrontal cortex: a functional mri study of task diculty and process specicity. Journal of Neuroscience, 15(9):5870{5878. Deverett, B., Kislin, M., Tank, D. W., Samuel, S., and Wang, H. 147 (2019). Cerebellar disruption impairs working memory during evi- dence accumulation. bioRxiv, page 521849. Deverett, B., Koay, S. A., Oostland, M., and Wang, S. S. (2018). Cere- bellar involvement in an evidence-accumulation decision-making task. Elife, 7:e36781. Dixon, M. L., Thiruchselvam, R., Todd, R., and Christo, K. (2017). Emotion and the prefrontal cortex: An integrative review. Psycho- logical bulletin, 143(10):1033. Domenech, P., Redout e, J., Koechlin, E., and Dreher, J.-C. (2017). The neuro-computational architecture of value-based selection in the human brain. Cerebral Cortex, 28(2):585{601. Dove, A., Pollmann, S., Schubert, T., Wiggins, C. J., and Von Cra- mon, D. Y. (2000). Prefrontal cortex activation in task switching: an event-related fmri study. Cognitive brain research, 9(1):103{109. Eckert, M. A., Menon, V., Walczak, A., Ahlstrom, J., Denslow, S., Horwitz, A., and Dubno, J. R. (2009). At the heart of the ventral attention system: the right anterior insula. Human brain mapping, 30(8):2530{2541. Fellows, L. K. and Farah, M. J. (2007). The role of ventromedial prefrontal cortex in decision making: judgment under uncertainty or judgment per se? Cerebral cortex, 17(11):2669{2674. Finucane, M. L., Mertz, C., Slovic, P., and Schmidt, E. S. (2005). Task complexity and older adults' decision-making competence. Psychol- ogy and aging, 20(1):71. Finucane, M. L., Slovic, P., Hibbard, J. H., Peters, E., Mertz, C., and MacGregor, D. G. (2002). Aging and decision-making compe- tence: An analysis of comprehension and consistency skills in older 148 versus younger adults considering health-plan options. Journal of Behavioral Decision Making, 15(2):141{164. FitzGerald, T. H., Seymour, B., and Dolan, R. J. (2009). The role of human orbitofrontal cortex in value comparison for incommensu- rable objects. Journal of Neuroscience, 29(26):8388{8395. Fjell, A. M., Westlye, L. T., Amlien, I., Espeseth, T., Reinvang, I., Raz, N., Agartz, I., Salat, D. H., Greve, D. N., Fischl, B., et al. (2009). High consistency of regional cortical thinning in aging across multiple samples. Cerebral cortex, 19(9):2001{2012. Fregni, F., Orsati, F., Pedrosa, W., Fecteau, S., Tome, F. A., Nitsche, M. A., Mecca, T., Macedo, E. C., Pascual-Leone, A., and Boggio, P. S. (2008). Transcranial direct current stimulation of the prefrontal cortex modulates the desire for specic foods. Appetite, 51(1):34{41. Fujiwara, J., Usui, N., Eifuku, S., Iijima, T., Taira, M., Tsutsui, K.-I., and Tobler, P. N. (2018). Ventrolateral prefrontal cortex updates chosen value according to choice set size. Journal of cognitive neu- roscience, 30(3):307{318. Goel, V., Gold, B., Kapur, S., and Houle, S. (1997). The seats of reason? an imaging study of deductive and inductive reasoning. NeuroReport, 8(5):1305{1310. Goldberg, T. E., Berman, K. F., Fleming, K., Ostrem, J., Van Horn, J. D., Esposito, G., Mattay, V. S., Gold, J. M., and Weinberger, D. R. (1998). Uncoupling cognitive workload and prefrontal cortical physiology: a pet rcbf study. Neuroimage, 7(4):296{303. Gottwald, B., Wilde, B., Mihajlovic, Z., and Mehdorn, H. (2004). Ev- idence for distinct cognitive decits after focal cerebellar lesions. Journal of Neurology, Neurosurgery & Psychiatry, 75(11):1524{ 1531. 149 Grady, C. L., Springer, M. V., Hongwanishkul, D., McIntosh, A. R., and Winocur, G. (2006). Age-related changes in brain activity across the adult lifespan. Journal of cognitive neuroscience, 18(2):227{241. Grimaldi, G. and Manto, M. (2012). Topography of cerebellar decits in humans. The Cerebellum, 11(2):336{351. Hampshire, A., Gruszka, A., Fallon, S. J., and Owen, A. M. (2008). In- eciency in self-organized attentional switching in the normal aging population is associated with decreased activity in the ventrolateral prefrontal cortex. Journal of Cognitive Neuroscience, 20(9):1670{ 1686. Harbaugh, W. T., Krause, K., and Berry, T. R. (2001). Garp for kids: On the development of rational choice behavior. American Economic Review, 91(5):1539{1545. Hare, T. A., Camerer, C. F., and Rangel, A. (2009). Self-control in decision-making involves modulation of the vmpfc valuation system. Science, 324(5927):646{648. Hare, T. A., Malmaud, J., and Rangel, A. (2011). Focusing attention on the health aspects of foods changes value signals in vmpfc and improves dietary choice. Journal of Neuroscience, 31(30):11077{ 11087. Hare, T. A., O'Doherty, J., Camerer, C. F., Schultz, W., and Rangel, A. (2008). Dissociating the role of the orbitofrontal cortex and the striatum in the computation of goal values and prediction errors. Journal of neuroscience, 28(22):5623{5630. Harris, A., Hare, T., and Rangel, A. (2013). Temporally dissocia- ble mechanisms of self-control: early attentional ltering versus late value modulation. Journal of Neuroscience, 33(48):18917{18931. 150 Henri-Bhargava, A., Simioni, A., and Fellows, L. K. (2012). Ventro- medial frontal lobe damage disrupts the accuracy, but not the speed, of value-based preference judgments. Neuropsychologia, 50(7):1536{ 1542. Hunt, L. T., Woolrich, M. W., Rushworth, M. F., and Behrens, T. E. (2013). Trial-type dependent frames of reference for value compari- son. PLoS computational biology, 9(9):e1003225. Hutcherson, C. A., Plassmann, H., Gross, J. J., and Rangel, A. (2012). Cognitive regulation during decision making shifts behavioral con- trol between ventromedial and dorsolateral prefrontal value systems. Journal of Neuroscience, 32(39):13543{13554. Jenkinson, M., Beckmann, C. F., Behrens, T. E., Woolrich, M. W., and Smith, S. M. (2012). Fsl. Neuroimage, 62(2):782{790. Jocham, G., Furlong, P. M., Kr oger, I. L., Kahn, M. C., Hunt, L. T., and Behrens, T. E. (2014). Dissociable contributions of ventrome- dial prefrontal and posterior parietal cortex to value-guided choice. Neuroimage, 100:498{506. Kable, J. W. and Glimcher, P. W. (2007). The neural correlates of subjective value during intertemporal choice. Nature neuroscience, 10(12):1625. Kahnt, T., Heinzle, J., Park, S. Q., and Haynes, J.-D. (2011). De- coding dierent roles for vmpfc and dlpfc in multi-attribute decision making. Neuroimage, 56(2):709{715. Kalpouzos, G., Ch etelat, G., Baron, J.-C., Landeau, B., Mevel, K., Godeau, C., Barr e, L., Constans, J.-M., Viader, F., Eustache, F., et al. (2009). Voxel-based mapping of brain gray matter volume and glucose metabolism proles in normal aging. Neurobiology of aging, 30(1):112{124. 151 Karatekin, C., Lazare, J. A., and Asarnow, R. F. (2000). Relevance of the cerebellar hemispheres for executive functions. Pediatric neu- rology, 22(2):106{112. Kim, H., Shimojo, S., and O'Doherty, J. P. (2010). Overlapping re- sponses for the expectation of juice and money rewards in human ventromedial prefrontal cortex. Cerebral cortex, 21(4):769{776. Konovalov, A. and Krajbich, I. (2019). Over a decade of neuroeco- nomics: what have we learned? Organizational Research Methods, 22(1):148{173. Kriegeskorte, N., Simmons, W. K., Bellgowan, P. S., and Baker, C. I. (2009). Circular analysis in systems neuroscience: the dangers of double dipping. Nature neuroscience, 12(5):535. Kroger, J. K., Sabb, F. W., Fales, C. L., Bookheimer, S. Y., Cohen, M. S., and Holyoak, K. J. (2002). Recruitment of anterior dorso- lateral prefrontal cortex in human reasoning: a parametric study of relational complexity. Cerebral cortex, 12(5):477{485. Leech, R., Kamourieh, S., Beckmann, C. F., and Sharp, D. J. (2011). Fractionating the default mode network: distinct contributions of the ventral and dorsal posterior cingulate cortex to cognitive control. Journal of Neuroscience, 31(9):3217{3224. Leech, R. and Sharp, D. J. (2013). The role of the posterior cingulate cortex in cognition and disease. Brain, 137(1):12{32. Levy, D. J. and Glimcher, P. W. (2011). Comparing apples and oranges: using reward-specic and reward-general subjective value representation in the brain. Journal of Neuroscience, 31(41):14693{ 14707. Lim, S.-L., O'Doherty, J. P., and Rangel, A. (2011). The decision value computations in the vmpfc and striatum use a relative value 152 code that is guided by visual attention. Journal of Neuroscience, 31(37):13214{13223. Lim, S.-L., O'Doherty, J. P., and Rangel, A. (2013). Stimulus value signals in ventromedial pfc re ect the integration of attribute value signals computed in fusiform gyrus and posterior superior temporal gyrus. Journal of Neuroscience, 33(20):8729{8741. Liston, C., Matalon, S., Hare, T. A., Davidson, M. C., and Casey, B. (2006). Anterior cingulate and posterior parietal cortices are sen- sitive to dissociable forms of con ict in a task-switching paradigm. Neuron, 50(4):643{653. Luo, S., Ainslie, G., Pollini, D., Giragosian, L., and Monterosso, J. R. (2012). Moderators of the association between brain activation and farsighted choice. Neuroimage, 59(2):1469{1477. MacPherson, S. E., Phillips, L. H., and Della Sala, S. (2002). Age, executive function and social decision making: a dorsolateral pre- frontal theory of cognitive aging. Psychology and aging, 17(4):598. Manes, F., Villamil, A. R., Ameriso, S., Roca, M., and Torralva, T. (2009). \real life" executive decits in patients with focal vascular lesions aecting the cerebellum. Journal of the neurological sciences, 283(1-2):95{98. Mather, M. (2012). The emotion paradox in the aging brain. Annals of the New York Academy of Sciences, 1251(1):33. McClure, S. M., Li, J., Tomlin, D., Cypert, K. S., Montague, L. M., and Montague, P. R. (2004). Neural correlates of behavioral prefer- ence for culturally familiar drinks. Neuron, 44(2):379{387. McCoy, A. N. and Platt, M. L. (2005). Risk-sensitive neurons in macaque posterior cingulate cortex. Nature neuroscience, 8(9):1220. 153 McFadden, B. R., Lusk, J. L., Crespi, J. M., Cherry, J. B. C., Mar- tin, L. E., Aupperle, R. L., and Bruce, A. S. (2015). Can neural activation in dorsolateral prefrontal cortex predict responsiveness to information? an application to egg production systems and cam- paign advertising. PloS one, 10(5):e0125243. McFadden, D. et al. (1973). Conditional logit analysis of qualitative choice behavior. McLaren, D. G., Ries, M. L., Xu, G., and Johnson, S. C. (2012). A generalized form of context-dependent psychophysiological interac- tions (gppi): a comparison to standard approaches. Neuroimage, 61(4):1277{1286. Monterosso, J. R., Ainslie, G., Xu, J., Cordova, X., Domier, C. P., and London, E. D. (2007). Frontoparietal cortical activity of methamphetamine-dependent and comparison subjects performing a delay discounting task. Human brain mapping, 28(5):383{393. Nichelli, P., Grafman, J., Pietrini, P., Alway, D., Carton, J. C., and Miletich, R. (1994). Brain activity in chess playing. Nature. O'Doherty, J. P., Buchanan, T. W., Seymour, B., and Dolan, R. J. (2006). Predictive neural coding of reward preference involves dis- sociable responses in human ventral midbrain and ventral striatum. Neuron, 49(1):157{166. O'Reilly, J. X., Woolrich, M. W., Behrens, T. E., Smith, S. M., and Johansen-Berg, H. (2012). Tools of the trade: psychophysiological interactions and functional connectivity. Social cognitive and aec- tive neuroscience, 7(5):604{609. Osherson, D., Perani, D., Cappa, S., Schnur, T., Grassi, F., and Fazio, F. (1998). Distinct brain loci in deductive versus probabilistic rea- soning. Neuropsychologia, 36(4):369{376. 154 Pearson, J. M., Heilbronner, S. R., Barack, D. L., Hayden, B. Y., and Platt, M. L. (2011). Posterior cingulate cortex: adapting behavior to a changing world. Trends in cognitive sciences, 15(4):143{151. Peters, E., Hess, T. M., V astfj all, D., and Auman, C. (2007). Adult age dierences in dual information processes: Implications for the role of aective and deliberative processes in older adults' decision making. Perspectives on Psychological Science, 2(1):1{23. Petrides, M. (1994). Frontal lobes and behaviour. Current opinion in neurobiology, 4(2):207{211. Plassmann, H., O'Doherty, J., and Rangel, A. (2007). Orbitofrontal cortex encodes willingness to pay in everyday economic transactions. Journal of neuroscience, 27(37):9984{9988. Rangel, A., Camerer, C., and Montague, P. R. (2008). A framework for studying the neurobiology of value-based decision making. Nature reviews neuroscience, 9(7):545. Raz, N., Lindenberger, U., Rodrigue, K. M., Kennedy, K. M., Head, D., Williamson, A., Dahle, C., Gerstorf, D., and Acker, J. D. (2005). Regional brain changes in aging healthy adults: general trends, indi- vidual dierences and modiers. Cerebral cortex, 15(11):1676{1689. Resnick, S. M., Pham, D. L., Kraut, M. A., Zonderman, A. B., and Davatzikos, C. (2003). Longitudinal magnetic resonance imaging studies of older adults: a shrinking brain. Journal of Neuroscience, 23(8):3295{3301. Rosenbloom, M. H., Schmahmann, J. D., and Price, B. H. (2012). The functional neuroanatomy of decision-making. The Journal of neuropsychiatry and clinical neurosciences, 24(3):266{277. Rudorf, S. and Hare, T. A. (2014). Interactions between dorsolat- eral and ventromedial prefrontal cortex underlie context-dependent 155 stimulus valuation in goal-directed choice. Journal of Neuroscience, 34(48):15988{15996. Rushworth, M., Walton, M. E., Kennerley, S. W., and Bannerman, D. (2004). Action sets and decisions in the medial frontal cortex. Trends in cognitive sciences, 8(9):410{417. Rypma, B., Prabhakaran, V., Desmond, J. E., and Gabrieli, J. D. (2001). Age dierences in prefrontal cortical activity in working memory. Psychology and aging, 16(3):371. Salat, D. H., Lee, S. Y., Van der Kouwe, A., Greve, D. N., Fischl, B., and Rosas, H. D. (2009). Age-associated alterations in cortical gray and white matter signal intensity and gray to white matter contrast. Neuroimage, 48(1):21{28. Seeley, W. W., Menon, V., Schatzberg, A. F., Keller, J., Glover, G. H., Kenna, H., Reiss, A. L., and Greicius, M. D. (2007). Dissociable intrinsic connectivity networks for salience processing and executive control. Journal of Neuroscience, 27(9):2349{2356. Serences, J. T. (2008). Value-based modulations in human visual cor- tex. Neuron, 60(6):1169{1181. Serences, J. T. and Saproo, S. (2010). Population response proles in early visual cortex are biased in favor of more valuable stimuli. Journal of neurophysiology, 104(1):76{87. Sippel, R. (1997). An experiment on the pure theory of consumer's behaviour. The Economic Journal, 107(444):1431{1444. Smith, C. D., Chebrolu, H., Wekstein, D. R., Schmitt, F. A., and Markesbery, W. R. (2007). Age and gender eects on human brain anatomy: a voxel-based morphometric study in healthy elderly. Neu- robiology of aging, 28(7):1075{1087. 156 Smith, S. M., Fox, P. T., Miller, K. L., Glahn, D. C., Fox, P. M., Mackay, C. E., Filippini, N., Watkins, K. E., Toro, R., Laird, A. R., et al. (2009). Correspondence of the brain's functional architecture during activation and rest. Proceedings of the National Academy of Sciences, 106(31):13040{13045. Sokol-Hessner, P., Hutcherson, C., Hare, T., and Rangel, A. (2012). Decision value computation in dlpfc and vmpfc adjusts to the avail- able decision time. European Journal of Neuroscience, 35(7):1065{ 1074. Train, K. E. (2009). Discrete choice methods with simulation. Cam- bridge university press. Tregellas, J. R., Davalos, D. B., and Rojas, D. C. (2006). Eect of task diculty on the functional anatomy of temporal processing. Neuroimage, 32(1):307{315. Uddin, L. Q. (2015). Salience processing and insular cortical function and dysfunction. Nature Reviews Neuroscience, 16(1):55{61. Uddin, L. Q., Nomi, J. S., H ebert-Seropian, B., Ghaziri, J., and Boucher, O. (2017). Structure and function of the human insula. Journal of clinical neurophysiology: ocial publication of the Amer- ican Electroencephalographic Society, 34(4):300. Winkler, A. M., Ridgway, G. R., Webster, M. A., Smith, S. M., and Nichols, T. E. (2014). Permutation inference for the general linear model. Neuroimage, 92:381{397. 157

Abstract (if available)

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Neuroeconomic mechanisms for valuing complex options

PDF

The evolution of decision-making quality over the life cycle: evidence from behavioral and neuroeconomic experiments with different age groups

PDF

Homeostatic imbalance and monetary delay discounting: effects of feeding on RT, choice, and brain response

PDF

Behabioral and neural evidence of state-like variance in intertemporal decisions

PDF

Sequential decisions on time preference: evidence for non-independence

PDF

Insula activity during safe-sex decision-making in sexually risky men suggests negative urgency and fear of rejection drives risky sexual behavior

PDF

The acute impact of glucose and sucralose on food decisions and brain responses to visual food cues

PDF

Dedifferentiation of emotion regulation strategies in the aging brain: an MVPA investigation

PDF

Brain and behavior correlates of intrinsic motivation and skill learning

PDF

Deciphering the variability of goal-directed and habitual decision-making

PDF

The effect of present moment awareness and value intervention of ACT on impulsive decision-making and impulsive disinhibition

PDF

Functional models of fMRI BOLD signal in the visual cortex

PDF

Decoding information about human-agent negotiations from brain patterns

PDF

Behavioral and neural evidence of incentive bias for immediate rewards relative to preference-matched delayed rewards

PDF

Affective neuropsychiatric symptoms and neural connectivity in the early stages of Alzheimer’s disease

PDF

Functional magnetic resonance imaging characterization of peripheral form vision

PDF

Spatial and temporal patterns of brain activity associated with emotions in music

PDF

Value based purchasing: decision-making processes underlying hospital acquisitions of orthopedic devices

PDF

Heart, brain, and breath: studies on the neuromodulation of interoceptive systems

PDF

Age differences in diffusivity in the locus coeruleus and its ascending noradrenergic tract

Asset Metadata

Creator Zyuzin, Jekaterina (author)

Core Title Value-based decision-making in complex choice: brain regions involved and implications of age

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Neuroscience

Publication Date 03/30/2020

Defense Date 12/03/2019

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

Tag Age,complex choice,fMRI,OAI-PMH Harvest,value-based decision-making

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Monterosso, John (committee chair), Brocas, Isabelle (committee member), Coricelli, Giorgio (committee member), Page, Kathleen Alanna (committee member)

Creator Email kot.jushkina@gmail.com,zyuzin@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c89-277555

Unique identifier UC11673370

Identifier etd-ZyuzinJeka-8234.pdf (filename),usctheses-c89-277555 (legacy record id)

Legacy Identifier etd-ZyuzinJeka-8234.pdf

Dmrecord 277555

Document Type Dissertation

Rights Zyuzin, Jekaterina

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