Differential functional connectivity underlying asymmetric reward-related activity in human and non-human primates

The orbitofrontal cortex (OFC) is a key brain region involved in complex cognitive functions such as reward processing and decision-making. Neuroimaging studies have shown unilateral OFC response to reward-related variables, however, those studies rarely discussed the lateralization of this effect. Yet, some lesion studies suggest that the left and right OFC contribute differently to cognitive processes. We hypothesized that the OFC asymmetrical response to reward could reflect underlying hemispherical difference in OFC functional connectivity. Using restingstate and reward-related MRI data from humans and from rhesus macaques, we first identified a specific asymmetrical response of the lateral OFC to reward in both species. Crucially, the subregion showing the highest reward-related asymmetry (RRA) overlapped with the region showing the highest functional connectivity asymmetry (FCA). Furthermore, the two types of functional asymmetries were found to be significantly correlated across humans. Altogether, our results suggest a similar pattern of functional specialization between the left and right OFC is present in two primate species.

The orbitofrontal cortex (OFC) is a key brain region involved in complex cognitive 22 functions such as reward processing and decision-making. Neuroimaging studies 23 have shown unilateral OFC response to reward-related variables, however, those 24 studies rarely discussed the lateralization of this effect. Yet, some lesion studies 25 suggest that the left and right OFC contribute differently to cognitive processes. We 26 hypothesized that the OFC asymmetrical response to reward could reflect 27 underlying hemispherical difference in OFC functional connectivity. Using resting-28 state and reward-related MRI data from humans and from rhesus macaques, we first 29 identified a specific asymmetrical response of the lateral OFC to reward in both 30 species. Crucially, the subregion showing the highest reward-related asymmetry 31 (RRA) overlapped with the region showing the highest functional connectivity 32 asymmetry (FCA). Furthermore, the two types of functional asymmetries were 33 found to be significantly correlated across humans. Altogether, our results suggest a 34 similar pattern of functional specialization between the left and right OFC is present 35 in two primate species. 36 37 Introduction 38 39 The orbitofrontal cortex (OFC) is a key brain region involved in complex behavior 40 such as value-based decision-making (1), cognitive flexibility (2) and state space 41 representation (3). This brain region is heterogenous and can be subdivided on the 42 basis of cytoarchitecture, connectivity, or function (4-8). The large majority of 43 studies investigating the functional organization of the OFC consider it to be 44 symmetrically organized between hemispheres (1,(9)(10)(11)(12). Some unilateral lesion and 45 stimulation studies have nevertheless shown differential behavioral effects. For 46 instance, direct intracortical stimulation in humans showed a left lateralization of 47 negative experience compared to neutral experience (13). Patients with right OFC 48 lesions were more impaired in the Iowa Gambling Task than those with left lesions 49 (14). Asymmetrical OFC responses in healthy subjects have also been reported in 50 fMRI studies (for meta-analyses, see (15,16)). However, this result has rarely been 51 discussed. 52 Lateralization of functions in the prefrontal cortex has been shown previously, in 53 particular for language processing (17), visuo-spatial attention (18), but also for 54 relational integration reasoning (15). In humans, reductions in asymmetry have been 55 associated with impaired cognitive functions (19) and hemispheric specialization is 56 suggested to increase processing abilities by reducing bilateral redundancy (20) 57 indicating that there may be some benefit when homotypical areas in each 58 hemisphere specialize. Lateralization of functions has also been reported in non-59 human primates in the context of audition and vocalization (21-24), or attention (25). 60 Yet, lateralization in other contexts, such as reward processing, has not received 61 much attention in any species. 62 Using data from the Human Connectome Project, and data collected in rhesus 63 macaques (Macaca mulatta), we assessed the nature of the asymmetrical OFC 64 response during reward tasks. First, we identified an asymmetrical response to 65 reward in a specific area of the OFC in both species. Second, we observed that the 66 connectivity of the OFC with the rest of the brain was significantly different between 67 hemispheres. Interestingly, the brain region responding differentially in the reward 68 task was the same as the brain region showing asymmetrical whole-brain 69 connectivity. Moreover, the two types of functional asymmetry were correlated 70 across individuals. Together, our results suggest that the left and right OFC might 71 support different functions -that remain to be characterized, due to an intrinsic 72 difference in their connectivity to the rest of the brain. Humans 80 We selected 57 subjects from the Human Connectome Project for which rs-fMRI had 81 been obtained at 7T and who participated in a gambling task designed to assess 82 reward processing and decision-making (26). Participants had to guess whether a 83 hidden card was higher or lower than a visible card. They received positive, neutral 84 or negative monetary feedback according to the correctness of the response (see 85 Methods). In the fMRI data, we focused on the contrast 'Reward versus Punishment' 86 to localize the reward-related activity in the whole brain ( Figure 1A). Replicating 87 previous results from a larger dataset (26), this contrast also revealed higher activity 88 for reward compared to punishment in the ventromedial prefrontal cortex (vmPFC) 89 and in the ventral striatum. Interestingly, a significant cluster was found in the right 90 OFC, but not in the left OFC (cluster-corrected, cluster size > 150 voxels). Note that 91 the uncorrected map did not reveal a response in the left OFC either ( Figure 1A). 92 To assess whether this hemispherical difference was significant, we mapped the 93 individual z-maps onto the individual MSMAll surfaces, that are registered on the 94 symmetric MNI 152 template (27). We mirrored the data of the left hemisphere so 95 they could be compared to the data on the right hemisphere. We computed the 96 unsigned left versus right difference in the contrast 'Reward versus Punishment' for 97 every subject and tested for significant effect at the group level in a large Orbital and 98 Medial Prefrontal Cortex (OMPFC) mask (see Methods). We found a significant 99 difference between left and right OMPFC for reward-related activity in the OFC 100 (pcorr=0.012) ( Figure 1B). This result reveals asymmetric reward-related activity at the 101 intersection of the lateral orbitofrontal sulcus (LOS) and transverse orbitofrontal 102 sulcus (TOS). 103 Figure 1 -Neural responses to reward and hemispheric differences in reward responses in humans and macaques. A. Statistical maps relating to the contrast 'reward versus punishment' in humans. Clusters in yellow show significant positive effect (FWE corrected, p<0.05). Clusters in dark red indicate uncorrected effect at p<0.001. B. Unsigned difference between the sizes of the effects illustrated in A in the left and in the right hemispheres. Color code indicates z-statistics at the group level, the map is restricted to the OMPFC and cluster corrected (cluster-level p<0.05, permutation tests). C. Average of the individual session statistical maps relating to various reward contrasts in macaques (see Methods). Because of the large difference in the number of human and macaque individuals tested, the map is arbitrarily thresholded to illustrate similarity of response with human data. D. Unsigned difference between the sizes of the effects illustrated in C in the left and in the right hemisphere. Color code indicates z-statistics at the group level, the map is restricted to the OMPFC and show clusters larger than 10 vertices.

104
Macaques 105 106 Reward-related asymmetry in macaques was investigated in fMRI data collected 107 from previous studies (see Methods). Eight monkeys who performed different types 108 of reward-related tasks were included in the analyses. For each monkey, we used the 109 reward-related contrasts (see Methods) of each session and averaged them across 110 sessions and individuals to obtain a whole-brain map of reward-related activity 111 ( Figure 1C). As in human participants, we projected each session map to a common 112 surface and computed the unsigned left versus right difference in all available 113 contrasts. We found two large clusters (larger than 10 vertices) of reward-related 114 asymmetry (z>2.3) in the OMPFC. First, we observed asymmetric reward-related 115 activity close to the medial orbital sulcus. Second, we also identified a cluster close to 116 the LOS, just posterior to its intersection with the TOS.

118
In summary, this second area of asymmetry in macaques lies in a similar location 119 with respect to sulcal landmarks in the two species ( Figure 1D). In humans it 120 corresponds to the caudal part of area 11l, extending into area a47r according to the 121 parcellation of Glasser et al, 2016 (28). This location corresponds to the caudal part of 122 47/12m in both humans and macaques in the standard cytoarchitectonic framework 123 proposed by Mackey and Petrides (29). 124 125 126 127

129
To determine whether this asymmetry could be explained by an asymmetry in the 130 functional connectivity of the OMPFC, we compared the connectivity profiles of the 131 left and right OMPFC. In both humans (n=57) and macaques (n=14), for each vertex 132 of the OMPFC, we extracted the connectivity (strength of correlation between time 133 series) with all vertices in the brain, from the group-level time series dataset 134 (computed with MIGP, see Methods). The procedure was repeated for the left and 135 the right OMPFC and in each case it was repeated to measure connectivity with 136 ipsilateral and contralateral hemispheres (thereby creating two maps illustrated in 137 figure 2A). The procedure was then repeated a further two times to examine the 138 connectivity of left and right OMPFC with the left hemisphere (regardless of 139 whether the left hemisphere was ipsilateral or contralateral) and the right 140 hemisphere (again, regardless of whether it was ipsilateral or contralateral). It was 141 then possible to assess whether there was any asymmetry in OMPFC connectivity 142 with either the ipsilateral or contralateral hemisphere or with either the left or the 143 right hemisphere. ( Figure 2B and C, see Methods). We found in each of the four 144 resulting maps of human OMPFC functional connectivity at least one cluster in the 145 OFC with a particularly high asymmetry. The conjunction of the four maps revealed 146 a unique cluster ( Figure 2D). In the following analyses, the FCA measure 147 corresponds to the average of the four types of asymmetry measures. We confirmed 148 the significance of FCA in this cluster at the group level in humans (t(56)=12.29, 149 p=2.10 -17 ). The same analysis conducted in macaque data revealed very similar 150 results; the conjunction analysis showed a single cluster in the OFC, with a 151 significant FCA at the group level (t(13)=3.01, p=0.01). 152

154
Overlap between reward-related cluster and functional connectivity cluster 155 To examine the link between reward-related asymmetry and functional connectivity 156 asymmetry, we projected the results from the two previous sets of analyses onto a 157 common surface (Figure 3). We observed partial overlap of the two clusters in the 158 lateral OFC, in both humans and macaques, indicating unique hotspots of functional 159 asymmetry, as defined by both reward-related activity and by functional 160 connectivity, in the OFC in both species. We computed the coefficient of functional 161 connectivity asymmetry (FCA) in the reward-related asymmetry (RRA) lateral 162 clusters and found that it was significantly higher than in the rest of the OMPFC  Moreover, we extracted the individual participants' RRA and FCA coefficients from 170 the OFC cluster resulting from the conjunction of the two asymmetry analyses 171 (labeled 'Functional Asymmetry cluster' or FA cluster). We found that the two 172 measures of asymmetry, based on RRA and FCA, were strongly correlated in 173 humans (r=0.35, p=8.10 -3 ). In macaques, in order to increase the statistical power of 174 the analysis, we decomposed the 14 individual RRA points into experimental data 175 points (18 different contrasts from 4 protocols, see table 1) and again found a 176 significant correlation between RRA and FCA measures (r=0.49, p=0.038). Together, 177 these results suggest that asymmetry in functional connectivity might explain 178 asymmetry of results in task-related activity in both species.

180
Functional connectivity characteristics 181 Finally, we compared the functional connectivity of the left and right FA cluster with 182 the whole brain in order to characterize their differences. In humans, we observed 183 that the left FA cluster shows a negative functional connectivity with a network 184 including anterior cingulate cortex (ACC), posterior cingulate cortex (PCC), and 185 temporoparietal junction (TPJ). We will refer to this network as the Default Mode 186 Network (DMN). We also observed that both seeds were positively connected to a 187 frontoparietal network, which we refer to as the Executive Network (ExN, Figure 4). 188 To quantify this difference, we extracted the functional connectivity of each seed interaction were also significant (Seed summarized in Figure 4. 222 223

Morphological characteristics in humans 224
Given the richness of the HCP data, we were able to further explore some 225 morphological features of the asymmetric OFC FA cluster. We checked whether it 226 was characterized by particular morphological features and found no specific 227 pattern of myelination, gyrification (curvature) or cortical thickness ( Figure 5). We 228 compared such features in the left and right FA cluster and found that the 229 myelination of the right FA cluster was higher than in the left FA cluster (t (56)

247
In the present study, we provide evidence for functional lateralization in OFC.

248
Lateralization in the frontal cortex has been considered most often in relation to 249 language processes and praxis (30-32) but also linked to attention (33) and emotional 250 regulation (34). Although the adaptive consequences of lateralized functions are not 251 well understood, it is thought that hemispheric specialization could increase 252 processing abilities by reducing bilateral redundancy (20). Reward-related 253 asymmetry in the OFC is consistent with many previous studies reporting unilateral 254 responses in the OFC (35-41), there has only rarely been acknowledgement that this 255 is the case (42, 43). Crucially we show an interrelationship across subjects between 256 the reward related asymmetry (RRA) and a functional connectivity-related 257 asymmetry (FCA). Differences between connectivity patterns in the left and right 258 OFC are notably related to their coupling with a set of brain regions often referred to 259 as the DMN. The right OFC was found to be more strongly connected to the DMN 260 than the left OFC. In addition, we observed a similar functional lateralization in the 261 OFC in non-human primates. This result suggests that this asymmetry could have 262 been present in the last common ancestor of humans and old-world monkeys 263 around 29 million years ago. A recent study found an inter-hemispheric OFC 264 asymmetry in rodents in a reversal learning task (44), with the right OFC being more 265 recruited in the task than the left OFC. In tandem with the current results this 266 suggests that reward-related asymmetry in or near OFC might have been a feature of 267 the mammalian brain present since the last common ancestor of rodents and 268 primates more than 100 million years ago.

270
To our knowledge, this is the first time that the functional asymmetry of the OFC 271 response to reward has been investigated in relation to the same region's 272 asymmetrical functional connectivity, in both humans and macaques. The reward 273 gambling task used in humans as part of the HCP has some limitations; the simple 274 condition contrast "reward vs punishment" is not ideal for investigating finer 275 aspects of the reward representation. It is therefore difficult to interpret the impact of 276 this OFC lateralization on cognitive processes and behavior. It is possible that the 277 results of studies employing causal approaches such as stimulation or investigation 278 of the effect of brain lesions that have also noted differences in effects in the two 279 hemispheres (13, 14, 45) reflect the same underlying asymmetry as investigated here.

281
It should, nevertheless be remembered that some studies have reported no effect of 282 OFC lesion laterality (46) or a bilateral OFC responses to reward (47, 48). Therefore, 283 it is important to mention that we do not claim an absolute and total functional 284 dissociation between left and right OFC but rather a graded difference between the 285 contributions that they make. If that is the case, then lateralization in reward-related 286 processing in OFC would resemble lateralization in the language system. It is 287 possible that the relative contribution of each hemisphere's OFC might differ 288 depending on the requirements of the experimental paradigm. For instance, some 289 studies only report the left OFC to represent outcome information (20), while others 290 only report the right OFC to respond to identity-specific value (19, 20).

292
It may be worth noting that in our study reward responsivity was investigated in the 293 context of decision making. Intriguingly a recent meta-analysis of lateralization of 294 function suggested that decision-making rather emotion, communication, or 295 perception/action is associated with the OFC lateralization (16). Intracranial 296 electrophysiological recordings in humans have shown that risk-taking biases are 297 driven by a lateralized push-pull neural response, with an increase of high 298 frequency activity in the right hemisphere biasing subjects toward risky bets (43).

299
Alternatively it has been suggested that OFC lateralization might be considered 300 within an exploration/exploitation framework (38). One possibility might be that 301 OFC lateralization is associated with the valence of feedback but no evidence has 302 been found that this is the case (38).

304
Given that connectivity constrains and partly determines the functions that could be 305 supported by a given brain region (49), one might use rs-fMRI results to further 306 speculate about the nature of the functional differences between the left and right 307 OFC. DMN has been shown to strongly overlap with the social brain network (50). 308 However, responses to social feedbacks, if anything appear stronger in the left OFC 309 than in the right OFC (51). DMN has also been associated with self-referential mental 310 activity, and recollection of prior experiences (52). It might therefore be 311 hypothesized that that it is an internally driven valuation process, i.e. a value 312 assignment that requires individuals to remember or simulate (such as the taste of a 313 cake), that underlies right OFC lateralization. On the other hand, a valuation process 314 linked to external features such as color combination in a painting could recruit the 315 left OFC more. Future investigations will aim at testing this specific hypothesis. 316

317
In summary, OFC lateralization has been overlooked or mentioned only in passing 318 in many functional studies. Here, we provide evidence for lateralization in terms of 319 reward-related function and in terms of functional connectivity both in humans and 320 in macaques. Therefore, we strongly encourage future studies to report relative 321 variation in activation in the left and right OFC, and to take into account differences 322 between hemispheres when interpreting the results in OFC.

328
Humans 329 The data used in this study are released as part of the Human Connectome Project 330 (WU-Minn Consortium: Human Connectome Project, RRID: SCR_008749, 331 http://db.humanconnectome.org) (51). We selected the S900 subject release with 7T 332 structural and resting-state MRI (rs-MRI) data. The data were preprocessed 333 according to the HCP pipeline (52). Of the 73 subjects in this specific HCP release, 16 334 subjects were excluded because of family ties with other subjects in the database. 335 The data analysis was therefore based on 57 subjects (37 females).

337
Analyses were conducted on the data aligned using areal-feature-based registration 338 (called "MSMAll" for "Multimodal Surface Matching" (29)). This procedure aligns 339 vertices on the cortical surface across subjects not only according to gross folding 340 morphology, but also takes into account the subject-specific functional features, such 341 as the location and distribution of resting-state networks. The MSMAll approach 342 dramatically improves the functional alignment of cortical areas over and above 343 registration based solely on volumetric or surface-based morphological registration. 344 This type of registration is referred to as "area-based" registration and is sometimes 345 considered a near optimal functional alignment (29 Gambling task in humans 362 Reward-related BOLD signal was recorded with fMRI during a card-guessing 363 gambling task played for monetary reward that has been previously described (26). 364 Participants completed a card-guessing game where they were required to guess the 365 number (ranging from 1 to 9) on a mystery card in order to win or lose money. 366 Participants were instructed to guess if the unknown card number was more or less 367 than 5 by pressing one of two buttons on a response box. Feedback was given as the 368 revealed card number with a cue to inform the participants if they received a 369 monetary reward, monetary loss or nothing (neutral no reward/loss outcome 370 received for number 5) trial. The task was presented in blocks of eight trials that 371 were either mostly rewarded (six reward trials pseudo-randomly interleaved with 372 neutral and/or loss trials) or mostly loss (six loss trials interleaved with reward 373 and/or loss trials). For each of the two runs, there were two mostly reward and two 374 mostly loss blocks, interleaved with four fixation blocks (15 s duration).

376
Protocol 1 in monkeys: Object Discrimination Reversal Task 377 The experimental task used in Protocol 1 is described in detail elsewhere (39, 53). 378 Briefly, the task was designed to investigate contingent learning mechanisms and 379 specifically how and where in the brain associations between choice options and 380 outcomes (i.e. reception of reward) resulting from choosing them are formed. Four 381 macaques had to choose between pairs of abstract visual stimuli while in the 382 magnetic resonance imaging (MRI) scanner. On each trial, the two stimuli available 383 for choice (available options) were drawn from a set of three, each associated with 384 distinct reward probabilities. The rewards were delivered probabilistically in a 385 manner that fluctuated across the session, with two of the options reversing toward 386 the middle of a session. Each stimulus' reward probability was uncorrelated from 387 that of the others. On each trial one of the two available options was chosen by the 388 monkey, the other was unchosen and a third option was invisible and unavailable 389 for choice. In our study, we focused on the receipt of the reward.

391
Protocol 2 and 3 in monkeys: Decision to act Task 392 The experimental task used in Protocol 2 is described in detail elsewhere (54). 393 Briefly, the task was designed to investigate how contextual factors and internal 394 state, shaped by present and past environment are integrated to influence whether 395 and when to act. 4 monkeys initially performed this task but we only included the 396 two monkeys (13 and 14) who also performed the resting-state fMRI data 397 acquisition. In that task, macaques were trained to track the number of dots on a 398 screen while in the MRI scanner. Dots appeared one at a time on a screen and 399 animals could decide to make a response, at a time of their choice, by tapping on a 400 response pad in front of them. The number of dots on the screen at the time of 401 response determined the probability of reward. Reward probability was drawn from 402 a sigmoid function: the longer the animals waited before responding, more dots 403 appeared on the screen, and the higher was the probability of reward. Different 404 levels of reward magnitude were associated with different dot colors, and the 405 reward magnitude varied from trial-to-trial. Once the monkeys responded, they 406 received drops of juice or no juice according to the reward probability distribution 407 and the time of their response. There was a 4 second delay between the response and 408 the outcome. In the context of our study, two events on each trial were of special 409 interest: the onset of the stimulus (dots), since the color is indicating the expected 410 level of reward, and the outcome (0, 1, 2 or 3 drops of blackcurrant juice).

412
Data from protocol 3 has not been published yet. However, the task is exactly the 413 same except that the frequency of all the good offers increased and of all the bad 414 offers decreased (i.e., there were more trials with high reward magnitude and less 415 trials with low reward magnitude in protocol 3 compared to protocol 2).

417
Protocol 4: Stimulus-reward association task 418 The data and results from the experimental task used in Protocol 4 have not been 419 published yet. Briefly, the control task used here investigated how a monkey would 420 respond to visual cues indicative of how much reward could be obtained, or lost (i.e. 421 poured into a visible plastic jar). 4 male rhesus macaques were trained to associate a 422 set of 10 stimuli with various reward magnitudes (i.e. from 0 to 2 drops of reward 423 smoothie that could be either obtained or discarded). On any trial one stimulus was 424 presented on the screen. The monkey had 10 seconds to respond by putting his hand 425 over a homemade infrared sensor. Once selected the stimulus was replaced by a 426 hollow white frame. After a 3.5 to 4.5s delay, the stimulus was presented back 427 (feedback) and the reward delivered. If the monkey did not respond within 10 428 seconds, the trial was aborted and the same stimulus was presented again after the 429 inter-trial interval. The stimulus-outcome association was probabilistic. In 24% of the 430 trials, the feedback was different from the cue. Parameter estimates (BOLD) for the contrast (reward > punishment; cope6.feat) were 450 available for 57 participants. We chose this contrast to establish relationships with 451 reward. As the paradigm was a card-guessing task, the contrast corresponded to 452 reward receipt and did not include an anticipation phase. 453 To obtain group statistics, second level (group) analysis on volumes was conducted 454 using FLAME (FMRIB's Local Analysis of Mixed Effects) stage 1, part of FSL (version 455 5.0.8 http://fsl.fmrib.ox.ac.uk/). The main contrast of interest, "Reward versus 456 Punishment", of each participant was entered into a second level random-effects 457 analysis using a one-sample t-test. The main effect images are all cluster-corrected 458 results with the standard threshold of z>2.3.

460
For clarity in the data visualization and for a better visual comparison with resting-461 state data, we then projected the volume result on the averaged MSMAll 462 midthickness surface of all participants, using the 'wb command′ and 'volume to 463 surface mapping' functions from the connectome-workbench 464 (https://www.humanconnectome.org/software/connectome-workbench.html).

466
To test the asymmetry of reward-related activity, each individual z-stat map 467 corresponding to the 'reward vs punishment' contrast was projected onto its 468 corresponding MSMAll surface. Then, the left and right data were extracted from 469 each hemisphere in the OMPFC. The individual unsigned difference between the left 470 and right z-statistics in the OMPFC were computed and then assessed for 471 significance at the group level using permutation tests (see below). weighted images were acquired using a gradient-refocused echo (GRE) sequence 488 with a 1.5 x 1.5 x 1.5 mm resolution, TR 10 ms, TE 2.52 ms, and flip angle 25°. These 489 images were later used for offline MRI reconstruction.

491
Preprocessing was performed using tools from FMRIB Software Library (FSL) (58), 492 Advanced performance were reconstructed by an offline-SENSE method that achieved higher 498 signal-to-noise and lower ghost levels than conventional online reconstruction (61) 499 (Offline_SENSE GUI, Windmiller Kolster Scientific, Fresno, CA). A low-noise EPI 500 reference image was created for each session, to which all volumes were non-linearly 501 registered on a slice-by-slice basis along the phase-encoding direction to correct for 502 time-varying distortions in the main magnetic field due to body and limb motion. 503 The aligned and distortion-corrected functional images were then non-linearly 504 registered to each animal's high-resolution structural images. A group specific 505 template was constructed by registering each animal's structural image to the 506 CARET macaque F99 space (61). Finally, the functional images were temporally 507 filtered (high-pass temporal filtering, 3-dB cutoff of 100s) and spatially smoothed 508 (Gaussian spatial smoothing, full-width half maximum of 3mm). 509 510 fMRI data analysis in macaques 511 To perform whole-brain statistical analyses we used a univariate generalized linear 512 model (GLM) framework as implemented in FSL FEAT (62). At the first level, we 513 constructed a GLM to compute the parameter estimates (PEs) for each regressor. The 514 GLMs were constructed based on the specific questions raised in each protocol: 515 516 -GLM1 (Protocol 1): = 0 + 1 DEC + 2 choV + 3 uncV + 4 unpV + 5 choT-517 uncT + 6 unpCT + 7 locT + 8 REW + 9 NOREW + 10 cClo + 11 rewTreward 518 + 12 rewTnoreward + 13 + 14 ℎ + ε 519 520 -GLM2 (Protocol 2): Regressors of interest: 536 -REW and NOREW: constant regressors were time-locked to onset of feedback, for 537 receipt or non-receipt of the reward 538 -expectedReward: parametric regressor with up to four levels (depending on 539 protocol), which represents expected reward magnitude 540 -levelOut: parametric regressor with three or four levels representing the reward 541 outcome on the current trial 542 543 Regressors of non-interest: rightwards responses 578 -mouth: distortion due to mouth movements 579 580 Regressors in bold are the contrasts linked to reward that we included in our 581 analyses. For each protocol and each contrast, the first-level z-statistics of each 582 session in every monkey were extracted to compute the main effect of reward (fixed 583 effect analysis on volumes). Then, each z-statistic volume was projected onto left and 584 right surfaces and used to compute the asymmetry of reward representation in the 585 OMPFC (linear mixed-effect models that include random factor for protocol and 586 monkeys). 587 588 rs-MRI data acquisition and processing in humans 589 The preprocessed 7T data were downloaded from the HCP website. We selected the 590 package called ' rs-MRI data acquisition and processing in macaques 603 The 14 monkeys were scanned under anesthesia to acquire resting-state data. fMRI 604 and anatomical scans were collected according to previously used protocols (67).

605
Anesthesia was induced using intramuscular injection of ketamine (10 mg/kg) 606 combined with either xylazine (0.125-0.25 mg/kg) or midazolam (0.1 mg/kg) and 607 buprenorphine (0.01 mg/kg). Macaques also received injections of atropine (0.05 608 mg/kg), meloxicam (0.2 mg/kg), and ranitidine (0.05mg/kg). Anesthesia was 609 maintained with isoflurane. Isoflurane was selected because it has been 610 demonstrated that resting-state networks are still present using this agent for 611 anesthesia (68). The anesthetized animals were placed in an MRI-compatible 612 stereotactic frame (Crist Instrument) in a sphinx position within a horizontal 3T MRI 613 scanner with a full-size bore. The same coils as for awake scans (see fMRI data 614 acquisition) were used for data acquisition. Whole-brain BOLD fMRI data were 615 collected using the following parameters: 36 axial slices, resolution of 1.5 × 1.5 mm, 616 slice thickness of 1.5 mm, TR of 2280 ms, TE of 30 ms, 1600 volumes. Structural scans 617 were acquired in the same session using a T1-weighted MP-rage sequence (no slice 618 gap, 0.5 × 0.5 × 0.5 mm, TR of 2500 ms, TE of 4.01 ms and 128 slices). 619 620 The detailed preprocessing pipeline for the resting-state fMRI has been described 621 elsewhere (69, 70). Briefly, after reorientation to the same convention for all 622 functional EPI datasets, the first volumes were discarded to ensure a steady radio 623 frequency excitation state. EPI timeseries were motion corrected using MCFLIRT 624 (71). Brain extraction, bias-correction, and registration were achieved for the 625 functional EPI datasets in an iterative manner, the mean of each functional dataset 626 was registered to its corresponding T1w image using rigid-body boundary-based 627 registration (FLIRT, (71, 72))). EPI signal noise was reduced both in the frequency 628 and temporal domain. The functional timeseries were high-pass filtered with a 629 frequency cut-off at 2000 s. Temporally cyclical noise, for example originating from 630 the respiration apparatus, was removed using band-stop filters set dynamically to 631 noise peaks in the frequency domain of the first three principal components of the 632 timeseries. To account for remaining global signal confounds we considered the 633 signal timeseries in white matter and meningeal compartments, there confound 634 parameters were regressed out of the BOLD signal for each voxel. Following this 635 confound cleaning step, the timeseries were low-pass filtered with a cut-off at 10 s. 636 The data were transformed to F99 and spatially smoothed using a 2 mm FWHM 637 Gaussian kernel. Lastly, the data timeseries were demeaned to prepare for functional 638 connectivity analyses. In macaques, the networks were defined from the connectivity of bilateral seeds in 664 the anterior cingulate sulcus (DMN) and the mid-cingulate sulcus (ExN). 665 666

ROI definition 667
Regions of interest (ROI) were drawn manually on the ventro-medial prefrontal 668 cortex (vmPFC) and the orbitofrontal cortex (OFC), to cover a large portion of the 669 orbito-medial prefrontal cortex (OMPFC). The dorsal medial boundary was 670 delineated by an arbitrary horizontal line that runs from the front of the brain to the 671 genu of the corpus callosum. The ventral surface of the frontal lobe was included 672 from the frontal pole rostrally to the anterior perforated substance caudally.

674
Functional Connectivity Asymmetry coefficient 675 To investigate the asymmetry of connectivity between the left and the right OMPFC, 676 four measures of asymmetry were used: 677 • Ipsilateral Functional Connectivity Asymmetry (FCAIpsi): Difference between 678 the connectivity of the left OMPFC (OL) with the left hemisphere (HL) and 679 the right OMPFC (OR) with the right hemisphere (HR). The statistical validity of our results was assessed by extracting variables of interest 703 from each subject and testing for significance at the group level using one-sample t-704 tests. When assessing significance of clusters on resting-state MRI data, each FCA 705 map was computed for every subject. The main effect was then tested using one-706 sample student t-test (two-tailed). 707 708 To assess the statistical validity of the RRA clusters in both humans and monkeys, 709 we used the Fisher randomization test (74) with 10000 randomizations of the RRA 710 values (z-scored) of each subject. The maximal cluster-level statistics (the sum of t-711 values across contiguous points passing a significance threshold of 0.01 (z=2.3)) were 712 extracted for each shuffle to compute a 'null' distribution of effect size across the 713 OMPFC mask. For each significant cluster in the original (non-shuffled) data, we 714 computed the proportion of clusters with higher statistics in the null distribution, 715 which is reported as the 'cluster corrected' p-value (pcorr) (75).

717
Anatomical MRI data acquisition and analyses 718 The preprocessed anatomical 7T data were downloaded from the HCP website. We 719 selected the package called 'Structural Preprocessed for 7T (1.6mm/59k mesh)', 720 which contained 1.6mm resolution data, collected at 3T. In this package, myelin, 721 curvature and cortical thickness maps are available for each subject, registered on 722 MSM-All, making those maps comparable with the connectivity maps. When 723 investigating the morphological features of the OMPFC, we extracted the values of 724 those maps for each subject and computed the mean of each feature.