Neural signals related to movement can be measured from intracranial recordings and used in brain-machine interface devices (BMI) to restore physical function in impaired patients. In this study, we explore the use of more abstract neural signals related to economic value in a BMI context. Using data collected from the orbitofrontal cortex in non-human primates, we develop deep learning-based neural decoders that can predict the monkey's choice in a value-based decision-making task. Out-of-sample performance was improved by augmenting the training set with synthesized data, showing the feasibility of using limited training data. We further demonstrate that we can predict the monkey's choice sooner using a neural forecasting module that is equipped with task-related information. These findings support the feasibility of user preference-informed neuroengineering devices that leverage abstract cognitive signals.

ABSTRACTThe orbitofrontal cortex (OFC) plays a crucial role in value-based decision-making. While previous research has focused on spiking activity in OFC neurons, the role of OFC local field potentials (LFPs) in decision-making remains unclear. LFPs are important because they can reflect synaptic and subthreshold activity not directly coupled to spiking, and because they are potential targets for less invasive forms of brain-machine interface (BMI). We recorded LFPs and spiking activity using multi-channel vertical probes while monkeys performed a two-option value-based decision-making task. We compared the value- and decision-coding properties of high-gamma range LFPs (HG, 50-150 Hz) to the coding properties of spiking multi-unit activity (MUA) recorded concurrently on the same electrodes. Results show that HG and MUA both represent the values of decision targets, and that their representations have similar temporal profiles in a trial. However, we also identified value-coding properties of HG that were dissociable from the concurrently-measured MUA. On average across channels, HG amplitude increased monotonically with value, whereas the average value encoding in MUA was net neutral. HG also encoded a signal consistent with a comparison between the values of the two targets, a signal which was much weaker in MUA. In individual channels, HG was better able to predict choice outcomes than MUA; however, when simultaneously recorded channels were combined in population-based decoder, MUA provided more accurate predictions than HG. Interestingly, HG value representations were accentuated in channels in or near shallow cortical layers, suggesting a dissociation between neuronal sources of HG and MUA. In summary, we find that HG signals are dissociable from MUA with respect to cognitive variables encoded in prefrontal cortex, evident in the monotonic encoding of value, stronger encoding of value comparisons, and more accurate predictions about behavior. High-frequency LFPs may therefore be a viable – or even preferable – target for BMIs to assist cognitive function, opening the possibility for less invasive access to mental contents that would otherwise be observable only with spike-based measures.

Full text Figures and data Side by side Abstract Editor's evaluation eLife digest Introduction Results Discussion Methods Data availability References Decision letter Author response Article and author information Metrics Abstract In economic decision-making individuals choose between items based on their perceived value. For both humans and nonhuman primates, these decisions are often carried out while shifting gaze between the available options. Recent studies in humans suggest that these shifts in gaze actively influence choice, manifesting as a bias in favor of the items that are viewed first, viewed last, or viewed for the overall longest duration in a given trial. This suggests a mechanism that links gaze behavior to the neural computations underlying value-based choices. In order to identify this mechanism, it is first necessary to develop and validate a suitable animal model of this behavior. To this end, we have created a novel value-based choice task for macaque monkeys that captures the essential features of the human paradigms in which gaze biases have been observed. Using this task, we identified gaze biases in the monkeys that were both qualitatively and quantitatively similar to those in humans. In addition, the monkeys' gaze biases were well-explained using a sequential sampling model framework previously used to describe gaze biases in humans—the first time this framework has been used to assess value-based decision mechanisms in nonhuman primates. Together, these findings suggest a common mechanism that can explain gaze-related choice biases across species, and open the way for mechanistic studies to identify the neural origins of this behavior. Editor's evaluation This study analyzed viewing behavior in monkeys during value-based decision-making to determine whether relationships between gaze patterns and choices previously described in humans are also present in monkeys. The study used a clever task design and sophisticated modeling approaches to reveal robust evidence for similarities to extant human data. This is important to the field because it suggests common neural mechanisms linking viewing behavior and decision-making, which can now be further explored across species. https://doi.org/10.7554/eLife.78205.sa0 Decision letter Reviews on Sciety eLife's review process eLife digest When we choose between two items, we might expect to spend more time looking at the one we have a pre-existing preference for. For example, at the grocery store, you might assume that someone who likes grapes better than bananas would spend a longer time looking at the grapes. Surprisingly, a series of studies on human decision-making have shown that the opposite relationship is also true: the more time we spend looking at an item, the more likely we are to pick it. This 'gaze bias' occurs in many real-life and laboratory decision settings, and it is especially evident for choices between two equally preferred options. However, examining the brain circuits that underpin this behavior has so far been difficult due to a lack of animal models in which to study them. In response, Lupkin and McGinty proposed that rhesus macaques may be the ideal species in which to study gaze biases, as these animals likely rely on the same brain regions as humans when gazing and making decisions. To test this hypothesis, a computer-based decision game similar to the ones used for humans was designed for the monkeys. It involved the animals having to choose between two icons that were associated with different amounts of a juice reward. Analysing how long the macaques had spent looking at each icon before making their choice revealed that they indeed tended to select the icon they had looked at for longer – including when the two icons indicated equal rewards. Other types of gaze biases present in humans were also detected, such as choosing the icon that was viewed first or last in a trial. Additional analyses using computer simulations confirmed that the gaze biases of humans and monkeys were comparable and, critically, that they could be explained by similar underlying brain processes. These strong similarities suggest that rhesus macaques could be used to study the neural basis for decision-making in both humans and nonhuman primates, potentially making it easier to examine the harmful changes in decision-making that occur in conditions like substance abuse or depression. Introduction Economic decisions are ubiquitous in the natural world, and in natural settings humans and other primates often evaluate the available options by moving their eyes, rapidly shifting the focus of their gaze between the options as they deliberate. Over the past decade, studies in humans have documented an association between gaze and value-based choices. In particular, when offered a choice between multiple items, people are more likely to select the first item that they view, the item they look at just prior to indicating their decision, and the item that they spend the longest time viewing over the course of their deliberation. Further, choice biases can be induced through manipulations of gaze (Gwinn et al., 2019; Liu et al., 2020; Liu et al., 2021; Pärnamets et al., 2015; Sui et al., 2020; Tavares et al., 2017). Together, these studies suggest that gaze plays an active role in the decision process, such that simply viewing a given option makes us more likely to choose it (see Krajbich, 2019, for a review). However, despite increased interest in the role of gaze in decision-making, knowledge of the underlying neural mechanisms is still limited (c.f., Krajbich et al., 2021; Lim et al., 2011). One critical barrier to identifying these mechanisms is the lack of a suitable animal model of this behavior, and consequently a lack of studies addressing neural mechanisms at the cellular level. To this end, the aim of the present study was twofold: first, to develop a novel behavioral paradigm for macaque monkeys that captures the essential features of tasks previously used in humans; and second, to determine whether monkeys exhibit gaze-based choice biases that are similar to those of humans. While the neural mechanisms linking gaze and economic choice are not fully known, the underlying computations have been explored using sequential sampling models, a highly successful class of models used to explain simple decision behaviors (Ratcliff and McKoon, 2008). In these models, a stream of noisy evidence for each option is accumulated until one item accrues enough evidence to reach a pre-defined threshold, and a decision is rendered in favor of that item. For value-based tasks, the evidence accumulation rate can be formulated as a function of the relative value of the possible outcomes—the larger the difference in relative value, the faster that evidence accumulates for the higher valued item. Over the last decade, the basic sequential sampling model framework has been extended to account for the influence of gaze behavior (i.e. overt attention) on the decision process. These models, collectively referred to as attentional sequential sampling models (aSSMs), assume that evidence for a given item accumulates at a faster rate while that item is being viewed, and accumulates slower when it is not being viewed (Krajbich et al., 2010; Krajbich and Rangel, 2011; Smith and Krajbich, 2018; Tavares et al., 2017; Thomas et al., 2019). Overall, these models have been successful in accounting for human gaze biases across several decision-making domains (Krajbich, 2019; Manohar and Husain, 2013; Smith and Krajbich, 2018; Thomas et al., 2019; Westbrook et al., 2020), supporting the idea that gaze modulates the decision process by biasing an evidence accumulation process. To identify candidate neural substrates, intracranial recordings in nonhuman primates (NHPs) are an ideal approach, as they combine an animal model with human-like oculomotor behavior with the high temporal resolution necessary to accommodate the rapid pace of natural eye movements (Kravitz et al., 2013; Ongür and Price, 2000; Wise, 2008). Moreover, NHPs are commonly used as animal models in neuroeconomics. However, existing paradigms contain elements that make them unsuitable for testing the relationship between gaze and choice behavior. More specifically, many NHP studies restrict eye movements and/or use eye movements as a means of reporting choice, rather than for inspecting the choice options, (Cavanagh et al., 2019; Hunt et al., 2018; Rich and Wallis, 2016). These common restrictions can interfere with natural gaze patterns and therefore limit the ability to identify the relationship between gaze behavior and choice. To address the need for an animal model of gaze biases, we have developed a novel behavioral paradigm in which monkeys are able to freely deploy their gaze to the targets on the screen and to indicate their choice by a manual response. Using this paradigm, we replicated the three core behavioral signatures of gaze-related choice biases: We found that the monkeys were more likely to choose the first item that they viewed, the last item that they viewed before deciding, and the items that received more overall gaze time. In addition, we use a computational model previously used to assess human gaze biases (Thomas et al., 2019) to determine whether similar computational mechanisms can explain choices in NHPs. Results Basic psychometrics We trained two monkeys to perform a value-based decision task in which they used eye movements to view the choice options and used a manual response lever to indicate their decision (Figure 1A). To begin a trial, the monkeys were required to fixate on the center of the task display while holding down the center of three response levers. After the required eye fixation and lever hold period was satisfied, two targets were presented on the left and right of the task display, each associated with a reward ranging from 1 to 5 drops of juice. Following the initial fixation period, gaze was unrestricted, allowing the monkeys to look at each target as many times and for as long as they liked before indicating a choice. To encourage the monkeys to look directly at the targets, and to limit their ability to perceive the targets using peripheral vision, the targets were initially masked until the first saccade away from the central fixation point was detected (see Methods). To indicate their choice, the monkeys first lifted their hand off the centrally-located response lever, and then pressed the right or left lever to indicate which of the two targets they chose. The reaction time (RT) in each trial was defined as the period between the onset of the targets and the lift of the center lever. We collected data from a total of 54 sessions: 29 from Monkey C (N = 15,613) and 25 for Monkey K (N = 14,433). Figure 1 with 1 supplement see all Download asset Open asset Decision-making task and performance in two monkeys. (A) Abbreviated task sequence; see Figure 1—figure supplement 1 for full task sequence. The interval between target onset and center lever lift defines the decision reaction time (RT). The yellow and blue glyphs are choice targets, and the gray '+' shapes indicate the location of visual crowders designed to obscure the targets until they are viewed (fixated) directly. For clarity, in this panel crowders are shown in gray and at a reduced scale; on the actual task display, crowders were multicolored and the same size as the targets, as in panel C. (B) An example set of 12 choice targets, organized into 5 groups corresponding to the 5 levels of juice reward. (C) Close-up view of a single target array, consisting of a central yellow choice target surrounded by six non-task-relevant visual crowders. Two such arrays appear on the display (panel A), each located 7.5° from the display center. (D–F). Task performance. (D) Fraction of left choices as a function of the left minus right target value (in units of juice drops). (E) Reaction Time (RT) decreases as a function of difficulty, defined as the absolute difference between target values; (F) Number of fixations per trial decreases as a function of difficulty. Filled circles show the mean values and smooth lines show the logistic (D) or linear (E–F) regression fits derived from mixed effects models (Supplementary file 1, rows 1–3, respectively). Error bars are too small to be plotted. The data in blue are from Monkey C (N = 15,613), and in red, from Monkey K (N = 14,433). © 2021, McGinty and Lupkin. Figure 1A-C are reproduced from Figure 1A-C of McGinty and Lupkin, 2021, bioRxiv, published under the CC-BY-NC 4.0 International license (https://creativecommons.org/licenses/by-nc/4.0/). Overall choice performance is illustrated in Figure 1D. Choices were nearly optimal for the easiest trials (value difference of +/-3 or 4 drops of juice,>98% of choices in favor of higher value target) and occasionally suboptimal for more difficult trials (value difference of +/-1 or 2 drops of juice, 86% higher-value choices for both Monkey C and Monkey K). When the targets were of equal value, the probability of the monkeys choosing the left or right target was near chance (50% left choices for Monkey C, 46% for Monkey K). This pattern of choices indicates that the monkeys were behaving according to the incentive structure of the task, with a very low rate of guesses or lapses (Fetsch, 2016). Further, choice behavior was well-described by a logistic function parameterized by the difference between the values of the left and right items (fit lines in Figure 1D are from a logit mixed effects regression: βvalue-difference = 1.37, CI = [1.34, 1.40]; F (1,30044)=8411.60, p<1e-10). Additionally, we found that both the average RT and the average number of fixations per trial were modulated by choice difficulty: the monkeys responded slower and made more fixations when the two targets were near in value, compared to when the value difference between the targets was large (RT: Figure 1E, linear mixed effects regression: βdifficulty = -0.032, CI: [-0.039,–0.026]; F (1, 30044)=106.09, p<1e-10. Number of Fixations: Figure 1F, Linear mixed effects regression: βdifficulty = -0.10, CI = [-0.12,–0.09]; F(1, 30044)=126.7, p<1e-10). Together these results are consistent with behavioral patterns seen in both humans and NHPs performing 2-alternative forced choice tasks (e.g. Gold and Shadlen, 2007; Thomas et al., 2019). Effects of difficulty and chosen value on reaction times and the number of fixations In the preceding section, we show that both RT and the number of fixations increase as a function of trial difficulty. However, in high-performing subjects trial difficulty is correlated with chosen value, because the trials with large value differences (low difficulty) are also trials where higher-value offers are chosen. Therefore, the RT and fixation effects originally attributed to difficulty may be explained in part by chosen value. To determine the unique contributions of chosen value and difficulty to RT and the total number of fixations, we used the approach of Balewski et al., 2022. Briefly, we modeled each behavioral metric as a function of one predictor, and then regressed the other predictor against the residuals of the first model. For example, to measure the unique contribution of chosen value to RT, we first ran a linear mixed effects model explaining RT as a function of difficulty. We then took the residuals from this model, which represent the RT after accounting for difficulty, and ran a second mixed effects model explaining these residuals as a function of chosen value. Consistent with the findings of Balewski and colleagues, RT was modulated more by chosen value than by difficulty (βchosen-value = -0.026, CI = [-0.034,–0.019]; F (1,30044)=46.35, p<1e-10; CPD = 8%; βdifficulty = -0.011, CI = [-0.012,–0.010]; F (1,30044)=318.44, p<1e-10; CPD = 1.4% Figure 2A/C, Supplementary file 1 rows 4–5). Interestingly, we found the opposite pattern for the total number of fixations per trial: this metric was strongly modulated by difficulty (βdifficulty = -0.065, CI = [-0.086,–0.043]; F (1,30044)=34.46, p=4.41e-9; CPD = 1.8%; Figure 2C), and showed no effect of chosen value (βchosen-value = -0.028, CI = [–0.095, 0.39]; F (1,30044)=0.66, p=0.42; CPD = 0.69%; Figure 2D, Supplementary file 1 rows 4–5). In sum RTs appear to reflect the size of the upcoming reward (faster responses for higher rewards), whereas the number of fixations reflect choice difficulty, with more fixations associated with more difficult trials. Figure 2 Download asset Open asset Differential contributions of difficulty and chosen value to decision reaction times and the number of fixations per trial. (A) The effect of trial difficulty on reaction time after accounting for the effect of chosen value. (B) The effect of trial difficulty on number of fixations after accounting for chosen value. (C) The effect of chosen value on reaction time after accounting for difficulty. (D) The effect of chosen value on number of fixations after accounting for difficulty. Y-axes give the dependent measures (RT or number of fixations) after being residualized by either difficulty or chosen value, as indicated (see Methods). Dots indicate the mean of each dependent measure for each session. Black lines show linear model fits to the plotted dots. The betas and coefficients of partial determination (CPD) of these models are in the top right corner. Blue dots show the session means from Monkey C (N=25 sessions, 15,613 trials), red dots are from Monkey K (N=29 Sessions,14,433 trials). Gaze behavior and basic fixation properties The following sections characterize the number and durations of the monkeys' fixations onto the choice targets throughout each trial. A fixation was defined as a period of stable gaze upon one of the two targets. Consistent with fixation analyses in human studies, consecutive fixations upon the same target were merged into a single fixation epoch, and fixations onto non-target locations were not included in the analysis (See Behavior Analysis for details). Following conventions from human studies, final fixation durations are defined as the interval between fixation onset and the center lever lift; in other words, final fixation durations are delimited by the reaction time in each trial (e.g. Krajbich et al., 2010). For non-final fixations, the duration is defined in the conventional way, as the interval between fixation onset and the initiation of the next saccade. Number of fixations per trial In our task, information about the choice targets was obscured unless the monkey was directly fixating on them. Therefore, the optimal strategy involves a minimum of two fixations, one to each target. Consistent with this, the monkeys almost always looked at both targets on each trial (97.2% of trials for Monkey C, 91.6% for Monkey K). The fraction of trials with 1, 2, 3, or ≥4 total fixations before the RT for Monkey C was 2.8%, 41.2%, 49.2%, and 6.8%; and for Monkey K was 8.4%, 43.0%, 43.8%, 4.8%. Thus, in the vast majority of trials the monkeys either looked at each target exactly once (2 total fixations) or looked at one target once and the other twice (3 total fixations). The average number of fixations per trial for Monkey K was 2.45 SEM 0.02 and for Monkey C was 2.57 SEM 0.04. This average is smaller than in human binary choice tasks, in which the average number of fixations per trial is usually 3 or more. As we discuss below, the difference between human and monkey total fixation numbers has implications for comparing fixation durations between species. Fixation durations depend on relative and absolute order in the trial Fixations can be defined by both their absolute and relative order in a trial. Absolute order refers to the serial position in the trial: first, second, third, etc. Relative order refers to when the fixation occurs with regards to the trial start or end: initial, middle, final. Note that these definitions are not independent or mutually exclusive. Instead, they depend on the total number of fixations in the trial. For example, in a trial with a total of two fixations, the second fixation is also the 'final' fixation (there are no middle fixations). However, in a trial with three total fixations, the second fixation is considered a 'middle' fixation. Because the monkeys usually make either two or three total fixations in a trial, relative and absolute order are conflated. (In contrast, humans typically make three or more fixations per trial, so that second fixations are almost always 'middle' fixations.) Consequently, it is necessary to independently analyze the effects of both absolute and relative fixation order. To assess the effects of absolute order (first, second, third, etc.) independent of relative order (final/non-final), we first fit a linear regression explaining fixation duration as a function of relative order. We then regressed absolute order against the residuals from the first model. Consistent with human fixation patterns, we found that fixation durations decreased as a function of absolute order, after controlling for relative order (βrelativeOrder = -0.049, CI = [-0.065,–0.033]; F(1, 7762)=36.06, p<1.92e-9; Figure 3). The decrease is especially notable at the third fixation, which in a binary task is the first 're-fixation' onto a previously sampled target, and is consistent with human fixation patterns (Manohar and Husain, 2013). Next, to assess the effects of relative order independent of absolute order, we fit a linear mixed-effects model comparing fixation durations across absolute positions, treating absolute position as a categorical variable. We then fit a second model explaining the residuals as a function of whether or not the fixation was a final fixation (i.e., relative order). After accounting for serial position, we found that non-final fixations were shorter than final fixations (βrelativeOrder = -0.047, CI = [-0.057,–0.037]; F(1, 7762)=86.34, p<1e-10; Figure 3). In sum, we identified two independent order-based effects on duration: a decrease in duration with successive fixations in a trial, as well as a tendency to fixate longer on final fixations. Figure 3 Download asset Open asset Distribution of fixation durations across absolute and relative position in the trial. Violins show distribution of fixation durations across absolute position in the trial (x-axis) and split according to whether they were final (dark gray) or non-final (light gray) fixations in the trial. 1st fixations: N=30,046 non-final and 1,652 final; 2nd fixations: N=15,753 non-final and 12,641 final; 3rd fixations: N=1,749 non-final and 14,004 final; 4th fixations or greater: N=80 non-final (not shown) and 1,749 final. The monkeys' prolonged final fixations seem to contradict prior human studies, which typically report shorter final fixations. However, this discrepancy may be explained by the relatively larger number of fixations that humans make (mean ~3.7, computed from freely available data from the study described in Krajbich et al., 2010), compared to the monkeys in the present study (mean ~2.5). Consequently, humans' final fixations tend to have higher absolute order and are thus nearly always re-fixations onto a target—both factors that are associated with shorter fixation durations, as noted above. In contrast, the monkeys' final fixations are overwhelmingly either the second or third in the trial. Thus, not only do their final fixations have a lower absolute order, but they are also comprised of a larger fraction of longer initial fixations onto a target (i.e. when the final fixation is the 2nd fixation). Fixation durations depend on target value The average fixation duration was dependent on the value of the target being fixated, and upon the order in the trial in which the fixation occurred. For the first fixation in each trial, duration increased as a function of fixated target value (βfixValue = 0.003, CI = [0.0011,0.0040]; F(1, 28392)=11.37, p=0.0007, Supplementary file 1, row 8). Intuitively, the effect size can be interpreted as the influence of a one-drop change in juice volume; for example, for initial fixations the regression estimate of 0.003 corresponds to a 3ms increase in fixation duration per drop of juice. For middle fixations, defined as any fixation that was neither first nor final, the average duration decreased as a function of the value of the fixated target (βfixValue = -0.0076, CI = [-0.0086,–0.0067]; F(1, 17580)=231.37, p<1e-10; Supplementary file 1, row 9). At the same time, middle fixation durations were also influenced by the value of the target not currently being viewed (βnofixValue = -0.035, CI=[-0.040,–0.030]; F(1, 17580)=197.78, p<1e-10; Supplementary file 1, row 10); because most middle fixations were the second fixations in the trial, this means that the value of the first-viewed item 'carried over' into the second fixation to influence its duration. Combining the influences of fixated and non-fixated values into a single variable, middle fixation durations increased as a function of relative value, defined here as the value of the fixated target minus the value of the non-fixated target (βrelativeFixVal = 0.016, CI = [0.012, 0.021]; F(1, 17580)=48.55, p<1e-10; Supplementary file 1, row 11). Finally, we find that durations decreased as function of the absolute difference in target values (βdifficulty = -0.019, CI = [-0.025,–0.0154]; F(1, 17580)=47.17, p<1e-10; Supplementary file 1, row 12). Together, these results support the notion that fixation behavior and valuation are not independent processes, and that the effects of value on viewing durations must be accounted for when computing and interpreting gaze biases (see Cumulative Gaze-Time Bias, below). Gaze-related choice biases Humans show three core choice biases related to gaze behavior. First, they tend to choose the item that they viewed first (initial fixation bias). Second, they tend to choose the item they spend more time looking at over the course of the trial (cumulative gaze-time bias). Third, they tend to choose in favor of the target they are viewing at the moment they indicate their choice. All three of these biases are evident in our monkey subjects, as detailed below. However, before these biases can be quantified, we must first address the question of whether monkeys show evidence for a two-stage form of gaze bias recently identified in human behavior. Evidence for a two-stage model of gaze biases To explain how these gaze biases come about, prior human studies have used modified sequential sampling models (aSSMs) that provide an increase in evidence accumulation for a given item while it is being viewed (see Krajbich, 2019, for review). This bias generally takes one of two forms: Some studies suggest that the gaze amplifies the value of attended targets so that the increase in evidence accumulation is proportional to the attended value (i.e. is multiplicative). Others suggest that gaze acts by providing a fixed increase in evidence accumulation to attended targets, independent of target values (i.e. is additive). Studies directly comparing these two mechanisms are not fully conclusive (Smith and Krajbich, 2019; Thomas et al., 2019), perpetuating debates as to which is 'correct'. A third, hybrid modeling framework offers a compromise (Manohar and Husain, 2013; Westbrook et al., 2020). Under this framework, a trial can be divided into two stages. The first stage is the decision stage, where evidence accrues for the decision offers, and gaze increases evidence accumulation for attended items in multiplicative fashion. The second stage begins once an internal decision boundary is reached (a choice is made) and continues until the choice is physically reported. During this interval the gaze becomes drawn to the to-be-chosen item, which is captured with an additive gaze bias. In other words, in the first stage, gaze influences choice, whereas in the second stage, the latent choice influences gaze. Two patterns in the monkeys' fixation data are consistent with this two-stage model. First, as shown in Figure 3, final fixation durations are longer than non-final fixations even after accounting for absolute order in the trial, suggesting some process that occurs at the end of each trial that prolongs fixation durations. Second, the monkeys' final fixations were on the to-be-chosen target on nearly every trial (~97% of trials), consistent with gaze being drawn to the to-be-chosen items. To more directly test the hybrid, two-stage, model in our data, we reasoned that the relative fit of a model with an additive gaze bias compared to one with a multiplicative gaze bias would indicate whether choice reflects one mechanism alone, or a mixture of both. Using the Gaze-Weighted Linear Accumulator Model (GLAM, Thomas et al., 2019), we found that while the multiplicative model provided a slightly better fit than the additive model, the difference was not significant (Figure 4, 0 on the x-axis, Supplementary file 3). This suggests that our data likely reflect a mixture of additive and multiplicative gaze biases, consistent with the hybrid model. Figure 4 Download asset Open asset Relative fit of multiplicative and additive models across NDT truncation values. Using an aSSM modeling framework, we computed the posterior probability (y-axis) of a model with either a multiplicative gaze bias parameter (black circles) or an additive parameter (white circles) being more likely to have generated the data. Probabilities (x-axis) were computed using data in which the final portion of the trial was truncated by 0ms (N = 30,046 trials), 50ms (N = 30,045 trials), 100ms (N = 30,

ABSTRACTThe primate orbitofrontal cortex (OFC) has long been recognized for its role in value-based decisions; however, the exact mechanism linking OFC value representations to decision outcomes has remained elusive. To address this question we show for the first time that trial-wise variability in choices can be explained by variability in value signals decoded from many simultaneously recorded OFC neurons. Mechanistically, this relationship is consistent with the projection of activity within a low-dimensional value-encoding subspace onto a potentially higher-dimensional, behaviorally-potent output subspace.

In economic decision-making individuals choose between items based on their perceived value. For both humans and nonhuman primates, these decisions are often carried out while shifting gaze between the available options. Recent studies in humans suggest that these shifts in gaze actively influence choice, manifesting as a bias in favor of the items that are viewed first, viewed last, or viewed for the overall longest duration in a given trial. This suggests a mechanism that links gaze behavior to the neural computations underlying value-based choices. In order to identify this mechanism, it is first necessary to develop and validate a suitable animal model of this behavior. To this end, we have created a novel value-based choice task for macaque monkeys that captures the essential features of the human paradigms in which gaze biases have been observed. Using this task, we identified gaze biases in the monkeys that were both qualitatively and quantitatively similar to those in humans. In addition, the monkeys' gaze biases were well-explained using a sequential sampling model framework previously used to describe gaze biases in humans-the first time this framework has been used to assess value-based decision mechanisms in nonhuman primates. Together, these findings suggest a common mechanism that can explain gaze-related choice biases across species, and open the way for mechanistic studies to identify the neural origins of this behavior.

Neuroeconomics seeks to explain how neural activity contributes to decision behavior. For value-based decisions, the primate orbitofrontal cortex (OFC) is thought to have a key role; however, the mechanism by which single OFC cells contribute to choices is still unclear. Here, we show for the first time a trial-to-trial relationship between choices and population-level value representations in OFC, defined by the weighted sum of activity from many individual value-coding neurons.

Rapid Eye Movement (REM) sleep has been shown to modulate the consolidation of fear memories, a process that may contribute to the development of Post-Traumatic Stress Disorder (PTSD). However, contradictory findings have been reported regarding the direction of this modulation and its differential effects on recall versus generalization. In two complementary experiments, we addressed this by employing sleep deprivation protocols together with a novel fear-conditioning paradigm that required the discrimination between coexisting threat and safety signals. Using skin conductance responses and functional imaging (fMRI), we found two opposing effects of REM sleep: While REM impaired recall of the original threat memories, it improved the ability to generalize these memories to novel situations that emphasized the discrimination between threat and safety signals. These results, as well as previous findings in healthy participants and patients diagnosed with PTSD, could be explained by the degree to which the balance between threat and safety signals for a given stimulus was predictive of threat. We suggest that this account can be integrated with contemporary theories of sleep and fear learning, such as the REM recalibration hypothesis.

Sleep, and particularly rapid eye movement sleep (REM), has been implicated in the modulation of neural activity following fear conditioning and extinction in both human and animal studies. It has long been presumed that such effects play a role in the formation and persistence of posttraumatic stress disorder, of which sleep impairments are a core feature. However, to date, few studies have thoroughly examined the potential effects of sleep prior to conditioning on subsequent acquisition of fear learning in humans. Furthermore, these studies have been restricted to analyzing the effects of a single night of sleep-thus assuming a state-like relationship between the two. In the current study, we used long-term mobile sleep monitoring and functional neuroimaging (fMRI) to explore whether trait-like variations in sleep patterns, measured in advance in both male and female participants, predict subsequent patterns of neural activity during fear learning. Our results indicate that higher baseline levels of REM sleep predict reduced fear-related activity in, and connectivity between, the hippocampus, amygdala and ventromedial PFC during conditioning. Additionally, skin conductance responses (SCRs) were weakly correlated to the activity in the amygdala. Conversely, there was no direct correlation between REM sleep and SCRs, indicating that REM may only modulate fear acquisition indirectly. In a follow-up experiment, we show that these results are replicable, though to a lesser extent, when measuring sleep over a single night just before conditioning. As such, baseline sleep parameters may be able to serve as biomarkers for resilience, or lack thereof, to trauma.SIGNIFICANCE STATEMENT Numerous studies over the past two decades have established a clear role of sleep in fear-learning processes. However, previous work has focused on the effects of sleep following fear acquisition, thus neglecting the potential effects of baseline sleep levels on the acquisition itself. The current study provides the first evidence in humans of such an effect. Specifically, the results of this study suggest that baseline rapid eye movement (REM) sleep may serve a protective function against enhanced fear encoding through the modulation of connectivity between the hippocampus, amygdala, and the ventromedial PFC. Building on this finding, baseline REM measurements may serve as a noninvasive biomarker for resilience to trauma or, conversely, to the potential development of posttraumatic stress disorder following trauma.

Human studies of sleep and cognition have established thatdifferent sleep stages contribute to distinct aspects of cognitive and emotional processing. However, since the majority of these findings are based on single-night studies, it is difficult to determine whether such effects arise due to individual, between-subject differences in sleep patterns, or from within-subject variations in sleep over time. In the current study, weinvestigated the longitudinal relationship between sleep patterns and cognitive performance by monitoring both in parallel, daily, for a week. Using two cognitive tasks - one assessing emotional reactivity to facial expressions and the other evaluating learning abilities in a probabilistic categorization task - we found that between-subjectdifferences in the average time spent in particular sleep stages predicted performance in these tasks far more than within-subject daily variations. Specifically, the typical time individualsspent in Rapid-Eye Movement (REM) sleep and Slow-Wave Sleep (SWS) was correlated to their characteristic measures of emotional reactivity, whereas the typical time spent in SWS and non-REM stages 1 and 2 was correlated to their success in category learning. These effects were maintained even when sleep properties werebased onbaseline measures taken prior to the experimental week. In contrast, within-subject daily variations in sleep patterns only contributed to overnight difference in one particular measure of emotional reactivity. Thus, we conclude that the effects of natural sleep onemotional cognition and categorylearning are more trait-dependent than state-dependent, and suggest ways to reconcile these results with previous findings in the literature.