Retention of Value Representations Across Time in People with Schizophrenia and Healthy Controls

Abstract

Background:

The current study aimed to further etiological understanding of the psychological mechanisms underlying negative symptoms in people with schizophrenia. Specifically, we tested whether negative symptom severity is associated with reduced retention of reward-related information over time, and thus a degraded ability to utilize such information to guide future action selection.

Methods:

44 patients with a diagnosis of schizophrenia or schizoaffective disorder and 28 healthy volunteers performed a probabilistic reinforcement-learning task involving stimulus pairs in which choices resulted in reward or in loss avoidance. Following training, participants indicated their valuation of learned stimuli in a test/transfer phase. The test/transfer phase was administered immediately following training and one-week later. Percent retention was defined as accuracy at weeklong delay divided by accuracy at immediate delay.

Results:

Both healthy controls and people with schizophrenia showed similarly robust retention of reinforcement learning over a one-week delay interval. However, in the schizophrenia group, negative symptom severity was associated with reduced retention of information regarding the value of actions across a weeklong interval. This pattern was particularly notable for stimuli associated with reward compared to loss avoidance.

Conclusions:

Our results show that, although individuals with schizophrenia may initially learn about rewarding aspects of their environment, such learning decays at a more rapid rate in patients with severe negative symptoms. Thus, previously learned reward-related information may be more difficult to access to guide future decision-making and to motivate action selection.

Introduction:

Negative symptoms such as abnormalities in motivation, are a core feature of schizophrenia (1). Such symptoms are both debilitating and resistant to current interventions (2). Thus, negative symptoms represent a critical and unmet clinical therapeutic target. In recent years, two lines of research have revealed insights into the psychological mechanisms underlying negative symptoms in schizophrenia and their associated neural correlates. First, people with schizophrenia show hedonic responses similar to controls when experiencing rewarding stimuli, but diminished responses, compared to controls (3-4), when anticipating future rewarding experiences (5-8). Thus, intact experiences of reward do not appear to have the expected impact on the subsequent initiation of goal-directed action. Second, people with schizophrenia have difficulty using the experience of reward receipt to generate and update mental representations of prospective value for actions (9-11). The results from both of these lines of research suggest difficulty using information about previous experiences to facilitate later decision-making and action initiation (12). In the current manuscript, we argue that difficulty using previously rewarding experiences to guide future behavior may be linked to poor long-term memory retention of positively valenced reward information in schizophrenia. Specifically, we test whether learning decays at a more rapid rate in patients with high negative symptom severity and thus is not fully available to guide future decision-making and behavior.

In attempting to understand how representations of value for particular actions are generated, modified, and utilized for future decision-making, researchers have relied on reinforcement learning (RL) frameworks (13). Generally, in RL models, values for actions are shaped by reward prediction errors (RPEs) – differences between expected and actual outcomes. RPEs have been closely tied to the activity of dopaminergic cells in the midbrain, with cells firing more rapidly for better-than-expected outcomes (positive RPEs) and ceasing to fire for worse-than-expected outcomes (negative RPEs) (14). Multiple studies have examined RL in schizophrenia (15-27,27-33). Results typically show that simple and/or implicit forms of RL are largely intact in schizophrenia (15), and that deficits emerge when learning is explicit and task conditions are difficult. Further, some (21,34-36) but not all (37-39) studies of RL have suggested impaired learning from rewarding stimuli but intact learning to avoid loss in schizophrenia. This dissociation provides an intriguing account of reduced goal-oriented behavior, such that individuals fail to seek rewarding aspects of their environment while also avoiding aspects associated with loss.

There has been remarkably little work in human subjects documenting the long-term retention of RL (40-42). This lack of evidence is surprising as, in real life, action preferences are often shaped by repeated outcomes across days, weeks, or months and the durability of this knowledge is basically presumed rather than actually being demonstrated. In one exception, Wimmer et al., conducted a study examining memory for value representations at a 3-week delay in healthy individuals (42). Results, at delay, showed greater memory for the value of trained stimuli when learning occurred in short, condensed sessions distributed over the course of weeks compared to a single “massed” training session. We are unaware of other studies documenting long-term retention of RL in healthy human subjects. In the schizophrenia literature, Herbener implemented an implicit preference conditioning task which associated visual patterns with particular frequencies of reward (41). People with schizophrenia and healthy controls both showed preference for the more frequently rewarded stimulus immediately following training. While controls maintained this preference at a 24-hour delay, people with schizophrenia did not, suggestive of difficulty in maintaining stimulus-reward relationships over time. Negative symptoms did not relate to delay performance, possibly due to the implicit nature of the learning paradigm. In other studies, individuals with schizophrenia have shown relatively intact performance on implicit RL paradigms and no systematic relationships between RL performance and negative symptom severity (15,43).

Several years ago, our group published a study examining probabilistic RL in schizophrenia (21). In a training phase, participants were presented with four separate pairs of stimuli, one at a time. Two of the stimulus pairs were associated with potential gain and the other two pairs were associated with avoiding a potential loss. Immediately following training, a test/transfer phase was administered. In this phase, participants were presented with both the original training pairs, as well as novel pairing (i.e., pairings of two stimuli, each from a separate training pair) and asked to choose the stimulus they preferred. Participants did not receive feedback in the test/transfer phase. People with schizophrenia with severe negative symptoms showed impaired learning of rewarding pairs compared to controls; however, the groups were similar in learning to avoid losses. Further, in the test/transfer phase, controls showed a clear preference for stimuli that frequently led to reward versus stimuli that frequently avoided loss, whereas people with schizophrenia did not. Finally, using a hybrid computational model with the ability to assess the relative contributions of actor-critic learning (thought to reflect basal-ganglia function) versus Q-learning (thought to reflect orbital frontal cortex function) to decision-making, results showed reduced reliance on Q-learning in people with schizophrenia. Taken together, these results suggested impairment in learning representations of positive value in schizophrenia, and difficulty using such information when making decisions.

Given the small number of studies assessing long-term retention of RL and the potential importance of value representation maintenance to negative symptoms in schizophrenia, we aimed to extend the results of Gold et al. (2012) by analyzing a previously unreported task condition of the original sample. Specifically, the goal of the current study was to examine whether schizophrenia and negative symptom severity are associated with a failure to retain information regarding the expected value of actions over time. To this end, we assessed preferences for task stimuli immediately following learning (reported in the original paper) and at a one-week delay (unreported). We hypothesized that that individuals with schizophrenia would show reduced retention of RL across a weeklong interval compared to controls. Further, we hypothesized that reductions in retention would be most pronounced among people with high negative symptom severity.

Methods:

Participants

Forty-four outpatients meeting DSM-IV criteria for schizophrenia (SZ) or schizoaffective disorder and 28 demographically similar healthy control (HC) subjects participated in the study. SZ subjects had been on a stable medication regimen for at least 4 weeks and were considered clinically-stable at the time of testing. Patients were recruited from the Maryland Psychiatric Research Center and from local outpatient psychiatric clinics.

The HC subjects were recruited from the community via random phone number dialing, Internet and newspaper advertisements, and word of mouth among recruited participants. They had no current Axis I or Axis II diagnoses, as established by the Structured Clinical Interview for DSM-IV-Axis I Disorders (44), and Structured Interview for DSM-IV Personality (45). Control subjects also reported no family history of psychosis and were not taking psychotropic medications. All participants had no history of significant neurological injury or disease and reported no significant medical or substance use disorders. Participants provided informed consent for a protocol approved by the University of Maryland Institutional Review Board.

Symptom and Neuropsychological Assessment:

Negative symptoms were assessed in participants with schizophrenia using a 22-item version of the Scale for the Assessment of Negative Symptoms (SANS) (46). This version omits Attention items, which have been shown to relate more to cognition than negative symptoms (47). Items from the SANS were summed to create an overall score (SANS Total). From the SANS, we also extracted factors assessing the severity of experiential (asociality, anhedonia, and avolition) and expressive (blunted affect and alogia) negative symptoms (21). We report both SANS total, as well as correlations between experiential and expressive symptom domains. The Brief Psychiatric Rating Scale (BPRS) was administered to participants with schizophrenia in order to determine severity of positive symptoms and global expression of symptomatology (48).

Both HC and schizophrenia participants completed measures of word reading (Wide Range Achievement Test – 4^th Edition) and the MATRICS (Measurement and Treatment Research to Improve Cognition in Schizophrenia Consensus Cognitive) battery (49). The groups did not significantly differ in age, sex, ethnicity, or parental education. The SZ group reported significantly less personal education than the HC group (Table 1). Symptom severity, medication information, and neuropsychological test scores are also listed in Table 1.

Table 1:

Demographic, Symptom, and Neuropsychological Variables

Variable	HC Group (n= 28)	SZ Group (n = 44)	Test Statistic	p-value
Demographic and Clinical Characteristics
Age, mean (SD)	41.2 (9.4)	41.8 (10.0)	t = −0.3	0.8
Education, mean (SD)
Participant	14.8 (1.9)	12.3 (1.7)	t = 5.9	<0.01
Maternal	13.1 (2.2)	13.2 (3.0)	t = −0.1	0.92
Paternal	13.5 (2.7)	13.3 (3.4)	t = 0.3	0.59
Sex, No.
Male	17	32	χ2 = 1.14	0.31
Female	11	12
Race/Ethnicity, No.
Black	14	21	χ2 = 0.04	0.85
White	14	23
Current Prescribed Psychiatric Medications
Antipsychotics No.
Atypical		30
Typical		8
Typical and Atypical		6
Antidepressant No.		11
Mood Stabilizers No.		7
Anxiolytic No.		5
Anticholinergic No.		4
Neuropsychological Test and Symptom Data
Clinical Rating Scale, mean (SD)
BPRS total		65.0 (37.5)
BPRS positive cluster		10.0 (4.8)
SANS Total		36.6 (15.9)
Neuropsychology score
WRAT	104.5 (14.5)	96.7 (13.8)	t = 2.3	0.03
WTAR	105.5 (13.8)	97.2 (13.1)	t = 2.6	0.01
MATRICS Composite	48.5 (12.1)	29.4 (12.8)	t = 6.3	< 0.01
MATRICS visual memory	46.8 (13.5)	36.9 (14.5)	t = 2.9	0.01

Reinforcement-Learning Task:

Training Phase

Participants completed a probabilistic incentive learning task (21). During this task, participants were presented with 4 pairs of landscape stimuli, one pair at a time. Two of the pairs were associated with potential gains; the optimal response was reinforced on 90% of the trials for one pair and 80% of the trials for the other pair. Two of the pairs involved learning to avoid losses; the optimal response avoided losing on 90% of the trials for one pair and 80% of the trials for the other pair. All participants completed a brief 12-trial practice session to ensure task comprehension, followed by 160 learning trials (each pair presented 40 times) with all pair types presented in a randomized order.

Test/Transfer Phase

A test/transfer phase was administered both immediately following the learning phase and at approximately one-week delay (mean = 7.14 days, SD = 0.84). In these 64 trials, the original 4 training pairs were each presented 4 times (16 total trials), and 24 novel pairings were each presented twice (48 total trials). For novel pairings, each stimulus from the training phase was presented with every other stimulus. No feedback was administered during the test/transfer phase. Instead, participants were instructed to pick the item in the pair that they thought was “best” based on their earlier learning. With exception of separate randomization of trial presentation, structure of the immediate and weeklong test/transfer phases was identical. Participants were not told during the initial study visit that they would be asked to repeat aspects of the task one-week later. A manuscript describing the learning and immediate test/transfer phase has already been published (21). The current report focused on percent retention at one-week delay, although we do report broad measures from the training phase (e.g., total accuracy during learning) in order to better characterize retention effects.

For analysis purposes, specific individual pairings in the test/transfer phase were grouped based on whether they appeared in the training phase and according to the frequency and types of outcomes with which individual stimuli were associated in the training phase. The purpose of these groupings was to reduce noise within the data by increasing the number of trials within a condition. First, presentations of the 4 original training pairs (16 presentations) were averaged, creating a single measure of learning for test item pairings (test pairs). Second, presentations where an optimal stimulus (i.e., 80% or 90% winner, 80% or 90% loss avoider) was paired with a suboptimal stimulus (i.e., 20% or 10% winner, 20% or 10% loss avoider) that had been presented with a different stimulus during the initial training phase were averaged, creating a single measure of generalization of trained items to novel decision-making contexts (transfer pairs). Given previous evidence of intact learning from losses but impaired learning from gains in schizophrenia (21, 34-36), we conducted follow-up analyses of stimuli associated with gain and loss avoidance separately.

Statistical Analysis:

Diagnostic group differences were assessed using an independent samples t-test with percent retention (weeklong delay accuracy divided by immediate delay accuracy) as a dependent variable. Analyses examining effects of negative symptoms were conducted both dichotomously (as we have done previously (21)) and continuously. For dichotomous analyses, we conducted an ANOVA with a 3-level between-subjects factor of group (HC, Low-Neg, High Neg) with SZ groups separated based on a median split of the experiential items of the SANS (21) and a dependent variable of percent retention. For continuous analyses, we conducted spearman rank-order correlations between percent retention and symptom severity measures. Follow-up analyses were conducted examining gain and loss avoidance pairings separately. We also conducted a hierarchical logistic regression in order to determine whether choice behavior at weeklong delay was predicted by the difference between frequencies of selecting particular stimuli during training or the difference between the experienced values of stimuli during training (supplement). In this analysis, the dependent variable was performance at one-week delay in order to observe the precise type of information that was retained at delay. Finally, we conducted follow-up analyses excluding individuals whose accuracy measures were not significantly above chance at immediate delay (supplement).

Results:

Figure 1 shows test/transfer phase performance at both immediate and one-week delay. Both groups showed surprisingly high levels of retention for previously optimal stimuli at weeklong delay. Unexpectedly, test-pair percent retention, relative to immediate performance, was similar for the HC and SZ groups (88.2% for HCs; 86.2% for the SZ group; t(70) = 0.52, p = 0.61; Figure 1A). Percent retention for transfer pairs was also similar between groups (t(70) = −0.25, p = 0.81; Figure 1B). Results were similar when using a 3-level group factor (after splitting the patient group according to negative symptom severity), suggesting that retention did not differ between HCs, low negative symptom patients, and high negative symptom patients when analyzed dichotomously (supplement).

Figure 1: Accuracy at immediate and weeklong delay for trained and transfer pairs.

Error bars represent standard error of the mean.

Figure 2 shows the relationship between continuous negative symptom severity (SANS Total) and percent retention. Given the non-normal distribution of study variables spearman rank-order correlations were performed. For test pairs, negative symptom severity was significantly associated with percent retention (Figure 2A; rho = −0.38, p = 0.01), such that individuals with more severe negative symptoms retained less information over time. This effect held when controlling for severity of positive symptoms. The effect was in the same direction but was not significant for transfer pairs (Figure 2D; rho = −0.14; p = 0.38). Associations between positive symptoms, aggregate symptom severity, and percent retention were not significant (p-values > 0.07), suggesting a relationship specifically with negative symptoms (supplement). Negative symptom effects were similar to the aforementioned results when separated into experiential and expressive subdomains, suggesting relations between retention and general negative symptom severity (supplement). When restricting the sample to participants who performed significantly above chance at immediate delay, associations between test-pair retention and negative symptoms was similar to the overall sample but failed to reach significance (supplement).

Figure 2: Scatterplots illustrating associations between information loss and negative symptom severity.

Note removal of outliers does not significantly alter the significance of associations

Effects of Valence - Gain vs. Loss Avoidance

Because several previous studies (including our own) have reported impaired learning from rewarding outcomes but intact learning to avoid losses in people with schizophrenia (21,34-36), we conducted follow-up analyses examining effects of feedback with respect to the negative symptom associations observed above (Figure 3). For test pairs (Figure 2B & 2E), associations between negative symptom severity and retention were similar for gain and loss avoidance (Gain: rho = −0.31; p = 0.04; Loss avoidance: rho = −0.33; p = 0.03; William’s t-test = −0.14, p = 0.89). For transfer pairs (Figure 2C & 2F), associations between negative symptom severity and retention were stronger for gain compared to loss at a trend level (Gain: rho = −0.37; p = 0.01; Loss avoidance: rho = −0.02; p = 0.91; William’s test = −1.8, p = 0.074). Negative symptom effects were similar experiential and expressive subdomains (supplement).

Figure 3: Accuracy at immediate and weeklong delay, separated by feedback type, for test and transfer pairs.

Error bars represent standard error of the mean.

Correlations with Cognition and Learning Phase Task Performance

We addressed contributions of cognitive impairment and learning phase accuracy in two ways. First, we conducted spearman correlations between percent retention and 1) standardized neurocognitive measures (MATRICS Composite Score, WRAT4 Standard Score), 2) performance during the training phase (total accuracy). These analyses revealed that neurocognitive measures and training phase performance were not significantly correlated with percent retention (Table 2). However, training phase performance, immediate delay accuracy, and weeklong delay accuracy were significantly correlated (supplement). Second, we conducted a multiple regression analysis including immediate test/transfer phase accuracy, training phase accuracy, MATRICS Composite score, WRAT4, and SANS Total Score as simultaneous predictors of percent retention. In these analyses, negative symptoms remained a significant predictor of percent retention (Table 3). Correlations between negative symptoms and percent retention remained significant when including equivalent dose as a covariate (supplement).

Table 2:

Spearman Correlations between Training Phase Performance, Neurocognitive measures, and percent retention.

	Healthy Control
	Test Pairs						Transfer Pairs
	All		Gain		Loss		All		Gain		Loss
	rho	p	rho	p	rho	p	rho	p	rho	p	rho	p
LEARNING ACC	−0.26	0.19	−0.05	0.81	−0.25	0.20	0.07	0.74	0.08	0.69	−0.15	0.44
MATRICS TOTAL	0.03	0.88	0.16	0.43	0.09	0.65	0.01	0.95	0.12	0.56	−0.10	0.60
WRAT4SS	−0.23	0.24	−0.14	0.49	−0.05	0.80	−0.18	0.35	0.04	0.83	−0.23	0.23
	Schizophrenia
	Test Pairs						Transfer Pairs
	All		Gain		Loss		All		Gain		Loss
	rho	p	rho	p	rho	p	rho	p	rho	p	rho	p
LEARNING ACC	−0.24	0.13	−0.12	0.44	−0.28	0.07	−0.07	0.65	−0.05	0.76	−0.10	0.53
MATRICS TOTAL	−0.05	0.74	−0.01	0.94	−0.13	0.43	−0.01	0.93	0.13	0.39	−0.12	0.45
WRAT4SS	0.16	0.31	0.21	0.17	0.03	0.88	0.17	0.28	0.12	0.46	0.13	0.42

Table 3:

Multiple Regression Analyses predicting long delay performance from negative symptoms, learning phase performance, immediate delay performance, and neurocognitive measures.

ALL TEST PAIRS (DEPENDENT VARIABLE PERCENT RETENTION)
	Beta	SE	t-value	p-value
SANS TOTAL	−0.008	0.003	−2.255	0.03
IMMEDIATE DELAY	−0.002	0.005	−0.296	0.769
MATRICS OVERALL	0.001	0.004	0.191	0.85
WRAT4SS	−0.152	0.308	−0.495	0.624
LEARNING PHASE ACCURACY	0.373	0.469	0.796	0.431
GAIN TRANSFER PAIRS (DEPENDENT VARIABLE PERCENT RETENTION)
	Beta	SE	t-value	p-value
SANS TOTAL	−0.009	0.003	−3.017	0.005
MATRICS OVERALL	−0.003	0.004	−0.629	0.533
WRAT4SS	0.003	0.004	0.71	0.482
IMMEDIATE DELAY	−0.903	0.367	−2.461	0.019
LEARNING PHASE ACCURACY	0.525	0.544	0.964	0.341

Experienced Value vs. Frequency of Choice

Because retention of value-based choices over long delays has not been extensively studied, and the above analyses do not indicate exactly what was retained, it is plausible that rather than retaining the learned value of each stimulus, participants simply remembered which stimulus they had selected most often during the initial training session and used that information to make decisions. Such stimulus-response learning would facilitate appropriate generalization in most cases but is nevertheless dissociable from learned values, given individual participant histories of choices and outcomes. Indeed, logistic regression analysis confirmed that both prior frequency of selection and experienced expected values were independent predictors of week-long transfer performance (supplement). Notably, however, negative symptoms were parametrically and selectively related to decreased reliance on expected value to make decisions at delay (Figure 4). This effect was similar for expressive and experiential negative symptoms.

Figure 4: Correlations between Experienced Value and Frequency and Negative Symptoms:

Individual dots represent the effect of experienced value (left) and frequency (right) on choice behavior at weeklong delay for each individual subject. Results showed that the effect of experienced value, but not frequency, varied parametrically as a function of negative symptom severity. Thus, individuals with high negative symptom severity do not appear to be utilizing experienced value information, to the same degree as healthy controls, when making decisions at weeklong delay.

Discussion:

We document several important findings from both basic and clinical science perspectives. First, we show robust percent retention of RL over a one-week delay interval. This was true across groups. While strong long-term retention of action-outcome contingencies is well-documented (50-52), particularly in the motor and procedural learning literatures, remarkably little evidence of such effects has been provided using choice-based, RL paradigms in human participants (42). The current result addresses this lacuna in the literature, demonstrating that individuals not only retain stimulus-response contingencies, but also mental representations of value for particular stimuli.

From a clinical perspective, we show that individual differences in percent retention, for rewarding stimuli, are related to negative symptom severity. Such findings are consistent with recent theories suggesting that people with schizophrenia show difficulty in using information about previously rewarding experiences to facilitate future decision-making (3,9,10,12). Specifically, we provide preliminary evidence for a hypothesis that these difficulties might arise, in part, due to a failure to retain high-fidelity representations of the value of actions across time, and thus a degraded ability to utilize such representations to guide future action selection. Finally, it is striking that we did not observe significant diagnostic group differences. We consider this noteworthy as people with schizophrenia typically perform more poorly than HCs across most demanding cognitive tasks. Thus, our failure to find between-group effects was unexpected.

As stated in the introduction, the current work was largely motivated by Gold et al., 2012 who found that negative symptoms were associated with deficit representations of positive expected value (21). Our manuscript extends this finding by showing that high negative symptom patients have difficulties maintaining reward-related information over the course of a week. Further, we show that people with high negative symptom severity are less likely to use expected value to make decisions at weeklong delay, similar to findings of reduced reliance on Q-learning (thought to be driven by orbital frontal cortex function) in schizophrenia. Despite these consistencies with prior results, some discrepancies must also be noted. First, while we observed significant correlations between negative symptom severity and percent retention using continuous measures, we failed to find significant relationships between retention and negative symptom severity when using a dichotomous approach. Second, while Gold et al., 2012 reported correlations between RL measures and experiential negative symptoms, the current analyses reveal associations between negative symptom severity and percent retention of RL more broadly with relatively equivalent correlations for both experiential and expressive domains.

The current findings are also partly consistent with Herbener (2009) demonstrating reduced long-term retention of preferences for implicitly learned stimuli at a 24-hour delay in schizophrenia. While we failed to observe such group differences, we did find associations between negative symptom severity and percent retention of explicitly-learned preferences. The observed differences between our results and the findings of Herbener may be due to considerable differences in task design. First, Herbener assessed implicit RL (i.e., participants were not aware of associations between stimuli and probability of reward receipt), whereas our design assessed explicit RL. Importantly, implicit RL has not been shown to be associated with negative symptoms in previous reports (43, 53). Second, feedback in the Herbener paradigm was predetermined regardless of the participant’s choices, whereas feedback in our design depended on participant choice. Third, Herbener assessed retention at a 24-hour delay, whereas we assessed retention at one-week delay.

The current manuscript is also partially consistent with an older literature examining emotional memory in schizophrenia (54). Although the findings from this literature are mixed, studies assessing memory at longer (>24 hrs) delay (41, 55-58) report deficits in retention of emotional information for people with schizophrenia compared to HCs more consistently than those assessed at shorter delays (56, 59-62). Several of these reports have shown enhanced recall of negatively-valenced stimuli than for positively-valenced stimuli in SZ (54). Such findings are consistent with the current result of poor percent retention of gain information in individuals with severe negative symptoms.

Despite the large body of RL research in human subjects, there has been remarkably little work regarding how value representations are retained beyond a single testing session (40-42,55-58). The current report extends such findings by suggesting that representations of value for learned stimuli are retained at very high levels, over the course of a weeklong delay. While the current findings are behavioral, it is important to consider the potential neural substrates that may be underlying reduced retention of value representations in high negative symptom patients. We might speculate, given previous evidence of degraded expected value representations in people with schizophrenia with severe negative symptoms (21), that reduced top-down frontal communication with the basal ganglia could account for reduced retention. Consistent with this account, high negative symptom patients showed a weaker effect of expected value on decision-making at one-week delay. However, reduced retention could reflect deficit dopamine-dependent mechanisms in the basal ganglia and frontal cortex, memory encoding mechanisms within the hippocampus, or an interaction among frontal, hippocampal, and striatal systems. Interestingly, a recent study in healthy participants found that striatal RPE signaling strengthens formation of episodic memories, suggesting interplay between striatal and hippocampal systems during RL (63). Thus, future clinical studies may benefit from leveraging work in the basic sciences that attempts to link RL with memory consolidation (63-66).

Another important direction for future research would be to determine the potential roles of incentive salience in modulating observed effects. In the current study, participants were not told that they would repeat portions of the experimental task at a one-week delay. Recent work has suggested that the relative contributions of various RL systems to decision-making varies depending on incentive contexts (67). Further, individual differences in relationships between these incentive effects and trait dimensions relevant to psychopathology have been noted (68). It will be important for future work to determine whether the retention effects seen in the current study may be modulated by incentive context in healthy populations, and if such modulation may vary with aspects of psychopathology

Limitations

Our sample size was modest. Future work will be needed to replicate the current findings in a larger sample. Second, the SANS was used to assess negative symptoms. Newer measures are available that better reflect the field’s current conceptualization of negative symptoms (69,70). We were unable to use such measures as the data presented in the current study were collected prior to the development of these measures. Third, the current RL paradigm did not include a non-reinforced, “neutral”, learning pair. Thus, we are unable to fully account for the extent to which the current findings are representative of general retention difficulties versus difficulties unique to learning and retaining rewards. Fourth, all participants with schizophrenia were taking anti-psychotic medications at the time of study completion, which may have influenced choice behavior due to influence on dopamine systems.

Summary

Our results show that, although individuals with schizophrenia may initially learn about rewarding actions, such learning decays at a more rapid rate in patients with severe negative symptoms and may be more difficult to access to guide future decision-making and to motivate action selection. Further work is necessary to replicate these findings and examine the interaction between RL and memory processes to gain a better mechanistic account of reduced maintenance of value representations in people with schizophrenia.