Range-Adaptive Value Representation in Different Stages of Schizophrenia: A Proof of Concept Study

Abstract

Amotivation is related to value representation. A comprehensive account of amotivation requires a mechanistic understanding of how the brain exploits external information to represent value. To achieve maximal value discriminability, brain valuation system will dynamically adapt its coding sensitivity to the range of values available in any given condition, so-called range adaptive coding. We administered an experimental task to 30 patients with chronic schizophrenia (C-SCZ), 30 first-episode schizophrenia (FE-SCZ), 34 individuals with high social anhedonia (HSoA), and their paired controls to assess range adaptation ability. C-SCZ patients exhibited over-adaptation and their performances were negatively correlated with avolition symptoms and positive symptoms and positively correlated with blunted-affect symptoms and self-reported consummatory interpersonal pleasure scores, though the results were non-significant. FE-SCZ patients exhibited reduced adaptation, which was significantly and negatively correlated with avolition symptoms and positively correlated with the overall proportion of choosing to exert more effort. Although HSoA participants exhibited comparable range adaptation to controls, their performances were significantly and negatively correlated with the proportion of choosing to exert more effort under the lowest value condition. Our results suggest that different stages of schizophrenia spectrum showed distinct range adaptation patterns. Range adaptation impairments may index a possible underlying mechanism for amotivation symptoms in FE-SCZ and more complicated and pervasive effects on clinical symptoms in C-SCZ.

Introduction

Negative symptoms of schizophrenia (SCZ) are closely related to poor functional outcomes,^1–3 and therefore constitute major unmet therapeutic needs. From both the theoretical and empirical perspectives, amotivation has been proposed to be the core domain of negative symptoms.^4–6 However, little is known regarding the pathophysiology of amotivation, which becomes a major barrier to effective treatments for negative symptoms.

Amotivation refers to the reduced volition to initiate or maintain goal-directed behavior, and such notion has promoted research focusing on reward processing. Reward processing involves two core components, ie, wanting and liking,⁷ and both involve representation of value.^8,9 The accuracy of the value representation plays a key role in motivating goal-directed behavior. Besides, implicit or explicit decision-making, based on the represented value, is the fundamental cognitive operation essential for motivated behavior. Representation of value precedes and influences decision-making,^10–12 but its translation into motivated behavior could only be mediated through decision-making processes.^9,13 In other words, only when the differentiation of different represented values is sufficiently strong, value-based decision-making can be made, so as to motivate the pursuit of the most suitable behavior. The ability to represent values of different possible actions “discriminatingly” is, therefore, a fundamental process underpinning motivated behavior.

Evidence has shown that the value signals are represented using an adaptive/normalized code, which promotes value discrimination.^14–17 Theoretically, the neural response for a given value would always be the same under an absolute coding system. However, under an adaptive coding system, the neural response for a given value could vary in different situations, depending on its relative position among other different represented values (ie, among a range of values). This particular type of adaptive coding is termed “range adaptive coding” or “range adaptation”.¹⁶ In research on sensory processing, adaptation implicates adjustments and re-adjustments of neuronal sensitivity to the range of sensory inputs.^18,19 Consequently, the slope as a function indicating neuronal responses would vary with the input range. It would be steeper in settings of smaller ranges, and flatter in settings of larger ranges. Adaptation to the statistics of sensory stimuli (such as range and mean) could improve sensory representation discrimination and enhance the effectiveness of the neural operating range.^20,21 For example, the human retina adapts to the mean luminance of the visual stimuli, and such adaptation improves predictive coding in the new context.²²

Recently, this type of adaptation has been explored in the field of value-based decision-making. Range adaptation could lead to maximum value discrimination within each range and thus the optimal coding efficiency.^23,24 Range adaptation to prediction errors also improves adaptive behavior and promotes efficient learning.^25–27 Conversely, impaired range adaptation could cause dysregulated value discrimination and undermine motivation.²⁸ The deep neuronal basis of range adaptive coding may implicate the cellular basis of the psychopathology of negative symptoms and inform future developments of early identification and treatments of emerging negative symptoms.

The association between SCZ, negative symptom, and range adaptation has been studied. Empirical findings suggest that dopamine modulates the operation of range adaption to prediction error signals in the human midbrain and striatum.²⁹ Dopamine may mediate the association between range adaptation and SCZ. Importantly, another study found an impaired range adaptation to prediction error signals in SCZ patients, and showed that such range adaption is correlated with levels of schizotypal traits in the health population.²⁷ Moreover, two fMRI studies directly assessed the capacity of range adaptation to represented values among SCZ patients and individuals with high schizotypal traits,^30,31 and found an attenuated range adaptation in the right caudate during representations of the outcome values in both SCZ and schizotypal samples. In SCZ patients, this attenuated range adaptive coding signal of the right caudate was negatively correlated with negative symptoms. Although previous neuroimaging evidence^30,31 for neural substrates of range adaptive coding are very insightful, behavioral data is much needed for bridging the distant link between neural mechanisms and clinical manifestations. Using behavioral paradigms which are convenient to administer, research on range adaptive coding could bring scientific contributions by (1) ascertaining replicability of the proposed proposition^30,31 at different levels of the brain-behavior-symptoms complex, (2) utilizing samples larger than previous neuroimaging studies,^30,31 and (3) extending such proposition to different subclinical and clinical cohorts along the SCZ spectrum which are more specifically related to transdiagnostic negative symptoms.

Therefore, this study administered the behavioral paradigm of the effort-based pleasure experience task (E-pet)³² to investigate range adaptation, and valid instruments to thoroughly measure the symptom of amotivation.^32–35 To investigate possible evolutions of range adaptation impairments along the course of SCZ, and to follow the SCZ spectrum approach, we utilized samples at different stages of the spectrum, spanning from people with high social anhedonia (HSoA) to first-episode SCZ (FE-SCZ) to chronic SCZ(C-SCZ). This study aimed to test the range adaptation capacity and examined its relationship with the multi-facet construct of motivation in the SCZ spectrum. We hypothesized that clinical and subclinical individuals would have impaired range adaptation compared with controls, and range adaptation impairments would be correlated with levels of amotivation.

Method

Participants

This study recruited three different samples. Sample 1 comprised 30 patients with C-SCZ (duration of illness more than 10 years) and 30 matched healthy controls (C-Control). Sample 2 comprised 30 patients with FE-SCZ and 30 matched healthy controls (FE-Control). The operational criteria for FE-SCZ are shown in Supplementary Materials. Clinical participants were recruited from clinics at Castle Peak Hospital, and controls were recruited from the neighboring community.

Sample 3 comprised 34 individuals with HSoA and 34 individuals with low levels of social anhedonia as controls (HSoA-Control), recruited from a pool of 119 university students in Beijing and Shanghai, based on their ratings on the Chapman Social Anhedonia Scale (RCSAS).^36,37 The students completed the RCSAS online, and those who scored >=17 on the RCSAS were recruited as participants with HSoA, whereas those who scored <=7 were recruited as HSoA-control. The cut-off scores were defined based on our previous study.³⁶

For all samples, the exclusion criteria included a history of head injury, drug or alcohol abuse. For controls, there should be no history or family history of mental disorder.

Measures

The E-pet.

The E-pet paradigm has been detailed elsewhere.³⁵ In brief, the E-pet comprised three different phases, ie, decision-making, anticipatory pleasure, and consummatory pleasure (see figure 1). Participants were first asked to choose between a high-effort (grip level: 75% of the strongest grip) and a low-effort task (grip level: 25% of the strongest grip), based on the offered reward magnitude and probability. The E-pet set the minimum efforts required for completion of the low-effort and high-effort tasks on a participant-by-participant basis. In the low-effort tasks, the reward value range was fixed to¥0.5–¥1. The high-effort task has two potential value ranges of ¥1.5–¥2.4 and¥5–¥8. The more effort the participants exerted, the higher outcome value they would receive given a certain probability. During the consummatory pleasure rating phase, participants were presented with reward value they have earned, and they rated their pleasantness on a 9-point Likert scale. In this study, we mainly focused on the consummatory ratings since it measures the participants’ subjective valuation of the outcome value and thus depends on the range adaptation of value.

Procedure.

Participants first completed the ACIPS, then undertook the PANSS and the SANS (only for clinical participants) (see Supplementary Method), and lastly the E-pet. Lower scores of the ACIPS indicate greater impairment to experience interpersonal pleasure, whilst lower scores of the SANS and PANSS indicate less severe symptoms. Written informed consent was obtained from each participant. The study was approved by the Ethics Committees of the Institute of Psychology, the Chinese Academy of Sciences (H16015, H15042), and the New Territories West Cluster Clinical and Research Ethics Committee (NTWC/CREC/15065).

Data Analysis.

At the neural level, dependent variable for range adaptive coding is the neuronal firing rate.³⁸ Range adaptive coding is operationally defined as the slope inversely proportional to value ranges,³⁸ ie, the slope decreases as the range enlarges (see figure 2a). Thus, we quantified the degree of adaptation by taking the difference between the response slopes of the two different reward ranges for high-effort task (slope narrow—slope wide). The similar methods have been used in other studies, albeit in the context of fMRI data analysis.^30,31

In this process, we first filtered the data by picking out the trials where high-effort task was chosen by each participant during the decision-making phase. Within these chosen high-effort trials, there would be two different value ranges (narrow and wide). Then, we conducted a linear regression model with the consummatory rating scores in these trials as the dependent variable, on a participant-by-participant basis. In the regression model, three predictors were involved. The first two predictors were the predefined value range (narrow or wide) in each trial and the actual outcome value received in each trial (determined by both the value range and the participants’ actual exerted effort level). The last predictor was the interaction variable of “value range x outcome value.” By including this interaction variable, we tried to get a numerical index of the estimated difference between the two response slopes of two value ranges (see Supplementary Methods). The regression coefficients (∆beta value) for the interaction variable were taken as a proxy index reflecting the difference in response slopes between two value ranges. The validity of using ∆beta value as measurements of range adaption is well-supported (see Supplementary Methods and Results).

For estimation of the slopes at the individual level, poor fitting might occur. To filter the data, participants who received less than two different outcome values within either the narrow or the wide range were excluded. Besides, participants whose standard errors of the estimated ∆beta value were more than 2.5 standard deviations from their group means were excluded, resulting in 24 C-SCZ vs 29 C-controls, 28 FE-SCZ vs 28 FE-controls, and 34 HSoA vs 34 HSoA-controls.

For group difference analysis, each participant’s ∆beta value was entered into an independent-sample t-test to compare the group difference. We also directly compared and plotted the ∆beta values of all the six groups.

Then, the ∆beta values were correlated with participants’ clinical symptoms (ie, all the SANS subscales scores and the PANSS subscale scores). In particular, instead of using the PANSS negative subscale, we examined the correlation between ∆beta values and the PANSS “amotivation factor” based on the factor structure found by Khan et al, 2017³⁹. The correlations of ∆beta values with the ACIPS subscale scores, the proportion of choosing high-effort task in E-pet, and the antipsychotic medication dosage in chlorpromazine equivalence were also examined. The correlation analyses were first conducted in the C-SCZ, FE-SCZ, and HSoA cohorts separately. Then, defying the diagnostic boundaries,⁴⁰ we merged the three cohorts for additional correlation analyses. Notably, it would be misleading to conclude that the results of two separate tests are comparatively different, merely based on the observation that one result is significant but the other is not.^41,42 Therefore, we directly contrasted the correlations between range adaptation and SANS avolition symptom with those between range adaptation and other clinical symptoms (see Supplementary Materials). The FDR corrections were applied to the correlational analyses.

Results

Demographic Information

Table 1 summarizes the demographics, self-report anhedonia traits, and psychopathology of the SCZ samples and respective controls. The results did not show any significant differences in demographics and estimated IQ between patients and controls in Samples 1 and 2. Details of medication information are shown in Supplementary Materials. Table 2 summarizes the demographics and self-reported anhedonia traits of Sample 3, and the results showed that participants with HSoA and HSoA-controls did not differ in demographics and estimated IQ.

Range Adaptation Performance Among Individuals at Different Stages of SCZ and Controls

For group difference analysis, there was a significant group effect in ∆beta value between C-SCZ and C-controls (t_37.911= 3.424, p = .001, Cohen’s d = 0.964, 95% Confidence Interval (CI) = [0.287, 1.115], see figures 2b and 2c). The result showed a heightened range adaptation among the C-SCZ group, suggesting over-adaptation. As the value range became wider, the extents of slope decreases were larger in C-SCZ than C-controls.

The group difference in ∆beta value between FE-SCZ and FE-controls was also significant (t₅₄ = -2.066, p = .044, Cohen’s d = -0.551, CI = [–0.654, –0.010], see figures 2d and e), indicating that the FE-SCZ group showed a significantly reduced range adaptation for the outcome value, ie, a sign of insufficient adaptation. The ∆beta values were comparable between the HSoA and HSoA-control groups (t₆₆= –0.277, p = .783, Cohen’s d = 0.067, CI = [–0.388, 0.293]).

The exploratory analysis on the three clinical and subclinical samples found a significantly group effect in ∆beta values, with the direction of FE-SCZ < HSoA < C-SCZ (F_2,83 = 15.833, p < .001, Partial η² = 0.276).The ∆beta values derived from all the six groups were plotted in Figure S2 to detail the range adaptation performance in all the present samples.

Relationship of Range Adaptation Performance with Negative Symptoms, Other Clinical and Subclinical Symptoms, and Medications

For C-SCZ participants, we found non-significant correlations of range adaptation with the SANS avolition symptom (r₂₄ = –0.248, p = .242, CI = [–0.592, 0.172], p_{FDR corrected} = 0.487), PANSS positive symptom (r₂₄ = –0.271, p = .199, CI = [–0.608, 0.148], p_{FDR corrected} = 0.487), SANS blunted-affect symptom (r₂₄ = 0.325, p = .121, CI = [–0.090, 0.644], p_{FDR corrected} = 0.487) and ACIPS consummatory subscale score (r₂₄ = 0.340, p = .105, CI = [-0.074, 0.653], p_{FDR corrected} = 0.487, see figure 3a and Supplementary table S1 for details).

For FE-SCZ participants, we found significant correlations of the ∆beta value with the SANS avolition symptom (r₂₈ = –0.387, p = .042, CI = [–0.664, –0.016], p_{FDR corrected} = .294, see figure 3b) and the total proportion of choosing the high-effort task in E-pet (r₂₈ = 0.376, p = .049, CI = [0.003, 0.657], p_{FDR corrected} = .294, see figure 3c).

To contrast the correlations between the ∆beta value and avolition symptoms with those between the ∆beta value and other clinical symptoms, we conducted an exploratory analysis to test whether the relationship between range adaptation and amotivation symptoms would be relatively specific. We found that the correlation of ∆beta value with the SANS avolition score was significantly stronger than its correlations with the other SANS subscales and the PANSS general subscale scores in FE-SCZ participants (see Supplementary table S2). However, for C-SCZ participants, the correlation of ∆beta value with the SANS avolition symptom was comparable to its correlations with the PANSS positive symptom, the PANSS general psychopathology, and the SANS alogia symptom (see Supplementary table S2).

Then, we pooled the data from the FE-SCZ and C-SCZ samples together and repeated the correlation analyses, while entering the duration of illness and the medication dose as covariates. The correlations of ∆beta value with the SANS avolition subscale score (r₄₈ = –0.321, p = .023, CI = [–0.528, –0.041], p_{FDR corrected} = .253) and the ACIPS consummatory subscale score (r₄₈ = 0.282, p = .048, CI = [0.002, 0.406], p_{FDR corrected} = .264) were significant (see Supplementary figures S1a and S1b). When we merged the clinical and subclinical samples together, we found a significant negative relationship between ∆beta value and the willingness of choosing high-effort task under the highest rewarded condition(r₈₆ =-0.259, p =.016, CI=[0.001,0.416]). Moreover, the result remained significant when we merged all the six present samples including the healthy controls (see Supplementary Results).

Among the HSoA group, a significant correlation between ∆beta value and the proportion of choosing high-effort task in the lowest reward condition reached significance (r₃₄ = -0.412, p = .015, CI = [–0.659, –0.086], p_{FDR corrected} = .075, see figure 3d).

Discussion

The present study investigated whether range adaptation was associated with amotivation symptoms in C-SCZ and FE-SCZ patients, and individuals with HSoA. Our findings suggested that the three groups showed distinct range adaptation pattern. Patients with C-SCZ appeared to have over-adaptation with their differences in response slopes for a narrow and a wide range enlarged, whereas patients with FE-SCZ showed insufficient adaptation as their slope differences reduced as compared to healthy individuals. People with HSoA, however, exhibited intact adaptation comparable to healthy individuals. Moreover, range adaptation was negatively (but non-significantly) correlated with amotivation symptoms and positive symptoms, and positively (but non-significantly) correlated with affective blunting and self-reported consummatory interpersonal pleasure scores in patients with C-SCZ. In patients with FE-SCZ, range adaptation was significantly and negatively correlated with levels of avolition, and positively correlated with the willingness to choose high-effort tasks. In individuals with HSoA, range adaptation was significantly and negatively correlated with the willingness to choose high-effort tasks under the lowest reward condition. Despite our small sample size and relatively modest findings, this investigation suggests that varied patterns of range adaptation ability exist across the SCZ spectrum. Moreover, range adaptation ability was associated with both the clinical and subclinical manifestations of amotivation in FE-SCZ. Yet, in C-SCZ, the impaired range adaptation might have more complicated and pervasive effects on clinical symptoms.

Performance of Range Adaptation to Outcome Value in Different Stages of SCZ

Range adaptation to prediction errors are impaired in SCZ patients, and such impairment is also correlated with schizotypal traits in the general population.²⁷ Moreover, attenuated range adaptation signals in the right caudate have been found in SCZ patients.^30,31 Using a behavioral paradigm, our findings apparently extended earlier evidence^27,30,31 to support that distinct range adaptation patterns would manifest at the behavioral level in SCZ patients and negative schizotypal people who are at risk of developing SCZ.

More specifically, a distinct pattern of range adaptation performance showed that healthy individuals have medium extent of range adaptation, but SCZ patients exhibit different levels of impairments in range adaptation (see Supplementary figure S2). For those at the risk of onset, range adaptation ability was comparable to controls, but with the illness processing, range adaptation impairments emerged. More extreme range adaption ability, at both the low and high ends of it, maybe associated with SCZ. Our findings implicate the existence of a “normal range” of range adaptation ability. Indeed, range adaptation heavily relies on the balance between excitatory and inhibitory neural activity.^43–45 If the balance shifts to either direction, adaptation performance would vary and become polarized to the extremes. Our findings supported this notion of balanced neural circuits underlying range adaptation. Given that patients with established SCZ exhibited relatively extreme range adaptation ability, the role of excitatory and inhibitory neural activity should be further studied, and the balance of two opposing signals may be related to the onset of SCZ. The observed difference between FE-SCZ and C-SCZ patients in range adaptation may also be attributable to the long-term effects of D2-blocking agents commonly found in C-SCZ patients. In fact, evidence suggests that pure dopamine antagonists (such as sulpiride) could dampen range adaptation to prediction error signals,^27,29 and it is plausible that this confound can modulate range adaptation to outcome values.

Our findings that individuals with HSoA showed normal range adaptation ability appear to suggest that this subclinical population may have certain protective factors, protecting them from developing impairments of range adaptation, and these protective factors may also be important in determining the risk of developing SCZ. Although our study was cross-sectional in design, the impaired range adaptation ability may heighten the risk of crossing of diagnostic boundary between subclinical and clinical status along the SCZ continuum.⁴⁶

The Relationship of Range Adaptation to Outcome Value with Clinical and Subclinical Symptoms

The range adaptation performance is correlated with level of amotivation in the FE-SCZ, the HSoA, and the C-SCZ group, though not all of results were significant. Moreover, by following the spectrum approach and merging the clinical and subclinical samples, we found a significant relationship between range adaptation and amotivation across the SCZ continuum. Our findings are consistent with prior evidence. For example, previous studies reported correlations between value representation ability and negative symptoms in SCZ patients.^47,48 Other studies demonstrated the association between negative symptoms and neural correlates of range adaptation in SCZ patients at early and chronic stages.^30,31 Our study extended these earlier findings by demonstrating the association between range adaptation and level of amotivation in first-episode, chronic SCZ (though non-significantly) as well as negative schizotypal samples. What’s more, the correlational results further supported the “normal range” of range adaptation ability. Both extreme heightened and reduced adaptation were coupled with reduced motivation and effortful behavior. Notably, range adaptation performance is correlated with different measures of amotivation (ie, the ACIPS, the SANS, and E-PET) in different samples. In fact, amotivation is believed to be a multi-facet construct.⁴⁹ Range adaptation to outcome values is one of the fundamental processes which underlies value-based decision making and motivated behavior,¹⁶ and therefore impaired range adaptation can affect various facets of motivation.

A significant negative correlation was found between range adaptation and the willingness to exert more effort under the lowest rewarded condition in people with HSoA. This negative correlation was, at first glance, contrary to our assumption. However, we purposed a possible explanation. The lowest rewarded condition in E-pet is 12% chance of winning¥1.5–¥2.4 for a high-effort task, which means an extra 50% effort could lead to an extra reward whose expected value is 0.12–0.168. The net value of choosing high-effort task under this condition would be negative (–0.5+(0.12–0.168)) and suboptimal. For people with higher range adaptation, their value discriminability towards the small value range could be higher. Thus, they would be unlikely to make such suboptimal choices.

Using an exploratory analysis to compare the correlations between range adaptation and amotivation against those between range adaptation and other clinical symptoms, we attempted to examine whether range adaptation would be specifically related to negative symptoms rather than other symptoms. Although our findings appear to partially support such a specific link between range adaptation and amotivation in FE-SCZ patients, the evidence for such a specific association remains unclear, in particular in C-SCZ patients. Many other different factors, such as long-term medication effects and long-term cognitive impairments, may influence range adaptation in C-SCZ patients. Having said that, our findings apparently suggest that the impaired range adaptation may be a possible intervention target to enhance patients’ motivated behavior, in particular for earlier stages of SCZ.

Furthermore, given its deep roots in the neural computation, range adaptation shows great promise in offering a framework within which we can understand the causative disturbances at the computational, neurobiological, and behavioral levels for amotivation. It concurs with the recent attempts to explain symptoms of SCZ in terms of “false inference,” a form of computations.⁵⁰ However, computational research on SCZ has been focused mostly on positive rather than negative symptoms. A plausible and testable computational model for negative symptoms is very important. Our investigations on range adaptation in SCZ explored this novel research area, which would provide a unique opportunity to develop a viable computational model for negative symptoms.

Several limitations should be borne in mind. First, our sample size was small, and many of our results were modest in magnitudes. Although we filtered the data using several methods to minimize the outliers’ effect on our small sample, future research should clarify our findings with a large sample. Second, we only recruited individuals with HSoA as the subclinical sample. However, following the full-dimensional approach, ultra-high-risk populations or unaffected first-degree relatives of SCZ patients should be included. Besides, we did not explore the effects of different types of negative symptoms (primary vs secondary, persistent vs transient) on range adaptation. Lastly, multiple testing of correlations between range adaptation and clinical and subclinical symptoms were conducted, and may lead to many biases.^41,42 Despite our efforts to address these problems using FDR corrections and direct comparison among correlation coefficients, our findings should be interpreted with caution.

To conclude, our study investigated the relationship between range adaptation and amotivation using two clinical samples with SCZ at different illness stages and one subclinical sample with high levels of social anhedonia. Across the SCZ spectrum, the three groups showed distinct range adaptation ability. We provided empirical evidence for different patterns of range adaptation along the SCZ spectrum, and laid important groundwork for future studies examining how range adaptation influences clinical symptoms, especially amotivation symptoms.

Funding

This study was supported by a grant from the National Key Research and Development Programme (2016YFC0906402), the Jiangsu Provincial Key Research and Development Program (BE2020661), Beijing Municipal Science & Technology Commission Grant (Z161100000216138), and the CAS Key Laboratory of Mental Health, Institute of Psychology.

Acknowledgments

All authors report no competing interests.

References