Relational Memory in the Early Stage of Psychosis: A 2-Year Follow-up Study

Abstract

Background

Relational memory, the ability to bind information into complex memories, is moderately impaired in early psychosis and severely impaired in chronic schizophrenia, suggesting relational memory may worsen throughout the course of illness.

Methods

We examined relational memory in 66 early psychosis patients and 64 healthy control subjects, with 59 patients and 52 control subjects assessed longitudinally at baseline and 2-year follow-up. Relational memory was assessed with 2 complementary tasks, to test how individuals learn relationships between items (face-scene binding task) and make inferences about trained relationships (associative inference task).

Results

The early psychosis group showed impaired relational memory in both tasks relative to the healthy control group. The ability to learn relationships between items remained impaired in early psychosis patients, while the ability to make inferences about trained relationships improved, although never reaching the level of healthy control performance. Early psychosis patients who did not progress to schizophrenia at follow-up had better relational memory than patients who did.

Conclusions

Relational memory impairments, some of which improve and are less severe in patients who do not progress to schizophrenia, are a target for intervention in early psychosis.

Introduction

Schizophrenia is associated with cognitive deficits.^1–4 Memory deficits are especially prominent^5–9: they are present early in the illness,^8,10,11 progress with illness duration,⁸ and are strong predictors of functional outcomes.^12–14 One form of memory, relational memory, is strikingly impaired in schizophrenia^15–29 and is of particular interest in schizophrenia due to its critical dependence on an intact prefrontal-hippocampal network.^20,29–35

Relational memory refers to the ability to bind individual elements of an experience to create a unified representation of discrete events. Our memories are not “snapshots” of our experience, but flexible representations that bind events into relational networks.^36–38 Hippocampal lesion patients show profound deficits in relational memory,^32,39–41 while other forms of memory, including memory for individual items, remain relatively spared.^30,42,43 Similarly, relational memory is preferentially impaired in schizophrenia^15,29 and is associated with abnormal hippocampal function.^{18,23,28,29,44} The hippocampus, in coordination with the prefrontal cortex,^{29,34,35,45–47} also supports the ability to draw novel inferences about trained relationships.^36–38 Schizophrenia patients show decreased ability to make novel inferences based on prior learning,^17,22,29 and these deficits have been linked to hippocampal and prefrontal cortex dysfunction.^28,29

Relational memory deficits may parallel the progression of hippocampal-prefrontal neuropathology observed in schizophrenia. Relational memory impairments are more pronounced in chronic stages of schizophrenia^{2,18,22,48–61} than in early psychosis^21,24,62,63 or in clinical high-risk states.⁶² Similarly, the hippocampus and prefrontal cortex show marked structural and functional deficits in chronic schizophrenia,^64–66 with lesser neuropathology observed in the earlier stages of illness.^{49,60,67–71} This led Lepage et al²⁰ to propose relational memory deficits as a proxy for the progression of neuropathology in schizophrenia. In support of a proposed decline, a recent longitudinal study of relational memory found significant declines over 5 to 11 years of illness.⁷² However, whether relational memory deficits progress during the initial years of illness remains unclear.

To test this hypothesis, we used a longitudinal design to study relational memory during the early stage of psychosis. Patients completed a baseline assessment within the first 2 years of illness and a follow-up assessment 2 years later. Participants completed 2 complementary tasks to assess the distinct aspects of relational memory: the face-scene binding (FSB) task,⁴⁰ which uses implicit eye movements to measure the ability to bind together 2 arbitrary items (faces and scenes), and the associative inference (AI) task,⁷³ which assesses the ability to make explicit inferences using bound items (faces with a house), requiring the flexible use of prior relational memories (figure 1). The AI task was recommended by the Cognitive Neuroscience Treatment Research to Improve Cognition in Schizophrenia (CNTRICS) initiative as the preferred task for investigating relational memory in schizophrenia.¹⁹

Fig. 1.

Relational memory tasks. (A) Face-scene binding (FSB) task: subjects learned to associate a face with a background scene. During testing, participants were asked to find, with their eyes, the one face among 3 equally familiar faces that had been previously paired with the background scene while their eye movements were recorded. (B) Associative inference (AI) task: subjects learn 3 sets of paired associates: House with Face (H-F1), same House with new Face (H-F2), and 2 new Faces (F3-F4). In testing, memory for the trained pairs and a novel set of inferential face-face pairs (F1-F2), which could only be solved by overlapping associations with the house, were tested.

We previously found marked relational memory impairment in both the FSB and AI tasks in chronic schizophrenia samples with 16 to 22 years of illness duration^18,22,23 and less prominent deficits in a sample of psychosis patients with less than 2 years of psychosis.^21,24 This is the first longitudinal study of the 2 tasks. We hypothesized that relational memory impairments would progress over 2 years of illness. In a secondary analysis, we tested the hypothesis that relational memory is more impaired in early psychosis patients who progress to schizophrenia than in patients who do not. Because these tasks test the distinct aspects of relational memory and employ different methodologies, which may affect discriminatory power between tasks,⁷⁴ we report these tasks separately.

Methods

Participants

One-hundred thirty participants (66 early psychosis patients and 64 healthy control subjects) were enrolled in a prospective longitudinal study of relational memory between May 2013 and August 2017 and followed for 2 years (figure 2). Fifty-nine (89%) patients and 52 (81%) healthy control participants completed the study. Cross-sectional baseline data from some of our participants have already been reported (110 subjects in Avery et al²¹; 91 subjects in Armstrong et al²⁴), but this is the first report of follow-up data.

Fig. 2.

Flow diagram of study participants. All participants completed the associative inference (AI) task (right) at baseline and most participants also completed the face-scene binding (FSB) task (left).

Patients in the early stage of a psychotic disorder were recruited from the outpatient clinics and inpatient units of Vanderbilt Psychiatric Hospital. Inclusion criteria included: premorbid IQ above 75, <2 years of psychotic illness, and meeting at least 2 criteria A for schizophrenia.^75,76 Early psychosis participants were excluded if they reported active substance use or dependence in the past month or if a psychotic disorder due to a medical condition was diagnosed. Additionally, only participants meeting criteria for either schizophreniform disorder or schizophrenia/schizoaffective disorder (this combined group will be called schizophrenia going forward) at 2-year follow-up were included in this analysis.

Healthy control participants were recruited from the community via advertisement and included if they did not meet the criteria for any Axis I disorder^75,76 at enrollment and at the end of the study. All diagnoses were ascertained using the Structured Clinical Interview for DSM-IV-TR (SCID),⁷⁵ and groups were recruited with the goal to match for mean age, gender, race, and parental education (table 1).

Table 1.

Demographic and Clinical Characteristics

Demographics	Healthy Control FSB, n = 58 / AI, n = 64		Early Psychosis FSB, n = 60 / AI, n = 66		Healthy Control vs Early Psychosis
	Mean	SD	Mean	SD	Statistic	df	P
Age	22/22	2.88/2.92	21/21	3.19/3.16	t = 1.26	128	.21
Gender, % male	74%/73%	—	75%/77%	—	Χ 2 = 0.26	1	.61
Race	46 W, 12 O / 50 W, 14 O	—	47 W, 12 O / 50 W, 16 O	—	Χ 2 = 0.10	1	.75
IQ, WTAR	112/112	10.58/10.66	103/103	16.06/16.07	t = 3.88	110.14	<.001*
Participant education	14.7/14.6	1.86/1.87	13.3/13.3	2.23/2.16	t = 3.75	128	<.001*
Parental education	14.9/14.9	2.27/2.22	15.3/15.2	2.95/2.83	t = 0.72	122.69	.47
Clinical	Baseline (n = 66)			2-year follow-up (n = 59)
	Mean ± SD		Range	Mean ± SD		Range
Diagnosis	4 BP/44 SZF/18 SZ		—	0 BP / 16 SZF / 43 SZ		—
Duration of psychosis, weeks	28 ± 24.72		1.1–103.1	138 ± 25.86		104.7–210.7
CPZ equivalent, mg/day	322 ± 142.04		0–750	317 ± 207.71		0–900
HAM-D 17A	9 ± 6.68		1–25	5 ± 4.15		0–15
YMRS	2 ± 3.80		0–20	2 ± 2.90		0–15
PANSS Total	67 ± 20.60		34–118	51 ± 15.15		32–104
PANSS—Positive	17 ± 8.15		7–36	13 ± 4.76		7–27
PANSS—Negative	17 ± 8.15		7–37	12 ± 6.31		7–32
PANSS—General	33 ± 9.18		18–59	26 ± 7.01		16–46

Note: FSB, Face-Scene Binding Task, AI, Associative Inference Task; WTAR, Wechsler Test of Adult Reading; SD, Standard Deviation; BP, Bipolar Disorder with Psychotic Features; SZF, Schizophreniform; SZ, Schizophrenia; HAM-D, Hamilton Depression Rating Scale; YMRS, Young Mania Rating Scale; PANSS, Positive and Negative Syndrome Scale.

*Denotes significant P-values (P < .05).

All participants provided written informed consent and received monetary compensation for their time. The study was approved by the Vanderbilt University Institutional Review Board, Nashville, TN.

Clinical Assessment and Measures

To fully characterize the trajectory of illness, we collected extensive clinical data during in-person interviews at enrollment and at the end of the study (supplementary table S1). Early psychosis participants completed a thorough clinical interview, including the SCID,⁷⁵ Positive and Negative Syndrome Scale (PANSS),⁷⁷ Depression Rating Scale (HAM-D),⁷⁸ Young Mania Rating Scale (YMRS),⁷⁹ Screen for Cognitive Impairment in Psychiatry (SCIP),⁸⁰ Personal and Social Performance scale (PSP),⁸¹ and Quality of Life scale (QoL).⁸² At baseline, early psychosis participants completed 2 additional assessments: the Symptom Onset in Schizophrenia Inventory (SOS)⁸³ and the Premorbid Adjustment Scale (PAS).⁸⁴ The duration of untreated psychosis (DUP) was calculated as the date of onset of psychosis, determined with the SOS, to the date of first antipsychotic treatment. The duration of psychosis was calculated as the date of onset of psychosis until the date of study enrollment, including DUP. All data gathered during the in-person interviews were augmented by an extensive review of all available medical records. Diagnostic consensus meetings were held and final diagnoses were made by psychiatrist S.H.

Experimental Design: FSB and AI tasks

Relational binding was assessed using the FSB task.^18,40 The ability to make inferences about trained relationships was assessed using the AI task.^22,24,73Figure 1 summarizes the main features of the 2 tasks and the supplementary Methods and Results provide complete details of the study procedure.

Data Analysis

Task #1: Face-Scene Binding

Relational memory for face-scene pairs was characterized as the change in the proportion of time spent viewing the Match face (viewing slope). Because hippocampally driven preferential viewing occurs rapidly,⁸⁵ we examined the first 2 seconds of each trial, similar to our previous studies.^18,21,86 To calculate viewing slopes, the proportion of time spent viewing the Match face relative to other display elements (eg, Non-Match faces, background scene) was calculated over 250 ms time bins (eg, 0–250, 250–500, 500–750 ms, ... 1750–2000) and corrected for total screen viewing time. The proportion of Match face viewing was averaged across test trials and modeled for each participant using regression, where the proportion of viewing is predicted by the log-transformed time bin. The natural log transform linearizes the viewing curve, which is steepest during early viewing, enabling linear regression analysis. Viewing slopes were corrected for between-subject differences in initial eye position (first 250 ms time bin) following face presentation.^87,88

Task #2: Associative Inference

Our primary analysis was inferential memory performance (F1-F2 accuracy) over time. A secondary analysis compared inferential memory performance (F1-F2) in 2 patient groups (schizophreniform disorder and schizophrenia). We complemented primary and secondary analyses with a comparison of accuracy for inferential (F1-F2) and non-inferential (F3-F4) pairs, similar to our previous studies.^22–24 We included all trials in subjects with baseline and/or follow-up data and included pair type as a third independent variable (F1-F2 and F3-F4). Finally, for all 3 analyses above, we selected only those inferential (F1-F2) pairs for which 3 of 4 or the final 2 premise pairs (H-F1 and H-F2) were correctly identified. We included subjects in such correct trials-only analyses if they contributed a minimum of 10 trials to the analysis. This resulted in the exclusion of 2 early psychosis participants at baseline and 3 early psychosis participants at 2-year follow-up. The average number of trials included was higher in the early psychosis group relative to healthy controls at baseline (29 and 25, t_67.8 = 4.86, P < .001) and at follow-up (29 and 26, t_92.6 = 2.64, P = .01).

Statistical Analysis

Between-group analyses were completed using RStudio v.1.2⁸⁹ and the lme4 package.⁹⁰ To enable comparison across tasks, we report a robust effect size estimate⁹¹ for all tests.

Task #1: Face-Scene Binding

We analyzed the viewing slope over time. For the primary analysis, a linear mixed-effects model tested for effects of group (healthy control vs early psychosis) and time (months between study visits) as fixed factors on relational memory slopes (dependent variable), with participant as a random factor. For the secondary analysis, a linear mixed-effects model tested for the effects of progression of illness, with effects of patient group (schizophreniform disorder vs schizophrenia) and duration of illness (months) as fixed factors on relational memory slopes (dependent variable) and with participant as a random factor.

Task #2: Associative Inference

For the primary analysis of F1-F2 accuracy over time, we used a logistic regression model with count data of correct and incorrect responses as the outcome variable, as F1-F2 accuracy was highly skewed toward ceiling performance in healthy control participants. The model tested for effects of group (health control vs early psychosis) and time (months between study visits) as fixed factors on F1-F2 accuracy (dependent variable), with participant as a random factor. For the secondary analysis, we used a logistic regression model to test for effects of progression of illness, with effects of patient group (schizophreniform disorder vs schizophrenia) and duration of illness (months) as fixed factors and F1-F2 accuracy as the dependent variable, with participant as a random factor.

Results

Clinical Sample

The patient sample was young (mean age 21, range 16–31), primarily male (77%), Caucasian (78%), and educated (13 years of education, range 9–22) (table 1; supplementary table S1).

Fifty-nine out of 66 early psychosis participants completed the 2-year follow-up visit. The majority (43, 73%) were diagnosed with schizophrenia or schizoaffective disorder after 2 years of follow-up. Within this group, 17 already met criteria at study entry, 22 progressed from an initial diagnosis of schizophreniform disorder, and 4 converted from an initial diagnosis of psychotic bipolar disorder. The minority (16, 27%) were diagnosed with schizophreniform disorder at study entry and did not progress to another diagnosis within 2 years (supplementary table S1). The 7 early psychosis participants without a 2-year follow-up visit had been diagnosed with schizophrenia (n = 1) and schizophreniform disorder (n = 6) at study entry and were not demographically and clinically different from early psychosis participants who completed both visits (P’s > .05). Details of medication status are reported in the supplementary Methods and Results.

Task #1: Face-Scene Binding

Overall, the healthy control group performed better than early psychosis group, as measured by a more rapid increase in the slope of preferentially viewing the correct face (main effect of group, t_170.42 = 3.53, β = 7.792, SE = 1.549, effect size = 0.62, P < .001; figure 3A). Viewing slopes were similar over time in both groups (no main effect of time, P = .14; no group by time interaction, P = .13), indicating relational memory remained impaired in early psychosis participants over the course of the study. The healthy control group also showed better explicit accuracy than the early psychosis group during a forced-choice test (supplementary table S2). An analysis of viewing slopes limited to correct trials only found similar results to the all-trials analysis (supplementary table S3).

Fig. 3.

Relational memory at baseline and follow-up. (A) The early psychosis group performed worse on the face-scene binding (FSB) task (smaller viewing slope) than the healthy control group and remained impaired at follow-up. (B) FSB task performance was similar across 16 schizophreniform disorder participants and 43 schizophrenia participants. (C) Associative inference (AI) performance (F1-F2 accuracy) improved over the 2 years, with better performance in the healthy control group than the early psychosis group. (D) AI performance at baseline and follow-up in schizophreniform disorder and schizophrenia. Schizophreniform disorder participants (N = 16) were more accurate than schizophrenia participants (N =43).

In a secondary analysis, we compared 16 schizophreniform disorder participants with 43 schizophrenia participants (figures 3B and 4A). Patient groups had similar viewing slopes (no main effect of group, P = .45) and did not differ over time (no main effect of time, P = .45; no group by time interaction, P = .20), indicating that the progression of psychosis was not associated with further relational memory decline. Results were similar when considering correct trials only (supplementary table S4).

Fig. 4.

Individual relational memory performance in the schizophreniform disorder and schizophrenia groups. (A) Face-scene binding (FSB) performance was heterogeneous in both patient groups. (B) The majority of schizophreniform disorder participants showed good associative inference (AI) performance, while the schizophrenia group was more heterogeneous.

Task #2: Associative Inference

Overall, the healthy control group had better inferential memory performance than the early psychosis group (main effect of group, β = −1.54, SE = 0.24, effect size = 1.12, P < .001). Both groups similarly improved at follow-up (main effect of time, β = .011, SE = 0.005, effect size = 0.32, P = .042; no group by time interaction, P = .74; figure 3C). An analysis limited to correct trials only revealed similar results (supplementary Methods and Results). Comparing inferential and non-inferential performances, we again found that the healthy control group performed better than the early psychosis group; both groups showed the expected inferential memory effect of greater accuracy on F1-F2 relative to F3-F4 pairs, which did not change over time (supplementary Results and figure S1A).

In a secondary analysis, we compared 16 schizophreniform disorder participants with 43 schizophrenia participants (figures 3D and 4B). While performance in both groups improved at follow-up (main effect of time, β = .024, SE = 0.008, effect size = 0.66, P = .004), the schizophreniform disorder group performed significantly better (main effect of diagnosis, β = −.978, SE = 0.361, effect size = 0.70, P = .007). The schizophrenia group showed considerable heterogeneity; in contrast, only one schizophreniform disorder participant showed poor relational memory (<66% accuracy) at follow-up (figure 4B). The time by diagnosis interaction was not significant (β = −.012, SE = 0.009, effect size = 0.24, P = .18). A correct trials-only analysis revealed similar results (supplementary Results). Comparing inferential and non-inferential memory showed similar findings, as inferential vs non-inferential memory was higher for both groups and the overall accuracy improved over time (supplementary Results). Both groups demonstrated the inferential relational memory effect, but the schizophreniform group had a stronger pattern than the schizophrenia group (supplementary Results and figure S1B).

Cognitive and Clinical Correlates

To determine whether improvements over time in AI performance could be explained by cognitive or clinical differences between the 2 patient groups, we conducted exploratory post hoc comparisons. The schizophreniform disorder group had higher premorbid IQ than the schizophrenia group (Wechsler Test of Adult Reading [WTAR]: 111.5 ± 10.35 vs 98.76 ± 16.57; t_43.63 = 3.48, P = .003). At 2-year follow-up, but not at baseline, the schizophreniform disorder group had better cognition scores when using a gross measure of cognitive ability (SCIP: −0.08 ± 0.57 vs −0.74 ± 0.99; t_46.36 = 3.19, P = .003). Both IQ and cognition were positively correlated with relational memory performance at 2-year follow-up (WTAR: AI, r = .35, P = .008; FSB, r = .55, P < .001; SCIP: AI, r = .62, P < .001; FSB, r = .47, P < .001).

Post hoc comparisons of clinical symptoms showed that PANSS total scores were not different between the diagnostic groups at baseline (schizophreniform: 65.38 ± 17.20; schizophrenia: 66.30 ± 21.52; t₅₇ = 0.15, P = .88) but were significantly different at 2-year follow-up (schizophreniform: 39.00 ± 5.88; schizophrenia: 55.77 ± 15.06; t_56.78 = 4.31, P < .001). PANSS scores at both baseline and 2-year follow-up were negatively correlated with AI performance (baseline: r = −.35, P = .004; follow-up: r = −.37, P = .004).

Discussion

This study provides novel and compelling evidence for timing and specificity of relational memory deficits in the early stage of psychosis. Employing 2 complementary tasks, we found relational binding deficits to be stable over 2 years, while inferential ability showed significant improvements. Secondary analyses revealed inferential ability to be less impaired in schizophreniform disorder participants who did not progress to schizophrenia.

Cohort studies had shown relational memory to be moderately impaired in early psychosis^21,22 and severely impaired in chronic schizophrenia,^18,22,23 suggesting relational memory deficits to progress.⁹² Using a longitudinal study design, we found no deterioration of relational memory ability in the first 2 years of a schizophrenia spectrum disorder. While longer-term follow-up is warranted, our findings indicate that the early stage of psychosis may be a window for intervention.

A strength of our study is the use of 2 relational memory tasks. Relational memory allows us to form rich, complex memories of our past experience, by binding together information that is experienced close in time. To capture complementary aspects of relational memory, we measured both the ability to bind items and the ability to make inferences about learned relationships. While the ability to bind items into relational memories is critically dependent on the hippocampus,^37,41,93 the ability to make inferences about relationships involves a broader hippocampal-prefrontal network.^33,45,46,94 We found that hippocampal-based relational binding deficits did not change, but that inferential performance improved in the first 2 years of a schizophrenia spectrum disorder. This supports the hypothesis that prefrontal-hippocampal connections are a crucial mediator of clinical and cognitive performance in early psychosis,^{29,53,67,95–101} whereas intra-hippocampal circuits are more closely linked to pervasive features of psychosis.¹⁰² Additional studies examining distinct aspects of relational memory and their prefrontal and hippocampal mechanisms will be essential in clarifying this relationship.

Our comparison of relational memory tasks revealed that schizophreniform disorder patients maintain the ability to make novel inferences but are as impaired as schizophrenia patients in relational binding. This suggests that prefrontal-dependent inferential ability is uniquely preserved in patients who maintain a diagnosis of schizophreniform disorder over the first 2 years of illness. These findings are consistent with prior evidence that memory deficits mirror the progression of illness,^{8,92,103–108} with fewer deficits predictive of lower diagnostic conversion¹⁰⁹ and better diagnostic outcome.² Using the same AI task, we previously found that half of the early psychosis patients performed within a normal range when making novel inferences,²⁴ while less than 20% of chronic schizophrenia patients perform within the normal range.^22–24 We now extend these findings and show that superior relational memory at the onset of psychosis indicates better outcomes after 2 years of illness.

The initial course of psychotic disorders varies tremendously, with significant heterogeneity in both the initial diagnosis and progression and conversion of diagnoses.^110,111 A quarter of our early psychosis participants had not progressed beyond schizophreniform disorder after 2 years, a ratio that is in line with recent meta-analyses.^112,113 In addition to better inferential memory performance, they also displayed higher premorbid IQ than the schizophrenia group. Although general cognition was similar at baseline between the 2 patient groups, the schizophreniform disorder group displayed improved cognition compared with the schizophrenia group at follow-up. This may indicate both a premorbid protective factor of higher cognitive reserve and cognitive recovery following a psychotic episode. But our sample of non-converters was small and future studies need to confirm the role of prefrontal-hippocampal networks for progression from early psychosis to schizophrenia.

Inferential performance was also correlated with clinical and cognitive improvement across the early psychosis group during the 2-year period. In patients, psychotic symptoms improved between baseline and follow-up, suggesting that the acute psychotic episode contributed to the larger inferential memory impairments seen at baseline. This finding is consistent with recent evidence indicating relatively high clinical remission and recovery rates in the early stage of psychosis.¹¹⁴

Similarly, our findings of the overall stability or improvement in relational memory over 2 years are in line with a meta-analysis of longitudinal cognitive functioning in first-episode psychosis, which found general neurocognitive improvements that did not differ in the length of follow-up from 1 to 5 years.³ However, longitudinal studies have found a mixed pattern for memory, ranging from memory decline to amelioration.^{106,115–121} Our current results aid in clarifying these mixed findings and provide further indication that the years following the initial onset of psychosis may be a relatively stable neurocognitive period that precedes further decline.

Our study has several limitations. First, participants lost to follow-up may bias the sample toward a higher-functioning phenotype. This concern is largely mitigated by the high retention rate in our early psychosis (89%) and healthy control (81%) groups. We also did not find any demographic or clinical differences between participants who completed or did not complete the study. Second, most participants with early psychosis were medicated at baseline and follow-up, although fewer at follow-up (64%) than baseline (85%), which may affect relational memory performance. However, there were no associations between relational memory performance and medication dosage. Third, the healthy control group showed higher average IQ; future studies should test relational memory performance in early psychosis patients with less impaired IQ, eg, non-affective psychosis patients.⁸⁶

We conclude that relational memory is impaired in the early stage of psychosis and does not decline further over the next 2 years. We found compelling evidence of better inferential memory in early psychosis patients who have better clinical outcomes. Parsing the heterogeneity of memory dysfunction will be critical for determining who will or will not progress to schizophrenia.^122,123 Our findings suggest that retention of inferential ability over the early course of illness may be a valuable marker of both illness severity and clinical progression.

Funding

This work was supported by the Charlotte and Donald Test Fund, National Institute of Mental Health (NIMH) grant (R01-MH70560 to Dr Heckers), the Vanderbilt Psychiatric Genotype/Phenotype Project, and the Vanderbilt Institute for Clinical and Translational Research (through grant 1-UL-1-TR000445 from the National Center for Research Resources [NIH]).