Effect of citalopram on hippocampal volume in first-episode schizophrenia: Structural MRI results from the DECIFER trial

Abstract

Hippocampal volume loss is prominent in first episode schizophrenia (FES) and has been associated with poor clinical outcomes and with BDNF genotype; antidepressants are believed to reverse hippocampal volume loss via release of BDNF. In a 12-month, placebo-controlled add-on trial of the antidepressant, citalopram, during the maintenance phase of FES, negative symptoms were improved with citalopram. We now report results of structural brain imaging at baseline and 6 months in 63 FES patients (34 in citalopram group) from the trial to assess whether protection against hippocampal volume loss contributed to improved negative symptoms with citalopram. Hippocampal volumetric integrity (HVI) did not change significantly in the citalopram or placebo group and did not differ between treatment groups, whereas citalopram was associated with greater volume loss of the right CA1 subfield. Change in cortical thickness was associated with SANS change in 4 regions (left rostral anterior cingulate, right frontal pole, right cuneus, and right transverse temporal) but none differed between treatment groups. Our findings suggest that minimal hippocampal volume loss occurs after stabilization on antipsychotic treatment and that citalopram’s potential benefit for negative symptoms is unlikely to result from protection against hippocampal volume loss or cortical thinning.

1. Introduction:

Hippocampal volume loss is prominent early in the course of schizophrenia (Adriano et al., 2012) and has been associated with poor outcomes (Bois et al., 2016; Ho et al., 2017a). In a sample of medication-naïve individuals with first episode schizophrenia, we recently found that a reduction during the first 8 weeks of treatment in left hippocampal volumetric integrity (HVI), which is inversely related to volume loss, predicted poor response of negative symptoms and was associated with brain-derived neurotrophic factor (BDNF) genotype (Goff et al., 2018). Mondelli and colleagues (Mondelli et al., 2011) also found that BDNF genotype predicted hippocampal volume loss in early psychosis. Hippocampal volume loss has also been attributed additional factors (Goff et al., 2018) including inflammation and oxidative stress (Aleksovska et al., 2014) which may be exacerbated by an initial antipsychotic effect (Rocchetti et al., 2015) and might not extend into the maintenance phase of treatment.

Based on the finding by Mondelli and colleagues (Mondelli et al., 2011) that BDNF genotype predicted hippocampal volume loss in early psychosis and on our replication of this relationship (Goff et al., 2018), we conducted the Depression and Citalopram in First Episode Recovery (DECIFER) Study, a 12-month, placebo-controlled trial of add-on citalopram in first episode schizophrenia patients who had been stabilized on second generation antipsychotics (Goff et al., 2019). Citalopram was selected because antidepressants have been shown to affect the hippocampus in part via release of BDNF (Bjorkholm and Monteggia, 2016) and citalopram has minimal pharmacokinetic interaction with antipsychotic drugs. We found that citalopram improved negative symptom severity and that a longer DUP predicted benefit from citalopram compared to placebo (Goff et al., 2019). We now present analyses of structural brain imaging of participants in this study to examine the relationships among hippocampal volume loss, citalopram treatment, DUP, and clinical outcomes. Given the above literature and our previous findings in this study, we predicted 1) that the rate of hippocampal volume loss after initial stabilization on antipsychotic medication in participants receiving add-on placebo would be less than the median annualized 4% rate of hippocampal volume loss that we previously observed during the first 8 weeks of treatment in medication-naïve first episode schizophrenia participants (Goff et al., 2018); 2.) that citalopram would protect against further hippocampal volume loss compared to placebo and that this effect would be modulated by DUP; and 3.) that negative symptoms would be associated with hippocampal volume loss. In exploratory analyses, we measured volumes of hippocampal subfields CA1 and dentate gyrus for a more detailed examination of citalopram effects on the hippocampus and measured cortical thickness in other brain regions to identify other potential sites of citalopram’s therapeutic effect on negative symptoms. We also examined whether BDNF genotype predicted hippocampal volume loss with placebo versus citalopram.

2. Methods:

2.1. Study Design:

As previously described (Goff et al., 2019), the Depression and Citalopram in First Episode Recovery (DECIFER) Study was a placebo-controlled 12-month trial of add-on citalopram in nonaffective first episode schizophrenia patients stabilized on clinician-determined second generation antipsychotic for a mean of 10.3 weeks. MRI scans were performed at baseline, 24 weeks, and 52 weeks. The current analysis examines outcomes at 24 weeks in order to maximize sample size (n=66). A total of 44 subjects (21 placebo, 23 citalopram) completed both 24-week and 52-week follow-ups. Exploratory results using linear mixed model examining this smaller subset is presented in Supplement-15.

2.1.1. Participants and randomization:

The protocol for the DECIFER Study (ClinicalTrials.gov: NCT01041274) was approved by IRBs at the four participating sites: New York University Medical Center, the Massachusetts General Hospital, Shanghai Mental Health Center, and the Second Xiangya Hospital of Central South 2014. All participants provided written informed consent. Eligibility criteria included a first episode of schizophrenia or schizophreniform disorder based on the Structured Clinical Interview for DSM IV-TR (SCID)(First, 1994), age 15–40, onset of psychosis before age 35, cumulative antipsychotic treatment for at least 4 weeks and fewer than 24 weeks, no antidepressant treatment within four weeks, a score less than 7 on the Calgary Depression Scale for Schizophrenia (CDSS; 6 is a threshold for probable major depression) (Addington et al., 2014) (Addington et al., 1993), and a score less than 3 (moderate) on the Clinical Global Impression Scale for Severity of Suicidality (CGI-SS) (Lindenmayer et al., 2003). In addition, participants had to have been free of any substance abuse for 3 months (except for cannabis, in which case only moderate or severe cannabis use disorders were excluded), not have unstable medical illness, and have a corrected QT interval on electrocardiogram less than 500 msec. Participants remained on their clinician-determined second generation antipsychotic and were randomized in parallel to add-on citalopram or placebo in a 1:1 ratio stratified by site. Within each stratum, the sequence of treatment assignments was constructed by a statistician using permuted random blocks with variable block sizes; study drug and placebo were prepared in identical capsules and labelled by a research pharmacist. All other research personnel remained blinded until the end of the study.

2.1.2. Study drug:

Citalopram was initiated at a dose of 20 mg once daily and increased after one week to a target dose of 40 mg if tolerated. Dose adjustment within a range of 10–40 mg daily was allowed during the trial based on clinical assessment of side effects. Subjects were assessed by a psychiatrist every week for the first four weeks and then every two weeks thereafter.

2.1.3. Psychoeducation:

All participants received 16 sessions of weekly individual psychoeducation and relapse prevention planning followed by 8 monthly sessions. The intervention was provided by a doctoral level clinician with extensive experience treating individuals with first episode psychosis.

2.1.4. Cognitive–behavioral therapy (CBT):

Participants who scored 3 (moderate suicidality) on the CGI-SS or 7 or greater on the CDSS at any point during the trial were treated with a standard 12 session CBT approach to depression consisting primarily of behavioral activation and skills training in the identification and modification of depressogenic cognitions with cognitive restructuring. Participants who continued to meet these criteria after 4 weeks or who scored a 4 (severe) or higher on the CGI-SS or 10 or greater (severe) on the CDSS were dropped from double-blind treatment and could be openly prescribed an antidepressant.

2.2. Measurements:

2.2.1. MRI procedure:

The imaging procedure was harmonized across all four study sites. Participants were scanned twice (two images) at each of two time-points (baseline and 24 weeks) on 3T Siemens scanners at all sites using similar pulse-sequences. Detailed MRI sequences can be found in Supplement Table S11.

2.2.2. Hippocampal volumetric integrity:

HVI, a measure of hippocampal intactness expressed as the parenchymal fraction of a standardized volume of interest that is expected to encompass the hippocampus in a normal brain, was calculated from longitudinal T1-weighted images using a fully automated procedure (Ardekani et al., 2016b; Goff et al., 2018). This procedure has demonstrated excellent test-retest reliability (ICC = .998 in the same session; ICC = .995 in two different sessions 1 week apart, Figure S2), superior performance compared to FreeSurfer v5.3.0 in discriminating individuals with Alzheimer’s disease from healthy age- and sex-matched controls on the basis of hippocampal volume change (Ardekani et al., 2016a; Ardekani et al., 2017; Ardekani et al., 2016b), and superior performance compared to FreeSurfer v6.0 in differentiating FEP subjects from healthy controls and in identifying hippocampal volume change early in treatment (Goff et al., 2018).

2.2.3. Hippocampal CA1 and dentate gyrus subfield volumes:

The longitudinal analysis protocol of the FreeSurfer v6.0 automated hippocampal subfield extraction tool was used to segment the right and left hippocampal subfields CA1 and dentate gyrus (these two volumes were used from among the standard 12 substructures yielded by this tool). In the longitudinal analysis protocol, an unbiased within-subject template image is created by registering images from a given participant across scanning time points (here baseline and 24 weeks). Information from this template is then used to standardize segmentations performed on each time point, increasing the robustness of detecting longitudinal change (Reuter et al., 2012). All segmentations were visually inspected for accuracy by an experienced rater. No subfield segmentations required manual adjustment.

2.2.4. Cortical Thickness:

Cortical thickness was measured using the cortical surface stream of FreeSurfer v6.0 based on the T1-weighed structural MRI described in detail in prior literature published by Dale et al.(Dale et al., 1999) Briefly, the method finds the surface between white and gray matter and between gray and pia. The distance between the white/gray and the gray/pial surfaces gives the thickness at each location of cortex (Fischl and Dale, 2000). We used the cortical parcellation based on the Desikan-Kiliany atlas to determine the average cortical thickness in 70 cortical regions (34 regions per hemisphere plus the mean thickness of each hemisphere).

2.2.5. Rating scales:

Negative symptoms were measured by the Scale for the Assessment of Negative Symptoms (SANS) modified by elimination of items measuring inattention and inappropriate affect (Blanchard and Cohen, 2006). The Brief Psychiatric Rating Scale (BPRS) (Lukoff et al., 1986) positive symptoms subscale (Ventura et al., 2000) was used to measure severity of positive symptoms. The CDSS was used to measure depressive symptoms (Addington et al., 1996). Cognition was measured by the MATRICS Consensus Cognitive Battery (MCCB) composite score (Nuechterlein and Green, 2006). Baseline and 24-week data were used for analysis.

Raters at all sites were trained in-person on all rating scales and in the administration of the MCCB and met inter-rater reliability criteria (85% agreement within 1 point difference from the gold starndard rating) on three videotaped interviews for the BPRS, CDSS and SANS.

DUP was estimated based on history provided by the participant and by family members and was defined as the number of weeks elapsed since the onset of at least one persistent psychotic symptom of moderate or greater severity prior to initiation of antipsychotic medication. Duration of antipsychotic treatment at baseline was calculated as the difference between the start date of antipsychotic treatment and the date of the baseline MRI scan.

2.2.6. DNA:

A whole blood sample was obtained and DNA extraction was performed and stored at −80 °C. BDNF Val66Met genotyping (rs6265/G196A) was conducted with the Sequenom MassArray Platform (Sequenom Inc, San Diego, CA) using primers that have been previously described (Ho et al., 2007).

2.3. Statistical Analysis:

Statistical analyses were performed using SAS (version 9.3), except for exploratory path analysis using Mplus. All data were screened for normality of distribution using Shapiro-Wilk normality tests to determine the proper model. For normally distributed data, we used Pearson correlations to examine the relationship between baseline variables and independent sample t-tests to examine between-group differences. Otherwise, we used Spearman rank order correlations and the Wilcoxon rank sum tests. ANCOVA was used to examine interactions between citalopram treatment and other baseline variables on HVI, hippocampal subfield volumes, and symptom change. Intracranial volume (ICV), sex, age, and baseline antipsychotic dose (in chlorpromazine equivalents) were controlled for in analyses of hippocampal subfield volumes. In order to minimize outlier effects, we used median values of independent variables to divide subjects into high and low groups, and conducted sensitivity analyses using this approach for all ANCOVA analyses. In exploratory analysis, least absolute shrinkage and selection operator (LASSO) was used to select areas where changes of cortical thickness predicted SANS score changes. All 70 Desikan-Kiliany atlas cortical regions were used in the analysis. The areas selected by LASSO were then used to conduct a path analysis exploring the effect of treatment on SANS score change through changes in cortical thickness in these areas. Detailed methodology for conducting LASSO and path analysis can be found in the Supplement.

3. Results:

3.1. Baseline characteristics:

A total of 129 subjects were screened and 95 were randomized (49 to citalopram and 46 to placebo), with 66 subjects completing evaluations at both baseline and week 24. For CONSORT diagram and information on study attrition, please see the original manuscript.(Goff et al., 2019) Imaging results were available and met quality standards for 87 participants at baseline and 65 at week 24. Participants received antipsychotic for a mean duration of 10.3 (SD = 5.2) weeks prior to baseline evaluations. Placebo and citalopram groups did not differ on any demographic or clinical characteristic at baseline (Table 1), nor did drop-outs differ from completers on demographic or clinical characteristics (Supplement Table S5.1). Information on adverse effects of citalopram were included in the original publication on this trial.(Goff et al., 2019)

Table 1:

Baseline characteristics of placebo and citalopram groups

		Placebo		Citalopram	Test statistic
	N	Mean (SD)	N	Mean (SD)	(df)	p value
Age	45	23.69 (4.63)	49	23.20 (5.09)	t(92) = .48	.63
Ethnic Background White, No. (%) Black Asian Other	46	13 (28.3) 8 (17.4) 23 (50.0) 2 (4.3)	49	16 (32.7) 4 (8.2) 27 (55.1) 2 (4.1)	χ2(8) = 1.87	.60
Women, No. (%)	46	18 (39.1)	49	17 (34.7)	χ2(4) = 3.15	.53
DAT (weeks)	37	10.30 (5.15)	43	10.60 (5.30)	t(80) = −.26	.79
DUP (weeks)	42	29.36 (39.79)	45	53.71 (90.06)	t(61) = −1.65	.10
Baseline antipsychotic dose (mg CPZ equivalent)	42	394.03 (195.04)	47	367.76(167.25)	t(81) = .68	.50
BPRS Positive Symptoms	45	11.11 (4.62)	46	11.48 (5.02)	t(89) = −.36	.72
CDSS	45	2.58 (2.99)	46	1.83 (1.79)	t(72) = −1.45	.15
SANS Total Score	45	18.91 (12.08)	46	21.37 (14.25)	t(87) = −.89	.38
ICV (cc)	42	1537.61 (149.51)	48	1575.70 (184.60)	t(88) = −1.08	.28
LHVI, median (IQR)	41	.9380 (.9107 −.9476)	47	.9344 (.9062 −.9519)	Z = −.11	.46
RHVI, median (IQR)	41	.9411 (.9100 − .9600)	47	.9378 (.9135 − .9498)	Z = −.65	.56
LCA1	40	619.13 (62.72)	47	630.56 (74.69)	β = 4.562	.71
RCA1	40	650.19 (68.02)	47	673.81 (71.73)	β = 16.636	.16
LDG	40	316.14 (36.00)	47	316.46 (36.38)	β =−2.228	.74
RDG	40	328.98 (42.98)	47	333.09 (39.35)	β = .4050	.96

DAT: duration of antipsychotic treatment, DUP: duration of untreated psychosis, CPZ: chlorpromazine, BPRS: Brief Psychiatric Rating Scale, CDSS: Calgary Depression Scale for Schizophrenia, SANS: Scale for the Assessment of Negative Symptoms, ICV: intracranial volume, LHVI: left hippocampal volume integrity, RHVI: right hippocampal volume integrity, LCA1: left CA1, RCA1: right CA1, LDG: left dentate gyrus, RDG: right dentate gyrus.

3.2. Longitudinal HVI change over 6 months:

Baseline to followup changes in HVI are shown in Table 2 and Figure 1. In the placebo group, left HVI decreased over 24 weeks at a median annualized rate of −.00503 (IQR −.01542 to .00880), or by .52% (of baseline HVI), and on the right at a median annualized rate of −.00232 (IQR −.01816 to .00221), or by .26%. Neither differed significantly from 0 (left, t = −1.30, p = .21; right, Z = −1.07, p = .14). The annualized mean rate of reduction of LHVI in the placebo group differed significantly from a previously reported annualized mean reduction rate of 6.2% during the first 8 weeks after intiation of antipsychotics (t = 11.54, p < .0001)(Goff et al., 2018). In the citalopram group, HVI decreased on the left by .01186 (IQR −.02185 to .00413), or by 1.31%, and on the right by .01256 (IQR −.02770 to .00155), or by 1.37%. Change in left and right HVI did not differ significantly between citalopram and placebo groups (left, Z = −.33, p =.37; right, Z = −.63, p =.26). HVI results by study site can be found in Supplement Table S4.

Table 2:

Annualized Change in hippocampal volume integrity and subfield volumes at 24-week follow-up

	Placebo			Citalopram			Test statistic
	N	M(SD)/median(IQR)	Pecentage	N	M(SD)/median(IQR)	Pecentage	Z/ β	p value
LHVI	29	−.00503 (−.01542, .00880)	−.52% (−1.69%, .97%)	34	−.01186 (−.021853, .00413 0)	−1.31% (−2.31%, .45%)	Z = −.33	.37
RHVI	29	−.002316 (−.0181587, .002 214)	−.26% (−1.89%, .26%)	34	−.01256 (−.02770, .00155)	−1.37% (−2.97%, .17%)	Z = −.63	.26
LCA1	29	−5.164889 (32.18700)	−.68% (5.10%)	34	−22.239006 (37.1 2184)	−3.16% (5.82)	β = −17.46	.06
RCA1	29	−2.11922 (36.715 01)	−.35% (5.57%)	34	−21.69342 (41.80 975)	−3.10% (6.28%)	β = −20.62	.04
LDG	29	−3.862075 (14.46 023)	−1.08% (4.54%)	34	−11.477437 (21.2 5999)	−3.35% (6.80%)	β = −8.093	.09
RDG	29	−.5726257 (25.22 090)	−.08% (7.52%)	34	−9.2337609 (30.0 9944)	−2.44% (8.70%)	β = −8.809	.23

LHVI: left hippocampal volume integrity, RHVI: right hippocampal volume integrity, LCA1: left CA1, RCA1: right CA1, LDG: left dentate gyrus, RDG: right dentate gyrus.

Figure 1.

Box Plot of HVI at baseline, 24-week, and 52-week in citalopram and placebo groups

3.3. Effect of DUP on HVI:

Baseline LHVI and RHVI and the change from baseline in LHVI and RHVI were not significantly associated with DUP or duration of antipsychotic treatment (p > 0.1) and DUP did not interact with treatment group in predicting change in HVI (left, p = .57; right, p = .86). The interaction remained nonsignificant after performing a median split to divide DUP into high vs. low groups to minimize outlier effects (left, p =.23; right, p =.17) (Supplement Table S7).

3.4. Association between HVI and clinical outcomes:

At baseline, SANS total scores were positively correlated with LHVI (r=.23, p=.03) but not with RHVI (r=.13, p =.24). While HVI at baseline did not predict 24-week SANS (left, p = .58, right, p =.30), baseline LHVI had a negative interaction with treatment group such that subjects with low baseline LHVI had greater SANS score reductions at week 24 with citalopram treatment compared to placebo (p=.04, also see Figure 2). There was no significant association between change in LHVI or RHVI and change in SANS score.

Figure 2.

Effect of interaction between baseline LHVI and treatment group on 24-week SANS score change

3.5. Effect of BDNF genotype on HVI and hippocampal subfields:

BDNF GG genotype was not significantly associated with baseline HVI (left, Z = −.82, p = .21; right, Z=1.68, p=.95), nor with HVI change over 24 weeks (left, Z = −1.08, p = .14; right, Z =−.32, p=.37). Participants with the BDNF GG genotype had reduced left dentate gyrus (DG) volume reduction (left, β = 22.28, p = .007; right, β = 24.00, p = .071) compared to BDNF non-GG genotype carriers (AA, AG), and this effect was not modified by citalopram (left, p=.31, right, p = .65).

3.6. Exploratory analyses of CA1 and DG subfield volumes and cortical thickness:

Annualized changes in CA1 and DG can be found in Table 2. Changes in left and right CA1 and DG in the placebo group did not significantly differ from 0 (Supplement). Participants in the citalopram group had a greater decrease in right CA1 volume at week 24 compared to the placebo group after controlling for age, sex, ICV, and antipsychotic dosage (p=.04; Table 2). Changes in left CA1 and in left and right DG volume did not differ significantly between treatment groups. Associations between baseline CA1 and DG, changes in CA1 and DG, and symptoms can be found in the supplement; none were significant (p > 0.1). After controlling for ICV, age, and sex, changes in cortical thickness of left rostral anterior cingulate, right frontal pole, and right cuneus were negatively associated with SANS score change, and change in right transverse temporal cortical thickness was positively associated with SANS score change at 24 weeks, but none differed between treatment groups. Structural model testing that examined treatment effect through these cortical areas did not show any significant role of citalopram treatment in the model (Supplement Table S8).

4. Discussion:

Structural imaging at baseline and week 24 during this randomized, placebo-controlled trial of add-on citalopram in stabilized first episode schizophrenia patients produced several notable findings. First, we found in the group receiving antipsychotic plus placebo a non-significant median annualized rate of reduction of 0.5% in LHVI which, consistent with our hypothesis, was markedly less than the median annualized 4% reduction in LHVI that we previously observed during the initial 8 weeks of antipsychotic treatment in medication-naïve patients, suggesting that stabilization on antipsychotic treatment may be associated with attenuation of the rate of hippocampal volume loss. Second, contrary to our hypothesis, citalopram treatment was not associated with a further reduction in hippocampal volume loss and, paradoxically, was associated with significantly greater volume reduction in the right CA1 subfield. Third, and also contrary to expectation, left HVI at baseline was positively associated with SANS total score, indicating that less volume loss was associated with greater severity of negative symptoms. Greater atrophy at baseline was also associated with a greater reduction in negative symptoms with citalopram. Whereas DUP predicted hippocampal volume loss in previous medication-naïve patients (Goff et al., 2018) and predicted response of negative symptoms to citalopram in the current trial (Goff et al., 2019), DUP was not associated with baseline HVI or change in HVI, nor did it interact with treatment group in predicting HVI. Association between hippocampal volume and clinical symptoms may be complex. A recent study showed that hippocampal volume loss was indirectly associated with increased negative syptoms via verbal memory impairment (Duan et al., 2020). Another study found that volumes of multiple hippocampal subfields negatively correlated with cognitive functions in medication-naïve FES patients (Xiu et al., 2020). These findings suggest that relationships between hippocampal volume, DUP and symptoms may differ according to stage of illness and treatment and volume loss may be associated with improved function under certain conditions. In our study during the first year of maintenance treatment, preservation of hippocampal volume did not mediate citalopram’s therapeutic effect for negative symptoms.

Hippocampal volume change during initial antipsychotic treatment has been well-studied (Goff et al., 2018; Nakahara et al., 2018), whereas change during the period immediately following initial stabilization on antipsychotic medication has rarely been studied. Our finding of non-significant volume reduction during early maintenance treatment is consistent with a finding from a similar study in which Breier and colleagues (Breier et al., 2018) randomized 60 stabilized first episode psychosis patients to add-on NAC or placebo and found improvement of negative symptoms with NAC but no change in hippocampal volume over 24 months in the placebo group and no effect of NAC on hippocampal volume. While we must be cautious in comparing the rates of hippocampal volume loss between our previous study of medication-naïve patients (Goff et al., 2019) and our current study, the median 4% annualized reduction in LHVI that we observed during the first 8 weeks of antipsychotic treatment in medication-naïve patients at Shanghai Mental Health Center in our previous study was substantially greater than the 0.5% annualized reduction that we observed in all participants receiving placebo in the current study and greater than the 1.5% (n = 9) reduction observed in patients receiving placebo studied at Shanghai Mental Health Center in the current study. Participants from this site in the current study were quite similar in age, sex, and antipsychotic treatment, and were assessed with the same MRI scanner as the patients in the prior study (Goff et al., 2019).

We did not observe significant change over 6 months in DG or CA1 subfield volumes in the placebo group. Longitudinal change in hippocampal subfields may also be stage-specific, as Li and colleagues (Li et al., 2018) found significant reductions in total hippocampal and CA1 and DG volumes during the first 6 weeks of antipsychotic treatment in medication-naïve patients and Ho and colleagues (Ho et al., 2017b) found a mean 6% annualized reduction in CA1 volume in a sample of patients with a mean duration of illness of 6 years. Ho and colleagues (Ho et al., 2017b) also found that left CA1 volume loss correlated with negative symptoms whereas right CA1 volume loss correlated with positive symptoms.

We did not observe a “neuroprotective” effect of citalopram on hippocampal volume, although the absence of significant volume reduction over 6 months in the placebo group limited our ability to test a neuroprotective effect. Paradoxically, when we examined hippocampal subfields in an exploratory analysis, we found that the citalopram group exhibited greater volume loss in the right CA1 subfield compared to placebo, despite improvement of negative symptoms. This finding is contrary to our hypothesis, which was based on SSRI effects in patients with depression. Of note, Willard and colleagues (Willard et al., 2015) found that sertraline increased right anterior hippocampal volume in depressed cynomegulus monkeys and decreased right anterior hippocampal volume in nondepressed cynomegolus monkeys, while improving affiliative behaviors in the non-depressed monkeys.

Our findings overall suggest that citalopram’s therapeutic effect on negative symptoms was unlikely to be mediated by preservation of brain volume. Other mechanisms could be underlying the therapeutic process. The SSRI fluoxetine was found to normalize hippocampal functional connectivity in a mouse model of Down’s syndrome (Stagni et al., 2013). Studies of sertraline in individuals with obsessive compulsive disorder (OCD) have found increases in functional connectivity associated with improvement in OCD symptoms (Bernstein et al., 2019; Shin et al., 2014); the mechanism by which SSRIs affect functional connectivity is not clear, but this may be a more promising model for examining the therapeutic effects of citalopram on negative symptoms. Alternatively, SSRIs are hypothesized to return the brain to an enhanced state of plasticity characteristic of early development (Kobayashi et al., 2010); which might contribute to improvement of negative symptoms by remodeling of microcircuits (Castren, 2013). During brain development, BDNF release is permissive for synaptic pruning (Singh et al., 2008) and in the presence of an impairment of synaptic pruning, as has been postulated in schizophrenia, it is possible that citalopram might promote pruning and hence be associated with volume loss. Thus, not all brain volume loss observed during treatment in schizophrenia may be pathological—it may vary depending on the stage of treatment and the molecular charateristics of the medication (Lesh et al., 2015; Li et al., 2018).

Our exploratory analysis found 4 cortical regions associated with change in negative symptoms. Some of these regions were previously reported to be associated with domains of cognitive and affective impairment in schizophrenia, such as error processing (Laurens et al., 2003) and positive subjective experience(Wacker et al., 2009) (rostral anterior cingulate cortex), affective appraisal (Polli et al., 2007), social cognition and decision making (Hiser and Koenigs, 2018) (frontal pole) and visual (cuneus) and auditory processing (transverse temporal cortex) (Mørch-Johnsen et al., 2016; Pantelis et al., 2003). Because citalopram treatment did not modify cortical thickness change in these regions, it is unlikely that citalopram reduced negative symptom burden by protecting against brain atrophy in these regions.

4.1. Limitations:

We did not correct for multiple comparisons in exploratory analyses. Therefore, results related to hippocampal subfields and cortical thickness should be considered preliminary. In addition, we were limited by sample size in our power to detect treatment effects on hippocampal volume loss, and by the heterogeneity introduced by four study sites. However, results were consistent across study sites.

4.2. Conclusion:

Consistent with our hypothesis, we found that hippocampal volume loss was attenuated in first episode psychosis patients during the first 6 months after stabilization on antipsychotic medication compared to a previous sample of medication naïve patients studied during the first 8 weeks of treatment. This finding suggests that early maintenance treatment with antipsychotic medication may reduce hippocampal volume loss, consistent with a neuroprotective effect. We did not find evidence that citalopram’s benefit for negative symptoms was mediated by an additional preservation of hippocampal volume and did not find significant associations between hippocampal volume loss, negative symptoms and DUP, suggesting that previously-reported associations between these factors may reflect the stage of treatment.

Highlights.

• Hippocampal atrophy is prominent in early psychosis and has been linked to negative symptoms.

• Add-on citalopram during the maintenance phase of treatment improved negative symptoms

• Citalopram did not affect hippocampal volume over 6 months

• Cortical thickness in 4 regions correlated with negative symptoms, unrelated to citalopram.

• Effects on brain volume did not account for citalopram’s improvement of negative symptoms