Ventral prefrontal serotonin 1A receptor binding: a neural marker of vulnerability for mood disorder and suicidal behavior?

Abstract

Mood disorders and suicidal behavior have moderate heritability and are associated with altered corticolimbic serotonin 1A receptor (5-HT_1A) brain binding. However, it is unclear whether this reflects genetic effects or epigenetic effects of childhood adversity, compensatory mechanisms, or illness stress-related changes. We sought to separate such effects on 5-HT_1A binding by examining high familial risk individuals (HR) who have passed through the age of greatest risk for psychopathology onset with and without developing mood disorder or suicidal behavior. PET imaging quantified 5-HT_1A binding potential BP_ND using [11C]CUMI-101 in healthy volunteers (HV, N=23) and three groups with one or more relatives manifesting early-onset mood disorder and suicide attempt: 1. unaffected HR (N=23); 2. HR with lifetime mood disorder and no suicide attempt (HR-MOOD, N=26); and 3. HR-MOOD with previous suicide attempt (HR-MOOD+SA, N=20). Findings were tested in an independent cohort not selected for family history (HV, MOOD, and MOOD+SA, total N=185). We tested for regional BP_ND differences and whether brain-wide patterns distinguished between groups. Low ventral prefrontal 5-HT_1A BP_ND was associated with lifetime mood disorder diagnosis and suicide attempt, but only in subjects with a family history of mood disorder and suicide attempt. Brain-wide 5-HT_1A BP_ND patterns including low ventral prefrontal and mesiotemporal cortical binding distinguished HR-MOOD+SA from HV. A biological endophenotype associated with resilience was not observed. Low ventral prefrontal 5-HT_1A BP_ND may reflect familial mood disorder and suicide-related pathology. Further studies are needed to determine if higher ventral prefrontal 5-HT_1A BP_ND confers resilience, reducing risk of suicidal behavior in the context of familial risk, and thereby offer a potential prevention target.

Introduction

Mood disorders including unipolar and bipolar depression (MOOD) and suicidal behavior are associated with morphological brain and serotonin (5-HT) system abnormalities that may mediate genomic liability (1–5). The offspring of individuals with early-onset depression are at higher risk of developing mood disorder and suicidal behavior (6,7). The identification of potentially heritable, biological markers of risk or resilience for mood disorders and suicide is important. Such markers may help to identify persons at elevated risk of developing a mood disorder or attempting suicide and thereby, can facilitate prevention efforts.

The pathology of mood disorders and suicide risk have been consistently linked to alterations in brain serotonin 1A receptor (5-HT_1A) binding potential using Positron Emission Tomography (PET) and postmortem brain studies of suicide decedents. Most studies demonstrating lower corticolimbic 5-HT_1A receptor binding in depressed patients have expressed the binding in terms of binding potential BP_ND (5,8). We have previously reported that binding potential BP_F was elevated in depressed patients both during an acute episode (9,10) and while in remission, suggesting this is a biologic trait (11). Subsequently, we reported a pilot study that suggested altered brain 5-HT_1A binding potential may be familially transmitted from depressed parents to their children and may be an endophenotype differentiating individuals at higher risk for developing major depressive disorder (12). To our knowledge, this was the first study to examine brain 5-HT_1A binding potential in a small sample of unaffected 1st degree relatives of individuals with a mood disorder and lifetime history of suicide attempt. This study suggests that 5-HT_1A receptor binding abnormalities are not purely a consequence of illness or treatment. However, a paucity of PET studies of the serotonin system in unaffected relatives means it is unknown whether alterations in brain 5-HT_1A binding potential are familial traits that confer risk or resilience for developing a mood disorder and suicidality.

Notably, the current study used a different radioligand than our previous related work mentioned above. Regarding differences in radiotracers, [¹¹C]WAY-100635, used in our previous papers (9–12), is a 5-HT_1A antagonist that bind with equal affinity to the coupled and uncoupled conformations, while [¹¹C]CUMI-101, used here, as a biased agonist, may predominantly bind only to receptors in the high affinity conformation (13,14). An agonist and an antagonist tracer may measure different 1A receptor spatial distributions and binding affinity.

To distinguish differences associated with risk for developing psychiatric disorder from differences related to compensatory adaptation possibly marking resilience or illness related changes, the current study examined regional differences in regional 5-HT_1A binding potential using [11C]CUMI-101 in low risk healthy volunteers (HV) and in medication-free high-risk individuals, having a first or second degree relative with both an early onset mood disorder and a history of suicidal behavior. High risk individuals were either unaffected despite being HR and recruited when largely beyond the age of risk for mood and suicidal behavior), or affected with lifetime history of mood disorder alone (HR-MOOD) or mood disorder plus one or more lifetime suicide attempts (HR-MOOD+SA). Based on previous findings from our group and others (5,9–11,15,16), we hypothesized altered regional brain 5-HT_1A receptor binding potential in both HR-MOOD and HR-MOOD+SA relative to HV, and explored whether HR would display different alterations suggesting a biological phenotype related to resilience for mood disorders or suicidal behavior. We also tested whether statistical learning applied to whole-brain patterns of 5-HT_1A receptor binding potential could distinguish HVs from unaffected and affected high risk individuals. This complementary analysis increased power to detect and characterize patterns of brain-wide abnormalities in 5-HT_1A receptor binding potential that may further elucidate the pathophysiology of mood disorder and suicidal behavior. Positive findings may have clinical utility in identifying high risk individuals before onset of psychopathology.

To address these objectives, an exploratory, voxel-wise analysis was conducted in the primary “discovery” dataset (FAMPATH) to identify candidate clusters showing group differences in 5-HT_1A receptor binding potential. These clusters were then employed as regions of interest (ROIs) to be tested for replication in ROI-level analyses in an independent dataset of healthy volunteers and individuals affected with current or lifetime mood disorder and often also lifetime history of suicide attempt, that was available from two additional studies using the same [11C]-CUMI-101 PET tracer. The subjects in this dataset had current or lifetime history of major depressive disorder and were not recruited with the stringent family history requirements for FAMPATH. Similarly, multivoxel pattern analysis (MVPA) was applied to voxel-wise PET estimated parametric maps of binding potential to distinguish between the four groups in the primary discovery FAMPATH dataset (HV, HR, HR-MOOD, and HR-MOOD+SA), and the best performing model was applied to the second independent dataset to test its generalizability. Finally, we compared the results to those in the HR group to infer whether observed differences reflected putative markers of risk, resilience or compensatory mechanisms for developing a mood disorder and suicidal behavior.

Methods and Materials

Inclusion/exclusion criteria for the primary dataset: FAMPATH

The FAMPATH sample consisted of high-risk individuals and healthy volunteers (HV) who were 25 years of age or older, an age by which mood disorder and suicidal behavior in our familial transmission studies are generally phenotypically clinically expressed (17,18). All high-risk subjects were categorized as being at-risk for a mood disorder and suicide because they had a first or second degree relative with early-onset (<40 years) mood disorder and a lifetime history of suicide attempt. The subjects in the high-risk group were separated into vulnerable and resilient groups based on their phenotype: individuals with a lifetime history (current or past) of mood disorders and a history of suicide attempt (HR-MOOD+SA, n=20); individuals with a lifetime history (current or past) of mood disorders but no suicide attempt (HR-MOOD, n=26); and resilient (HR, n=23) individuals with no personal history of mood disorder or suicide attempt. Qualifying mood disorders included major depressive disorder, bipolar disorder, dysthymia and depressive disorder not otherwise specified. The HR-MOOD sample consisted of 22 participants with major depressive disorder and 4 participants with other diagnoses. The HR-MOOD+SA sample consisted of 16 participants with major depressive disorder, and 4 with bipolar disorder. Healthy volunteers (HVs, n=23) had no personal or family history of psychiatric disorders or suicidal behavior. The sample was recruited at two sites, University of Pittsburgh Medical Center and New York State Psychiatric Institute, in protocols approved by the respective Institutional Review Boards (IRB) at the University of Pittsburgh and The New York State Psychiatric Institute. Participants were recruited through outpatient clinics, clinician referrals and IRB-approved advertisements and all participants provided written, informed consent. See Supplementary Methods for assessments and additional inclusion/exclusion criteria for FAMPATH and the second independent dataset.

Final sample sizes across the three included studies

The primary discovery dataset (FAMPATH) consisted of 23 healthy volunteers (HV), 23 high familial risk (HR) participants, 26 HR with a lifetime history of mood disorder (HR-MOOD), and 20 HR-MOOD with one or more lifetime suicide attempts (HR-MOOD+SA). The second independent dataset included 36 HV, 38 individuals with current mood disorder (MOOD), and 19 participants with MOOD and one or more lifetime suicide attempts (MOOD+SA). The total sample size of the second independent dataset is N=93, while the total across both sets is N=185. Note that data in the second dataset was acquired and analyzed the same way as was the FAMPATH cohort in terms of PET imaging acquisition, binding potential quantification and subsequent analysis. Our group has not previously published on 5-HT_1A PET imaging findings in this second cohort.

PET/MR scanning, image preprocessing and binding quantification

See Supplementary Methods for PET/MR scanning and imaging preprocessing procedures. PET procedures to quantify 5-HT_1A binding potential BP_ND were described in detail elsewhere (19). Our primary outcome measure is the binding potential BP_ND= f_NDBavail/K_D, where f_ND is the fraction of free tracer in the non-displaceable tissue compartment, B_avail is the concentration of 5-HT_1A receptor that are unoccupied or available to bind to the tracer, and K_D is the tracer equilibrium dissociation constant. BP_ND is used as its quantification does not require blood sampling. BP_ND was quantified at the voxel level using the reference tissue version of the likelihood estimation in graphical analysis (19) which at the voxel level proved to be the method providing the closest estimation of BP_ND to values obtained based on arterial blood for [¹¹C]CUMI101 (19). See Supplementary Methods for quantification of BP_P and V_T for subjects with usable venous samples (~65% of the total sample).

Imaging processing and analysis

Voxel-wise univariate analyses:

Analyses were conducted using SPM12 (www.fil.ion.ucl.ac.uk/spm/) and implemented in MATLAB version 7.13 on Ubuntu Linux OS 14.04. Voxel-level PET binding maps were spatially normalized to a 2 × 2 × 2 mm MNI-space template using Advanced Normalization Tools (ANTs) (20) and smoothed with a 8 mm full width at half maximum FWHM Gaussian filter. The images were submitted to a 1-way ANCOVA with four levels (HV, HR, HR-MOOD, HR-MOOD+SA) in the primary FAMPATH dataset and three levels (HV, MOOD, MOOD+SA) in the second dataset. An absolute threshold was applied to remove voxels with BP_ND values below 0.05, and non-gray matter voxels were excluded from analyses via a gray matter mask generated by thresholding a tissue probability map in MNI space (provided with SPM8) at > 0.2. Sex, age and scan site were included as nuisance covariates for both datasets.

A cluster-determining threshold (CDT) of p<0.01 and k>150 was used to identify candidate regions in the FAMPATH primary discovery dataset that differed between groups (HV, HR, HR-MOOD, and HR-MOOD+SA). These clusters were then used to define ROIs for ROI analysis in the second independent dataset. Two types of ROI analyses were conducted for each of the clusters identified in the primary discovery dataset. The first analysis used Marsbar v0.44 (http://marsbar.sourceforge.net) and the signal (i.e. voxel-wise beta estimates from 1st level analyses) within each ROI was first averaged together before model estimation. In the second analysis, an anatomical ROI mask generated using WFU Pickatlas, was used for small volume correction (SVC) in which models are estimated across all voxels within the mask. For the latter, correction for multiple comparisons was conducted using 3dClustSim (compiled December 11, 2018) with the -acf option (input parameters estimated using residuals from the SPM design file) and a cluster determining threshold of p < 0.05 for cluster sizes defined using N=1 nearest neighbors and 2-sided testing. This approach was based on recent recommendations to reduce false-positive rates when using cluster-extent correction (21,22) and includes a more accurate estimate of the noise smoothness values using a mixed model (Guassian plus a monoexponential) (23). Results were also confirmed using nonparametric permutation for cluster-extent correction (FSL’s randomise function with 2000 permutations and cluster determining threshold t-value=1.96).

Voxel-wise multivariate statistical learning analyses:

Multivoxel pattern analysis (MVPA) analyzed the joint BP_ND signal across multiple regions in a single subject in order to predict the diagnostic class of that subject. Briefly, binary classification tasks using a linear kernel Support Vector Machine SVM (24) with a filter feature selection (t-test) and leave-one-out cross-validation was applied using the Spider v1.71 MATLAB toolbox (http://people.kyb.tuebingen.mpg.de/spider/) with default regularization parameter C = 1. Classification performance for each top N selected features (voxels) was assessed using non-parametric permutation. See Supplementary Methods for more details.

The above procedures were repeated within the FAMPATH and second independent dataset for each pairwise group comparison within each cohort. In addition, to test whether the models learned in the FAMPATH dataset are generalizable to individuals not selected for family history, a model trained over the full FAMPATH sample (using the top N features that gave the best classification performance) was applied to the second independent dataset.

Results

Demographic and clinical results

Within both the FAMPATH and second dataset, the subgroups did not differ in sex ratio (p-values > 0.2). In the second dataset the MOOD+SA were younger, and the MOOD older, compared with the HV group (ANOVA omnibus p<0.001, see Table 1). The second dataset had a smaller MOOD+SA group because one study from which the sample was drawn did not actively recruit individuals with one or more lifetime suicide attempts. The HR-MOOD and HR-MOOD+SA groups in the primary FAMPATH dataset had relatively lower current Beck Depression Inventory (BDI) and Hamilton Depression Rating Scale (HAM17) scores than in the second dataset. Participants in the FAMPATH were not required to be currently ill because we previously reported (11) that altered 5-HT1A binding was a trait that was present during acute depressive episodes and when in remission between episodes (Table 1). The second dataset MOOD and MOOD+SA subgroups had higher BDI and HAM17 scores because the sample involved currently depressed subjects (Table 1).

Table 1.

Demographics and behavioral data.

FAMPATH	HV (N=23)	HR (N=23)	HR-MOOD (N=26)	HR-MOOD+SA (N=20)	pval
Age: mean (iqr)	39.91 (15.25)	39.22 (20.50)	36.46 (15.00)	37.30 (17.00)	0.62
HAM17: mean (iqr)	0.91 (1.75)	1.13 (1.75)	4.04 (6.00)	10.30 (12.50)	0
BDI: mean (iqr)	1.35 (1.75)	2.35 (4.75)	6.69 (10.00)	17.35 (15.00)	0
Sex: count (%)	7/16 (0.30/0.70)	7/16 (0.30/0.70)	9/17 (0.35/0.65)	7/13 (0.35/0.65)	0.98
FamDep: count (%)	23/0 (1.00/0.00)	4/18 (0.18/0.82)	4/22 (0.15/0.85)	1/16 (0.06/0.94)	0
FamSui: count (%)	23/0 (1.00/0.00)	4/18 (0.18/0.82)	4/21 (0.16/0.84)	5/12 (0.29/0.71)	0
Site: count (%)	13/10 (0.57/0.43)	7/16 (0.30/0.70)	10/16 (0.38/0.62)	9/11 (0.45/0.55)	0.33
SECOND DATASET	HV (N=36)	N/A	MOOD (N=38)	MOOD+SA (N=19)	pval
Age: mean (iqr)	32.14 (13.00)		35.74 (18.00)	27.21 (12.50)	0
HAM17: mean (iqr)	0.92 (1.00)		17.63 (6.00)	17.00 (11.00)	0
BDI: mean (iqr)	0.92 (1.00)		25.11 (9.00)	23.22 (6.00)	0
Sex: count (%)	14/22 (0.39/0.61)		16/22 (0.42/0.58)	8/11 (0.42/0.58)	0.95
FamDep: count (%)	35/0 (1.00/0.00)		16/18 (0.47/0.53)	10/9 (0.53/0.47)	0
FamSui: count (%)	31/0 (1.00/0.00)		31/2 (0.94/0.06)	16/3 (0.84/0.16)	0.07
site: count (%)	7/29 (0.19/0.81)		10/28 (0.26/0.74)	1/18 (0.05/0.95)	0.17

Group differences were assessed using Kruskal–Wallis one-way ANOVA (continuous variables) or chi² test (categorical variables). Note that FAMPATH required HR participants have a first-degree or second-degree relative with both a qualifying mood disorder and history of suicidal behavior. The FamDep and FamSui variables indicate number of participants with one or more first-degree relatives with MDD or suicide attempt, respectively. FamDep and FamSui are reported here since these variables were the same and accessible across both datasets. iqr=interquartile range.Table 1.

As expected, the FAMPATH HR-MOOD and HR-MOOD+SA subgroups had higher rates of familial mood disorder relative to the second dataset which was not selected for family history (85–94% vs. 47–53% subjects with one or more first degree relatives with major depressive disorder). They also had higher rates of familial suicidal behavior (~71–84% vs. 6–16% of subjects with one or more first degree relatives with lifetime suicide attempt). Among the HR subgroup of FAMPATH, 82% had at least one first degree relative with MDD, and 82% had at least one first degree relative who made a nonfatal suicide attempt; the remainder had at least one second degree relative with lifetime MOOD and suicide attempt (see Table 1).

Voxel-based 5-HT1A Receptor Binding Univariate Analyses

A whole-brain omnibus F-test thresholded at p<0.01, k>150 testing for differences between the four groups (HV, HR, HR-MOOD and HR-MOOD+SA) in the FAMPATH dataset revealed two clusters: a cluster in the ventromedial prefrontal cortex (vmPFC, Figure 1A, top row) and a rostral cluster in medial orbitofrontal cortex (mOFC, see Figure 1A, bottom row). Post hoc t-tests show lower 5HT_1A BP_ND in vmPFC and mOFC in HR-MOOD vs. HV (t(85)=3.12, p=0.001 and t(85)=3.67, p=0.0002), HR-MOOD+SA vs. HV (t(85)=3.50, p=0.0004 and t(85)=3.08, p=0.001), HR-MOOD vs. HR (t(85)=2.44, p=0.008 and t(85)=2.27, p=0.01), and HR-MOOD+SA vs. HR (t(85)=2.83, p=0.003 and t(85)=1.75, p=0.04) (see Table 2).

Figure 1. Individuals with both lifetime and family history of mood disorder and suicide attempt have lower 5-HT_1A binding in ventromedial prefrontal and orbitofrontal cortex with respect to healthy volunteers and patients without family history for these traits.

A) A voxel-wise, whole brain omnibus F-test in the discovery dataset (FAMPATH) identified two clusters with different 5-HT_1A binding potential (BP_ND) among four groups: 1. healthy volunteers (HV, N=23) and three groups with one or more relatives with early-onset mood disorder and suicide attempt: 2. unaffected high risk individuals (HR, N=23), 3. individuals with lifetime mood disorder (HR-MOOD, N=26), and 4. HR-MOOD individuals with lifetime history of suicide attempt (HR-MOOD+SA, N=20) (p<0.01 uncorrected, k>150, left column). One cluster was in ventromedial prefrontal cortex (vmPFC, top row), and the other was located more ventrally in medial orbitofrontal cortex (mOFC, bottom row). Each cluster was used to define ROIs for ROI-level analyses in the second independent dataset (middle column). The second dataset included HV (N=36), individuals with mood disorder (MOOD, N=38), and depressed individuals with one or more lifetime suicide attempt (MOOD+SA, N=19). Group differences in 5-HT_1A binding potential in either cluster were not observed in this dataset (p-values > 0.2 uncorrected). Five of the MOOD and MOOD+SA subjects in this second dataset were considered “high familial risk” (“HR-patients”) on the basis of having at least one first degree relative with MDD and at least one first degree relative with lifetime suicide attempt. The HV and HR non-patients combined had higher median 5-HT_1A binding potential than the 5 HR-patients in both regions, though neither cluster reached statistical significance based on Mann Whitney U-test (right column). B) An anatomical mask defined using the WFU Pickatlas (left) was used for a voxel-wise small volume cluster-extent corrected ROI analysis. Lower 5-HT_1A binding potential in the replication sample positive family history-subjects vs. HV and non-familial depressed and suicide attempter subjects was observed in subgenual anterior cingulate cortex at a one tailed level of statistical significance (p=0.07 corrected, right).

Table 2.

Post hoc t-tests for pairwise contrasts between groups using ROI analysis (Marsbar, see methods).

STUDY AND ROI	Contrast 1		Contrast 2		Contrast 3		Contrast 4		Contrast 5		Contrast 6
FAMPATH	HV > HR		HV > HR-MOOD		HV > HR-MOOD+SA		HR > HR-MOOD		HR > HR-MOOD+SA		HR-MOOD > HR-MOOD+SA
	t(85)	p-val	t(85)	p-val	t(85)	p-val	t(85)	p-val	t(85)	p-val	t(85)	p-val
vmPFC	0.69	0.25	3.12	0.001	3.50	0.0004	2.44	0.008	2.83	0.003	0.57	0.29
mOFC	1.37	0.09	3.67	0.0002	3.08	0.001	2.27	0.01	1.75	0.04	−0.40	0.65
SECOND DATASET	HV and non-HR patients > “HR”*		HV > MOOD		HV > MOOD+SA		N/A		N/A		MOOD > MOOD+SA
	diff	p-val	t(87)	p-val	t(87)	p-val					t(87)	p-val
vmPFC	0.11	0.16	0.21	0.42	−0.14	0.56					0.11	0.46
mOFC	0.08	0.37	−0.71	0.76	0.29	0.39					0.42	0.34

^*The diff column lists the difference in medians between the two groups, and p-val is based on a Mann-Whitney U-test. See Figure 1A (right panel) and Figure 1B for voxel-wise results when using an anatomic ROI and small volume cluster-extent correction.

vmPFC=dorsal cluster in Figure 1, top row; mOFC=ventral cluster in Figure 1, bottom row; All reported p-values in the table are one-sided since we hypothesized specific directions in the second independent dataset based on FAMPATH dataset results.

The two clusters were used as ROIs to test for group differences (HV, MOOD, MOOD+SA) in the second dataset. In this dataset, an omnibus F-test found no evidence for group differences in either cluster (p-values > 0.2, Figure 1A, middle panel). Since the participants in this cohort were not selected for family history (in contrast to FAMPATH) we tested whether group differences might emerge when the participants were stratified by familial risk (“HR”) status. For this analysis in the second dataset, 5 subjects (2 MOOD and 3 MOOD+SA) had at least one first-degree relative with MDD and at least one 1st degree relative with lifetime suicide attempt (HR-patients, see Table 1). The median 5-HT_1A BP_ND in the vmPFC and mOFC clusters for the 5 subjects was compared to all subjects in the second dataset without such family history (HV and non-HR MOOD and MOOD+SA, or non-HR, patients). No significant differences in 5-HT_1A BP_ND were found in either cluster (vmPFC: HR-patients median=0.77, HV and non-HR patients median=0.88, Mann-Whitney U test p=0.16; mOFC: HR-patients median=1.00, HV and non-HR patients median=1.08, Mann-Whitney U test p=0.37, see Figure 1A, right panel). We also tested for lower binding in positive family history subjects in ventral prefrontal cortex using voxel-wise testing constricted to an anatomical ROI mask and small volume cluster-extent correction. This analysis revealed a cluster in ventral prefrontal cortex in the vicinity of subgenual anterior cingulate cortex that showed lower 5-HT_1A BP_ND in positive family history subjects relative to HV and non-HR patients that was statistically significant in a one-tailed comparison (p=0.07 cluster extent corrected, Figure 1B, right).

Voxel-based multivariate learning analysis results

Multivoxel pattern analysis (MVPA) was applied to distinguish between the four diagnostic groups in the primary FAMPATH dataset using the whole-brain, voxel-wise 5-HT_1A binding potential PET maps. Of the six total pairwise comparisons (HV vs. HR, HV vs. HR-MOOD, HV vs. HR-MOOD+SA, HR-MOOD vs. HR-MOOD+SA, HR vs. HR-MOOD and HR vs. HR-MOOD+SA), significant classification performance was achieved for HR-MOOD+SA vs. HV (p<0.001 corrected). The peak performance was attained when using the top 200 features/voxels (AUC=0.74, sensitivity=0.78, specificity=0.70, Figure 2, left panel). Regions that figured heavily into the discrimination included medial orbitofrontal cortex, ventromedial prefrontal cortex, inferior and anterior temporal cortex (Figure 2, right panel). A best performing model was trained over the full FAMPATH dataset using the top 200 selected features (voxels). The learned model was then applied to the second dataset to classify HV vs. MOOD+SA. Classification performance was greater than the theoretical null classification (AUC=0.55) but did not reach statistical significance (p=0.15, data not shown).

Figure 2. Voxel-wise PET binding distinguishes healthy volunteers from depressed suicide attempters.

(A) Out-of-sample classification performance (Area under the ROC curve, or AUC) in the primary FAMPATH dataset was plotted vs. number of features that have been ranked by their absolute t-score (in the training data). Three of the six total pairwise comparisons are plotted (HV vs. HR-MOOD, HV vs. MOOD+SA, and HR-MOOD vs. HR-MOOD+SA, the rest are not shown). Above chance classification for HV vs. MOOD+SA was achieved across all top N features (p<0.001 corrected for multiple comparisons), with peak performance when using top 200 selected features (highlighted box, AUC=0.74, sensitivity=0.78, specificity=0.70). Solid gray line plots the mean of the null distribution and error bars represent the spread such that 95% of the null values fall within the lower and upper bounds). (B) The top 200 features (voxels) over the entire training (primary FAMPATH) dataset were used to train a classification model and plot the informative features on the brain. Informative regions included medial orbitofrontal cortex, ventromedial prefrontal cortex, inferior and anterior temporal cortex (top two rows). Each voxel was color coded according to its weight score from the SVM model learned from the full dataset using the top 200 features. Red indicates positive weights, blue indicates negative weights.

Discussion

Across two independent datasets, we observed evidence for lower 5-HT_1A BP_ND receptor binding in high familial risk MOOD and MOOD+SA (vs. HV) individuals in ventromedial prefrontal and medial orbitofrontal cortex. The directionality of these findings is consistent with studies by other groups that have reported lower 5-HT_1A binding potential BP_ND in limbic, frontal and temporal cortices in major depression (15,25–27) and in remitted depressed patients (28). Our findings are also consistent with a 2016 meta-analysis of 10 studies (including 218 patients with MDD and 261 HV) that found lower 5-HT_1A receptor binding potential BP_ND in various corticolimbic ROIs, although vmPFC and mOFC did not appear to be included in the meta-analysis (5). In the current study, low ventral prefrontal 5-HT_1A BP_ND binding potential was found in groups with mood disorder alone or with mood disorder and suicidal behavior, but only in individuals with a family history of both mood disorder and suicidal behavior. Therefore, low ventral prefrontal 5-HT_1A BP_ND may reflect suicide- and mood disorder-related neuropathology in the context of high familial risk for suicide.

A related paper by our group (Melhem et al. 2021) that used the same FAMPATH cohort found no group differences in 5-HT_1A BP_ND (29). The key methodological difference with Melhem et a. 2021 is that this study estimated and examined group differences in voxel-wise 5-HT_1A BP_ND, whereas Melhem et al. 2021 used ROI-level quantification of 5-HT1A BP_ND in a priori regions, where only the average binding within a specified region of interest is estimated. Melhem 2021 et al. did not detect group differences in any of the ROIs tested. An anatomical ROI-based analysis is less likely to detect group differences if the signal is confined to a relatively small area within a given ROI parcellation, since the signal would be diluted by other voxels within the ROI that do not show a group difference. This problem is worse for larger ROIs like the prefrontal cortex.

The vmPFC and mOFC are involved in depression-relevant processes, including emotional regulation, reward processing, and negative self-appraisal (30) and have been previously implicated with pathophysiology of both mood disorders and suicide across a range of analysis techniques including neuroimaging, neuropathologic and lesion analysis (31,32). Our findings also converge with a recent comprehensive review of two decades of suicide neuroimaging studies, that suggests that impairments in ventral prefrontal cortex play a role in suicide ideation and development of suicidal behavior (4).

According to a model proposed in (33), neural differences shared between affected and unaffected HR individuals compared to low-risk HV are putative markers of risk, whereas neural changes differentiating affected and unaffected HR may reflect compensatory, protective, or resilience markers. In our primary FAMPATH dataset, the 5-HT_1A BP_ND in vmPFC and mOFC was lower in HR-MOOD and HR-MOOD+SA compared with HR, and the HR levels were comparable to those observed in the low-risk HV group (see Table 2). Moreover, the MVPA did not distinguish HR from the HV group. Therefore, low ventral prefrontal 5-HT_1A BP_ND is not an abnormality associated with familial risk or resilience since it is absent in the HR group.

Low ventral prefrontal 5-HT_1A BP_ND may have a causal relationship or be part of the pathogenesis of MOOD and MOOD+SA in individuals with familial risk. Because we did not observe lower ventral prefrontal 5-HT_1A BP_ND in MOOD+SA without a family history, it appears that low 5-HT_1A BP_ND in this brain region may mediate familial causal factors responsible for MOOD and MOOD+SA. It should be noted that without a longitudinal study of high-risk individuals that shows these biologic changes are present before the development of symptoms, one cannot rule out the possibility that these brain changes in binding are a consequence of the illness in terms of being either damage due to the illness or a homeostatic response. However, that would not explain the familial transmission and we previously reported altered 5-HT_1A binding in the brains of asymptomatic offspring of depressed parents and those with the most abnormal binding went on to develop a subsequent depression (12). If the findings of this pilot study can be replicated and generalized to MOOD+SA, that could rule out homeostatic responses or biologic change as a response to active illness as an explanation of our findings. Further studies are needed to determine if higher 5-HT_1A binding potential in vmPFC/mOFC or other brain regions may confer resilience to developing suicidal behavior in the context of mood disorder. If so, it could be a potential mood disorder-suicide prevention target.

Consistent with the univariate analysis which found the largest and most consistent group differences between HV and HR-MOOD+SA, MVPA distinguished between HV vs. HR-MOOD+SA with above chance performance, with informative voxels located in inferior mesiotemporal cortex and vmPFC/mOFC. The MVPA could not distinguish between any of the other group pairs. Moreover, the best fitting model for HV vs. HR-MOOD+SA did not distinguish HV vs. MOOD+SA in the replication dataset. A possible reason that MVPA could not distinguish between the other group pairs and that the FAMPATH HV vs. HR-MOOD+SA model did not generalize to the second HV vs. MOOD+SA dataset is that the MVPA picked up signals in the brain-wide 5-HT_1A binding potential patterns that were specific to having high familial risk for both mood disorder and suicidal behavior. Future studies of HR-MOOD+SA are required to test whether our multivariate pattern classification results generalize to other samples.

Limitations

Previous studies, including by our group, have reported findings that contradict the findings of the current study. Specifically, we previously reported higher 5-HT_1A receptor binding potential BP_P and/or BP_F during a major depressive episode (9,10), in remitted depressed patients when between episodes and unmedicated (11), and in HR offspring of depressed patients (Milak et al., 2018). The discrepancy with our current results is likely explained by differences in outcome measure (8,34) and choice of reference region (white matter cerebellum for our previous results vs. gray matter cerebellum here), and by differences in the characteristics of the radiotracers used to quantify 5-HT_1A binding potential. Our previous studies used BP_F, while this study used BP_ND as the primary outcome measure since only venous plasma measures, required for calculation of BP_F, BP_P and V_T outcomes, were available only for a subset (~65%) of the subjects and we wanted to maximize sample sizes. When using the subset of subjects in FAMPATH with BP_P and V_T images, the main findings (higher 5-HT_1A binding in HV and HR vs. HR-MOOD and HR-MOOD+SA) were consistent with results obtained using BP_ND in the full sample in terms of the direction of the findings, but with higher p-values (HV+HR > HR-MOOD+HR-MOOD+SA, p-values between 0.04 and 0.25 for BP_P and V_T in both vmPFC and mOFC clusters, data not shown).

Our MVPA analysis demonstrates alterations in the pattern of brain-wide 5-HT1A BP_ND that distinguishes HR-MOOD+SA from low risk HV. The model did not generalize to MOOD+SA with low familial risk in the second dataset. Future studies with larger sample sizes are needed to determine if our findings can be extended to lower risk suicidal behavior in the context of mood disorder.

Finally, we did not find direct evidence for a risk or resilience biological endophenotype. It is possible that other biologic systems, perhaps not involving the serotonin system, provide the resilience seen clinically in the HR group we identified who had no evidence of having developed mood disorder or manifested suicidal behavior.

Conclusion

In summary, our findings suggest lower ventral prefrontal and orbitofrontal 5-HT_1A BP_ND in individuals with both lifetime and family history of suicide attempt and mood disorder. Differences in ventral prefrontal 5-HT_1A BP_ND were not observed in patients without family history for these conditions, suggesting that both high familial risk and psychopathology account for the finding. Whole-brain pattern classification analysis showed promise in distinguishing healthy volunteers from both high familial risk suicide attempters with mood disorder, suggesting suicidal behavior in the context of mood disorder and familial risk is associated with altered spatial pattern of 5-HT_1A BP_ND in ventromedial prefrontal and mesiotemporal cortex. If our results are confirmed in future studies, low ventral prefrontal 5-HT_1A BP_ND may be a biomarker of risk for suicidal behavior and mood disorders. Future studies could then seek the biological phenotype of resilience and consider if modifying binding in this brain region can reduce the risk of familial transmission of depression and suicidal behavior. 5-HT_1A BP_ND could be targeted with pharmacological and/or cognitive behavioral interventions. For example, Gray et al. (35) found chronic SSRI treatment alters 5-HT_1A receptor binding potential. Future longitudinal studies should also examine temporal relationships between suicidal behavior, mood disorder course, and ventral prefrontal serotonin 1A binding.

Data availability statement

The derived, de-identified data that support the findings of this study are available upon reasonable request from the corresponding authors. The raw data are not publicly available as participants of this study did not agree for their data to be shared publicly.