ABSTRACT
Suicidal behaviors are strongly linked with mood disorders, but the specific neurobiological and functional gene-expression correlates for this linkage remain elusive. We therefore tested the hypothesis that a convergent neuroanatomical and gene-expression signature will underlie mood disorder associated psychiatric morbidity and related suicide mortality. To do so, first, we applied an anatomical likelihood estimation (ALE) MRI meta-analysis across 72 voxel-based morphometry (VBM) studies including 2387 (living) participants that identified a selectively reduced anterior insula cortex gray matter volume (GMV) as a potential neuroanatomical signature for mood disorder. This neuroanatomical signature was then specifically used to guide postmortem RNA-Sequencing studies of 100 independent donor brains with a life-time history of major depressive disorder (N=30), bipolar disorder (N=37) and non-affected controls (N=33) using a sample from the National Institute of Mental Health Brain collection core. In this latter study, factor analysis first identified a higher-order factor representing number of Axis-1 diagnoses (i.e. morbidity) and suicide-completion (i.e. suicide-mortality). Using this higher-order factor as a contrast variable, differential gene-expression changes were examined in high psychiatric morbidity and related suicide mortality versus low psychiatric morbidity and related suicide mortality in mood disorder cohorts and controls. We identified in immune, inflammasome, neurodevelopmental, and transcriptional pathways and a weighted gene co-activation network analysis identified co-activated gene modules for psychiatric morbidity and suicide-mortality outcomes. These results provide a functional gene-expression link between mood disorder associated psychiatric disease morbidity and suicide-mortality.
INTRODUCTION
Major depressive disorder and bipolar disorder ⎯ here together referred to as mood disorders, are the third leading cause of the global disease burden (Collins et al. 2011; Murray et al. 2012). Mood disorders account for the majority of completed suicides (Waern et al. 2002; Marangell et al. 2006) and they were linked to ∼47,000 suicides in the United States in 2017 alone (American Foundation for Suicide Prevention, 2019). However, the convergent neurobiological basis for mood symptoms/syndromes and suicide is unknown, limiting advances in developing novel interventions.
Neuroimaging studies have identified reduction in gray matter volume (GMV) in the anterior insular cortex and anterior cingulate cortex (ACC) in association with diagnosis of psychiatric disorder in general (Goodkind et al. 2015), and the regional GMV volume reductions in these anterior insula and ACC network have been especially implicated in mood disorder diagnoses in particular (Wise et al. 2017). Neurobiological integrity of the right anterior insular cortex is shown to (a) predict mood diagnostic severity (Hatton et al. 2012), (b) modulate subjective responses to distress, pain, and psychosocial adversity (Wager et al. 2013; Eisenberger 2015), (c) regulate affective interoception (Craig 2009; Slavich et al. 2010), (d) associate with stress-related inflammatory markers (Khalsa et al. 2018), and (e) predict psycho- and pharmaco-therapeutic efficacy in mood disorders (McGrath et al. 2013). Anterior insula cortex-ACC functional connectivity during affective processing differentiated mood disorder suicide-attempters from non-attempters (Pan et al. 2013). Furthermore, abnormalities in anterior insula GMV and synaptic abnormalities are linked to suicidal-behavior in mood disorder (Wagner et al. 2012; Mathew et al. 2013). Anterior insula response to stress is shown to impact hypothalamic-pituitary-adrenal (HPA) axis-driven inflammatory responses (Khalsa et al. 2013), which may serve to exacerbate mood disorder associated psychiatric morbidity and suicidal-behavior (Oquendo et al. 2014; Wohleb et al. 2016). Although a preponderance of evidence supports abnormal anterior insula integrity in psychosocial distress (Schneidman 1998; Mee et al. 2011; Wager et al. 2013) and mood/comorbid psychiatric symptoms, an underlying functional genetic contribution in terms of functional gene-expression changes for these abnormalities remains largely unknown.
The lack of a well-defined relationship between aberrant brain structure and function with underlying molecular changes within this brain region is an impediment to understanding pathophysiology. Moreover, evidence for shared genetic mechanisms underlying psychiatric diagnoses (Brainstorm consortium, Anttila et al. 2018) is not well-integrated with brain imaging correlates of psychiatric disease-morbidity and specific behaviors, in this case, suicide. In the present study, MRI meta-analysis was used to test the hypothesis that 1) reduced anterior insula volume will be the most prominent neuroanatomical signature for mood disorder diagnoses. We confirmed this hypothesis with our meta-analytic findings and then used this anatomical hallmark to guide dissection of postmortem brain tissue for analyses of molecular/gene-expression signatures that could pave the way for precision profiling of gene functions underlying mood symptoms across diagnoses in clinically-relevant brain sub-regions. This approach enabled us to further test the hypothesis that 2) gene-expression signatures for psychiatric disease morbidity and related suicide-mortality will share similar molecular profiles in the voxel-based morphometry (VBM) meta-analysis defined-postmortem anterior insula of mood disorder individuals, thereby providing a neurobiological framework for characterizing convergent neural-and-gene expression signatures for behavioral brain diseases.
METHODS PARTICIPANTS
The imaging meta-analysis provided a consolidation of the current mood disorder VBM work by quantitatively integrating all the published results of volumetric comparisons of interest between controls and mood disorder participants, or correlations of volumetric measures with mood disorder symptom-specific measures that amounted to 72 previously published studies (i.e. unique volumetric comparisons or experiments) of differences in GMV measures between individuals with major depression or bipolar disorder versus healthy subjects. After selecting only publications with more than 20 subjects per comparison samples, data were included from 26 publications consisting of 43 experiments examining major depressive disorder<controls whole-brain GMV changes (demographics in Table 1A); and 21 publications of 29 experiments examining bipolar disorder<controls (Table 1A) together involving 2387 imaging participants in the meta-analysis.
The 47 studies included in our meta-analysis, as well as relevant studies that ended up not being included based on the above-mentioned inclusion criterion, are listed in supplementary Table S2. While 3 of the 72 studies/experiments included in the meta-analysis assessed suicidal behavior in relation to volumetric changes in mood disorder, suicidal phenotypes was not a specific selection criteria for study inclusion as there were very few studies in the BrainMap database that specifically assessed the relationship between suicidal phenotypes and brain volume. In study 2, RNA samples were extracted from the ALE-defined anterior insular cortex sub-region (in study 1) postmortem tissue of 100 donors from NIMH brain bank (Table 1B).
DESIGN
Based on the principle that neural structure subserves functional control of complex behavioral repertoires (Koechlin 2016), we localized the structural brain signature for mood disorders across samples and methods in study 1. Experiments of GMV changes associated with mood disorder diagnoses in defined stereotaxic space were included for analysis of localized GMV changes across studies in major depressive disorder, bipolar disorder, and major depressive disorder and bipolar disorder versus controls using the well-established ALE algorithm (Eickhoff et al. 2009). This signature guided localized anatomical-dissection of postmortem tissue for whole-transcriptome characterization of differential gene-expression and weighted gene co-expression network analysis (WGCNA).
NEUROIMAGING VBM META-ANALYSIS
Here, models the spatial uncertainty associated with each reported location of significant between-group differences in GMV changes (Eickhoff et al., 2009; 2012) and performed ALE assessment of GMV changes in (i) major depressive disorder<controls, (ii) bipolar disorder<controls, and (iii) mood disorders in general (i.e., pooled across major depressive disorder and bipolar disorder) versus controls.
BRAIN DISSECTION and RNA-EXTRACTION
The NIMH Human Brain Collection Core (HBCC) provided the Postmortem samples for which informed consents are acquired according to NIH IRB guidelines. Clinical characterization, neuropathology screening, and toxicology analyses followed previous protocols (Martin et al. 2006). For blocks of brain tissue that were dissected using a cutting-mold, the anterior insula in pre-existing slabs numbered as 3, 4, and (depending on brain size) slab 5 were dissected systematically dissected with the aid of visualization of several (about ∼16) electronic images of anatomical slices and related regional anatomical landmarks of the anterior insula meta-analytic GMV images (see sample images in Fig 1A and Fig 1B). For blocks of brain tissue that were dissected using free hand, the sections were removed from slabs containing the right anterior insula encompassing the identified reduced volume in the completed meta-analysis by specifically starting at the anatomical landmark where the caudate and putamen are approximately equal in size and extending to the most anterior portion of the insula (see Fig 1C showing a sample section before and after anterior insula dissection). All dissected tissues were pulverized and 50mg were aliquoted and used for standardized total RNA extraction and only samples with RNA integrity numbers (RIN) which is a measure of RNA postmortem RNA quality were included in the study (see Table 1B for average RIN values per group).
Illumina-Sequencing, Read-Mapping and Gene-Quantification
Total RNA was extracted and only samples with RNA integrity numbers > 5 were used. Ribosomal RNA was depleted using RiboMinus Eukaryote kit from Life Technologies (Foster City, CA, USA) for RNA-Seq and confirmed using an Agilent Technologies’ Bioanalyzer (Santa Clara, CA, USA). The 100 samples were processed using TruSeq RNA Library Prep Kit v2 and sequenced on the Illumina HiSeq 4000 at the Genome Sequencing and Analysis Facility (GSAF) at the University of Texas, Austin, USA. Paired-end libraries with average insert sizes of 200bp were obtained using NEBNext Ultra II Directions Library Prep kit from New England BioLabs and mRNA selection was done used the Poly(A) purist kit from Thermofisher. 30 million paired-end reads per sample (150 base pairs in length) were generated by sequencing every sample on 4 lanes of the sequencer. Sequenced reads were assessed for quality with Fastqc (Andrews 2010). The reads were pseudo-aligned to the human reference transcriptome (GRCh38-gencode) using kallisto (Kallisto, 2019), and gene-level abundances were obtained. The abundances were normalized using DESeq2, and transformed with variance stabilizing transformation (a transformation to yield counts that are approximately homoscedastic, having a constant variance regardless the mean expression value). Principal Component Analysis was performed using 25% of the highest variance genes in order to look at the underlying structure of the data and to identify the largest sources of variance.
STATISTICAL ANALYSIS
VBM Meta-analysis
Convergence across the findings reported in previous VBM studies was assessed using ALE, which in brief consists of first modelling the spatial uncertainty associated with each reported location for significant between-group differences (Eickhoff et al. 2009; Turkeltaub et al. 2012), computing the convergence across experiments by the union of the ensuing probabilistic model relative to a null-distribution reflecting a random spatial association between the findings of different experiments (Eickhoff et al. 2012) and finally statistical inference for a whole-brain corrected significance level of p<0.001 using threshold free cluster enhancement (Smith & Nichols 2009). We performed ALE separately focusing on GMV changes in (i) major depressive disorder, (ii) bipolar disorder, and (iii) mood disorders in general (i.e., pooled across major depressive disorder and bipolar disorder) relative to healthy controls.
Postmortem variable factor-analysis
The postmortem variables included mood disorder-diagnoses; # of lifetime-Axis-I diagnostic-occurrences (Axis-I-load); # of lifetime-Axis-III diagnoses; manner of death (natural, suicides/homicides/accidents) and cause of death from medical examiner reports; demographics (race, age at death, sex, years of education, number of children/fecundity, and marital records); technical variables (brain-weight, postmortem-index, pH, and RIN); and toxicology (blood alcohol/blood narcotics levels). We applied Principal Axis Factoring (Oblimin Rotation with Kaizer Normalization) (Costello & Osborne 2005) to identify higher-order factors explaining the differences in postmortem variables and included those variables with communalities of ≥ 0.5.
Differential Expression Analysis
Because mood disorder-diagnoses, Axis I diagnostic-load (i.e. total number of psychiatric diagnoses), and manner of death (together comprising the factor we refer to here as the psychiatric morbidity and mortality component) predominantly explains variability in (i) psychiatric disease morbidity (data on any Axis-III/medical morbidity like cardiovascular disorders or cancer have not been accounted for in our analysis) and (ii) suicide mortality (i.e. mortality or completion of suicide that can be directly linked to lifetime mood disorder symptoms), we binned samples into two groups of low versus high psychiatric disease morbidity and suicide-mortality for profiling gene-expression in the anterior insula cortex. Specifically, differential gene-expression between samples differing in psychiatric morbidity and suicide-mortality status was assessed based on the negative binomial distribution for modeled gene counts using DESeq2 (Anders 7 Huber 2010). RIN scores were included in the DESeq2 design matrix as a covariate to control for its potentially confounding effects.
We first defined low versus high scores of psychiatric morbidity using the psychiatric morbidity and suicide-mortality scores derived from the factor analytic scores into ≤ 0.82 for low versus ≥ 0.82 for high scores using a split-half method of dividing the maximum score across groups by 2 (i.e. 1.64/2). We then compared 1) high versus low psychiatric morbidity and suicide-mortality scores across major depressed and control groups (i.e. major depressed with high scores versus major depressed and all controls with low scores); 2) high versus low psychiatric morbidity and suicide-mortality scores across bipolar and control groups (i.e. bipolar with high scores versus bipolar and all controls with low scores); 3) high versus low psychiatric morbidity and suicide-mortality scores across all groups (i.e. bipolar and major depressed with high scores versus bipolar, major depressed and all controls with low scores).
To assess the relationship between suicide mortality and mood disorders specifically, we binned our groups into 1) low (non-suicidal deaths across all samples to include normal controls who were by definition all non-suicide deaths) versus 2) high (suicide-completion across major depression and bipolar samples) to assess the presence of specific gene-expression patterns associated with suicide-mortality across our total postmortem population. This analysis was performed separately across a) major depressive disorder suicides vs. major depressive disorder non-suicides & controls, b) bipolar disorder suicides vs. bipolar disorder non-suicides and controls, and c) combined mood disorder, i.e. major depressive disorder and bipolar disorder suicides vs. major depressive disorder and bipolar disorder non-suicides alone in line with a similar approach previously reported by Pantazatos SP et al. 2017). Controls were omitted in the last comparison (i.e. comparison c) to identify genes specifically linked to severity of suicide in persons diagnosed with psychiatric disorders. Only genes with corrected p-value (after benjamini-hochberg multiple testing correction) ≤ 0.1 and absolute fold changes ≥1.5 are reported as significantly differentially expressed. Pathways and gene-ontology (GO) terms enriched in these genes were identified using Enrichr (Chen et al. 2013; Kuleshov et al. 2016).
Weighted Gene Co-expression Network Analysis (WGCNA)
Scale-free co-expression networks were constructed with gene-abundances using the WGCNA package in R (Langfelder & Horvath 2008). WGCNA provides a global perspective and allows identification of co-expressed gene-modules. It avoids relying on arbitrary-cutoffs involved in selecting differentially-expressed genes and instead identifies a group of genes that are changing in the same direction and magnitude, even if these changes are smaller in magnitude. WGCNA thereby identifies genes that are potentially co-regulated or belong to the same functional pathway, using a dynamic tree-cutting algorithm based on hierarchical clustering (i.e. minimum module size=30). To identify co-expressed gene modules of interest, we incorporated covariate information and selected those co-expressed gene modules correlating significantly with diagnostic, and suicide-linked variables. Driver genes (i.e. genes within co-expressed gene modules whose singular expression patterns are similar to the overall expression profile of the entire co-expressed modules) within these modules were used to identify pathobiological functions associated with each module.
RESULTS
Identification of a Mood Disorder Neuroanatomical Signature in Living Brains
The study 1 VBM meta-analysis (N=2387) revealed reduced GMV in the right anterior insular cortex in mood disorders (p<0.0001 corrected) (Fig 1A) consistent across both major depressive disorder and bipolar disorder, since major depressive disorder and bipolar disorder groups did not differ significantly (Fig 1A). The localized reduced anterior insular neuroanatomical-signature for mood disorders was manually segmented in ITKSNAP (http://www.itksnap.org/pmwiki/pmwiki.php) (Fig 1B) and the segmented volume guided postmortem dissection of tissue used in RNA-seq characterization of gene-expression (Fig 1C).
Postmortem Group Differences and Factor Analysis
Demographic variables did not differ among groups, except for race: major depressive disorder and bipolar disorder groups had more Caucasian donors whereas healthy subjects had more African American donors (p<0.0001, F=12). Covarying for race in subsequent ANOVAs, Axis-I-load (p<0.0001, F=30) (Fig 2A) and suicide-completion (p<0.0001; F=39.7) were higher in major depressive disorder and bipolar disorder donors than controls (Fig 2B). Using Bonferroni correction, post-hoc pairwise-comparisons of the original postmortem variable yielded group-differences in Axis-I-load in major depressive disorder>controls (p<0.0001); bipolar disorder>controls (p<0.0001); and bipolar disorder>major depressive disorder (p = 0.004); while suicide-completion differed in major depressive disorder>controls (p<0.0001) and in bipolar disorder>controls (p<0.0001), but not between major depressive disorder and bipolar disorder. Linear regression analysis showed that Axis-I-load predicted suicide-completion (B=.195, t=5.2, p<0.0001, d=1.408), but not age at death.
Post-hoc multiple comparisons of the high-order factor-analytic variables yielded group-differences in psychiatric morbidity and suicide-mortality: using Bonferroni correction, psychiatric morbidity and suicide-mortality was highest in bipolar disorder>controls (p<0.0001, Fig 2C); bipolar disorder>major depressive disorder (p<0.0001, Fig 2C), and major depressive disorder>controls (p<0.0001, Fig 2C). Linear regression revealed that psychiatric morbidity and suicide-mortality negatively predicted (a) RIN-scores (B=-2.1, t=-3.3, p=0.001) across groups; (b) fecundity (B=-3.17, t=-2, p=0.041, d=1.79) across groups; and (c) age at death (B=-8.7, t=-.27, p=0.025, d=2.1) only across major depressive disorder and bipolar disorder. These findings prescribed our subsequent analytical focus on psychiatric morbidity and suicide-mortality.
Differential-Expression & WGCNA Identified Enriched Postmortem Anterior Insula Gene-Expression Signatures
Differential gene-expression analyses assessed transcriptomic profiles associating with variability in psychiatric morbidity and suicide-mortality. We binned mood disorder associated psychiatric morbidity and suicide-mortality scores into ≤ 0.82 for low versus ≥ 0.82 for high scores using a split-half method of dividing the maximum score across groups by 2 (i.e. 1.64/2) and found differentially-expressed immune and inflammatory-pathways, toll-like receptor-signaling, nuclear factor kappa-light-chain-enhancer of activated B cells (NF-kb-signaling), chemokine-signaling, and cytokine-cytokine receptor interactive-pathway genes (Fig 3A-D, and Table SA-1A-F).
Specifically, within major depressive disorder and controls (i.e. major depressive disorder cases scoring high on psychiatric morbidity and suicide-mortality versus low scores and controls), we found 9 under-expressed inflammatory cytokine and AKT-signaling (CCL3 & CCL4); innate immunity/antigen recognition and elimination of bound antigens (IGHV2-5 and IGHV2-70); mRNA splicing and enzyme-binding (HSPA6); cellular development and homeostasis (HSPA7, PCSK5, & SERPINH1) genes; and 1 mitochondrially encoded cytochrome-C-oxidase lncRNA-pseudogene (MTCO1P12); and an over-expressed lncRNA-pseudogene (MTCO2P12) (Fig 3A-D, Table SA-1A).
Similar differential gene-expression analysis in bipolar disorder individuals with higher psychiatric morbidity and suicide-mortality scores versus those with lower scores and controls was associated with 4 under-expressed innate immunity (IGHV2-70); neuroprotection, neurodevelopment and CNS-diseases including major depressive disorder (CXCL11, RP11-7461.1, & SELE) pathway genes (Fig 3A-D, Table SA-1B). Analysis of high psychiatric morbidity and suicide-mortality scores in combined major depressive disorder and bipolar disorder versus low scores and controls yielded 10 under-expressed inflammatory cytokine and AKT-signaling (CCL4); innate immunity (IGHV2-70 & IGKV2-28); cell-neurodevelopment and CNS diseases (CXCL11, SELE, PCSK5, & HSPA7); and transcriptional regulatory-RNA (MIR5190); but also 2 over-expressed innate immunity (RP11-566K19.8) pathway genes; and a lncRNA-pseudogene (MTCO2P12) (Fig 3A-D, Table SA-1C).
We then examined gene-expression profiles related to suicide post-hoc by binning scores ≥1 for completed-suicides in major depressive disorder and 0 for non-suicides in major depressive disorder and controls (i.e. comparing gene-expression in major depressive disorder suicides versus major depressive disorder non-suicide and controls), and found 3 over-expressed WNT-signaling (FZD8); transcriptional regulation of adaptive responses to oxygen tension/hypoxia, DNA-binding transcriptional activity/co-activation (HIF3A); and dioxygenase activity (PHYHD) pathway genes (Table SA-1D). Binning suicide completion in bipolar disorder-suicides versus non-suicidal bipolar disorder and controls, we found 5 under-expressed innate immunity (IGHV2-5, IGHV2-70, IGHV3-7, IGHV4-39, & IGHV3-15); and CNS-disease (SORD2P) pathway genes (Table SA-1E).
Assessing gene-expression profiles associated with suicide-completion in the pooled major depressive disorder and bipolar disorder completed-suicides versus major depressive disorder and bipolar disorder non-suicidal deaths yielded 21 under-expressed innate immune and inflammatory-cytokine (CRISPLD1, CHI3L1, P2RY6, & SECTM1); protein-protein interaction regulatory (MT1A, HILPDA, HELZ2, FOSB, FAM198A, SOCS3, & TPST1); neurodegeneration (RP11-155G14.6, SLC39A14, & SERPINA3); cellular-neurodevelopmental and transcriptional (LIMK2, SFN, & EDN3) pathway genes (Table SA-1F); as well as uncharacterized genes/pseudogenes (MTND2P28, BAALC-AS1, RP11.420L9.5, & RP11.435J9.2). We also found 5 over-expressed inflammatory (RP11.1100L3.8); intracellular protein transport (TBC1D3E); cell fate and apoptosis regulation (GZMA); and transcriptional, embryonic/forebrain cell development and defect (CTD-2207O23.3, & TDGF1) pathway genes (Fig 3D, Table 1A-A-F).
WGCNA characterized the potential co-regulated gene-modules involved in mood disorder morbidity and suicide-mortality. After correcting for covariates and filtering out low-mean genes, we found 2 dominant co-expression modules (coded as blue and black Figs. 4-5; Fig SA1A-B) strongly associating with psychiatric morbidity and suicide-mortality variables. Using the 30 most highly connected (i.e. hub genes) in these two modules, the blue module showed enrichment in infection, addiction and cell signaling pathways, among others. The black module is enriched in major depression, dopaminergic synapse, metabolism and addiction pathways.
DISCUSSION
Using a neuroimaging meta-analysis to refine a structurally reduced anterior insula region of interest in major depression and bipolar disorder, we identified a postmortem across-diagnostic mood disorder linked psychiatric morbidity and suicide-mortality associated gene-expression signature within this neuroimaging meta-analysis identified reduced anterior insular cortical gray matter signature. Given this anterior insula sub-region’s documented role in regulating affective and physical pain/distress, general bodily homeostatic and interoceptive salience, our convergent structural neurobiological and functional gene-expression findings of (a) a reduced right anterior insula cortical gray matter signature in living mood disorder patients, coupled with (b) a preponderance of under-expressed, but also albeit to a lesser extent, over-expressed gene-expression signatures within the identified anterior insula-locale in mood disorder postmortem brains, identified an anatomically-precise functional gene-expression basis for mood pathologies.
In light of the right anterior insula’s role in sensing emotional/physical pain associated with social isolation and disconnection (Eisenberg 2015), these results provides a potential gene-regulatory windows into the neuropathological ramifications for increased mood disorder associated psychiatric morbidity (O’Connor & Nock 2014; Nock et al. 2018), which could compound suicidal outcomes (Nock et al. 2018). The right anterior insula’s role in coding affective salience and psychological pain (Schneidman 1998), and the possible collective modulatory impact of these states on maladaptive impulses such as the urge to escape unbearable misery via suicide, provides a putative anatomical framework for mood disorder associated psychiatric morbidity and suicide-mortality (Schneidman 1998; Craig 2009; Slavich et al. 2010; Mee et al. 2011; Hatton et al. 2012; McGrath et al. 2013; Wager et al. 2013; Eisenberg 2015; Goodkind et al. 2015; Koechlin 2016; Wise et al. 2017; Khalsa et al. 2018). The identified association between mood and associated psychiatric morbidity and suicide-mortality scores with fecundity, and with predominant under-expressed gene-expressive functions suggests an evolutionary significance of the current results. For instance, genetic abnormalities governing aberrant anterior insula-mediated social deficits (Jabbi et al. 2012) and severe mood disorders could attenuate reproductive prowess (Mullins et al. 2017).
Within the identified reduced anterior insula signature, our differential gene-expression analyses identified predominant down-regulations in innate immune functions, inflammation pathways, and AKT-signaling related to mood-associated psychiatric morbidity and suicidal-mortality. Further differential-expressions involving cellular-homeostatic, neurodevelopmental, and transcriptional pathways in the anterior insula cortex were found to be associated with psychiatric morbidity and suicidal-mortality in less directionally-specific (i.e. up/down-regulatory) patterns. While the cellular-origins of our findings of predominantly downregulated immune-related and neurodevelopmental gene-expression changes cannot be specified with our bulk tissue RNA-seq methods, astrocyte-derived cytokine functions have been documented to induce synapse engulfment/elimination and thereby influence synaptic pruning (Vianchtien et al. 2018; Bennett & Molofsky 2019). Furthermore, immune pathway mediation of mood dysfunctions and related psychiatric diagnoses are proposed to be likely multifaceted as part of the brain’s immune-related response repertoire such as toll-like receptor signaling, can be influenced by both a) pathogen-associated molecular patterns (Kawai & Akira 2007) and b) danger-associated molecular patterns (Klune et al. 2008; Piccinini et al. 2010). This perspective on the multifaceted nature of brain immune signaling in relation to behavioral dysfunctions like mood disorders deserves further analysis especially in light of 1) the strong link between endogenous danger-associated immune/inflammatory cellular functions in promoting homeostasis (Klune et al. 2008) and 2) the potential impacts of environmental stress experiences on endogenous cellular stress as well as inflammatory responses (Slavich et al. 2010).
Existing findings of both over-expressed immune-related signals (Pandey GN 2017), and our current imaging-guided anterior insula postmortem findings that clearly replicated previous results of predominantly under-expressed immune/inflammatory function (e.g. chemokine ligand 2 CCL4) and genes/pseudogenes implicated in regulatory cellular functions (e.g. a serpin peptidase inhibitor & HSPA7) in postmortem dorsolateral prefrontal cortex BA9 brains of mood disorder individuals with and without suicide completion (Pantazatos et al. 2017) needs to be better contextualized. For instance whether these differences in the directionality of regulatory patterns of gene-expression findings (i.e. over-expressed versus under-expressed immune expressive genes) in mood disorder postmortem studies is somewhat related to methodological differences in in terms of targeted micro-RNA assays versus whole transcriptome sequencing approaches, or differences in sample sizes, sample selection criteria, or qualitative postmortem material differences between studies, needs to be examined more carefully.
Together, our current findings are consistent with data on the role of immune dysfunctions in CNS diseases (Oquendo et al. 2014; Wohleb et al. 2016; Pantazatos et al. 2017; Butovsky et al. 2018), and inflammasome functional prediction of major depressive disorder treatment outcomes (Syed et al. 2018). The results further lend credence to the hypothesis that neurodevelopmental and transcription-factor genes are critical mediators of complex adaptive brain functions (Changeaux et al. 2017); especially within the context of the anterior insula’s integration of affective and physiological feeling states (Craig 2009; Slavich et al. 2010; Kurth et al. 2010; Eisenberg 2015; Khalsa et al. 2018) ‘including homeostatic maintenances in sickness and health’ (Craig 2009; Khalsa et al. 2018), that are likely not entirely independent of both pathogen-associated molecular patterns (Kawai & Akira 2007) and danger-associated molecular patterns (Klune et al. 2008; Piccinini et al. 2010) known to induce brain immune signaling.
At the systems level, the toll-like receptor (TLR) pathway genes found to be under-expressed here are documented to recognize conserved motifs in microorganisms (Akira 2003) and stimulation of TLRs are shown to mediate acute-immune defense and cytokine production/release (Perkins 2007). Second, our identified under-expressed NF-κB pathway genes are implicated in controlling DNA transcription, cytokine production and cell survival (Meffert et al. 2003), and are essential for cellular-immune response to infection, stress-related shocks (Van Amerongen et al. 2009), and synaptic plasticity and memory (Meffert et al. 2003; Van Amerongen et al. 2009). Third, the identified under-expressed chemokine-signaling pathways govern critical spatiotemporal cell-positioning during developmental coordination and translational guidance of cell-locomotion and migration (Turner et al. 2014). Fourth, the identified under-expressed cytokines are implicated in cell-specific innate and adaptive inflammatory host defenses, cellular-development, cell-death, angiogenesis, and maintenance of cellular homeostasis (Syed et al. 2018). Conversely, the Wnt-β-catenin signaling pathway found to be over-expressed in major depressive disorder suicides is an evolutionarily conserved inter-cellular communication system that mediates stem cell renewal, cell-proliferation and differentiation during embryogenesis and adulthood (Meffert et al. 2003). Moreover, our WGCNA results showing co-activated gene modules for a) lifetime mood disorder-diagnoses, b) lifetime Axis-I diagnoses, and c) suicide-completion status and the lethality of the committed suicide methods in (i) immune, (ii) major depression diagnosis, and (iii) dopaminergic-pathways, suggest that multi-genic influences may be impacting mood disorder disease burden and suicides. Taken together, our observations of convergent under-expressed TLRs, NF-κB, chemokine, and cytokine-cytokine interactive pathways transcriptomic signatures for psychiatric morbidity and suicide-mortality; and suicide-mortality-specific over-expressed Wnt signaling pathway, suggests possible dysregulatory mechanisms for aberrant cellular processes very early in development. These processes may negatively shape adaptive immune, inflammasome and chemokine-cytokine responses to adverse socio-emotional and environmental distress, with a prolonged experience of these adverse circumstances likely leading to compromised anterior insula anatomical and physiological integrity, and associated maladaptive rupture in regulatory mood states.
Unlike cardiovascular disease and cancer research, where pathobiological measures are causally linked to disease morbidity and end-point mortality, the causal neurobiological root causes of mental illnesses are unknown, limiting measurable biological predictability of suicidal-mortality. Our findings of convergent structural neurobiologically defined functional gene-expression signatures for mood disorder associated psychiatric morbidity and suicide-mortality across major depressive disorder and bipolar disorder supports shared heritable neurogenetic pathologies underlying comorbid neuropsychiatric symptoms (Anttila et al. 2018; Gandal et al. 2018). While the cell-type specific aberrations and their relationship with differential gene expression profiles needs to be studied to better understand the molecular mechanisms underlying abnormal neuroanatomical signatures for mood symptoms, especially in-terms of diagnostic specificity between major depressive disorder and bipolar disorder, our results represent a step towards developing brain region-specific functional gene-expression blueprints for therapeutic targeting of broad/specified molecular pathways. Furthermore, the effects of medication on gross neuroanatomical measures and gene-expression profiles also needs to be assessed in future studies. In sum, our findings bridging convergent neuroanatomical and gene-expression signatures for measures of the degree of comorbid psychiatric symptoms in mood disorders and suicides, represents a framework for discoveries of novel biomarkers for brain diseases.
SUPPLEMENTARY APPENDIX
RESULTS
Acknowledgements
The NIMH Human Brain Collection Core provided RNA-samples for all 100 postmortem data and we thank the NIMH and Drs. Barbara Lipska, Stefano Marenco, Pavan Auluck and HBCC colleagues for providing the studied samples. We thank Wade Weber of Dell Medical School Psychiatry Department, UT Austin for assistance in preparing the manuscript, Nicole Elmer of the Biomedical research support for help with gene-expression result figures, and Jessica Podnar and GSAF colleagues for RNA-seq support. This work was supported by the Dell Medical School, UT Austin funds for MJabbi, and Hans Hofmann is supported by NSF-DEB 1638861 and NSF-IOS 1326187.