Stress and its impact on the transcriptome

Abstract

Exposure to stress during the course of a lifetime is inevitable in the animal kingdom. It is the response to stress, the valence of the exposure and the developmental time point that largely determine the consequences to the initial and subsequent exposures. The versatility of transcriptomic methods to yield rich, high resolution, information laden datasets from entire brain regions to single cells makes it a powerful approach to investigate the effects of stress from several angles. Dysregulation of the transcriptome is now a phenotypic signature of many neuropsychiatric disorders. New insight has been gained from examining stress-induced changes in gene expression at a global scale. Human postmortem datasets from depression and posttraumatic stress disorder studies have identified major gene expression changes in the diseased brain including sex-specific changes and marked differences in male and female molecular profiles for the same disorder. Extensions of this work into animal models has explored the impact of transcriptomic dysregulation on early life stress, chronic and transgenerational impact of stress. Here we explore the findings of human postmortem genomic studies of neuropsychiatric disorders and comparable animal models through the lens of transcriptomic dysregulation and how these findings have contributed to our understanding of stress.

Introduction

The important role of psychosocial stress as a major precipitating factor in the development of psychiatric disorders, particularly depressive, fear and anxiety disorders are widely recognized. Understanding the neurobiological consequences of stress exposure has therefore been a major focus of psychiatric neuroscience research for several decades. Among the tools and technologies that have strongly aided the quest for insight into the molecular actions of stress, gene expression analyses have played a prominent role. The progression from single gene methods to genome-wide approaches is remarkable. The ability to quantify global gene regulation in diverse tissues, ranging from brain regions to neuroanatomical cell layers and even single cells has enabled scientists in the field to investigate stress-induced changes at multiple levels. Examining the complete set of RNA transcripts or transcriptome from a variety of sources in a single high throughput experiment has become feasible due to the progress in instrumentation, technologies and bioinformatic analysis. Due to the maturity and robustness of the technology, elegant in vivo manipulations can be followed by transcriptome analysis to provide deep, high resolution mechanistic insight.

A major advantage of transcriptomics followed by bioinformatics analyses is that the results are unbiased, and the inference and interpretation does not hinge on the regulation of a single or few molecules but rather on a statistically validated and biologically connected network of candidates. For example, seemingly different stress paradigms could yield a proinflammatory gene signature but differ in the specific genes that are dysregulated. Furthermore, relationships such as co-regulation and reciprocal regulation can be explored to provide new understanding. The ability to uncover key hub molecules supports the testing of new hypotheses in a highly informed manner. In this review we discuss how recent studies have shed new light on well-studied stress regulated molecules such as glucocorticoids and brain derived neurotrophic factor (BDNF) as well as emerging stress research topics including transcriptomic interrogation of brain circuits. We draw attention to the results of postmortem major depressive disorder (MDD) and posttraumatic stress disorder (PTSD) RNA analysis given the goal of several stress paradigms is to gain insight into these complex psychiatric disorders.

Human Postmortem transcriptomic studies of major depressive disorder

Postmortem studies have become crucial in characterizing the molecular underpinnings of neuropsychiatric disorders. Human postmortem research allows for integration of genetic information with human disease state, which is an increasingly promising strategy to identify novel targets. By integrating multiple layers of genomic information (genetic, transcriptomic, and epigenomics) of specific brain regions we can gain unparalleled insight into the molecular determinants of disease. There is consistent evidence of morphometric changes in the prefrontal cortex of subjects with MDD (1, 2), including reduced neuronal body size (3). Atrophy of hippocampal pyramidal neurons and reductions in volume have also been observed in MDD (4, 5). As a result, the focus of most functional genomic studies of stress disorders using postmortem brain have focused on the prefrontal cortex and the hippocampus. From the onset, the majority of these studies used whole-genome microarray technologies to interrogate the transcriptome. But the advent of high throughput, deep sequencing technologies have largely replaced traditional microarrays as they allow for a complete snapshot of the transcriptome including all coding and non-coding RNAs, their splicing characteristics, and expression levels.

One of the first studies to examine the molecular changes in MDD brain used whole-genome microarray expression analysis on two microdissected hippocampal subfields, the dentate gyrus (DG) and the CA1 pyramidal cell layer (6). This study identified the kinase MKP-1 (mitogen activated protein kinase-coded by the gene DUSP1) as significantly upregulated in both hippocampal regions. Several other DUSP gene family members were also dysregulated though only in a single region. MKP-1 is a negative regulator of neurotrophic factor gene expression cascades and consequently this study identified marked decreases in brain-derived neurotrophic factor (BDNF), VGF nerve growth factor and vascular endothelial growth factor (VEGFA) in their depression cohort, a finding consistent with the preclinical animal studies of stress (7). Subject matched dorsolateral prefrontal cortex tissue from this same cohort found reduced expression of functional synaptic genes (8). This decrease was matched to electron microscopic stereology, revealing reduced spine density in this region. A follow-up study (9) by the same group found increased expression of REDD1, an inhibitor of mTORC signaling with a likely functional role in mediating the neuronal loss and atrophy in PFC of depressed subjects. Taken together, these initial postmortem studies point to disruptions in synaptic function caused by dendritic spine loss driven by deficits in growth factor signaling as the primary molecular pathology in MDD brain.

Recent evidence has pointed to sex-specific transcriptomic differences in subjects with MDD. Historically, females were excluded from most preclinical studies of neuropsychiatric disorders. While differences in male versus female biology were cited as reasons to exclude females, this is in fact the best reason to add females to these studies (10). A land mark paper in 2017 identified a pattern of sex-specific transcriptional signatures in 6 brain regions of MDD subjects that were more divergent than convergent (11). Further they identified sex-specific key driver genes in postmortem tissue that when expressed in rodents of the opposite sex had no effect on depressive behaviors while expression in the same sex did, thus supporting the sex-specific functionality of these findings. Additionally, a meta-analysis of eight previously published human MDD postmortem microarray datasets revealed a similar pattern of sexual dimorphism (12) and a recent study using an independent MDD cohort as a psychiatric control for PTSD observed a similar effect (13). These studies have set the stage for future work investigating sex-specific differences in other neuropsychiatric disorders and makes clear the need for development of novel therapeutics for men and women.

Most postmortem studies have used intact or “bulk” tissue. One goal of gene regulation biology is to understand the cell type-specific contribution to gene expression. The human brain is made up of myriad cell types and subtypes and parsing out the individual contributions of these cells to the transcriptomes of diseased tissue is critical to understanding the underlying molecular pathology. Due to challenges associated with extracting single cell mRNA from postmortem tissue, most studies use isolated nuclei (single nuclei) mRNA for sequencing. Several studies have shown a high degree of correlation between nuclear and cytoplasmic mRNA (14). As of this review, only one study has employed single nuclei RNA-sequencing on postmortem brain tissue of MDD subjects (15). This study examined the transcriptomes of ~80,000 nuclei from dorsolateral prefrontal cortex of an all-male MDD cohort. They identified 26 distinct cell types and expression changes in over 60% of them. Interestingly, they observed the greatest degree of expression change in excitatory neurons and oligodendrocyte precursor cells, cell types with limited prior implication in MDD; a finding that highlights the need to explore single cell-type involvement in postmortem biology.

Human Postmortem transcriptomic studies of Posttraumatic Stress Disorder

PTSD occurs in a subset of individuals who have experienced a traumatic event in life. The risk of developing PTSD after a traumatic event ranges widely from 5–30% with certain populations such as refugees and combat veterans developing at higher rates (16, 17). PTSD is a psychiatric disorder best understood from the standpoint of epigenetic changes as it requires an external “environmental” event to trigger it. Why some people who experience a traumatic event develop PTSD while others don’t is not well understood but is likely part genetic, part life history and part premorbid psychology (18–20). Women are twice as likely to develop PTSD after a traumatic event and are at a higher general rate of developing PTSD than men. Traumatic stress experiences activate several regions of the brain including the prefrontal cortex, the hippocampus, and the amygdala. Collectively, these regions comprise the “fear” circuit and is one of the most well understood and studied circuits of the brain (21, 22).

To date most PTSD transcriptomic studies have focused on peripheral tissues such as blood-specifically, white blood cells (23). They have identified altered expression levels of pro-inflammatory cytokines and genes related to glucocorticoid activity. Few studies have examined the brain postmortem biology of PTSD mostly due to a lack of available tissue. As such, PTSD postmortem studies lag behind other psychiatric disorders such as depression and schizophrenia. The first whole genome postmortem gene expression study was a pilot study examining the differential gene expression changes in dorsolateral prefrontal cortex in a small cohort of PTSD subjects (24). They identified down regulation of glucocorticoid related genes serum/glucocorticoid kinase 1 (SGK1) and FK506 binding protein 5 (FKBP5). Overexpression of a dominant negative Sgk1 virus (dnSGK1) during auditory fear conditioning caused higher levels of freezing in the contextual memory recall test three days later, suggesting that SGK1 inhibition enhances memory of contextual cues associated with fear conditioning. These findings point to a functional role in fear memory processing for SGK1.

A powerful strategy for identifying biomarkers of a disorder is to combine postmortem transcriptomic studies with human neuroimaging. Two elegant studies used photon emission topography (PET) to measure differential protein binding in patients with PTSD. PET imaging showed mGluR5 binding in PFC is increased in patients with PTSD, suggesting a dysfunction in glutamate cycling and transmission (25). Previous studies have implicated mGluR5 receptor regulates stress-induced fear conditioning (26). This increased availability of mGluR5 is thought to be associated with avoidance symptoms in PTSD. A targeted transcriptomic study of the glutamatergic markers in PTSD postmortem subgenual anterior cingulate (BA 25) identified significant increases in SH3 and multiple ankyrin repeat domains 1 (Shank1) transcription. Shank1 protein is responsible for tethering mGluR5 to the cell surface and could be responsible for the increased mGluR5 binding observed in PTSD patients. A recent study used PET imaging of the 18-kDa translocator protein (TSPO), a microglial marker, on PTSD and health controls with and without trauma exposure (27). They identified prefrontal cortical TSPO availability negatively associated with PTSD symptoms and was significantly lower than normal controls. These findings were extended to a PTSD postmortem cohort which identified lower transcript levels of TSPO activating protein and microglial marker TNFRSF14 (tumor necrosis factor receptor superfamily Member 14). These findings point to decreased microglial activation associating with PTSD pathophysiology.

A recent initiative by the US Department of Veterans Affairs National Center for PTSD has begun development of a PTSD brain bank populated with PTSD, non-PTSD psychiatric controls (major depressive disorder) and neurotypical controls. Recently, the first well powered transcriptomic study emerged from this important resource (13). This study performed differential gene expression and weighted co-expression analysis of 4 prefrontal cortex subregions, matched for sex and age across both diagnoses (PTSD and MDD) and controls. They found a highly connected down regulated interneuron co-expression module in which one of the key driver genes, ELFN1(Extracellular leucine rich repeat and fibronectin type III Domain Containing 1), was identified in a transcriptome wide-association study using the genotype data from the largest PTSD GWAS analysis (28) thus linking risk with disease state. Comparable to findings in MDD, the authors also found marked sexual dimorphism in PTSD with little overlap between the male and female transcriptomes. This study employed a unique strategy of including a psychiatric control-major depressive disorder. As approximately 50% of newly diagnosed PTSD patients are also comorbid for MDD (29, 30) this allowed for the study to disentangle the molecular pathology of the disorders. Remarkably, despite their high comorbidity their transcriptomic organization was largely divergent with only a handful of overlapping DEGs and co-expression modules.

Transcriptomic convergence of human MDD and animal models of depression

Much of our understanding of the molecular pathology of MDD is derived from animal models of chronic stress which give rise to depressive-like behaviors. There are several animal models for generating a depressive-like phenotype including chronic social defeat stress (CSDS) (31, 32) chronic variable stress (33), adult isolation (34, 35), and chronic unpredictable stress (36). These models have been shown to approximate aspects of depressive behavior including anhedonia, anxiety, and social avoidance. Major depression includes a wide array of causes and symptoms with multiple subtypes and therefore it is unlikely that any one animal model can effectively recapitulate MDD on a molecular level.

Similar to studies of human MDD brain tissue, global analyses have been performed on targeted brain regions of animals after chronic stress. However, few studies have performed comprehensive integration of animal and human transcriptomic data to find overlapping genes and pathways. Several recent studies have attempted to tackle this issue and compare data from animal models and humans to identify any overlap. One study systematically compared bulk-tissue transcriptional signatures in medial prefrontal cortex and nucleus accumbens of humans with MDD with 3 models of chronic stress (37). They compared differential gene expression and interspecies weighted gene co-expression analysis to identify overlapping transcriptomic organizational patterns between chronic stress and MDD. They identified significant overlap between transcriptomic alterations in both regions between human MDD and the 3 mouse models with no model capturing the entirety of the MDD transcriptomic landscape. The authors suggest that specific models should be used to capture specific aspects of the depression phenotype. A separate study employed single cell FACS sorted RNA-seq of mouse mPFC to identify cell type-specific expression changes after chronic unpredictable stress (38). This study identified sex-specific differences between male and female SST interneuron transcriptomes before and after chronic stress. Gene set enrichment identified dysregulation of the EIF2 signaling pathway in SST interneurons. Remarkably, the chronic stress transcriptomic changes observed in SST interneurons recapitulated much of the MDD Brodmann area 25 transcriptome in terms of genes regulated and direction of change. Additionally, the study identified signaling pathways that significantly overlapped between the human and mouse datasets. This study points to the need to examine single cell-type contribution of gene expression changes in both animal models and human postmortem tissue.

Stress enhanced fear learning (SEFL) has been proposed as a model of PTSD-related behaviors (anxiety-like and deficits in fear extinction) with greater etiological validity to study the individual differences in stress response in rodents. A recent study using this model has highlighted that genes previously implicated in PTSD (Drd1, Drd2 and Adcyap1) were altered in the amygdala, only in susceptible but not resilient mice subjected to SEFL (39), suggesting the utility of this model to study susceptibility vs. resilient mechanisms related to PTSD. Further, sub-chronic variable stress (SCVS) model, which comprises of daily variable and unpredictable stressors for one week, induces depressive-like behaviors only in females suggesting increased stress susceptibility of female mice (11). A similar increased susceptibility of females is observed in MDD patients (40). Development of animal behavioral assays that accurately model specific disease domains can enable rigorous testing of disease genes and therapeutic agents.

Glucocorticoid transcriptome

An elevation in glucocorticoids (GCs) is an essential and critical endocrine response to stress exposure. Understanding the contribution of GCs to the overall consequences of stress has been of major interest in stress research. The molecular actions of GCs are mediated via binding to GC receptors (GR) that are functional ligand-activated transcription factors. Transcriptomic analyses of GC and GR-induced gene regulation has enabled the field to obtain insight into the molecular mechanisms of stress. Understanding the detrimental effects of early stress exposure and determining the direct contributions of GCs can be helpful in developing focused protective interventions. Of the ~4500 genes regulated by GR overexpression in the hippocampal dentate gyrus, there was substantial overlap (62%) of regulated transcripts, genes involved in axonal guidance, growth, synaptogenesis, neurotransmitter signaling, development and cognition, between early life (3 weeks after birth) and lifetime GR overexpression (41). However, overall number of genes regulated in the nucleus accumbens was 4-fold lower than in the DG and only sparse overlap between early and lifetime exposure to amplified GR signaling. The shared high anxiety phenotype in both groups is likely a result of GR-induced gene regulation in the DG that was altered early in life and remained unchanged by continued GR signaling.

GC transcriptome studies have consistently reported the regulation of certain genes, including Serum glucocorticoid-regulated kinase 1(Sgk1), Fk506 binding protein (Fkbp5), glucocorticoid receptor (Nr3C1), Kruppel like factor 9 (Klf9), ERBB receptor feedback inhibitor 1 (Errf1) and Regulated in development and DNA damage responses 1 (REDD1)/DNA damage-inducible transcript 4 (Ddit4) (42). A brief discussion of their functional roles is helpful in appreciating the diverse effects that can ensue downstream of alterations in GC levels. Sgk1 produces differential stress effects based on the brain regions where it is regulated. A decrease in the prefrontal cortex leads to deficits consistent with trauma-associated behavior (24) while an increase in the hippocampus mediates the GC-induced decrease in neurogenesis (43). Fkbp5/Fk506bp, named due to its ability to bind the immunosuppressant drug FK506, associates with heat shock protein 90 (Hsp90) and steroid receptors, playing key roles in GR signaling and nuclear translocation (44). Fkbp5 polymorphisms have attracted significant attention as a stress modulating gene that links exposure to stress with PTSD (45–47). Nr3C1/GR, which has a 10-fold lower affinity to corticosterone than the mineralocorticoid receptor, is the primary mediator of GCs effects when levels are high as during stress (48). Hippocampal GR expression levels are strongly associated with negative feedback regulation of GC levels and HPA axis activation. A substantial body of animal and human studies have shown a strong relationship between adverse early life social environment and GR gene DNA methylation, providing insight on how early life experiences can result in lasting gene expression and behavioral changes via epigenetic mechanisms (49). Klf9 is a transcription factor and a major feed forward regulator of chronic GC-induced gene transcription (50). Interestingly, forebrain deletion of Klf9 was protective against stress-induced fear behaviors by preventing stress and chronic GC-induced spine enlargement in the ventral hippocampus (51). REDD1 is rapidly induced by metabolic and cellular stress and is a negative regulator of mammalian target of rapamycin (mTOR) (52). Chronic but not acute stress increased REDD1 gene expression in the rat prefrontal cortex, causing a reduction in spine density (9). It is noteworthy that REDD1 was elevated in MDD dorsolateral prefrontal cortex (9). In addition to disease pathophysiology mechanisms, these results are also important from the standpoint of antidepressant activity as ketamine produces rapid antidepressant effects by activating mTOR signal transduction (53).

BDNF alters the stress gene profile

Differential regulation of hippocampal BDNF by stress and antidepressants is foundational to the widely popular neurotrophic hypothesis of depression/antidepressant action. An intertwining of BDNF and GC signaling is therefore rather intriguing (54, 55). Primary cortical neurons treated simultaneously with BDNF and dexamethasone (Dex, synthetic glucocorticoid) doubles the number of regulated genes (933) in comparison to either Dex (435) or BDNF (456) alone (56). About half of these genes (455 of 933) were regulated only by co-treatment with BDNF and Dex, and were not regulated independently by either treatment, indicating a complex relationship in the modulation of gene transcription when signaling cascades driven by both ligands are activated simultaneously. The combined (BDNF + Dex) profile is more strongly represented by synaptic plasticity, neurotransmission and axon guidance genes than with either DEX or BDNF alone. Such an interaction between BDNF and GC signaling has implications for understanding the stressed brain because it indicates that BDNF works in dependent partnership with the GR, driving its phosphorylation to overhaul the stress gene profile and modify behavior (54). Microarray analysis of BDNF gene deletion revealed a pronounced reduction in the expression of several immediate early genes (IEGs), including Egr1, Fos, Arc and DUSP in the frontal cortex (57). As these IEGs play key roles in neuroplasticity, in the absence of BDNF, the brain’s molecular stress response mechanisms are likely to be affected (58). The BDNF Val66Met polymorphism, which affects the secretion of BDNF protein, has been actively studied to decipher its role in psychiatric disorders. Transgenic mice with the Met allele exhibit a hippocampal CA3 stress gene profile that is vastly different from wildtype animals (59). The 3-fold higher gene expression changes in the Met allele carriers indicates a heightened stress sensitivity, which is consistent with higher anxiety (60), stress vulnerability (61) and impaired plasticity (62) with this polymorphism. Met allele mice exhibit elevated despair behavior at baseline but are resistant to further adverse effects of chronic corticosterone that are seen in control mice. Future studies testing the role of neurotrophic signaling in specific neural circuits can yield important data towards understanding neuropsychiatric disease mechanisms.

Stress-induced gene regulation in neural circuits

There has been a growing interest in the field to understand psychiatric disorders from a neural circuitry angle. Examining stress-induced gene regulation and gene networks in well characterized circuits has the potential to identify dysregulated disease circuits and hub molecules. Stress during the critical postnatal period (P10–20) increases susceptibility to developing behavioral deficits upon subsequent adult stress exposure (63). This occurs by uniquely altering reward circuitry (VTA, NAc and PFC) gene profiles in a manner different from either early or adult stress, indicating that the stress-primed circuit responds differently to later exposure than a normal circuit (64). Understanding the currently unknown relationship between the stress-primed transcriptome and the second insult will likely include a role for chromatin remodeling and regulatory regions.

A Stress-induced synchrony in gene regulation between the nucleus accumbens and the PFC was noted in resilience to social defeat and a comparable pattern was observed between the ventral hippocampus and PFC in stress vulnerability (65). Employing a gene coexpression network analysis enabled the construction of stress susceptible and resilient modules to identify important hub genes. Interestingly, susceptibility hub molecules such as Dkk1 and Sdk1, that were identified and functionally validated, were not themselves regulated at the transcript level in any of the circuit brain regions.

The lateral habenula (LHb), a key processing center for negative reward signals and goal-directed behavior, is activated in depression (66, 67) and by stress (68). LHb neurocircuitry includes projections to the VTA (dopaminergic) and dorsal raphe (DR, serotonergic). Transcriptomic dissection of circuitry revealed a direct relationship between LHb →VTA but not LHb →DR neurons in regulating stress-induced passive coping behavior (69). Interestingly, the LHb →VTA stress signature, that elevated glutamate receptors and potassium channel genes, had no impact on anxiety or anhedonia measures (69). These studies show the importance of precisely determining circuit specificity in regulating particular maladaptive behaviors and their molecular correlates. It is conceivable that stress impacts multiple circuits in the brain and the translation of knowledge from animal assays to an understanding of psychiatric disorders would require methods to integrate and model multidimensional data.

Conclusions

The significant recent progress in obtaining transcript level data from a diverse body of stress research has led to important new insight into the molecular neurobiology of stress. Research directions in the field are likely to include the integration of genomics, proteomics and epigenomics data to understand in further detail how stress impacts the brain and is transduced into behavioral response. Future postmortem analyses of stress disorders will likely depend on integration of GWAS and transcriptomics to identify functional targets of risk genes. As GWA studies have needed to increase in size, so will postmortem transcriptomic studies. While small sample sizes were used even as recent as a few years ago, it is now common for most postmortem studies to have >50 to 100 subjects per group. The recent development of transcriptome wide-association studies for identifying cis-regulating genes from GWAS studies has made this even more important as TWAS weights calculated from eQTLs require large numbers to perform meaningful computations.

To date only one study has examined single cell contribution of gene expression dysregulation in MDD and none for PTSD. Investigations are moving toward single cells in both human postmortem and animal models of stress and emerging single cell -type spatial transcriptomics technology is on the horizon. The caveat to current single cell work is the limited number of cells captured and the predilection of the technology to only capture the top ~20% of expressed transcripts. Still single cell transcriptomic technologies continue to develop rapidly, and future systems will likely capture a greater number of cells and transcripts. Understanding the single cell-type molecular consequences of stress will identify novel pathways, gene networks, and disease transcript modules underlying stress disorders and potentially yield viable targets for diagnostic and therapeutic intervention.

Figure 1: Sex specificity in postmortem transcriptomes of MDD and PTSD.

Divergent profiles were observed between male and female transcriptomes in MDD and PTSD. Comparative analysis of MDD and PTSD revealed sparse overlap in transcriptome profiles despite high incidence of MDD in PTSD diagnoses. Animal model studies of genes dysregulated in male or female MDD exhibit depressive-like behavior only in animals of the matching and not the opposite sex.

Figure 2: Model of stress actions in psychiatric disease.

Stress-induced molecular signaling in brain regions such as the hippocampus and prefrontal cortex elevate vulnerability to development of major depressive and posttraumatic stress disorder. Disease progression can continue to drive dysfunction in these brain regions.