How spatial omics approaches can be used to map the biological impacts of stress in psychiatric disorders: a perspective, overview and technical guide

Abstract

Exposure to significant levels of stress and trauma throughout life is a leading risk factor for the development of major psychiatric disorders. Despite this, we do not have a comprehensive understanding of the mechanisms that explain how stress raises psychiatric disorder risk. Stress in humans is complex and produces variable molecular outcomes depending on the stress type, timing, and duration. Deciphering how stress increases disorder risk has consequently been challenging to address with the traditional single-target experimental approaches primarily utilized to date. Importantly, the molecular processes that occur following stress are not fully understood but are needed to find novel treatment targets. Sequencing-based omics technologies, allowing for an unbiased investigation of physiological changes induced by stress, are rapidly accelerating our knowledge of the molecular sequelae of stress at a single-cell resolution. Spatial multi-omics technologies are now also emerging, allowing for simultaneous analysis of functional molecular layers, from epigenome to proteome, with anatomical context. The technology has immense potential to transform our understanding of how disorders develop, which we believe will significantly propel our understanding of how specific risk factors, such as stress, contribute to disease course. Here, we provide our perspective of how we believe these technologies will transform our understanding of the neurobiology of stress, and also provided a technical guide to assist molecular psychiatry and stress researchers who wish to implement spatial omics approaches in their own research. Finally, we identify potential future directions using multi-omics technology in stress research.

Keywords:

• Stress
• psychiatry
• spatial
• omics
• spatial transcriptomics
• spatial proteomics
• visium
• xenium
• cosmx
• geomx

Stress as a risk factor for psychopathology and evidence of its impacts

While genetics explains a proportion of the risk for psychiatric disorders, environmental risk factors also heavily contribute to disorder development, progression, and severity. For some disorders, such as depression, environmental exposures contribute overwhelmingly to the risk of disorder development (Penner-Goeke & Binder, 2019). Of these, experiencing significant levels of psychological stress is the leading risk factor (Basu & Banerjee, 2020). For instance, the prevalence of major depressive disorder in individuals with traumatic exposures in their lives is increased compared to individuals without these stress exposures (24% versus 12% prevalence, respectively), suggesting that genetics alone does not account for the risk (Coleman et al., 2020). Significant childhood trauma has also been reported in 75.6% of 349 chronically depressed patients, with the severity and the repeated experience of traumatic events (for example, childhood abuse or neglect) being correlated with symptom severity (Negele et al., 2015). However, the cellular and molecular mechanisms that underpin the increased risk of disorder development from stress exposure are not well understood.

Our limited understanding of the molecular mechanisms responsible for increasing risk is partly because of the complexity of stress. The type of stress, duration, and amount of stress exerts differential effects that produce variable molecular outcomes. These effects are virtually impossible to control or account for in a clinical setting. Further, examining stress in humans is complex as people experience multiple unique stressors over their lifetime, sometimes concurrently, making it challenging to quantify and compare individuals accurately. To this end, animal models have been particularly valuable in examining the impacts of stress on the brain with experimental control, helping us to understand both the biological mechanisms that underpin specific stress exposures and the behavioral outputs (Cattane et al., 2022). Rodent studies generally report that various models of stress exposure (e.g. social defeat stress, unpredictable stress, restraint stress, etc.) cause reductions in neuronal arborization, dendritic spines, and molecules that enhance synaptic plasticity. The cellular alterations are reported across several brain areas, including the prefrontal cortex and hippocampus (see 6 for an extensive review). However, the effects appear to be brain region-dependent, given that minimal alterations to neuronal branching and spine numbers are reported in the amygdala in response to chronic stress paradigms (Leem et al., 2020; Patel et al., 2018).

In humans, stress exposures (such as abuse or neglect) have a direct effect on emotional and cognitive processing, as extensively studied and reviewed elsewhere (Sandi & Haller, 2015; von Dawans et al., 2021). Additionally, human neuroimaging studies using structural magnetic resonance imaging (sMRI) have demonstrated that stress can elicit persistent changes in the volumes of several brain structures, including the prefrontal cortex, hippocampus, and amygdala (Kaul et al., 2021). Interestingly, these studies indicate that the type of stress experienced has specific downstream impacts on the brain structures that perceive and process particular stressful stimuli. For instance, women who had experienced childhood sexual abuse presented with significant thinning in the genital representation field of the primary somatosensory cortex, whereas women who experienced childhood emotional abuse had cortical thinning of regions involved in self-awareness and evaluation (the left precuneus and left anterior and posterior cingulate cortex) (Heim et al., 2013). Additionally, witnessing domestic violence during childhood was reported to reduce grey matter thickness and volume in the right lingual gyrus, as well as thickness in the secondary visual cortex bilaterally and the left occipital pole (Tomoda et al., 2012). Considering the presented studies, a key question is, what do these structural changes represent at the cellular and molecular level, where novel drugs could be designed to intervene?

Several studies have attempted to address this, with evidence that components of the underlying cytoarchitecture are affected. For example, reduced volume or cortical thinning is hypothesized to be representative of reductions in neuronal numbers and changes in neuronal morphology, such as altered dendritic architecture (Chen et al., 2020; Cobb et al., 2013; Kaul et al., 2020; Schoenfeld et al., 2017; Stockmeier & Rajkowska, 2004). The timing of stress is also important, with significant reductions in dendritic spines observed in the postmortem orbitofrontal cortex of psychiatric cases exposed to significant stress in childhood compared to stress experienced in adulthood (Kaul et al., 2020). Cytoarchitectural changes are essential to understand as they are directly related to molecular composition and ultimately responsible for regional and global tissue functioning (Chklovskii, 2004; Shepherd et al., 2005). This raises the question, what are the molecular drivers of these cytoarchitectural changes? This has been a key question driving the field.

The influence of multi-omics research in understanding stress and psychopathology

Historically, molecular characterization of biological tissues, including the brain, has been limited to examining single or very few targets according to a priori hypotheses, for example, using experimental techniques such as western blot, immunohistochemistry and quantitative polymerase chain reaction (qPCR). Specifically in psychiatry and understanding the role of risk factors (such as stress exposure), approaching scientific questions with a narrow focus on a single target is unfavorable as psychiatric disorders and stress responses are inherently complex, affect the whole brain, and are highly variable between individuals. Omics technologies have advantageously allowed for unbiased investigation at multiple levels of regulation, for example, the genome, epigenome, transcriptome, proteome, and metabolome. These technological advancements have rapidly accelerated our biological understanding of the brain in health and disease in a much shorter time than otherwise possible, providing a more holistic view of the molecular sequelae that occur in the brains of individuals with psychiatric disorders.

For example, large-scale efforts to better understand the pathophysiology of psychiatric disorders have been conducted by the PsychENCODE Consortium, involving the implementation of omics technologies to identify modules of risk genes and how they are co-expressed (that is, how they are not only altered but how they relate to each other) (Ashley-Koch et al., 2018). In a paper published as part of the consortium in 2018, psychiatric disease-associated alterations were mapped at the isoform level in brain samples from the frontal and temporal cortices using bulk tissue RNA sequencing (RNA-seq) method and co-expression network analyses. Alterations to the splicing of 48 RNA-binding proteins and splicing factors with unknown functions were identified (10%; FDR = 8.8 × 10⁻⁴), and close to 1000 noncoding RNAs were found to be dysregulated in disorders (Gandal et al., 2018). Altered disorder-specific pathways were also identified, with splicing dysfunction in angiotensin receptor signaling identified in bipolar disorder and splicing alterations impacting guanosine triphosphatase receptor activity, neuron development, and the actin cytoskeleton identified in schizophrenia (Gandal et al., 2018). Interestingly, expression of microglia-, astrocyte-, excitatory neuron, and interneuron-specific genes were also altered in a disorder-specific manner, suggesting that cell type-specific alterations occur across psychopathologies (Gandal et al., 2018). To explore this further, gene expression changes were examined cell type specifically by integrating bulk tissue RNA-seq with single-cell and single-nucleus data from multiple brain regions (hippocampus, striatum, amygdala, cerebellum, thalamus, and 11 neocortical areas), revealing that excitatory neurons across the neocortex of schizophrenia patients were enriched for the expression of psychiatric disorder risk genes (Li et al., 2018). Identifying excitatory neurons as a key site of increased expression of schizophrenia risk genes highlights that single-cell technologies can pinpoint molecular changes at a much greater resolution than previously possible, identifying the molecular changes and in which cell types they occur. Single-cell omics techniques have the potential to greatly aid in highly targeted drug development in the future, especially as such studies begin to explore not only disorder-related alterations but also the impacts of risk factors such as stress (Van de Sande et al., 2023).

In the stress research field, omics profiling of stress in rodents has been particularly valuable for improving our understanding of how stress increases the risk for psychiatric disorders over time in different complex brain areas. For instance, one study examining the hippocampus of C57Bl/6J mice exposed to different stress paradigms (novelty stress, cold swim stress, and restraint stress) with unbiased proteomics and transcriptomics identified that more than 20% of genes and proteins were differentially expressed between the ventral and dorsal regions (Floriou-Servou et al., 2018). Building on our knowledge that unique stress events differentially impact specific brain regions, distinctive gene expression profiles were observed between the different stress paradigms brain-area specifically (Floriou-Servou et al., 2018). Additionally, 215 similarly expressed genes were identified, with upregulated genes related to transcriptional regulation and downregulated genes related to cell adhesion molecules and cell junctions, indicating these pathways are altered in a stress-type independent manner (Floriou-Servou et al., 2018).

Building on the above studies, many omics methods can now be conducted at single-cell resolution, enabling a richer and more meaningful context of observations. Examples of studies that have used single-cell sequencing analyses to improve how we understand stress include profiling of animals (Brivio et al., 2023; Kos et al., 2023; Lopez et al., 2021; Miranda et al., 2023) and in vitro stress models (Cruceanu et al., 2022; Seah et al., 2022), as well as postmortem brain tissues derived from individuals with a history of childhood abuse (Tanti et al., 2022; Warhaftig et al., 2023). These reports have not only improved our understanding of how stress increases the risk for psychiatric disorders but also narrowed down cell-type specific mechanisms, such as oligodendrocytes as the target for sex differences in gene expression in the paraventricular nucleus (Brivio et al., 2023), subpopulations of Abcb1-expressing cells in the adrenal gland involved in stress adaptations (Lopez et al., 2021), and changes in transcriptional profiles of glutamatergic and GABAergic neurons in the ventral hippocampus (Kos et al., 2023). Novel therapeutic targets with high specificity, such as regulators of the highly stress-responsive gene FKBP5 (Matosin et al., 2023), have also been identified and characterized (Seah et al., 2022). Altogether, single-cell sequencing studies have highlighted that heterogeneity within major cell types and their anatomical location in the body is essential to fully understand the functional outcomes of stress.

It should be noted that single-cell and -nucleus sequencing methods rely on dissociating dissected tissue, thus removing the spatial context within the tissue to a certain extent and limiting the ability to evaluate particular subregions or cell-cell interactions fundamental to multicellular systems’ functioning. The anatomical context of cellular and molecular hierarchy, from genome to metabolome, is essential for cell functioning (Vandereyken et al., 2023). Cell function is modulated through the extracellular environment, consisting of microenvironmental factors (such as pH or chemical gradients) and neighboring cells (Bloom & Zaman, 2014; Custódio et al., 2014; Vandereyken et al., 2023). Surrounding cells influence function through paracrine-secreted signaling molecules or juxtracrine signaling (Vandereyken et al., 2023). The spatial organization of cells influences cell differentiation, cell state, cell adhesion, cell migration, and cell signaling, all of which impact brain function (Bloom & Zaman, 2014) and are likely to be differentially responsive to stress and psychopathology, providing insight into the disease course. Thus, it is paramount that we not only build an understanding of the brain at a single-cell resolution but that we do so with spatial context.

To our knowledge, no studies to date have investigated the molecular and cellular impacts of stress on the brain using a spatial omics approach, neither in rodents nor humans, which positions spatial omics as the next frontier in our field. Despite not yet being implemented to examine the biology of stress, spatial omics studies are emerging in other fields. For example, Navarro and colleagues recently implemented Visium spatial transcriptomics (10x Genomics) in the hippocampus and olfactory bulbs of a genetic Alzheimer disease mouse model to investigate the impact of stress on Alzheimer’s pathology development (Navarro et al., 2020). The authors identified that a diverse set of genes involved in the stress response were prominently expressed in both the hippocampus and olfactory bulb of the Alzheimer’s mouse model, elucidating novel drug and mechanistic targets (Navarro et al., 2020). Navarro and colleagues provide an excellent example of how spatial omics can be used in the field of molecular psychiatry to understand stress as a risk factor for psychiatric disorders.

The development and benefits of spatial omics

The earliest spatially resolved technologies were focused on visualizing mRNA and originated in the 1960s with the development of radioactive in situ hybridization (ISH) of ribosomal RNA (Moses & Pachter, 2022). Over the past ∼60 years, this core technology has been built upon and morphed through various methods of multiplexed ISH of RNA with an increasing number of targets and sensitivity over time (Moses & Pachter, 2022). Now, the technologies have dramatically improved resolution and allow the analysis of the entire transcriptome in a largely unbiased manner, at close to single-cell resolution with anatomical context (Ståhl et al., 2016). Spatial methods have shown great promise, with even early versions of spatial transcriptomics methods being paradigm-shifting at the time despite having lower resolution and sensitivity (Batiuk et al., 2022; Maynard et al., 2021; Method of the Year 2020: spatially resolved transcriptomics, 2021).

The field has since been fast-moving toward multiple other spatial omics approaches to explore a wide range of biological processes. For instance, we can now visualize multiple levels of biomolecules in their spatial context, from DNA regulation and gene expression to translational changes, protein expression and post-translational modifications of proteins (Table 1). Although there have been enormous strides in technological development in the last decade, these methods are still in their infancy in their utilization in stress and psychology research, with only two studies to date having used these technologies to further the understanding of psychopathological conditions.

Table 1. Summary of the most commonly used spatial multi-omics platforms and technical specifications.

Spatial omics platforms	Tissue type compatibility: Fresh frozen (FF) or formalin-fixed paraffin-embedded (FFPE)	Biased or unbiased target capture	Sample dimension (mm)	Further analysis possible in the same section	Resolution
Spatial Transcriptomics
Visium (10x Genomics, 2023)	FF and FFPE	Unbiased (Captures Poly(A) tail)	6.5 × 6.5	No	55 µm
Xenium (Janesick et al., 2022)	FF and FFPE	Biased (In situ sequencing – maximum 300 target genes)	12 × 24	Yes	Subcellular
Slide-seqV2 (Stickels et al., 2021)	FF	Unbiased	Scalable	No	10 µm
MERSCOPE (Vizgen, 2023)	FF and FFPE	Biased (smFISH based - up to 500 targets)	1500 × 1500	Yes	Subcellular
Spatial transcriptomics and proteomics
CosMx (He et al., 2022)	FF and FFPE	Biased (smFISH – up to 1,000 RNA and 100 protein targets)	100 × 100	Yes	Single-cell and subcellular
GeoMx DSP (Merritt et al., 2020)	FF and FFPE	Biased transcriptome (over 18,000 targets) Biased proteome (∼ 100 targets)	10 × 10	Yes	Single-cell
DBiT-seq (Liu et al., 2020)	FF and FFPE	Unbiased RNA Biased proteome (up to 22 targets)	∼ 1000 × 1000 (depends on channel width)	No	10 − 50 um
Spatial Epigenetics
Spatial-ATAC-seq (Deng et al., 2022)	FF	Unbiased	∼ 1000 × 1000 (depends on channel width)	No	10 − 50 um
Spatial-CUT&Tag (Deng et al., 2022)	FF	Unbiased	∼ 1000 × 1000 (depends on channel width)	No	20-50 um
Spatial Proteomics
CODEX (Black et al., 2021)	FF, formalin-fixed, and FFPE	Biased (up to 60 targets)	NA	Yes	Single-cell and Subcellular
Immune-SABER (Saka et al., 2019)	FF and FFPE	Biased (∼10 targets)	NA	Yes	Single-cell and subcellular
Spatial Metabolomics, Lipidomics, and Proteomics
MALDI-MSI (Zhu et al., 2021)	FF and FFPE	Biased	NA	Yes	5-20 um

In the first study, Batuik and colleagues performed extensive single-nucleus RNA sequencing (snRNA-seq) in the dorsolateral prefrontal cortex of schizophrenia subjects and matched controls to identify case-control cell-type specific differences (Batiuk et al., 2022). The authors then combined the snRNAseq data with Visium spatial transcriptomics, both to improve the resolution of each dataset and subsequently to validate the differential expression observed by snRNAseq in the upper cortical layers and the gene ontology terms first observed in the snRNAseq (Batiuk et al., 2022). From the analysis of the snRNA-seq data, they observed a reduction in interneuron populations and an increase in excitatory neurons in the upper cortical layer neuronal subtypes, which was validated histologically (Batiuk et al., 2022). Consistent results of neuronal alterations in schizophrenia patients were also reported in the upper layers using spatial transcriptomics, with this region being highly implicated in the symptomology of schizophrenia (Batiuk et al., 2022). This study provides an important proof-of-concept for how droplet-based snRNAseq can be combined with spatial transcriptomics methods to increase the resolution of spatial methods and identify spatial domains of disorder-specific alterations to gene expression. Analysis of gene expression changes with spatial context is essential to adapt to stress-related psychopathology, especially considering the epigenetic nature of the disease development (such as assessing chromatin accessibility using an assay for transposase-accessible chromatin with sequencing (ATACseq)) and progression, which has been evidenced and discussed in depth (Klengel & Binder Elisabeth, 2015).

In the second spatial transcriptomics study, Maynard and colleagues used spatial transcriptomics of the dorsolateral prefrontal cortex to explore the laminar distribution of enriched genes implicated in schizophrenia, bipolar disorder, and major depressive disorder (Maynard et al., 2021). The authors elegantly combined spatial transcriptomics with other large-scale genomics data in this study to improve the sensitivity and contextualization of acquired data. Specifically, the team integrated risk gene data sets from psychiatric disorder cohorts with the spatial transcriptomics data from control subjects to identify cortical layers where these risk genes were enriched (Maynard et al., 2021). Layer 2 and layer 5 were enriched in schizophrenia risk gene expression, and layer 2 was enriched in bipolar disorder risk gene expression (Maynard et al., 2021). Although the analysis was carried out in controls, the ability to characterize risk gene expression in specific cortical layers provides spatial molecular information and, thus, inferred functional information within the region. The next step in illuminating biologically relevant pathology is expanding spatial multi-omics analysis to encompass environmental risk factors and their contribution to psychiatric disorder development.

Spatial multi-omics approaches: technical aspects

Given that the impacts of stress are largely mediated through epigenetic alterations (Matosin et al., 2017; 2018), there is a particular need to analyze the interaction of environmental factors (including stress) and gene regulation and how these mechanisms manifest as the symptomology of psychiatric disorders. Many spatial omics technologies have been developed to address questions at different levels of molecular regulation, with the combination of the data produced providing comprehensive insights into disease. For stress research, these combined approaches will provide a powerful toolkit for unraveling the complexity of molecular responses to stress at a spatial and molecular level. The application of spatial multi-omics has been successful in other fields, such as regenerative wound healing research, with the molecular mechanisms behind fibrosis finally elucidated (Foster et al., 2021). In cancer research, niches of spatially significant transcriptional changes influenced by glioblastoma microenvironments have also been identified, giving insight into tumor development and potential targets for treatment (Ravi et al., 2022). With the success in other fields, this provides a strong proof-of-concept that these methods will be effective in furthering our understanding of the pathoaetiology of psychiatric disorders.

A recent advancement that is predicted to largely impact the success of spatial multi-omics in stress research has been technologies that enable measurement of multiple levels of omics information (e.g., both the transcriptome and proteome) on the same tissue section (Table 1). For example, Nanostring has two platforms (GeoMx and CosMx) that enable proteome and transcriptome analysis in a single tissue section. GeoMx (Merritt et al., 2020) uses uniquely spatially barcoded oligonucleotides that covalently bind via a UV-photocleavable linker to mRNA (over 18,000 targets) and proteins (up to 100 targets) in the same tissue section. When exposed to UV light, the covalent bond is broken, and the barcoded oligonucleotides are released, collected, and sequenced using next generation sequencing (NGS). CosMx (He et al., 2022) is a similar platform that measures a large number of mRNA targets (up to 1,000) and proteins (up to 64 validated targets) at a 3D subcellular resolution in a sample using in situ hybridization of fluorescent molecular barcodes. DBiT-seq (Liu et al., 2020) is another method that utilizes microfluidics to tag mRNA (unbiased) and proteins (up to 22 targets); this method requires a specially designed microfluidics device to deliver the barcodes. In brief, the samples are stained using antibody-delivered-DNA-tags that contain a polyadenylated (polyA) tail and a unique barcode. Microfluidic flow barcoding is then carried out by passing channel-specific probes through microchannels (10 µm width). The probes contain an oligo-dT sequence that binds to mRNA polyA tail or antibody-delivered-DNA-tags, which can then be processed using NGS. Adapting the methods to stress-related psychopathology research makes spatially resolved transcriptome and proteome analysis concurrently on the same tissue section possible. Therefore, stress-related changes to transcription and the direct impacts on the surrounding protein levels can be delineated, highlighting the significance of the technology in the field. Due to the epigenetic nature of the effects of stress, the multi-omic approach would provide insight into the alterations in gene expression and subsequent protein profiles within specific brain regions.

Recent studies have also reported methods to examine ­spatially resolved epigenomics and transcriptomics in the same section, which, in stress research, is particularly needed as ­epigenetic changes underpin the increased risk for psychiatric disorder development and progression (Klengel & Binder Elisabeth, 2015). Two methods include the spatial assay for transposase-accessible chromatin with sequencing (SpatialATAC&RNAseq) (Deng et al., 2022) and the spatial cleavage under targets and tagmentation (Spatial-CUT&Tag) methods (Deng et al., 2022; Zhang et al., 2023). Both methods combine the microfluidics transcriptomic tags of DBiT-seq with the probes of the epigenomic methods ATAC-seq and CUT&Tag. Spatial ATAC&RNAseq is an assay for assessing chromatin accessibility using DNA oligomers inserted into the accessible genome loci by transposase (Tn5). The spatially barcoded oligomers are introduced to the sample using microfluidic channels, and following amplification, the samples are sequenced using NGS. Spatial-CUT&Tag uses antibodies that target histone modifications followed by antibody-targeted tagmentation with Tn5 fused with protein A and spatial barcoding with oligomers through microfluidics.

Some single-omics spatial methods can also be carried out sequentially in the same section. For example, Visium (10x Genomics) (10x Genomics, 2023) can be used following Xenium (10x Genomics) (Janesick et al., 2022) or mass spectrometry imaging (Vicari et al., 2023) to provide a high-resolution, unbiased map of the transcriptome in a sample, coupled with target proteins or the metabolome. Visium uses spatially unique poly-dT oligonucleotides to capture mRNA unbiasedly, with a significant limitation in spatial resolution of approximately 55 µm. In contrast, Xenium requires a panel of RNA and proteins to be preselected; however, the spatial resolution is subcellular. Together these break-through technologies demonstrate the potential for transformation of our molecular understanding of psychiatric disorders, although there are considerations with every application that need to be carefully weighed during the project design phase.

Considerations for experimental design

Although spatial multi-omics methods are invaluable in the analysis they facilitate, there are important considerations when applying them to stress and psychopathology cohorts, as outlined in Table 1. For example, does the sample remain intact following analysis? Depending on the research question, this may be a significant limitation. For instance, if the research question is: how do epigenetic changes translate to transcriptomic outcomes in the hippocampus of stress-related psychiatric disorder patients? Spatial-ATAC-seq or Spatial-ATAC-seq would be used to assess the epigenetic changes; however, both methods also destroy the sample. Therefore, unbiased transcriptomic analysis using Visium would not be possible in the same tissue section. The biased approaches of Xenium or MERSCOPE would be the only viable options for sequential epigenetic and transcriptomic analysis in the same tissue section. Another consideration is tissue compatibility, with several methods such as Spatial-CUT&Tag, Spatial-ATAC-seq, and Slide-seqV2 (Stickels et al., 2021) only compatible with fresh frozen tissue, which may limit the availability of samples. For semi-biased analysis methods such as Xenium, CosMx, GeoMx, CODEX (Black et al., 2021), and Immuno-SABER, the targets must also be preselected (Saka et al., 2019).

Additionally, probes that detect the desired targets simultaneously may or may not be available when using biased approaches. Biased approaches may very well be favorable for some stress research questions; however, usually targets will be limited (e.g., up to 300 targets for Xenium). There is also often a tradeoff between resolution (single-cell or subcellular level) and how biased the approach is, with unbiased approaches having a lower resolution (55 µm for Visium). The specific structures in the tissue samples being analyzed and the sample dimensions must also be checked for compatibility with each technology. For example, Visium has a capture area of 6.5 mm². Splitting the sample across multiple capture areas could be necessary if the tissue is too large, although doing so would dramatically increase the cost and the processing time. There are improvements to spatial transcriptomics techniques on the horizon with higher resolution unbiased platforms being developed and recently released (e.g., VisiumHD (Porterfield, 2022), which enables spatial transcriptomics at single-cell resolution).

Several other considerations that can help improve the reproducibility of the data should be taken into account when planning multi-omics experiments, summarized in further detail by Vandereyken, Sifrim (Vandereyken et al., 2023), Krassowski, Das (Krassowski et al., 2020), and Emmert-Streib (Emmert-Streib, 2022). Briefly, the key to a successful multi-omics study is in the robustness of the experimental design. Considerations include sample size and statistical power; sources of variation, confounders (e.g. PMI) and biases; quality control and assurance measures for the chosen omics and statistical analysis algorithms are included in the design; and cross-validation in cases where biases are unavoidable (Krassowski et al., 2020). A severe testing framework can also be implemented, which provides a computational safeguard for biological relevance and is recommended in conjunction with the measures outlined above (Emmert-Streib, 2022). Other considerations are the high cost and low throughput of current spatial omics platforms, and ideally, we hope to move toward reversal of this balance (low cost and high throughput).

Lastly, it is always important to assess if existing datasets, atlases, and consortia may be adequate to address a particular research question before generating new data. Some sources include the Allen Brain Atlas (https://portal.brain-map.org/) and data from the Common Mind Consortium (https://www.nimhgenetics.org/resources/commonmind) and the PsychENCODE Consortium (https://www.nimhgenetics.org/resources/psychencode). Consortium efforts such as these provide a rich resource of different omics data from psychopathology samples, which have utility for discovery analyses as well as for validation and replication of new findings.

Considerations for integration and analysis of multiple layers of spatial information

As outlined above, several considerations need to be made when planning spatial multi-omics experiments, especially when using human postmortem brain tissue derived from stress-specific psychopathology cohorts with a finite amount of tissue available. Another consideration when designing experiments is how the data will be processed. Whilst the power and utility of spatial multi-omics technologies are evident, the processing and integration of the resulting large datasets are equally important. Computational pipelines for data analysis have been developed alongside spatial technology. However, the analytical pipelines have a number of limitations when integrating the multiple layers of spatial information (data integration). Several biological assumptions in the current integration strategies require resolution, especially methods for integrating multiple datasets (Argelaguet et al., 2021; Vandereyken et al., 2023). To this end, anchors are an important concept used to achieve data integration. Anchors are specifically defined as patterns from two or more omics datasets that are identified as relatively consistent between the samples compared to alternate patterns (Chen et al., 2023). At present, integration strategies for multi-omics can be divided into three categories based on the anchor chosen: horizontal, vertical or diagonal integration (Argelaguet et al., 2021). The most consistent patterns form a link between different spatial data modalities and are used as a reference point when comparing samples.

In summary, integration methods encompass three main methods each which have pros and cons, as summarized in Table 2. Horizontal integration describes the integration of two or more omics datasets derived using the same omics method in different samples (e.g. integrating Visium experiments profiling gene expression from the same tissue region across different donors). In this case, the anchor would be the similarities in expression patterns of captured coding mRNA inherent to the particular tissue region chosen. Note that similar cell states or molecules across different samples are implied in horizontal integration, with the algorithm forcing non-matching patterns, which can mask relevant biological variability during batch correction (Argelaguet et al., 2021; Vandereyken et al., 2023). Vertical integration describes the integration of data from the same sample processed using different omics methods (e.g. integrating Visium and CODEX on the same tissue section). Vertical integration is typically used when processing spatial multi-omics data, as adjacent tissue sections from the same sample are usually used for analysis, with horizontal integration subsequently used in batch correction (Long et al., 2023; Vandereyken et al., 2023). There are a number of challenges associated with vertical integration which are expanded on by Argelaguet & Cuomo (Argelaguet et al., 2021), including integrating molecular readouts from different data modalities, which each have different statistical assumptions (Argelaguet et al., 2021). Diagonal integration is used when there is no anchor (e.g. integrating spatial ATAC-seq and Xenium data from different donors) (Argelaguet et al., 2021; Vandereyken et al., 2023; Xu & McCord, 2022). Diagonal integration is the most difficult to integrate as the data is unpaired. However, this can be addressed by converting the integration into a vertical/horizontal integration task or manifold alignment (Argelaguet et al., 2021; Xu & McCord, 2022). The conversion of diagonal integration relies on strong assumptions (see examples in (Argelaguet et al., 2021)). Manifold alignment, for example, assumes the data generated from the different omics methods has similar distributions or that the data was produced through a similar experimental process (Xu & McCord, 2022).

Table 2. Summary of bioinformatics integration strategies.

Method	Advantages	Disadvantages
Horizontal integration Integration of two or more omics datasets derived using the same omics method in different samples (e.g. integrating Visium experiments profiling gene expression from the same tissue region across different donors)	Relatively straightforward integration Allows for direct comparison of similar samples across different donors – disease-control comparison Experimental consistency	Can mask relevant biological variability during batch correction Forces non-matching patterns, potentially masking biological variability Limited scope into molecular processes as integrated datasets are derived from the same omics method
Vertical integration integration of data from the same sample processed using different omics methods (e.g. integrating Visium and CODEX on the same tissue section)	Integrates different omics methods on the same sample providing comprehensive insights into molecular mechanisms Particularly useful for spatial multi-omics studies where adjacent tissue sections are analyzed Reduces variability introduced by sample-to-sample differences, batch effects, or technical artifacts	Challenges in integrating molecular readouts with different statistical assumptions Assumed data compatibility that may introduce biases or confounders that affect the accuracy and reliability Experimental inconsistency - Does not allow for disease-control comparison
Diagonal integration Used when there is no anchor (e.g. integrating spatial ATAC-seq and Xenium data from different donors)	Can be converted to vertical/horizontal tasks Allows for exploring interactions or associations involving multiple molecular layers, enhancing the depth and breadth of omics analyses Allows for disease-control comparison using multiple omics modalities Enhances compatibility of diverse datasets	Data is unpaired, making integration difficult and reliant on strong assumptions regarding data distributions and experimental processes Experimental inconsistency Reduced statistical power

In addition to integration, methods to readily compare spatially resolved molecular levels (e.g. mRNA and protein) between two or more comparative experimental groups are not well developed. This bioinformatics area requires prompt attention if we are to use these methods to determine condition- and state-related changes. Further, data reproducibility and validation of the findings in spatial multi-omics studies can be difficult, with type 1-error and generalizability significantly contributing to low reproducibility (Perng & Aslibekyan, 2020). In human samples, additional contributions to reproducibility issues in omics studies can be primarily a result of demographic and clinical effects, which are difficult to control. Examples include those related to tissue quality (RNA integrity, postmortem interval or the tissue fixation method used) and the individual’s lifestyle (e.g. smoking and alcohol consumption). In the case of examining the impacts of stress exposures and psychopathological processes, variables such as the type, duration, and severity of stressors (e.g. neglect or abuse), single or comorbid diagnosis, manner of death, and age of disease onset are important to consider. Although the abundance of variables increases the difficulty of delineating biologically relevant or technical differences, the value of these studies for disease understanding and subsequent treatment improvement is high, and we should continue to find the most robust data analysis strategies possible to analyze them.

Future directions

Current spatial omics and multi-omics technologies are powerful in their ability to analyze biologically intricate tissue samples and provide an immense depth of information for complex pathologies. The technologies will be particularly promising for improving our understanding of stress neurobiology and the role of stress as a major risk factor for psychiatric disorders. In the coming years, we envisage the development and more common implementation of novel platforms that bring together multiple levels of spatially resolved omics information (such as epigenetics, transcriptomics and proteomics) at subcellular resolution on a single tissue slice. Applying these techniques in case/treatment-control experimental design, in both animal models and human tissues, will propel our understanding of the consequences of stress and how it contributes to the development of psychiatric disorders, ultimately leading to the identification of novel therapeutic targets for treatment and intervention. For this to become a reality, we must develop effective data integration strategies and methods for accurate case-control comparisons and perform experiments in sample sizes with power to identify significant differences with confidence.

Many questions remain in the stress research field for which spatial multi-omics can provide invaluable answers. Investigating where epigenetic changes are localized anatomically, what regulatory processes are causing them, and how they impact the surrounding transcriptome and proteome will be essential to identifying the mechanisms of stress in psychopathology and potential treatment targets. Spatial multi-omics also allows stress-specific cell subtype dysfunction to be identified with concurrent spatial co-expression network analysis, revealing spatially coordinated cell and molecular pathways involved in stress-related psychiatric disorders. The use of spatial multi-omic approaches in animal models and 3D cell culture models are also needed to provide insights that cannot be gained from postmortem analysis, such as the dynamics of stress and molecular changes occurring over time in longitudinal studies, with a controllable environment.

As aforementioned, we highlight that each currently available spatial technology has limitations, and thus, each experimental design should be carefully planned according to the scientific question. Increasing access to spatial omics technologies through lowering cost, data sharing, increasing throughput capacity, and improving the reproducibility of findings will be ground-breaking for many research fields, including molecular psychiatry. The advancements in molecular psychiatry will continue as improvements are made to data integration computational methods across spatial multi-omics platforms and the ability to compare experimental groups readily. In the future, the ability to analyze complex disorders without bias using spatial multi-omics at single-cell resolution (such as the recently made available Visium HD) and even at subcellular resolution (e.g. with platforms such as Xenium being capable of capturing significant portions of transcriptome and proteome expression) will provide invaluable insight into biological mechanisms underlying risk factors in disorder development. Using these methods to characterize the molecules altered by risk factors, such as stress, and their effects on the microenvironment of the brain can bring us closer to identifying biomarkers for early intervention and delineating biologically distinct disorders that will benefit from defined treatment strategies.

Conclusion

In conclusion, spatial omics and multi-omics technologies hold immense promise for advancing our understanding of stress neurobiology and its profound impact on the development, progression and prognosis of psychiatric disorders. By elucidating the intricate molecular mechanisms underlying stress-related pathologies, these cutting-edge approaches pave the way for identifying novel therapeutic targets and personalized treatment strategies. As we continue to refine data integration strategies, enhance experimental design, and expand access to spatial omics technologies, the field of molecular psychiatry stands poised for ground-breaking discoveries. With the ability to analyze complex disorders at unprecedented resolutions, such as single-cell and even subcellular levels, we are on the precipice of uncovering crucial biomarkers for early intervention and refining our understanding of biologically distinct disorders. Through innovation and collaboration, spatial omics technologies could revolutionize the landscape of psychiatric research, offering hope for improved outcomes and quality of life for individuals affected by these debilitating conditions.

Acknowledgements

The authors would like to thank Dr Cristiana Cruceanu for her helpful comments and feedback.

Disclosure statement

The authors report that there are no competing interests to declare.

Additional information

Funding

This work was supported by the Rebecca L Cooper Medical Research Foundation with a Project Grant and the Al & Val Rosenstrauss Fellowship awarded to Dr Matosin.

Notes on contributors

Amber R. Curry

Amber R. Curry is interested in the neurobiology of stress, mental illness and the human hippocampus.

Lezanne Ooi

Dr. Lezanne Ooi research considers the development of cellular imaging techniques to understand neurodegenerative disease.

Natalie Matosin

Dr. Natalie Matosin is interested in how stress contributes to the pathology of psychiatric illnesses, examining human brain samples with cutting-edge omics technologies.