Observational evidence

Meta-analyses synthesising evidence from cross-sectional studies have demonstrated alterations in levels of a wide range of immune-related markers in people with depression and psychotic disorders (both acute and chronic). ^{Reference Osimo, Pillinger, Rodriguez, Khandaker, Pariante and Howes1} Importantly, meta-analytical evidence demonstrated elevated levels of interleukin-6 (IL-6), tumour necrosis factor-α, interferon-γ, transforming growth factor-β and interleukin-17 (IL-17) in antipsychotic-naïve people with first-episode psychosis, ^{Reference Pillinger, Osimo, Brugger, Mondelli, McCutcheon and Howes2} even before the potentially confounding cardiometabolic effects of antipsychotics and chronicity could manifest.

Longitudinal designs help to clarify the temporal order of events, thereby lending support to causality if alterations in inflammatory markers precede the onset of depression and psychosis. A meta-analysis of longitudinal associations between C-reactive protein (CRP) levels and psychosis showed inconclusive evidence that high baseline CRP levels might increase the risk of subsequent psychosis. ^{Reference Osimo, Baxter, Stochl, Perry, Metcalf and Kunutsor3} CRP is a readily available immune biomarker but is noisy and non-specific. Like many biomarker associations from observational studies, the association of CRP with depression and psychosis may reflect (a) immune changes causally contributing to depression and psychosis; (b) immune changes and depression and psychosis co-occurring as a consequence of shared upstream processes (e.g. early life adversity); or (c) the downstream immune effects of pre-diagnosis depression and psychosis symptoms on lifestyle (e.g. nutrition, exercise). Furthermore, many longitudinal studies measure only immune-related phenotypes at a single time-point, limiting our ability to test for reciprocal (bidirectional) causal associations between immune dysregulation and depression and psychosis.

Genetic and genetic epidemiology evidence

Genetic analyses are a powerful tool for probing causal associations between depression and psychosis and inflammation, because genetic variants are randomly assorted and fixed at conception, thereby reducing the possibility of confounded results.

In Genome-Wide Association Studies (GWAS), people who have the target phenotype or trait (e.g. depression) and healthy controls are recruited to test which genetic variants across the genome – usually single-nucleotide polymorphisms – are significantly associated with the phenotype or trait. Both depression and psychosis have shown associations with regions of the genome related to immune function. ^{Reference McGrath and Lynall4} However, these findings must be interpreted carefully, because most genes are multi-functional with roles in multiple tissues. Genes are often labelled ‘immune’ simply because they have been studied in greater depth in the (more accessible) immune system.

Another way to build on GWAS findings is through Mendelian randomisation. This approach uses genetic variants that are associated with a biological exposure as natural proxies (‘instruments’) to estimate whether that exposure plays a causal role in an outcome. Studies employing Mendelian randomisation have shown that genetically predicted increases in IL-6, CRP and some immune cell genes and phenotypes are associated with higher risk of depression and psychosis, ^{Reference Perry, Upthegrove, Kappelmann, Jones, Burgess and Khandaker5} suggesting causal associations between immune-related markers and depression and psychosis. However, when the associations of genetic variants with the exposure of interest are weak, the ability to make causal inferences in a Mendelian randomisation analysis is limited. GWAS studies of immunophenotypes that are biologically more specific than cytokines, but more experimentally intensive to assay (e.g. cell subset-specific proteins), often have smaller sample sizes, making robust inference about the role of specific cellular functions challenging.

Causal inference and novel drug repurposing approaches

Observational studies and secondary analysis of clinical trials have shown some evidence of an association between people taking immunomodulatory treatments for inflammatory conditions and improvements in psychiatric symptoms. ^{Reference Kappelmann, Lewis, Dantzer, Jones and Khandaker6} However, many of the observational studies have important limitations, such as the comparison group consisting of people who would be never eligible for the target immunomodulatory treatment in clinical practice. Target trial emulation addresses this by explicitly specifying a hypothetical randomised controlled trial (RCT), and carefully emulating its components using observational data. For instance, it helps to identify scenarios in which it is possible to compare individuals prescribed immunomodulators and those who could plausibly have received them but were given alternative treatments. There are now significant ongoing efforts to apply this methodology to data from large-scale administrative registers to assess the impact of immunosuppressants on psychiatric outcomes. In the future, linkage of these administrative registers to biomarker data could allow stratification of people by their inflammatory status.

Complementary to this, genetically anchored methods such as drug target Mendelian randomisation allow to estimate the potential effects of medications on a disorder. Unlike traditional Mendelian randomisation, which models the effect of genetic variants associated with a biological phenotype, drug target Mendelian randomisation identifies variants that mimic the biological action of a drug. Applying this to immunomodulators and psychiatric outcomes could clarify causal pathways and identify novel therapeutic targets. Ongoing work in this area aims to refine these methods and extend them to a broader range of immune targets.

Cellular immunopathology

Until recently, much of the field’s focus has been on acute phase reactants (e.g. CRP) and peripherally measured cytokines. However, it is unclear whether increased IL-6, for example, reflects IL-6 secretion from the brain, from innate immune cells, from adaptive immune cells, from endothelial cells or from adipose tissue, all of which could be contributing to the pool of IL-6 measured in blood. Genetic variants predisposing to multiple psychiatric disorders tend to localise in regions of the genome that are active in brain cells, but also in adaptive immune cells such as T, B and natural killer cells, ^{Reference Lynall, Soskic, Hayhurst, Schwartzentruber, Levey and Pathak7} suggesting an important transdiagnostic causal role for these cells. Recent advances and falling costs in high-dimensional cytometry, single-cell RNA sequencing and other ‘omics’ technologies now facilitate detailed profiling of cellular proteins, gene expression and epigenetic states from the same blood sample. These tools enable investigation at scale of cellular immunopathology at the level of specific immune subsets, or even single cells. We predict that approaches that integrate information from psychiatric GWAS with functional genomic data on how immune cells respond to different contexts will deepen our understanding of how environmental exposures, such as stress and infection, interact with genetic risk to alter disease onset and trajectories. ^{Reference McGrath and Lynall4}

Anti-brain antibodies

A parallel strand of research investigates the association of schizophrenia (and other psychotic disorders) with anti-neuronal antibodies, particularly the N-methyl-D-aspartate receptor (NMDAR). Meta-analyses show that 2–4% of patients with first-episode psychosis have serum NMDAR IgG antibodies, with widely variable rates reported in the general population, ^{Reference Cullen, Palmer-Cooper, Hardwick, Vaggers, Crowley and Pollak8} potentially due to significant differences between live versus fixed cell-based assays. The causal role of these antibodies is well established in autoimmune encephalitis, where psychiatric symptoms often precede neurological manifestations. However, in psychosis cohorts, antibody-positive individuals show only subtle differences in clinical presentation or short-term outcomes compared with antibody-negative peers, ^{Reference Al-Diwani, Handel, Townsend, Pollak, Leite and Harrison9} and most lack overt neurological signs, posing a challenge for the selection of patients who would most benefit from antibody screening.

Identifying practical, meaningful biomarkers

A major methodological obstacle for immunopsychiatry is identifying a sensitive, specific, standardised and temporally stable (at least over days) marker of immune dysfunction that is relevant for the prediction of prognosis or treatment response. To date, no immune stratification variable has demonstrated robust reproducibility and clinical utility across cohorts; efforts such as Advanced Stratification of People with Depression Based on Inflammation (https://www.gamian.eu/advanced-stratification-of-people-with-depression-based-on-inflammation-aspire/), integrating information across trials about baseline immune predictors of antidepressant response, will be critical. An unresolved issue in the field is whether the pathophysiological immune contribution to depression and psychosis is important in a discrete subgroup of individuals, or occurs along a more continuous spectrum. This uncertainty complicates approaches to detection and treatment stratification. Either way, precision immunopsychiatry is most likely to be successful if there is a subgroup of individuals, or a tail of individuals in a normal distribution of immunopathogenesis, in whom immunity plays a central role; mechanistically relevant biomarkers could then support the use of immunomodulatory medications. It will be also critical to understand how the well-described effects of demographic, lifestyle, socioeconomic and environmental factors intersect with psychiatric genetic risk in leading to symptoms.

The field is somewhat in a methodological double-bind: without good biomarkers, it is difficult to conduct well-motivated stratified RCTs, yet the most informative biomarkers of treatment–response may emerge only from such trials. To resolve this, exploratory biomarker studies and clinical trials must proceed in tandem, ideally within the same cohorts. Initiatives such as the United Kingdom and Republic of Ireland (UKRI) ImmunoMind Hub (https://www.mentalhealthplatform.ac.uk/research-hubs/immunomind) and UK Mental Health Goals omics (https://www.gov.uk/government/publications/life-sciences-healthcare-goals/mental-health-goals) programmes exemplify this integrated approach, combining deep multiomic blood phenotyping with clinical outcomes and including built-in potential for patient recall for stratified interventions.

The antibody-positive psychosis literature offers a proof of concept for this integrated approach: here, anti-neuronal antibodies already function as a stratifying variable for trials and may define a pathophysiologically distinct subgroup. It is currently unclear whether immune-modulating therapies outperform standard care for patients carrying these antibodies; multiple randomised trials of B cell-depleting therapies for antibody-positive psychosis are under way.