Introduction
Schizophrenia is a severe mental illness affecting 1% of the population worldwide, and often results in lifelong disability (Saha, Chant, & McGrath, Reference Saha, Chant and McGrath2007). It is heterogeneous in etiology, symptomatology, and prognosis, which makes it difficult to understand and treat. Identification of promising biological entities could support investigations of schizophrenia pathophysiology and advance the development of novel and individualized therapeutic approaches tailored to patients with specific disease etiologies or molecular characteristics (Zhang, Sweeney, Bishop, Gong, & Lui, Reference Zhang, Sweeney, Bishop, Gong and Lui2023).
Identifying the underlying sources of heterogeneity and their biologically meaningful differentiations is an ongoing challenge. Neuroinflammation hypothesis and related immunological pathways have received considerable attention in recent years with a strongly established presence in schizophrenia (Benros, Eaton, & Mortensen, Reference Benros, Eaton and Mortensen2014; Schizophrenia Working Group of the Psychiatric Genomics, 2014). Using post-mortem brain or peripheral inflammatory markers, studies have identified an elevated inflammatory subtype of schizophrenia patients (Boerrigter et al., Reference Boerrigter, Weickert, Lenroot, O'Donnell, Galletly, Liu and Weickert2017; Fillman et al., Reference Fillman, Weickert, Lenroot, Catts, Bruggemann, Catts and Weickert2016; Hoang et al., Reference Hoang, Xu, Lutz, Bannai, Zeng, Bishop and Lizano2022; Lizano et al., Reference Lizano, Lutz, Xu, Rubin, Paskowitz, Lee and Bishop2021), suggesting the identified inflammatory marker differences between cases and controls are driven by a subset of people with psychosis. Based on peripheral epigenetic profiles, our group recently also identified a discrete subtype of schizophrenia patients with notably altered methylation levels of genes related to immune cell function (Luo et al., Reference Luo, Pi, Hu, Wang, Xiao, Li and Lui2021). This immune-related subtype is associated with more severe symptomatology, cognitive impairment, and abnormal brain anatomy (Lizano et al., Reference Lizano, Lutz, Xu, Rubin, Paskowitz, Lee and Bishop2021; Luo et al., Reference Luo, Pi, Hu, Wang, Xiao, Li and Lui2021). Overactivation of the immune system in these individuals is thought to be a pathway for increased brain alteration and increased psychosis risk (Pape, Tamouza, Leboyer, & Zipp, Reference Pape, Tamouza, Leboyer and Zipp2019). Such findings support the feasibility of subtyping the schizophrenia syndrome on biological grounds and the promise of the future treatment developments targeting immune dysregulation in psychosis.
Most previous subtyping studies have recruited patients with chronic disease and on long-term anti-psychotic treatments, which may introduce a number of confounding factors related to long illness duration and anti-psychotic drug exposure (Xiao et al., Reference Xiao, Sun, Shi, Jiang, Tao, Zhao and Lui2018; Zhang et al., Reference Zhang, Deng, Yao, Xiao, Li, Liu and Gong2015). For example, exposure to anti-psychotic drugs could significantly alter levels of cytokines and other inflammatory markers in schizophrenia (de Witte et al., Reference de Witte, Tomasik, Schwarz, Guest, Rahmoune, Kahn and Bahn2014), possibly through metabolic disturbance (Prestwood et al., Reference Prestwood, Asgariroozbehani, Wu, Agarwal, Logan, Ballon and Freyberg2021). Anti-psychotics have also been shown to be epigenetic modifiers via action at the promoters of candidate genes and global change to the methylome (Burghardt, Khoury, Msallaty, Yi, & Seyoum, Reference Burghardt, Khoury, Msallaty, Yi and Seyoum2020). The validity and reliability of the immune subtype in first-episode psychosis have been questioned (Pillinger et al., Reference Pillinger, Osimo, Brugger, Mondelli, McCutcheon and Howes2019). It remains unclear whether there is an immune subtype evident in first-episode psychosis as we have seen in longer-term ill patients. It is crucial to demonstrate the biologically derived subtypes of schizophrenia are detectable and reliable irrespective of the illness stage and treatment status, given its promise to guide future treatment developments and patient stratification,
Here, we conducted a study parsing heterogeneity of schizophrenia based on peripheral blood DNA methylation profiles in a sample of patients with first-episode schizophrenia (FES). The consistency of this classification structure with that of our previously reported cohort of treated, longer-term ill patients with schizophrenia (LTS) was tested. We hypothesized that a similar classification structure would be found in both FES and LTS patients, with high correspondence in molecular, cerebral, and clinical features.
Materials and methods
The study was approved by the West China Hospital research ethics committee of Sichuan University, and written informed consent was obtained from all study participants or their legal guardians before enrolment. All procedures contributing to this work complied with the ethical standard of Helsinki Declaration of 1975.
Participants
Fifty-nine antipsychotic naïve patients with FES (age 30.69 ± 11.34 years, 21 males) were recruited from psychiatric clinics in Chengdu, China. A diagnosis of schizophrenia was established using the Structured Clinical Interview for DSM-IV Axis I Disorders (SCID I) (First, Spitzer, Gibbon, & Williams, Reference First, Spitzer, Gibbon and Williams1995). Illness onset was determined using the Nottingham Onset Schedule (Singh et al., Reference Singh, Cooper, Fisher, Tarrant, Lloyd, Banjo and Jones2005) with information provided by patients, family members, and medical records. Psychopathology ratings were obtained using the Positive and Negative Syndrome Scale (PANSS) (Kay, Opler, & Lindenmayer, Reference Kay, Opler and Lindenmayer1989). Cognitive ability was measured using the Brief Assessment of Cognition in Schizophrenia (BACS) (Keefe et al., Reference Keefe, Goldberg, Harvey, Gold, Poe and Coughenour2004). The BACS composite z-score was calculated by comparing each participant's performance on each measure to the performance of a healthy Chinese comparison group of similar age, recruited from the same geographical region. Fifty-six healthy subjects (age 24.57 ± 2.30 years, 14 males) were recruited from the same geographical region. The non-patient edition of the SCID was used to confirm a lifetime absence of major psychiatric illness, and healthy participants reported no known family history of serious mental illness in their first-degree relatives.
The following exclusion criteria were applied to all participants: (1) neurological disorder or systemic medical illness; (2) alcohol or substance abuse disorder; (3) intellectual disability; and (4) ongoing medical therapies with known impact on central nervous system function or anatomy or immune system function. MR scans (described below) were reviewed by an experienced neuroradiologist to exclude individuals with gross brain abnormalities. Participants were all of Han ancestry and right-handed.
DNA extraction and Methylation profiling
Peripheral blood (5 ml, EDTA 1 mg:5 ml) was collected and immediately frozen at −80 °C until further processing. Genomic DNA was extracted by a phenol-chloroform extraction. DNA concentration and integrity were assessed by a NanoDrop 2000 spectrophotometer (Thermo Fisher Scientific, Waltham, MA, USA) and agarose gel electrophoresis, respectively.
DNA methylation profiles were generated by combining bisulfite conversion of genomic DNA and whole-genome amplification with direct, array-based capture and scoring of the CpG (cytosine-guanine) loci. DNA samples were processed and hybridized to the Illumina Infinium Human-Methylation EPIC 850K BeadChip (Illumina Inc., San Diego, CA), developed to quantitatively assay more than 850 000 methylation sites across the genome at single-nucleotide resolution, following the Infinium HD Methylation Assay Protocol. Hybridized Bead Chips were imaged on an Illumina iScan system following the manufacturer's recommendations. Data were normalized by subtracting the background value, which was determined by averaging the signals of built-in negative control bead types. The normalized data were then used to calculate DNA methylation levels, which were displayed as β-values ranging from 0 to 1, corresponding to unmethylated and fully methylated sites, respectively. Gene annotation was performed using Illumina's annotation of probes.
Methylation data processing
Raw data was processed using the R package minfi (Aryee et al., Reference Aryee, Jaffe, Corrada-Bravo, Ladd-Acosta, Feinberg, Hansen and Irizarry2014; Fortin, Triche, & Hansen, Reference Fortin, Triche and Hansen2017), and probes and samples were filtered by ChAMP (Tian et al., Reference Tian, Morris, Webster, Yang, Beck, Feber and Teschendorff2017) default parameters. Then, quality control on probes and samples was performed. Briefly, probes were excluded if they met one of the following criteria: (1) detection p-value above 0.01; (2) < 3 beads in at least 5% of samples; (3) located on chromosome 23 or non-CpG sites; (4) related to SNP; (5) had multiple hits. Samples with above 10% undetected probes were excluded. Normalization of the methylation beta value was conducted by the BMIQ method (Teschendorff et al., Reference Teschendorff, Marabita, Lechner, Bartlett, Tegner, Gomez-Cabrero and Beck2013). The batch effect of plate, well, sentrix id and sentrix position and potential confounding effects of age and gender on the methylation data were checked by single value decomposition (SVD) analysis (Teschendorff et al., Reference Teschendorff, Menon, Gentry-Maharaj, Ramus, Gayther, Apostolidou and Widschwendter2009). Details are summarized in online Supplementary Material. Technical factors that showed significant contribution to the variation of the dataset were corrected using the Combat function in the sva package (Leek, Johnson, Parker, Jaffe, & Storey, Reference Leek, Johnson, Parker, Jaffe and Storey2012).
Subtype identification
Non-negative matrix factorization (NMF) and k-means clustering were applied together on the top 10 000 CpG sites with greatest variance among patients to identify methylation-related subtypes. NMF is an algorithm that reduces the dimensionality of complex data and can discover subclasses based on feature patterns (Brunet, Tamayo, Golub, & Mesirov, Reference Brunet, Tamayo, Golub and Mesirov2004) and R package NMF was utilized. The output in this study is a decomposition of methylation data into k subtypes with k metagenes, wherein k is the named factorization rank in the algorithm. To find the optimal value of k, rank value from 2 to 10 was tested over 200 runs. K means clustering was then applied with the optimal k as the cluster number on the coefficient matrix to assign samples to subtypes.
Differential methylation probes and region over-representation analysis
Differentially methylated probes (DMPs) were detected between each subtype and their matched controls via R package limma with Benjamini-Hochberg correction. DMPs that met all of the following criteria were kept and the rest were filtered out: (1) average expression value above 0.2; (2) adjusted p value below 0.01; (3) no SNP on the probe site; and (4) a minimum difference of 0.05 in beta value between groups. Genes associated with DMPs were subjected to representation analysis in terms of Gene Ontology Biological Process (GO BP) by R package cluster Profiler (Yu, Wang, Han, & He, Reference Yu, Wang, Han and He2012).
Subtype validation in LTS cohort
In our previous work (Luo et al., Reference Luo, Pi, Hu, Wang, Xiao, Li and Lui2021), we identified a distinct subtype of schizophrenia with unique molecular, cerebral, and clinical features associated with immune dysfunction in a cohort of LTS patients. The consistency of our present classification structure with that previously reported cohort was tested by clustering patients in LTS cohort using features relevant for classification in FES cohort. CpG features relevant for the specific subtype classification were extracted by the featureScore and extractFeature method in R package NMF. Specifically, 1192 CpG features were selected by the NMF classification as subtype relevant features in FES cohort. Then, a Random Forest (RF) model (Python module sklearn) was built based on the 1192 CpG features, and was used to predict the subtypes of LTS patients. Dice similarity coefficients (DSC) were used to evaluate the consistency between previous subtyping and subtyping based on the classification model developed with FES patients. DSC was calculated by multiplying the number of shared bands between two patterns by two and then dividing by the sum of the number of bands in each pattern. Permutation tests (n = 5000) were used to create empirical distributions of DSC to assess the statistical significance of DSC differences between subtypes.
Shared DMPs between two cohorts
The shared DMPs between the subtype with widespread methylation alterations in FES cohort and that subtype in previous LTS cohort were further evaluated. Shared DMPs were examined on methylation sites that met the following criteria in both cohorts: (1) average expression value above 0.2; (2) adjusted p-value below 0.01; and (3) a minimum difference of 0.05 in beta value between group.
The compactness and separation of subtypes and controls were tested based on the shared DMPs. We calculated pairwise Pearson correlation coefficient of samples on the beta methylation value of the shared DMP sites were calculated and determined the sample distance as one minus the absolute value of the coefficient value. Multidimensional scaling (MDS) was applied to obtain 2D coordinates of each sample. An average subtype similarity value distributions between and within subtypes were derived by randomly permuting subtype labels on the 2D plane for 10 000 permutations. p values were calculated based on the permutation.
Cell-type decomposition analysis
A reference-based method was used via GLINT (Rahmani et al., Reference Rahmani, Yedidim, Shenhav, Schweiger, Weissbrod, Zaitlen and Halperin2017) on the DNA methylation profile to estimate blood cell-type proportion in each sample. This method estimates the proportion of monocytes, CD8 + T-cells, CD4 + T-cells, NK cells, B cells, neutrophils, and eosinophils according to the Houseman model (Houseman et al., Reference Houseman, Accomando, Koestler, Christensen, Marsit, Nelson and Kelsey2012), using a panel of 300 highly informative methylation sites in blood and reference data collected from sorted blood cells as reference (Reinius et al., Reference Reinius, Acevedo, Joerink, Pershagen, Dahlen, Greco and Kere2012). It provides prediction of cell type proportion rather than absolute counts.
To test whether the difference in cell-type proportions between P2 subtype and control was consistent in both FES and LTS cohort, and to provide a comprehensive and precise estimate of the effect size, a random effects meta-analysis was used to combine differences in cell-type proportions between P2 subtype and controls across FES and LTS cohorts. Meta-analysis of cell-type proportion in FES cohort and that in LTS cohort was performed using metagen function in meta package (Schwarzer, Reference Schwarzer2007). Estimated effects and standard errors were calculated by linear regression between controls and the subtype of patients across two cohorts.
MRI acquisition and image processing
MRI examinations were performed on 3T SIEMENS TrioTim scanner equipped with a 32-chanel head coil located at the West China Hospital, Sichuan University. High-resolution T1 weighted images and diffusion-weighted images were acquired. Acquisition and processing details are summarized in online Supplementary Material.
Results
Patient subtyping based on DNA methylation profiles in FES cohort
After quality control on the raw data, 726 002 out of 853 307 probes were retained and no subject was dropped. NMF identified two patient subtypes based on the top 10 000 methylation probes with greatest variance among patients, according to the cophenetic coefficient and silhouette value (Fig. 1a and b). We labeled the two patient subtypes as FES-P1 and FES-P2. Thirty-one (52.5%) of 59 FES patients were identified as FES-P1 and 28 (47.5%) of them were identified as FES-P2.
(a) Cophenetic coefficient value and silhouette value summarized through NMF rank value from 2 to 10, consensus was made for each rank over 200 runs in FES cohort. (b) Consensus clustering matrix over 200 runs computed at k = 2–4 by NMF. Samples were hierarchically clustered based on the consensus clustering matrix. Value 1 (dark red) indicates samples were always in one cluster, and value 0 (dark blue) indicates samples were never in one cluster. Both highest cophenetic value and silhouette value occurred at rank = 2. (c) Over-representation analysis of DMPs (between FES-P2 and controls) associated genes in terms of gene ontology biological process. Only top 20 over-represented terms were displayed. NMF, non-negative matrix factorization; FES, patients with first-episode schizophrenia; DMPs, differentially methylated probes.
DMPs in FES cohort
DMPs detection analysis revealed that FES-P2 was more divergent from healthy controls than FES-P1 in terms of blood methylation levels. We observed 16 961 DMPs between FES-P2 and controls and only 86 DMPs between FES-P1 and controls, with the between group differences of beta values above 0.05. Over-representation analysis of genes associated with FES-P2 DMPs via GO BP indicated an enrichment of genes involved in neutrophil activity (Fig. 1c).
Clinical characteristics of subtypes in FES cohort
Clinical characteristics of patients with the two subtypes are shown in Table 1. FES-P2 patients had significantly higher PANSS total and general psychopathology scores ((p = 0.001 and p = 0.003), when compared to FES-P1 subtype. FES-P2 also showed a trend of higher PANSS positive symptom score (p = 0.071) and lower BACS composite scores (p = 0.088) than FES-P1. FES-P1 and FES-P2 patients differed significantly in age with higher mean age of FES-P2 subtype, but not in gender, illness duration or BMI.
Clinical characteristics of methylation subtypes in FES cohort
Abbreviation: BMI, body mass index; PANSS, score: PS, positive symptoms, NS, negative symptoms, GP, general psychopathology, T, total score.
Neuroanatomical features of subtypes in FES cohort
There was no significant group difference in fractional anisotropy across the examined fiber tracts and in subcortical volumes. We explored cortical thickness features at the threshold of uncorrected p < 0.05 at the vertex-level. Patients in FES-P2 had relatively more prevalent cortical thickness increases, predominantly localized in cingulate, parietal, and temporal cortical regions (online Supplementary Fig. S2 in Supplementary Material).
Subtype validation in LTS cohort
The above findings were consistent with our previous work (Luo et al., Reference Luo, Pi, Hu, Wang, Xiao, Li and Lui2021) with LTS patients in which we identified a subtype (we refer it as LTS-P2 subtype in this paper) exhibiting widespread methylation level alterations among genes enriched in with immune cell activity and a significantly higher proportion of neutrophils. We tested the consistency of the subtypes identified in two cohorts by clustering patients in the LTS samples using features that defined subtypes in the FES cohort. Among 38 patients in LTS-P1, 34 patients were classified as P1 subtype (LTS-P1′) based on the FES subtype features, and 4 patients were incorrectly classified as P2 subtype. All patients in LTS-P2 were again placed in the P2 subtype (LTS-P2′) using the FES-based model. The Dice similarity coefficient between LTS-P1 and LTS-P1′ was 0.944, and between LTS-P2 and LTS-P2′ was 0.926, demonstrating the similar classification structure in the FES and LTS samples (p < 0.001).
Shared DMPs across FES and LTS cohorts
We observed 14 311 shared DMPs in FES-P2 and LTS-P2 (Fig. 2a). We tested the compactness and separation of subtypes and controls based on the shared DMPs. MDS plot showed that P2 subtype in both cohorts had a discrete distribution from P1 subtype and controls, while P1 subtype and controls had relatively overlapped distribution (Fig. 2b). The P2 subtype had significantly high compactness both in FES (p < 0.0001) and LTS cohorts (p < 0.0001), significant high separation of LTS-P2 from other patients in the LTS cohort (p = 0.005) and relatively high separation of FES-P2 from other patients in FES cohort (p = 0.072). Enrichment analysis of genes associated with shared DMPs via GO BP indicated an enrichment related to biological process involved in immune cell activity, especially neutrophil activity (Fig. 2c). Those findings demonstrated there is high correspondence in epigenetic features between FES-P2 and LTS-P2 patients.
(a) Shared DMPs of subtype P2 in FES and LTS cohorts. (b) Multidimensional scaling (MDS) plot of pairwise distance across cohorts. The distance was calculated by Pearson correlations based on beta methylation value of shared DMP sites. (c) Enrichment analysis of genes contained shared DMPs via Gene Ontology Biological Process. (d) Enrichment analysis of genes contained DMPs discovered between P2 subtype in FES and LTS cohorts via gene ontology biological process. Only top 20 over-represented terms were displayed. FES, patients with first-episode schizophrenia; LTS, patients with long-term ill schizophrenia; DMPs, differentially methylated probes.
DMPs between P2 subtype in FES and LTS
Differences in methylation profile between P2 subtype in FES and LTS were also explored. We observed 13353 DMPs between FES-P2 and FES-P2 patients, with between group differences of beta values above 0.05. Over-representation analysis of genes associated with DMPs via GO BP indicated an enrichment of genes top-ranking involved in axonogenesis and neuron development (Fig. 2d).
Differential estimates of cell-type proportions
The proportion of specific cell types in peripheral blood was estimated using measures derived from DNA methylation data. Cell-type decomposition analysis identified a significantly higher proportion of neutrophils and significantly lower proportion of lymphocytes, including CD4 + T-cells, CD8 + T-cells, and natural killer (NK) cells in FES-P2 patients compared to both healthy controls and FES-P1 (Fig. 3a). FES-P1 patients showed higher proportion of B cells compared to both FES-P2 and controls.
(a) Cell-type proportions derived from DNA methylation data in FES cohort. Pairwise comparison between control and the two FES patient subtypes were applied. *p < 0.05, **p < 0.01. (b) Forest plots from meta-analyses of differences in cell-type proportions between P2 subtype and controls across FES and LTS cohorts. TE, the effect (the mean difference between P2 subtype cases and controls); s.e. TE, standard error of the effect, FES, patients with first-episode schizophrenia; LTS, patients with long-term ill schizophrenia.
A random effects meta-analysis was done to provide comparable estimates of effect size in cell-type proportions between P2 subtype and control across two cohorts. We found that P2 subtype cases had significantly elevated estimated proportions of neutrophils (mean difference = 0.160; p = 0.008), and significantly lower proportions of CD4 + T-cells (mean difference = −0.060; p < 0.001), CD8 + T-cells (mean difference = −0.056; p = 0.006), and NK cells (mean difference = −0.039; p = 0.010) (Fig. 3b, and online Supplementary Table S1 in supplementary material). Interestingly, the differences in neutrophils, CD4 + T-cells, CD8 + T-cells, and NK cells, were more apparent in LTS cohort, with FES cohort characterized by weaker effects.
Discussion
Based on peripheral epigenetic profiles in a sample of FES patients and a second comparison group of LTS patients, we demonstrated a distinct subtype of schizophrenia with molecular, cerebral, and clinical features associated with dysregulated immune function. To the best of our knowledge, this is the first study using epigenetic data and a data-driven analytic method to identify reliable biologically defined subtypes of schizophrenia across FES and LTS patients. This clustering of patients shows high correspondence across two patient samples, suggesting that our reported patient classification was robust irrespective of treatment status or stage of illness.
Heterogeneity has been established in multiple domains of deficit in schizophrenia, including electrophysiology (Clementz et al., Reference Clementz, Parker, Trotti, McDowell, Keedy, Keshavan and Tamminga2022) and neuroanatomy (Ivleva, Turkozer, & Sweeney, Reference Ivleva, Turkozer and Sweeney2020; Xiao et al., Reference Xiao, Liao, Long, Tao, Zhao, Luo and Gong2022). There is a growing and compelling body of evidence that subclinical inflammation in the central nervous system and immune dysregulation play a role in the etiopathogenesis of schizophrenia, but only in a subgroup of patients (Miller & Goldsmith, Reference Miller and Goldsmith2017). Across studies using somewhat different measures, 30 to 50% of patients with schizophrenia demonstrated an elevated inflammatory status (Boerrigter et al., Reference Boerrigter, Weickert, Lenroot, O'Donnell, Galletly, Liu and Weickert2017; Fillman, Sinclair, Fung, Webster, & Shannon Weickert, Reference Fillman, Sinclair, Fung, Webster and Shannon Weickert2014; Fillman et al., Reference Fillman, Weickert, Lenroot, Catts, Bruggemann, Catts and Weickert2016; Lizano et al., Reference Lizano, Lutz, Xu, Rubin, Paskowitz, Lee and Bishop2021). In the context of gene-environment interplay, epigenetic mechanisms represent a plausible molecular source of etiopathological heterogeneity and have received substantial attention in schizophrenia (Smigielski, Jagannath, Rössler, Walitza, & Grünblatt, Reference Smigielski, Jagannath, Rössler, Walitza and Grünblatt2020). Focusing on the biological heterogeneity of epigenetic features, our study provides novel evidence for a biologically distinct subtype of schizophrenia defined by altered methylation profiles related to immune system dysregulation that is present both at illness onset and after years of illness and treatment. Moreover, subgroup with widespread methylation level alterations showed greater symptom severity than patients with limited alteration in methylation profiles, resembling previous observations that elevated inflammatory status is associated with more severe symptom manifestation (Lizano et al., Reference Lizano, Lutz, Xu, Rubin, Paskowitz, Lee and Bishop2021). The association with clinical features and consistent findings across FES and LTS cohorts provides independent validation for the robustness and clinical relevance of our findings.
It is noteworthy that the shared epigenetic features of the immune-related subtype in both cohorts showed high associations with neutrophil activity. Though by indirect measures, our cell-type decomposition analysis revealed significantly higher proportion of neutrophils and lower proportion of lymphocytes in this specific subtype than the other one or healthy controls. Elevated neutrophil counts or neutrophil-to-lymphocyte ratio (NLR) has been considered as a proxy marker for inflammation and has been reported in patient samples of schizophrenia (Bhikram & Sandor, Reference Bhikram and Sandor2022; Moody & Miller, Reference Moody and Miller2018; Steiner et al., Reference Steiner, Frodl, Schiltz, Dobrowolny, Jacobs, Fernandes and Bernstein2020). Specifically, findings suggest that NLR is elevated during the FES, remains elevated throughout the course of the disorder, decreases during remission, and is associated with symptom severity. Our results suggest abnormal neutrophil activity as the core features of this distinct subtype of patients and support a role of inflammation in schizophrenia. Moreover, those defining features do not form later in the course of illness or in relation to treatment but are present close to illness onset before treatment. Thus, they are more likely to be fundamental to illness etiology than perhaps representing consequences of illness.
The concept that schizophrenia involves a propensity toward overactivated immune system suggests an anti-inflammatory or immunomodulatory treatment may be clinically beneficial. Some studies have shown beneficial results of some anti-inflammatory drugs on symptom severity in patients with schizophrenia (Cakici, van Beveren, Judge-Hundal, Koola, & Sommer, Reference Cakici, van Beveren, Judge-Hundal, Koola and Sommer2019), though most give the drugs to all patients rather than those with immune activation. Studies are needed to treat the select patients with immune alteration. Deciding how to best target the immune activation in schizophrenia may benefit from a better understanding of the specific epigenetic and immunological features associated with its presence.
We did not find significantly greater abnormalities of white matter tracts and subcortical volume in the immune-related subtype of FES cohort. It is noteworthy that the subtype with alterations in methylation and immune features exhibited nominally significant cortical thickness increases (not significant after multiple comparison corrections), potentially reflecting neuro-inflammation and pseudothickening of gray matter (Lizano et al., Reference Lizano, Lutz, Xu, Rubin, Paskowitz, Lee and Bishop2021; Pasternak et al., Reference Pasternak, Westin, Bouix, Seidman, Goldstein, Woo and Kubicki2012). The lack of significant structural alterations could be related to the relatively early disease stage of those patients, as we and others have repeatedly shown that brain alterations are often more pronounced in longer-term ill patients which could make subgroups more readily distinguishable. This interpretation was supported by our observation of a smaller effect size of difference in proportions of neutrophils in the FES cohort, and greater amount of DMPs in our previous LTS cohort compared with its matched controls (14311 DMPs in FES-P2 VS 63720 DMPs in LTS -P2). Another possibility is that the presence of immune activation over time contributes to a progressive alteration of brain anatomy seen in some patients. Directed comparison of methylation profile between the immune subtype of two cohorts revealed DMPs with an enrichment of genes top-ranking involved in axonogenesis and neuron development, indicating greater involvement of deficient neurodevelopment in the later course of disease. This is in accordance with progressive cerebral alterations reported in schizophrenia (Velthorst et al., Reference Velthorst, Fett, Reichenberg, Perlman, van Os, Bromet and Kotov2017), which could be related to either longer-term course of illness or anti-psychotic drug exposure. Free from those effects, alterations in FES patients are more likely to reflect the intrinsic pathological changes of the disease.
Several issues need to be considered when interpreting our findings. First, while we demonstrated the existence of an immune-related subtype in FES and LTS cohorts, our study is cross-sectional. Longitudinal data are needed to elucidate how this subtype may develop in disease course and how this classification is consistent within subjects over time. Second, the direct relation between methylation modifications and both protein expression and immune system biology is yet to be established. Lastly, in FES cohort, the mean age of immune-related subtype was older than the other one. To explore the age effect on methylation profiles, we performed an additional analysis with age included as covariates. The results still hold when age is accounted for (details are provided in the online Supplementary material), indicating limited effect of age on difference of methylation profiles between subtypes. The finding that FES-P2 group was older but more clinically impaired may reflect that the effects related to immune dysregulation are more robust than the effects of age in this regard.
In summary, the current study demonstrated the existence of an immune subtype is derivable and robust in both untreated FES and LTS patients. It is a particularly important outcome, considering that FES cohort is an entirely new sample of psychosis and healthy individuals, separated from the recruitment of LTS cohort. The remarkable similarities across two patient samples illustrate the promise of biologically-based illness subtyping and potentially of novel treatment strategies targeting immune dysregulation for a subpopulation of schizophrenia patients.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0033291724000667
Data availability statement
Data are available upon request to the corresponding authors.
Author contributions
Lui S, Gong Q, Xie D, and Luo C contributed to study design. Luo C, Xiao Y, and Hu N contributed to acquisition of data. Luo C, Pi X, and Zhang Q contributed to analysis and interpretation of the data. Luo C and Pi X contributed to the drafting of the manuscript. Sweeney J, Bishop J, Lui S, and Xie D made critical revision to the manuscript for important intellectual content. Lui S, Gong Q, and Xie D had full access to all data in the study and took responsibility for the integrity of the data and the accuracy of the data analysis.
Funding statements
This study was supported by National Key R&D Program of China (Project Nos.2022YFC2009901, 2022YFC2009903, 2022YFC2009900), the National Natural Science Foundation of China (Project Nos. 82120108014, 82102000, 82071908, 81901702), Sichuan Science and Technology Program (Project Nos. 2021JDTD0002), Chengdu Science and Technology Office, major technology application demonstration project（Project Nos. 2022-YF09-00062-SN, 2022-GH03-00017-HZ）and 1.3.5 Project for Disciplines of Excellence, West China Hospital, Sichuan University (Project Nos. ZYGD23003).
Competing interests
None.
Code availability
Relevant code is available at https://github.com/pixuenan/FES_methy.
