Volunteers aged 18–60 from Budapest and Greater Manchester were recruited to participate in the NewMood study (New Molecules in Mood Disorders, Sixth Framework Program of the European Union LHSM-CT-2004-503474) without receiving any compensation via advertisements, an online platform, and general practices. After providing written informed consent, participants received a saliva-based genetic sampling kit for genotyping and a questionnaire pack, consisting of a detailed background evaluation and multiple sub-units for in-depth phenotyping of several depression-related markers and endophenotypes (23, 24). Of the NewMood dataset, comprehensive mapping to enable exclusion of related individuals and those with incomplete relevant phenotypic and genotypic data yielded the present sample of 1,762 unrelated adults with self-reported European ethnic white origin, who were genotyped for the IL6 gene.

Ethical approval was obtained from both local ethics committees (North Manchester Local Research Ethics Committee, Manchester, UK, and Scientific and Research Ethics Committee of the Medical Research Council, Budapest, Hungary). The study was carried out in accordance with the Declaration of Helsinki.

Participants provided buccal mucosa cells via a genetic sampling kit including a cytology brush (Cytobrush plus C0012, Durbin PLC) sent by mail to detect DNA. Extraction of genomic DNA was carried out according to the protocol of Freeman et al. (29). Genotyping was performed using Illumina CoreExom PsychChip. Genotyping was carried out in accordance with ISO 9001:2000 quality management requirements, and was blinded regarding phenotype.

Variant annotation was carried out based on the GRCh37/hg19 human assembly. For the purpose of phasing SHAPEIT was used to estimate additional haplotypes, followed by imputation of missing genotypes via IMPUTE2. This process yielded 186 SNPs on the IL6 gene with boundaries extended by 10 kilobase (kb) pairs at both sides.

Genotyping provided a dataset including 1,762 individuals genotyped for 186 SNPs in the IL6 gene (with boundaries extended by 10 kb) in the NewMood database. For SNPs we applied a 3-step quality control protocol including calculation of Hardy-Weinberg Equilibrium (HWE; >1 × 10⁻⁵), missingness rates (MF; <0.05), and minor allele frequencies (MAF; >0.01) along the criteria suggested by Coleman et al. (30). For quality control purposes and main effect/interaction models Plink v1 0.9 was used. 86 SNPs survived quality control and were entered into analyses.

Logistic and linear regression models were applied to explore the effect of genotype and gene-environment interactions on lifetime suicide attempts (SUIC) as a dichotomous variable and on current suicidal ideation (SI-BSI03), hopelessness (H-BSI18) and current thoughts of death (ToD-BSI21) as continuous outcome variables, respectively. First, main effects of genetic variation on all phenotypes were analysed (Supplementary Table 1), followed by gene-environment interaction models with early childhood adversities (CHA) in case of suicide attempts (SUIC), current suicidal ideation (SI-BSI03), hopelessness (H-BSI18) and current thoughts of death (ToD-BSI21) (Supplementary Table 2); and with recent negative life events (RLE) in case of current suicidal ideation (SI-BSI03), hopelessness (H-BSI18) and current thoughts of death (ToD-BSI21) (Supplementary Table 3).

Next, all regression models, including main effect and interaction, were followed up by a clumping process using the CLUMP function in Plink. Clumping is a statistical method for yielding clumps of intercorrelated SNPs based on empirical estimates of their linkage disequilibrium (LD), stepping beyond independent significance levels, identifying connected SNPs and their index or top SNP (the one with the highest significance). The four parameters used for clumping were as follows: (1) maximum p-value of the clump's top SNP was 0.001; (2) maximum p-value for secondary SNPs was 0.05; (3) minimum LD R² was 0.5; and (4) physical distance threshold was 250 kilobases.

As secondary analyses, we also analysed the interaction effects of different types of recent life events including financial difficulties (RLE-financial), personal problems (RLE-personal), intimate relationship problems (RLE-relationship), and social network disturbances (RLE-social) (28) on current indicators of suicide risk. Intercorrelations between the subscales have also been previously reported, and were either negligibly weak or non-significant (31).

All analyses were run according to additive, dominant, and recessive models. In all statistical models, population (Manchester or Budapest), gender and age were entered as covariates. In addition, in the case of gene-environment interaction models, the main effects of both the SNP and the stressors (CHA/RLE) were also included as covariates.

Nominal significance threshold was set at p < 0.05. Bonferroni-method was used to correct for multiple comparisons. Assuming 33 models [in main effect: 4 outcome variables (SUIC/SI-BSI03/H-BSI18/ToD-BSI21), × 3 genetic models (additive/dominant/recessive) = 12 models; in interaction: 1 outcome variable (SUIC) × 1 environmental variable (CHA) × 3 genetic models (additive/dominant/recessive) = 3; and 3 outcome variables (SI-BSI03/H-BSI18/ToD-BSI21) × 2 environmental variables (CHA/RLE) × 3 genetic models (additive/dominant/recessive) = 18 models] the corrected significance threshold was set at p = 0.0015. For the secondary analyses, assuming 36 models [3 outcome variables (SI-BSI03/H-BSI18/ToD-BSI21) × 4 environmental variables (RLE intimate/financial/personal/social) × 3 genetic models (additive/dominant/recessive) = 36 models], significance threshold was set at p = 0.0014. The main steps of genotyping and the statistical analyses are illustrated in Figure 1.

In the above analyses, Plink v1.90 was used to calculate missingness rate (MR; <0.05), Hardy-Weinberg equilibrium (HWE; >1 × 10⁻⁵) and minor allele frequency (MAF; >0.01) as part of quality control steps prior to the analyses; for clumping; and for building linear and logistic regression models to test for main and interaction effects of genetic variation in the IL6 gene. Analyses were supported by scripts individually written in R 3.0.2 (R Core Team, 2013). R was also used to illustrate the effects of significant findings (version 4.0.3 with the ggplot2 package). Descriptive statistics were run using IBM SPSS Statistics 25.

All data used in the study were openly shared and are available in the Figshare depository at https://figshare.com/s/6d0f6f781466c78a7c0a.

Descriptive statistics of the analysed sample are shown in Table 1.

Logistic regression models with SUIC (for lifetime suicide attempts) and linear regression models with SI-BSI03, H-BSI18, and ToD-BSI21 (for suicidal ideation and other markers of current suicidal risk) identified several individual SNPs reaching nominal significance threshold, however, as all p-values exceeded the maximum threshold specified for clumping (p = 0.001), clumps could not be identified.

Logistic regression models on the interaction between IL6 and childhood adversities on lifetime suicide attempts yielded three significant clumps (one for additive and two for dominant models), surviving correction for multiple testing (Table 2; Figure 2). In the additive model, a clump containing three SNPs emerged with rs2069835 as the top SNP and the minor C allele as a protective allele (p = 0.0004; Figure 2A). In the dominant models, one identified clump contained four SNPs with rs2069835 as the top SNP and the minor C allele as a protective allele (p = 0.0004; Figure 2B), while the other contained 3 SNPs, with rs1880241 as the top SNP and the minor G allele as a risk allele (p = 7.08 × 10⁻⁵; Figure 2C).

Linear regression models on the interaction between IL6 and childhood adversities on current suicidal ideation (SI-BSI03) yielded three significant clumps (two for additive and one for dominant models) all consisting of one SNP each, surviving correction for multiple testing (Table 3; Figure 3). In the additive model, one identified clump contained rs2069837 with the minor allele G as a risk allele (p = 0.0006; Figure 3A). Furthermore, one clump containing only one SNP, rs7458109, was significant in both additive and dominant models, with the minor C as a risk allele (p = 0.0006 and p = 0.0004 for additive and dominant models, respectively) (Figures 3B,C).

Linear regression models on the interaction between IL6 and childhood adversities on current hopelessness (H-BSI18) yielded no significant clumps.

Linear regression models on the interaction between IL6 and childhood adversities on current thoughts of death (ToD-BSI21) yielded three significant clumps (one for additive, dominant and recessive models) surviving correction for multiple testing (Table 4; Figure 4). In case of the additive model, a clump containing 24 SNPs with rs1474348 as top SNP and the minor C allele as a protective allele emerged (p = 0.0005; Figure 4A). In case of the dominant model, a clump containing 5 SNPs with top SNP rs4719714 and the minor T allele as a risk allele was found (p = 0.0007; Figure 4B). In the recessive model, a clump containing 19 SNPs with rs2069845 as top SNP and the minor G allele as a protective allele was identified (p = 0.0002; Figure 4C).

Linear regression models on the interaction between IL6 and recent negative life events (RLE) on current suicidal ideation (SI-BSI03) yielded three significant clumps (one for dominant and two for recessive models) surviving correction for multiple testing (Table 5; Figure 5). In the dominant model, a clump containing seven SNPs with top SNP rs4719713 and the minor G allele as a risk allele was identified (p = 0.0003; Figure 5A). In the recessive model, one identified clump contained 4 SNPs with top SNP rs7805828 and the minor A allele as a protective allele (p = 7.39 × 10⁻⁵; Figure 5B); while the other clump contained six SNPs with top SNP rs60056354 and the minor T allele as a protective allele (p = 0.0007; Figure 5C).

No significant clumps emerged in linear regression models on the interaction between IL6 and recent negative life events (RLE) either on current hopelessness (H-BSI18), or current thoughts of death (ToD-BSI21).

Linear regression models on the interaction between IL6 and recent negative life events related to financial problems (RLE-financialC2) on current suicidal ideation (SI-BSI03) yielded two significant clumps surviving correction for multiple testing, one of which was significant in both additive and dominant models. In the additive and dominant models, a clump containing three SNPs with top SNP exm609179 and the minor A allele as a risk allele was identified (p = 0.0002), while in the recessive model a clump with 15 SNPs including top SNP rs1554606 and the minor T allele as a protective was found (p = 0.0002). In interaction with life events related to personal problems (RLE-personalC3) linear regression identified one clump in the IL6 gene which survived correction for multiple testing in both additive and dominant models, containing 6 SNPs including top SNP rs34328912 with the minor C allele as a risk allele (p = 0.0003). Life events related to intimate relationships (RLE-intimateC1) or social network (RLE-socialC4) did not interact with IL6 variation on current suicidal ideation.

On the other hand, recent intimate relationship-related life events and social network-related life events interacted with IL6 variation on current thoughts of death (ToD-BSI21). In case of recent intimate relationship life events (RLE-intimateC1) two interacting clumps surviving significance for multiple testing were identified. The clump included 16 SNPs in the additive and 14 SNPs in the dominant model; top SNP was rs7802307 for both models with the minor T allele as a protective allele (p = 0.0008 and p = 0.0009 for additive and dominant models, respectively). There was also a significant interaction between IL6 variation and recent life events related to the social network (RLE-socialC4) according to the recessive model surviving correction for multiple testing yielding one clump including 2 SNPs with top SNP rs7802307 and the minor T allele as a protective allele (p = 0.0005). In case of current thoughts of death (ToD-BSI21), no significant interaction effects emerged in case of the other two types of life events, financial problems (RLE-financialC2) or personal problems (RLE-personalC3).

In case of hopelessness (H-BSI18), no significant interaction effects were found with any recent life event subtypes.

Genomic location of significant SNPs and top SNPs identified in the clumping procedure are shown in Figure 6. Figures are based on diagrams generated by LDlink, LDmatrix Tool (https://ldlink.nci.nih.gov/?tab=ldmatrix).

To detect the functional effect of the top significant SNPs, we utilised FuncPred (https://snpinfo.niehs.nih.gov), and SNPNexus (https://www.snp-nexus.org/v4/guide/#cons), including DeepSea to predict significance of predicted chromatin effect and evolutionary conservation; FunSeq2 to calculate a non-coding score for the identified variants; GWAVA to calculate functional probability, and Genomic Evolutionary Rate Profiling. Results are shown in Supplementary Tables 4–12 separately for all analyses where significant clumps were identified.

Cytokines are a heterogeneous group of regulatory polypeptides, released by immunocompetent cells such as lymphocytes and macrophages, playing a role in host defence and repair processes of tissues. and besides being able to cross the blood-brain barrier by active transport, are also produced in the brain by neurons, astrocytes and activated microglial cells, thus providing a direct link with the immune system (32). Cytokines are increasingly recognised as messenger molecules, that modulate neuroendocrine functions, participate in neuroinflammatory and neurodegenerative processes, and have also been implicated in neurodevelopmental disorders and in the neurobiology stress-related conditions such as mood disorders or suicide (6, 15, 33, 34). Proinflammatory cytokines IL-6, IL-1, and TNF-alpha appear to be the most relevant considering their actions in the brain (34), and one of their most remarkable and relevant characteristics is their pleiotropy to bind different cell target types (6).

IL-6 is a pleiotropic cytokine mediator of immune reactions with functions including production of acute phase proteins in liver, B-lymphocyte proliferation and differentiation, osteoclast activation, and fever induction in brain. There are two types of IL-6 signalling: the anti-inflammatory or classical pathway, and the proinflammatory pathway or transsignalling way. IL-6 binds to membrane-bound IL-6R in the classical pathway, and to soluble IL-6R in the proinflammatory pathway. The proinflammatory pathway is used by various cell types within the brain (34). IL-6 can directly bind to receptors on neurons in the central nervous system and has been shown to directly impact generation of action potentials, which may influence changes in emotion and behaviour including increased risk for suicidal intent and behaviour (35).

Cytokines including IL-6 contribute to changes in the brain and behaviour via several mechanisms. Cytokine receptors, including IL-6 receptors can be found in specific brain regions and can thus have a direct effect on neuronal function and induce specific effects on behaviour and emotion; for example, IL-6 receptors are located on serotonin neurons in the medulla oblongata, hypothalamus, hippocampus, cerebellum, and prefrontal cortex. Cytokines also modulate concentration and function of monoamine neurotransmitters and especially serotonin and metabolites in various brain regions implicated in the pathophysiology of suicide including the prefrontal cortex, hippocampus and amygdala (5, 14). Cytokines also induce changes in the function of the HPA-axis, the other major system implicated in suicidal behaviour (14). Furthermore, inflammation and increased cytokine levels are also associated with activation of the kynurenine pathway (36) with downstream production of metabolites with effects on glutamate neurotransmission (37). Ultimately, disruption of regulatory corticostriatal circuits as a consequence of increased cytokine levels lead to abnormalities in executive function, emotion regulation, hopelessness and impulsivity contributing to suicide vulnerability (38). However, given their role in the central nervous system, cytokines including IL-6 quite likely exert a profound influence via their effect on foetal and later neurodevelopment (39).

Genetic component of suicidal behaviour is estimated at 43% (40) with 30–50% for a broader suicidality phenotype (41). In a previous GWAS, strongest candidate genes of suicidal behaviour were linked to inflammatory response (42), and recent studies identifying top ranking molecular markers predicting suicidal ideation lead to 8 genes including IL6 (besides SAT1, SKA2, SLC4A4, KIF2C, MBP, JUN, and KLHDC3) (16, 43, 44). The imbalance between anti- and proinflammatory cytokines in suicidal behaviour also suggests that cytokine genes may be putative contributors of suicide risk (45), with heritability for IL-6 levels estimated in twin studies ranging between 0.25 and 0.26 when taking into account confounding factors such as BMI (46).

IL6 gene in humans is mapped to region p15-21 on chromosome 7. It includes four introns, five exons, encoding a precursor protein with a total length of 212 amino acids, yielding a 186-amino acid mature segment and a 28-amino acid signal sequence (34). A very recent study using an SNP-based correlation and Mendelian randomisation approach showed a consistent association between increased IL-6 signalling using different genetic proxies and suicidality (47).

In spite of its suspected role in suicidal behaviour, only a few studies directly investigated genetic variation in IL6 in association with suicide. One study assessed two variants in suicide attempters, completers and controls, selected based on previous studies showing their role in IL6 expression regulation (48, 49). While no difference was found in rs2069845 allele frequencies in the three groups, the C allele of promoter-based rs1800795, possibly influencing histone acetylation and transcription factor binding and associated with higher IL-6 levels (48, 49), was more common in completed than attempted suicide, and was also associated with the lethality of suicide attempts (45) suggesting that higher IL-6 levels may mediate this association. Interestingly, in a previous study the G allele of rs1800795 was found to be neuroprotective considering its role on hippocampal volume in healthy subjects (50). Considering that smaller hippocampal volume predicts suicide attempts in depressed patients (51), this association may account for one of the effects of IL6 gene on suicide risk.

Remarkably, both of the candidate variants rs2069845 and rs1800795 investigated by the study of Eftekharian et al. emerged as members of significant clumps in our present analyses, which, in contrast, investigated all 186 SNPs available in the NewMood database without hypothesis-based variant pre-selection. However, in our analyses the mentioned variants did not influence suicidal behaviours as a main effect, but only in interaction with environmental events, and rs1800795 showed an opposite effect, as in our study the C allele appeared as a protective variant.

Our finding that effect of IL6 variation was only observable in those exposed to some forms of environmental influences, either early traumas or recent negative life events, falls in line with previous observations that in several psychiatric conditions, including depression (31, 52–55), genetic factors do not directly influence manifestations, but rather exert their effects via influencing sensitivity towards environmental factors, and it seems to be the case also for both suicide and inflammation.

Stress has a well-known impact on neuropsychiatric disease which may in part be mediated via immunological mechanisms especially in the case of mood and anxiety disorders (4). Exposure to stressful life events and social stressors such as social defeat, loss, separation or rejection may cause impairments in various aspects of immune function, including stress-related changes in cytokine levels and function, and proinflammatory cytokine upregulation in both human and animal models (56). Inflammation increases threat-related neural sensitivity to negative social experiences to enhance sensitivity to threats to well-being or safety in order to avoid them, and also enhances reward-related neural sensitivity to positive social experiences to increase approach-related motivations towards others who might provide support and care (56). Psychological stress in clinical studies leads to elevated inflammation (57), with IL-6 levels up to four times higher in chronic stress compared to no stress (34, 58–61). Acute stress (62) and life-event related stress (63) also correlates with increased blood IL-6 levels (34).

Genetic variability likely interacts with environmental stressors leading to increased inflammatory markers in genetically susceptible individuals (14). However, only a few studies looked at the GxE effect in the background of increased IL-6 levels in the face of stress, and no previous studies investigated the effect of IL6 variation on suicidal manifestations in a GxE approach. In one study, older adults losing a spouse did not only show a two-fold increase IL-6 levels, but this effect was moderated by the G allele of candidate variant rs1800795 in the IL6 gene (64), the same variant implicated in the aforementioned candidate gene study on the association between IL6 SNPs and suicide (45), and identified also in our analyses in two variant clumps. Similarly to the study of Schultze-Florey et al. in our models rs1800795 interacted with environmental influences, but with early childhood adversities rather than recent life events. Furthermore, in line with the results of Schultze-Florey et al. the C allele which was associated with an unchanged IL-6 level in the face of stress emerged as a protective allele in our models. Finally, we observed no main effect of this variant, similarly to the results of Schultze-Florey et al. where genotypes alone did not account for observed differences in circulating IL-6 in bereaved subjects and non-bereavers (64).

While this SNP did not show an effect in our study in interaction with recent stress, we found several clumps of variants interacting with negative life events occurring in the past year on current suicidal ideation. We could not identify clumps of variants interacting with recent stress on other suicide risk markers such as current thoughts of death or hopelessness. However, as specific genes and variants may mediate the effects of only specific types of stressors (53–55, 65), we also looked at the interaction between IL6 variation and distinct recent stress types. Our models identified separate variant clumps significantly interacting with recent financial problems or personal problems on suicidal ideation, and with intimate relationship problems and social network problems on current thoughts of death, which once again suggests that different types of stressors may exert their effect via the mediation of different genetic variants.

Early life adversity has been described as a major contributor to enhanced vulnerability to lifetime suicide (66). Childhood maltreatment affects immune response predisposing to heightened inflammatory states lasting into adult life with long-term consequences on both brain and behaviour (67), including abnormal cortisol stress responses, persistently increased levels of inflammatory cytokines including IL-6 and TNFα, low-grade elevations in other proinflammatory markers such as CRP, as well as greater inflammatory responses to later psychosocial stress (68). In a study investigating depressed patients raised inflammatory gene expression including cytokine-producing genes was linked to childhood adversity coupled with high suicide risk, and reduced gene expression was observed in those without childhood adversity and low suicide risk, suggesting the involvement of the immune system as a mediator between childhood adversity and increased suicide risk through a long-lasting activation of inflammation (69). In our study, variation in IL6 showed robust interaction effects with childhood adversity on both lifetime suicide attempts and current suicidal ideation and current thoughts of death, emphasising the role of inflammation and as well as the role of IL6 variation in predisposing to suicide risk in those exposed to early adversities, which might be exploited for risk screening and identification of those particularly in need of preventive interventions. As mentioned before, it is also notable that we identified in our clumps two variants previously investigated in association with suicide in candidate variant studies, providing further support to their putative role.

Hopelessness has long been identified as one of the key and independent risk factor of suicidal behaviour (70). In our study we found no association between IL6 variation and hopelessness, either as a main effect or in interaction with recent stress or childhood adversities. This is surprising in the light of previous studies which found positive correlations between IL-6 levels and hopelessness in serum and saliva samples taken at different times of the day (71), reported associations between hopelessness and IL-6 levels in depressed patients (72), and found that adolescents exhibiting higher levels of hopelessness reflecting cognitive vulnerability responded with higher increase of IL-6 levels to laboratory-based acute stress (73).

While in our study we focused on the effects of linkage disequilibrium-based clumps of variants in the IL6 gene rather than on single SNPs, and used top SNPs only to characterise the identified significant clumps, several of the member SNPs of significant clumps have previously been implicated in the literature in association with possibly relevant phenotypes.

Most importantly, as mentioned above, rs1800795, a member of two significant clumps interacting with childhood adversities on current thoughts of death, have previously been implicated in association with suicide (45), although with another direction; and with higher circulating IL-6 levels for G allele carriers in interaction with recent stress (64), which corresponds to our results where the minor C was associated with a protective effect. It is also noteworthy that in another study we have previously reported that the rs1800795 in interaction with recent stress increased depression risk and had an impact on depression severity (74). These convergent findings underscore the role of this particular variant in mediating the effects of stress on suicide and related phenotypes. rs1800795 has also been associated with a number of somatic conditions including cancer, inflammatory and autoimmune disorders. rs1800795 is located in the putative promoter of IL6 and affects binding of transcription factors GATA1, GATA2, and YYI as well as NF1 and Sp1 (49) with likely functional consequences as reflected by the positive associations for this variant in divergent studies.

Rs7801617, a member SNP in a clump significantly interacting with childhood adversities on previous suicide attempts has previously been implicated in response to escitalopram in the GENDEP study (75). Another SNP in the same clump, rs2069824, has also been associated with treatment response in major depression and suggested to be a candidate for further investigation (76). These results may once again converge with our findings suggesting further studies on the association of these IL6 variants with the efficacy of antidepressants in reducing the risk of suicidal behaviour.

Rs1880241, a top SNP of a clump significantly interacting with childhood adversities on lifetime suicide attempts has been found to be related to level of circulating CRP in a study revealing a protective effect of CRP on schizophrenia and a risk effect of CRP on bipolar disorder (77). IL-6 is an upstream stimulator of CRP production (35, 47), and many of its effects on relevant phenotypes may be mediated by CRP. Finally, rs4719714, top SNP of a clump significantly interacting with childhood adversities on current thoughts of death have been implicated in association with fatigue and sleep disturbances (78, 79).

Several of our identified variants have been implicated in other conditions such as somatic illnesses, references to these and results of functional prediction are shown in Supplementary Tables. However, the functional consequences or the effect on cytokine levels or function in case of the majority of these variants are unknown.

Understanding the role of inflammation and the contribution of the IL6 gene has several potential clinical implications. IL6 genotype may be utilised as a potential biomarker to identify specific subgroups of depressed patients with increased vulnerability for inflammation; and to detect at-risk subjects especially following early childhood adversities or recent stress and possibly target them with preventive measures focusing on inflammation (6). In this regard it is important that besides anti-inflammatory agents, psychological therapies, such as cognitive-behaviour therapy (80) or supportive-expressive dynamic psychotherapy (SEDP) has been shown to decrease IL-6 levels (81), without the side-effects of pharmacological agents, making subjects who carry the risk variants and exposed to stress suitable candidates for such intervention. The association of IL6 genotype with the efficacy of such therapies on decreasing both IL-6 levels and suicide risk might also be tested. One further important question arising partly from our study is that whether childhood adversities in interaction with IL6 genotype have a harmful effect on multiple measures of suicide risk, then how viable it is to offset or remediate the dire consequences of childhood trauma and maltreatment (67) before the onset of clinical symptoms by psychosocial or pharmacological interventions focusing on inflammation.

Besides finding potential biomarkers of acute and long-term suicide risk, a novel approach to treatment of suicide with novel treatment targets is also needed, especially as current treatments focus solely on treating the primary psychiatric diagnosis (5). Specific effects of cytokines on emotion and behaviour in different brain areas could be exploited for more optimal treatment targets, by identifying upstream triggers of inflammation and downstream neurobiological effectors and moderators conveying vulnerability or resilience (5). There is significant therapeutic promise in addressing different facets of the immune system (4), and immunotherapies may be helpful for depressed patients with low-grade inflammation or associated risk factors or suicide (47). Besides potential new targets, several anti-inflammatory treatments clinically approved for other indications might be tested in suicidal patients. Once more, use of potential biomarkers such as IL6 variation may help identify those who may benefit from such agents even in the face of the abovementioned limiting factors.

Several limitations of our study must be taken into account when interpreting the impact of our results. First, lifetime suicide attempts, as well as both early childhood adversities and recent life events occurring in the past year were retrospectively assessed based on self-report, and are thus subject to recall and reporting biases. Current suicidal ideations, thoughts of death and hopelessness are similarly self-reported. Second, assessment of childhood adversity and recent negative life events does not take into consideration the differing severity and subjective impact of individual life events. Thirdly, we could not exclude the contribution of depression to our suicidal phenotypes. Finally, besides IL6, IL6R encoding the IL-6 receptor is also very important in determining IL-6 levels and functions, thus variation in this gene should also be covered by future broader studies to understand the impact of IL6. Nevertheless, our study also has several strengths, including consideration of several variants without hypothesis-based candidate SNP selection along the IL6 gene with a linkage disequilibrium-based clumping method, employing different manifestations of suicidal behaviour along the suicide spectrum as phenotypes, and using a GxE paradigm with two etiologically different types of stressors.