Dissecting the genetic overlap between severe mental disorders and markers of cellular aging: Identification of pleiotropic genes and druggable targets

Abstract

Patients with severe mental disorders such as bipolar disorder (BD), schizophrenia (SCZ) and major depressive disorder (MDD) show a substantial reduction in life expectancy, increased incidence of comorbid medical conditions commonly observed with advanced age and alterations of aging hallmarks. While severe mental disorders are heritable, the extent to which genetic predisposition might contribute to accelerated cellular aging is not known. We used bivariate causal mixture models to quantify the trait-specific and shared architecture of mental disorders and 2 aging hallmarks (leukocyte telomere length [LTL] and mitochondrial DNA copy number), and the conjunctional false discovery rate method to detect shared genetic loci. We integrated gene expression data from brain regions from GTEx and used different tools to functionally annotate identified loci and investigate their druggability. Aging hallmarks showed low polygenicity compared with severe mental disorders. We observed a significant negative global genetic correlation between MDD and LTL (r_g = −0.14, p = 6.5E−10), and no significant results for other severe mental disorders or for mtDNA-cn. However, conditional QQ plots and bivariate causal mixture models pointed to significant pleiotropy among all severe mental disorders and aging hallmarks. We identified genetic variants significantly shared between LTL and BD (n = 17), SCZ (n = 55) or MDD (n = 19), or mtDNA-cn and BD (n = 4), SCZ (n = 12) or MDD (n = 1), with mixed direction of effects. The exonic rs7909129 variant in the SORCS3 gene, encoding a member of the retromer complex involved in protein trafficking and intracellular/intercellular signaling, was associated with shorter LTL and increased predisposition to all severe mental disorders. Genetic variants underlying risk of SCZ or MDD and shorter LTL modulate expression of several druggable genes in different brain regions. Genistein, a phytoestrogen with anti-inflammatory and antioxidant effects, was an upstream regulator of 2 genes modulated by variants associated with risk of MDD and shorter LTL. While our results suggest that shared heritability might play a limited role in contributing to accelerated cellular aging in severe mental disorders, we identified shared genetic determinants and prioritized different druggable targets and compounds.

Introduction

Patients with severe mental disorders such as bipolar disorder (BD), schizophrenia (SCZ) and major depressive disorder (MDD) show premature mortality and decreased life expectancy, up to 10–20 years compared with the general population, mainly due to increased incidence of aging-related conditions such as cardiovascular and metabolic disorders [1]. In addition, severe mental disorders have also been associated with a steeper age-related declined in executive function, as well as with changes in brain structure commonly associated with aging, such as ventricular enlargement, loss of grey matter in the cerebellum and hippocampus, and decreased volume of the prefrontal cortex [2–4] These findings led to hypothesize that severe mental disorders might be characterized by accelerated biological aging, a condition in which the rate of biological aging is increased as compared to chronological aging. Accordingly, a large number of studies reported alterations of markers of accelerated cellular aging, such as leukocyte telomere length (LTL) and mitochondrial DNA copy number (mtDNA-cn) in patients with severe mental disorders [5]. Telomeres are specialized and highly-conserved DNA-protein structures consisting in multiple (TTAGGG)n repeats and associated proteins called shelterins. These specialized structures prevent chromosome shortening and chromosome end fusion, ensuring chromosome stability. While telomeres shorten after each cell division in most somatic tissues, with this progressive decline leading to cell senescence, this physiological shortening can be accelerated by different stressors such as inflammation or oxidative stress [6]. Most studies to date investigated LTL, a practical measure that correlates with telomere length in different tissues [7]. Mitochondrial dysfunction, which results in alteration in energy production and increased levels of reactive oxygen species, is another widely investigated hallmark of cellular aging [8, 9]. mtDNA-cn is a biomarker of mitochondrial function which has been repeatedly associated with overall mortality and age-related diseases. This double-stranded DNA molecule is strongly affected by aging-associated mutations due to its high replicative index, oxidative environment with reactive oxygen species generated by mitochondria during ATP synthesis, and limited DNA repair capacity due to its lack of protective histones [8]. While under physiological conditions the mtDNA-cn remains relatively stable, a decrease in mtDNA-cn has been described in different disorders characterized by oxidative stress and dysfunctions of the respiratory chain in mitochondria [9]. However, an increase in mtDNA-cn in these conditions has also been described, possibly due to a compensative upregulation aimed at overcoming the bioenergetic defects caused by some mtDNA mutations [10].

While a growing number of studies suggested that patients with severe mental disorders show biological features suggestive of accelerated cellular aging such as shorter LTL or altered mtDNA-cn, contrasting results have also been reported and there is scarce knowledge on the factors that might accelerate or rather counteract accelerated cellular aging in these patients. For instance, some studies showed longer LTL in patients with BD compared with non-psychiatric controls, with this finding potentially explained by a putative protective effect of the mood stabilizer lithium against accelerated cellular aging [11–13]. In the case of mtDNA-cn, while some studies reported a reduction in patients with severe mental disorders compared with controls [14], other studies reported results in the opposite direction or no significant difference [15, 16].

Both mental disorders and hallmarks of aging have a genetic component. The largest GWAS of LTL was conducted in 472,174 well-characterized participants in the UK Biobank [7]. This study identified 197 variants associated with LTL at 138 genomic loci (108 of which were novel). In addition, genetically determined LTL was associated with multiple biological traits (e.g. bone marrow function) as well as a number of diseases spanning neoplastic, vascular and inflammatory disorders [7]. A recent study that derived mtDNA-cn in 383,476 UK Biobank participants of European ancestry using the AutoMitoC pipeline, which leverages single nucleotide polymorphisms (SNP) array probe intensities, identified 71 loci significant at a genome-wide threshold [17]. Few studies have tried to assess whether the putative relationship between mental disorders and hallmarks of aging might be affected by genetic determinants. In a previous study, we used two-sample mendelian randomization analysis to evaluate the bidirectional association between genetically determined LTL (using summary statistics from a GWAS including genetic data for 37,684 individuals [18]) and predisposition to BD (using data from PGC BD freeze 2, including 20,352 cases and 31,358 controls [19]) and reported negative results [12]. Consistently, another study did not find any significant association between polygenic risk for BD, MDD or SCZ and LTL in a sample including 351 participants characterized for depression using self-reported measures [20]. However, a recent study conducted in the larger UK Biobank cohort for which LTL measurements are now available, reported a polygenic risk score (PRS) for depression to be associated with shorter telomeres (β = −0.006, adjusted p = 0.001) [21], suggesting the existence of a possible link between mental disorders and genetically predicted LTL. To our knowledge, no study explored shared genetic bases between severe mental disorders and mtDNA-cn.

Thanks to the development of analytical methods based on pleiotropy, novel genetic variants associated with severe mental disorders as well as genetic determinants shared with related traits, have been identified [22–26]. However, to date, this approach has never been applied to LTL or mtDNA-cn. In this study, we used state-of-the-art approaches to quantify the genetic overlap between severe mental disorders and LTL or mtDNA-cn and identify shared genetic determinants and their direction of effect (i.e. whether they predispose patients to accelerated cellular aging or rather play a counteractive role). We used MiXeR to quantify the trait-specific and shared architecture of mental disorders and hallmarks of aging and the conditional/conjunctional false discovery rate (condFDR/conjFDR) method to detect shared genetic loci. Finally, we integrated gene expression data from brain regions and used different tools to functionally annotate and prioritize identified loci and investigate their druggability. Our results elucidate genetic markers shared between severe mental disorders and markers of cellular aging.

Materials and Methods

GWAS samples

We conducted a cross-trait analysis using the largest publicly available releases of GWAS summary statistics from PGC for BD and SCZ, or PGC and iPSYCH for MDD [27–29] and for two markers of biological aging (LTL and mtDNA-cn) from UK Biobank [7, 17]. For BD and MDD, to avoid sample overlap, we used the GWAS summary statistics excluding participants from UK Biobank. The BD sample included 40,463 cases from 56 cohorts collected in Europe, North America and Australia and 313,436 controls of European origin [27]. The SCZ sample included 53,386 cases and 77,258 controls of European origin [28], while the MDD sample 166,773 cases and 507,679 controls of European origin [29]. GWAS summary statistics for genetically determined LTL [7] and for mtDNA-cn (ascertained using the AutoMitoC pipeline, which leverages SNP array probe intensities) [17] were obtained for 472,174 and 383,476 UK Biobank participants, respectively. For all GWAS datasets, ethical approval was obtained by the original GWAS studies and quality control procedures, including adjustment for population stratification, were performed by the original studies. Analyses were conducted on autosomal variants common to GWAS on mental disorders and LTL or mtDNA-cn, after the exclusion of ambiguous variants (A/T and C/G) or variants located in regions characterized by strong LD such as the MHC region (chr6:25119106-33854733), chromosome 8p23.1 (chr8:7200000–12500000) and the MAPT gene (chr17:40000000–47000000).

Global genetic correlation analysis

Global genetic correlation analysis was conducted using LDSC [30]. To this aim, summary statistics were converted into the LDSC format, while linkage disequilibrium scores were computed using 1000 Genomes European data [30, 31]. The cross-trait LDSC method represents an extension of single-trait LDSC to estimate heritability and genetic correlation from GWAS summary statistics. This method allows to study the genetic correlation globally, considering the average of the shared signals across the genome, including the contribution of SNPs that do not reach genome-wide significance [31], taking into account possible sample overlap and population stratification.

Bivariate causal mixture model (MiXeR)

We used bivariate causal mixture model (MiXeR, v. 1.3) to study shared genetic bases between severe mental disorders and LTL or mtDNA-cn beyond genetic correlation. This method quantifies the genetic overlap between two traits estimating the total number of shared and trait-specific genetic variants [32]. We first constructed conditional QQ plots, which represent modified versions of the standard QQ plots that allow to visualize the cross-trait polygenic enrichment. These plots are constructed by creating subsets of SNPs based on the level of association with the secondary phenotype (using three thresholds p ≤ 0.10, p ≤ 0.01 and p ≤ 0.001). Under the null hypothesis, nominal p-values follow the straight line, while under cross-trait polygenic enrichment they show progressive leftward deflections. Next, we applied causal mixture models to quantify the genetic overlap between severe mental disorders and markers of biological aging. We first computed univariate estimates to quantify trait-influencing loci for each trait of interest. Next, we computed bivariate estimates of genetic overlap between each mental disorder and markers of biological aging. Detailed information on the modelling of genetic effects is available in the original MiXeR article [32]. Briefly, additive genetic effects are modeled as a mixture of components that are plotted in Venn diagrams and that represent SNPs not associated with any of the two traits, SNPs associated with only one trait or with both. In addition, we computed the Dice coefficient (DC), which represents the proportion of SNPs shared between two traits out of the total number of SNPs estimated to be associated with both traits.

Conditional and conjunctional false discovery rate analysis

To identify shared loci between severe mental disorders and markers of biological aging, we used the condFDR/conjFDR method implemented in pleioFDR [33, 34]. This method represents a complementary approach compared with MiXeR as the latter offers information about the total amount the genetic overlap between pairs of traits, while the condFDR/conjFDR method allows to identify the specific shared genetic variants. Specifically, this method allows to re-adjust the GWAS statistics in a primary phenotype (e.g. BD) by leveraging pleiotropic enrichment with a GWAS in a secondary phenotype (e.g. LTL). For each p-value in the primary phenotype, condFDR estimates are obtained by calculating the stratified empirical cumulative distribution function of the p-values [35]. The strata are obtained by the enrichment of SNP associations depending on increased p-values in a secondary phenotype [35]. The conjFDR method is an extension of condFDR aimed at discovering SNPs associated with two phenotypes simultaneously. After inverting the roles of the primary and secondary phenotypes, the conjFDR is defined as the maximum of the two condFDR values. A conjFDR < 0.05 was considered to be significant, as in previous studies [35–38].

Definition of genetic loci and functional enrichment

Independent significant genetic loci were defined according to the FUMA protocol [39]. Lead SNPs were defined by double clumping (a clumping of SNPs significant and independent at r² < 0.6, and a secondary clumping of these SNPs at r² < 0.1). Loci separated by a distance lower than 250 kb were merged. 1000 genome phase 3 was used as a reference panel to compute linkage disequilibrium in FUMA. The direction of allelic effects for significant variants was evaluated by comparing betas reported in the original GWAS. Positional and functional annotation of lead SNPs was performed using different tools. The Combined Annotation-dependent Depletion (CADD) score, which predicts how deleterious a variant is on protein structure/function by contrasting variants that survived natural selection with simulated mutations, was computed in FUMA [40]. RegulomeDB rank (from 1 to 7, with 1 being associated with the highest evidence of functional effects) was calculated using RegulomeDB in FUMA [41] based on known and predicted regulatory elements including regions of DNase hypersensitivity, binding sites of transcription factors and promoter regions. Next, we searched whether SNPs acted as eQTLs based on genotyping and gene expression data (obtained from a range of 114 - 209 samples) from Genotype-Tissue Expression (GTEx) v.8 in brain regions. While aging is multifactorial process associated with changes at a multiple-organ level, we chose to specifically focus on genes predicted to be modulated in brain regions in order to select targets with the highest probability of playing a relevant role for both phenotypes under study (aging and severe mental disorders), also based on the fact that this list of genes was used as input to investigate druggability. In the GTEx project, gene expression was measured with Illumina TrueSeq RNA sequencing or Affymetrix Human Gene 1.1 ST Expression Array, while genotyping data were obtained with whole genome sequencing, whole exome sequencing, Illumina OMNI 5 M, 2.5 M or Exome SNP arrays [42]. We reported cis eQTLs in a +/− 1 Mb cis window around the transcription start site (TSS) and significant based on FDR. For genes for which expression levels are modulated by SNPs associated with increased risk of mental disorders and shorter LTL or lower mtDNA-cn, we tested the functional enrichment for gene ontology (GO) terms using enrichR [43], adjusting results based on FDR. In addition, we investigated whether proteins encoded by the identified genes showed significant PPI enrichment using STRING [44]. A significant PPI indicates that the identified proteins have more interactions among themselves than would be expected for a random set of proteins of the same size and degree distribution drawn from the genome. Genes modulated by SNPs associated with increased predisposition to mental disorders and shorter LTL or lower mtDNA-cn were searched in DGIdb [45] to assess whether they are known targets of existing drugs (drug-gene interactions) or ‘potentially druggable’ based on their involvement in selected pathways, molecular functions or gene families [45]. The DGIdb database classifies genes in categories based on information retrieved from different drug target repositories (DrugBank, PharmGKB, Chembl, Drug Target Commons, Therapeutic Target Database and others). In addition, we searched for upstream regulators of prioritized genes using IPA (Ingenuity System Inc, USA). Upstream regulators are defined as genes, microRNAs, transcription factors or chemical compounds that affect the genes of interest through effects on expression, transcription, activation, molecular modification, transport or binding events according to the Ingenuity Knowledge Base, a large collection of observations in various experimental contexts. P-values of overlap for each upstream regulator based on Fisher’s exact test calculated by IPA were adjusted for multiple testing with FDR (a q-value < 0.05 was considered to be significant).

Results

Pleiotropic enrichment and quantification of genetic overlap

Among mental disorders, only MDD showed a significant negative global genetic correlation with LTL (r_g = −0.14, p = 6.5E−10). Conversely, the global genetic correlation between LTL and BD (r_g = −0.03, p = 0.18) or SCZ (r_g = 0.00, p = 0.98) was not significant. LTL and mtDNA-cn were significantly positively correlated (r_g = 0.12, p = 9.4E−04), while mtDNA-cn was not significantly correlated with any mental disorder (BD: r_g = −0.005, p = 0.88; SCZ: r_g = 0.02, p = 0.82; MDD: r_g = 0.01, p = 0.74).

On the other hand, conditional QQ plots suggested cross-phenotype polygenic enrichment between all severe mental disorders and genetically determined LTL (Supplementary Fig. 1). As shown in Supplementary Fig. 2, analyses conducted with MiXeR showed a much larger polygenicity for severe mental disorders than for markers of aging. Nonetheless, based on the positive AIC reported in Table 1, for both LTL and mtDNA-cn, the model fitted by MiXeR was found to explain the GWAS signal better than a model considering the maximum possible polygenic overlap given a trait’s architecture (i.e. a model in which the causal variants of the least polygenic trait form a subset of the causal variants of the most polygenic trait, best vs max AIC) or a model with the minimal possible polygenic overlap (i.e. a model constrained to a specific value of the genetic correlation, best vs. min AIC), supporting the existence of a polygenic overlap.

Table 1.

Proportion of overlapping genetic variants on the total polygenicity of each trait.

Trait 1	Trait 2	Dice, mean (se)	% of variants associated with mental disorders and shared with aging markers	% of variants associated with aging markers and shared with mental disorders	best vs min AIC	best vs max AIC
BD	LTL	0.05 (0.02)	3%	66%	12.91	10.03
BD	mtDNA-cn	0.02 (0.01)	1%	30%	1.87	3.45
SCZ	LTL	0.03 (0.01)	2%	40%	6.03	9.08
SCZ	mtDNA-cn	0.01 (0.01)	1%	27%	2.44	6.91
MDD	LTL	0.07 (0.01)	3%	91%	2.09	3.45
MDD	mtDNA-cn	0.01 (0.01)	1%	27%	2.06	5.33

best vs max AIC: comparison of the best model fitted by MiXeR vs the model with maximum possible polygenic overlap given trait’s genetic architecture. A positive value means that best model explains the observed GWAS signal better than the max model, despite its additional complexity (due to the fact that MiXeR has to find the value of the polygenic overlap). best vs min AIC: best model vs the model with the minimal possible polygenic overlap. AIC Akaike Information Criterion, BD bipolar disorder, LTL leukocyte telomere length, MDD major depressive disorder, mtDNA-cn mitochondrial DNA copy number se standard error, SCZ schizophrenia.

A large proportion of variants estimated to be associated with markers of biological aging were also associated with severe mental disorders (Table 1). Namely, of variants associated with LTL, 66%, 40% and 91% were also shared with BD, SCZ and MDD, respectively. A lower but still substantial proportion of variants associated with mtDNA-cn was shared with severe mental disorders (from 27% for SCZ and MDD to 30% for BD. Conversely, when considering variants associated with severe mental disorders, only a limited proportion (from 1% to 3%) was also shared with markers of biological aging (Table 1).

Genetic loci shared between severe mental disorders and markers of aging

Fifteen genetic loci including 17 lead SNPs were shared between BD and LTL at a conjFDR < 0.05 (Supplementary Table 1). Of the 17 lead SNPs, 6 were associated with an increased risk of BD and shorter LTL (while 11 with increased risk of BD and increased LTL). Six of the 17 lead SNPs were found to act as expression quantitative trait loci (eQTL) for different genes in at least one brain region (Supplementary Table 1). However, only the intergenic SNP rs10822056, suggested to act as an eQTL for the ADO gene in the caudate brain region, was associated with increased predisposition to BD and shorter LTL, while all the other SNPs acting as eQTLs were associated with BD and longer LTL. Four genetic loci including 4 lead SNPs were shared between BD and mtDNA-cn at a conjFDR < 0.05 (Supplementary Table 1). The allele associated with increased predisposition to BD was associated with lower mtDNA-cn for 2 SNPs and with higher mtDNA-cn for the other 2 SNPs. Only one SNP (rs3208937) was found to act as an eQTL for different genes in different brain regions (Supplementary Table 1).

A total of 52 loci were shared between SCZ and LTL at a conjFDR < 0.05 (Supplementary Table 2). Of the 55 lead SNPs, 33 were associated with an increased risk of SCZ and shorter LTL. Twenty-eight of the 55 lead SNPs were found to act as eQTLs for different genes in at least one brain region (Supplementary Table 2). Of these, 16 SNPs were associated with increased predisposition to SCZ and shorter LTL, while the other SNPs acting as eQTLs with increased predisposition to SCZ and longer LTL. Genes modulated by SNPs associated with increased predisposition to SCZ and shorter LTL were enriched for 2 cellular components and 9 molecular GO terms (Supplementary Table 3). Proteins encoded by the identified genes also showed a significant protein-protein interaction (PPI) enrichment (PPI enrichment p-value: 1.0E-05, Supplementary Figure 3). Twelve genetic loci were shared between SCZ and mtDNA-cn at a conjFDR < 0.05 (Supplementary Table 2). Of these, 3 SNPs were associated with increased risk of SCZ and shorter mtDNA-cn and all acted as eQTLs for different genes in at least one brain region (Supplementary Table 2). Genes modulated by these SNPs were enriched for 4 biological processes GO terms: Negative Regulation Of Cell Population Proliferation (GO:0008285, p = 7.3E−04, FDR = 0.04, OR = 22.4, genes: EIF2AK2, CDK10, GAS8); Negative Regulation Of Cellular Process (GO:0048523, p = 0.002, FDR = 0.04, OR = 15.6, genes: EIF2AK2, CDK10, GAS8); Regulation Of Cell Population Proliferation (GO:0042127, p = 0.006, FDR = 0.04, OR = 10.8, genes: EIF2AK2, CDK10, GAS8); Cellular Component Assembly (GO:0022607, p = 0.007, FDR = 0.04, OR = 19.1, genes: FANCA, GAS8).

Finally, 19 loci were shared between MDD and LTL at a conjFDR < 0.05 (Supplementary Table 4). Of the 19 lead SNPs, 13 were associated with increased risk of MDD and shorter LTL and 6 were significant eQTLs for different genes in at least one brain region (Supplementary Table 4). Four of the 6 eQTLs were associated with increased predisposition to MDD and shorter LTL. Genes modulated by these SNPs were significantly enriched for 5 biological processes with different GO terms (Supplementary Table 5). Only 1 locus with rs4955411 as the lead SNP was shared between MDD and mtDNA-cn. This SNP was found to be a significant eQTL for several genes in different brain regions (Supplementary Table 4).

Druggable genes and upstream regulators

The ADO gene, which was modulated by SNPs associated with BD and increased biological aging, is not part of the druggable genome based on the Drug Gene Interaction Database (DGIdb). Conversely, 14 and 7 genes modulated by SNPs associated with shorter LTL and SCZ (Table 2) or MDD (Table 3), respectively, were druggable.

Table 2.

Druggable genes modulated by SNPs associated with SCZ and shorter LTL or lower mtDNA-cn.

Chr	Lead SNP	Location	EA/ OA	beta SCZ	beta LTL	eQTL in brain regions (GTEx v. 8)	Druggable genes
1	rs4844621	C1orf132	A/G	0.05	−0.01	CD46 (amygdala, anterior cingulate cortex, caudate, cerebellum, cortex, frontal cortex, hyppocampus, hypothalamus, nucleus accumbens, putamen, substantia nigra)	CD46
10	rs17883150	GSTO1	G/A	0.04	−0.01	RP11-127L20.3 (amygdala, anterior cingulate cortex, caudate, cerebellum, cortex, frontal cortex, hippocampus, hypothalamus, nucleus accumbens, putamen); GSTO2 (amygdala, anterior cingulate cortex, caudate, cerebellum, cortex, frontal cortex, hippocampus, hypothalamus, nucleus accumbens, putamen); ITPRIP (cortex)	GSTO2
11	rs10838843	Intergenic	G/T	−0.07	0.01	FOLH1 (caudate, cerebellum, nucleus accumbens, putamen)	FOLH1
12	rs73230058	PITPNM2	C/T	0.06	−0.01	ABCB9 (cerebellum); KMT5A (cerebellum)	ABCB9
15	rs16969894	IREB2	C/T	0.07	−0.01	CHRNA3 (caudate, nucleus accumbens); CHRNA5 (nucleus accumbens)	CHRNA3, CHRNA5
15	rs4702	FURIN	G/A	0.08	−0.01	FURIN (frontal cortex)	FURIN
20	rs310653	SRMS	T/C	−0.04	0.01	PTK6 (caudate)	PTK6
22	rs4822000	RP11-12M9.4	T/C	−0.04	0.01	MCHR1 (cerebellum); RP11-12M9.4 (cerebellum); SLC25A17 (cerebellum, nucleus accumbens); ZC3H7B (cerebellum)	MCHR1, SLC25A17
22	rs138832	BRD1	A/G	−0.05	0.01	ALG12 (hippocampus, nucleus accumbens); CRELD2 (hypothalamus); RP3-522J7.6 (cerebellum); ZBED4 (cerebellum)	CRELD2

Chr	Lead SNP	Location	EA / OA	beta SCZ	beta mtDNA-cn	eQTL in brain regions (GTEx v. 8)
2	rs11899117	EIF2AK2	G/A	0.04	−0.01	NDUFAF7 (cerebellum); EIF2AK2 (cerebellum)	EIF2AK2
16	rs164749	Intergenic	T/G	−0.04	0.01	SPATA33 (anterior cingulate cortex, caudate, cerebellum, cortex, hippocampus); LINC02166 (anterior cingulate cortex, caudate, cortex, nucleus accumbens); GAS8 (anterior cingulate cortex, cerebellum, cortex, nucleus accumbens, putamen); CDK10 (cerebellum); FANCA (cerebellum); VPS9D1 (cortex)	CDK10
22	rs41297816	TBX1	A/G	−0.05	0.01	DGCR12 (amygdala); TXNRD2 (caudate)	TXNRD2

Table 3.

Druggable genes modulated by SNPs associated with MDD and shorter LTL.

Chr	Lead SNP	Gene	EA/ OA	beta MDD	beta LTL	eQTL in brain regions (GTEx v. 8)	Druggable genes
1	rs4653448	Intergenic	A/G	0.02	−0.03	PARP1 (cerebellum)	PARP1
3	rs34614773	RNF123	T/G	0.03	−0.01	FAM212A (cerebellum); GMPPB (cerebellum, cortex, hyppocampus, putamen); HYAL3 (cerebellum), MST1R (caudate, cerebellum, cortex); RBM6 (amygdala, anterior cingulate cortex, caudate, cerebellum, cortex, hippocampus, hypothalamus, nucleus accumbens, putamen, substantia nigra); RNF123 (amygdala, anterior cingulate cortex, caudate, cerebellum, cortex, hypothalamus, nucleus accumbens)	HYAL3, MST1R, RNF123
11	rs112181005	C11orf49	G/A	0.03	−0.01	LPR4 (caudate, nucleus accumbens, putamen); MADD (cerebellum); NR1H3 (caudate, nucleus accumbens)	NR1H3
22	rs7290458	Intergenic	T/C	-0.02	0.01	ZC3H7B (cerebellum); MCHR1 (cerebellum); RP11-12M9.4 (cerebellum); SLC25A17 (nucleus accumbens)	MCHR1, SLC25A17

After multiple testing correction, significant upstream regulators of genes modulated by SNPs associated with SCZ and shorter LTL included 1 drug, 1 microRNA and 5 other genes (Supplementary Table 6). A network of these regulators and the 6 modulated genes is shown in Fig. 1.

Fig. 1. Network of upstream regulators of genes modulated by SNPs associated with increased risk of SCZ and shorter LTL.

The figure shows a network constructed based on significant upstream regulators of genes modulated by SNPs associated with increased risk of SCZ and shorter LTL (shown in blue). Among upstream regulators, drugs are shown in yellow while other regulators in grey.

Significant upstream regulators of genes modulated by SNPs associated with MDD and shorter LTL included 20 drugs, 20 enzymes, 5 transcription regulators, 30 chemical compounds and 37 other genes (Supplementary Table 7). Among upstream regulator drugs, genistein has been suggested to exert anti-inflammatory and antioxidant effects, as well as to be a promising therapeutic compound in different age-related disorders. A subset of the network including this drug, the 2 genes of our list suggested to be modulated by this drug (HYAL3 and CTSF) as well as other significant modulators or transcription regulators of these 2 genes are shown in Fig. 2.

Fig. 2. Network of upstream regulators of genes modulated by SNPs associated with increased risk of MDD and shorter LTL.

The figure shows a network constructed based on the drug genistein, the 2 genes in our dataset (shown in blue) predicted to be modulated by this compound as well as other upstream regulators of these genes. Among upstream regulators, drugs are shown in yellow while genes in grey.

Discussion

In this study, we explored pleiotropy between severe mental disorders and two hallmarks of aging. Using global genetic correlation, we found only MDD to be significantly associated with shorter LTL (r_g = −0.14, p = 6.5E−10). This result is in line with a recent study suggesting shorter LTL in individuals with depression in UK Biobank as well as a significant association between a PRS for depression and shorter LTL [21]. However, using a method able to identify specific genetic loci between two traits, rather than just investigating genetic correlation at a global level, we identified significant cross-trait enrichment between all the investigated severe mental disorders and genetically determined LTL or mtDNA-cn. Interestingly, a high percentage of the identified loci showed lead SNPs with a direction of effect unexpected based on the hypothesis that severe mental disorders are characterized by accelerated cellular aging. Specifically, only 35%, (6/17), 60% (33/55) and 68% (13/19) of lead SNPs were associated with increased risk of BD, SCZ or MDD, respectively, and shorter LTL. BD showed the highest percentage of loci with an unexpected direction of effect. Interestingly, among mental disorders, BD is the one for which the most conflicting results have been reported as regards to shorter LTL compared with individual without mental illness. Indeed, a number of studies has observed longer LTL in patients with BD compared with controls [12, 46]. While treatment with the mood stabilizer lithium might exert a potential counteractive effect on telomere shortening [11, 47], it might also be the case that some genetic variants associated with increased risk of BD might protect patients from accelerated telomere shortening. Among lead SNPs associated with shorter LTL but reduced risk of BD, we found the rs12919664 SNP, located in the TERF2 gene. This gene encodes a telomere-specific protein which is a component of the telomere nucleoprotein complex, plays a key role in the protective activity of telomeres and is a negative regulator of telomere length. The C allele, that we found to be associated with increased risk of BD and longer LTL, reduces levels of the TERF2 gene in the cerebellum and in the cortex (Supplementary Table 1).

Among SNPs with a functional effect associated with a higher risk of BD and cellular aging, the rs11556924 SNP in the ZC3HC1 gene and the rs7909129 SNP in the SORCS3 gene, were predicted to exert a detrimental effect on proteins based on the CADD score. In particular, the rs11556924 in the ZC3HC1 gene is an exonic non-synonymous variant with a CADD score of 30 (variants with scores above 20 are predicted to be among the 1.0% most deleterious substitutions in the human genome) [40, 48]. In our data, the rs11556924 C allele was associated with a reduced risk of BD and reduced LTL. The latter observation is in accordance with results from a recent study linking the in ZC3HC1 rs56179563 SNP (which is in linkage disequilibrium with rs11556924, D’ = 0.95, R² = 0.87) with overall survival. Namely, the rs56179563 A allele, which is correlated with the rs11556924 T allele, was associated with increased lifespan in a mortality risk-factor-informed GWAS [49]. The other SNP with a potential detrimental effect, located in the SORCS3 gene, was associated with shorter LTL and increased predisposition to all three mental disorders (Supplementary Tables 1–3). This gene has been previously implicated in different mental disorders [50–52] as well as in Alzheimer’s disease [52, 53]. The protein encoded by this gene is a member of the vacuolar protein sorting 10 (Vps10) family of receptors, which represent cargos of the retromer complex and are involved in protein trafficking and intracellular/intercellular signalling in neuronal and non-neuronal cells [54, 55]. The retromer is a complex of proteins that control the reverse transport of molecules from the endosomes trans-Golgi network or to the cell surface, thus potentially playing a relevant role in different neurodegenerative diseases. Besides their roles as cargo proteins of the retromer complex, members of the Vps10 receptor family such as SORCS3 have been shown to modulate neurotrophic signaling pathways [55]. Based on this evidence, SORCS3 represents an interesting target that might be involved in the molecular mechanisms underlying accelerated cellular aging in severe mental disorders. In addition, four loci with rs11588837, rs11638445, rs12629701 and rs2345964 as lead SNPs, were associated with longer LTL and predisposition to both BD and SCZ. Three of these variants are significant eQTLs for a variety of genes in different brain regions. Among genes modulated by these variants, rs11588837 can affect the expression of VPS45 (a gene encoding another protein involved in trafficking through the endosomal system [56]) in the frontal cortex.

SCZ was the severe mental disorder for which we identified the highest number of genetic loci shared with LTL (52 loci with 55 lead SNPs) or mtDNA-cn (12 loci with 12 lead SNPs), with around half of these SNPs found to act as significant eQTLs for a variety of genes in at least one brain region (Supplementary Table 2). Genes modulated by SNPs associated with increased predisposition to SCZ and shorter LTL were enriched for GO terms related to Acetylcholine Receptor Activity and S-adenosylmethionine-dependent Methyltransferase Activity. Genes included in this GO term encoded two subunits of the cholinergic receptor (CHRNA3 and CHRNA5) that we found to be modulated by the rs16969894 SNP in brain regions. Specifically, the rs16969894 C allele, that we found to be associated with increased risk of SCZ and reduced LTL, increases expression of CHRNA3 in the caudate and of both genes in the nucleus accumbens. Analyses conducted with IPA showed proteins included in the neuregulin (NRG) family to be significant upstream regulators of both CHRNA3 and CHRNA5 (Supplementary Table 6). Genetic variation at the CHRNA3/5 locus has been previously associated with lifespan [57, 58]. However, it is unclear whether this effect might be entirely or only partially mediated by nicotine dependence, by an increased vulnerability to smoking effects [57], or also by other factors.

MDD was the disorder for which most SNPs showed the expected direction of effect (the allele associated with increased predisposition to MDD was also associated with shorter LTL). This observation is in line with the finding of a significant negative global genetic correlation between MDD and LTL. Using the list of 7 genes modulated by SNPs associated with MDD and shorter LTL as input, we found genistein (4’,5,7-trihydroxyisoflavone) as a common upstream regulator of HYAL3 and NR1H3, two genes for which the MDD risk allele induced lower expression in either cerebellum (HYAL3) or caudate and nucleus accumbens (NR1H3) (Supplementary Table 5). Genistein is a plant-derived phytoestrogen part of the flavonoid family. Due to its chemical structure similar to that of the mammalian estrogens, this compound can modulate endogenous estrogens via binding to the estrogen receptors [59, 60]. Genistein is gaining high interest due to its suggested antiproliferative [61, 62], anti-inflammatory [59] and antioxidant [63] effects in in-vitro models and in preclinical studies, leading to explore its potential pharmacological effects in a range of disorders characterized by a state of chronic inflammation or microinflammation, such as metabolic [64], cardiovascular [65] and neurodegenerative diseases [66]. Indeed, a chronic state of low-grade inflammation and oxidative stress have been hypothesized to play a role in the pathogenesis of severe mental disorders as well as of age-related disorders. A recent study using a system biology approach based on the analysis of metabolomic and gut microbiota data, suggested genistein to be among the most potentially promising novel compounds to treat MDD [67], supporting previous observation of antidepressants effects in rats [68]. While randomized human clinical trials will be needed to establish the potential clinical utility and safety of genistein, it might be speculated that this molecule with anti-inflammatory properties might be especially useful for patients with MDD with high genetic load for accelerated cellular aging.

We conducted a large cross-trait analysis between severe mental disorders and markers of cellular aging using state-of-the-art methods. Some limitations have to be taken into consideration when interpreting our results. Firstly, our analyses were restricted to participants of European origin. It is therefore possible that these results might not be directly transferable to participants with a different origin. Secondly, aging is a multifactorial process driven by genetic and environmental factors, including lifestyle factors that have not been the focus of this study. Future expansions of this work might include taking into account environmental exposure to factors previously shown to be associated with accelerated aging, as well as testing the association between severe mental disorders and epigenetic measures of aging. Finally, while we evaluated the functional effect of the identified genetic variants on gene expression in brain regions, our analyses were restricted to peripheral markers of biological aging. Future expansions of this work might focus on other aging hallmarks with brain relevance, such as brain age. Our study leveraged the largest publicly available GWAS summary statistics for BD, SCZ and MDD to gain insights on the genetic determinants specifically shared between each of these severe mental disorders and markers of aging. However, genetic determinants of mental disorders have been shown to transcend diagnostic boundaries, thus supporting the utility of cross-disorder analyses to enable the discovery of novel variants [69] and dissect molecular mechanisms underlying different disorders. A recent cross-disorder study reported that genes shared between different mental disorders are disproportionately associated with biological pathways related to neurodevelopment and show distinctive gene expression patterns [70]. Since similar differences might also be observed in the association between mental disorders and cellular aging, future expansions of our study will include evaluating the association between cross-disorder genes and markers of aging, in order to discover novel genetic determinants potentially underlying accelerated cellular aging in mental disorders as well as better dissect differences between cross-disorder and disorder-specific mechanisms.

While the present study has leveraged large GWAS summary statistics, future studies to be conducted in datasets with individual genotype data might allow to explore of additional relevant aspects related to cellular aging in severe mental disorders such as the role of gene-gene interactions as well as the aggregated effects of sets of genetic variants. In addition, while we conducted in-silico analyses to identify genetic variants acting as eQTLs, it would be useful to assess whether genes predicted to be modulated by these variants show differences in gene expression, either in peripheral or brain tissues, in patients with severe mental disorders compared with non-psychiatric controls. Finally, the increased availability of cellular models such as neurons and neural precursors cells derived from induced pluripotent stem lines might allow to gain insights into the potential brain relevance of identified targets. Our results support the existence of shared genetic loci between severe mental disorders and markers of biological aging and prioritize functional loci and druggable targets for further investigation. Based on the direction of effect of identified variants, shorter LTL in patients with SCZ or BD could be at least partly counteracted by genetic factors. In addition, our data point to the retromer complex and intracellular protein trafficking as interesting cross-disorder molecular mechanisms potentially underlying the shared genetic bases between severe mental disorders and markers of biological aging.

Author contributions

CP, conceptualization, data curation, data analysis, visualization, writing and review; DC, writing – review & editing; AM, writing – review & editing; PP, visualization, writing – review & editing; GPP, writing – review & editing; GS, writing – review & editing; RA, writing – review & editing; CC, writing – review & editing; MM, visualization, writing – review & editing; AS, conceptualization, writing – review & editing. All authors read and approved the final manuscript.

Funding

None.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41386-024-01822-5.