Epidemiological and Genetic Analyses of Schizophrenia and Breast Cancer

Abstract

Background and Hypothesis

While the phenotypic association between schizophrenia and breast cancer has been observed, the underlying intrinsic link is not adequately understood. We aim to conduct a comprehensive interrogation on both phenotypic and genetic relationships between schizophrenia and breast cancer.

Study Design

We first used data from UK Biobank to evaluate a phenotypic association and performed an updated meta-analysis incorporating existing cohort studies. We then leveraged genomic data to explore the shared genetic architecture through a genome-wide cross-trait design.

Study Results

Incorporating results of our observational analysis, meta-analysis of cohort studies suggested a significantly increased incidence of breast cancer among women with schizophrenia (RR = 1.30, 95% CIs = 1.14–1.48). A positive genomic correlation between schizophrenia and overall breast cancer was observed ($r_g$ = 0.12, P = 1.80 × 10⁻¹⁰), consistent across ER+ ($r_g$ = 0.10, P = 5.74 × 10⁻⁷) and ER– subtypes ($r_g$ = 0.09, P = .003). This was further corroborated by four local signals. Cross-trait meta-analysis identified 23 pleiotropic loci between schizophrenia and breast cancer, including five novel loci. Gene-based analysis revealed 27 shared genes. Mendelian randomization demonstrated a significantly increased risk of overall breast cancer (OR = 1.07, P = 4.81 × 10⁻¹⁰) for genetically predisposed schizophrenia, which remained robust in subgroup analysis (ER+: OR = 1.10, P = 7.26 × 10⁻¹²; ER–: OR = 1.08, P = 3.50 × 10⁻⁶). No mediation effect and reverse causality was found.

Conclusions

Our study demonstrates an intrinsic link underlying schizophrenia and breast cancer, which may inform tailored screening and management of breast cancer in schizophrenia.

INTRODUCTION

Individuals of psychiatric disorders often exhibit an excess comorbidity of subsequent physical diseases. One example lies in schizophrenia and breast cancer.¹ Despite 48% decreased odds of undergoing screening via mammography or breast examination in women with schizophrenia compared to their general population counterpart,² a consistent schizophrenia-breast cancer phenotypic link has been well-established. Aggregating data from 12 cohort studies involving 125 760 women of mixed populations, a meta-analysis demonstrates a 31% significantly heightened incidence of breast cancer among patients with schizophrenia (95% confidence intervals, 95% CIs = 1.14–1.50).³

The phenotypic relationship has been argued to be originated from shared risk factors as several well-known breast cancer determinants are frequently observed in schizophrenia patients, including smoking, drinking, obesity, and reduced physical activity.⁴ It has also been hypothesized that using antipsychotic medications, many of which increase prolactin, a hormone involved in cellular differentiation of mammary gland, plays a mediating role in breast cancer development.⁵ The observational nature of conventional epidemiological studies, however, hinders causal inference as results might be hampered by confounders and reverse causality,⁶ prompting investigations into the intrinsic relationship.

Leveraging summary statistics of genome-wide association studies (GWAS), a significant schizophrenia-breast cancer genetic correlation ($r_g$ = 0.14) has been reported.⁷ One shared genetic locus (GATAD2A) has further been identified⁸ as influencing the risk of both traits. In addition, by adopting Mendelian randomization (MR) analysis, an approach utilizing genetic variants as instrumental variables (IVs) to examine putative causal relationships, three existing studies have consistently identified a significant causal effect of genetically predisposed schizophrenia on breast cancer risk (odds ratios (ORs) ranging from 1.04 to 1.09).^9–11 Nevertheless, several major gaps from these existing studies remained to be narrowed, including a lack of systemic investigation into the genetic relationship and an absence of breast cancer subtypes characterized by hormone receptor status. In addition, previous MRs did not perform sensitivity analysis to guarantee core model assumptions, hampering the robustness of results.¹²

As epidemiological evidence continues to accumulate and sample size of GWAS continues to increase, it is timely to extend previous findings by conducting a comprehensive interrogation on the phenotypic and genetic relationships underlying schizophrenia and breast cancer.⁶ Therefore, in the current study, we first evaluated a phenotypic association using data from the UK Biobank involving more than 230 000 female participants, based on which we then performed an updated meta-analysis of cohort studies to systematically evaluate the incidence of breast cancer in schizophrenia patients. Next, leveraging summary statistics from the hitherto largest GWAS, we performed a large-scale genome-wide cross-trait analysis to quantify overall and local genetic correlations, pleiotropic loci, as well as putative causal relationships between these two complex conditions. The overall study design is shown in figure 1.

Fig. 1.

A schematic framework on the overall design of current study. ER, estrogen receptor.

Methods

We first used data from UK Biobank to estimate the phenotypic association and performed an updated meta-analysis to evaluate the subsequent risk of breast cancer in women with schizophrenia. We then leveraged information on the hitherto largest GWAS of schizophrenia¹³ and breast cancer,^14,15 as well as an array of cutting-edge statistical approaches to quantify shared genetic basis, pleiotropic loci, and causal associations. Specifically, we estimated global genetic correlation applying the cross-trait linkage disequilibrium (LD) score regression (LDSC).¹⁶ We then calculated pairwise local genetic correlation for schizophrenia and breast cancer using SUPERGNOVA.¹⁷ The genetic links were further classified into horizontal pleiotropy and vertical pleiotropy, we performed a cross-phenotypic association (CPASSOC)¹⁸ analysis to test the hypothesis that a SNP affected both traits, performed a gene-based association analysis using Multi-marker Analysis of GenoMic Annotation (MAGMA) v1.10¹⁹ to investigate the shared genes, and performed a bidirectional two-sample MR²⁰ analysis to detect the putative causal relationship. In addition, we conducted tissue enrichment analysis based on the Genotype-Tissue Expression (GTEx) project²¹ and functional mapping and annotation of genome-wide association studies (FUMA)²² to gain biological insight into the genetic components shared by schizophrenia and BC. All participants were of European ancestry. For details please see Methods in Supplementary Materials.

RESULTS

Phenotypic Association Analysis

We included 288 exposed individuals with schizophrenia, together with 1433 birth year-, and recruitment center-matched unexposed individuals, with no differences regarding basic characteristics (Supplementary table 8). With a total of 20 673 accumulated person-years, the mean follow-up time was 11.6 ± 2.5 and 12.7 ± 2.2 years for exposed and matched unexposed individuals, respectively. The Kaplan Meier curves was shown in Supplementary figure 2. After adjusting for birth year and recruitment center, we observed a positive association (albeit nonsignificant) between schizophrenia and subsequent risk of breast cancer (HR = 1.06, 95% CIs = 0.38–2.92). Fully-adjusted model yielded to similar finding (HR = 1.16, 95% CIs = 0.40–3.37) (Supplementary table 9).

The null association between schizophrenia and incidence of breast cancer is perhaps not surprising due to a limited number of individuals in the exposed group (N = 288). Integrating our results with results from 9 existing cohort studies, the updated meta-analysis including over 129 143 European women with schizophrenia derived a significant association (RR = 1.30, 95% CIs = 1.14–1.48), despite a pronounced heterogeneity (P < .001, I² = 86%) (figure 2). Sensitivity analysis omitting one study at a time yielded to similar findings, with RRs ranging from 1.21 to 1.33 (all P < .05).

Fig. 2.

Forest plot of pooled relative risk of incident breast cancer in women with schizophrenia. Square represents the estimate of rate ratio for each study, with the size of square representing the weight of each study with respect to the overall estimate; horizontal line represents the 95% confidence intervals, and diamond represents the overall estimate and its 95% confidence intervals. RR, rate ratio. Tang et al*, our observational analysis. NA, number of schizophrenia cases are not available.

Global and Local Genetic Correlation Analysis

Motivated by a significant phenotypic association, we continued to explore a genetic link. As shown in table 1, a substantial genetic effect was identified for both schizophrenia and breast cancer, with a SNP-heritability of 37% and 13%, respectively. The SNP-heritability was quantified as 15% and 7% for ER+ and ER– subtypes. After adjusting for multiple testing, we found a significant genetic correlation between schizophrenia and overall breast cancer ($r_g$ = 0.12, P = 1.80 × 10⁻¹⁰), which was observed in both ER+ ($r_g$ = 0.10, P = 5.74 × 10⁻⁷) and ER– subtypes ($r_g$ = 0.09, P = .003).

Table 1.

Genome-wide genetic correlation between schizophrenia and breast cancer.

Trait 1	$h_{trait1}^2$	Trait 2	$h_{\mathbf{t}\mathbf{r}\mathbf{a}\mathbf{i}\mathbf{t}2}^2$	$r_g$ (95% CI)	P-value
SCZ	0.37	BC	0.13	0.12 (0.08, 0.16)	1.80 × 10−10
SCZ	0.37	ER+ BC	0.15	0.10 (0.06, 0.14)	5.74 × 10−7
SCZ	0.37	ER– BC	0.07	0.09 (0.03, 0.14)	0.003

SCZ, schizophrenia; BC, breast cancer; $r_g$, genetic correlation; $h^2$, SNP-heritability; ER, estrogen receptor.

Partitioning the whole genome into LD-independent regions, we identified four genomic regions showing significant local genetic correlations, including two regions shared by schizophrenia and overall breast cancer (12p13.33 and 19p13.11), four regions shared by schizophrenia and ER+ breast cancer (8q23.3, 9p13.2, 12p13.33, and 19p13.11), and one region shared by schizophrenia and ER– breast cancer (19p13.11) (Supplementary figure 3). Notably, chr19: 18506815-19873269 at 19p13.11 was identified as a significant region across all three analyses, harboring GATAD2A a previously reported risk gene for both schizophrenia²³ and breast cancer.²⁴ Chr12: 2176265-2886299 at 12p13.33 was identified as shared by schizophrenia with both overall and ER+ breast cancer, harboring a previously reported schizophrenia locus CACNA1C.²⁵ Two regions were identified exclusively for schizophrenia with ER+ breast cancer, including chr8: 116095815-117130004 harboring a previously reported breast cancer locus TRPS1,¹⁴ and chr9: 36714252-37678932 harboring PAX5, implicated in several psychiatric disorders including autism spectrum disorder, bipolar disorder, and major depression.²⁶

Cross-trait Meta-analysis

In pairwise CPASSOC, a total of 23 independent loci reached genome-wide significance (P_CPASSOC < 5 × 10⁻⁸, P_{schizophrenia} < 1 × 10⁻⁵, and P_BC < 1 × 10⁻⁵), including 14 loci shared between schizophrenia and overall breast cancer and 17 loci shared between schizophrenia and ER+ breast cancer (table 2). No significant shared locus was found for schizophrenia with ER– breast cancer.

Table 2.

Significant pleiotropic SNPs identified by cross-trait meta-analysis (P_CPASSOC < 5 × 10⁻⁸ and P_single-trait < 1 × 10⁻⁵, clumping $r^2$ = 0.2).

SNP	CHR	BP	A1	A2	BETA		P SCZ	P BC	P CPASSOC	Gene
					SCZ	BC
Schizophrenia and breast cancer overall
Known shared SNPs (PSCZ < 5 × 10−8 and PBC< 5 × 10−8)
rs7702731	5	44914285	A	G	0.06	0.06	6.97 × 10−10	1.25 × 10−18	3.11 × 10−28	MRPS30, RP11-357F12.1
rs6451798	5	45387854	C	T	−0.07	−0.06	1.40 × 10−09	4.36 × 10−13	2.66 × 10−21	HCN1
rs62061733	17	44018399	A	G	0.06	0.05	1.57 × 10−08	3.50 × 10−10	6.81 × 10−17	MAPT
rs2905432	19	19484295	G	A	0.07	0.04	6.25 × 10−16	2.09 × 10−10	5.16 × 10−22	GATAD2A, MAU2
Single-trait driven shared SNPs (PSCZ < 5 × 10−8or PBC< 5 × 10−8)
rs6805189	3	71532113	T	C	0.04	0.04	7.21 × 10−07	2.68 × 10−10	9.07 × 10−16	FOXP1
rs13198474	6	25874423	G	A	0.17	0.06	1.82 × 10−22	7.94 × 10−07	9.54 × 10−24	SLC17A3
rs13195402	6	26463575	G	T	0.21	0.06	5.74 × 10−36	5.80 × 10−08	9.93 × 10−39	BTN2A1
rs13195636	6	27509493	A	C	0.21	0.05	6.55 × 10−40	3.52 × 10−06	4.09 × 10−43	HNRNPA1P1, ZNF184
rs67981811	6	28354835	C	G	0.21	0.06	1.86 × 10−38	4.17 × 10−07	1.69 × 10−41	ZSCAN12
rs77642095	12	14408698	C	T	0.05	−0.06	2.07 × 10−06	4.90 × 10−15	6.84 × 10−21	GNAI2P1, RN7SL676P
rs4702	15	91426560	G	A	0.08	0.03	2.79 × 10−21	1.01 × 10−06	1.97 × 10−22	FURIN
Novel shared SNPs (5×10 −8 < P SCZ/BC < 1 × 10−5and not in LD with established index SNPs (r2 < 0.2)
rs2957468	8	106325360	A	G	0.04	−0.04	6.02 × 10−06	6.68 × 10−08	3.79 × 10−12	ZFPM2
rs3809114	12	57848639	G	A	−0.04	−0.03	2.99 × 10−06	2.98 × 10−06	2.61 × 10−10	INHBE
rs199498	17	44865603	T	C	0.06	0.04	1.24 × 10−07	2.56 × 10−07	1.31 × 10−12	WNT3
Schizophrenia and ER+ breast cancer
Known shared SNPs (PSCZ < 5 × 10−8and PBC < 5 × 10−8)
rs10035564	5	45252500	A	G	−0.07	−0.06	4.38 × 10−13	1.64 × 10−14	1.08 × 10−27	RP11-357F12.1, HCN1
rs67981811	6	28354835	C	G	0.21	0.08	1.86 × 10−38	4.40 × 10−08	4.83 × 10−42	ZSCAN12
Single-trait driven shared SNPs (PSCZ < 5 × 10−8or PBC< 5 × 10−8)
rs6805189	3	71532113	T	C	0.04	0.04	7.21 × 10−07	3.76 × 10−08	6.05 × 10−14	FOXP1
rs7447717	5	45770252	A	G	0.05	0.04	5.95 × 10−10	2.52 × 10−07	8.30 × 10−16	HCN1, CTD-2013M15.1
rs13198474	6	25874423	G	A	0.17	0.08	1.82 × 10−22	2.68 × 10−07	1.56 × 10−25	SLC17A3
rs13195402	6	26463575	G	T	0.21	0.08	5.74 × 10−36	1.58 × 10−07	3.17 × 10−39	BTN2A1
rs13195636	6	27509493	A	C	0.21	0.08	6.55 × 10−40	6.91 × 10−08	1.09 × 10−43	HNRNPA1P1, ZNF184
rs9257566	6	29144532	C	T	0.19	0.06	1.27 × 10−36	9.22 × 10−06	5.75 × 10−40	OR2J2, OR2J4P
rs2957468	8	106325360	A	G	0.04	−0.05	6.02 × 10−06	1.85 × 10−08	2.47 × 10−13	ZFPM2
rs75905550	12	14408554	G	C	0.05	−0.05	2.15 × 10−06	1.49 × 10−08	6.49 × 10−14	GNAI2P1, RN7SL676P
rs4702	15	91426560	G	A	0.08	0.03	2.79 × 10−21	3.97 × 10−06	1.10 × 10−22	FURIN
rs1724390	17	43663247	C	A	0.06	0.05	2.00 × 10−08	2.56 × 10−07	1.95 × 10−14
rs2532363	17	44333423	G	T	0.07	0.05	2.52 × 10−09	3.33 × 10−07	4.23 × 10−15	RP11-259G18.2, RP11-259G18.3
rs2965183	19	19545696	G	A	−0.07	−0.04	3.78 × 10−16	1.06 × 10−07	1.96 × 10−21	GATAD2A
Novel shared SNPs (5 × 10−8< PSCZ/BC< 1 × 10−5and not in LD with established index SNPs (r2< 0.2)
rs4865721	5	49441601	G	T	0.06	0.04	5.50 × 10−07	9.64 × 10−06	3.64 × 10−11	EMB, CTD-2013M1
rs10156333	8	117055269	T	C	0.04	−0.03	9.27 × 10−06	6.72 × 10−06	3.12 × 10−10	LINC00536
rs199498	17	44865603	T	C	0.06	0.04	1.24 × 10−07	2.18 × 10−06	1.49 × 10−12	WNT3

SNP, single nucleotide polymorphisms; CHR, chromosome; BP, physical position of SNP, build GRCh37; A1, effect allele; A2, alternative allele; BETA, effect allele beta coefficient; SCZ, schizophrenia; BC, breast cancer; P_SCZ, P-value for schizophrenia; P_BC, P-value for breast cancer; P_CPASSOC, P-value for cross-phenotype association; Gene, linear closest genes of index SNPs, mapped using SNPnexus; ER, estrogen receptor.

Among the 14 pleiotropic SNPs affecting both schizophrenia and overall breast cancer, four were “known” pleiotropic SNPs (significantly associated with both traits in previous GWAS), seven were “single-trait-driven” pleiotropic SNPs (significantly associated with one of the two traits in previous GWAS). Here we highlighted the three novel shared SNPs. The most significant novel shared SNP was rs199498 (P_CPASSOC = 1.31 × 10⁻¹²) located at WNT3, a gene associated with breast cancer metastasis.²⁷ The second most significant novel shared SNP was rs2957468 (P_CPASSOC = 3.79 × 10⁻¹²) located at ZFPM2, a gene associated with antipsychotic-induced parkinsonism.²⁸ The third most significant novel shared SNP was rs3809114 (P_CPASSOC = 2.61 × 10⁻¹⁰) mapped to INHBE, a gene involved in TGF-β signaling which was associated with schizophrenia.²⁹

For schizophrenia and ER+ breast cancer, apart from the eight pleiotropic SNPs also identified for schizophrenia and overall breast cancer, nine additional shared SNPs were identified, totaling 17 pleiotropic SNPs (table 2). Among these SNPs, two were “known” pleiotropic SNPs and 12 were “single-trait-driven” pleiotropic SNPs. Three novel shared SNPs were rs199498 (P_CPASSOC = 1.49 × 10⁻¹²) which was also identified as novel for schizophrenia and overall breast cancer, rs4865721 (P_CPASSOC = 3.64 × 10⁻¹¹) located upstream of EMB, which was associated with schizophrenia in mixed population,³⁰ and rs10156333 (P_CPASSOC = 3.12 × 10⁻¹⁰) located close to a long intergenic nonprotein coding RNA LINC00536.

Gene-based Analysis

Through MAGMA, an independent set of significant associated genes (P < 2.57 × 10⁻⁶) were first identified for each trait, including 520 genes for schizophrenia, 286 genes for overall breast cancer, 184 genes for ER+ breast cancer, and 49 genes for ER– breast cancer (Supplementary tables 10-13). Among these identified genes, there were 24 overlapped genes between schizophrenia and overall breast cancer, 16 genes between schizophrenia and ER+ breast cancer, and one gene between schizophrenia and ER– breast cancer.

For the 24 genes shared between schizophrenia and overall breast cancer, 13 genes (THOC7, ATXN7, PSMD6, FOXP1, MRPS30, HCN1, ZNF365, TM6SF2, SUGP1, MAU2, GATAD2A, TSSK6, and YJEFN3) were previously reported to associate with both traits by GWAS,^{14,15,31–37} nine genes (PDE4D, CRHR1, SPPL2C, MAPT, STH, KANSL1, LRRC37A, ARL17B, and NSF, eight out of which located at 17q21.31) only with breast cancer,¹⁵ and one gene (OR2B2) only with schizophrenia.³⁸ The remaining gene WNT3, located at 17q21.31; however, has not been found significant in either schizophrenia or breast cancer GWAS (Supplementary table 14).

For the 16 genes shared between schizophrenia and ER+ breast cancer, apart from the 13 genes also identified for schizophrenia and overall breast cancer, three uniquely shared genes (HIST1H2BN, HIST1H1B, and HIST1H4L) are histone genes clustered on chromosome 6p22.1, from which a LD region was observed to associate with schizophrenia.³⁹

For schizophrenia and ER– breast cancer, one shared gene ZNF365 was observed, which was also identified as shared by schizophrenia with both overall and ER+ breast cancer.

Tissue Enrichment Analysis

We sought to understand the underlying biological mechanisms of the identified genes by performing tissue-enrichment analyses using data from GTEx. Most genes identified by CPASSOC or MAGMA expressed significantly in brain, breast, and whole blood sample. For schizophrenia and overall breast cancer, brain-hypothalamus was identified as significantly enriched for the expression of CPASSOC-identified shared genes, withstanding multiple correction (Supplementary figure 4, Supplementary tables 15-16). Similar enrichments in brain-related tissues (e.g., brain-hippocampus, brain-cerebellar hemisphere, brain-cortex, brain-amygdala, brain-putamen, and brain-hypothalamus) were observed for MAGMA-identified shared genes although failed to pass multiple correction (Supplementary figure 5, Supplementary tables 17-18).

Mendelian Randomization

We finally conducted a bidirectional two-sample MR to test causal relationship (figure 3). Using IVW, genetically predisposed schizophrenia was significantly associated with an increased risk of overall breast cancer (OR = 1.07, P = 4.81 × 10⁻¹⁰), which remained consistent in MR-Egger regression (OR = 1.24, P = 1.37 × 10⁻⁸), weighted median approach (OR = 1.06, P = 3.81 × 10⁻⁶), MR-PRESSO (OR = 1.07, P = 3.17 × 10⁻⁹), and MR-CAUSE (OR = 1.03, P = 1.18 × 10⁻⁴). The estimates remained directional consistent in weighted mode approach, despite larger statistical uncertainties. Similar associations were generated after removing palindromic SNPs (OR = 1.07, P = 4.73 × 10⁻⁸) or pleiotropic SNPs (OR = 1.08, P = 6.92 × 10⁻⁹). Subgroup analysis further identified such an association in both ER+ (IVW OR = 1.10, P = 7.26 × 10⁻¹²) and ER– subtypes (IVW OR = 1.08, P = 3.50 × 10⁻⁶). Multivariable MR taking into consideration potential confounders yielded similar results with even more pronounced magnitude and significance, implying that the causality of schizophrenia on breast cancer and subgroup was not influenced by common confounders (Supplementary figure 6). The two-step MR analysis suggested that cigarette smoking was not a mediator in the schizophrenia-breast cancer causal pathway, with a nonsignificant proportion mediated (4.0%, 95%CI = −6.6% to 7.4%).

Fig. 3.

Bidirectional causal relationship underlying schizophrenia and breast cancer. Panel A (schizophrenia on overall breast cancer), Panel B (schizophrenia on ER+ breast cancer), and Panel C (schizophrenia on ER– breast cancer) show the estimates of causal associations. Panel D shows the estimates of reverse-direction causal associations (overall, ER+, and ER– breast cancer on schizophrenia). Square represents the point estimates of causal effect, and horizontal line represents the 95% confidence intervals. Inverse variance weighted approach was used as the primary analysis, MR-Egger, weighted median, weighted mode, MR-PRESSO, MR-CAUSE, excluding pleiotropic SNPs, and excluding palindromic SNPs were used as sensitivity analysis. ER, estrogen receptor.

No evidence of reverse causality was found according to the Steiger directionality test (schizophrenia on breast cancer: P = 3.37 × 10⁻⁷⁶; schizophrenia on ER+ breast cancer: P = 9.49 × 10⁻⁶⁶; schizophrenia on ER- breast cancer: P = 3.72 × 10⁻⁵⁶), and we performed reverse-direction MR analysis confirmed this conclusion (figure 3). The mean F-statistics for all exposures were larger than 10 (ranging from 49.45 to 98.06), indicating strong instruments (Supplementary table 19).

DISCUSSION

As far as we understand, this is the most comprehensive phenotypic and genetic analyses that systematically investigated the phenotypic association, genetic correlation, pleiotropic loci, and causal relationship underlying schizophrenia and breast cancer. Our phenotypic association analysis demonstrated a positive association of schizophrenia with breast cancer risk, which was corroborated by findings of genetic study. We found evidence supporting a significant shared genetic basis, both globally and regionally, suggesting an intrinsic link underlying these two traits. This genetic overlap was further classified into horizontal pleiotropy and vertical pleiotropy, reflected by the novel pleiotropic loci identified in CPASSOC, the shared genes revealed by MAGMA, and the putative causal relationship demonstrated by MR. Extending to breast cancer subtypes, similar patterns of results were found for ER+ subtype. These findings advance our understandings to the complex link underlying a mental health condition and a physical malignancy, providing important implications in the prevention for breast cancer in woman with schizophrenia.

The positive genome-wide genetic correlation between schizophrenia and breast cancer, estimated as around one-eighth of the shared genetic contribution by the two traits, was largely in line with previously reported estimate of 0.14.^7,8 Such a magnitude of effect, albeit moderate, is noteworthy in comparison with corresponding estimates evaluated for schizophrenia with other cancer types, for example, colorectal ($r_g$ = 0.01, P = .75), lung ($r_g$ = 0.07, P = .06), and ovarian ($r_g$ = 0.11, P = .06).⁷ The overall shared genetic basis was further corroborated by local signals identified at genomic regions 8q23.3, 9p13.2, 12p13.33, and 19p13.11. While CACNA1C (12p13.33), PAX5 (9p13.2), and GATAD2A (19p13.11) are directly implicated in multiple psychiatric disorders^23,26,40 as well as in the initiation or progression of breast cancer,^24,41,42TRPS1 (8q23.3) modulates SHBG,⁴³ a well-established risk factor for both schizophrenia and breast cancer. These findings collectively highlight a nonnegligible etiological link underlying schizophrenia and breast cancer which deserves further investigation.

The 23 shared loci identified by cross-trait meta-analysis are largely validated and supported by existing literatures. For example, the shared SNP rs2905432 (19p13.11) identified by us was in strong LD with rs2965183 ($r^2$ = .93) and rs2905426 ($r^2$ = 1.00), two loci priorly reported as shared by schizophrenia and breast cancer.⁸ Multiple shared loci were mapped to genes implicated in smoking,⁴⁴ educational attainment,⁴⁵ BMI (e.g., MRPS30, HCN1, FOXP1, and SLC17A3), or psychiatric disorders (e.g., BTN2A1, ZNF184, ZSCAN12, FURIN, ZSCAN12, BTN2A1, and OR2J2), all were relevant to or risk factors of schizophrenia and breast cancer.

An advantage of cross-trait meta-analysis is that evidence of association in combination with multiple studies can reveal signals which might not have reached genome-wide significance in a single-trait analysis. We highlight five genes (ZFPM2, EMB, INHBE, LINC00536, and WNT3) mapped to novel shared loci in CPASSOC. Both ZFPM2 and EMB show strong evidence as influencing important risk factors of breast cancer or schizophrenia including BMI and high-density lipoprotein cholesterol. ZFPM2 has been found to highly express in human cerebellum tissue by GTEx, FANTOM5, and HPA dataset, the dysfunction of which is implicated in several psychiatric disorders including schizophrenia.⁴⁶INHBE is involved in autism spectrum disorder, major depressive disorder and schizophrenia.³⁴ In addition, small somatic deletions of INHBE lead to INHBE-GLI1 fusions and activate the sonic hedgehog pathway, which plays an important role in benign hepatocyte proliferation.⁴⁷LINC00536 plays an essential role in promoting tumor progression, and a prior study found that LINC00536 was upregulated in breast cancer patients.⁴⁸WNT3 influences smoking behavior and levels of SHBG,⁴⁹ both of which are risk factors for breast cancer and schizophrenia. It has been widely discussed that the overexpression of WNT3 activates wnt/β-catenin signaling pathway implicated in neurodevelopmental processes and/or psychiatric disorders.⁵⁰ Moreover, wnt/β-catenin signaling pathway suppresses the antitumor immune response and participates in the metastatic process of breast cancer.⁵¹

With the identification of 27 shared genes, our MAGMA analysis further supports the existence of pleiotropy at the level of genes. Specifically, five genes (FOXP1, GATAD2A, HCN1, MAPT, MRPS30, and WNT3) were identified by both CPASSOC and MAGMA showing a direct or indirect association with breast cancer and/or schizophrenia. In addition, most of the shared genes correspond well with established knowledge. For example, six shared genes at 19p13.11 (TM6SF2, SUGP1, MAU2, GATAD2A, TSSK6, and YJEFN3) were reported in association with breast cancer and schizophrenia, and 19p13.11 was identified by both our local genetic correlation and CPASSOC. More importantly, this locus was also identified as shared by three hormone-related cancers including breast, ovarian, and prostate ($P$ = 3.4 × 10⁻¹⁰).²⁴

Despite limited by the number of input genes, hypothalamus in brain appeared as a significantly enriched tissue for the expression of CPASSOC-identified shared genes. Genetic mutations associated with schizophrenia are found to display corresponding changes in the activity of a specific subregion of the hypothalamus.⁵² Moreover, estrogen activity plays a crucial role in the development and progression of breast cancer, which in premenopausal women is regulated mainly through the hypothalamic-pituitary-gonadal axis.⁵³

Utilizing a comprehensive two-sample MR design, our results suggest that genetic predisposition to schizophrenia is significantly associated with an increased risk of breast cancer which is consistent with our meta-analysis. While concordant with results from existing MRs,^9–11 our study greatly extended previous findings in several critical aspects. We applied an enlarged number of genetic instruments (186 IVs vs. 142 IVs) which were derived from the largest GWAS of schizophrenia, increasing the statistical power to detect causal associations. The 186 IVs explained altogether 6.6% of the phenotypic variance, with an F-statistic estimated at 49.45, suggesting a minimal weak instrument bias. Furthermore, the sensitivity analyses conducted in tandem to the main analyses served to test the assumptions of MR and provide evidence for the robustness of primary findings. For example, we excluded SNPs associated with potential confounders of the exposure-outcome relationship to comply with the “exclusive restriction” assumption. Our results, in the context of other epidemiological and genetic findings, confirm that schizophrenia is a causal risk factor for breast cancer, which provide support for a personalized breast cancer screening to improve further prevention in vulnerable individuals. In addition, leveraging a reverse directional MR, we did not identify a causal association between genetically predicted breast cancer with schizophrenia risk. While a previous cohort study assessed the risk of incident schizophrenia following breast cancer by cross-linkage of nationwide patient registers, the effect size was small in inpatient diagnosis of schizophrenia (HR = 1.29, P = .003) which attenuated to nonsignificant when including outpatient diagnosis of schizophrenia (HR = 1.13, P = .218).⁸ Future large-scale longitudinal studies are needed to establish or rule out such a relationship.

We acknowledge several limitations. First, our findings were restricted to European population. As schizophrenia and breast cancer incidence show significant racial differences,^54,55 future studies are expected to extend to other ancestry groups to fully disclose the biological mechanisms underlying these two traits. Second, we did not investigate additional breast cancer subtypes based on other hormonal receptor expression due to limited sample size, for instance, human epidermal growth factor receptor 2 breast cancer cases (N = 2,884) and triple-negative breast cancer cases (N = 8602) analyzed by Zhang et al.¹⁴ Larger and more powerful subtype-specific GWAS are needed to extend our findings. Third, we assessed tissue enrichments based on 54 tissue types available in GTEx, which may restrict the understanding of gene regulation mechanisms due to limited tissue availability.

In conclusion, the current study advances our understanding of the relationship between schizophrenia and breast cancer through an observational analysis, a meta-analysis of all available cohort studies and genetic analysis using data from the largest GWAS conducted in European ancestry. Findings from this study demonstrates the genetic etiology underlying the phenotypic link of schizophrenia and breast cancer and deliver important public health message on that the management of schizophrenia in women would potentially help mitigate the long-term burden of malignant diseases.

Funding

This work was supported by the National Key R&D Program of China (2022YFC3600600, 2022YFC3600604), the National Natural Science Foundation of China (U22A20359, 81874283, 81673255, 82204170), the Recruitment Program for Young Professionals of China, the Promotion Plan for Basic Medical Sciences and the Development Plan for Cutting-Edge Disciplines, Sichuan University, and other Projects from West China School of Public Health and West China Fourth Hospital, Sichuan University.

Conflict of interest statement

The authors have declared that there are no conflicts of interest in relation to the subject of this study.