Pleiotropic Association of CACNA1C Variants With Neuropsychiatric Disorders

Abstract

Background

Neuropsychiatric disorders are highly heritable and have overlapping genetic underpinnings. Single nucleotide polymorphisms (SNPs) in the gene CACNA1C have been associated with several neuropsychiatric disorders, across multiple genome-wide association studies.

Method

A total of 70,711 subjects from 37 independent cohorts with 13 different neuropsychiatric disorders were meta-analyzed to identify overlap of disorder-associated SNPs within CACNA1C. The differential expression of CACNA1C mRNA in five independent postmortem brain cohorts was examined. Finally, the associations of disease-sharing risk alleles with total intracranial volume (ICV), gray matter volumes (GMVs) of subcortical structures, cortical surface area (SA), and average cortical thickness (TH) were tested.

Results

Eighteen SNPs within CACNA1C were nominally associated with more than one neuropsychiatric disorder (P < .05); the associations shared among schizophrenia, bipolar disorder, and alcohol use disorder survived false discovery rate correction (five SNPs with P < 7.3 × 10⁻⁴ and q < 0.05). CACNA1C mRNA was differentially expressed in brains from individuals with schizophrenia, bipolar disorder, and Parkinson’s disease, relative to controls (three SNPs with P < .01). Risk alleles shared by schizophrenia, bipolar disorder, substance dependence, and Parkinson’s disease were significantly associated with ICV, GMVs, SA, or TH (one SNP with P ≤ 7.1 × 10⁻³ and q < 0.05).

Conclusion

Integrating multiple levels of analyses, we identified CACNA1C variants associated with multiple psychiatric disorders, and schizophrenia and bipolar disorder were most strongly implicated. CACNA1C variants may contribute to shared risk and pathophysiology in these conditions.

Introduction

Genome-wide association studies (GWAS) provide insight into common genetic variation contributing to the risk of complex disorders. GWAS has proven successful in recent years in identifying single nucleotide polymorphisms (SNPs) underlying a range of neuropsychiatric diseases.¹ Most neuropsychiatric disorders are polygenic; each associated allele contributes a small amount to overall risk. Some of this risk is shared between diagnoses.^2,3 The regulatory effects of these shared risk alleles may underlie common symptoms, comorbidity, overlapping response to therapeutics, or other features shared among diagnoses.

Here we focus on the gene CACNA1C (12p13.3), which has emerged as a risk locus in GWAS studies of multiple neuropsychiatric conditions. CACNA1C encodes the α-1C subunit of the voltage-gated L-type calcium channel (LTCC). Variants in CACNA1C have been associated with bipolar disorder and schizophrenia,^4,5 major depressive disorder,⁶ autism spectrum disorder (ASD) and attention deficit hyperactivity disorder (ADHD),^1,7 Parkinson’s disease,⁸ substance use disorders,⁹ and stroke.¹⁰ Additionally, CACNA1C variants have been associated with some subclinical and transdiagnostic phenotypes, such as schizotypy.^11,12 Several specific risk variants are shared among these disorders. For example, rs1006737, the best-known functional variant in CACNA1C,⁴ is a risk variant for schizophrenia, bipolar disorder, major depressive disorder, and ASD.^13,14 rs4765905 is shared by ASD and schizophrenia.¹⁴ rs7297582 is shared by bipolar disorder and major depressive disorder.⁶ The Psychiatric Genetics Cross Disorder Consortium has also reported a shared SNP, rs1024582, across schizophrenia, bipolar disorder, major depressive disorder, ASD, and ADHD.¹

Most previous studies examining CACNA1C have investigated associations with a single neuropsychiatric disorder.¹⁴ A few have tested associations with two neuropsychiatric disorders^6,15,16; two have looked across more than two conditions.^13,17 Here we comprehensively examined the association of all known CACNA1C SNPs with 13 different neuropsychiatric disorders from 37 independent cohorts, comprising 70,711 subjects, to identify shared risk alleles and their influence on CACNA1C expression in the brain and on measures of brain structure in over 40,000 structural brain scans from the ENIGMA consortium. Such a comprehensive analysis, using harmonized methods across conditions, can shed new light on shared genetic contributors to neuropsychiatric illnesses, including revealing the core CACNA1C-relevant conditions.

Materials and Methods

Subjects

A total of 70,711 subjects from 37 independent cohorts with 13 different neuropsychiatric disorders were analyzed: schizophrenia (6 cohorts), bipolar disorder (3), major depressive disorder (1), ASD (1), alcohol dependence (4), nicotine dependence (8), ADHD (2), Alzheimer’s disease (2), Parkinson’s disease (3), multiple sclerosis (3), amyotrophic lateral sclerosis (1), stroke (2), and ischemic stroke (1). The studies included case-control, case-only, and family-based designs. Study procedures were reviewed and approved by the Human Investigation Committee of the participating institutions. All participants, or their parents in the case of minor children, signed informed consent prior to participation. The dataset numbers, dataset names, phenotypes, diagnosis criteria, ethnicities, study designs, sample sizes, genotyping platforms, grant numbers, principal investigators’ names, associated publications, and dbGaP accession numbers for these cohorts are shown in supplementary table S1. Detailed demographics of these cohorts have been published before (see references in supplementary table S1).

Genotyping and Imputation

The cohorts included in this analysis were genotyped using a number of different microarray panels (detailed in supplementary table S1). To derive consistent genetic marker sets across all cohorts, we imputed untyped SNPs across the CACNA1C region. The CACNA1C region examined extends from Chr12:1,898,079 (5ʹ) to Chr12:2,774,397 (3ʹ) (Genome Build 36), including a 5ʹ flanking region that extends to the boundary of the neighboring gene CACNA2D4 and a 3ʹ flanking region that extends to the boundary of FKBP4.

To improve the accuracy and success rate of imputation, we employed the 1000 Genome Project and HapMap 3 panels as references, separating the CEU and YRI ethnicities during imputation. We used IMPUTE2,¹⁸ which employs an MCMC algorithm to derive full posterior probabilities for genotypes of each SNP, rather than the “best-guess.” The imputation parameters were set at burnin = 10 000, iteration = 10 000, k = 100, Ne = 11 500, and confidence level = 0.99. To increase sample sizes and marker density for imputation, we merged datasets within the same ethnicity. Imputed data were cleaned stringently, including the removal of SNPs with minor allele frequency (MAF) < 5%, imputation information score (INFO) < 0.8, and imputed allelic R² < 0.5. Families with Mendelian errors and SNPs with errors were excluded, and for SNPs that were directly genotyped, we used direct genotypes rather than imputed data.^19,20

Study Design and Analytic Strategies

figure 1 illustrates the relationships examined in this work. Statistical associations between diagnoses and SNPs in the CACNA1C region were tested first, using association analysis within each cohort followed by meta-analysis for each disorder and then analysis of shared risk across disorders. Next, we examined the relationships of the identified SNPs with potential biological effects, including (1) potential regulatory effects on RNA (ncRNA and mRNA) expression by bioinformatic analysis, and (2) potential regulatory effects on brain structural measurements (ICV/GMV/SA/TH) by quantitative trait locus (QTL) analysis, which could validate that the SNP-disease associations were biological but not only statistical ones. Finally, differential mRNA expression analysis in brains between diseases and controls was performed, which could also help to support a biological SNP-disease association.

Fig. 1.

Regulatory pathway from SNPs to diseases. (Solid lines, pair-wise associations examined in the present study; dash lines, potential pair-wise associations predicted by literatures or bioinformatic analyses; arrows, regulatory effect directions; ICV, intracranial volume; GMV, subcortical gray matter volume; SA, cortical surface area; TH, cortical thickness; ncRNA, noncoding RNAs including LncRNAs, snRNAs or miRNAs).

In most analyses we corrected for multiple comparisons using the false discovery rate (FDR): the SNP-disease association analysis in each cohort (supplementary table S1); the meta-analysis for each disorder (supplementary table S2); the analysis of shared risk across disorders (table 1 and supplementary table S3) ; the analysis of potential regulatory effects of pleiotropic risk variants in each cohort (supplementary table S4); and the analysis of potential regulatory effects replicable between cohorts ­(tables 2 and 3). Q-values for these analyses that adjusted P-values based on an optimized FDR approach were computed using the R package QVALUE. A q-value < 0.05 indicates statistical significance.

Table 1.

Summary of cross-disorder risk SNPs within CACNA1C

SNP	Neuropsychiatric disorders	Ethnicity	Meta dataset#(Table S1)	Risk allele (frequency in cases)	Z score	OR (95% CI)	I 2	P value	Q value
rs7297582	Schizophrenia	EA	29,30,32,33	C: 0.402	2.03	1.08 (1.00, 1.17)	0.31	.040	>0.05
	Alcoholism	AA	1,3	C: 0.096	3.33	1.41 (1.15, 1.73)	0	1.3 × 10 −5	0.012
rs2238072	Schizophrenia	EA	29,30,32,33	T: 0.259	2.00	1.09 (1.00, 1.19)	0.05	.040	>0.05
	Bipolar disorder	EA	11,12	T: 0.075	2.22	1.48 (1.05, 2.09)	0.38	.030	>0.05
rs17801265	Schizophrenia	EA	29,30,32,33	A: 0.174	2.07	1.09 (1.00, 1.18)	0	.040	>0.05
	Alcoholism	EA	2,4	A: 0.139	2.39	1.20 (1.03, 1.38)	0	.020	>0.05
rs216016	Schizophrenia	AA	31,34	C: 0.103	4.60	1.52 (1.27, 1.82)	0	1.8 × 10 −6	1.6 × 10 −3
	Autism	EA	10	C: 0.075	—	1.14	0	.017	>0.05
rs216017	Schizophrenia	AA	31,34	G: 0.114	3.96	1.42 (1.19, 1.69)	0	6.4 × 10 −5	0.029
	Autism	EA	10	G: 0.075	—	1.14	0	.014	>0.05
rs2239079	Schizophrenia	AA	31,34	A: 0.150	2.63	1.24 (1.06, 1.45)	0	.009	>0.05
	Bipolar disorder	EA	11,12	A: 0.345	2.87	1.18 (1.05, 1.32)	0	.004	>0.05
rs12305678	Schizophrenia	AA	31,34	G: 0.314	1.92	1.13 (1.00, 1.29)	0	.049	>0.05
	Alcoholism	EA	2,4	G: 0.054	2.25	1.32 (1.04, 1.68)	0.37	.020	>0.05
rs115773686	Schizophrenia	AA	31,34	G: 0.143	2.63	1.63 (1.18, 5.86)	0.20	.020	>0.05
	Bipolar disorder	AA	13	G: 0.064	5.53	4.04 (2.46, 6.63)	0	.030	>0.05
rs11062086	Nicotine dependence	EA	19,21,23,25	T: 0.151	2.17	—	0	.030	>0.05
	Bipolar disorder	AA	13	T: 0.284	2.98	1.53 (1.16, 2.03)	0	.003	>0.05
rs723672	Nicotine dependence	EA	19,21,23,25	C: 0.439	2.17	—	0	.030	>0.05
	Bipolar disorder	AA	13	C: 0.362	4.31	1.82 (1.39, 2.40)	0	.025	>0.05
rs4765687	Nicotine dependence	EA	19,21,23,25	G: 0.463	1.97	—	0	.049	>0.05
	Bipolar disorder	EA	11,12	G: 0.430	3.23	1.19 (1.07, 1.33)	0	.001	>0.05
rs11062277	Nicotine dependence	EA	19,21,23,25	G: 0.170	1.97	—	0	.048	>0.05
	Bipolar disorder	EA	11,12	G: 0.097	3.55	1.56 (1.22, 2.00)	0	.004	>0.05
rs10848680	Nicotine dependence	EA	19,21,23,25	G: 0.273	2.00	—	0	.047	>0.05
	Bipolar disorder	EA	11,12	G: 0.115	2.64	1.53 (1.12, 2.10)	0.01	.008	>0.05
rs2239131	Nicotine dependence	AA	18,20,22,24	G: 0.386	2.09	—	0	.037	>0.05
	Major depressive disorder	EA	14	G: 0.277	2.35	1.13 (1.02, 1.26)	0	.020	>0.05
rs2370413	Autism	EA	10	C: 0.186	—	1.37	0	.002	>0.05
	Major depressive disorder	EA	14	C: 0.470	2.92	1.15 (1.05, 1.26)	0	.003	>0.05
rs11062297	Autism	EA	10	C: 0.070	—	1.22	0	5.5 × 10 −4	0.059
	Bipolar disorder	EA	11,12	C: 0.050	2.64	1.53 (1.12, 2.10)	0.20	.008	>0.05
rs11062091	Bipolar disorder	AA	13	G: 0.075	3.99	2.58 (1.62, 4.12)	0	1.8 × 10 −5	0.005
	Parkinson’s disease	EA	26,27,28	G: 0.149	2.5	1.17 (1.07, 1.29)	0	.010	>0.05
rs2299657	Bipolar disorder	AA	13	G: 0.113	4.27	2.34 (1.58, 3.45)	0	5.7 × 10 −6	0.005
	Parkinson’s disease	EA	26,27,28	G: 0.207	2.5	1.12 (1.04, 1.21)	0	.002	>0.05

The dataset# corresponds to Table S1. All SNPs were independent of one another except for rs216016 and rs216017. EA, European; AA, African–American; CI, confidence interval; OR, odds ratio; Z score, from meta-analysis. I² heterogeneity of meta-analysis; Bold P values are < Bonferroni-corrected α (=7.25 × 10⁻⁴ in EA and 4.55 × 10⁻⁴ in AA). Bold q values are <0.05.

Table 2.

P-values/q-values (effective alleles) for the replicable regulatory effects of disease-risk SNPs on ICV and different subcortical GMVs

SNP	Position	Disease-risk allele	Sharing diseases	Target region	CHARGE-ENIGMA	ENIGMA2
rs11062091	2032630	G	Bipolar disorder_AA/Parkinson’s disease_EA	Hippocampus	0.036 (G)	0.050 (G)
rs2238072	2326968	T	Schizophrenia_EA/Bipolar disorder_EA	ICV	7.7 × 10 −4 /0.014 (C)	2.6 × 10 −4 /1.5 × 10−3 (C)
rs2238072	2326968	T	Schizophrenia_EA/Bipolar disorder_EA	Pallidum	0.044 (C)	0.034 (C)

ICV, Total Intracranial Volume. Bold P values/q values correspond to q < 0.05.

Table 3.

P-values/q values (effective alleles) for the replicable regulatory effects of disease-risk SNPs on different cortical surface area (SA) and thickness (TH)

					SA	TH	TH
SNP	Position	Risk allele	Sharing phenotypes	Target region	ENIGMA3	ENIGMA3	UKBB
rs2299657	2036217	G	Bipolar disorder_AA/Parkinson’s disease_EA	Temporal pole		0.048 (T)↓	0.024 (T)↓
rs2239131	2099320	G	Nicotine dependence_AA/Major depressive disorder_EA	Lateral orbitofrontal	0.033 (A)↓	0.036 (G)↑	0.044 (G)↑
rs2238072	2326968	T	Schizophrenia_EA/Bipolar disorder_EA	Frontal pole	0.033 (C)↓	0.018 (T)↑
rs2238072	2326968	T	Schizophrenia_EA/Bipolar disorder_EA	Lateral occipital	0.039 (T)↑		0.007/0.028 (C)↓
rs2239079	2421079	A	Schizophrenia_AA/Bipolar disorder_EA	Bankssts	0.022 (G)↓	0.012 (G)↓
rs17801265	2468876	A	Schizophrenia_EA/Alcoholism_EA	Temporal pole		7.8 × 10-3 (A)↑	0.015 (A)↑
rs17801265	2468876	A	Schizophrenia_EA/Alcoholism_EA	Entorhinal	0.024 (G)↓	0.012 (A)↑
rs10848680	2635439	G	Nicotine dependence_EA/Bipolar disorder_EA	Fusiform	0.030 (A)↓	0.024 (G)↑

EA, European–Americans; AA, African–American; Bankssts, banks of the superior temporal sulcus. ↑, the effect allele consistent with the risk allele; ↓, the effect allele opposite to the risk allele. Bold P value/q value corresponds to q < 0.05.

In an exploratory analysis, performed for validation of the primary FDR-corrected analysis, we also corrected for multiple comparisons in all of the above SNP-disease association analyses (supplementary tables S1, S2, S3 and 1) using the Bonferroni correction for the number of independent variant blocks (Dʹ > 0.8 within each block).^21–23 In the differential CACNA1C mRNA expression analysis, in which FDR correction was not feasible due to nonnormal distribution of the P values across cohorts, we used only Bonferroni correction, by the number of independent cohorts (table 4).

Table 4.

Differential expression of CACNA1C mRNA in brain

	Expression level		t-Test		References
Phenotypes	Cases	Controls	P	Region	PMID#
Schizophrenia	225.7 ± 30.9	249.1 ± 28.1	.007	anterior prefrontal	19255580
Bipolar disorder	44.8 ± 7.5	40.8 ± 7.0	.036	dorsolateral prefrontal	16894394
Bipolar disorder	47.0 ± 9.6	35.2 ± 4.2	.002	orbitofrontal	16894394
Alcoholism	6.4 ± 0.2	5.9 ± 0.7	.035	hippocampus	23981442
Parkinson’s dis.	2.4 ± 0.1	2.3 ± 0.0	.004	lateral substantia nigra	16344956
Parkinson’s dis.	2.6 ± 0.1	2.5 ± 0.1	.035	lateral substantia nigra	16344956

Bold P values are < Bonferroni-corrected α (=0.01).

Data Quality Control

Prior to the association analysis, we applied stringent criteria to “clean up” phenotype and genotype data, as described previously.^24,25 In brief, the subjects with allele discordance, missing race, a mismatch between self-identified and genetically inferred race, or a missing genotype call rate ≥ 2% across all SNPs were excluded. We excluded monomorphic SNPs and SNPs with allele discordance, Mendelian errors, or an overall missing genotype call rate ≥ 2%. SNPs with MAF ≤ 0.05 in each cohort or in Hardy-Weinberg disequilibrium (P < .0001) in the unaffected subjects were also excluded.

SNP-Diagnosis Association Analysis

For case-control and case-only cohorts, associations between alleles and diagnoses were analyzed using logistic regression (for binary traits) or linear regression analysis (for quantitative traits) as implemented in the program PLINK.²⁶ Diagnosis served as dependent variable, alleles of each SNP served as independent variables, and the first 10 principle components (PCs) of ancestry (to control for population stratification and admixture effects),²⁴ sex, and age served as covariates. For family cohorts, we used DFAM (for binary traits) or QFAM (for quantitative traits), as implemented in PLINK,²⁶ to test associations. All associations were simulated 10 000 000 times.

SNP P values for each phenotype and racial group were meta-analyzed to generate combined P values and heterogeneity (I²) using METAL²⁷ or RevMan (https://revman.cochrane.org). The I² statistic estimates the percentage of variance due to racial heterogeneity across the cohorts included in meta-analysis; heterogeneity is categorized as low (0%–25%), moderate (25%–50%), large (50%–75%), or extreme (75%–100%).

Functional Analysis by Bioinformatics

We conducted a series of bioinformatic analyses to predict the biological functions of disease-sharing risk SNPs using FuncPred²⁸ and VE!P²⁹ characterizing whether the identified SNPs were located in LncRNAs, snRNA or miRNA, in open chromatin regions, in transcription factor binding sites (TFBS), in exonic splicing silencers (ESS) or enhancers (ESE), within methylated CpG islands, or within copy number variations (CNVs), whether these SNPs were predicted to affect protein function or structure, and whether they had regulatory potential, high conservation scores, and/or predicted significance in pathogenesis.

Differential Expression Analysis of CACNA1C mRNA in Postmortem Human Brain

We analyzed the CACNA1C mRNA expression in five independent cohorts of postmortem brain tissue. The cohorts comprised:

• anterior prefrontal cortex—28 schizophrenia and 23 healthy controls³⁰;

• dorsolateral prefrontal cortex—30 bipolar disorder and 31 healthy controls³¹;

• orbitofrontal cortex—10 bipolar disorder and 11 healthy controls³¹;

• hippocampus—6 alcoholism and 6 healthy controls (all female)³²; and

• lateral substantia nigra—9 Parkinson’s disease and 7 healthy controls.³³

Characteristics of these postmortem tissue samples, which have all been described before, are summarized in supplementary table S5. Mean age of these cohorts ranged from 43.5 to 80.0 years; and age and sex distributions were roughly matched between cases and controls. Mean postmortem interval (PMI) ranged from 8.1 to 37.2 h. RNA integrity numbers (RIN) were all >7.0. mRNA expression was assessed using microarrays and validated by qPCR; details of array and probe names and references are listed in supplementary table S5; details of quality control (QC) steps for microarray data and qPCR experiment are described in supplementary methods, and other details regarding demographics, tissue collection sources and procedure, RNA extraction methods, and quality of RNA have been published elsewhere.^30–33 Normalized and log-transformed CACNA1C mRNA expression levels were compared between cases and controls by ANCOVA, in which the scores from expression PC1 were taken to be used as a covariate to account for variability in data quality due to RNA loading of the samples, sample degradation, and other factors. The PCA was performed on the QC metrics for expression data such as average signal, background, standard deviation of the background, number present, raw q, scale factor, GAPDH 3ʹ/5ʹ ratio, β-actin 3ʹ/5ʹ ratio. Other covariates in ANCOVA included age and gender.

Potential Regulatory Effect of Disease-Sharing Risk Variants on ICV and on the GMVs of Subcortical Structures

Brain structural measures were evaluated in two cohorts with European ancestry admixtures, CHAGE-ENIGMA³⁴ and ENIGMA2.^35,36 Intracranial volume (ICV) was measured in >37 000 subjects (26 577 from the CHAGE-ENIGMA, 11 373 ENIGMA2). Gray matter volume (GMV) of basal ganglia (caudate, putamen, pallidum, and nucleus accumbens), amygdala, hippocampus, and thalamus was measured using FIRST³⁷ or FreeSurfer³⁸ in >50 000 subjects (37 542 CHAGE-ENIGMA, 12 994 ENIGMA2). All subjects were genotyped using microarrays, with imputation of missing data based on the 1000 Genome Project panels. Genetic homogeneity was assessed in each subject using multi-dimensional scaling (MDS).

The potential regulatory effects of the disease-sharing risk variants identified above on ICV and GMV measures were analyzed using multiple linear regression analysis, controlling for age, sex, four MDS components, ICV (for non-ICV phenotypes), diagnosis (where applicable; most subjects were free of neurodegenerative and neuropsychiatric disorders), study site (for multi-site studies only), and dummy variables for scanner (when multiple scanners were used at the same site).

Potential Regulatory Effect of Disease-Sharing Risk Variants on Cortical SA and Average TH

Cortical surface area (SA) and thickness (TH) were derived from in vivo whole brain T1-weighted MRI scans using FreeSurfer³⁸ in two independent cohorts, the ENIGMA3 cohort (36 936 subjects of both European and non-European ancestry admixtures) and the UK Biobank cohort (10 083 subjects with European ancestry admixtures).³⁹ SA and TH were quantified for each subject across the whole cortex and within 34 distinct gyral-defined regions in each hemisphere according to the Desikan-Killiany atlas.⁴⁰ SA was measured at the grey-white matter boundary. TH was measured as the average distance between the white matter and pial surfaces. The SA and TH were analyzed separately based on the radial unit hypothesis⁴¹ that these measures are governed by different developmental processes. The total SA and average TH of each hemisphere were computed separately.

We analyzed the potential regulatory effects of the disease-sharing risk variants identified above on a total of 70 traits (total SA, average TH, and SA and TH of 34 cortical regions averaged across right and left hemispheres) using multiple linear regression analyses, adjusting for the effects of sex, linear and nonlinear age effects, interactions between age and sex, ancestry (the first four MDS components), diagnostic status (when the cohort followed a case-control design), MRI acquisition orientation, study site (for multi-site studies only), and dummy variables for scanner (when multiple scanners were used at the same site). For analyses of regional SA and TH, total SA or average TH was added as an additional covariate, to isolate regionally specific genetic influences.

Results

SNP-Disease Association

SNP-disease association in each cohort

A total of 2545 SNPs across the CACNA1C region were examined in all 37 GWAS cohorts The number of SNP-disease associations with P < .05 (N = 13–128/cohort) or q < 0.05 (N = 0–91/cohort] and the minimal P value for the most significant SNP in each cohort (1.7 × 10⁻⁵ ≤ min(P) ≤ 6.7 × 10⁻³) are listed in supplementary table S1. The disorders, at individual cohort level, that survived FDR correction (on 2545 P-values per cohort) included schizophrenia, bipolar disorder, ADHD, Autism, substance dependence (alcoholism or nicotine dependence), Parkinson’s disease, amyotrophic lateral sclerosis, and multiple sclerosis (q < 0.05; supplementary table S1).

These 2545 SNPs were located within 69 (in European ancestry admixtures) or 110 (in African ancestry admixtures) independent variant blocks (Dʹ > 0.8 within each block; supplementary figures S1a and S1b, generated by Haploview⁴²). The corresponding Bonferroni-corrected α is 7.25 × 10⁻⁴ (European) or 4.55 × 10⁻⁴ (African). The number of SNP-disease associations with P < α (N = 0–9/cohort) is listed in supplementary table S1.

SNP-disease association for each disease

Meta-analysis of these 2545 SNP-disease associations for the 13 neuropsychiatric disorders analyzed, separated by ethnicity, identified 2 SNPs significantly associated with schizophrenia (1.6 × 10⁻³ ≤ q ≤ 0.029), 3 with bipolar disorder (4.7 × 10⁻³ ≤ q ≤ 5.0 × 10⁻³), 1 with alcoholism (q = 0.012), and 4 with autism (0.015 ≤ q ≤ 0.050; supplementary table S2). All of these associations (1.8 × 10⁻⁶ ≤ P ≤ 3.1 × 10⁻⁴; supplementary table S2) also survived Bonferroni correction (α = 4.55 × 10⁻⁴). I² values for these meta-analyses ranged from 0 to 0.38 (table 1).

SNP-disease associations shared between disease pairs

Eighteen SNPs were associated with two diagnoses with the same direction of effect (1.8 × 10⁻⁶ ≤ P ≤ .049; table 1). Schizophrenia shared three risk variants with mood disorders, two with autism, and three with substance dependence. Mood disorders additionally shared two risk variants with autism and six with substance dependence. Parkinson’s disease shared two risk variants only with bipolar disorder (table 1; figure 2).

Fig. 2.

Neuropsychiatric disorders sharing risk CACNA1C variants. (Schizophrenia and mood disorders are the center of this map because all other neuropsychiatric diseases share risk variants with them. The thickness of the connecting lines corresponds to the number of sharing SNPs. The solid lines correspond to FDR correction survival, but the dash lines do not. Abbreviations: SCZ, schizophrenia; BP, bipolar disorder; MDD, major depressive disorder; ND, nicotine dependence; AD, alcohol dependence; PD, Parkinson’s disease).

These risk variants spanned the CACNA1C region. The pairwise r² values between two neighboring SNPs ranged from 0.000 to 1.000, which included twelve <0.3, two between 0.3 and 0.6, one between 0.6 and 0.8, one between 0.8 and 1, and one = 1.0 (between rs216016 and rs216017) (supplementary figure S2). Shared risk SNPs for schizophrenia, bipolar disorder, and alcoholism survived both FDR (1.6 × 10⁻³ ≤ q ≤ 0.029) and Bonferroni corrections (1.8 × 10⁻⁶ ≤ P ≤ 6.4 × 10⁻⁵; table 1). Shared risk SNPs for autism survived Bonferroni (P = 5.5 × 10⁻⁴) but not FDR correction (q = 0.059). No risk allele was shared by ≥3 disorders. Ten SNPs were nominally associated with two diagnoses with opposite directions of effect; none of these survived FDR correction (all q > 0.05; supplementary table S3).

Functional Analysis by Bioinformatics

Among the 18 risk SNPs identified in table 1, the only exonic variant (rs11062091) is categorized as likely benign but is predicted to have regulatory potential. Four SNPs are located in predicted transcription factor binding sites (TFBS). One SNP is located in an open chromatin region and may be associated with regulatory factor binding. Four SNPs located in LncRNAs and seven SNPs located in enhancers may regulate gene expression. Two SNPs occur at highly conserved positions. (supplementary table S6).

Differential Expression of CACNA1C mRNA in Brains

CACNA1C mRNA was differentially expressed in the brains between cases and controls across five postmortem transcriptomic datasets (table 4). CACNA1C expression was negatively associated with schizophrenia in anterior prefrontal cortex (BA10), relative to matched controls (P = .007). It was positively associated with bipolar disorder in dorsolateral prefrontal (P = .036) and orbitofrontal (P = .002) cortices, with alcoholism in hippocampus in females (P = .035), and with Parkinson’s disease in lateral substantia nigra (0.004 ≤ P ≤ 0.035). These associations survived Bonferroni correction (α = 0.01 to account for five independent cohorts) in anterior prefrontal cortex in schizophrenia (negatively), orbitofrontal cortex in bipolar disorder (positively), and lateral substantia nigra in Parkinson’s disease (positively).

Cross-Diagnosis (Disease-Sharing) Risk Variants Were Associated With Alterations in ICV and the GMV of Pallidum and Hippocampus

The association of the 18 identified cross-diagnosis risk variants (table 1) with ICV and with GMV of seven subcortical structures were analyzed in two large structural MRI datasets. After FDR correction, two SNPs (rs115773686 and rs2238072; shared by schizophrenia and bipolar disorder) were significantly associated with ICV (2.6 × 10⁻⁴ ≤ P ≤ 5.6 × 10⁻³; 1.5 × 10⁻³ ≤ q ≤ 0.016; supplementary table S4).

The risk allele (T) of rs2238072 was significantly (q < 0.05) negatively associated with ICV in both CHARGE-ENIGMA (P = 7.7 × 10⁻⁴; q = 0.014) and ENIGMA2 (P = 2.6 × 10⁻⁴; q = 1.5 × 10⁻³) cohorts; this association survived FDR correction. rs2238072 was nominally (P < .05; q > 0.05) negatively associated with the GMVs of pallidum in both CHARGE-ENIGMA and ENIGMA2 cohorts (P = .044 and .034, respectively). The risk allele (G) of rs11062091 was nominally (P < .05; q > 0.05) positively associated with the GMV of hippocampus in both CHARGE-ENIGMA and ENIGMA2 cohorts (P = .036 and .050, respectively; table 2).

Cross-Diagnosis Risk Variants Were Associated With Cortical SA and TH

The association of the 18 identified cross-diagnosis risk variants with whole-brain and regional cortical SA and TH was analyzed in two large structural neuroimaging datasets (see Methods). After FDR correction for 70 comparisons per SNP (see Methods), two SNPs were significantly associated with SA of cuneus and precuneus (5.3 × 10⁻³ ≤ P ≤ 5.6 × 10⁻³; 0.029 ≤ q ≤ 0.037) and six SNPs were significantly associated with the TH of frontal pole, lateral orbitofrontal, rostral middle frontal, parstriangularis, transverse temporal, banks of the superior temporal sulcus (bankssts), and lateral occipital cortices (1.4 × 10⁻³ ≤ P ≤ 7.1 × 10⁻³; 0.027 ≤ q ≤ 0.038; supplementary table S4). These eight SNPs were shared by schizophrenia, bipolar disorder, alcohol or nicotine dependence, and Parkinson’s disease (table 1).

Most disease-sharing risk alleles were nominally (P < .05) negatively associated with SA but positively associated with TH, with some exceptions (table 3). The risk allele of rs2238072 was positively associated with the SA (P = .039) but negatively associated with the TH (P = .007) of the lateral occipital cortex; the latter association survived FDR correction (P = .007; q = 0.028).

Discussion

We studied variants in the CACNA1C gene in 37 different patient cohorts, comprising over 70 000 subjects with 13 different neuropsychiatric disorders. Variants in the CACNA1C region were nominally associated with all 13 disorders, both in individual cohorts and after meta-analysis for each disorder (P < .05). After FDR correction, variants in CACNA1C were significantly associated with schizophrenia, bipolar disorder, ADHD, Autism, substance dependence (either alcoholism or nicotine dependence), Parkinson’s disease, amyotrophic lateral sclerosis, and multiple sclerosis in individual cohorts, and with schizophrenia, bipolar disorder, alcoholism, and ASD after disease-specific meta-analysis (q < 0.05). These findings support the hypothesis that variation in CACNA1C is associated with neuropsychiatric illness across diagnoses.

Eighteen risk alleles in the CACNA1C region were nominally shared among schizophrenia, mood disorders (bipolar disorder and depression), substance dependence (alcohol or nicotine dependence), autism, and Parkinson’s disease (table 1). Among these seven disorders, schizophrenia and mood disorders stand out: all CACNA1C variants associated with risk across two diagnoses entailed risk for either schizophrenia or mood disorders, or both. Shared associations with schizophrenia, mood disorders, and alcoholism remained significant after FDR correction. Shared risk between mood disorders and substance use disorders was particularly notable, with six implicated alleles. No single risk allele was shared by more than two diagnoses.

Several subsequent analyses support the conclusion that this variation has relevant biological consequences. Bioinformatic analysis suggested that 13 SNPs of the identified SNPs may regulate gene expression. We found CACNA1C mRNA expression negatively associated with schizophrenia in the anterior prefrontal cortex (BA10); this finding remained significant after Bonferroni correction. Barnes et al have described disruption of many pathways and processes underpinning synaptic plasticity in anterior prefrontal region in schizophrenia.⁴³ Synaptic plasticity can be regulated by LTCCs,^44,45 providing a potential mechanistic link between our findings and this hypothesis; this merits further study. Interestingly, previous work has reported a trend-level reduction in CACNA1C mRNA expression in BA10 in bipolar disorders⁴⁶ and major depression,⁴⁶ consistent with our findings in schizophrenia, suggesting a shared role of CACNA1C mRNA in schizophrenia, bipolar disorder and depression. In contrast, CACNA1C expression was positively associated with bipolar disorder in the orbitofrontal cortex, a structure implicated in the pathophysiology of bipolar disorder,^47,48 and in Parkinson’s disease in the lateral substantia nigra, which is implicated in that disorder’s pathophysiology.⁴⁹

The risk alleles of rs2238072 and rs2239079, which are shared by schizophrenia and bipolar disorder, were negatively associated with ICV and SA and TH of the banks of the superior temporal sulcus (bankssts). Similar structural abnormalities—reduced ICV and reduced GMV of bankssts—have been reported in both schizophrenia and bipolar.^50–55 The risk allele of rs2299657, which is shared by bipolar disorder and Parkinson’s disease, was negatively associated with the TH of the temporal pole, consistent with the reduced GMV of temporal pole previously reported in both of these conditions.^56,57 The associations of disease-associated CACNA1C alleles with ICV and lateral occipital TH survived FDR correction. These various lines of evidence suggest biological mechanisms whereby CACNA1C SNPs may contribute to pathophysiology, especially in schizophrenia and bipolar disorder.

Many neuropsychiatric diagnoses have overlapping clinical symptoms.⁵⁸ For example, schizophrenia and bipolar disorder can both present with psychotic episodes and disorganized thinking. Genetic factors that contribute to both diagnoses, such as those we describe in CACNA1C, may underlie such shared symptomatology. Indeed, CACNA1C has previously been shown to play an important role in overlapping symptoms shared amongst schizophrenia, mood disorders, and autism.¹⁴ Schizophrenia, mood disorders, autism, and substance dependence have high rates of comorbidity.^59–61 Bipolar disorder has been suggested to be a risk factor for the development of Parkinson’s disease.⁶²CACNA1C variants may contribute to this comorbidity.

The association of CACNA1C alleles with multiple neuropsychiatric conditions, especially schizophrenia and bipolar disorder, suggests that the gene may play an important role in fundamental neuropathophysiological processes that contribute to multiple disorders. CACNA1C has been implicated in genomic effects, alterations in synaptic plasticity, regulation of the dopamine D1-receptor^63,64 and CREB/BDNF signaling pathways,^65,66 all of which have been implicated in neuropsychiatric pathogenesis. These are likely to be fruitful areas for future work better elucidating the neurobiological consequences of CACNA1C variation.

This study has limitations. We examined only thirteen neuropsychiatric disorders and only one of the promising risk genes. Allelic variation in other genes will of course contribute similarly to overlapping pathophysiological processes. Additionally, heterogeneity between cohorts, arrays, and/or quality of imputation is likely and may bias or attenuate our findings. Finally, only samples of European and African ancestry admixtures were analyzed in this study, and future examination of additional ancestry admixtures, such as Asian ancestry admixtures,⁴ could increase the generalizability of findings and improve rigor.

Conclusion

We identified 18 risk CACNA1C SNPs that were shared by pairs of seven neuropsychiatric disorders. Schizophrenia and bipolar disorder are most strongly implicated. CACNA1C variants appear to contribute to shared neuropathogenic mechanisms across these neuropsychiatric disorders and merit further study.

Supplementary Material

Supplementary material is available at https://academic.oup.com/schizophreniabulletin/.

Supplementary Figure S1. LD map of all CACNA1C variants in Europeans (a) or Africans (b). (These LD maps are generated using Haploview. There are 69 variant blocks in Europeans (a) and 110 in Africans (b) across the entire CACNA1C. Dʹ > 0.8 within each block).

Supplementary Figure S2. LD map of 18 disease-sharing risk CACNA1C variants. (Red and blue squares indicate r² and Dʹ values, respectively. Pairwise r² values between neighboring SNPs are presented. This LD map is generated using LDMatrix).

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship and/or publication of this article: This study is also funded by the National Natural Science Foundation of China (81201057; 81771452), Shanghai Natural Science Foundation (20ZR1448400), Shanghai Municipal Health Bureau Project (20124109), Chinese Medical Association, Psychiatry—Servier Youth Research Fund, Shanghai Mental Health Center international cooperation project (2013-) and Shanghai Municipal Center for Mental Health Clinical Research Program.

Conflict of Interest Statement

There is no conflict of interest to declare.