Copy Number Variants and Their Association With Intracerebral Hemorrhage Risk: A Case–Control Study

ABSTRACT

Introduction

Intracerebral Hemorrhage (ICH) is a leading cause of morbidity and mortality worldwide and lacks effective therapeutic interventions. Despite previous studies, the genetic underpinnings of ICH remain poorly understood. We sought to investigate the role of copy number variants (CNVs) in ICH pathophysiology to identify novel etiological mechanisms and therapeutic targets.

Methods

We analyzed microarray data from ICH cases and controls using PennCNV software, followed by rigorous quality control and CNV validation. This case–control study explored associations between large, rare CNVs and ICH risk, examining CNVs that are genic, contain ohnologs, or span multiple genes. Functional enrichment analysis was also conducted to identify overrepresented CNVs in biological processes relevant to ICH.

Results

The study included 649 cases and 437 controls who passed quality control. Cases were more likely to have genic CNVs compared to controls (39.6% vs. 32.5%, p value = 0.02) after adjusting for age, sex, hypertension, and admixture. In pathway analysis, CNVs associated with cholesterol biosynthesis (6.3% vs. 3.2%, p value = 0.04) and blood vessel development were more common in cases (8.2% vs. 5.3%, p value = 0.06). While associations were observed at nominal significance (p < 0.05), none withstood correction for multiple comparisons.

Discussion

Our study represents the first CNV analysis of ICH risk. While significant associations between large, rare CNVs and ICH risk were not established, the findings indicate potential pathways for further research. The identified trends warrant further investigation, and our methodological framework provides a foundation for future large‐scale studies aimed at elucidating the role of CNVs in ICH pathogenesis and outcomes.

1. Introduction

Intracerebral hemorrhage (ICH) accounts for 10%–20% of all strokes worldwide and is the deadliest subtype of stroke, with a 40% 30‐day case fatality rate [1]. Previous studies have suggested a combination of both environmental and genetic factors in the pathophysiology of ICH [2, 3]. However, no demographic (e.g., age, sex, ancestry, race/ethnicity, family history), clinical (e.g., hypertension), or genetic profile fully explains the risk of ICH. The heritability estimate of ICH has been calculated to be at least 29%, and its pathophysiology is thought to be driven in part by a polygenic architecture [4]. Despite previous efforts to perform large‐scale genome‐wide association studies (GWAS) of ICH [5], our knowledge surrounding the genetic underpinnings of the disease remains very limited. Exploring the role of genetic structural variants such as copy number variants (CNVs) in the etiology of ICH could contribute to our understanding of this complex condition and possibly lead to the development of new prevention and treatment strategies through the identification of novel contributing pathways and motivate drug targets.

CNVs are segments of the genome that have been either duplicated or deleted, contributing significantly to genetic diversity among individuals. These structural variations can impact numerous genes and are particularly implicated in neurodevelopmental and neurodegenerative disorders [6, 7, 8, 9]. CNVs arise through several genomic mechanisms such as non‐allelic homologous recombination, microhomology‐mediated events, and non‐homologous repair [10]. On average, CNVs affect approximately 4.8%–9.5% of the human genome [11]. While SNPs are more numerous in the genome, CNVs typically affect a larger number of nucleotides per variant, making them substantial contributors to genomic structural diversity between individuals [6, 12]. These variants can range in size from 50 base pairs to entire chromosomes [13]. They tend to cluster in specific “genomic hotspots” such as regions of complex genetic architecture (where a significant number of Locus Control Regions (LCRs) are found), in pericentromeric and subtelomeric regions, and in other regions that are determined by specific chromosomal structure [11].

The specific function of CNVs and therefore their clinical interpretation has not yet been fully elucidated partly due to the complexity of the CNV analysis workflow. However, studies suggest that the length of a CNV is less important in determining its clinical impact than the complex interplay between the protein‐coding genes, dosage‐sensitive genes, and regulatory elements involved [13]. For example, ohnologs represent dosage‐sensitive genes that have been rarely duplicated and are therefore more likely to be overrepresented in pathogenic CNVs [14]. Deletions generally tend to be more deleterious to human health than duplications [15]. The frequency of a CNV in the population also affects its association with diseases, with rare CNVs (frequency less than 1%) significantly contributing to common polygenic diseases [16]. Conversely, only a small number of prevalent CNVs have been associated with diseases [17]. Various CNV types have been linked to Mendelian, sporadic, and complex diseases, including schizophrenia, autism, HIV susceptibility [18], epilepsy [19], and vascular conditions [20, 21]. No studies have yet examined the relationship between CNVs and ICH.

Given the substantial heritability of ICH, the unexplored potential of CNVs in large samples, and their demonstrated impact across the genome and association with various complex traits, this study aims to investigate the role of rare CNVs in a multi‐center case–control cohort. By focusing on CNVs associated with dosage‐sensitive genes and relevant biological processes, we seek to identify novel genetic factors that could contribute to the risk of ICH, potentially leading to new preventive and therapeutic strategies.

2. Methods

2.1. Sample Description

This study of CNV analysis included patients previously enrolled in the Genetic and Environmental Risk Factors for Hemorrhagic Stroke (GERFHS) and the Genetics of Cerebral Hemorrhage with Anticoagulation (GOCHA) studies along with cases from the following sites: Hospital del Mar Intracerebral Hemorrhage study and Vall d'Hebron Hospital ICH study in Barcelona, Spain; Jagiellonian University Hemorrhagic Stroke Study in Krakow, Poland; and the Lund Stroke Register study in Lund, Sweden. Because of their limited sample sizes, data from the four European studies (ESs) were analyzed together for the purposes of quality control. Selection of cases and controls for these studies has been previously described in the corresponding genome‐wide association study that forms the base of this study [21]. Briefly, GERFHS is a population‐based, case–control study that enrolled Black and non‐Hispanic White (NHW) participants within a 50‐mile radius of the University of Cincinnati and population‐based controls matched by age (±5 years), race, and sex from the same population. The GOCHA study enrolled ICH patients of European ancestry aged > 55 years from multiple academic centers. Controls enrolled in the GOCHA study must not have had a history of ICH.

Control subjects were ICH‐free individuals enrolled from the same population that gave rise to the case subjects at each participating study site, aged > 55 years (GOCHA) and > 18 years (GERFHS and ESs). Control subjects were sampled by random digit dialing in GERFHS and from ambulatory clinics in the remainder of the studies.

Demographic variables were systematically obtained from structured participant and family member interviews at each site, along with demographic variables including self‐identified race/ethnicity, age, sex, medical history, smoking status, and alcohol use. European ancestry was confirmed via the GWAS genotype data.

All studies were approved by the Institutional Review Board or ethics committee at each participating site. Participants provided informed consent; when subjects lacked capacity or competency, consent was obtained from their legal proxies. Individuals without baseline demographic and clinical data collected were excluded from this analysis.

2.2. Quality Control of the Microarray Samples, CNV Analysis and Mapping

CNVs were detected using high‐density single‐nucleotide polymorphism microarray data from 649 ICH patients and 437 controls (Affymetrix Human SNP Array 6.0). The quality control process for our genotype data has been previously described [5]. Prior to CNV‐level analysis, each sample underwent rigorous quality control using noise‐free‐CNV software (https://noise‐free‐cnv.sourceforge.net/). This software was employed to mitigate genomic wave noise (noise attributed to the GC content of the DNA) and individual probe set deviations. Samples exhibiting excessive GC‐wave or SNP‐level noise were excluded. Additionally, samples showing potential cryptic relatedness, as indicated by distogram clusters, were also removed. Finally, samples with an outlier number of CNV calls (> 125 or < 15 CNVs) were excluded.

CNVs were visually inspected for accuracy based on the log R ratio (LRR) and B allele frequency (BAF) and analyzed with the PennCNV software [22]. To confirm the authenticity of CNV calls, LRR values were compared against a reference sample. This step was crucial in identifying somatic mosaicism and correcting “split calls”—cases where CNVs were erroneously divided into multiple segments. Known CNVs from the Database of Genomic Variants (DGV) [23] were consulted to check the frequency and previously reported CNVs within the population. For our analysis of “large, rare” CNVs, we defined “large” as those spanning ≥ 20 consecutive SNP probes (typically corresponding to regions > 100 kb in length) and “rare” as those with a minor allele frequency < 1% in both our study population and reference databases. Exclusions were made for CNVs not meeting these criteria, as well as those without coding sequences, particularly in complex regions like pericentromic or heterochromatic zones, within the MHC region of chromosome 6, and the Y chromosome, or within segmental duplications (as detection of CNVs in those repetitive areas of the genome is challenging).

The breakpoints of each validated CNV were mapped onto the human genome (GRCh38) via the Ensembl database (https://www.ensembl.org/Homo_sapiens) to ascertain their dimensions and impacted genes. CNVs were classified as genic if they disrupted protein‐coding regions or contained gene fusion breakpoints as previously described [23]. More specifically, a CNV was classified as “genic” if it involved either the deletion of one or more coding exons or the duplication of any coding region, including internal exons [24]. Conversely, non‐genic CNVs, which did not affect coding sequences (e.g., intronic or intragenic), were excluded. CNVs containing ohnologs—genes preserved through evolution and sensitive to dosage variations—were identified using the Ohnolog Repository (https://ohnologs.curie.fr/) and were included in the analysis [25].

2.3. Functional Enrichment Analysis

The individuals were further classified depending on whether their CNVs encompassed genes linked to any of four a priori selected biological processes involved in ICH pathophysiology [26, 27, 28, 29]: cholesterol biosynthetic process, extracellular matrix organization, blood vessel development, or coagulation. The analysis was constrained to predefined gene ontology categories to be mindful of the study's sample size. We used as input all of the CNV‐involved genes of each individual and tested for enrichment of each of the previously mentioned biological processes utilizing the GO::TermFinder [30] (https://go.princeton.edu/cgi‐bin/GOTermMapper). The number and proportion of individuals with CNV genetic contents enriched in the above processes were compared between cases and controls.

2.4. CNV and ICH Risk Statistical Analysis

Participant characteristics are reported as mean, standard deviation, and percentages. A logistic regression model was computed to test for an association between ICH and rare CNVs. The primary exposure variable was defined as the presence (yes/no) of specific CNV categories in each individual: CNVs involving at least one gene, CNVs spanning more than five genes, CNVs containing at least one ohnolog, and CNVs enriched in biological processes relevant to ICH. The logistic regression model adjusted for age, sex, hypertension status, and population substructure based on the Admixture analysis. Since 59 individuals had missing data for hypertension status, a sensitivity analysis was computed, dropping hypertension as a covariate in the model.

All statistical analyses were performed using R studio, version 2024.04.2 + 764, with an FDR‐adjusted p value threshold of less than 0.05 for statistical significance.

3. Results

The overall study design is presented in Figure 1. We analyzed 1087 individuals (649 cases and 437 controls, median age 71.3 vs. 69.0, respectively) that met inclusionary criteria for this analysis as described in the methods. Demographics and ICH risk factors of the cohort can be found in Table 1. We excluded 207 samples (19% of the total cohort) due to either containing an outlier number of Penn CNV calls (> 125 or < 15), duplicate samples, or containing a high level of genomic wave or per‐SNP noise after visual inspection. The most common reason for excluding individual samples was a high level of GC‐wave noise identified through visual inspection in the noise‐free‐CNV software.

FIGURE 1.

Study overview. This diagram outlines the sequential steps undertaken in our study to analyze copy number variants (CNVs) in a cohort of 649 Intracerebral Hemorrhage (ICH) cases and 437 control subjects.

TABLE 1.

Descriptive characteristics and APOE genotype of study participants.

Variable	Control (N = 437)	Case (N = 649)
Age (years)	67.8 ± 13.5	69.6 ± 13.9
Female	221 (50.6%)	309 (47.6%)
History of hypertension	216 (49.4%)	473 (72.9%)
History of diabetes	61 (14.0%)	145 (22.3%)
History of high cholesterol	199 (45.5%)	288 (44.4%)
History of Afib EKG	35 (8.0%)	89 (13.7%)
Smoking—Current	49 (11.2%)	85 (13.7%)
Smoking—Former	183 (41.9%)	278 (44.7%)
Smoking—Never	205 (46.9%)	259 (41.6%)
Current alcohol use	188 (46.7%)	165 (32.7%)
E2 Alleles—0	328 (83.7%)	501 (77.9%)
E2 Alleles—1	63 (16.1%)	131 (20.4%)
E2 Alleles—2	1 (0.3%)	11 (1.7%)
E3 Alleles—0	21 (5.4%)	57 (8.9%)
E3 Alleles—1	140 (35.7%)	255 (39.7%)
E3 Alleles—2	231 (58.9%)	331 (51.5%)
E4 Alleles—0	286 (73.0%)	450 (70.0%)
E4 Alleles—1	95 (24.2%)	170 (26.4%)
E4 Alleles—2	11 (2.8%)	23 (3.6%)

Following CNV filtering and validation, we performed a comparative analysis between cases and controls and adjusted for age, sex, hypertension, and admixture (Table 2). Of the entire cohort, 36.8% carried at least one genic CNV, with ICH patients carrying a greater percentage than controls (39.6% vs. 32.5%; OR = 1.38, p = 0.02). No significant difference was observed between cases and controls in the presence of CNVs in five or more genic regions or CNVs containing ohnologs either before or after adjustment for predetermined variables. Cases demonstrated enrichment for CNVs in the cholesterol biosynthesis pathway (6.3% vs. 3.2%, p = 0.04) and a trend towards CNVs in the blood vessel development pathway (8.2% vs. 5.3%, p = 0.06), but not for CNVs containing ohnologs (19.0% vs. 17.6%, p = 0.57). Neither CNVs spanning extracellular matrix organization genes (2.5% vs. 3.0%, p = 0.79) nor spanning coagulation biological process genes (0.9% vs. 2.3%, p = 0.12) exhibited CNV enrichment in ICH cases. None of the pathways tested met an FDR‐adjusted threshold of significance. Exclusion of hypertension as a covariate in the logistic model did not alter inferences. Furthermore, a case‐only analysis of the two ICH subtypes—lobar and non‐lobar—revealed no significant variations in CNV characteristics between the cohorts (p > 0.20 for all comparisons; Table S1).

TABLE 2.

Comparative analysis of copy number variant (CNV) characteristics in patients and controls adjusted for age, sex, hypertension, and admixture.

Variable	Control (N = 437)	Case (N = 649)	p value	OR	95% CI	FDR p value
Genic CNVs	142 (32.5%)	257 (39.6%)	0.02	1.38	1.06–1.79	0.12
CNVs affecting > 5 genes	26 (5.9%)	25 (3.9%)	0.15	0.65	0.36–1.16	0.20
CNVs associated with blood vessel development	23 (5.3%)	53 (8.2%)	0.07	1.63	0.97–2.76	0.15
CNVs associated with extracellular matrix organization	13 (3.0%)	16 (2.5%)	0.55	0.79	0.37–1.71	0.57
CNVs associated with cholesterol biosynthetic process	14 (3.2%)	41 (6.3%)	0.04	1.93	1.02–3.66	0.15
CNVs associated with coagulation biological process	10 (2.3%)	6 (0.9%)	0.12	0.44	0.15–1.25	0.20
CNVs containing ohnologs	77 (17.6%)	123 (19.0%)	0.57	1.10	0.79–1.52	0.57

4. Discussion

Our study tested for an association between large, rare, genic CNVs and the risk of ICH. Utilizing microarray samples of historically collected ICH cases and controls of European ancestry, as confirmed by the genetic data itself, we explored known CNVs using well‐established software, augmented by a stringent quality control step to confirm our findings. We conclude that although patients carry more genic CNVs and CNVs involved in relevant biological processes, these findings did not reach statistical significance using FDR. This study represents the first effort to rigorously examine the role of CNVs in the pathophysiology of ICH and an important negative report that greater power is needed and should be followed by further, larger‐scale studies.

Although this is the first study to evaluate the association between CNVs and ICH, previous studies have examined the role of CNVs in ischemic stroke and its risk factors. For instance, CNVs were found in patients with cervical artery dissection, including the MYH11/ABCC6 locus [21]. People with lower copy numbers within the LpA gene have twice the risk of ischemic stroke compared to those with higher copy numbers [17]. Regarding ischemic stroke outcomes, patients with a higher number of large, gene‐containing CNVs experienced poorer functional outcomes at 3 months [31]. A few single‐center cohorts have studied the association between CNVs and subarachnoid hemorrhage. A study of 23 Korean patients found no clear risk association [32], while a study of 191 Japanese patients identified two common variants associated with aSAH risk [33].

Notably, none of the CNVs identified in our study physically encompassed SNPs previously associated with ICH risk in GWAS, including those at the 1q22 locus. This suggests CNV analysis may uncover distinct genomic regions contributing to ICH pathophysiology through mechanisms not captured by SNP‐based approaches, highlighting the complementary value of diverse genetic methodologies in characterizing the genetic architecture of ICH.

Our study has several strengths that increase the reliability of our results. An important component of the study is the visual inspection of noise‐reduced datasets, enabling a critical evaluation of PennCNV findings, including their presence, their length, and the position of their breakpoints. Manual curation of software CNV calls is important as currently available CNV calling algorithms might not accurately differentiate between true and artificial CNV calls and often fail to reduce the GC‐wave and residual noise of the samples [34, 35]. Our study also contains a large cohort of primary ICH patients whose phenotype is well characterized. Additionally, we use large, publicly available databases of known CNVs to corroborate their existence in the population. Finally, apart from the number of CNVs, we consider important features of the CNVs such as their frequency in the population, their involvement in biological processes, and their genic content, including the involvement of ohnologs.

Our study also has limitations. First, we employed a focused analytical strategy by examining only large, rare, genic CNVs. This targeted approach was adopted due to our limited sample size, allowing us to maximize statistical power for detecting biologically relevant associations. However, it necessarily excluded potentially important non‐coding CNVs that could affect regulatory elements and contribute to ICH pathophysiology. There were several rationales for our focused approach: (1) The effect of common, small, non‐coding CNVs is less likely to contribute to human disease [36]; (2) Common, small, and non‐coding CNVs are typically in linkage disequilibrium (LD) with common SNPs [37], meaning their associations could potentially be detected by existing GWAS; and 3) The microarray SNP data from which these CNVs are derived were specifically designed to identify risk genes, making a “genic” focus logical. Nevertheless, recent evidence suggests that structural variants in non‐coding regions can significantly impact gene expression and disease risk through disruption of enhancers, promoters, and three‐dimensional chromatin organization [36, 37], and these elements were not covered in our analysis. The functional interpretation of such variants requires additional genomic annotations and substantially larger sample sizes than were available for this study.

We acknowledge that the inclusion of individuals from separate studies might introduce batch effects. Finally, it would appear that the effect sizes observed remain modest and will require larger sample sizes in well‐phenotyped studies to detect CNV associations and subsequent associations will likely need to be robustly tested in independent samples, both qualifiers are as we see in other genetic association studies. Whether our associations did not reach multiple comparisons thresholds of significance due to sample size or the innate heterogeneity in ICH (e.g., cause, location) will require additional investigations. Additionally, it's important to expand future studies to include individuals of different ancestries and ethnicities due to their unique genetic and cultural characteristics.

5. Conclusion

In conclusion, our study represents the first effort to elucidate the role of CNVs in the pathophysiology of ICH. Although we are unable to report significant associations between rare CNVs and ICH risk, our findings demonstrate proof of concept preliminary findings that CNVs with specific contents and characteristics may contribute to ICH development. Therefore, our study underscores the necessity for further research with substantially larger datasets. Such well‐curated CNV analyses could augment our understanding of ICH and pave the way for novel therapeutic strategies.

Author Contributions

S.P., C.G.‐G., D.W., and S.L.D. contributed to the conception and design of the article; C.D.L., P.S., M.C., and T.H. contributed to the acquisition and/or analysis of data; T.N.K., C.B., H.I.H., J.R., and C.D.A. contributed to drafting the text or preparing the figures.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Table S1: acn370230‐sup‐0001‐TableS1.docx.

Funding Statement

This work was funded by Clinical Center grant R01NS042147; National Institutes of Health grants R01NS059727, R01NS36695, and U01NS036695.

Data Availability Statement

The data supporting the findings of this study may be made available to other qualified investigators upon reasonable request.