Genetics of Spontaneous Intracerebral Hemorrhage

Introduction

The epidemiology of spontaneous intracerebral hemorrhage (ICH) has rapidly progressed in the past few decades, with important genetic and non-genetic discoveries.¹ Hypertension has long been associated with ICH and evidence now demonstrates that untreated hypertension produces substantial increases in risk of this condition.^2,3 Unlike cardiovascular disease, cholesterol is inversely associated with ICH, with low levels of total and LDL cholesterol having an association with increased risk. ^4,5 Along these lines, the Stroke Prevention by Aggressive Reduction in Cholesterol Levels (SPARCL) study⁶ demonstrated that, in secondary prevention of ischemic stroke, statin use was associated with a small increase in risk of ICH, although large meta-analyses did not confirm this correlation in the general population.⁷ Moderate alcohol consumption has consistently been identified as associated with a lower ICH risk, while heavy alcohol consumption has been found to associate with an elevated risk of this condition.⁸ Anticoagulant treatment, largely in relationship to warfarin use, has also been consistently associated with ICH risk with a particular predilection for cerebellar ICH.^9,10 Finally, cerebral amyloid angiopathy, first recognized as an ICH mechanism in the 1990s, is a common feature of lobar hemorrhages.¹¹

While these risk factors explain an important proportion of the variance in ICH risk, a significant portion of this variation remains unexplained. In addition, in the absence of proven acute treatments for ICH, novel targets for therapeutic interventions are urgently needed. Population genetics can contribute to solve these two unanswered questions, as heritability estimates based on genome-wide data from unrelated individuals indicate that up to 30% of ICH risk can be explained by common and rare genetic variation.¹² Further, through studies that combine environmental and genetic risk factors, genomic analysis can also help us understand how individual susceptibility to environmental risk factors varies across a given population.¹³ Here, we review the most significant findings to date in the area of ICH genetics (Table 1).

Table 1.

Non-Mendelian genetic contribution to intracerebral hemorrhage.

Ref.	Study type	Stroke subtype	Chr.	Gene landmark	Strongest association	Risk allele	Risk allele frequency	Odds ratio	p value
14	Candidate gene	Lobar ICH	19	APOE	rs429358 / rs7412	e2	0.12	1.82	6.6 × 10-10
14	Candidate gene	Lobar ICH	19	APOE	rs429358 / rs7412	e4	0.12	2.20	2.4 × 10-11
15	GWAS	Nonlobar ICH	1	PMF1 / SLC25A44	rs2984613	C	0.32	1.33	2.2 × 10-10
16	GWAS	Nonlobar ICH	13	COL4A2	rs9588151	T	0.34	1.23	5.54 × 10-9
17	Candidate gene Sequencing	Lobar + Nonlobar ICH	13	COL4A1	c.C1055T	T	<0.01	-	-
					c.C1612G	G
18	Candidate gene Sequencing	Lobar + Nonlobar ICH	13	COL4A2	c.C3448A	A	<0.01	-	-
					c.G5068A	A
					c.A3368G	G

Ref. = Reference; Chr. = chromosome; ICH = intracerebral hemorrhage. Odds ratio and p-value are not reported for the last two rows because these sequencing studies found rare variants to be present in a few cases and none of the controls, precluding formal association testing.

Paradigm Shift: Candidate gene to genome wide and sequencing studies

A candidate gene approach evaluates the possible association between genetic variation at a specific gene or genomic region and risk of disease.¹⁹ Candidates are selected based on the cellular function of encoded proteins, known associations with environmental risk factors, or known association of the proposed gene and related phenotypes. While easy to perform with small sample sizes, much of their success depends on candidate selection, variation within populations, and the potential for biases leading to false positive and negative results.

Advances in genotyping technology led to the ability to screen millions of genetic markers across the genome.²⁰ Genome-wide association studies (GWAS) utilize this technology to agnostically interrogate the entire human genome. Genome-wide genotyping chips include 500,00 to 1 million pre-selected genetic markers that are common in the general population. Through linkage disequilibrium and imputation (see below), the number of evaluated variants increases to several millions, resulting in ~90% coverage of common variation across the genome in Caucasian populations. Because these several million variants evaluate up to one million independent genetic loci, correction for multiple testing leads to the stringent statistical significance threshold of p<5×10^-8. As a result, genetic studies utilizing this technology require large sample sizes.

Achieving appropriate sample sizes has been particularly difficult for ICH due to its low incidence (15% of all strokes) and heterogeneous biology, the latter leading to different risk factor profiles when considering different locations of the hemorrhage within the brain.²¹ To attain the necessary sample sizes, avoid false positive results, and overcome phenomena like the winner’s curse, modern GWASs combine data from several different studies and include an independent replication. This massive amount of data can only be collected through the coordinated efforts of large, international, multi-disciplinary research collaborations; for cerebrovascular diseases, the International Stroke Genomics Consortium (ISGC) is currently the largest network of stroke genetic investigators around the world. In addition to large consortia, the advancement of population genetics has been further supported by the commitment of the scientific community to open access and free data sharing through structures like dbGap, MEGASTROKE, the CHARGE Consortium, and the Cerebrovascular Disease Knowledge Portal.

Technical and analytical terms

The field of population and medical genetics utilizes several terms that are highly specific. Below, we provide a succinct summary of some of these terms.²²

Mutation / Genetic variant / DNA sequence variant / Allele

These terms are usually used interchangeably.

Minor allele frequency (MAF)

The single nucleotide polymorphism (SNP) is the most widely studied mutation in population genetics. As the name implies, this mutation involves only one base pair position. Most SNPs involve only two alleles: one found in a minor proportion of the studied population (the minor allele) and another found in most people (the major allele). Based on the minor allele frequency (MAF), SNPs are classified as common (MAF >5%), low-frequency (MAF 0.5-5%), rare (MAF <0.5%) and private (minor allele present in a few families).

Penetrance

Percentage of individuals with a mutation that have the disease of interest.

Mendelian pattern of inheritance

Diseases with this type of inheritance are caused by private, highly penetrant mutations; they are also called Mendelian, monogenic or familial conditions and are classified as autosomal dominant, autosomal recessive, and x-linked. In Mendelian conditions, the genotype predicts the disease status with high accuracy.

Non-Mendelian (non-familial) genetic contribution

Common mutations with low penetrance can result, at a population level, in a statistically significant, albeit small, increase in risk of disease. Most recent genetic findings in ICH involve this type of genetic effect. The genotype status for these low-penetrance mutations does not predict if the carrier will develop the disease.

Linkage disequilibrium (LD)

At a population level, mutations are correlated with other contiguous mutations. These mutations are in LD, whereas genetic variants that arise independently from one another are in linkage equilibrium.

Susceptibility locus / genomic (or genetic) locus

A set of highly correlated mutations is referred to as an LD block. In GWAS, an LD block that is highly associated with the disease of interest is called a susceptibility locus. Genetic association studies can identify susceptibly loci, but are unable to differentiate the specific mutation within the LD block causally related to disease.

Phenotyping based on biology

Accurate phenotyping has been crucial to the success of genetic studies of ICH. ICH is the final manifestation of different types of cerebral small vessel disease that develop over several years before the brain bleed takes place. Existing phenotyping approaches for ICH aim to reflect these biological differences in underlying mechanism. Similar to most clinical trials and observational studies, genetic studies classify ICH based on the location of the hematoma within the brain. Utilized categories include lobar (cortex and junction between cortex and white matter) and nonlobar (basal ganglia, thalamus, brainstem and cerebellum) ICH, the latter also called deep. Lobar hemorrhages are mostly related to cerebral amyloid angiopathy (CAA),¹¹ whereas nonlobar hemorrhages are mostly related to arteriosclerosis (also referred to as hypertension-related vasculopathy), a small vessel disease linked to long-standing hypertension.²³

The epsilon variants within APOE: an unusually strong genetic risk factor for ICH

Candidate genes studies have identified the epsilon variants within APOE as an important genetic risk factor for ICH. APOE is located in chromosome 19, consists of four exons and three introns totaling 3597 base pairs, and clusters with other genes related with lipid metabolism, including APOC1 and APOC2. APOE codes for Apolipoprotein E (APOE), a 299 amino acids long protein that transports lipoproteins, fat-soluble vitamins, and cholesterol into the lymph system and blood stream. It is synthesized principally in the liver, but has also been found in brain, kidneys, and spleen. APOE is a polymorphic gene with 3 alleles frequently encountered in the general population. These alleles are haplotypes constructed by two different SNPs: rs7412 and rs429358. These two SNPs produce the following alleles: epsilon 2 (rs7412-T, rs429358-T), epsilon 3 (rs7412-T, rs429358-T) and epsilon 4 (rs7412-C, rs429358-C), with allele frequencies of approximately 7%, 81% and 14%, respectively.²⁴ These three alleles produce proteins that differ in two amino acids: APOE epsilon 2 corresponds to cys112/cys158, epsilon 3 to cys112/ arg158 and epsilon 4 to arg112/arg158.²⁴

Converging lines of research indicate that APOE epsilon 2 and 4 substantially increase the risk of first and recurrent lobar ICH, in both spontaneous and warfarin-related hemorrhages. In the Cincinnati/Northern Kentucky study, these two epsilon alleles were strongly associated with an elevated risk of lobar ICH, with a population attributable risk of 30%.²⁵ A small candidate gene study of patients that survived a lobar ICH found that subjects who were homozygous for epsilon 2 and 4 had recurrence risks of 41% and 27%, respectively, substantially higher than the 10% recurrence risk estimated for individuals who are homozygous for epsilon 3.²⁶ By achieving association results that were genome-wide significant, a large multi-center, international candidate gene study provided definitive evidence for a role of these variants in lobar ICH. Using APOE epsilon 3 homozygosity as the reference category, epsilon 2 and 4 were associated with odds ratios of 1.8 and 2.2, respectively.¹⁴ Of note, similar results were obtained by a follow-up study that focused on warfarin-related hemorrhages.²⁷ Interestingly, while APOE epsilon 4 is robustly associated with cerebral amyloid angiopathy, the specific small vessel disease found in 70% of lobar ICH cases, the evidence for a link between epsilon 2 and CAA is inconclusive.

1q22: The first non-familial genetic risk factor for deep ICH

The ISGC completed the first GWAS of ICH in 2014.¹⁵ This multicenter study enrolled 1,545 cases (664 lobar and 881 nonlobar) and 1,481 controls, and identified 1q22 as the first non-familial genetic risk locus for nonlobar ICH. As stated above, genomic risk loci identified by GWAS correspond to genomic regions that contain several genetic variants in high LD. 1q22 corresponds to a narrow 48 kilobase region located in the long arm of chromosome 1 that contains PMF1 and SLC25A44. The highest association within this locus corresponded to rs2984613, a SNP with a MAF of 32% that associated with a 33% increased risk of deep ICH. Providing further confirmation of these results, a large scale GWAS of white matter hyperintensities, a neuroimaging biomarker of cerebral small vessel disease, identified 1q22 as a genome-wide significant susceptibility risk loci.

The biological mechanism underlying the association between 1q22 and nonlobar ICH remains to be elucidated. Polyamine metabolism and oxidative phosphorylation in the mitochondria are two possible processes that could mediate the observed association. Polyamine-modulated factor 1 (PMF1), the protein product of PMF1, plays an important role in the alignment of chromosomes and the formation kinetochores during mitosis. PMF1 exerts its effect in the cell nucleus and its function is modulated by polyamines. Closing a feedback loop, PMF1 also produces transcriptional induction of an acetyltransferase responsible for the rate-limiting step in the catabolic pathway of polyamines.²⁸ Observational evidence indicates that this could be the cellular process mediating the observed association, as polyamines are elevated in stroke patients^29,28,27and associate with breakdown of the blood-brain barrier in animal models.³⁰ SLC25A44 encodes the solute carrier family 25-member 44, a mitochondrial carrier protein.³¹ Prior studies showed that the burden of genetic variation across mitochondrial genes whose products intervene in oxidative phosphorylation associate with risk of nonlobar ICH.³²

COL4A1 and COL4A2: an example of rare and common genetic contribution to ICH risk

Several studies indicate that genetic variation in COL4A1 and COL4A2 associates with an elevated risk of ICH. A candidate gene study¹⁷ that used targeted sequencing in 96 unrelated cases and 145 controls identified two rare mutations within COL4A1 that were present in ICH cases only. These variants, COL4A1(P352L) and COL4A1(R538G), produced missense changes in amino acids with high conservation across species. Because COL4A1 is physically close to COL4A2, and COL4A1 and COL4A2 (their respective protein products) are structurally and functionally related, a second study¹⁸ evaluated rare genetic variation in the same cohort of individuals. It found three rare non-synonymous coding variants (COL4A2(E1123G), COL4A2(Q1150K), COL4A2(A1690T)) that were present in ICH cases only. Importantly, the latter study also explored possible mechanistic pathways using cellular essays, finding that these 3 mutations produced cytotoxicity by triggering intracellular accumulation of both COL4A1 and COL4A2.

Common genetic variation within COL4A1 and COL4A2 is also linked to ICH. An international, multicenter candidate gene study¹⁶ evaluated the role of common mutations in these genes in nonlobar ICH, small vessel ischemic stroke and magnetic resonance imaging-defined white matter hyperintensities. An intronic locus within COL4A2 was strongly associated specifically with risk of nonlobar ICH, with rs9521733, the top associated SNP within this locus, resulting in a 29% increase in risk of this type of brain bleed. It is possible, although still unproven, that these common variants lead to an elevated risk of ICH by dysregulating the cellular mechanism identified by the sequencing studies described above.

Mendelian (monogentic) diseases that manifest with ICH

The genetic variants described thus far correspond to common and rare mutations with incomplete penetrance that associate with an increased likelihood for sustaining an ICH, but do not result in a distinguishable Mendelian pattern of inheritance when analyzing pedigrees. A number of Mendelian (monogenic or familial) diseases caused by rare (MAF<1%) or private (only present in one or a few families) mutations manifest clinically as ICH. From an epidemiological standpoint, Mendelian forms of ICH represent only a small fraction of the total number of cases of this condition, tend to appear at a younger age, and affect Whites more often than other ethnic groups. Below, we describe two stereotypic examples of Mendelian ICH:

Familial cerebral amyloid angiopathy (CAA)

Several rare mutations resulting in familial CAA and lobar ICH have been described.¹¹ ICH related to familial types of CAA appear at younger ages and have a severer course and an earlier age of death. APP, the beta-amyloid precursor protein gene, is usually the affected gene, with mutations clustering around the A-beta-coding region (exons 16 and 17). Private mutations in other genes can also lead to familial CAA, including CST3, BRI, TTR, ITM2B, GSN, PRNP and ITM2B.³³

COL4A1-related intracerebral hemorrhage

Private mutations within this gene cause autosomal dominant syndromes manifesting with perinatal intracerebral hemorrhage and porencephaly, adult-onset ICH, microbleeds, lacunar strokes, and leukoaraiosis.³⁴ The mechanism linking COL4A1 mutations to ICH involves an alteration of the structural properties of basal membranes caused by an inhibition of collagen heterotrimers deposition.

Clinical implications

Early identification of familial forms of ICH caused by rare mutations that produce Mendelian patterns of inheritance is paramount to provide appropriate genetic counseling to patients planning to have children.³⁵ Likewise, for these familial forms, identification of the small vessel disease that leads to ICH may become important when making management decisions, for example when estimating the risk of brain bleeds when prescribing anticoagulant or antiplatelet therapies.³⁶ While no definite clinical application is currently available for common, incompletely penetrant mutations, several efforts are underway to evaluate whether the epsilon variants within APOE could play a role in precision medicine strategies aimed to stratify patients based on their prospective risk of sustaining an ICH.

Future directions

While rapid advancement has taken place using available genome wide association data, the field of ICH genetics faces several challenges. Perhaps foremost is the scale of ongoing analyses, currently in the thousands of cases, with limited discovery power for novel susceptibility risk loci and an even more limited ability to tackle the critically important evaluation of gene-gene and gene-environment interactions.³⁷ In comparison, for ischemic stroke, tens of thousands of samples are being analyzed, leading to a substantial increase in the number of susceptibility loci identified for this type of stroke. Another important challenge is the study of ICH genetics in disproportionately affected minority populations, which may have different risk factor profiles, particularly when considering the location of the hemorrhage.

Future studies will likely use sequencing technologies.³⁸ Rather than ascertaining pre-selected SNPs, these technologies yield complete (continuous) reads of DNA, either for coding regions (whole-exome sequencing) or the entire genome (whole-genome sequencing). This level of detail permits analysis of rare variants, usually through the calculation of the “risk burden.” Because each rare variant occurs in only a small proportion of cases, the risk burden expresses the cumulative burden of rare genetic variation across a gene or region of interest, subsequently becoming the tested exposure.³⁹ Sequencing strategies also allow the evaluation of RNA, including gene expression, splicing anomalies, gene silencing through methylation, microRNA, transfer RNA and ribosomal RNA.⁴⁰ It should be nonetheless noted that RNA studies require the collection of new samples to be used in by dedicated RNA processing pipelines.