Genome-Wide Association Study Identifies Genetic Risk Factors for Spastic Cerebral Palsy

Abstract

BACKGROUND

Although many clinical risk factors of spastic cerebral palsy (CP) have been identified, the genetic basis of spastic CP is largely unknown. Here, using whole-genome genetic information linked to a deidentified electronic health record (BioVU) with replication in the UK Biobank and FinnGen, we perform the first genome-wide association study (GWAS) for spastic CP.

OBJECTIVE

To define the genetic basis of spastic CP.

METHODS

Whole-genome data were obtained using the multi-ethnic genotyping array (MEGA) genotyping array capturing single-nucleotide polymorphisms (SNPs), minor allele frequency (MAF) > 0.01, and imputation quality score (r²) > 0.3, imputed based on the 1000 genomes phase 3 reference panel. Threshold for genome-wide significance was defined after Bonferroni correction for the total number of SNPs tested (P < 5.0 × 10^–8). Replication analysis (defined as P < .05) was performed in the UK Biobank and FinnGen.

RESULTS

We identify 1 SNP (rs78686911) reaching genome-wide significance with spastic CP. Expression quantitative trait loci (eQTL) analysis suggests that rs78686911 decreases expression of GRIK4, a gene that encodes a high-affinity kainate glutamatergic receptor of largely unknown function. Replication analysis in the UK Biobank and FinnGen reveals additional SNPs in the GRIK4 loci associated with CP.

CONCLUSION

To our knowledge, we perform the first GWAS of spastic CP. Our study indicates that genetic variation contributes to CP risk.

ABBREVIATIONS

CP

cerebral palsy

EHR

electronic health records

eQTL

expression quantitative trait loci

GMFCS

Gross Motor Function Classification Systems

GTEx

genotype-tissue expression

GWAS

genome-wide association study

ICD

International Classification of Disease

MAF

minor allele frequency

MEGA

multi-ethnic genotyping array

NES

Normalized enrichment score

SD

synthetic derivative

SNPs

single-nucleotide polymorphisms

TPM

transcripts per million

Cerebral palsy (CP) is a heterogeneous disorder characterized by impairment of motor function, spasticity, and developmental delay.¹ Additional comorbidities of CP include deficits in sensory perception, cognition, and behavior, highlighting the phenotypic spectrum of the disease.² The spastic CP subtype is the most common motor disorder in children, with estimates ranging from 1.5 to 4 per 1000 children born in the United States.³ Although many clinical risk factors have been associated with spastic CP, including prematurity, perinatal hypoxic-ischemic injury, and maternal infection, among others,² the role of genetic variation is less understood.⁴ Although numerous genome-wide association study (GWAS) have been performed for related CP traits such as autism and epilepsy,^5,6 there is a paucity of genetic data and genome-wide analysis of spastic CP.

Numerous approaches to identify rare de novo mutations, copy number changes, or hypothesis-driven candidate gene analysis have revealed some insight into the genetic basis of CP,^7-9 but these studies are limited by relatively small sample sizes and targeted approach. Furthermore, CP subtype-specific genetic architecture,¹⁰ with the exception of select findings for the ataxic subtype of CP,^11-13 has been difficult to elucidate largely because of relatively small and heterogeneous cohorts. To overcome these limitations, we perform a GWAS of spastic CP using samples obtained from BioVU,¹⁴ a DNA

Biobank linked to a deidentified electronic health records (EHR) system. Although additional human genetics studies and functional mechanistic validation are needed, our study lays the groundwork for genome-wide approaches to understand the role of genetics in conferring spastic CP risk.

METHODS

Patient Cohort Description

Demographic, clinical, pharmacological, and surgical data were collected from the synthetic derivative (SD), the deidentified electronic medical records component of BioVU.¹⁴ Since 2004, BioVU has collected “left-over” blood from routine clinical care to perform large-scale genetics studies. Until 2015, these samples have been accrued on an “opt-out” basis, but now are obtained on an “opt-in” basis. Patient consent was obtained at this time. For ongoing information pertaining BioVU, please refer to this link: https://victr.vumc.org/what-is-biovu/. Policies around the construction, ethics, and infrastructure of BioVU can be found here.¹⁴

Cases of spastic CP were identified using International Classification of Disease (ICD) codes including 343.9 (ICD-9) and G80.9 (ICD-10). Patients with athetoid CP (333.71 or G80.3) or ataxic CP (G80.4) were excluded. Etiology of CP was determined by review of deidentified documentation from at least 2 independent clinical encounters with a neurologist or neurosurgeon. Patients with diplegic, hemiplegic, quadriplegic, or tetraplegic forms of CP were included. If available, the radiologist's report, but not primary imaging study (because of BioVU policies related to patient privacy),¹⁴ was also reviewed. Cases of CP secondary to traumatic brain injury, nonaccidental trauma, or stroke were also excluded based on a review of clinical documentation and radiology reports. Patients with a diagnosis of primary dystonia or any other primary movement disorder were excluded. The following variables were collected from the SD: (1) corrected gestational age; (2) history of premature birth (<37 wk); (3) history of hypoxic-ischemic injury during birth (defined as radiographic features of hypoxic-ischemic encephalopathy on magnetic resonance imaging determined by the final radiologist interpretation; cases of acute ischemic stroke were excluded); (4) history of neonatal sepsis (defined as positive blood culture and/or unstable vital signs and clinical suspicion of sepsis, and/or C-reactive protein elevation over 10 mg/L, with 5 or more days of broad-spectrum antibiotics); (5) history of neonatal meningitis (defined as positive cerebrospinal culture or viral panel and clinical suspicion of meningitis, with 14 or more days of broad-spectrum antibiotics); (6) Gross Motor Function Classification Systems (GMFCS) scores; and (7) whether or not the patient received surgery for treatment of CP (baclofen pump placement or dorsal rhizotomy). Institutional review board approval was granted from Vanderbilt University.

Genome-Wide Association Study Analysis

We used BioVU,¹⁴ a deidentified health record linked to genetic data, to identify 604 patients with spastic CP and 9798 unaffected control individuals. Whole-genome data were obtained using the MEGA array using the 1000 genome reference panel,¹⁵ minor allele frequency (MAF) > 0.01. GWAS was performed using PLINK v1.90b6.7¹⁶ and ANNOVAR.¹⁷ Age, sex, and the first 5 principal components were included as covariates in logistic regression models to identify CP-associated single-nucleotide polymorphisms (SNPs). Analysis was restricted to individuals of European ancestry. Threshold for genome-wide significance was defined after Bonferroni correction for the total number of SNPs tested (P < 5.0 × 10^–8). Assuming a disease prevalence for spastic CP of 4 in 1000,³ rs78686911 allele frequency of 0.04, and genotype relative risk of ∼2.4, our study is appropriately powered to 0.84. STREGA reporting guidelines for genetic association studies were followed.¹⁸

GWAS Replication in the UK Biobank and FinnGen

Replication in FinnGen¹⁹ was performed using Illumina and Affymetrix chip arrays. Imputation was performed using the SISu v3 imputation reference panel of 3775 whole genomes. SNP associations were determined after controlling for age, sex, 10 principal components, and genotype batch as covariates. Cases for replication were assigned the phenotype “Cerebral palsy and other paralytic syndromes” (G6_CPETAL, 1025 cases and 134 613 controls; http://r3.finngen.fi/pheno/G6_CPETAL). For additional information, see https://finngen.gitbook.io/documentation/. Despite the larger sample size than our discovery cohort, this cohort was chosen as replication because the cases were not selected based on scrutinous interrogation of the medical record. Independent replication in the UK Biobank²⁰ was performed using Illumina HumanCoreExome array genotyping and imputed using the Haplotype Reference Consortium panel (46 cases and 15 582 controls; http://pheweb.sph.umich.edu/pheno/343). Associations were determined using saddlepoint approximation, adjusted for age, sex, and principal components 1 to 4. We considered infantile CP (Phecode 343, phewascatalog.org) as the replication phenotype. Statistical threshold for replication was defined as P < .05.

Genotype-Tissue Expression Project Analysis

The latest data release of genotype-tissue expression (GTEx) (version 8) was used.²¹ The GTEx portal (gtexportal.org) was used for data visualization. GRIK4 expression across tissues was assess using transcripts per million (TPM), which is calculated from a gene model where isoforms are collapsed to a single gene without additional normalization.²² The box plots represent the median and the 25th and 75th percentile and outliers were defined as above/below 1.5 times the interquartile range. Expression quantitative trait loci (eQTL) analysis was performed as previously described.²¹ Normalized enrichment score (NES) was used to determine the magnitude of effect.²¹ The m-value is the posterior probability from MetaSoft, which is the probability than an eQTL effect exists in the tested cross-tissue analysis.^23,24 An m-value < 0.01 is highly suggestive that an eQTL effect does not exist, whereas an m-value > 0.9 is highly predictive of an eQTL effect. All data visualization was performed using gtexportal.org.

RESULTS

We identified 604 cases of spastic CP and 2558 control individuals with whole-genome genetic information linked to a deidentified electronic health records (EHR) system (BioVU).¹⁴ We collected relevant CP risk factors information for descriptive purposes including premature birth (<37 wk, 38% of our cases), history of maternal infection (6% of our cohort), hypoxic-ischemic injury at birth (38% of our cases), history of neonatal sepsis (14% of our cases), history of neonatal meningitis (6% of our cases), and surgical treatment for spastic CP (baclofen pump placement or rhizotomy, 20% of our cases).^25,26 There was no statistically significant difference (P > .05) in these clinical risk factors between our case and control cohorts. These data are listed in Table 1.

TABLE 1.

Patient Cohort Description of Spastic CP Cases (n = 604)

	Spastic CP (n = 604)
Demographic
Premature birth (<37 wk)
No	189 (31%)
Yes	227 (38%)
Unknown	188 (31%)
Maternal infection
No	263 (44%)
Yes	36 (6%)
Unknown	305 (50%)
Hypoxic injury
No	180 (30%)
Yes	232 (38%)
Unknown	192 (32%)
Neonatal sepsis
No	337 (56%)
Yes	82 (14%)
Unknown	185 (30%)
Neonatal meningitis
No	380 (63%)
Yes	39 (6%)
Unknown	185 (31%)
GMFCS
1	2 (<1%)
2	3 (<1%)
3	6 (<1%)
4	27 (4%)
5	81 (13%)
Unknown	485 (80%)
Surgery (baclofen pump or rhizotomy)
No	479 (79%)
Yes	125 (21%)

GMFCS, Gross Motor Function Classification System.

Controlling for age, ethnicity, sex, and 5 principal components, we identify a singleton SNP reaching genome-wide significance with spastic CP (rs78686911; chromosome 11; Figure 1A and 1B). The imputation quality score (r²) for rs78686911 is strong (>0.8) and the MAF for this variant on the MEGA array (0.02755, across 90 813 total individuals) is very similar to that seen in the GnomAD (0.02226) and 1000 Genome (0.0106) databases. All carriers for rs78686911 were heterozygous for this SNP (C/T); no homozygous individuals (T/T) were identified. As a quantile-quantile (Q-Q) plot demonstrates, population stratification was sufficiently controlled (Figure 1C). This SNP is a singleton variant, as it is not in significant linkage disequilibrium (r² > 0.8) with any other SNPs in the International Genome Sample Resource, an updated assembly of the 1000 genomes project.¹⁵ Carrier status for rs78686911 was not disproportionately observed (P > .05) among patients with a history of hypoxic-ischemic injury, premature birth, maternal infection, or neonatal infection. Furthermore, rs78686911 genotype was not correlated with disease severity as determined by GMFCS score or need for surgery.

FIGURE 1.

A, Manhattan plot of GWAS for spastic CP (604 cases, 2558 controls). Individual dots represent SNPs associated with spastic CP, MAF > 0.01. The horizontal red line represents the threshold for genome-wide significance correcting for the total number of SNPs tested, P = 5.0 × 10^–8. The horizontal blue line represents a lower threshold for statistical significance set at P = 1.0 × 10^–5. B, LocusZoom plot of chromosome 11q23.3 containing the singleton variant rs78686911. C, Quantile-quantile (Q-Q) plot of CP-associated SNPs.

rs78686911 is located within an intronic region the gene glutamate ionotropic receptor kainite type subunit (GRIK4) according to the most recent 1000 genomes (phase 3) data release.¹⁵ Thus, to delineate the potential impact of rs78686911 on GRIK4 expression, we performed eQTL analysis. In fact, rs78686911 is associated with decreased expression of GRIK4 by eQTL analysis in (appropriately powered) thyroid tissue (n = 574 samples, Figure 2A, P = 2.2 × 10^–5, normalized effect size = –0.34) and among select brain tissues (Figure 2A). In fact, GRIK4 is highly expressed at the gene and exon levels across brain regions compared to other tissues available in GTEx (Figure 2B and 2C) and GRIK4 brain-tissue expression clusters with thyroid expression (Figure 2C), suggesting potentially similar regulatory mechanisms. However, additional studies are needed to delineate the precise mechanism by which rs78686911 contributes to CP risk.

FIGURE 2.

A, Single tissue level eQTL analysis for rs78686911 in GTEx (version 8). NES = normalized enrichment score. The m-value is the posterior probability from MetaSoft, which is the probability than an eQTL effect exists in the tested cross-tissue analysis.^23,24 An m-value < 0.01 is highly suggestive that an eQTL effect does not exist, whereas an m-value > 0.9 is highly predictive of an eQTL effect. Expression of GRIK4 across tissues available in GTEx (version 8). Gene B and exon C level expression of GRIK4 across tissues available in GTEx (gtexportal.org). TPM indicates TPM calculated from a gene model with isoforms collapsed to a single gene without additional normalization.²² The box plots represent the median and the 25th and 75th percentile. Individual points are marked as outliers if they are above/below 1.5 times the interquartile range. Data visualized were performed using gtexportal.org.

To strengthen our rationale that alteration in GRIK4 confers risk of CP, we sought independent GWAS replication of GRIK4 in the UK Biobank and FinnGen consortia (Table 2).^19,20 In the UK Biobank, we identify 3 SNPs near GRIK4: rs111800793 (MAF = 0.034, P = 2.3 × 10^–4), rs75195561 (MAF = 0.019, P = 3.5 × 10^–4), and rs17123976 (MAF = 0.019, P = 3.8 × 10^–4) associated with infantile CP (Phecode 343). In addition, we identify a locus (chromosome 11, position 120 858 021) located within an intron of GRIK4 (OR = 7.4 + 0.51, P = 9.6 × 10^–5) associated with CP and other paralytic syndromes in FinnGen. Of note, this locus is the most significant GWAS finding for this phenotype. In addition, rs78686911 is not in linkage disequilibrium (r² >0.8) with any of the variants in these loci. Collectively, these data suggest that variation in the GRIK4 loci confers risk to CP in independent populations.

TABLE 2.

Replication of SNPs Within the GRIK4 Locus in the UK Biobank and FinnGen Consortia

Phenotype	Biobank	Variant	MAF	P value
Infantile CP	UK Biobank	11:120399111 A > T (rs111800793)	3.4 × 10–2	2.3 × 10–4
Infantile CP	UK Biobank	11:120397891 C > T (rs75195561)	1.9 × 10–2	3.5 × 10–4
Infantile CP	UK Biobank	11:120396329 T > C (rs17123976)	1.9 × 10–2	3.8 × 10–4
CP and other paralytic disorders	FinnGen	11:120858021	2.7 × 10–3	9.6 × 10–5

DISCUSSION

To our knowledge, we perform the first GWAS of spastic CP and identify rs78686911 reaching genome-wide significance (P = 5.0 × 10^–8). rs78686911 is a common genetic variant (MAF = 0.04) and is located within an intronic region of GRIK4, a high-affinity kainate glutamatergic receptor of largely unknown function.²⁷ eQTL analysis suggests that rs78686911 decreases expression of GRIK4. Our data highlight the role of genetic variation in conferring risk to spastic CP.

Although rs78686911 is predicted to decrease GRIK4 expression by eQTL analysis, functional studies are needed to determine causality. Nonetheless, based on what is known about GRIK4, dysfunction of this gene could reasonably underly the pathogenesis of CP. Deletion of GRIK isoforms in mice results in hindlimb clasping, contracture, and motor deficits,²⁸ recapitulating phenotypes observed in CP patients. Overexpression of forebrain-specific GRIK4 in mice has been shown to alter synaptic activity and connectivity, features that are hypothesized to underlie CP pathophysiology²⁹ and suggest potential mechanistic consequences of GRIK4 dysfunction in spastic CP.^30,31 Intriguingly, in addition, gene duplication within the rs78686911 region (11q23.3-q24.1) has been associated with autism,³² a disease with overlapping features of CP and is highly comorbid with spastic CP.³³ Furthermore, variants within the GRIK4 locus reaching genome-wide significance have also been associated with decreased educational attainment, decreased self-reported math ability, and inversely correlate with highest level of math completed,³⁴ all comorbidities associated with CP.

Our current understanding of CP genetics has been hindered by inconsistent reporting, unclear selection criteria, and lack of appropriate clinical risk-factor control information. These limitations are perhaps why replication of genetic findings across studies has been tenuous.³⁵ Various approaches have been taken to understand the genetic etiologies of CP, including familial studies,³⁶ copy number/single-gene analysis,¹⁰ and whole-exome sequencing.^7,37 For example, the APOE4 allele, most widely known for its association with increased risk of Alzheimer disease, has been associated with CP.^38,39 However, the best evidence for this association comes from a small cross-sectional study (255 cases) in Norway,⁴⁰ a genetically homogeneous region largely due to geographic isolation and limited gene flow.⁴¹ Thus, it perhaps is not surprising that independent studies in much larger cohorts in Australia and China populations did not reproduce these findings.^42,43 Here, we overcome variability associated with CP diagnosis by performing genetic analysis linked to EHRs information, significantly reducing the possibility of tenuous case-control definitions, disparities, and barriers associated with self-referral studies common in the genetics literature.⁴⁴ Because our study was performed in patients of European ancestry, future genetic studies across diverse populations are needed to further resolve the genetic architecture of CP.

It is being increasingly appreciated that complex neurodevelopmental disorders are highly polygenic and that common genetic variants (MAF > 0.01) play a significant role and even exceed the contribution of rare variants.⁴⁵ Thus far, efforts to understand the role of genetics in CP have largely focused on identifying rare deleterious variants (de novo or inherited) with large effect sizes, as well as identifying copy number variations using whole-exome sequencing.^7,9,46 Most recently, an enrichment of damaging de novo mutations in TUBA1A and CTNNB1 as well as monogenic etiologies (FBXO31 and RHOB) have been described.³⁷ Intriguingly, pathway enrichment analysis of implicated genes also identified substantial overlap with comorbid neurodevelopmental risk genes, highlighting potential pathogenic similarities among related disorders.³⁷ Although these classic approaches have been successfully deployed to identify causal genes for Mendelian disorders and monogenic forms of CP,^37,47 the genetic architecture of CP may be more complex, polygenic, and involve regulatory regions of the genome. Even in the case of clearly “monogenic” disorders, common variants play a substantial role in phenotypic heterogeneity.⁴⁵ Thus, it is important to note that many variants across the genome are significantly associated with CP (P < 10^–5), but do not reach genome-wide significance (Figure 1). These data suggest that CP is a highly polygenic disorder that likely arises from the summative contribution of many genetic variants, consistent with other complex diseases.⁴⁸ GWAS on larger cohorts are essential to clarifying the role of the genetic variants implicated here and identification of additional risk loci.

Because incorporation of genome-wide genetic information is scalable and feasible in newborn screening,⁴⁹ future genetic models may be used to identify patients most at risk of developing spastic CP leading to opportunities for early intervention. Furthermore, because ∼23% of spastic CP patients in our cohort had undergone formal clinical-genetics evaluations, of which ∼37% had findings of unclear significance and all evaluations were targeted (rather than unbiased), integrating genetic information into the clinical setting may be feasible. However, given the paucity of data on diverse populations due to a preponderance of largely European-ancestry studies (including ours), we must exercise caution in interpreting these data and actively avoid exacerbated further health disparities across diverse populations.⁵⁰

CONCLUSION

To our knowledge, we perform the first GWAS for spastic CP. We report a SNP within GRIK4 (rs78686911) reaching genome-wide significance with spastic CP. eQTL analysis suggests that rs78686911 causes decreased GRIK4 expression. We provide replication of 4 additional variants within the GRIK4 loci in 2 independent biobanks. Elucidating the genetic determinants of spastic CP will lead to pathophysiological insights and potentially new pharmacologic strategies.

Funding

No funders had a role in the design and conduct in the study. Dr Hale is supported by the National Institutes of Health (F30HL143826) and Vanderbilt University Medical Scientist Training Program (5T32GM007347). Support for processing of genetic data is provided by a grant from the Vanderbilt Institute for Clinical and Translational Research (VICTR, VR54461, to Dr Hale). Funding sources for the BioVU resource can be found at https://victr.vanderbilt.edu/pub/biovu/.

Disclosures

The authors have no personal, financial, or institutional interest in any of the drugs, materials, or devices described in this article.