Sequencing of the IL6 gene in a case–control study of cerebral palsy in children

We studied 250 infants with CP and 305 randomly selected control infants born at < = 36 weeks gestation within the Kaiser Permanente Medical Care Program (KPMCP) during1991-2002, as previously described [17]. Briefly, we searched KPMCP records for physician diagnoses of “cerebral palsy,” “paresis,” “gait abnormality,” or “cerebral degeneration.” A single child neurologist (author YWW) reviewed medical records to confirm the diagnosis of CP. We defined CP as a non-progressive congenital motor dysfunction with examination findings of increased tone (spasticity, rigidity, dystonia) or choreoathetosis. Children with hypotonia, ataxia, myopathy, neural tube defect, genetic syndrome, and chromosomal anomaly were excluded. Infants with CP whose blood samples were unavailable for study (47), whose blood samples were taken after having received a blood transfusion (5), or whose blood samples were mislabeled (1), were excluded. A random sample of 305 infants from the study population with available blood samples comprised the control group of this study. All blood samples for DNA purification came from heel-stick dried blood collected in the neonatal period for purposes of newborn screening as part of the California Newborn Screening Program (NBS), unused portions of blood spots are stored in archives at the California Department of Public Health with standard procedure to avoid cross contamination [17]. For this study, punch biopsies of unused portions of blood spots were used for DNA extraction. We received a waiver of consent from the state of California IRB and the IRBs of the individual institutions. To prevent cross contamination between biopsies, the stainless punch was soaked in ethanol, then the ethanol-soaked punch was placed over open flame to burn off residual ethanol and DNA. DNA extraction and amplification was performed using QIAamp 96 DNA kits supplied by Qiagen (Chatsworth, CA), and RepliPHI Phi 29 Reagent Sets (Epicentre Technologies, Madison, WI) as previously described [17]. Ethnicity data was determined from chart review, using pre-defined categories within the KPMCP system: African American (26/22 = Cases/Controls), Asian (26/47), Hispanic (48/51), white (138/163), and others including mixed and unknown (12/22). Study procedures were approved by the institutional review boards at KPMCP, University of California San Francisco, and the California Committee for Protection of Human Subjects.

A combination of long and short amplicons was used to sequence 5000 base pairs covering the IL6 gene starting 300 bases upstream from the transcription start site. In total, 9 long amplicons and49 short amplicons were generated and later sequenced using standard di-deoxy terminator chemistry and capillary electrophoresis technology in an ABI 3730XL instrument (Applied Biosystems, Foster City, CA). Calls were made using Sequencher software (Gene Codes Corporation, Ann Arbor, MI). Genotype calls at important loci of interest were confirmed by manual calls made by two independent operators looking at peaks on an electropherogram using standard procedure. Any discrepant calls were labeled missing (Additional file 2: Table S2). SNP functionality was assessed using the poly-phen2 database [24] and the UCSC genome browser [25]. Hardy-Weinberg equilibrium was measured in controls from each population. In accordance with previous studies, the zero position is defined as the start of the transcribed portion of the gene, with numbers increasing in the direction of transcription, such that the +174 position aligns with rs1800795. Calls were made on the transcribed strand.

Linkage disequilibrium (LD) between the rs1800795 SNP and neighboring SNPs has been described in at least 9 studies, as detailed in Additional file 1: Table S1. The haplotype block of interest is defined differently in each of these studies, making it impossible to define a single consensus haplotype block of interest containing rs1800795. To resolve ambiguity regarding haplotype block definition in IL6, particularly the block containing rs1800795 (Additional file 1: Table S1), we began by investigating haplotype structure in healthy controls from each population. After reproducing the association between CP and rs1800795, we performed focused tests for association between CP and the haplotype containing rs1800795, followed by tests for association between CP and likely functional variants. Finally, we performed an unbiased search for association between CP and all other variants and haplotype blocks. For all association testing, the four populations were pooled; for association testing for common variants, population subgroup analysis was also performed. The R base package was used for statistical analysis except as stated. To elucidate the effect of known CP-associated SNPs and haplotypes, two simultaneous tests were performed. The effect of a single copy of variant allele was compared to wildtype using a Fisher’s exact test on the 2 × 2 table (heterozygote vs. homozygote wildtype), and the effect of two copies of the variant allele was measured using a Fisher’s exact test on a 2 × 2 table (homozygote variant vs. homozygote wildtype). This two-part test has the ability to capture a completely dominant model of inheritance (where both tests will be positive), a completely recessive model of inheritance (where only the second test is positive), a pure dose model of inheritance (where the effect size of the first test will be half the effect size of the second test), or any other model within the spectrum. Screening for other variants in IL6 that have not been previously described was performed as follows. For common variants, a “genotypic” test was performed on the 2 × 3 table of genotype versus case status, as well as an allele frequency test both using Plink software. For rare variants, the allele frequency test was performed, but the “genotypic” test was replaced by a “dominant” test on a 2 × 2 table of genotypes ((homozygous variant + heterozygote) vs. wildtype). Logistic regression was also performed, yielded similar results, and is not shown.

Variants were classified into haplotype blocks in each of the four large populations as follows. Pair-wise linkage disequilibrium between variants was calculated using r². Variants were re-ordered by unsupervised hierarchical clustering using 1-r² for distance and the Ward [26] algorithm of agglomeration. Branches of the dendrogram defined groups of non-sequential SNPs were grouped into haplotype blocks with intra-haplotype r² < 0.3 (Additional file 3: Figure S1). In most cases, long sequencing reads provided haplotype (phased genotype) information for any given individual. When phasing could not be determined by direct sequencing results, a maximum likelihood model was used to infer haplotypes. An individual’s genotype at the haplotype block is defined on each chromosome. A haplotype is called wildtype if all SNPs called within the haplotype are wildtype and at least half of the SNPs within the haplotype are called. A haplotype is called variant if all SNPs within the haplotype are variant and at least half of the SNPs within the haplotype are called. Hardy-Weinberg equilibrium for the haplotype was measured in controls from each population.

In total, 61 variable loci were identified in the dataset (Figure 1, Additional file 4: Table S3). Fifteen of these were common polymorphisms in the population (defined by minor allele frequency (MAF) < 5% in controls), including previously described SNP rs1800795 (−174G < C) [17]. Uncommon variants (allele frequency less than 5%) included two frameshift mutations. One frameshift mutation at position 316 likely disrupts gene function by generating a stop codon near the 5′ end of the first exon of the gene, effectively removing nearly all protein domains. The other frameshift mutation at position 2494 is unlikely to be damaging because it occurs at the 3' end of the last exon of some transcript variants and does not affect other transcript variants.

Variants were grouped into haplotype blocks in each population using r² as a measure of linkage disequilibrium, using Ward hierarchical clustering to re-order SNPs. The greatest linkage disequilibrium (LD) was seen in whites, followed closely by Hispanics and Asians, with African Americans showing the least LD (Additional file 3: Figure S1). In white controls, five haplotype blocks were identified with intra-block r² < 0.3 and the inter-block r² < 0.3 (Additional file 3: Figure S1). These five haplotype blocks included 18 of 61 variants. The rs1800795 SNP was included in a 7-SNP haplotype block: rs1800795 (−174), rs2069832 (615), rs2069833 (846), rs1474348 (1090), rs1474347 (1306), rs1554606 (1889), and rs2069845 (3331). In Hispanic controls, three haplotype blocks were identified, which included 13 variants. These blocks were identical to three of the five haplotype blocks identified in white controls. This includes the same 7-SNP haplotype block for rs1800795. In Asian controls, three haplotype blocks were identified including 11 variants. One of these blocks was again the same 7-SNP haplotype of rs1800795. Another block was identical to one found in Hispanics and whites. The third block found in Asians was a smaller version of the other block found in Hispanics and whites. Finally, in African Americans, five blocks were identified containing 12 SNPs. This time the rs1800795 SNP was part of a smaller 4-SNP haplotype, a subset of the 7-SNP haplotype found in the other three populations. The other four blocks found in African Americans included one block that was identified in all populations (1401 and 1431) and three novel blocks.

The 7-SNP block, which includes rs1800795 and was identified in whites, Hispanics, and Asians, includes an intronic SNP rs20698455 that disrupts a C_pG methylation [27] along with 5 other intronic SNPs. The smaller 4-SNP block of rs1800795 identified in African Americans is a subset of the 7-SNP block but does not include the rs20698455 SNP. Overlap between the7-SNP haplotype and previously reported haplotypes is shown in red text in Additional file 1: Table S1.

The rs1800795 SNP was identified as part of a 7-SNP haplotype block in whites, Hispanics, and Asians, and as part of a smaller 4-SNP block in African Americans. Among control infants, the variant haplotype allele was more common among whites (26% and 27% for the 7-SNP and 4-SNP haplotypes respectively) than among Hispanics (10% and 13%), Asians (7% and 7%), or African Americans (4% and 5%). There was no evidence of Hardy-Weinberg disequilibrium for either the haplotypes or for any of the SNPs in controls of each population.

The association between the 7-SNP haplotype and CP was stronger than the association seen between any individual SNP and CP, or between the 4-SNP haplotype and CP. Five of seven SNPs in this block showed significant association with CP with a recessive pattern of inheritance (Table 1). The odds ratio (OR) of association for homozygote versus wildtype at individual SNPs was at most 2.7. However, the OR for homozygotes for the variant at the full 7-SNP haplotype block variant was 4.3 (95% confidence interval (CI) = [2.0-10.1], Fisher exact test p = 0.00007, Bonferroni corrected p = 0.0004). The OR for homozygotes at the smaller 4-SNP haplotype was 3.0 (CI = [1.5-6.0], p = 0.0007, corrected p = 0.004). There was no evidence of increased risk for heterozygous carriers of any of these variant SNPs or for heterozygous carriers of the variant 4-SNP or 7-SNP haplotype.

Population subgroup analysis of the 7-SNP haplotype block revealed a significant odds ratio for homozygous white (OR = 3.18, CI = [1.3-8.4], one-sided p = 0.007) and Hispanic infants (OR = 8.02, CI = [0.9-400], one-sided p = 0.045), with no evidence of increased risk for heterozygotes (Figure 2). Only two African American infants and one Asian infant were homozygotes; all three had CP. Population subgroup analysis of the 4-SNP haplotype block showed a similar pattern across populations, though effect sizes were proportionally reduced in relation to the 7-SNP haplotype (not shown). Subgroup analysis revealed significant odds ratios in males (OR = 4.6, CI = [1.6-15], one-sided p = 0.0006) and females (OR = 3.04, CI = [0.9-10.9], one-sided p = 0.03), and across body weights (there were no case or control homozygote variants with birth weight less than 2500 g).

As described above, one of the rare variants identified resulted in a frameshift at position 316 within the first exon and caused early protein truncation, thus warranting a focused analysis of this variant. Within this case–control dataset, the frameshift allele was found in four Asian infants (allele frequency = 2.5%), two Hispanic infants (allele frequency = 1%), three white infants (allele frequency = 0.6%), and one infant of other ethnicity. The lone homozygote was white. In total, eight of ten carriers of the frameshift allele were infants with CP (OR = 4.53, CI = [0.9-44], p-value = 0.053) including the lone homozygote carrier. Clinical characteristics of the eight infant frameshift carriers are shown in Table 2. Of note, three of eight carrier infants had perinatal stroke (versus 20.2% of non-carriers), including the lone homozygote carrier, though this result was not statistically significant. To control for possible confounding effects, individuals homozygous for the variant haplotype allele described above were excluded in the analysis of the frameshift mutation.

Finally, we carried out an unbiased analysis of the fifteen common variants identified in this population. When corrected for multiple hypotheses testing using the method of Bonferroni, we found no evidence of association (p < 0.05) with CP for any variants other than those described above. This was true for the entire dataset (Additional file 5: Table S4) and also for each population subgroup (not shown). Of the 46 rare variants, 20 were found in more than one individual. Unbiased analysis of these 20 rare variants using a dominant model revealed no significant association after correction for multiple testing (Additional file 6: Table S5). We deemed this analysis underpowered due to the small sample size and the large number of hypotheses tested.