Copy Number Variation in Bronchopulmonary Dysplasia

TO THE EDITOR

Two twin studies [Bhandari et al., 2006; Lavoie et al., 2008] have found relatively high heritability (53%–79%) of susceptibility to bronchopulmonary dysplasia (BPD), a severe disorder of the pulmonary and cardiovascular systems in very low birth weight (VLBW) infants. To identify genetic factors underlying BPD we carried out a California wide population-based case-control study (n=1,726) of > 2 million genome-wide markers. We recently reported findings from analyzing the association between individual single nucleotide polymorphisms (SNPs) and BPD in this study [Wang et al., 2013]. Here we evaluate the potential relationship between copy number variants (CNVs) and BPD, which is important since CNVs have been associated with other suspected heritable disorders (e.g., autism [Glessner et al., 2009]).

Our case-control study identified singleton very low birth weight (VLBW) infant births from the California Perinatal Quality Care Collaborative (CPQCC, http://www.cpqcc.org/) [Gould, 2010], which represents more than 90% of all NICU admissions in California. More detailed methodology is described in our previous publication [Wang et al., 2013]. In brief, inclusion criteria were gestational age (GA) 25⁰–29^6/7 weeks, birth weight (BW) < 1500 grams, and ≥ 3 days mechanical ventilation during their hospitalization up to 36 weeks postmenstrual age (PMA). The ≥ 3 days mechanical ventilation was an inclusion criterion so that both cases and controls would be exposed to this “environmental” factor. NIH/NICHD criteria were used to diagnosemild (supplemental oxygen at 28 days after birth but not at 36 weeks PMA, moderate (supplemental oxygen < 30%) and severe (supplemental oxygen > 30% or positive pressure support) BPD [Walsh et al., 2004; Jobe and Bancalari, 2001]. BPD cases were defined as infants requiring supplemental oxygen or positive pressure ventilator support at 36 weeks PMA whereas control infants were breathing room air at 36 weeks PMA. The practices of each NICU determined the need for supplemental oxygen and physiologic assessments [Walsh et al., 2013] were not routinely carried out. Exclusion criteria included multiple birth, major congenital abnormalities, major surgery (patent ductus arteriosus (PDA) ligation was not excluded), infant death or left hospital prior to 36 weeks PMA, or supplemental oxygen status at 36 weeks PMA not known.

Infants were linked to their newborn screening blood spot. Genomic DNA was extracted from bloodspots [St. Julien et al., 2013] and genotyped (Illumina HumanOmni2.5 beadchip, San Diego, CA). Non-amplified DNA was used and the genotype calls were made using GenomeStudio software [Illumina, 2011] after quality control (QC) procedures [Wang et al., 2013]. After the above steps we successfully genotyped 899 BPD cases and 827 controls (n=1,726). The Institutional Review Board (IRB) of Stanford University and the Health and the Welfare Agency Committee for the Protection of Human Subjects of the State of California approved this study.

For covariates and potential confounders, analyses of our data and findings from the literature indicate that both sex and BW are strong predictors of BPD. To address possible bias due to population stratification, we estimated genetic ancestry using a principal components (PCs) analysis [Wang et al., 2013]. We included in regression analyses of CNVs the first three PCs, self-reported ethnicity, sex, and BW to control for their potential confounding of associations with BPD.

After removing individuals that had incorrectly called information from one data plate (n = 59) and one additional individual who was missing substantial probe intensity information, we analyzed CNV data from 1,666 infants (866 BPD cases and 800 controls). We called CNVs via the software PennCNV [Wang et al., 2007] using GC-wave factor adjustment. We used the following established QC procedures for identifying CNVs. We first removed CNVs that had fewer than 10 SNPs or that were shorter than 50kb in length, and merged large CNVs with a gap between them which was less than 20% of their length. We then removed poor quality samples that had standard deviation of normalized intensity (LRR) > 0.35, B Allele Frequency (BAF) drift > 0.01, number of CNVs > 80, or GC-wave factor (WF) > 0.05. These QC criteria are similar to those previously applied by others when using PennCNV [Glessner et al., 2009, 2010; Need et al., 2009; Davis et al., 2011].

After these steps a total of 21,399 CNVs were called for 1631 individuals (848 BPD cases and 783 controls). Overall there was an average of 13.1 CNVs per infant, which was similar between the BPD cases (13.0) and controls (13.2). A formal test indicated no association between the logarithm of the number of CNVs and BPD (p=0.998, from logistic regression, controlling for ethnicity, sex, BW, and three principal components of ancestry). For BPD cases, CNVs were a median 95.4kb length (range 50.0kb–11,308.7kb) and median 41 SNPs (range 10–4401 SNPs); for controls, CNVs were a median 96.4kb length (range 50.0kb–37,391kb) and median 41 SNPs (range 10–30,938 SNPs). The CNVs comprised four different copy number states: homozygous deletions (CN=0); hemizygous deletions (CN=1); and two possible duplication states (CN=3 or 4). The case/control ratios of the frequency of these copy number states, tested by dichotomizing at greater than the median, were: homozygous deletions, (27 observed in BPD cases / 848 cases) / (18 in controls / 783 controls) = 1.38 (p=0.58); hemizygous deletions, (5,376 / 848) / (5,340 / 783) = 0.93 (p=0.54); duplications (CN=3), (5,579 / 848) / (4,938 / 783) = 1.04 (p=0.85); and duplications (CN=4), (74 / 848) / (49 / 783) = 1.39 (p=0.25).

Focusing on specific CNVs, we evaluated the association between BPD and each SNP, evaluating separately for each SNP as a deletion (CN=0, CN=1, CN=other) and a duplication (CN=4, CN=3, CN=other) as in [Glessner et al., 2013] requiring a count of at least 20 in the minor homozygote plus heterozygote. For each SNP, we evaluated the deletion/duplication frequency between BPD cases and controls with a logistic regression of BPD adjusted for the deletion/duplication, GA, sex, BW, PC₁, PC₂, and PC₃. Then, because of boundary truncation problems, we defined CNV regions (CNVRs) which collapsed SNPs within a 1 MB distance of each other if they were of comparable significance (+/− log 10 P-value) as in [Glessner et al., 2013], forming 74 deletions and 57 duplications. The CNVR results are shown in the Manhattan plot in Figure 1, and the top 5 CNVR results are shown in Table I. No SNP reached the suggested multiple-testing correction for CNVRs of 5×10⁻⁴ [Glessner et al., 2013].

Figure 1.

Manhattan plot of the association of bronchopulmonary dysplasia (BPD) and each copy number variant region (CNVR). The dotted line at 0.0005 indicates the genome-wide CNVR significance level. [Color figure can be viewed in the online issue, which is available at http://onlinelibrary.wiley.com/journal/10.1002/(ISSN)1552-4833.]

Table I.

Top five copy number variant region (CNVR) associated with BPD. None reach genome-wide CNVR significance (0.0005). Genes are given when contained in the CNVR region, and a range between two genes if located in between genes (e.g., POTEA – LINC00293).

CNVR	Gene	Type	Case Freq	Cont Freq	OR	P
chr19: 20601335-20715233	ZNF826P	Del	0.019	0.029	1.71 (1.06, 2.78)	0.0035
chr17: 79925150-79986974	ASPSCR1, LRRC45, STRA13	Dup	0.011	0.0026	0.23 (0.08, 0.70)	0.01
chr22: 50302618-50458513	CRELD2, ALG12, PIM3, IL17REL	Del	0.0077	0.0051	0.67 (0.26, 1.69)	0.012
chr8: 46842124-47467663	POTEA – LINC00293	Dup	0.025	0.022	0.89 (0.55, 1.45)	0.013
chr9: 139243790-139273402	CARD9, GPSM1, DNLZ, SNAPC4	Del	0.0029	0.01	3.37 (1.18, 9.66)	0.014

Del, deletion; Dup, duplication.

Thus, we found no evidence that BPD cases have a larger number of CNVs than controls. We also did not observe particular CNVs that were significantly associated with an increased risk of BPD. The top CNVRs were in different regions of the genome than the top associated SNPs we found in our previous GWAS results. These results, similar to our previous GWAS results, do not point to particular genomic loci as the explanation for the previously described heritability for BPD. This may be due to several reasons, as has been discussed [Wang et al., 2013]. First our study population differs from the twin studies that reported high heritability [Bhandari et al., 2006; Lavoie et al., 2008]. These twin studies did not report the race/ethnicity of the patients, though we speculate that they were Caucasian given the geographic location [Bhandari et al., 2006; Lavioe et al., 2008], in contrast to our cases and controls of predominantly Mexican-Hispanic origin [Wang et al., 2013]. This genetic heterogeneity may reduce power, as different race/ethnicity groups have different prevalence of BPD. Thirdly, our eligibility of cases and controls required ≥ 3 days mechanical ventilation, which not all studies have used. This was chosen to better define the BPD phenotype and decrease the “environmental” differences between cases and controls, as we hoped it would improve our ability to detect genetic factors. However, extremely premature infants who did not require mechanical ventilation sometimes have BPD. Lastly, there may have been unknown differences between the NICUs.