Genome Wide Association Study of Sepsis in Extremely Premature Infants

Abstract

Objective

To identify genetic variants associated with sepsis (early and late-onset) using a genome wide association (GWA) analysis in a cohort of extremely premature infants.

Study Design

Previously generated GWA data from the Neonatal Research Network’s anonymized genomic database biorepository of extremely premature infants were used for this study. Sepsis was defined as culture-positive early-onset or late-onset sepsis or culture-proven meningitis. Genomic and whole genome amplified DNA was genotyped for 1.2 million single nucleotide polymorphisms (SNPs); 91% of SNPs were successfully genotyped. We imputed 7.2 million additional SNPs. P values and false discovery rates were calculated from multivariate logistic regression analysis adjusting for gender, gestational age and ancestry. Target statistical value was p<10⁻⁵. Secondary analyses assessed associations of SNPs with pathogen type. Pathway analyses were also run on primary and secondary end points.

Results

Data from 757 extremely premature infants were included: 351 infants with sepsis and 406 infants without sepsis. No SNPs reached genome-wide significance levels (5×10⁻⁸); two SNPs in proximity to FOXC2 and FOXL1 genes achieved target levels of significance. In secondary analyses, SNPs for ELMO1, IRAK2 (Gram positive sepsis), RALA, IMMP2L (Gram negative sepsis) and PIEZO2 (fungal sepsis) met target significance levels. Pathways associated with sepsis and Gram negative sepsis included gap junctions, fibroblast growth factor receptors, regulators of cell division and Interleukin-1 associated receptor kinase 2 (p values<0.001 and FDR<20%).

Conclusions

No SNPs met genome-wide significance in this cohort of ELBW infants; however, areas of potential association and pathways meriting further study were identified.

BACKGROUND

Sepsis is a potentially life threatening illness, affecting 10–20% of neonates worldwide, with a higher incidence in preterm and very low birth weight infants.^1,2 Up to one-third of extremely premature infants develop sepsis, which carries a high mortality and morbidity.^1–3

Susceptibility to neonatal sepsis is mediated by complex interactions between environmental and maternal factors and the neonatal immune response, which may be modified by differences in genetic composition or function.^3–5 Twin studies and ethnic variations suggest that host genetic factors may contribute to susceptibility to sepsis.^4–8 Genetic studies of sepsis in newborns have previously used candidate gene approaches, with varied findings. ^9–14

While genome-wide association studies (GWAS) have been performed in adult sepsis, no GWAS has reported genetic variation in risk of sepsis in extremely preterm infants.^15–17 An agnostic GWAS approach rather than a study on known variant candidate genes has potential to identify gene variants across many genes (1 million or more variants), including disease associations with previously unknown genes. In addition, investigators do not possess sufficient knowledge of the function of all genes to predict candidate genes well. The high rate of sepsis in extremely low birth weight (ELBW) neonates establishes them as a unique population for unbiased study of genetic associations with sepsis.⁴

METHODS

Study population and data source

We performed a secondary analysis utilizing an anonymized DNA biorepository from a prior study of cytokines and neurodevelopmental outcomes in ELBW infants conducted by the Eunice Kennedy Shriver NICHD Neonatal Research Network (NRN).^18,19 Enrolled infants were <1000 grams birth weight and <72 hours old at enrollment; infants with major congenital anomalies were excluded. The NRN’s anonymized DNA database and biorepository was previously approved by all site institutional review boards (IRBs) for candidate gene analyses and genome wide analyses of diseases associated with extreme prematurity, including sepsis.²⁰ We received a waiver of IRB authorization for use of this anonymized data set. In this multi-racial and multi-ethnic study, blood spot samples were collected on more than 800 ELBW infants and analyzed using the Illumina Omni-1 Quad array.²⁰

Primary Outcome

Our primary outcome was the identification of single nucleotide polymorphisms (SNPs) significantly associated with the occurrence of all sepsis in extremely preterm infants.

Secondary Outcomes

Given that EOS and LOS have distinct etiologies and likely differences in mechanisms underpinning their pathogenesis, we planned a priori to identify SNPs significantly associated with late-onset sepsis (LOS) alone.^3,21 Due to the relatively low incidence of early-onset sepsis (EOS), it was not feasible to examine SNPs associated with EOS alone. We also identified SNPs associated with Gram positive, Gram negative, and fungal pathogens. While infants who experienced multiple episodes of sepsis with different bacterial pathogens were included only once in the primary analysis, they were included in each pathogen category that they fit in secondary analyses. Finally, we also reviewed SNPs tested in previous candidate gene association studies in neonatal sepsis.

Study definitions

Cases consisted of infants with one or more episodes of culture-proven sepsis (EOS or LOS) or meningitis, treated with antibiotics for ≥ 5 days, or, who died before treatment was completed.^1,2 Controls were infants not classified as cases. Infants with positive blood or CSF cultures were defined as having culture proven sepsis. EOS comprised infants with culture proven sepsis in the first 72 hours of life, while LOS comprised cases beyond that timeframe. All subjects included in the study were followed through the duration of their hospital course for occurrence of sepsis.

Sample source and storage

Our study utilized DNA data obtained from samples originally collected as part of a prospective multicenter cohort study of cytokines in premature infants (the Cytokine Study, performed in 17 centers of the National Institute of Child Health and Human Development Neonatal Research Network).¹⁸ In the parent study, whole blood spots were obtained on filter paper within 4 hours after birth and at serial timepoints (days 3, 7, 14, 21). These filter paper blood spots were frozen as soon as possible. For our study and other secondary genomic analyses performed utilizing this dataset, DNA was extracted from filter paper blood spots for the earliest age available for each subject, amplified and analyzed using the Illumina Omni-1 Quad array, as described below. Additional details regarding sample methods are available in prior studies utilizing this sample set.^18,20

Genomic analysis

Genomic and whole genome amplified DNA were genotyped on the Human-OMNI1-Quad_v1-0_B BeadChip platform (Illumina, San Diego, California). Initially, 1.2 million SNPs were tested, of which over 900,000 (91%) were successfully genotyped. We employed standard quality control thresholds, and included samples and SNPs with > 97% call rate, while excluding minor allele frequencies < 1%. Additionally, 7.2 million SNPs were imputed against the HapMap 3 cosmopolitan reference panel with Impute2.²⁰ All SNP-gene assignments were made based on the UCSC gene tracks within the GRCh37/hg19 genome build.^22,23 Genotypes, and a subset of subject information inclusive of key characteristic and outcome phenotype data have been stored, with no identifying links, in the National Human Genome Research Institute Database of Genotypes and Phenotypes (dbGaP Study Accession: phs000353.v1.p1).^19,20.

Data Analysis

Following standard quality checks, single and multiple logistic regression models were developed for the primary and secondary analyses. Covariates included in the multiple regression model included gestational age, small for gestational age (SGA) status, ancestry, surgery, C-section, and the first four eigenvalues for ancestry (described below). In previous analyses of this dataset, these covariates were determined to be the most important predictors of sepsis.⁸ To minimize false discoveries, we established a conservative significance level of 10⁻⁵ as the ‘target’ for reporting, while a p value of 5×10⁻⁸ is regarded as a “genome-wide” association. The latter value is widely accepted to be the threshold in genome wide association studies that would adequately adjust for the multiple comparisons involved, and is estimated to correspond to a p value of 0.05 in a classical epidemiological study testing a single hypothesis.^22–25

We used this two-tiered approach in order to screen for genes of interest, which may be missed if only the more stringent threshold were applied. When SNPs met threshold significance, we did not limit reporting based on FDR thresholds, in order to better identify any genes of putative biologic importance that should be examined closely in future validation studies. ²⁵ P values were calculated per SNP for a one degree of freedom additive model. For the secondary analysis of previously identified candidate SNPs, we reported p values <0.05 as trends of interest.

Correction for population stratification

Spurious associations may be obtained, or true associations overlooked, if allele frequencies differ between subpopulations and are represented unequally between cases and controls due to systematic ancestry differences. ^26–28 The first four eigenvalues derived from linkage disequilibrium pruned for genotyped SNPs were included as covariables in primary and secondary analyses, to adjust for individual ancestry (eigenvalues generated using GWAS-Tools). ²⁰

Exploratory pathway analysis

We assigned genes to biological pathways using the Reactome database (www.reactome.org), using the adaptive rank truncated product (ARTP) method (pathway tools package ARTP).^29–31 SNPs were notated to genes based on their location within 50 kb of the gene models defined in Refseq, and p-values assigned based on the results of the GWA analysis. A summary of the association between each gene and sepsis was generated, based on the multiple-testing adjusted P-value associated with the most significant SNP within that gene. The ARTP method was used to combine gene-level P-values into a test statistic for the pathway association with sepsis.³¹ Given the exploratory nature of this analysis, we designated pathways with p values <0.005 as the cut-off for reporting, and reported false discovery rates (FDR) of less than 20% to facilitate assessment of the strength of the evidence.²⁵

RESULTS

Data from 751 ELBW infants passed quality control checks and were included in the analysis. This cohort included 345 infants with sepsis (Supplementary Table 1). Most infants with sepsis had experienced LOS (320 infants, 393 episodes). Ninety-nine infants experienced more than one episode of sepsis. Gram positive pathogens were present in more than three-fourths of the LOS events (252 episodes), while Gram negative and fungal sepsis occurred with lower frequency (72 and 69 episodes respectively, Supplementary Table 1). Infants could be positive for more than one organism per sepsis event. Of the infants with sepsis, 42 were diagnosed with culture proven meningitis. Eighty-one infants in the overall cohort developed necrotizing enterocolitis (NEC), including 40 cases of surgical NEC. NEC events occurred more frequently in infants with sepsis compared to uninfected infants (14.1% versus 8.7%, p = 0.02, Chi square test); however the dataset lacked information on timing of NEC in relation to sepsis event.

Primary Outcome

In the primary analysis of overall sepsis, no SNPs reached a priori genome-wide significance levels (5×10⁻⁸); however, several achieved p values of 10⁻⁵ (Table 1). While these SNPs were generally noted to be in intergenic regions, several were located on chromosome 16, in proximity to genes for FOXC2 (forkhead box protein C2) and FOXL1 (forkhead box protein L1) (Supplementary Figure 1). The analysis limited to LOS demonstrated similar results.

Table 1.

SNPs with highest association with all sepsis^*

Chr	SNP	Location**	P value	Gene
16	rs13380717	86904135	1.08E-07	Intergenic
16	rs9933616	86903194	3.64E-07	Intergenic
16	rs59876150	86918734	6.80E-07	Intergenic
16	rs34528289	86907297	6.95E-07	Intergenic
4	rs2412930	59588228	1.16E-06	Intergenic
4	rs6837629	59583956	1.49E-06	Intergenic
6	rs9456883	1.64E+08	1.63E-06	Intergenic
2	rs41461846	2.19E+08	1.93E-06	Intergenic
2	chr2:219344165	2.19E+08	1.99E-06	Intergenic
2	rs6717433	2.19E+08	2.04E-06	Intergenic

^*Top ten SNPs for each analysis listed; covariates in analysis: gestational age, SGA, eigenvalues for ancestry, surgery, and Cesarean section; infants who died before 72 hours excluded from controls

^**Location in Base pairs from p telomere

Secondary Outcomes

Examining Gram positive LOS only, SNPs that achieved target level of associations (<10⁻⁵) included introns for ELMO1 (engulfment and cell motility 1) on chromosome 7, and introns for VIL1 (villin 1, chromosome 2) (Table 2, Supplementary Figures 2a–b). For Gram negative LOS, SNPs in introns for RALA (Ras-like protein A) and IMMP2L (inner mitochondrial membrane peptidase-2 like), located on chromosome 7, and several SNPs in the intergenic area of chromosome 2 achieved the highest levels of association (Table 3). In the analysis of fungal LOS, over-represented SNPs included introns for PIEZO2 (piezo-type mechanosensitive ion channel component 2, chromosome 18), and several SNPs in the intergenic area of chromosome 10 (Table 4). In the analysis of previously tested candidate gene SNPs, we identified 62 SNPs from previous gene association studies of neonatal sepsis. Several SNPs demonstrated p values <0.05, but none reached target or genome wide levels of significance (Supplementary Table 2).

Table 2.

SNPs with highest association with Gram positive LOS^*

Chr	SNP	Location	P value	Gene	Gene function
7	rs6462728	37071352	8.42E-07	Intron ELMO1	Encodes cell motility and engulfment protein. Interacts with cytokinesis proteins to promote phagocytosis, cell migration.43,44 Animal studies of Gram negative sepsis: bacterial internalization, phagocytosis mediated via ELMO1/Dock180 pathway, which triggers activation of Rac1, causing membrane invagination and pseudopod formation.
7	rs13227735	37073437	8.83E-07	Intron ELMO1	See above
16	rs13380717	86904135	9.37E-07	Intergenic
3	rs3844280	10212258	1.24E-06	Intron IRAK2	Encodes Interleukin-1 receptor associated-kinase 2. Upon stimulation, this kinase associates with IL-1 receptor and participates in IL-1 mediated up-regulation of NF-KB
1	rs3100127	2.02E+08	1.30E-06	Intergenic
13	rs11840143	93284661	1.76E-06	Intron GPC5	Encodes glypican 5, a cell surface proteoglycan
2	rs41461846	2.19E+08	1.83E-06	Intron VIL1	Encodes villin 1, a calcium regulated actin binding protein
2	chr2:219344165	2.19E+08	1.89E-06	Intron VIL1	See above
2	rs6717433	2.19E+08	1.93E-06	Intron VIL1	See above
1	rs2361422	2.02E+08	1.97E-06	5′ PTPRVP

^*Top ten SNPs for each analysis listed; covariates in analysis: gestational age, SGA, eigenvalues for ancestry, surgery and Cesarean section; infants who died before 72 hours excluded from controls and LOS

^**Location in base pairs from p telomere

Table 3.

SNPs with highest association with Gram negative LOS^*

Chr	SNP	Location**	P value	Gene	Gene function
2	chr2:146369734	1.46E+08	1.13E-07	Intergenic
8	rs513793	20774064	5.82E-07	Intergenic
2	chr2:146358394	1.46E+08	8.84E-07	Intergenic
2	chr2:146356963	1.46E+08	8.85E-07	Intergenic
7	rs2237499	39740846	1.06E-06	Intron RALA	Encodes protein belonging to Ras family of GTP binding proteins, implicated in filopodia formation for cell migration
2	rs58162439	1.46E+08	1.55E-06	Intergenic
16	rs59876150	86918734	1.99E-06	Intergenic
7	rs4730486	1.11E+08	3.10E-06	Intron IMMP2L	Encodes protein important for function of mitochondrial inner membrane peptidase (IMP) complex
7	rs6950974	37075828	3.20E-06	Intron RALA
7	rs7811308	1.11E+08	3.92E-06	Intron IMMP2L

^*Top ten SNPs for each analysis listed; covariates in analysis: gestational age, SGA, eigenvalues for ancestry, surgery and Cesarean section; infants who died before 72 hours excluded from controls and LOS

^**Location in base pairs from p telomere

Table 4.

SNPs with highest association with fungal LOS^*

Chr	SNP	Location**	P value	Gene	Gene function
18	rs645505	10682720	6.35E-07	Intron PIEZO2	Encodes protein which functions as part of mechanically-activated cation channels, connecting mechanical forces to biological signals
10	rs11597285	8589813	7.36E-07	Intergenic
10	rs2031558	8589408	7.49E-07	Intergenic
10	rs11596835	8589969	7.53E-07	Intergenic
10	chr10:113662222	1.14E+08	1.28E-06	Intergenic
10	rs11593542	8600339	1.37E-06	Intergenic
10	rs1325884	8599820	1.43E-06	Intergenic
12	rs16913666	18134955	1.73E-06	Intergenic
4	rs17599816	47164556	1.85E-06	Intron GABRB1	Encodes gamma-aminobutyric acid (GABA) A receptor, beta 1
3	rs12490944	50059651	1.90E-06	Intron RBM6	Encodes RNA binding motif protein 6 (previously implicated in various cancers)

^*Top ten SNPs for each analysis listed; covariates in analysis: gestational age, SGA, eigenvalues for ancestry, surgery and Cesarean section; infants who died before 72 hours excluded from controls and LOS

^**Location in base pairs from p telomere

Findings from pathway analyses

We tested 1308 Reactome pathways. For Gram negative sepsis, 30 pathways had a p value <0.005 and FDR<20% (Supplementary Table 3). Pathways identified involved gap junction genes, fibroblast growth factor receptor and insulin receptor signaling. No pathways for Gram positive sepsis or fungal sepsis met FDR thresholds for reporting.

DISCUSSION

We report the first GWA study of sepsis in ELBW infants. In this unique cohort, we identified several SNPs of potential association, although none met genome-wide significance. The results contain potential clues to newborn genetic susceptibility to sepsis, with several SNPs identified related to innate immunity. Over-represented SNPs were found in FOXC2, FOXL1 (all sepsis), ELMO1, VIL1 (Gram positive sepsis), RALA, IMMP2L (Gram negative sepsis) and PIEZO2 (fungal sepsis).

Sepsis is a major co-morbidity in the ELBW population, with up to 36% of infants experiencing at least one episode during their hospital stay.^1,32,33 Immature innate and adaptive immune defenses, and environmental factors, contribute to a complex multifactorial etiology.^1,8 Recent evidence also suggests genetic susceptibility to infection in preterm neonates. ^6–8

Epidemiologic and twin studies indicate a heritable component in certain infections and their complications in adult populations.^34,35 In neonates, a twin study suggested that genetic risk factors were responsible for 49% of variability in sepsis risk;⁷ however this was not confirmed in a subsequent study of multiple births.⁸ Several candidate gene studies examined the association of neonatal sepsis with biologically plausible SNPs in cytokines, cytokine receptors, pathogen recognition receptors, cluster differentiation molecules, endothelial and hemostasis markers.^{10–13,36–38} These studies demonstrated varying results, described in a recent systematic review.¹⁴

We performed this study to explore genetic associations in neonatal sepsis with an unbiased genome wide approach. The NRN’s anonymized genomic database provides a sizable cohort of ELBW infants with genotype and phenotype data. In the analysis of all sepsis, we found over-represented SNPs in proximity to genes encoding forkhead box protein family transcription factors FOXC2 and FOXL1. These genes are involved in intestinal mucosal structure and function, including development of Peyer’s patches; alterations could plausibly contribute to risk of LOS via microbial translocation, and NEC.^39–42

There are likely differences in diagnosis, and in the interaction between the environment and genome in EOS as compared to LOS.²¹ To assess LOS as a distinct disease entity, we performed a secondary analysis to identify SNPs associated with LOS alone. As LOS comprised the majority of sepsis cases, the findings of this analysis were very similar to those previously mentioned in all sepsis. We also examined SNPs associated with known specific pathogen categories in LOS. In the analysis of Gram positive LOS, over-represented SNPs were found in engulfment and cell motility 1 (ELMO1) (chromosome 7), which promotes phagocytosis and cell migration (Table 2).^43–45

Our exploratory analysis of SNPs that have been previously tested in candidate gene association studies yielded trends of interest in gene variants located in TNF, IL-6, IL-10RA, TLR-2 and TLR-4 (Supplementary Table 2). While none reached target levels of significance in our study, these trends warrant further exploration in replication cohorts.

Traditional GWAS and candidate gene analyses examine significance at the level of individual SNPs and genes. This approach risks missing modest but more meaningful changes in multiple genes involved in a biological pathway.⁴⁶ In contrast, a pathway analysis approach uncovers enrichment of ‘significant’ SNPs in groups of related genes.^29,30 The strongest pathway associations were noted in Gram negative sepsis, where pathways involving regulation of gap junctions and connexin function had low FDR estimates (Supplementary Table 3). Connexin proteins provide the framework for gap junction hemichannels, which play an important role in migration and intracellular signaling in immune cells.⁴⁷ Downregulation of gap junction channels results from inflammation and infection.^48–51 Several pathways regulating fibroblast growth factor receptors were also significantly enriched (Supplementary Table 3), and are implicated in attachment and replication of certain pathogens.^52–55 Pathways related to gap junction channels and fibroblast growth factor receptors may play important roles in mediating respiratory and gut mucosal integrity.^50,51,55 On analyzing expression quantitative trait loci (eQTL), ELMO1 levels were noted to be regulated mostly in the gastrointestinal tract, in addition to vasculature. This finding preliminarily suggests a role for alteration in gastrointestinal ELMO1 expression in sepsis. Lastly, IRAK-2, a known component of the classical pro-inflammatory signaling pathway activated by infection, was also identified as enriched in patients that developed Gram positive sepsis.⁵⁷

Our moderately powered study did not identify SNPs that achieved significance at the genome-wide level, indicating that genetic susceptibility to sepsis may be mediated by more than just one or a few SNPs. We did find several over-expressed SNPs at our targeted ‘discovery’ association levels (10⁻⁶ to 10⁻⁷) in genes related to cell migration, phagocytosis and cytokine function. We also identified pathways related to innate immune receptors, maintenance of mucosal homeostasis and barrier defense mechanisms, and cellular signaling, that were enriched in patients who developed sepsis. Adaptive immunity is poorly developed in extremely premature infants; the immune response to infectious insults is therefore reliant on innate immune pathways.⁵⁸ We speculate that the identified variants, and pathways that they are involved in, may regulate key aspects of the premature neonate’s innate immune response, leading to alterations in susceptibility to sepsis. Alteration of gut and respiratory epithelial integrity mediated by these pathways may be an important mechanism of increased sepsis and NEC risk.

The relatively modest size of our cohort imposes limitations on the statistical power of this GWAS. While adult GWA studies have been larger, this is the largest ELBW neonatal population with complete genotyping and sepsis-specific phenotype data available. Our study provides novel insights into potential heritable factors for sepsis predilection in this extremely vulnerable population. The smaller sample sizes of subgroups limited conclusions that could be drawn from subgroup analyses, and also restricted our ability to perform analyses of specific phenotypic subgroups such as meningitis without bacteremia. We limited our analysis to a stringent definition of sepsis (culture proven infection that was treated with a prolonged course of antibiotics). Other limitations include the multi-ethnic population (fairly representative of the US preterm birth population). To address this, we used eigenvalues to represent ancestry in the analysis, which allowed more complete utilization of the dataset; but precluded any inferences related to specific racial or ethnic subgroups. This dataset contained 70 identical twins, with separate clinical data on each; SNPs were run on the first sample in an identical twin set (with blinding to clinical phenotype), and SNP results were ascribed to both twins. As we used anonymized clinical data collected in a general database, we possessed limited information on the temporality of comorbidities in relation to the sepsis event. As a result, NEC and other co-morbidities could not be reliably accounted for as covariates in the analyses.

We provide data on the role of genetic variants in predisposition to sepsis in extremely premature neonates. While no genes achieved genome-wide significance, several SNPs in genes and pathways relevant to innate immunity were identified with our screening approach using ‘target’ levels of significance. Additional pathway analyses and gene expression profiling studies could help better understand interactions and downstream effects of these genetic variations in neonatal sepsis. It is essential to ascertain whether these SNPs and pathways related to them, continue to demonstrate an association with sepsis in additional cohorts of ELBW infants.

What is known about this topic

• Sepsis is a major co-morbidity in extremely low birth weight infants.

• Preterm infants may have genetic susceptibility to infection.

What this study adds

• In a a genome wide association study in extremely low birth weight infants with sepsis, we identified several over-represented single nucleotide polymorphisms (SNPs) related to innate immunity, although none met genome-wide significance.

• Alteration of gut and respiratory epithelial integrity mediated by these SNPs and related pathways may be an important mechanism of increased sepsis risk.