Airway Mucosal Host Defense Is Key to Genomic Regulation of Cystic Fibrosis Lung Disease Severity

Abstract

Rationale: The severity of cystic fibrosis (CF) lung disease varies widely, even for Phe508del homozygotes. Heritability studies show that more than 50% of the variability reflects non-cystic fibrosis transmembrane conductance regulator (CFTR) genetic variation; however, the full extent of the pertinent genetic variation is not known.

Objectives: We sought to identify novel CF disease-modifying mechanisms using an integrated approach based on analyzing “in vivo” CF airway epithelial gene expression complemented with genome-wide association study (GWAS) data.

Methods: Nasal mucosal RNA from 134 patients with CF was used for RNA sequencing. We tested for associations of transcriptomic (gene expression) data with a quantitative phenotype of CF lung disease severity. Pathway analysis of CF GWAS data (n = 5,659 patients) was performed to identify novel pathways and assess the concordance of genomic and transcriptomic data. Association of gene expression with previously identified CF GWAS risk alleles was also tested.

Measurements and Main Results: Significant evidence of heritable gene expression was identified. Gene expression pathways relevant to airway mucosal host defense were significantly associated with CF lung disease severity, including viral infection, inflammation/inflammatory signaling, lipid metabolism, apoptosis, ion transport, Phe508del CFTR processing, and innate immune responses, including HLA (human leukocyte antigen) genes. Ion transport and CFTR processing pathways, as well as HLA genes, were identified across differential gene expression and GWAS signals.

Conclusions: Transcriptomic analyses of CF airway epithelia, coupled to genomic (GWAS) analyses, highlight the role of heritable host defense variation in determining the pathophysiology of CF lung disease. The identification of these pathways provides opportunities to pursue targeted interventions to improve CF lung health.

At a Glance Commentary

Scientific Knowledge on the Subject

Although candidate gene modifiers of cystic fibrosis lung disease severity have been identified through genome-wide association studies, the full extent of the pertinent genetic variation is not known.

What This Study Adds to the Field

We demonstrate that cystic fibrosis lung disease severity is associated with increased airway epithelial expression of genes under genomic (heritable) influence in pathways involving airway mucosal host defense.

Cystic fibrosis (CF) (Online Mendelian Inheritance in Man catalogue number 219700) is an autosomal recessive disorder caused by mutations in the cystic fibrosis transmembrane conductance regulator (CFTR) gene. More than 1,800 mutations have been described in CFTR (1), with the most common mutation, Phe508del, accounting for approximately 66% of CFTR mutations worldwide. Patients with CF experience multiorgan system dysfunction, but lung disease, characterized by chronic (bacterial) infection and inflammation, remains the most common cause of morbidity and mortality, and preserving lung function is a key therapeutic priority. The severity of CF lung disease varies widely, even among Phe508del homozygotes. Twin/sibling studies have demonstrated that more than 50% of the variation in CF lung disease severity reflects non-CFTR genetic variation, with environmental factors also having a role (2–4). The recognition of this heritable variability has led to the search for genetic modifiers, with the hope of identifying genes and gene networks, or pathways, that are harmful or protective, thus providing targets for novel therapeutics.

Such efforts have culminated in a recently reported metaanalysis of genome-wide association studies (GWAS) comprising 6,365 individuals with CF from the International CF Gene Modifier Consortium. CF GWAS (5, 6) employed a standardized Consortium lung phenotype, termed the “Kulich Normal Residual Mortality Adjusted (KNoRMA)” lung disease phenotype, which is a quantitative phenotype that uses 3 years of FEV₁ measures per subject, normalized to a CF reference population (7), and also adjusts for disease survival (8). The development of the KNoRMA phenotype allowed for harmonization of lung disease severity across international cohorts and led to identification of five loci associated with severity of CF lung disease (5). Complementary studies of gene expression in lymphoblastoid cell lines from 754 patients with CF, using KNoRMA as an outcome phenotype, identified additional genetic signatures based on gene expression pathways associated with severity of CF lung disease (9). The success of these studies provides an opportunity for mechanistic exploration. However, GWAS associations account for only a small percentage of expected genetic influence, and gene expression studies in lymphoblastoid cell lines do not optimally reflect airway epithelial biology.

To build upon previous success, we sought to identify novel non-CFTR genetic modifiers of lung disease severity by directly assessing gene expression in respiratory epithelia. We used RNA sequencing (RNA-seq) of nasal epithelial tissue, a well-recognized surrogate for lower airway epithelial function (10–12), from 134 patients with CF with existing GWAS data and the quantitative KNoRMA lung phenotype. We hypothesized that differential gene expression associated with CF lung disease severity would reveal novel candidate gene networks. We also analyzed GWAS data to (1) identify associations of single-nucleotide polymorphism (SNP) variation with nasal epithelial gene expression (i.e., expression quantitative trait loci [eQTLs]), (2) determine overlap between nasal epithelial gene expression– and GWAS-associated gene networks (pathways), and (3) explore the link between significant GWAS loci and nasal epithelial gene expression pathways. Some of the results of this study were previously reported in the form of abstracts (13–15).

Methods

Study Population, Sampling, and RNA-Seq Pipeline

Extended methods for each aspect of the study and analysis plan are provided in the online supplement (see Figure E1). Briefly, we conducted a multicenter study of nasal mucosal curettage biopsies obtained from 134 GWAS subjects with CF (5, 6) with two pancreatic insufficient CFTR mutations (n = 122 Phe508del homozygotes) and a broad spectrum of age and lung disease severity (Table 1). To quantify mucosal inflammation at sampling, nasal lavage obtained just prior to biopsy was analyzed for cytokine levels (IL-8, IFN-γ-inducible protein 10, and IL-1Ra), and the first curettage sample was stained for differential cell counts. From the next nine curettages, we collected cells for RNA isolation. RNA was sequenced using the Illumina HiSeq 2000 sequencing system by Expression Analysis (currently Q² Solutions) following standard library preparation and achieving at least 25 million reads per sample. Fragments per kilobase of transcript per million mapped reads (FPKM) values were determined as described in the online supplement, and gene expression values were included in the data analysis if they met a minimum mean expression threshold level of at least 1 FPKM, based on the 95th percentile of mean Y-chromosome–specific gene expression observed in female samples.

Table 1.

Characteristics of Study Subjects by Research Site

Research Site*	No. of Subjects	Consortium Lung Phenotype† (Mean ± SD)	Age at Consortium Lung Phenotype† (yr) (Median; Range)	Male Sex (%)	BMI (Mean ± SD)	Pseudomonas aeruginosa Infection‡ (%)	CFRD§ (%)	European|| (%)	Phe508del Homozygous (%)
CWRU	38	0.9 ± 0.9	24.3; 11.4–49.2	46	21.8 ± 3.3	90	48	98	79
JHU	17	1.2 ± 0.8	27.1; 18.3–47.3	49	22.2 ± 3.7	92	30	100	77
TOR	35	0.8 ± 0.6	23.1; 10.4–42.6	63	22.0 ± 3.8	78	23	97	100
UNC	44	0.7 ± 0.8	23.8; 11.6–49.5	53	21.9 ± 3.8	91	35	100	100
Total	134	0.8 ± 0.8	26.5; 10.4–49.5	52	21.8 ± 3.3	86	35	99	91

Definition of abbreviations: BMI = body mass index; CFRD = cystic fibrosis–related diabetes; CWRU = Case Western Reserve University; JHU = Johns Hopkins University; KNoRMA = Kulich Normal Residual Mortality Adjusted; TOR = University of Toronto; UNC = University of North Carolina.

^*See Methods section of online supplement for participating sites and enrollment information.

^†Subjects were defined by the quantitative Consortium lung phenotype (KNoRMA) value (8).

^‡Positive lower respiratory culture within 2 years preceding study enrollment; percentage noted is based on data available for 94 subjects.

^§CFRD percentage noted is based on data available for 117 subjects.

^||Based on self-identified ancestry and principal component analysis via SNP genotypes.

Analyses

KNoRMA (Consortium lung phenotype), a standardized quantitative phenotype that uses 3 years of measures of FEV₁, was used as the lung phenotype to quantitate lung disease severity, as previously described (5, 8, 9). Linear models of gene expression as response variables, with clinically relevant covariates (sex, two genotype principal components [PCs], nine expression PCs, transplant history, nasal steroid use, azithromycin use, CD45 expression, and D statistic [mean pair-wise FPKM r² per sample]), were used to determine associations of differential gene expression with KNoRMA, as well as with risk alleles at the five previously identified significant GWAS loci (Table E1) (5). These studies were complemented with a surrogate variable approach (16) (Table E1). To identify eQTLs, we used SNPs with a minor allele frequency greater than 0.05 and gene expression data (FPKM ≥1) as inputs in the Matrix eQTL package (17), which establishes eQTL associations under false discovery rate (FDR) control. To identify pathways significantly associated with differential gene expression, we used Significance Analysis of Function and Expression (SAFE) (18) coupled to pathway annotation sources selected for coverage, accuracy, and relevance (see online supplement). SAFE uses a resampling-based method, testing gene expression association with phenotype through random permutation of phenotype, and performs multiple test correction over the number of pathways tested in each analysis. To test the heritability of genes in significant pathways, we tested the likelihood of genes enriched for significant pathways versus their estimated heritability score determined in an independent blood gene expression report (19). To identify pathways significantly associated with GWAS data, a gene- and pathway-testing approach (GeneSetScan version 0.021) was applied to GWAS data from previously genotyped individuals with CF (n = 5,659, including 134 individuals in the present study) (5). GeneSetScan provides resampling-based multiple comparison–corrected P values for the number of pathways tested. In all analyses, pathways were reported if the corrected P value was less than 0.15, an established threshold for hypothesis generation in the context of these studies (20).

Results

Study Subjects and Evaluation of Inflammation in Nasal Mucosal Samples

Patients with CF tested in this study had a broad range of ages and lung disease severity (KNoRMA), and most had chronic lung infection with Pseudomonas aeruginosa (Table 1). Nasal curettage samples had a median of 87% epithelial cells (interquartile range, 77 to 94%) and a median of 12% neutrophils (interquartile range, 5 to 24%). To address the potential that subjects with more severe lung disease (low KNoRMA) might have more inflammation in the nasal mucosa and thus might confound the analyses, we tested for correlation of KNoRMA with degree of inflammation at the time of sample collection. We observed no significant correlation between KNoRMA and degree of inflammation in the nasal passages, as indexed by quantitative nasal mucosal examination scores, prebiopsy nasal lavage cytokine concentrations (IL-8, IFN-γ-inducible protein 10, and IL-1Ra), neutrophil counts derived from Diff-Quik stains, and CD45 expression (an indicator of inflammatory cells) in nasal mucosal RNA (Figures E2 and E3). Because there was strong correlation between CD45 expression and other measures of inflammation (cytokines, neutrophil counts) (Figure E3), CD45 expression was deemed a pertinent covariate in the analysis to adjust for overall inflammatory state.

Features of Gene Expression

Using the FPKM greater than or equal to 1 threshold for gene expression, 14,548 (52%) of 27,939 annotated genes were called as expressed and used in analyses. eQTLs with significant expression (FDR <0.15) were abundant (n = 14,098), with a preponderance of significant eQTLs within 1 Mb (cis) of the target gene (Table E2).

Relating Lung Disease Severity (KNoRMA) to Gene Expression

Linear models with covariates (see Methods section above and Table E1) were used to identify associations between gene expression level and KNoRMA. No individual gene met the level of statistical significance for association (Table E3). To detect coordinated networks of genes with pathophysiological relevance, we pursued rigorous pathway analysis to identify gene signatures. The analysis, using SAFE, identified pathways associated with lung disease severity with FDR less than 0.15, including viral infection, inflammatory signaling, lipid metabolism, macrophage function, and innate immunity (including HLA [human leukocyte antigen] genes) (Tables 2 and E4). Genes within pathways that contributed most robustly to the pathway significance (gene level P < 0.10) are provided in Table 2. (For a full listing of genes, see Table E5, tabs A and B.)

Table 2.

Gene Expression Pathways Significantly Associated with Consortium Lung Phenotype (KNoRMA)

Pathways with FDR <0.15		Genes		Statistics
Identifier	Name	No. of Genes		P Value*	Q Value†	Increased Expression‡	Genes in the Pathway that Significantly Contribute to Pathway Signal (Gene-Level P < 0.10, Ordered by P Value)§
KEGG pathways, n = 329 tested
05160	Hepatitis C virus	104		0.0004	0.0476	Detrimental	CDKN1A, SCARB1, ARAF, STAT1, BRAF, NRAS, PIAS1, IRF9, TICAM1, NFKB1, CLDN3, PIK3R5, TLR3, TP53, MAPK9, OAS3, MAVS
05168	Herpes simplex virus infection	158		0.0005	0.0476	Detrimental	TLR2, PML, JUN, HLA-DRB1, HLA-DMB, STAT1, CD74, HLA-A, TAF4B, HLA-G, IRF9, TICAM1, HLA-F, NFKB1, FOS, EP300, TLR3, HLA-DRA, TP53, HLA-DMA, CCL5, TAP1, MAPK9, OAS3, HLA-B, HLA-E, MAVS, HCFC2, C3
04640	Hematopoietic cell lineage	56		0.0016	0.0960	Detrimental	IL1R1, HLA-DRB1, ITGAM, CD7, CSF1, HLA-DRA, TFRC
04115	p53 signaling pathway	64		0.0017	0.0960	Detrimental	CDKN1A, CCNG2, BID, GADD45A, TP73, SERPINB5, GADD45B, BBC3, TP53, TNFRSF10B, EI24
00592	α-Linolenic acid metabolism	13		0.0007	0.0981	Protective	PLA2G4F, PLA2G6
00591‖
05322	Systemic lupus erythematosus	62		0.0022	0.1042	Detrimental	HLA-DRB1, HLA-DMB, HIST1H2BG, C2, HLA-DRA, HLA-DMA, HIST1H2AE, HIST2H2BE, HIST4H4, HIST1H4H, TROVE2, C3
04514	Cell adhesion molecules (CAMs)	86		0.0026	0.1063	Detrimental	PTPRC, HLA-DRB1, HLA-DMB, ITGAM, HLA-A, CD276, HLA-G, HLA-F, CLDN3, ICAM1, HLA-DRA, HLA-DMA, ITGB8, HLA-B, HLA-E, ITGB2
04930	Type 2 diabetes mellitus	28		0.0040	0.1356	Detrimental	PIK3R5, PRKCD, MAPK9
05219	Bladder cancer	36		0.0045	0.1356	Detrimental	ARAF, BRAF, NRAS, DAPK1, TP53
05161	Hepatitis B virus	120		0.0047	0.1356	Detrimental	TLR2, CDKN1A, CREB3L2, JUN, STAT1, NRAS, TICAM1, NFKB1, FOS, EP300, PIK3R5, TLR3, TP53, MAPK9, CCNA2, SMAD4, MAVS
05323	Rheumatoid arthritis	69		0.0055	0.1445	Detrimental	JUN, HLA-DRB1, HLA-DMB, CCL3L1, ATP6V0A4, FOS, CSF1, ICAM1, FLT1, CCL3, HLA-DRA, HLA-DMA, CCL5, TNFSF13B, ITGB2
04620	Toll-like receptor signaling pathway	80		0.0068	0.1485	Detrimental	TLR2, JUN, STAT1, CCL3L1, TICAM1, NFKB1, FOS, PIK3R5, TLR3, CCL3, CCL5, MAPK9, CCL4
00061	Fatty acid biosynthesis	10		0.0072	0.1485	Detrimental	ACSL1, ACACA, ACACB
05164	Influenza A virus	145		0.0074	0.1485	Detrimental	PML, JUN, HLA-DRB1, HLA-DMB, STAT1, IRF9, TICAM1, NFKB1, EP300, PIK3R5, ICAM1, TLR3, HLA-DRA, HLA-DMA, SLC25A6, CCL5, MAPK9, TNFRSF10B, OAS3, MAVS
M00034	Methionine salvage pathway	10		0.0083	0.1485	Detrimental	AMD1, MTAP
05203	Viral carcinogenesis	167		0.0086	0.1485	Detrimental	JUN, PRKACB, HLA-A, CDKN2B, NRAS, HDAC9, HLA-G, HIST1H2BG, IRF9, HLA-F, NFKB1, EP300, PIK3R5, TP53, HIST2H2BE, CCNA2, HIST4H4, HIST1H4H, HLA-B, HLA-E, C3
M00676	PI3K-Akt signaling	13		0.0086	0.1485	Detrimental	PIK3R5, FOXO3
GO biological process pathways, n = 4,228 tested
0051591	Response to cAMP	57		0.0002	0.1261	Detrimental	JUN, IGFBP5, STAT1, EGR1, SREBF1, APEX1, BRAF, JUNB, FOS, DUSP1, AKAP6, COL1A1, SPARC, FOSL2, AKAP7
0014074‖
0046683‖
0070665	Positive regulation of leukocyte proliferation	79		0.0003	0.1261	Detrimental	IGFBP2, PTPRC, HHLA2, CDKN1A, HLA-DMB, CD74, HLA-A, CD276, BST1, TICAM1, CSF1, CCL5, HLA-E, TNFSF13B
0032946‖
0050671‖
0051155	Positive regulation of striated muscle cell differentiation	24		0.0003	0.1261	Detrimental	EDN1, FOXP1, CD53, AKAP6
0033631	Cell–cell adhesion mediated by integrin	11		0.0004	0.1493	Detrimental	FERMT3, CCL5
GO molecular function pathways, n = 779 tested
0050431	Transforming growth factor-β binding	15		0.0003	0.1048	Bidirectional	LTBP1¶, VASN¶, CD109¶, HYAL2, ENG, CD36, LTBP4
MetaMiner cystic fibrosis–specific pathways (GeneGo)**, n = 36 tested
	Cholesterol and sphingolipid transport/distribution to the intracellular membrane compartments (normal and CF)	11		0.0058	0.0909	Bidirectional	STARD4¶, NPC1¶, NPC2¶, RAB7A¶
CF-relevant custom pathways††, n = 74 tested
	EHF transcription factor–negative correlation; PMID 25414352	18		0.0007	0.0237	Detrimental	ACSL1, C10orf10, DMKN, ID2, H1F0
	Asthma-COPD (down); PMID 25611785	26		0.0023	0.0504	Detrimental	CCDC81, PTGFR, FOLR1, STEAP2, DAPK1, LTF, CYP4X1
	Macrophage specific: M1 (classic) activation markers; PMID 25204199; and Macrophage activation: combined M1 and M2 markers‖; PMID 19635926	52		0.0038	0.0632	Detrimental	TLR2, GBP3, IL1R1, GBP2, IL8, ICAM1, CCL3, CCL5, IL32, C3AR1, GBP5, CCL4, APOL3
	Hypoxia responses: HIF1 target hypoxia (up); PMID 19491311	188		0.0099	0.1219	Detrimental	STARD4, KLHL24, IGFBP2, EDN1, NDRG1, CXCR4, BRAF, BCL2L11, GAPDH, PTGS2, FNDC3B, PSD3, ARL5B, GADD45B, FOXO3, ATF3, C1orf51, PLOD2
	Hypoxia responses: DC hypoxia (up); PMID 21148811	85		0.0117	0.1219	Detrimental	TLR2, CHST15, LOC374443, PPIF, CD53, SYNJ2, GBP2, LGALS8, LCP2, CD109, CDCP1, SLC29A1, INSIG1, FCAR, ERRFI1
	Hypoxia responses: MCF7 hypoxia (up); PMID 16565084	163		0.0174	0.1355	Detrimental	PLIN2, KLHL24, DSC2, SCARB1, JUN, HLA-DRB1, NDRG1, SOX9, CXCR4, IGFBP5, CCNG2, EGR1, ADM, DDR1, PLAUR, FLNB, FOS, CAV1, GADD45B, GJA1, ATF3, DUSP1, KLF7, ATXN1, EMR2
	Nasal scrape CF (down); PMID 16614352	29		0.0183	0.1355	Detrimental	EPSTI1, CD74, PRKACB, HLA-G, HLA-F, RPS2
CFTR interactome pathways (none), n = 11 tested
HLA-specific pathways, n = 2 tested
	Class I and class II	30		0.0853	0.0747	Bidirectional	HLA-DRB1¶, HLA-DMB¶, HLA-H¶, HLA-A¶, HLA-G¶, HLA-F¶, HLA-DRA¶, HLA-DMA¶, TAP1¶, HLA-B¶, HLA-E¶, PSMB8¶
	Class I and class II	30		0.0093	0.0080	Detrimental	HLA-DRB1, HLA-DMB, HLA-H, HLA-A, HLA-G, HLA-F, HLA-DRA, HLA-DMA, TAP1, HLA-B, HLA-E, PSMB8
	Class II	16		0.0968	0.0577	Detrimental	HLA-DRB1, HLA-DMB, HLA-DRA, HLA-DMA

Definition of abbreviations: CF = cystic fibrosis; CFTR = cystic fibrosis transmembrane conductance regulator; COPD = chronic obstructive pulmonary disease; DC = dendritic cell; EHF = ETS homologous factor; FDR = false discovery rate; GO = Gene Ontology Consortium; HIF1 = hypoxia-inducible factor 1; HLA = human leukocyte antigen; KEGG = Kyoto Encyclopedia of Genes and Genomes database; KNoRMA = Kulich Normal Residual Mortality Adjusted; PI3K = phosphoinositide 3-kinase; PMID = PubMed reference number.

Pathways limited to those with at least 10 but less than or equal to 200 genes.

^*SAFE (Significance Analysis of Function and Expression) analysis used 10,000 permutations to establish significance thresholds (18).

^†Benjamini-Hochberg FDR for pathway testing within each pathway set; Q values less than 0.15 were included.

^‡Increased expression of genes in pathway are detrimental (associated with worse lung disease) or protective (associated with milder lung disease) or bidirectional (associated with either worse or milder lung disease).

^§See Table E5, tab A, for an inclusive list of genes for these pathways; see Table E3 for gene Online Mendelian Inheritance in Man catalogue numbers.

^‖These pathways are statistically significant and carry robust overlap of genes with first-listed pathway; see Table E5, tab A, for an inclusive list of genes in pathways.

^¶For bidirectional pathways, genes with increased expression associated with worse disease are noted.

^**MetaMiner CF-specific pathways represent a version of the Thomson Reuters (formerly GeneGo) MetaDiscovery suite that is enriched with content specific for cystic fibrosis.

^††CF-relevant custom pathways were developed (46) using human gene counterparts (Table E8).

Because multiple methods have been proposed to correct for uncontrolled technical and population stratification, we also performed a secondary analysis using two surrogate variables (16) in lieu of nine expression PCs (Table E1) to obtain gene-level data. Analyses of these gene-level data with SAFE methodology yielded pathways associated with KNoRMA (Table E6; Table E5, tab C), including pathways related to viral infection, inflammatory signaling, lipid metabolism, and innate immunity (including HLA genes), concordant with our primary findings. Restricting the study cohort to 122 Phe508del homozygous patients also supported the primary findings (Table E5, tabs D and E). Increased gene expression was associated with worse lung disease for a majority of the pathways (labeled “detrimental” in Table 2; Table E5, tab A), and two examples of this relationship are provided in genes (HLA-DRB1 and TLR2) that significantly contributed to pathway results (Figure E4).

Heritable Features of Nasal Epithelial Gene Expression

Many of the top-ranked pathways were related to infectious/environmental exposures, but these pathways also had genes with significant eQTLs, which suggested a heritable component. To test if the significant pathways showed evidence of underlying heritability, we performed logistic regression of gene membership in enriched pathways for lung disease phenotype versus estimated heritability (see Methods section in the online supplement). Using heritability estimates (or proportion of gene expression controlled by genetic variances) of blood gene expression from a previous twin-based study of individuals without CF (19), we demonstrated that genes in the enriched pathways with FDR less than 0.15 (Table 2) showed significantly greater evidence of being heritable than the complementary set of genes not represented in the pathways (P = 2.6 × 10⁻⁶). We conclude that lung disease severity is associated with gene expression pathways that reflect, in part, underlying heritable traits.

Repeatability of Sample Measures

We acknowledge that nasal gene expression is prone to dynamic changes related to environmental influences. To provide additional insights related to this issue, we obtained nasal mucosal biopsies in a random subset of the study cohort (n = 39) at a second study visit and obtained RNA-seq data. We tested sample–sample correlations across all genes in the 39 paired samples (mean r = 0.958), relative to a background distribution derived from all 8,911 unique pairwise combinations from the 134 unique samples (mean r = 0.924). We demonstrated (using t statistic and permutation testing to account for dependence) that the paired samples had significantly higher correlation than the unpaired samples (P < 0.0001) (Figure E5), confirming robust intrasubject correlation of nasal epithelial gene expression.

Relating Lung Disease Severity (KNoRMA) to GWAS Pathways

Gene analysis and pathway analysis (GeneSetScan version 0.021) (21) of GWAS data from the previously genotyped cohort (5) had not been performed, and we used this method to identify pathways arising from the GWAS associations with KNoRMA (Table 3; Table E5, tab F). Pathways identified in this analysis were related to airway mucosal host defense, including viral response, inflammation, mucin/goblet cell biology, and cilia function. Interestingly, several pathways with diverse functional annotations (goblet-cell–relevant pathways, cytokine production by Th17 cells, vasodilation, and CFTR interactome [22] pathways) contained CFTR itself.

Table 3.

Genome-Wide Association Study Data Pathways Significantly Associated with Consortium Lung Phenotype (KNoRMA)

Pathway		Genes (n)	Corrected P Value*	Genes with Gene-Level P Value <0.10 (Ordered by P Value)
Identifier	Name
Analyses included all available pathways†
KEGG pathways, n = 338 tested
00510	N-glycan biosynthesis	48	0.019	ALG12, MAN1C1, MGAT5B, MGAT4C, MGAT4A, TUSC3, ALG14, MAN1A1, MAN2A1, GANAB, DPM3, ALG6
05168	Herpes simplex virus infection	173	0.030	HLA-DQA1, HLA-DQB1, HLA-DRB1, PVRL2, PVRL1, PER2, CCL2, TLR2, SRSF2, TYK2, CCL5, POLR2A, IFNA6, TP53, C3, IFNA13, IFNA1, EIF2AK2, LTA, TNF, IFNA2, IFNA5, MCRS1, TBPL1, IFNA14, TLR3, IFNA8, TAF5, HLA-B, IFNA17, PPP1CC, HLA-DOB, TAP1, TAP2, MAPK9, HCFC2, ALYREF, TBPL2
00601	Glycosphingolipid biosynthesis lacto and neolacto series	24	0.030	ST3GAL6, B3GALT5, FUT3, FUT5, FUT6, FUT2, B3GALT2, GCNT2, ST3GAL3, FUT4
05310	Asthma	23	0.102	HLA-DQA1, HLA-DQB1, HLA-DRB1, FCER1A, IL13, IL4, TNF, CCL11, HLA-DOB, PRG2
04650	Natural killer cell–mediated cytotoxicity	123	0.120	FCGR3B, ICAM1, PRKCB, KRAS, VAV2, VAV3, IFNA6, VAV1, PIK3R2, TNFSF10, IFNA13, IFNA1, NCR3, TNF, RAC2, IFNA2, IFNA5, HCST, TYROBP, PRF1, IFNA14, LCP2, IFNA8, MAPK3, HLA-B, IFNA17, PIK3R3, ULBP3, FCGR3A, RAET1L, RAF1
GO cellular component pathways, n = 516 tested
0044448	Cell cortex part	114	0.057	EXOC3, CAPZA2, GYS2, TCHP, CAPZB, PCLO, EXOC4, CORO1A, MYH2, SPTAN1, EXOC7, TRPV4, SPTBN4, EXOC3L2, SPTBN2, SPTA1, CDH1, LLGL1, ANK1, GYPC, PRKCZ, CALD1
0009898	Cytoplasmic side of plasma membrane	152	0.114	FRK, TNK2, GNA12, ACP1, KRAS, TYK2, LDLRAP1, PTK6, LYN, GNAO1, NPHS2, GNG5, GNG7, RASA1, GNA14, CABP1, HTRA2, TEC, SRMS, SPTA1, PTPN7, CDH1, ALOX15, GNAI3
0098562‡
GO biological pathways, n = 4,670 tested
0032770	Positive regulation of monooxygenase activity	25	0.024	AGTR2, APOE, KRAS, TNF, CALM1, POR, TERF2
0051000‡
2000027	Regulation of organ morphogenesis	165	0.029	AGTR2, MET, POU5F1, FOXP2, HNF1B, CNTF, SFRP2, SIX4, SMAD4, SNAI2, SOX17, MSX1, IFT88, MMP20, HGF, DMRT3, CTHRC1, SFRP1, FGFR2, CAV3, XBP1, SIX1, EDNRA, GPC3, TNF, WNT9B, ZNRF3, CDH1, EDN1, FGF1, POR, TBX5
0042311	Vasodilation	67	0.058	AGTR2, APOE, MRVI1, NPR1, SMTNL1, ADCYAP1, CFTR, NPPB, UTS2B, ADORA1, MKKS, P2RY2, HMOX1, BDKRB2, NOS1
0035150‡
0050880‡
0003018‡
0001711	Endodermal cell fate commitment	16	0.098	POU5F1, HNF1B, SOX17, CDC73, EOMES
0042659‡
0031960	Response to corticosteroid	140	0.139	AGTR2, TRH, S100B, KRAS, AQP1, CCL2, ALPL, ADCYAP1, GHRHR, SCGB1A1, STAR, BMP6, CASP9, SPARC, TNF, CALM1, ALDH3A1, GBA, TPH2, EDN1, SSTR3, ACADS, SLC18A2
GO molecular function pathways, n = 910 tested
0044548	S100 protein binding	10	0.123	S100B, AHNAK, S100A1, ATP2A2, FGF1
0032794	GTPase-activating protein binding	11	0.144	PLCD1, TSC1, GNAO1, FMNL3, CDH1, GNAI3
CF-relevant custom pathways, n = 72 tested
	Goblet cell relevant	37	0.001	MUC4, TFF2, CFTR, FUT6, GALNT12, SCGB1A1, ERN2, B4GALNT2, ST6GALNAC1, XBP1, MUC1, GCNT3, FUT4
	Ciliary trafficking	157	0.006	RAB8B, TBC1D7, PTCH1, EFHC1, ARFGEF2, IFT88, TTC26, IFT74, KIF19, RAB4A, RP1, VMA21, GLI3, IFT122, TRAF3IP1, TRPV1, COPG2, DNAH2, MKKS, OFD1, HSPB11, ODF2, IFT81, SSTR3, PACS1, ARHGEF1, KLC3, PCM1, GLI2, SCLT1
	Mucin Calu3	12	0.024	MUC4, MUC20, MUC1
	MCF7 hypoxia (down)	162	0.054	OSTM1, ADAT1, CORO1A, SNRNP40, GAS2L1, SPAG1, POLR3K, RAB35, EEF1E1, GPATCH2, CALM1, ADORA1, KPNA1, PPIF, GDPD3, SLCO3A1, GYG1, PIK3R3, ARHGDIA
	HIF1si (up)/MCF7 hypoxia (down)‡
	Asthma-COPD (up)	36	0.059	CEP72, FAM110C, CD44, TMEM200A, S100A16, CSTA, GCNT3, IGF2BP3, CEACAM5, CDC42EP5
	EHF positive correlation	154	0.092	MUC20, SLC44A4, SH3YL1, RAB25, LCN2, LIMA1, FUT3, STAP2, CTNND1, CEACAM6, FUT6, PTK6, CHMP4C, SH2D3A, SPAG1, PIGR, ST6GALNAC1, S100A14, MYH14, RIPK4, FUT2, SPINT2, CDH1, C10orf99, YAP1, CEACAM5, CGN, CDC42EP5, SLC44A3
	COPD (up)	50	0.099	MUC4, CFB, NR4A1, LCN2, ARNTL2, FUT3, IRAK3, MTNR1A, GCNT3, IGF2BP3, CEACAM5
	Airway epithelium T-helper type 2	92	0.115	CEP72, SCGB2A1, TFF3, FAM110C, CD44, TMEM200A, S100A16, CSTA, GCNT3, ITLN1, IGF2BP3, ALOX15, CEACAM5, CDC42EP5, SLC18A2, SLC22A16
CFTR interactome pathways§, n = 11 tested
	loT1hr dCF; Table E7, 484 genes	466	0.055	LMNA, PRDX1, AHNAK, LIMA1, CAPZB, PDCD6, MCM6, CFTR, ACLY, STAU1, CLPTM1, PPP6R1, SDHA, MYH2, RDX, XRCC5, STRBP, SPTAN1, TPM1, TUBB6, ACSL4, TP53, RBBP4, C3, POLR2E, CNN2, UBR4, MYH13, MOV10, PPP1R12A, RPLP0, MMS19, YTHDF3, SAE1, CSTA, MYH14, SNX27
	loT6hr dCF; Table E8, 618 genes	592	0.059	LMNB2, ICAM1, LMNA, PRDX1, AHNAK, LIMA1, CAPZB, PDCD6, MCM6, THADA, STAU1, CLPTM1, EXOC4, PPP6R1, SDHA, MYH2, RDX, XRCC5, STRBP, SPTAN1, TPM1, TUBB6, ACSL4, TP53, RBBP4, C3, POLR2E, SLC35E1, UBR4, SLC27A3, MYH13, MOV10, PPP1R12A, RPLP0, DARS, MMS19, NUP155, SAE1, MSN, MYH14, SNX27
	Core CFTR interactome; Table E1, 638 genes	620	0.088	CAPZA2, LMNB2, HSPA1B, BLMH, HSPA1A, ICAM1, LMNA, PRDX1, AHNAK, LIMA1, CAPZB, MCM6, THADA, RGPD2, COG6, CFTR, ACLY, STAU1, SORCS1, CLPTM1, EXOC4, SDHA, MYH2, RDX, XRCC5, STRBP, SPTAN1, TPM1, TUBB6, EXOC7, ACSL4, TP53, RBBP4, C3, POLR2E, CNN2, YTHDF2, UBR4, MYH13, MOV10, PPP1R12A, RPLP0, DARS, MMS19, YTHDF3, SAE1, RAC2, CSTA, MSN, MYH14, SNX27
	SAHA dCF‡; Table E11, 681 genes
MetaMiner cystic fibrosis–specific pathways (GeneGo)‖, n = 36 tested
	Cytokine production by Th17 cells in CF	41	0.090	ICAM1, CFTR, RELB, IL12RB1, CXCL1, IL8, RORC, CXCL6
HLA-specific pathways, n = 2 tested
	Class I and class II	18	0.095	HLA-DQA1, HLA-DQB1, HLA-DRB1, HLA-B, HLA-DOB, PSMB8, PSMB9, TAP1, TAP2
	Class II	8	0.146	HLA-DQA1, HLA-DQB1, HLA-DRB1, HLA-DOB
Analyses confined to pathways significant for differential expression in nasal scrape samples¶, n = 37 tested
KEGG pathways
05168	Herpes simplex virus infection	173	0.008	HLA-DQA1, HLA-DQB1, HLA-DRB1, PVRL2, PVRL1, PER2, CCL2, TLR2, SRSF2, TYK2, CCL5, POLR2A, IFNA6, TP53, C3, IFNA13, IFNA1, EIF2AK2, LTA, TNF, IFNA2, IFNA5, MCRS1, TBPL1, IFNA14, TLR3, IFNA8, TAF5, HLA-B, IFNA17, PPP1CC, HLA-DOB, TAP1, TAP2, MAPK9, HCFC2, ALYREF, TBPL2
05164	Influenza A virus	165	0.080	HSPA1B, HLA-DQA1, HLA-DQB1, HLA-DRB1, HSPA1A, HSPA1L, PABPN1L, ICAM1, PRKCB, CCL2, TYK2, CCL5, IFNA6, PIK3R2, TNFSF10, IFNA13, IFNA1, CASP9, EIF2AK2, TNF, IFNA2, IFNA5, IFNA14, TLR3, KPNA1, IFNA8, MAPK3, IFNA17, DDX39B, PIK3R3, HLA-DOB, MAPK9, RAF1

Definition of abbreviations: CF = cystic fibrosis; CFTR = cystic fibrosis transmembrane conductance regulator; COPD = chronic obstructive pulmonary disease; dCF = Phe508del; EHF = ETS homologous factor; GO = Gene Ontology Consortium; GTPase = GTP (guanosine triphosphate) enzyme; HIF1si = HIF-1α siRNA; HLA = human leukocyte antigen; KEGG = Kyoto Encyclopedia of Genes and Genomes database; KNoRMA = Kulich Normal Residual Mortality Adjusted; SAHA = suberoylanilide hydroxamic acid.

Default parameters with 1,000 simulations were used, and pathways were limited to those that contained at least 10 but less than or equal to 200 genes. GeneSetScan uses mapping of genotyped single-nucleotide polymorphisms to 50 kb upstream and downstream of protein-coding genes based on ENSEMBL version 82 annotation and maps genes to annotated pathways and gene sets. CF relevant custom pathways were developed (46) using human gene counterparts (Table E8). See Table E3 for gene Online Mendelian Inheritance in Man catalogue numbers.

^*Multiple comparison corrected P values.

^†Gene level P values were calculated using family-wise error rate (all single-nucleotide polymorphisms, genes, and pathways tested) as provided by GeneSetScan. Pathways are listed if corrected P value is less than 0.15.

^‡These pathways are statistically significant and carry robust overlap of genes with first-listed pathway; see Table E5, tab F, for complete listing of pathway genes.

^§For gene sets containing more than 200 genes, genes with P < 0.05 are listed; see Table E5, tab F, for complete list of pathway genes.

^‖MetaMiner CF-specific pathways represent a version of the Thomson Reuters (formerly GeneGo) MetaDiscovery suite that is enriched with content specific for CF.

^¶Pathways listed in Table 2 were evaluated for association with genotype.

Identification of Functional Overlap and Differences between Expression and GWAS Data

Pathways (and genes) identified in differential expression analysis (Table 2) were similar in many biological respects to those identified using GWAS data (Table 3). To determine the overlap of differential expression and GWAS results, we assessed those genes with P values less than 0.10 contributing to both expression (Table 2) and GWAS (Table 3) pathways. This yielded 18 genes (Figure 1), which is significantly greater than expected by random chance (P = 3.6 × 10⁻⁰⁶). Strikingly, the biological functions of all 18 genes are highly reflective of the broader concept that airway mucosal host defense related to environmental stimuli contributes to lung phenotype (Table E7).

Figure 1.

Top-ranked genes (P < 0.10) common to significant pathways in both differential expression and genome-wide association study (GWAS) analyses. Eighteen genes with significance levels of P < 0.10 were observed in overlap of differential expression and GWAS analyses.

Integration of GWAS Signals with Nasal Epithelial Gene Expression

To further integrate GWAS signal with nasal epithelial gene expression, we tested risk alleles of SNPs at the top five loci in our GWAS (5) for association with gene expression pathways in our nasal epithelial RNA-seq data. We used SAFE and approximately 1,000 randomly selected SNPs to rigorously control for statistical error (Table 4; see also Methods section of online supplement). This analysis demonstrated a significant association between differential expression pathways and the risk allele at four of the five significant GWAS loci (chromosomes [chr] 11, 5, 6, and X). Notably, the chr11 top-ranked GWAS SNP (rs10742326) was significantly associated with multiple pathways relevant to CF pathogenesis (Table 4; Table E5, tab G), including two CFTR-related pathways (i.e., CFTR-dependent regulation of ion channels in airway epithelium and a CFTR interactome pathway specific to Phe508del) (22). HLA genes, lipid transport, and inflammatory signaling were also identified (Table 4).

Table 4.

Gene Expression Pathways Significantly Associated with Risk Alleles for Significant Cystic Fibrosis Genome-Wide Association Study Loci

Chr	SNP rs Number	Pathway Identifier		Genes (n)	Statistics		Minor Allele	Risk Allele	Association with Risk Allele‡	Genes with Gene-Level P Value <0.10 (Ordered by P Value)§
		Set	Name		P Value*	Q Value†
11	rs10742326	CF-relevant custom pathways	COPD signature; PMID 23471465	66	0.0001	0.0035	A	G	Decreased expression	MUC4, ATP10B, SAA2, TMPRSS11D, SLC26A2, SLC26A4, SLC5A8, SAA1, SAA4, IRAK3, C15orf48, SLCO1B3, SERPINB7, EPB41L2, TPRXL, TRIM31, CCDC81, MTNR1A
11	rs10742326	CF-relevant custom pathways	Asthma nitric oxide gene cluster 3; PMID 25338189	48	0.0002	0.0046	A	G	Decreased expression	DUOXA1, FER1L5, WDR90, C16orf93, STK36, ARHGAP33, CEP164, HGS, PDXDC2P, KIAA0895L, TMEM234, MAP4K4, FAM193B, FBXO31, LINC00479, SPPL2B, RAD9A, MYO15B
11	rs10742326	CF-relevant custom pathways	CF MI Lasso; PMID 22466613	21	0.0005	0.0087	A	G	Decreased expression	PROM1, SLC9A3, CD44, CTSB
11	rs10742326	CF-relevant custom pathways	COPD up; PMID 23471465	49	0.0011	0.0153	A	G	Decreased expression	MUC4, ATP10B, SAA2, TMPRSS11D, SLC26A2, SLC26A4, SLC5A8, SAA1, SAA4, IRAK3, C15orf48, SLCO1B3, TPRXL, TRIM31, MTNR1A
11	rs10742326	MetaMiner cystic fibrosis–specific pathways (GeneGo)	CFTR-dependent regulation of ion channels in airway epithelium (normal and CF)	33	0.0040	0.0605	A	G	Decreased expression	ITPR3, ABCC9, WBP1, PRSS8, SLC9A3R1, GNA11, NEDD4, KCNN4, SCNN1A
11	rs10742326	CFTR interactome pathways	Core increased dCF over WT; Table E6, 52 genes	50	0.0337	0.0993	A	G	Increased expression	PSMD3, PSMD4, UBXN1, PSMD8, PSMD11, PSMA2, LMAN2, UBAC2
6	rs116003090	HLA specific	HLA class II	16	0.0626	0.0527	C	C	Bidirectional	HLA-DQB1, HLA-DRB1‖, HLA-DRB4‖, HLA-DQA2‖, HLA-DQA1‖, HLA-DRB5, HLA-DOB‖, HLA-DQB2
6	rs116003090	CFTR interactome pathways	HDAC7 dCF; Table E13, 450 genes	410	0.0202	0.0343	C	C	Increased expression	RDX, APOL2, HSPH1, PPP2R2A, PPP2CA, DNAJA1, SLC25A22, SAMHD1, EZR, YWHAH, SPTLC2, HSPA8, ICAM1, LMNA, PHGDH, KRT7, YWHAE, DCTN2, GART, SFXN3, PPL, LGALS3BP, CDH1, TUBB6, PSMA4, ACTN4, TMEM40, RUVBL1, CAST, UBXN1, TPM4, TIMM50, HSPD1, KLHL22, PSMA6, LAMB3, ITGA3, TAPBP, VDAC2, ERAP1, TF, RAB18, PDXK, ILVBL, SFN, PSMA1, MARS, NCAPG2, AHSA1, YME1L1, CALR
			loT6hr dCF; Table E8, 618 genes¶
			SAHA dCF; Table E11, 681 genes¶
6	rs116003090	CFTR interactome pathways	loT24hr dCF; Table E9, 199 genes	175	0.0226	0.0343	C	C	Increased expression	APOL2, HSPH1, PPP2CA, DNAJA1, YWHAH, HSPA8, KRT7, YWHAE, LMO7, LGALS3BP, TUBB6, ZW10, PSMA4, ACTN4, RUVBL1, TPM4, TIMM50, HSPD1, LAMB3, ERAP1, SFN, MARS
			loT24hr rev dCF; Table E10, 199 genes¶
6	rs116003090	CFTR interactome pathways	Core dCF specific; Table E5, 208 genes	193	0.0344	0.0404	C	C	Increased expression	TMEM165, SAMHD1, CBR1, SEC24C, C9orf167, ICAM1, DCTN2, SFXN3, CDH1, MX1, ISG15, ZW10, PSMA4, TMEM40, CAST, UBXN1, MX2, MOV10, LAMB3, ITGA3, RFC2, PPA1, VDAC2, PDXK, AHSA1, YME1L1
5	rs57221529	MetaMiner cystic fibrosis–specific pathways (GeneGo)	Cholesterol and sphingolipids transport/distribution to the intracellular membrane compartments (normal and CF)	11	0.0002	0.0032	G	A**	Increased expression	RAB9A, SCP2
5	rs57221529	HLA specific	HLA class II	16	0.0320	0.0268	G	A**	Increased expression	HLA-DMA, HLA-DRA, HLA-DMB, HLA-DOA, HLA-DRB1, HLA-DPB1, HLA-DPA1
5	rs57221529	HLA specific	HLA class I and class II	30	0.0957	0.0554	G	A**	Increased expression	HLA-DMA, HLA-DRA, HLA-DMB, HLA-DOA, HLA-DRB1, HLA-DPB1, HLA-DPA1
X	rs5952223	KEGG: M00154	Cytochrome c oxidase	17	0.0007	0.0957	T	C	Decreased expression	COX7A2L, COX6C, COX5B

Definition of abbreviations: CF = cystic fibrosis; CFTR = cystic fibrosis transmembrane conductance regulator; COPD = chronic obstructive pulmonary disease; dCF = Phe508del; HDAC7 = histone deacetylase 7; KEGG = Kyoto Encyclopedia of Genes and Genomes database; MI = meconium ileus; PMID = PubMed reference number; SAHA = suberoylanilide hydroxamic acid; WT = wild type.

Pathways were limited to those with at least 10 but less than or equal to 200 genes. CF-relevant custom pathways were developed (46) using human gene counterparts (Table E8).

^*Significance Analysis of Function and Expression analysis used 10,000 permutations to establish significance thresholds (18).

^†Benjamini-Hochberg false discovery rate for pathway testing within each pathway set; Q values less than 0.15 were included.

^‡Risk alleles may be associated with increased expression, decreased expression, or bidirectional expression of genes in pathway.

^§See Table E5, tab G, for an inclusive list of genes for these pathways; see Table E3 for gene Online Mendelian Inheritance in Man catalogue numbers.

^‖For bidirectional pathways, genes with increased expression associated with CF genome-wide association study loci risk alleles are noted.

^¶These pathways are statistically significant and carry robust overlap of genes with first-listed pathway; see Table E5, tab G, for a complete list of pathway genes.

^**For this study cohort, risk allele differs from that reported in broader CF genome-wide association studies (5).

Discussion

Using unbiased transcriptomic and integrative genomic approaches, we performed a comprehensive analysis to identify modifier genes and mechanistic pathways modulating CF lung disease severity. Although no single gene was statistically significant in isolation, the primary transcriptomic analysis identified differentially expressed genes in pathways (Table 2) under genomic (heritable) influence and relevant to airway mucosal host defense. The pathways that emerged from the analysis, particularly as related to viral infection, inflammation, apoptosis, lipid metabolism, and innate immune responses, including HLA genes, reflect the known complexity of CF pathophysiology. Importantly, the direction and content of differentially expressed genes in these pathways bear striking relevance to what is known about the pathogenesis of CF lung disease. Almost all of the significant pathways in the differential expression analysis demonstrate that increased gene expression is associated with worse lung disease (“detrimental”) (Table 2), which is congruent with the concept that persistent “hyperinflammatory” responses to environmental stimuli (such as viral or bacterial infection) contribute to more severe CF lung disease (23, 24). Indeed, viral infections in CF are known to lead to pulmonary exacerbations and decreased lung function (25, 26), and dysregulated inflammation is believed to adversely affect CF lung disease (23, 26). Our findings are congruent with a previous microarray analysis of nasal brushings in a small study of patients with CF (n = 12) which demonstrated that subjects with severe lung disease had increased expression of genes linked to viral infection, including STAT1 (Table 2), which is critical in the host response to viral infection and transcriptional activation of IFN-induced genes (27).

Pathway (GeneSetScan) analysis of genomic data in 5,659 patients with CF yielded significant pathways containing genes related to viral infection and innate immune response (Table 3), complementing the transcriptomic findings. Of the 18 top-ranked (P < 0.10) genes that were common to both results, nearly all were associated with airway mucosal host defense (Figure 1 and Table E7). Overlapping genes in these analyses, including ICAM1, IL8, and HCFC2, point to heritable variation in the inflammatory response to bacterial and viral infection yielding downstream effects on CF lung disease. Furthermore, IL8 has previously been implicated as a modifier gene in CF lung disease (28). Similarly, overlap of genes integral to the innate immune response (e.g., C3, HLA-B, HLA-DRB1, TLR2, and TLR3) demonstrates that heritable variation in expression of genes related to host defense plays a significant role in determining CF lung disease severity.

GWAS pathway analysis also identified additional host defense mechanisms related to lung disease severity, including goblet cell, mucin production, cilia trafficking, and CFTR interactome pathways (Table 3). Taken together, these pathways point to heritable variation in mucociliary clearance, a critical first-line innate airway defense mechanism involving the interaction of well-hydrated mucus with functional cilia, and a key mucosal defense mechanism regulating CF lung disease severity. Finally, pathways revealed by this GWAS analysis included genes located at significant CF GWAS loci (i.e., AGTR2, EXOC3, MUC20, MUC4, CD44, and HLA genes) (5). Genomic variation in regions near these genes is known to correlate with lung disease severity on the basis of our previously reported findings (5, 6); the gene networks identified in the present analysis provide new insight into potential mechanisms for effect of these candidate modifier genes on CF lung phenotype.

To further explore the mechanism of association between genomic variation at significant GWAS loci and lung disease severity, we used a novel approach to test for association of gene expression pathways with SNP variation at significant CF GWAS loci (5) (Table 4). The chr11p12-p13 GWAS locus is between EHF (an epithelial transcription factor) and APIP (an inhibitor of apoptosis as well as an enzyme in the methionine salvage pathway) (5, 6), and the association at this locus with lung disease severity is determined by Phe508del homozygotes (29); however, the mechanism by which the region produces its phenotypic effects is unresolved. The EHF transcription factor is implicated in recent reports (30, 31), as well as our findings (Tables 2 and 3), whereas other findings support a role for APIP by means of MTAP (Table 2, “Methionine salvage pathway” row) (32, 33). Our analysis demonstrated a significant association of the chr11 risk allele with decreased expression of genes involved in CFTR-dependent regulation of ion channels, as well as other CF-relevant pathways including genes pertinent to chronic obstructive pulmonary disease and asthma (Table 4). Importantly, both the chr11p12-p13 (rs10742326) and chr6p21.3 (near HLA; rs116003090) (5, 6) risk alleles were associated with increased expression of genes in CFTR interactome pathways (22). For the first time, to our knowledge, we demonstrate that in vivo networks, or pathways, of differentially expressed genes in airways are related to established genomic (SNP) variation, where risk alleles are associated with CF lung disease severity (5, 6). Importantly, to our knowledge, these findings represent the first described association of non-CFTR genomic variation with CFTR production, processing, and/or function itself. Coupled with GWAS (5, 6), gene expression networks associated with significant CF GWAS variants provide novel insight into potential mechanisms of effect for candidate modifier genes, and future research will benefit from exploration of these hypotheses.

Our integrated analysis also highlights the need for deep exploration of the HLA region. HLA genes have consistently been implicated across multiple studies of modifier genes in CF, including GWAS (5, 6), differential expression studies in transformed lymphocytes (9), and the nasal mucosal transcriptomic plus genomic pathway analyses described here. The association of genomic variation at the HLA chr6p21.3 (rs116003090) region with expression of genes regulating CFTR processing (Table 4) provides the first glimpse into a novel potential mechanism of action for genetic variation at the chr6 locus to modify CF lung disease, in addition to established roles of HLA in numerous inflammatory and pathogen response pathways (Table 2). The complexity of the HLA region has thus far denied the scientific community of a clear pathogenic mechanism for association with CF lung disease. It is now clear that detailed, integrated analysis of HLA genetic, allelic, and gene expression variability is a critical next step, with findings likely to be highly relevant to other chronic lung diseases, such as asthma, where GWAS signals also reside in the HLA region (34, 35).

Our study has some limitations. First, whereas we characterized the percentage of participants known to have chronic P. aeruginosa infection near the time of sampling, we cannot entirely eliminate infection status as a confounding factor, because microbial culture was not conducted at the sampling date, and this was coupled to our inability to access all possible infections known to have roles in CF lung disease (36). Furthermore, the study does not include a replication cohort or functional validation of any specific pathway. However, validation across analyses for certain genes/pathways (Figure 1) provided evidence of robust signatures that serve as a basis for future replication/functional validation. Future investigations should consider use of effect sizes demonstrated in this study (Table E3).

Despite recent advances in the development of CFTR correctors and potentiators for treatment of CF (37, 38), there remains a critical need for antiinflammatory therapies to ameliorate/optimize airway mucosal host defense that can be applied broadly to patients with CF (39). Currently, there are extensive ongoing efforts to develop such “antiinflammatory” therapies (40), and the genetic and genomic data presented here provide compelling support for these efforts. We highlight one example where the gene expression results have potential therapeutic relevance. Transcriptomic evidence of increased inflammatory signaling in the methionine salvage pathway (Table 2) includes increased gene expression of AMD1, MTAP, and APIP. The expected increases in enzymatic activity of these genes would reduce levels of the antiinflammatory metabolite methylthioadenosine while generating proinflammatory polyamines (32, 41, 42). Recent mass spectrometric metabolomic analysis of bronchoalveolar lavage fluid from children with CF has shown that increased polyamine levels correlate with neutrophilic inflammation and worse lung function (43), and the direction of this finding is congruent with our gene expression findings. Because pharmacologic inhibitors of this pathway are available (44), we have begun exploring this pathway in animal studies to provide proof-of-concept support for such inhibitors as a CF therapy (45).

In conclusion, this study represents a rigorous effort to use gene expression data from the highly CF-relevant airway (nasal) epithelial cell, complemented by extensive genetic data, to identify modifying pathways relevant to CF lung disease severity. The transcriptomic data we report provide unique evidence of increased airway epithelial gene expression in biologically informative pathways, congruent with underlying concepts that hyperinflammatory responses are deleterious to CF lung disease. The presence of genes in both expression- and genomics-based analyses (GWAS and SNP pathway analyses) provides support for the genomic basis of modifier genes, even when mediated through changes in expression. Although association studies of differential gene expression cannot establish cause and effect, genes in our significant pathways demonstrate robust heritability. Taken together with the results of our heritability analysis, these findings suggest that heritable traits linked to increased expression of non-CFTR genes, particularly those regulating inflammatory responses to environmental stimuli, play a key role in CF lung disease severity. Candidate pathways and genes identified by these studies offer novel targets for precision therapies directed toward genes with heritable effects on lung disease severity in CF.