Epigenetic responses to rhinovirus exposure in airway epithelial cells are correlated with key transcriptional pathways in chronic rhinosinusitis

Abstract

Background

Viruses may drive immune mechanisms responsible for chronic rhinosinusitis with nasal polyposis (CRSwNP), but little is known about the underlying molecular mechanisms.

Objectives

To identify epigenetic and transcriptional responses to a common upper respiratory pathogen, rhinovirus (RV), that are specific to patients with CRSwNP using a primary sinonasal epithelial cell culture model.

Methods

Airway epithelial cells were collected at surgery from patients with CRSwNP (cases) and from controls without sinus disease, cultured, and then exposed to RV or vehicle for 48 hours. Differential gene expression and DNA methylation (DNAm) between cases and controls in response to RV were determined using linear mixed models. Weighted gene co-expression analysis (WGCNA) was used to identify a) co-regulated gene expression and DNAm signatures, and b) genes, pathways, and regulatory mechanisms specific to CRSwNP.

Results

We identified 5,585 differential transcriptional and 261 DNAm responses (FDR<0.10) to RV between CRSwNP cases and controls. These differential responses formed three co-expression/co-methylation modules that were related to CRSwNP and three that were related to RV (Bonferroni corrected P < 0.01). Most (95%) of the differentially methylated CpGs (DMCs) were in modules related to CRSwNP, whereas the differentially expressed genes (DEGs) were more equally distributed between the CRSwNP- and RV-related modules. Genes in the CRSwNP-related modules were enriched in known CRS and/or viral response immune pathways.

Conclusion

RV activates specific epigenetic programs and correlated transcriptional networks in the sinonasal epithelium of individuals with CRSwNP. These novel observations suggest epigenetic signatures specific to patients with CRSwNP modulate response to viral pathogens at the mucosal environmental interface. Determining how viral response pathways are involved in epithelial inflammation in CRSwNP could lead to therapeutic targets for this burdensome airway disorder.

Graphical Abstract

• This study identified differential gene expression and epigenetic profiles of the airway epithelium between CRSwNP cases and non-CRS controls in an in vitro cell model.

• Interaction effects indicated that the epigenetic response to RV infection significantly differed between cases and controls. Distinct co-regulated gene expression and DNAm modules were specific to CRSwNP and to RV infection.

• These data suggest that intrinsic epigenetic features in the airway epithelium of CRSwNP influence the response to RV infection and increase susceptibility to CRS.

Abbreviations: CRSwNP, chronic rhinosinusitis with nasal polyps; DNAm, DNA methylation; DTC, dynamic tree cut; MD, merged dynamic; RNA-Seq, RNA sequencing; RV, rhinovirus

INTRODUCTION

Chronic rhinosinusitis (CRS) is a common inflammatory disease of the paranasal sinus epithelium¹. This upper airway disorder is commonly classified clinically into CRS sans nasal polyps (CRSsNP) and CRS with nasal polyps (CRSwNP). These subtypes are associated with different biological profiles^2–4 and environmental risk factors^5–7, reflecting heterogeneity of clinical features and (potentially) reflecting distinct underlying mechanisms or endotypes^{8, 9}. A major objective in the field is to identify the genesis of the chronic epithelial inflammation that is the hallmark of this condition. Despite major public health burdens (billions in expenditures on surgery, medication, and testing, major costs for reduced productivity, and significantly impaired quality of life, little is known about underlying molecular mechanisms.

One approach to identify such pathways has focused on genetic susceptibility, but the few associations reported have not been replicated^10–13 among the relatively few genetic studies to date^{9, 10}, An alternative approach has been to consider environmental influences, which are plausible drivers of diseases of the upper airway because of its direct interactions with inhaled exposures. Recent evidence supports a such role in CRS^{9, 14}. Indeed, one major role of the nose is in mucosal immune defense against airborne pathogens. Failure to incorporate a gene-environment interaction perspective may explain (in part) why traditional genetic studies have not been particularly fruitful.

In this study, we examined viral infection as one such environmental trigger that may drive epithelial immune responses in CRS. Several clinical features support this concept. First, patients with CRS are often symptomatic and have exam findings consistent with infection, but bacterial cultures are frequently negative. Second, respiratory viruses in general¹⁵ and rhinovirus (RV) in particular are common in some CRS patients¹⁶, suggesting a potential role for viral infection in this disease^{16, 17}. Third, RV induces nasal symptoms (congestion, drainage, etc.) and other clinical features that overlap with CRS^{6, 15, 18, 19} Finally, recent epidemiologic data showed that seasonal patterns of RV infections and CRS exacerbations closely overlap^{20, 21}. Importantly, direct knowledge of genomic responses to viral infection as a trigger important in CRSwNP remains limited. Curiously, however, a small fraction of adults develop CRS despite billions of RV infections annually in the United States. Taken together, these observations suggest that intrinsic features of the sinus epithelium may drive specific responses to environmental stimuli such as RV infection and cause CRS to develop in susceptible individuals. Therefore, understanding the specific molecular responses to this pathogen could yield insights into disease biology and potentially lead targeted therapies.

Here, we report the results of a multi-omics study to identify molecular responses to viral infection specific to CRSwNP. To avoid challenges inherent to in vivo studies, we used an upper airway (sinonasal) epithelial cell culture model to identify gene expression and DNAm responses to RV.

METHODS

Ethics Statement

Sinonasal epithelial cells were collected by brushing from study participants between March 2012 and August 2015 during endoscopic surgeries at Northwestern University Feinberg School of Medicine in Chicago, IL. Informed written consent was obtained from each participant and randomly generated IDs were assigned to all samples to preserve privacy. This study was approved by the institutional review boards of both Northwestern University and the University of Chicago.

Cohort description and sample collection

We used a systems biology approach to identify molecular responses to RV that are specific to individuals with CRSwNP. To this end, we designed a cell culture model of RV infection in primary airway epithelial cells (Figure 1).

Fig 1. Study design and analytical workflow.

See Methods for additional details.

Sinonasal epithelial cell brushings of the uncinate process were collected from 104 adults during routine endoscopic surgery for CWSwNP (cases) or for other unrelated indications (adenoidectomy, dentigerous cysts, septoplasty, and tonsillectomies) (non-CRS controls). Because the CRSwNP is the most severe CRS phenotype, we focused the design and analysis of this study on a single subtype to reduce the variability in our data.

The overall cohort included 49 adults with CRSwNP and 55 controls without sinus disease. CRSwNP was diagnosed based on the European Position Paper on Rhinosinusitis and Nasal Polyps (EPOS) criteria¹. Because asthma is a common co-morbidity with CRS and to avoid identifying pathways that overlap with this lower airway condition, we excluded subjects with a current or previous physician diagnosis of asthma to reduce potential confounding effects. However, after excluding subjects with asthma, only one case and one control had a history of allergic rhinitis and were both included in subsequent analyses.

Based on these criteria, 21 subjects with CRSwNP (cases) and 39 controls (non-CRS) were included in our study. This reduction in sample size had a minimal impact on the statistical power of our analyses for gene expression and DNA methylation, maintaining our capacity to discern significant differences (see Supplementary Methods for details). Demographic characteristics of the study cohort are shown in Table 1.

Table 1.

Demographic composition. Ethnicity differences for whites vs other races were tested between CRSwNP and non-CRS participants. Differences in the distribution of sex and ethnicity were evaluated using a chi-square test, and age was evaluated using a 2-sided t-test.

	Combined	CRSwNP	non-CRS	P value
N	60	21	39	-
Sex (% Female)	38	14	51	4.45×10−6
Age, Median (IQR)	45 (25.25)	54 (14)	38 (22)	3.1×10−3
Race/Ethnicity (self-reported)				0.60
White	39	15	23	-
Black	13	4	9	-
Hispanic	7	2	5	-
Bi/Multiracial	2	-	2	-

There were more men than women among the CRSwNP cases, consistent with prior epidemiologic studies¹. Additionally, the controls were younger, likely reflecting the age distributions of patients requiring surgical intervention for unrelated non-CRS conditions compared to those electing CRS surgery. Age and sex were included as covariates in all subsequent analyses. Ancestry principal components (PCs) were included as covariates to account for genetic ancestry differences between the cases and controls (see below).

Airway epithelial cell culture, RV

The airway epithelial cell culture model of RV infection was previously described in detail¹³. Briefly, after isolation at surgery, airway epithelial cells were cultured in bronchial epithelial cell growth medium to near confluence, then cryopreserved at −80°C and stored in liquid nitrogen (range 8 days to 3 years). Airway epithelial cells were thawed and assessed for viability using trypan blue staining or lactate dehydrogenase (LDH) assays prior to further processing. Only cells demonstrating a viability exceeding 90% were selected for the study. The viable cells were then cultured to confluence, and then treated for 48 hours with RV (RV-16; RV) using a MOI of 2 or with a vehicle (bronchial epithelial cell basal medium (BEBM) + Gentamicin/Amphotericin) alone (2-hour treatment followed by a wash in media; 46 hours culture time after this). After cell culture, cells were lysed and stored at −80°C until they were batched for DNA and RNA extraction. Cells were tested for pre-existing RV infection by RT-PCR prior to infection with RV-16 and excluded if positive. Samples displaying significant cell death post RV-16 infection were excluded from the study to minimize the risk of skewed results due to the confounding effects of non-viable cells. RV-16 was selected because of the extensive experience with this common serotype in cell culture studies, and because its main receptor, intracellular adhesion molecule 1 (ICAM-1), is present on the sinonaasal epithelium.

Ancestry Principal Components

Genotyping was performed using the Illumina Infinium HumanCore Exome+Custom Array. The 676 ancestry informative markers²² on this array were used to estimate ancestry PCs for of each subject, as previously described¹³. The first three ancestry PCs were included in all analyses to correct for population structure and genetic ancestry imbalances between the cases and controls.

RNA extraction, sequencing, and QC

RNA was extracted from RV- or vehicle-treated cells. cDNA libraries were constructed using the Illumina TruSeq RNA Library Prep Kit v2 and sequenced on the Illumina HiSeq 2500 System (50 bp, single-end) at the University of Chicago Genomics Core. Sample contamination and swaps were not detected using the VerifyBamID software²³. RNAseq mapping and quality control were applied to the dataset, as described¹³. After quality control, 11,898 autosomal genes in the 20 CRS cases and 32 non-CRS controls remained for downstream analyses.

Principal components analysis (PCA) was used to identify biological and technical sources of variation in the normalized RNAseq dataset. While cryopreservation time (range: 8 days to 3 years) was not a significant source of variation, six technical effects contributed to sample variance: technician, days of cell culture, cell lysate batch, RNA concentration, sequencing pool, and percent of mapped RNAseq reads. Thirteen unknown sources of variation (surrogate variables [SV]) were estimated for the dataset using the SVA R package ²⁴, after protecting for CRS and RV or vehicle treatment. Corrections for technical effects and SVs in the analyses are described below. These approaches allowed us to remove unwanted variation from the data, enhancing the signal of specific responses of interest.

DNA extraction, methylation profiling, and QC

Of the 866,836 CpG probes on the EPIC array (Illumina), we removed 74,444 probes that were either located on the X or Y chromosomes, had detection P values greater than 0.01 in more than 10% of samples, or missingness < 5%. Quality control and array normalization were applied to each sample, as described¹³. There were 792,392 autosomal CpGs in 21 CRS cases and 39 non-CRS controls which remained after quality control. PCA identified biological and technical sources of variation in the normalized methylation dataset. Variability due to cell harvest date was the only significant technical effect and age, sex, and ancestral PCs 1–3 were significant biological variables. Cryopreservation time was not a significant source of variation. Unknown sources of variation were predicted with the SVA package²⁴ in R, which estimated 21 SVs after protecting for CRSwNP status and RV or vehicle treatment. These variables were included as covariates in the analysis adjusted for in our analyses, again improving our signal to noise ratio.

Differential gene expression and DNA methylation analysis

To identify transcribed genes and DNAm sites associated with CRSwNP, RV-response, or both, we used linear mixed effects models to identify differentially expressed genes (DEGs) and differentially methylated CpGs (DMCs), based on M-values, using the limma R package²⁵. The models used to identify DNAm and gene expression (Gx) differences in CRS cases and non-CRS controls in vehicle- and RV-treated cells, as well as RV-response genes in CRS and non-CRS samples for each gene or CpG site followed the general form:

Y_{(Gx or DNAm)} ~ 𝛽₀ + 𝛽₁X_CRS + 𝛽₂X_Treatment + covariates.

We used an FDR of 0.10 to control the false positive rate. Biological and technical sources of variation were included as covariates for their respective datasets (described above), as well as age, sex, and ancestry PCs 1–3. To test for interactions between RV-response and CRSwNP on gene expression and on DNAm, we included an interaction term as follows:

Y_{(Gx or DNAm)} ~ 𝛽₀ + 𝛽₁X_Treatment + 𝛽₂X_CRS + 𝛽₃X_{Treatment x CRS} + Age + Sex+ ancPCs + covariates

Because of the general sparsity of interaction effects detected in small samples, tests for interaction effects were limited to genes or DNAm sites that were DEGs or DMCs, respectively, in any of the above analyses.

Correlation network construction by WGCNA

Gene expression and DNAm correlation networks were constructed using a supervised weighted gene co-expression network analysis (WGCNA)²⁶, a network analysis method for evaluating the correlation structure of our data and to explore relationships between molecular phenotypes (i.e. gene expression, DNAm). For this analysis, we included the 7,474 DEGs and 6,254 DMCs identified in the four differential gene expression and DNAm analyses, two exploring differences between cases and controls in RV and vehicle treated cells (CRSwNP_RV vs non-CRSwNP_RV and CRSwNP_vehicle vs non-CRSwNP_vehicle) and two exploring differences in RV response in cases and controls (CRSwNP_RV vs CRSwNP_vehicle and non-CRSwNP_{RV vs} non-CRSwNP_vehicle) at an FDR<0.10 (described above; Figure 2). The covariate-adjusted residuals for the DEGs and DMCs were merged, quantile normalized, and then a weighted adjacency matrix was created for the combined gene expression and DNAm residuals (see Supplementary Methods for details).

Fig 2. Differential gene expression and DNA methylation analysis of cultured airway epithelial cells treated with RV and a vehicle control.

A-D Results of the differential gene expression analyses. A Volcano plots of differentially expressed genes (DEG) between cases and controls from the vehicle-treated cells (left) and the RV-treated cells (right). The red line in each volcano plot indicates a 0.10 FDR threshold. B Venn diagram of DEGs from the comparison of cases and controls from the vehicle- and RV-treated cells. C Volcano plots of DEGs of RV-response in the cases (left) and controls (right). D Venn diagram of RV-responsive genes in the cases and controls. E-H Results of the differential DNA methylation analyses. E Volcano plots of the differentially methylated CpGs (DMC) between cases and controls in the vehicle-treated (left) and RV-treated (right) cells. F Venn diagram of DMCs between cases and controls from the vehicle- and RV-treated cells. G Volcano plot of DMCs in response to RV treatment in cells from the cases (left) and controls (right). H Venn diagram of RV-responsive CpGs in cases and controls.

Protein-protein interaction network construction

The Search Tool for the Retrieval of Interacting Genes database (STRING) is a database of known and predicted protein-protein interactions collected from multiple sources including computational predictions and experimental data. The STRING ‘Multiple Proteins’ search tool (STRING version 11.5; https://string-db.org/) was used to construct protein-protein interaction (PPI) networks based on genes that were included in the WGCNA modules and enriched in any biological process with an adjusted P value < 0.10. The STRING network was then visualized with the Cytoscape software²⁷, and highly interconnected regions (clusters) of this network were identified using molecular complex detection (MCOD)²⁸. Clusters with a MCODE score > 6 were considered accurate subnetwork predictions (accuracy > 90%). These clusters represent known and/or predicted gene interactions corresponding to genes in the modules identified from WGCNA.

RESULTS

Molecular profiles related to CRSwNP

Our first goal was to identify the differences in gene expression (n=20 cases, 32 controls) and DNAm profiles (n=21 cases, 39 controls) between CRSwNP cases and controls and/or between RV- and vehicle-treated samples. To this end, we conducted four analyses each for gene expression and DNAm comparing CRSwNP cases to controls separately in vehicle- and RV-treated cells and comparing RV- to vehicle-treated cells separately in CRSwNP cases and in controls. Overall, 7,474 genes were differentially expressed and 6,254 CpGs were differentially methylated in at least one of these four analyses. The number of upregulated and downregulated genes and hypomethylated and hypermethylated CpGs in each analysis are shown in Figure 2. CRS x RV interactions were detected for 61 CpGs, reflecting different DNAm responses to RV in cultured airway epithelial cells (Fig 3).

Fig 3. DNAm sites tested for interaction effects for CRSwNP status and treatment (CRSwNP x treatment).

A Venn diagram showing the overlap of the 6,254 differentially methylated sites at an FDR<0.10. The numbers on the left and right of the Venn diagram indicate the number of CpGs that were identified to have interaction effects (FDR<0.10) and the differential analyses from which they were initially identified. B Box plots showing examples of DNA methylation interaction effects for four of the 61 interactive CpGs. The adjusted P values for the interaction effects (P_int) are shown in each box plot.

WGCNA identified co-regulated modules of gene expression and DNA methylation specific to CRSwNP

We next used WGCNA²⁶, a systems biology tool, to evaluate coordinated transcriptional and epigenetic responses to RV or between cases and controls. For this analysis, we included the 7,474 DEGs and 6,254 DMCs that were differentially expressed or methylated, respectively. in at least one of the four analyses (see Fig 2). After merging small, closely related modules (see Methods), WGCNA assigned all the DEGs and DMCs to one of six modules of co-regulated gene expression and/or DNAm (Fig. 4a). Three modules included both DEGs and DMCs (black, blue, turquoise), two included only DEGs (red and yellow), and one included only DMCs (brown) (Fig. 4b). The correlations (and P-values) between the eigenvector of each module with CRSwNP and RV treatment are shown in Fig. 4b, identifying three modules correlated with CRSwNP and three modules correlated with RV. Surprisingly, no modules were significantly correlated with both RV infection and CRSwNP after correcting for multiple testing (12 tests, Bonferroni corrected P < 4.1×10⁻³).

Fig 4. Network analysis dendrogram showing co-regulation modules of gene expression and DNA methylation profiles identified by weighted gene co-expression network analysis (WGCNA).

A Dendrogram showing co-regulated modules of co-expression and DNAm. Colored rows below the dendrogram indicate modules that were identified by WGCNA. The merged modules with highly correlated expression or DNAm profiles are shown on the bottom row. B The module-trait relationship. Each row corresponds to a module eigenvector while each column corresponds to either CRS status or treatment. Each cell contains corresponding correlations (top value) and P value (bottom value). Cells are colored blue for negative correlations while red cells indicate positive correlations from −1 to 1, respectively. The legend on the right indicates the number of co-regulated genes and DNAm sites in each module.

The brown, blue, and turquoise modules were significantly correlated with CRSwNP and the red, black, and yellow modules were significantly correlated with RV, indicating that different molecular changes are related to CRSwNP and RV infection. The three CRSwNP-associated modules were predominantly comprised of DMCs: 95.6% of all DMCs were in the CRSwNP-associated module eigenvectors. The three RV-associated modules included 56.3% of all DEGs, with the remaining 43.6% distributed among two of the CRSwNP-associated modules (blue and turquoise). These data suggested that DNAm patterns may reflect intrinsic properties of airway epithelial cells from CRSwNP patients that modulate the responses of genes to RV infection.

Module genes are enriched in CRS and microbial response-associated gene pathways

To infer molecular mechanisms from the genes within the modules, we performed gene enrichment analyses (Enrichr gene enrichment analysis tool²⁹) using biological pathways from the WikiPathways database³⁰ for the five gene-containing modules (Table 2; Supporting Information). The RV-associated modules (red, yellow, black) contained genes enriched in pathways representing particular innate immune responses and molecular signaling pathways involved in response to microbial infection. These pathways have been implicated in the regulation of RV replication in bronchial epithelial cells³¹, and are upregulated by viruses in order to meet the demand for viral structural elements³² (Table 2). In contrast, the blue and turquoise CRSwNP-associated modules were predominantly enriched in mitochondrial and cytokine signaling molecular pathways respectively. These pathways have been linked to CRS, including genes in the IL-5 and TNF-α signaling pathways^{8, 9} (CASP8, GRB2, MAPK3, PTPN11). Eosinophilic inflammation (where IL-5 plays a critical role³³) is intimately involved in CRSwNP and TNF-α is involved in initiated relevant immune responses in CRSwNP³⁴.

Table 2.

Module genes enriched in biological pathways. The top five pathways are shown for each module.

	Module (# of Pathways Padj<0.10)	Term	P value	Adjusted P value	Odds Ratio	Module Genes / Pathway Genes
RV-Response Pathways	Black (4) Cell Proliferation	Retinoblastoma Gene in Cancer	4.63×10−05	2.18×10−02	2.96	17/87
		DNA Replication	3.18×10−04	7.51×10−02	3.61	10/42
		Breast cancer pathway	5.01×10−04	7.88×10−02	2.16	22/154
		EGF/EGFR Signaling Pathway	1.00×10−03	9.45×10−02	2.06	22/162
		Signaling Pathways in Glioblastoma	8.88×10−04	1.05×10−01	2.59	14/82
	Red (67) Immune Response to Microbial Infection	The human immune response to tuberculosis	6.35×10−19	3.00×10−16	20.22	16/23
		Type II interferon signaling (IFNG)	3.5×10−17	8.27×10−15	14.14	18/37
		Retinoblastoma Gene in Cancer	1.24×10−14	1.94×10−12	7.69	23/87
		DNA IR-damage and cellular response via ATR	1.99×10−11	2.34×10−09	6.9	19/80
		Apoptosis	2.34×10−08	2.21×10−06	5.54	16/84
	Yellow (9) Lipid and Cholesterol Biosynthesis	Genes related to primary cilium development	8.29×10−13	3.91×10−10	3.45	39/103
		Cholesterol Biosynthesis Pathway	9.68×10−10	2.28×10−07	7.3	12/15
		Sterol Regulatory Element-Binding Proteins (SREBP) signaling	5.83×10−07	9.17×10−05	3.04	23/69
		Ciliary landscape	1.57×10−04	1.24×10−02	1.77	42/216
		Lipid Metabolism Pathway	1.41×10−04	1.33×10−02	3.46	11/29
CRS Gene Pathways	Blue (10) Oxidative Phosphorylation	Cytoplasmic Ribosomal Proteins	1.31×10−39	6.21×10−37	12.93	45/89
		Electron Transport Chain (OXPHOS system in mitochondria)	1.61×10−14	3.81×10−12	6.46	26/103
		Oxidative phosphorylation	7.66×10−10	1.21×10−07	6.82	16/60
		Mitochondrial complex I assembly model OXPHOS system	1.80×10−07	1.70×10−05	5.94	13/56
		Nonalcoholic fatty liver disease	1.68×10−07	1.98×10−05	3.63	22/155
	Turquoise (1) Leptin Signaling	Leptin signaling pathway	9.47×10−06	4.47×10−03	5.1	11/76
		IL-5 Signaling Pathway	8.62×10−04	1.02×10−01	5.28	6/40
		MET in type 1 papillary renal cell carcinoma	1.36×10−03	1.07×10−01	4.18	7/59
		TNF alpha Signaling Pathway	1.21×10−03	1.14×10−01	3.44	9/92
		Pathways Affected in Adenoid Cystic Carcinoma	4.90×10−04	1.16×10−01	4.33	8/65

DNAm sites with interaction effects are enriched in two modules

To further explore the possibility that some modules may reflect co-regulated pathways of genes and/or CpGs with differential responses to RV between CRSwNP cases and controls, we first assessed the distribution of the 61 DMCs with CRS x RV interaction effects among the four modules containing CpGs (turquoise, black, brown, blue) (Fig 5A). Most of the interactions (86%) included CpGs in two modules: one module was correlated with CRSwNP (turquoise, 32 CpGs) and one module was correlated with RV treatment (black, 21 CpGs). The remaining eight CpGs were in two modules correlated with CRSwNP (brown, 6 CpGs; blue, 2 CpGs). Two of the modules were significantly enriched for DMCs with interaction effects compared to DMCs without interaction effects (P_adj ≤ 0.05), with a 3.71-fold enrichment in the CRS-correlated brown module (P = 2.0×10⁻⁶; P_adj = 8.0×10⁻⁶; hypergeometric test) and 7.86-fold enrichment in the RV-associated black module (P = 3.5×10⁻¹⁴; P_adj = 1.4×10⁻¹³; Fig 5B). The black module was also enriched for genes involved in proliferative responses, which are important features of both RV response and CRS pathogenesis.

Fig 5. Enrichment analysis of DNAm sites with interaction effects in WGCNA modules.

A Bar plot of the distribution of DNAm sites with interaction effects in modules with co-methylated CpG sites. The x-axis indicates the WGCNA module while the y-axis indicates the number of DNAm sites with interaction effects; the number of DNAm with interaction effects is shown above each bar. B Bar plot of enrichments of interaction DNA methylation sites in WGCNA modules. The x-axis indicates the WGCNA module, and y-axis indicates the fold enrichment of interactive DNA methylation sites. Enrichment P values for each module are shown above each bar (hypergeometric test).

To investigate the regulatory potential of these 61 CpGs, we overlapped their genomic locations with ENCODE31 transcription factor binding sites (TFBS) and tested for enrichments of the CpGs involved in interactions at TFBS relative to CpGs without interaction effects. Indeed, CpGs involved in interactions were enriched 1.28-fold in TFBSs (P = 5.70×10⁻³; hypergeometric test). Overall, 46 of the 61 (75.4%) CpGs overlapped with binding sites for 135 transcription factors, suggesting their potential for influencing the binding of these transcription factors and the expression of their downstream genes. These results demonstrate that DNAm patterns intrinsic to the airway epithelium of CRSwNP patients drive responses to RV by influencing the binding of transcription factors and regulating the expression of coordinated networks of genes.

Network analysis shows CRS-associated and RV response gene interactions

The STRING protein-protein interaction (PPI) database was used to construct a gene interaction network of co-expressed genes that were members of the five modules and enriched in biological pathways with a FDR< 0.10. The interaction network was comprised of 508 nodes and 8,978 edges (Fig 6A). The molecular complex detection (MCODE) method identified eight clusters of dense protein interactions with scores > 6. Although most of these clusters contained module genes in either the CRSwNP-associated (Fig 6B) or the RV-response pathways (Fig 6C), cluster 3 contained module genes in both pathways, potentially linking RV-response (black, red, yellow modules) to CRSwNP pathogenesis (blue and turquoise modules) (Fig 6D). The gene membership in Cluster 3 suggests that of the two CRSwNP-associated gene modules, the turquoise module may contain genes that mitigate CRSwNP exacerbation in the presence of RV infection (Fig E4 shows the remaining five modules). The presence of genes identified in Cluster 3 and DMCs with CRSxRV interactions is further support intrinsic DNAm levels driving RV transcriptional responses in the airway epithelium of CRSwNP patients.

Fig 6. PPI network construction and network clusters.

(A) PPI network of genes from the five WGCNA modules. Edges show the interaction between two genes and nodes are colored according to the module from which they were assigned. Seven significant clusters identified from the PPI network were determined using MCODE with a score >= 6. The most significant clusters are shown: (B) Cluster 1 with a MCODE score = 52.6. (C) Cluster 2 with a MOCDE score = 32.8. (D) Cluster 3 with a MCODE score =22.6.

Discussion

In this study, we examined the role of a relevant airway pathogen as a trigger of molecular responses in sinonasal epithelium and characterized regulatory gene expression and epigenetic programs that are important in CRSwNP. Using an in vitro model, we showed for the first time that differential epigenetic remodeling occurs in the airway epithelium of CRSwNP patients in response to RV infection, possibly due to intrinsic differences in the epigenetic “wiring” of these critical cells that form the mucosal interface with the environment. Our results further suggested that these CRSwNP-related programs led to dysregulated immune responses to RV, a hallmark of this burdensome disorder. Overall, we identified over 6000 DMCs and 91 biological pathways that may be involved in the CRSwNP, indicating that an altered regulatory response to RV infection results from fundamental differences in the epigenetic landscape in airway epithelial cells and that these differences may drive chronic inflammation in CRSwNP.

Several lines of evidence support these conclusions. First, we observed many DMCs between cases and controls in the vehicle-treated samples (Fig 2E), indicating that epigenetic differences are present after cryopreservation and growth in culture, but in the absence of RV infection, potentially reflecting intrinsic properties of the sinonasal epithelium of individuals with CRSwNP. Second, the vast majority of DMCs were among the three CRSwNP-associated WGCNA modules and distinct from signatures associated with RV treatment. Third, the DMCs with significant CRS x RV interaction effects were enriched for TFBSs, suggesting that variation in these ‘preprogrammed’ DNA methylation profiles may influence the binding of over 100 different transcription factors³⁵ and downstream expression of their gene targets. Taken together, these findings revealed a central role for epigenetic signatures in the airway epithelium in the pathogenesis of CRSwNP.

Our systems biology approach identified signatures of co-regulated genes that are enriched in pathways previously implicated in CRS or microbial-immune response mechanisms. Of particular note are the two modules (black and brown) that were enriched for DMCs with CRS x RV interaction effects, indicating that the responses to RV significantly differed between the cases and controls. The enrichment for these DMCs at TFBSs suggest that this co-methylation network may direct gene pathways involved in CRSwNP that are triggered by RV. The lack of co-expressed genes in the brown module suggests that the 2,284 DMCs in this CRS-associated module may have temporal-specific effects on gene expression outside of the 48-hour treatment time point of our cell culture model or during the early stages of disease onset. That is, the DMCs in this module may represent stable, long-term epigenetic states that impact CRSwNP onset (etiology) or progression (pathogenesis) under specific conditions that differ from the context of RV treatment in our model. The finding that the DMCs with interaction effects were also enriched in the RV-correlated black module further supports the notion of an epigenetic regulatory mechanism for genes enriched in cell proliferative pathways in response to RV.

The design of this study provided some advantages for downstream analyses. First, using primary airway epithelial cells from CRSwNP and non-CRS controls to model transcriptional and epigenetic responses to RV infection allowed us to make novel observations of regulatory response mechanisms that are related to CRSwNP. Second, by using the same samples for gene expression and DNAm analysis ensured that the results from both analyses were directly comparable and reflect the same biological condition, thereby reducing the variability introduced by inter-individual differences and controls for potential confounding effects. However, our study had some limitations. First, although we focused on a CRS-relevant tissue cell type (epithelial cells), these studies were in isolation from the many cell types (immune, goblet, others) that may affect mucosal immune response and contribute to CRSwNP. As a result, our model only partially captured physiologic RV-responses and case-control differences that are present in vivo. Second, in vivo studies of molecular responses to RV in CRSwNP face challenges due to a) variability in exposure to pathogens which can confound or even mask inter-individual responses to RV infection and b) heterogeneity in clinical disease types (e.g. CRSsNP and CRSwNP, as well as additional subtypes such as allergic fungal rhinosinusitis [AFRS], aspirin-exacerbated respiratory disease [AERD], cystic fibrosis, primary ciliary dyskinesia, and immune deficiencies^{36, 37}). In vitro cell culture models address these limitations by allowing for controlled environmental exposures in relevant cell types, and for precise study designs that minimize confounders to an extent that would be impossible in ex vivo studies. Although the type of cell culture method could affect epigenetic responses (e.g., air-liquid interface cultures³⁸), the inherent epigenetic patterns in the mucosal epithelium from CRSwNP should be present in the same cell type in other cell culture methods. Moreover, while the EPIC array is highly accurate and provides valuable insights into the epigenetic patterns associated with RV infection in CRSwNP, analyses are limited to the CpGs and genomic regions that were selected. In particular, arrays have coverage of the human methylome and therefore reduced sensitivity compared to more comprehensive sequencing-based methods. Third, our sample sizes were small, and our power was limited to detect interaction effects (CRS x RV). It is possible therefore, and even likely, that many more interactions would be detected in larger samples. Finally, this cross-sectional study at a single time point does not allow us to determine whether the epigenetic patterns associated with differential response to RV that is related to CRSwNP preceded and contributed to the onset of disease or was a consequence of the disease state itself. Longitudinal studies that collect ex vivo tissues are needed to address this important question. Similarly, we could not determine whether epigenetic profiles related to RV infection preceded transcriptional changes or if activation of transcription after RV infection led to changes in DNA methylation, as has been shown in innate immune cells responses to bacterial infection³⁹.

In summary, we report for the first time that epigenetic programs inherent to the mucosal epithelium from CRSwNP patients may play an important role in pathogen response pathways and lead to chronic inflammation of the upper airway, which has great public health consequences⁴⁰. Focus on the transcriptional networks that are correlated with DNAm and the effects on transcription factor binding in the future could identify novel drug targets or therapeutic modalities specific to CRSwNP.