Genetic and non-genetic factors affecting the expression of COVID-19 relevant genes in the large airway epithelium

Particular host and environmental factors influence susceptibility to severe COVID-19. We analyzed RNA-sequencing data from bronchial epithelial brushings - a relevant tissue for SARS-CoV-2 infection - obtained from three cohorts of uninfected individuals, and investigated how non-genetic and genetic factors affect the regulation of host genes implicated in COVID-19. We found that ACE2 expression was higher in relation to active smoking, obesity, and hypertension that are known risk factors of COVID-19 severity, while an association with interferon-related inflammation was driven by the truncated, non-binding ACE2 isoform. We discovered that expression patterns of a suppressed airway immune response to early SARS-CoV-2 infection, compared to other viruses, are similar to patterns associated with obesity, hypertension, and cardiovascular disease, which may thus contribute to a COVID-19-susceptible airway environment. eQTL mapping identified regulatory variants for genes implicated in COVID-19, some of which had pheWAS evidence for their potential role in respiratory infections. These data provide evidence that clinically relevant variation in the expression of COVID-19-related genes is associated with host factors, environmental exposures, and likely host genetic variation.

SPIROMICS cohort 13 (n = 163), notable for the high burden of COVID-19-relevant comorbidities and rich phenotype and whole genome sequencing (WGS) data from the TOPMed Project 14 . For replication, we used two asthma RNA-seq datasets, SARP (n = 156) and MAST (n = 35) as well as eQTL data from GTEx 15 .
We first analyzed expression levels of ACE2, the receptor of the SARS-CoV-2 Spike protein that is the key host gene for viral entry 16,17 . Corroborating previous reports 12, [18][19][20] , we found that current smoking, when compared to non-smoking, had the largest overall effect on ACE2 expression of any phenotypic feature studied in SPIROMICS, before and after adjustments for covariates (log2 fold change (FC) = 0.30 ± 0.06, P = 1.7x10 -7 , Fig. 1a, Supplementary Table 1). This effect was absent in former smokers. In similarly adjusted models, we found no association between ACE2 levels and Chronic Obstructive Pulmonary Disease (COPD, Extended Data Fig. 2a), nor with asthma in MAST 20 (Extended Data Fig. 2b). In SARP, ACE2 levels were slightly lower in asthmatics compared to healthy controls (Extended Data Fig. 2c), which was largely driven by decreased expression of ACE2 only in asthmatics on oral steroids (Extended Data Fig. 2d).
African American race was associated with increased ACE2 expression in both SPIROMICS and SARP, but no association after adjusting for covariates suggests that this was due to a higher prevalence of comorbid conditions (Extended Data Fig. 2e-f). However, ACE2 expression was significantly higher across datasets in association with two relevant comorbidities, obesity and hypertension ( Fig. 1b-c, Extended Data Fig. 3a-e, Extended Data Fig. 4a-b). Of note, we further found that use of anti-hypertensives in SPIROMICS attenuates the association between ACE2 and hypertension towards levels seen in non-hypertensive participants (Fig. 1c). When stratified by anti-hypertensive class, angiotensin receptor blockers (ARBs) and diuretics, but not ACE inhibitors or calcium channel blockers, were associated with lower ACE2 levels, partially dependent on smoking status (Extended Data Fig. 4c). Counterintuitively, modest decreases in ACE2 expression were seen in SPIROMICS in association with age (log2 FC = -0.064 ± 0.02, P = 0.005 for every 10-year age increase, Fig. 1e) and male sex (log2 FC = -0.076 ± 0.035, P = 0.03, Fig. 1d) before and after adjustments, although similar associations were not seen in SARP or MAST.
. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) Boxplots showing that ACE2 log2 gene expression (x-axis) was increased in association with current but not former smoking as compared to never smokers (a), obesity (b, validated in the MAST and SARP cohorts, Extended Data Fig. 3a-b), hypertension (c, adjustments include anti-hypertensive treatment, validated in SARP, Extended Data Fig. 4a, data not collected in MAST), and female sex (d, not replicated in either MAST or SARP, data not shown). e, In SPIROMICS, ACE2 gene expression was modestly decreased in association with advancing age, but this finding was not replicated in either MAST or SARP (Supplementary Table 1A). f, Scatterplots showing that ACE2 gene expression increased in association with higher levels of our previously validated gene signatures of the airway epithelial response to interferon (left panel) and to IL-17 inflammation (right panel) after adjusting for smoking status. Both these findings were replicated in MAST and SARP (Supplementary Table 1B). g, Boxplots showing that ACE2 Exon 1c, which contributes to the truncated ACE2 transcript is differentially increased in association with our interferon signature while Exons 1a and 1b that contribute to the full length transcript are not associated. h, Scatterplots showing increased Exon 1a usage with advancing age that is not observed for Exon 1b and 1c. P-values indicated by: ****<0.0001, ***<0.001, **<0.01, *<0.05, ns=not significant in linear models adjusted for covariates.

Fig. 1: ACE2 gene expression associations in SPIROMICS. a-d,
. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted October 4, 2020. . https://doi.org/10.1101/2020. 10.01.20202820 doi: medRxiv preprint As chronic airway inflammation, prevalent but heterogeneous in the airway diseases studied in the included cohorts, can influence gene expression and the host response to infections, we next studied how stereotypic adaptive airway immune responses affect ACE2 expression. We used our previously validated gene expression signatures to quantify Type 2-, Interferon-, and IL-17associated inflammation [21][22][23] . We found that ACE2 expression was associated with increased interferon-related inflammation, as previously reported 10 , as well as IL-17-related but not Type 2 inflammation across datasets (Fig. 1f). Corroborating the association with IL-17 inflammation, genes highly co-expressed with ACE2 expression included genes in our IL-17 signature across datasets (Supplementary Table 2). Recent reports suggested that ACE2 induction by interferon stimulation may be explained by expression of a truncated ACE2 isoform (initiated from exon 1c instead of 1a/b) that does not bind the SARS-CoV-2 spike protein 24,25 . We first corroborated this finding, showing that our interferon-stimulated gene signature is associated with increased exon 1c but not exons 1a or 1b usage (Fig. 1g). We also identified an increase in exon 1a usage with age suggesting that despite a decrease in overall ACE2 expression in association with age, expression of the full length ACE2 transcript may not be decreased (Fig. 1h). Importantly, differential exon 1c usage was not associated with any other clinical/biological outcomes of interest.
These results overall indicate that smoking, obesity, and hypertension affect airway epithelial expression of functional ACE2 isoforms, as previously shown for smoking 12,18,19 . The ACE2 association with interferon-related inflammation appears to be explained by the truncated version of ACE2 24,25 . Together these findings suggest that smoking, obesity and hypertension may contribute to COVID-19 severity through an association with increased ACE2 expression, while other risk factors such as male sex and airway disease likely contribute via other mechanisms, corroborating recent evidence on sex differences in the immune response to COVID-19 26 .
As the host's ability to mount an appropriate response to respiratory viruses may alter susceptibility to severe infection, we next performed gene set enrichment analyses (GSEA) to determine whether clinical risk factors are associated with similar airway gene expression patterns indicative of a diminished immune response that we recently identified early in COVID-19 by nasal/oropharyngeal swab 27 . As we previously reported 27 , the genes differentially expressed in association with SARS-CoV-2 infection compared to other viruses at diagnosis indicate a diminished innate and adaptive immune response that may allow for unabated viral infection and account for the long pre-symptomatic period associated with COVID-19. We hypothesized that clinical risk factors uniquely associated with COVID-19 severity (e.g. cardiovascular disease, . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 4, 2020.  Table 3). GSEA was then performed using FGSEA 28 in which these gene sets were tested against gene lists ranked by their log fold change differential expression in association with comorbid clinical risk factors.
We found that the genes most downregulated in association with SARS-CoV-2 infection as compared to other viruses were significantly enriched amongst genes downregulated in association with obesity, hypertension, and cardiovascular disease in SPIROMICS (Fig. 2a-c).
Findings for obesity were replicated in SARP and MAST, and for hypertension in SARP infections, which often trigger airway disease exacerbations by potentiating the chronic airway inflammation associated with these diseases and smoking exposure. We found this same pattern in association with asthma in MAST but not when considering asthma overall in SARP, potentially due to heterogeneity of its asthma subjects. When considering just asthmatics with uncontrolled symptoms or those on inhaled compared to no steroids (a marker of severity) we did find this same enrichment of genes up and downregulated in association with non-COVID viral infections (pathway enrichment shown in Fig. 2g).
. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 4, 2020. .  CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 4, 2020. . statistic with a significant finding indicated when the line crosses the dashed line at either end of the plot. Genes downregulated by SARS-CoV-2 infection compared to other viruses were significantly enriched amongst genes downregulated in association with cardiovascular conditions overall (a), hypertension (b), and obesity (c), while in Current (d) and Former smoking (f) and in COPD (e) these downregulated genes in COVID-19 were enriched among upregulated genes in association with comorbidity. ** indicates FDR < 0.05. g, COVID-19-related pathway gene sets were generated from an IPA analysis of the genes up and downregulated by SARS-CoV-2 infection compared to other viruses. Gene set enrichment scores for gene sets enriched at FDR < 0.05 (columns) are shown in the heatmap plotted against comorbidities (rows) with gene sets enriched amongst downregulated and upregulated genes indicated in blue and yellow, respectively. All pathways not enriched at FDR < 0.05 were shrunk to zero (white). Euclidean distance with average linkage was used for clustering.
We used pathway gene set enrichment to determine the potential biological significance of these findings. We found across datasets that pathway gene sets derived from genes downregulated by SARS-CoV-2 infection as compared to other viruses were also enriched amongst genes downregulated in association with obesity, hypertension, cardiovascular disease, and aging (FDR < 0.05, Fig. 2g, Supplementary Table 4). Enriched downregulated pathways included those related to pro-inflammatory cytokines such as IL-6 and IL-17 as well as macrophage and granulocyte activation. Furthermore, pathways related to cardiovascular and metabolic disease signaling such as atherosclerosis and diabetes signaling were also enriched. We confirmed the enriched findings by separately performing IPA canonical pathway analyses on the genes differentially expressed (P < 0.05) in association with these comorbidities, finding similar results in these global/unsupervised analyses (Supplementary Table 5). Conversely, pro-inflammatory airway conditions such as smoking and COPD led to opposite effects. These findings suggest that obesity, hypertension, cardiovascular disease, and age are associated with a relative COVID-19-relevant immunosuppression at the airway epithelium, which, by stunting early anti-viral host responses, could contribute to increased susceptibility to SARS-CoV-2 infection and disease severity.
In order to map host genetic variants, we focused on 496 genes implicated in SARS-CoV-2 infection (Supplementary Table 6): ACE2 and TMPRSS2, key genes for viral entry 16 ; CTSL, CTSB, and BSG, which may have a role as alternative routes for viral entry 16,29 ; host genes with protein-protein interactions with viral proteins 30 ; differentially expressed genes as a response to the infection in cultured airway epithelial cells 31 ; genes involved in autophagy that might counteract viral infection 32 ; and other high interest genes from the COVID-19 Cell Atlas (www.covid19cellatlas.org). Our cis-eQTL mapping in SPIROMICS (n = 144) identified significant (genome-wide FDR < 0.05) genetic regulatory variation for 108 (21.8%) of these COVID-19-. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted October 4, 2020. . related genes (Fig. 3a, Supplementary Table 7), with many genes also having significant eQTLs in other tissues in GTEx 15 (Supplementary Table 8). Given the sample size, we have good power to discover the vast majority of eQTLs with >2-fold effect on gene expression 15 . Many of the genes have a substantial genetic effect on gene expression: for example, the MERS receptor DPP4 33 has a cis-regulatory variant rs6727102 where the alternative allele decreases expression by 3.3fold (Fig. 3a). In 16 genes, the genetic regulatory effects were >50% of the magnitude of the differential expression induced by SARS-CoV-2 infection (Fig. 3b). While the key genes ACE2 or TMPRSS2 did not have eQTLs in bronchial epithelium (Extended Data Fig. 6a-b), as previously reported 20 , TMPRSS2 has an eQTL in GTEx lung tissue. This is consistent with the lack of phenome-wide association signals 34 or COVID-19 GWAS association at these loci 9 , suggesting that genetic regulation of these two genes is unlikely to contribute to potential host genetic effects on COVID-19. Many of the genes analyzed for eQTLs had variation in expression associated to clinical factors and comorbidities, with current smoking associated with the highest number of upand down-regulated genes in association with comorbidity (Extended Data Fig. 7a-b). Compared to ACE2, the effect of current smoking on the expression of TMPRSS2 was modest (Extended Data Fig. 6c), and as previously reported 11 , expression levels of TMPRSS2 were higher in asthmatic than healthy controls, but not in COPD, and it decreased in association with steroid use (Extended Data Fig. 6d).
Cis-eQTLs from bronchial epithelium replicated at a high rate in those tissues from the GTEx v8  Table 9). In total of 143 genes with eQTLs in SPIROMICS were not tested in GTEx nor eQTLGen Consortium 35 , since bronchial epithelium is not well represented in previous eQTL catalogs. In addition to standard cis-eQTL mapping, we mapped cell type interacting eQTLs 36 but none were discovered for the COVID-19-related genes.
. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted October 4, 2020. . Fig. 3: eQTLs in bronchial epithelium. a, Effect size measured as allelic fold change (aFC, log2) of the significant eQTLs for COVID-19 candidate genes. Error bars denote 95% bootstrap confidence intervals. b, Comparison of the regulatory effects and the effect of SARS-CoV-2 infection on the transcription of COVID-19 candidate genes in normal bronchial epithelial cells from Blanco-Melo et al. 31 The graph shows regulatory effects as aFC as in a, and fold change (log2) of differential expression comparing the infected with mock-treated cells with error bars denoting the 95% confidence interval. Genes with adjusted P-value < 0.05 in the differential expression analysis are colored in black, genes with non-significant effect are colored in grey. Highlighted genes have eQTL effect size greater than 50% of the differential expression effect size on the absolute scale. DE -differential expression. c, Replication of cis-eQTLs from bronchial epithelium in GTEx v8 using the concordance rate (proportion of gene-variant pairs with the same direction . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted October 4, 2020. . To study the role of these regulatory variants in COVID-19 risk, we first analyzed eQTLs in the chromosome 3 locus with a significant association with hospitalization due to COVID-19 9 and severe COVID-19 with respiratory failure 6,8 . We found no significant eQTLs in the bronchial epithelium for any of the six genes in this locus (Extended Data Fig 6e), suggesting that this genetic association may be driven by other tissues or cell types with a role in COVID-19. Next, given that COVID-19 GWAS still have limited power, we analyzed how regulatory variants for COVID-19 relevant genes associate to other immune-or respiratory-related phenotypes in large GWAS. Indication of these variants affecting (respiratory) infections would provide hypotheses of variants that might play a role in COVID-19 risk and its comorbidities (Fig. 4a). Thus, we performed a pheWAS analysis by Phenoscanner v2 37,38 for the 108 lead cis-eQTLs for COVID19related genes and diverse set of phenotypes (Supplementary Table 10). Furthermore, we used the SPIROMICS phenotype data to study associations for 20 phenotypes (Supplementary Table   11). Of these loci, 44 were associated with at least one phenotype (P < 10 -5 ), with expected patterns -best powered GWAS traits having most associations and shared signals for highly correlated traits (Extended Data Fig. 9). We further used colocalization analysis to extract loci where the eQTL and GWAS signals are likely to share a causal variant, as opposed to spurious overlap, focusing on 20 loci with associations for hematological and respiratory system traits of which 12 colocalized (PP4 > 0.5, Fig. 4b, Supplementary Table 12). In Figure 4c, we show Interferon-induced transmembrane protein 3 (IFITM3) where the eQTL is associated with multiple blood cell traits of the immune system 39 and neutrophil counts in SPIROMICS (P < 0.002). This gene is upregulated by SARS-CoV-2 infection 31 and has a well-characterized role in the entry of multiple viruses, including coronaviruses 40 . Another interesting gene, ERMP1 (Fig. 4d) has an eQTL colocalizing with an asthma GWAS association in the UK Biobank. ERMP1 interacts with the SARS-CoV-2 protein Orf9c 30 , and severe asthma is a risk factor for COVID-19 hospitalization 6 and death 41 . An eQTL for the MEPCE gene that interacts with SARS-Cov-2 protein Nsp8 30 is associated with platelet parameters 39 (Fig. 4e). Interestingly, platelets are hyperactivated in . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted October 4, 2020. . https://doi.org/10.1101/2020.10.01.20202820 doi: medRxiv preprint COVID-19 42,43 and platelet count could be used as a prognostic biomarker in COVID-19 patients [44][45][46] . . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted October 4, 2020. . https://doi.org/10.1101/2020.10.01.20202820 doi: medRxiv preprint In summary, our findings demonstrate replicable associations between current smoking, obesity, hypertension and increased bronchial epithelial ACE2 expression potentially facilitating SARS-CoV-2 entry into host cells. While we do find evidence that the truncated ACE2 transcript is present in the bronchial epithelium, it does not appear to account for these associations. It is, however, likely that much of the inter-individual variation in COVID-19 is driven by a more complex molecular response to the virus in the airway epithelium than expression of ACE2 alone. This supposition is supported by our results demonstrating that obesity, hypertension, cardiovascular comorbidities, as well as aging, are associated with a downregulation of mucosal immune response pathways similar to that seen in early SARS-CoV-2 infection in comparison to other viral infections. Together with clinical data and Mendelian randomization analyses of the causal role of smoking and BMI on severe COVID-19 47 , our result suggest that these important comorbidities increase COVID-19 susceptibility and severity by creating an airway microenvironment in which SARS-CoV-2 can gain a foothold before an effective host response is mounted. Conversely, we find that inflammatory airway conditions increase both innate and adaptive immune responses, potentially priming individuals for airway disease exacerbations in response to other viruses but not SARS-CoV-2, which has a different immune profile and does not appear to trigger exacerbations of airway diseases. Furthermore, we show that host genetics has a biologically meaningful effect on the expression of many genes that may play an important role in COVID-19, including genes of interest as future drug targets. We pinpoint multiple COVID-19-interacting genes for which genetic regulatory variants associate with immune-or respiratory-related outcomes; these variants may be candidates for host genetic risk factors for COVID-19, or its severity. Altogether, our findings of genetic and non-genetic factors affecting the expression of COVID-19-related genes in bronchial epithelium provide essential insights for understanding inter-individual variation of COVID-19 and developing therapeutic targets for COVID-19.
. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted October 4, 2020. . https://doi.org/10.1101/2020.10.01.20202820 doi: medRxiv preprint  . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

Whole genome sequencing data
Trans-Omics for Precision Medicine (TOPMed) Project 14 data freeze 9 consist of whole genome sequences of 160,974 samples with at least 15x average coverage, including 2,710 individuals from the SPIROMICS study. We obtained unphased genotypes for all individuals from the SPIROMICS study at sites with at least 10x sequencing depth (minDP10 call set) aligned to the human reference genome build GRCh38. Details regarding the DNA sample handling, quality control, library construction, clustering and sequencing, read processing, and sequence data quality control are described on the TOPMed website (www.nhlbiwgs.org/genetic). Variants passing all quality control (QC) filters were retained.

Derivation of airway epithelial transcriptomic data in SPIROMICS, SARP, and MAST
Cytological brushings of the airway epithelium were obtained from lower lobe bronchi at the segmental or subsegmental carina. RNA was isolated with miRNeasy extraction kits (Qiagen Inc., Valencia, CA). RNA quantity and quality were evaluated using a NanoDrop Spectrophotometer

Differential expression analysis of ACE2 in relation to host/environmental factors
Visualization and analyses of single gene and gene signature analyses were done using RLE normalized and COMBAT batch corrected gene expression from the DESeq2 and SVA packages in R. Linear regression models were fitted to evaluate associations between ACE2 expression (based on normalized count) and clinical variables. In SPIROMICS unadjusted models were evaluated along with models adjusted for potential confounders including smoking status, age, sex, body mass index (BMI), and race, as appropriate. For analyses of hypertension, adjusted analyses included anti-hypertensives as a covariate, in addition to sex, age, smoking status and . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted October 4, 2020. . race. In SARP adjusted analyses included covariates for asthma, steroid use (inhaled steroids alone or inhaled plus oral steroids), age, sex, race, and BMI, as appropriate. Analyses of hypertension were done only in participants with asthma (due to availability of data), but adjustments for sex, age, BMI, and race were still done. In MAST analyses were adjusted for age, sex, asthma disease status, race, and BMI.

Differential exon usage
Following alignment, we indexed and sliced the SPIROMICS BAM files to include 51.6 kb of the ACE2 genomic region (chrX:15,556,393-15,608,016 in the hg38 genome build) using samtools 53 .
GTF files were manually curated to include the three exons that contribute to differential isoform expression of ACE2 24 . Full length ACE2 transcripts are generated from two independent first exons, Exons 1a and 1b, with Exon 1b shared between these transcripts. The truncated ACE2 transcript that does not bind the SARS-CoV-2 virus but is associated with an interferon-stimulated gene response in experimental models originates from Exon 1c. Coordinates from the pre-print manuscript by Onabajo, et al. 24  recommendations, raw exon counts were adjusted for gene counts to remove the signal from differential gene expression using the formula: (Exon Count in each sample*mean raw ACE2 count)/raw ACE2 gene count in that sample. To adjust for differences in sequencing depth between samples the transformed counts were then multiplied by the size factor variable generated by the DESeq2 package from the sequencing analysis. Linear models adjusting for batch were then used to analyze differences in exon usage in association with covariates of interest. The primary analysis was to evaluate whether ACE2 exon 1c differential usage was associated with increases in our interferon-stimulated gene signature. In secondary analyses we determined whether clinical covariates were associated with differential exon 1c usage.

Gene set enrichment analysis of expression changes induced by COVID-19
Differential expression and gene set enrichment analyses were done using the limma 55 and is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted October 4, 2020. . https://doi.org/10.1101/2020.10.01.20202820 doi: medRxiv preprint sets were built using the 100 genes most up-and downregulated in association with infection type. Biological pathway gene sets were built by inputting the genes differentially downregulated between SARS-CoV-2 infection and other viral illness (P < 0.05) into the Ingenuity Pathway Analysis canonical pathway function. The pathway assessments were limited to downregulated genes given the relationship between downregulated gene sets and comorbidities in the initial analyses of the 100 gene sets. Gene set enrichment analyses were then performed using FGSEA 28 and the CAMERA function 56 in limma against gene lists ranked by their log fold change differential expression in association with comorbid clinical risk factors. Barcode plots were made using CAMERA. Normalized enrichment scores for heatmaps were extracted from FGSEA (not available through CAMERA, but CAMERA and FGSEA statistical results were similar). As smoking so strongly influences gene expression, in SPIROMICS differential expression analyses input into GSEA algorithms to evaluate clinical factors such as obesity, hypertension, cardiovascular conditions, and sex were first done in former smokers only to limit the effect of smoking, adjusting for age, sex and BMI if appropriate. In sensitivity analyses, we repeated the analyses in all subjects, adjusting for smoking status as well and found similar results. As asthma and steroid use so strongly influence gene expression, in SARP differential expression analyses of these other clinical factors were limited to asthma participants on inhaled but not oral steroids.
Secondary analyses included all asthma participants adjusting for steroid use, with similar findings. Findings were considered significant at P < 0.05 and false discovery rate (FDR) < 0.05 if multiple corrections were necessary. For Extended Data Fig. 7, in which we evaluated COVID-19-related genes identified by experimental data from the SARS-CoV-2 ex vivo infection of primary human bronchial epithelial cells 31 or thought to be proteins that interact with SARS-CoV-2 30 , we reported findings at the less stringent P < 0.05 as these analyses were hypothesis generating only.

COVID-19-related genes
We mined the growing body of COVID-19 related literature to identify host genes implicated in SARS-CoV-2 infection discovered using different analytical approaches. The following studies were used to compose a list of COVID-19 candidate genes: 1) Hoffmann et al. 16 that identified ACE2 as the receptor to be exploited by the SARS-CoV-2 for cellular entry, and proteases TMPRSS2 and cathepsin B/L both to be used by SARS-CoV-2 for S protein priming, whilst only TMPRSS2 is essential for viral entry and viral spread; 2) Gordon et al. 30  . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted October 4, 2020. . https://doi.org/10.1101/2020.10.01.20202820 doi: medRxiv preprint Cell Atlas (www.covid19cellatlas.org) that highlights 17 genes including cathepsins and other viral receptors or receptor associated enzymes; 5) Gassen et al. 32 that showed the role of SARS-CoV-2 infection in restricting AMPK/mTORC1 activation and autophagy; 6) Wang et al. 29 that reported a mediating role of CD147 (also known as BSG) in SARS-CoV-2 viral invasion.
To narrow the list of differentially expressed genes following SARS-CoV-2 infection from Blanco-Melo et al. 31 , we focused on the results from the ex vivo infection of primary human bronchial epithelial cells. To include in our candidate list, we chose genes that 1) have adjusted P-value < 0.05 in the differential expression analysis from primary cells and either cell lines (Calu-3 or ACE2expressing A549 cells, low-MOI infection; excluded genes with adjusted P = 0) or samples derived from COVID-19 patients, and 2) log2 fold change > 0.5 in absolute scale in primary cells and log2 fold change > 1 in absolute scale in the other experiment.
In total, we selected 514 candidate genes implicated in COVID-19 from six different sources. Of them, 496 genes were expressed in bronchial epithelium.

Expression quantitative trait mapping
Expression quantitative trait (eQTL) mapping was performed in 144 unrelated individuals from the SPIROMICS bronchoscopy sub-study with WGS genotype data from TOPMed and gene expression from bronchial epithelium profiled with RNA-seq following the analysis pipeline from the Genotype-Tissue Expression (GTEx) Consortium 15 . Gene expression data was normalized as follows: 1) read counts were normalized between samples using TMM 57 with the edgeR package in R 58 , 2) genes with TPM ≥ 0.1 and unnormalized read count ≥ 6 in at least 20% of samples were retained, 3) expression values were transformed using rank-based inverse normal transformation across samples.
Next, Probabilistic Estimation of Expression Residuals 59 (PEER) was used to infer hidden determinants of variability in gene expression levels due to technical and biological sources.
According to the optimization analysis for selection of PEERs by sample size to maximize cis-eGene discovery done in GTEx 15 , 15 PEERs were chosen to be used as covariates in eQTL mapping together with 4 genotype PCs and sex imputed from genotype data.
To control population structure, principal component analysis (PCA) was conducted on postvariant QC genotype data from unrelated SPIROMICS individuals. More precisely, PCA was performed on a set of LD-independent autosomal biallelic single nucleotide polymorphisms from not long-range LD regions 60 with a call rate ≥ 99% and MAF ≥ 0.05 using smartpca from . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted October 4, 2020. permutations were used to correct for multiple testing, and false discovery rate (FDR) < 0.05 was used to identify genes with statistically significant eQTLs (eGenes). We also used tensorQTL to map conditionally independent cis-eQTLs.
Lead cis-eQTL effect size was quantified as allelic fold change 63 (aFC), ratio of expression of the haplotype carrying the alternative allele to expression of the haplotype carrying the reference allele of an eQTL. Gene expression data normalized with DESeq2 size factors 51 and log2transformed were used as input together with the processed genotype VCF file. aFC was calculated requiring at least 2 samples (--min_samps 2) and minimum 1 observation of each allele (--min_alleles 1), and adjusting for the same covariates as in cis-eQTL mapping. To calculate confidence intervals 100 bootstrap samples were used. aFC estimates that hit the absolute cap value (log2(100)) were set to missing.

Cell type interacting expression quantitative trait mapping
Firstly, we used xCell 64  Significance across genes was determined by adjusting eigenMT P-values using the Benjamini-Hochberg procedure with FDR of 0.05.

Replication of cis-eQTLs in GTEx
We performed replication of cis-eQTLs (gene-variant pairs) found from bronchial epithelium in the Genotype-Tissue Expression (GTEx) project v8 release 15 . Using cis-eQTL summary statistics across 49 tissues from GTEx, we calculated the proportion of true positives 66 , ! , to estimate the proportion of sharing of cis-eQTLs between bronchial epithelium and GTEx tissues. We assessed the allelic direction of the cis-eQTLs from bronchial epithelium and GTEx tissues by calculating concordance rate, the proportion of gene-variant pairs with the same allelic direction. This comparison was restricted to cis-eQTLs with nominal P-value < 1x10 -4 in the given GTEx tissue.
Next, we analyzed the replication and concordance measure as a function of sample size and median cell type enrichment scores for seven cell types 36 : adipocytes, epithelial cells, hepatocytes, keratinocytes, myocytes, neurons, and neutrophils. Tissues with median enrichment score > 0.1 were classified as being enriched for the given cell type. We used Wilcoxon rank sum test to estimate the difference in replication and concordance estimates between tissues enriched or not enriched for the given cell type.

cis-eQTLs not identified in previous large eQTL catalogs
To investigate the tissue-specificity of cis-eQTLs from bronchial epithelium, we performed genelevel lookup in GTEx v8 and eQTLGen Consortium 35 . We identified genes with significant regulatory effects in SPIROMICS (FDR < 0.05) that were tested in neither catalog.
Then, we used the functional profiling webtool g:GOSt (version e99_eg46_p14_f929183) from g:Profiler 67 to perform pathway analysis of the 492 significant eGenes in SPIROMICS not tested in GTEx v8 Lung. Method g:SCS was used for multiple testing correction corresponding to . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted October 4, 2020. . experiment-wide threshold of α = 0.05. Significant eGenes from SPIROMICS (n = 4,881) that have at least one annotation (option "Custom over annotated genes") were used as a background in the enrichment test.

pheWAS of lead COVID-19 cis-eQTLs in SPIROMICS
We performed phenome-wide association studies in 1,980 non-Hispanic White (NHW) and 696 individuals from other ethnic and racial groups from SPIROMICS for the 108 lead cis-eQTLs to evaluate for phenotypic associations with spirometric measures, cell count differentials, CT scan measures, eosinophil counts, and IgE concentrations were log-transformed. Significance threshold was set for the number of eQTLs tested across phenotypes (P < 4.63x10 -4 ).

Lookup of phenome-wide associations with PhenoScanner v2
PhenoScanner v2 37,38 was used to lookup phenotype associations for the cis-eQTL variants from large-scale genome-wide association studies (GWAS) with association P-value < 10 -5 . We queried PhenoScanner database based on the rs IDs of the lead cis-eQTLs obtained from dbSNP version 151 (GRCh38p7, including also former rs ID to query). The phenoscanner R package (github.com/phenoscanner/phenoscanner) was used to perform the queries. Query results were filtered to keep one association for each of the variants per trait, preferring summary statistics from newer studies, studies with larger sample size, or based on UK Biobank data (GWAS round 1 results from the Neale Lab). Description of Experimental Factor Ontology (EFO) terms and classification to EFO broader categories were obtained from the GWAS Catalog or by manually searching EMBL-EBI EFO webpage (www.ebi.ac.uk/efo/). The regulatory variants for CEP250, FAR2, and TLE3 have phenotypic associations with both body height and pulmonary function test (PFT) measures from PhenoScanner. As GWAS analyses from the Neale Lab using UK Biobank data do not include height as a covariate in the model, we used the results of the lung function GWAS by Shrine et al. 68 to confirm if the suggestive . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 4, 2020. . https://doi.org/10.1101/2020.10.01.20202820 doi: medRxiv preprint signal for PFT trait has been observed before or rather seems to be an artefact of incomplete adjustment for height. Of note, Shrine et al. 68 have discovered 279 lung functions signals in the meta-analysis of UK Biobank and the SpiroMeta Consortium. We looked up the nearest GWAS hits to the eQTL, and calculated LD between the variants in the African and European populations using LDpop 69 web tool.

Colocalization analysis
Multiple trait associations observed for a single variant do not necessarily translate into shared genetic causality. To assess evidence for shared causal variant of a cis-eQTL and a GWAS trait, we used the Bayesian statistical test for colocalization, coloc. We used the newer version of coloc 70 that allows conditioning and masking to overcome one single causal variant assumption (condmask branch of coloc from github.com/chr1swallace/coloc). We only tested colocalization for loci where the eQTL had at least one phenotypic association based on the lookup analysis with Phenoscanner from the following EFO parent categories: hematological measurement, pulmonary function measurement, respiratory disease. From each of the smaller EFO categories, we chose one trait with the smallest P-value for which we were able to find summary statistics using GWAS Catalog REST API or among the Neale Lab GWAS round 2 results (www.nealelab.is/uk-biobank/). Coloc was run on a 500-kb region centered on the lead cis-eQTL (+/-250 kb from the variant) with priors set to p1 = 10 -4 , p2 = 10 -4 , p3 = 5x10 -6 . We used the coloc.signals() function with mode = iterative and method = mask for GWAS traits with LD data from the 1000 Genomes Project to match the ancestry of the discovery population (e.g., choosing CEU for LD if the discovery population is of European ancestry). We allowed for a maximum of three variants to mask, with an r 2 threshold of 0.01 to call two signals independent and P-value threshold of 1x10 -5 to call the secondary signal significant. We used method = single for the eQTLs, because the corresponding eGenes did not have secondary independent signals. We prioritized eGenes with posterior probability for colocalization (PP4) > 0.5 as loci with evidence for colocalization.
. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 4, 2020. . the National Heart, Lung and Blood Institute (NHLBI). Genome Sequencing for "NHLBI TOPMed: SubPopulations and InteRmediate Outcome Measures In COPD Study" (phs001927) was performed at the Broad Institute Genomics Platform (HHSN268201600034I). Core support including centralized genomic read mapping and genotype calling, along with variant quality metrics and filtering were provided by the TOPMed Informatics Research Center (3R01HL-117626-02S1; contract HHSN268201800002I). Core support including phenotype harmonization, data management, sample-identity QC, and general program coordination were provided by the The authors thank the SPIROMICS participants and participating physicians, investigators and staff for making this research possible. More information about the study and how to access . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 4, 2020. . https://doi.org/10.1101/2020.10.01.20202820 doi: medRxiv preprint