Subject Enrollment and Cohort Characteristics
We enrolled 24 control and 39 PARDS subjects from April 1, 2018 to June 30, 2021. Subject characteristics are presented in Table 1, and Supplemental Table 1 provides patient-specific data and describes assays conducted on each specimen, and Supplemental Fig. 1 summarizes the analyzed specimens.
Our overall approach was (1) identify differences in the epigenetic state of PARDS vs. Control and between PARDS subjects, (2) characterize transcriptional differences, (3) compare the ability of nasal transcriptome vs. serum biomarkers to predict PARDS duration, and (4) identify potential regulatory nodes by reconciling transcriptomic and methylomic data.
The Methylation Patterns of Nasal and Bronchial Epithelial Cells Do Not Differ by Collection Site
We first compared DNA methylation of paired nasal and bronchial specimens. We analyzed 8 paired brushings from 3 PARDS and 1 control subjects. By site, 0.2% percent of regions were differentially methylated (DMRs) between the upper and lower airways with 48% corresponding to annotated CgP islands or shores, and 9% promoters (24 genes, Supplemental Table 2, Supplemental Fig. 2A&B). By k-means clustering, specimens were more similar by subject than by collection site (Supplemental Fig. 2C) indicating similarity of the nasal and bronchial methylomes.
PARDS Patients Have Hypomethylation of Promoters Related to Inflammation
We sought to identify patterns of DNA methylation in the nasal epithelium of PARDS and control subjects. Methylation analysis of day 1 specimens from 13 control and 20 PARDS patients showed segregation into two groups which we termed Methyl Subgroup 1 and Methyl Subgroup 2 (Fig. 1A&B). Methyl Subgroup 1 was comprised of 12 PARDS subjects and Methyl Subgroup 2 was composed of 8 PARDS and all control subjects (Table 2). There were no clear differences in age or sex between groups. Since exposure to of the nasal epithelium to oxygen might impact DNA methylation, we assessed which subjects had been treated with at least 24 hours of supplemental oxygen in the 7 days prior. Three Subgroup 1 and zero Subgroup 2 subjects had been exposed to > 24 hours of nasopharyngeal supplemental oxygen (i.e. most were intubated shortly after presentation). Among PARDS subjects, there were no substantial differences in PARDS severity, PARDS duration, or illness severity between Methyl Subgroups.
Methyl Subgroup 1 subjects had hypomethylation of genes potentially important in PARDS. Using a threshold of 25% differential methylation and an adjusted p-value threshold of 0.1, 0.8% of loci were differentially methylated. Compared to Methyl Subgroup 2, in Methyl Subgroup 1, 85 transcription start sites were hypo- and 24 were hypermethylated (Supplemental Table 3). Subgroup 1 hypomethylated genes were related to cell adhesion, and hypermethylated genes were related to cell metabolism (Fig. 1C). Thus, a subset of PARDS subjects have a localized increased chromatin accessibility for genes that might be important in inflammatory cell recruitment and epithelial cell function.
Methylation Changes Over Time
DNA methylation is considered a relatively stable epigenetic feature. We performed methylomic analysis on nasal brushings from control and PARDS subjects in which we had at least two specimens to assess Methylomic Subgroup stability. There were 9 control and 21 PARDS subjects with two or more evaluable specimens (70 total). Methyl Subgroup was stable in 15 of 21 PARDS and 9 of 9 control subjects with all control subjects remaining in Methyl Subgroup 2 (Fig. 2A). There was no consistent pattern in the six PARDS subjects with changes in Methylomic Subgroup (Fig. 2B). The nasal epigenome is stable in control subjects but changes in PARDS with respect to genes that might be important in PARDS pathobiology.
mRNA-Seq Data Filtering, Normalization and Batch Correction
One hundred forty-two nasal specimens from 56 subjects were sequenced, aligned, and normalized with removal of mitochondrial and ribosomal transcripts. Transcripts without at least five reads in half of specimens, and specimens without at least 100,000 reads and 5,000 unique transcripts were excluded from further analysis. Expression values of 10,846 transcripts in 129 specimens from 37 PARDS and 15 control subjects were normalized by batch prior to analysis (Supplemental Fig. 3A-C). Analyzing PARDS specimens suggested that six or seven principal component best characterized the data, and covariate analysis for PARDS severity, comorbidity, infectious etiology, direct lung injury, race, sex, and age only identified age as being significantly correlated with any component and this only explaining 0.6% of dataset variance (Supplemental Fig. 3D-E).
Nasal and Bronchial Transcriptomes are Dynamic
In comparing the transcriptomes of the nasal and bronchial epithelium, twelve specimens from five PARDS and one control subject were evaluable. By k-means clustering, principal component analysis, and Euclidean distance, specimens tended to cluster more closely by subject than by collection site (Supplemental Fig. 4). As in healthy adults , upper and lower conducting airway gene expression is similar in PARDS.
Four Different Transcriptional Patterns in the Nasal Epithelium of PARDS Subjects
We performed k-means clustering of each evaluable specimen and identified four subgroups within our dataset which we termed A, B, C, and D (Fig. 3A&B). Principal component 1 genes were largely related to ciliary cell function and Interleukin-4 (IL-4) signaling, and principal component 2 genes were largely related to inflammation and chemokine signaling (Supplemental Tables 4&5). We defined a differentially expressed gene (DEG) as one with 2-fold changed expression with an adjusted p-value of less than 0.1. Subgroups A, B, C, and D contained 1,375; 834; 841 and 2,038 DEGs and had substantial DEG overlap (Fig. 3C, Supplemental Table 6, Supplemental Fig. 5). GSEA showed Subgroup A enriched for IL-4, IL-10, and IL-13 signaling. Subgroup B had downregulation of ciliary function-related processes and upregulation of innate immune ones. Subgroup C had upregulation of these same ciliary processes and downregulation of IL-4, IL-10, and IL-13 signaling. Subgroup D had downregulation of both microtubule and innate immune processes and upregulation of processes related to epithelial integrity (Fig. 3D-F). In evaluating the abundance of cell-specific mRNAs, Subgroup A and B both had reduced abundance of ciliated cell mRNAs and Subgroup B had increased myeloid cell mRNAs. Subgroup C had a greater abundance of ciliated cell mRNAs compared to all others, and Subgroup D had reduction epithelial stem cell mRNAs (secretory and basal cells, Table 3). The only notable difference in clinical characteristics between the subgroups was a trend towards fewer ventilator free days in Subgroup B (Table 4).
In evaluating how PARDS Subgroups changed over time, Subgroups B and D were restricted to earlier time points with the proportion of specimens classified as Subgroup C increasing over time. Although the time intervals between specimens was not consistent, Subgroups B and D tended to transition to Subgroups A and C, and Subgroup C specimens remained consistent (Fig. 4A). Taken together, these data support a model in which Subgroups B and D represent two different modes of injury. Subgroup B is characterized by innate immune activation and ciliary dysfunction, and Subgroup D is characterized by epithelial dysfunction without inflammation. Both Subgroups transition to Subgroup A which is anti-inflammatory with increased mRNA levels of cytokines important in epithelial repair and differentiation (IL-4, IL-10, and IL-13). Subgroup C is homeostatic with restored epithelial function (Fig. 4B).
Comparison of PARDS and Control Nasal Transcriptomes
We reanalyzed the PARDS specimens and 27 control specimens from sixteen control subjects. Three control subjects developed lung injury (a new oxygen requirement) and one developed PARDS over the course of the study. This last subject clustered with Subgroup B, and 1, 3, and 3 specimens of the lung injury control subjects clustered with A, B, and C respectively. Among control subjects who did not develop lung injury, 3, 4, 11, and 1 clustered with A, B, C, or D respectively (Supplemental Fig. 6). These data suggest that the nasal transcriptome reflects lung injury in non-PARDS subjects and that the homeostatic state of the nasal epithelium is Subgroup C.
Microbiome Differences between Control and Two PARDS Subgroups
Although we did not identify any differences in viral infection between Transcriptomic Subgroups (Table 4) the presence of respiratory viruses might influence the nasal transcriptome. We first visualized day 1 and all PARDS nasal specimens by infectious agent (Supplemental Fig. 7A&B). For this analysis, we used a shotgun metagenomic approach in which read sequences are aligned to viral and bacterial genomes. In day 1 Subgroup A, B, C, and D specimens, 67%, 42%, 24%, and 40% had a diagnosed viral infection respectively; however, in analyzing all time points, 47%, 36%, 27%, and 33% had a diagnosed viral infection—consistent with dataset analysis showing that infectious agent did not explain a significant portion of data set variation (Supplemental Fig. 3). For more comprehensive assessment of the microbiome, we performed shotgun metagenomics with comparison by Transcriptomic Subgroup and PARDS severity. Microbial diversity was increased in nasal brushings of PARDS subjects, not different between Transcriptomic Subgroups, and greater in moderate PARDS compared to severe or no PARDS (Supplemental Fig. 7C&D). There were no differences in the fraction of bacterial or viral reads between Transcriptomic Subgroups (Supplemental Fig. 7E&F). The dissimilarity of control vs. Transcriptomic Subgroups was significant (p < 0.001) but limited to seven bacterial an no viral species (Supplemental Tables 7&8), and there were no significant microbiome differences by PARDS Severity. Neither bacteria nor specific viruses influence PARDS Transcriptomic Subgroup.
Nasal Transcriptomic Subgroup Predicts Prolonged PARDS Better than Serum Biomarkers
If our inflammation and epithelial cell injury model were correct, then we would expect longer PARDS duration in Subgroups B and D. Limiting our analysis to the first available specimen (Fig. 5A), we used days meeting PARDS criteria as a primary endpoint because a sizable number of subjects remained intubated after PARDS resolution (Fig. 5B) making ventilator free days a less-reliable measure of lung injury resolution. We found that the number of days meeting PARDS criteria was greater in Nasal Transcriptomic Subgroups B&D compared to Subgroups A & C (Fig. 5C&D, median 8 vs 2 days, p = 0.02).
There is much interest in serum biomarker panels for ARDS and PARDS subclassification. We quantified serum levels of seventeen serum biomarkers obtained at the time of nasal brushing. None of the seventeen assayed biomarkers had a significant association with Nasal Transcriptomic Subgroup (Supplemental Fig. 8). To compare the ability of serum biomarkers and nasal transcriptomic subgroups to predict continued PARDS, the optimal predictive threshold for each at each time point was determined by receiver operator characteristic, and sensitivity, specificity, positive predictive value, and negative predictive value were determined for each biomarker and Nasal Transcriptomic Subgroup B or D (Fig. 5E, Supplemental Fig. 9). The sensitivity of Nasal Transcriptomic Subgroup B or D to predict continued PARDS at up to 7 days was as good or better than any of the assayed serum biomarkers, although there was some diminishment over time. The specificity of Subgroup B or D with regards to continued PARDS was poor. The positive predictive value of Transcriptomic Subgroup was poor for continued PARDS at day 3 (60%), but gradually improved to 80% by day 10 while the positive predictive value of all serum biomarkers diminished over time. In contrast, the negative predictive value of Transcriptomic Subgroup at 3 days was very good at 92% but diminished to 38% in predicting continued PARDS at 10 days while biomarker negative predictive value improved for longer-duration PARDS. These data indicate that serum biomarkers and nasal transcriptomics yield different information with low serum biomarkers being specific for the rapid resolution of PARDS but high serum biomarkers lacking sensitivity for continued PARDS up to one week later. In contrast, transcriptomics lacks specificity but has good sensitivity for continued PARDS up to seven days and is more predictive of prolonged PARDS than serum biomarkers.
Functional Epigenetic Modules in the Nasal Epithelium of PARDS Subjects
To identify genes and functions that were coordinately controlled at the epigenetic and transcriptional levels, we performed functional enrichment for genes with increased mRNA and reduced methylated DNA or reduced mRNA and increased methylated DNA by Nasal Transcriptomic and Methylomic Subgroups. Although Methyl Subgroup 1 was composed entirely of PARDS subjects and Transcriptomic Subgroup B had longer PARDS duration, there was coordinate hypomethylation and increased mRNA levels of inflammatory genes in the combined B2 subgroup suggesting that epithelial inflammation is not regulated at the epigenetic level. In contrast, combined subgroup C2 had hypermethylation and reduced mRNA levels of immune-associated genes suggesting that this may be important for a return to homeostasis (Fig. 6A). This is consistent with hypomethylation and increased gene expression of cilia-associated genes in combined subgroup C2 (Fig. 6B). The genes for each combined subgroup analysis are listed in Supplemental Table 9 and additional GSEA analyses in Supplemental Fig. 10.