Lung Transcriptomics Link Emphysema to Barrier Dysfunction and Macrophage Subpopulations

Abstract

Rationale

Although many studies have examined gene expression in lung tissue, the gene regulatory processes underlying emphysema are still not well understood. Finding efficient nonimaging screening methods and disease-modifying therapies has been challenging, but knowledge of the transcriptomic features of emphysema may help in this effort.

Objectives

Our goals were to identify emphysema-associated biological pathways through transcriptomic analysis of bulk lung tissue, to determine the lung cell types in which these emphysema-associated pathways are altered, and to detect unique and overlapping transcriptomic signatures in blood and lung samples.

Methods

Using RNA-sequencing data from 446 samples in the Lung Tissue Research Consortium and 3,606 blood samples from the COPDGene study, we examined the transcriptomic features of chest computed tomography–quantified emphysema. We also leveraged publicly available lung single-cell RNA-sequencing data to identify cell types showing chronic obstructive pulmonary disease–associated differential expression of the emphysema pathways found in the bulk analyses.

Measurements and Main Results

In the bulk lung RNA-sequencing analysis, 1,087 differentially expressed genes and 34 dysregulated pathways were significantly associated with emphysema. We observed alternative splicing of several genes and increased activity in pluripotency and cell barrier function pathways. Lung tissue and blood samples shared differentially expressed genes and biological pathways. Multiple lung cell types displayed dysregulation of epithelial barrier function pathways, and distinct pathway activities were observed among various macrophage subpopulations.

Conclusions

This study identified emphysema-related changes in gene expression and alternative splicing, cell type–specific dysregulated pathways, and instances of shared pathway dysregulation between blood and lung.

At a Glance Commentary

Scientific Knowledge on the Subject

Prior studies have investigated the transcriptomic characteristics of emphysema and its associated biological pathways. However, less is known about alternative splicing mechanisms and cell type–specific transcriptional patterns in emphysema. In addition, a comparison between dysregulated genes and pathways in blood and lung tissues is needed to better understand the utility of noninvasive diagnostic and prognostic tools for emphysema.

What This Study Adds to the Field

Using lung samples from the Lung Tissue Research Consortium, we performed differential gene and alternative splicing association analyses for computed tomography–quantified emphysema. We then queried a previously published lung tissue single-cell RNA-sequencing atlas of patients with chronic obstructive pulmonary disease (COPD) and control subjects to determine lung cell type–specific expression patterns of the biological pathways identified from the bulk analyses. We also detected unique and overlapping transcriptomic signatures in Lung Tissue Research Consortium lung tissues and blood samples from the COPDGene study. We showed that multiple pathways, including oxidative phosphorylation and ribosomal function processes, were enriched in both blood and lung tissues. We also observed that in COPD, multiple lung cell types displayed dysregulation of epithelial barrier function pathways, and distinct pathway activities were observed among various macrophage subpopulations.

Chronic obstructive pulmonary disease (COPD) is a major source of morbidity and mortality (1). Emphysema, an important COPD phenotype, is independently associated with higher risks of cardiovascular disease, lung cancer, and mortality (2–5). Understanding the transcriptomic characteristics of emphysema may help develop efficient nonimaging screening methods and targeted therapies. Although linked to genes involved in transforming growth factor-β (TGF-β) signaling (6, 7), B cell immunity (6, 8, 9), and hypoxia (9–11), our knowledge of emphysema-associated alternative splicing mechanisms and cell type–specific pathways in human lung tissue remains limited.

Alternative splicing regulates gene expression by generating multiple protein isoforms from a single gene and is linked to many human diseases (12), including COPD (13–15) and emphysema (16). Despite the extensive use of animal models to study emphysema, the role of alternative splicing remains unaddressed, presenting a research gap. Understanding the impact of alternative splicing in emphysema could reveal disease mechanisms and new therapeutic targets. In addition, whereas many pathological processes have been linked to emphysema, it is unclear whether other biological processes may also be at play. It is also less obvious which cell types exhibit pathway dysregulation in the lungs of subjects with emphysema compared with control subjects, although it is widely recognized that many cell types, including neutrophils (17, 18) and T lymphocytes (19, 20), are drivers of the disease. Furthermore, comparing dysregulated genes and pathways in blood and lung tissues is necessary to better understand the utility of noninvasive diagnostic and prognostic tools for emphysema.

In the present study, we hypothesized that there are significant emphysema-associated transcriptomic biomarkers and pathways identified from lung tissue, cell type–specific signatures, and important similarities and differences between lung tissue and blood. To test these hypotheses, we used RNA-sequencing (RNA-seq) data from lung tissue samples in the Lung Tissue Research Consortium (LTRC) to identify genes and alternative splicing mechanisms associated with computed tomography (CT)-quantified emphysema. We then queried a previously published single-cell RNA-seq atlas of lung tissues of patients with COPD and control subjects to determine which cell types are associated with these pathways. Last, we compared emphysema-associated transcriptomics in LTRC lung tissue samples with those in whole-blood samples in the COPDGene (Genetic Epidemiology of COPD) study. Some of these results were previously reported as an abstract and a preprint (21, 22).

Methods

Study Description

We obtained lung tissue samples from the NHLBI LTRC (www.nhlbi.nih.gov/science/lung-tissue-research-consortium-ltrc, https://biolincc.nhlbi.nih.gov/studies/ltrc/). Details regarding subject recruitment were previously published (23). Participants included smokers and nonsmokers over the age of 21 who underwent surgical lung biopsy, lung volume reduction surgery, lung transplant, or lung nodule/mass resection. These subjects cover the entire spectrum of the Global Initiative for Chronic Obstructive Lung Disease spirometry grading system (24). We excluded subjects missing any of the following clinical data: CT-quantified emphysema, CT scanner model, FEV₁, body mass index (BMI), current smoking status, and pack-years of smoking. We also excluded subjects with idiopathic pulmonary fibrosis (IPF), defined as a consensus clinical diagnosis of IPF or a pathologic diagnosis of usual interstitial pneumonia or honeycomb lung without a clinical diagnosis of another interstitial lung disease (25). All samples were sequenced through the NHLBI TOPMed program.

All analyses conducted on LTRC data were also performed using whole-blood RNA-seq data from the COPDGene study, employing the same covariates for comparison (16). COPDGene is a longitudinal study investigating the epidemiologic and genomic characteristics of COPD. This study included 10,371 non-Hispanic White and African American subjects from 21 U.S. clinical centers, aged 45–80 at study enrollment, with an average of 44 pack-years of cigarette smoking (NCT00608764, www.copdgene.org) (26). Each study visit collected spirometry data, questionnaires, and chest CT scans. Whole-blood RNA-seq was obtained at the 5-year follow-up visit (visit 2). All subjects provided written informed consent, and all clinical sites received institutional review board approval.

Emphysema Quantification

We used Analyze 8.1 (www.analyzedirect.com) (LTRC) (27) and Thirona (www.thirona.eu) (COPD) to quantify emphysema, measured as the Hounsfield units at the 15th percentile of the CT density histogram at end inspiration + 1,000 (Perc15) (28, 29). Lower Perc15 values indicate more emphysema. Our primary analysis treated emphysema as a continuous variable, and we also categorized subjects into high- and low-severity groups on the basis of whether their adjusted Perc15 density was below or above the median.

Differential Expression and Usage Analyses

We used the limma-voom linear modeling (as implemented in limma version 3.46.0) to test associations between emphysema and RNA transcripts (30, 31). “Differential expression” refers to changes in absolute expression levels of a feature. “Differential usage” captures alternative splicing, indicating changes in relative expression levels of isoforms/exons within a given gene. All models were adjusted for age, race, sex, BMI, pack-years, current smoking status, FEV₁, CT scanner model, and library preparation batch. The COPDGene analysis also included white blood cell proportions as covariates.

Pathway Analysis

We used the Ensemble of Gene Set Enrichment Analyses (egsea) software (version 1.18.1) for gene set enrichment analysis (GSEA) on the differential gene expression (DGE) results (32), using Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway gene sets as the reference. The CAMERA approach was employed, which adjusts the gene set test statistic using intergene correlation (33, 34). Pathway directionality was determined by the majority of upregulated or downregulated genes (32). Genes corresponding to these pathways were mapped using the pathview and GAGE libraries in R (35, 36). Pathview summarizes multiple gene entries within a node using the sum method, displaying the summed log fold change values. Some enriched pathways could not be displayed because of the lack of manually drawn reference maps in the KEGG database for certain species.

Because egsea requires effect sizes and differentially used isoforms (DIUs) and differentially used exons (DEUs) analyses lack a singular effect size (because when one isoform increases, often another decreases), we used topGO Gene Ontology (GO) enrichment for the DIU and DEU results because it does not need effect sizes (37). We also performed topGO for the DGE results for comparison. We reported GO pathways with at least three significant genes and P values <0.05.

Comparison of Our Lung DGE Results with Previous Literature

We compared our DGE results with recent COPD and emphysema transcriptomics studies by Ghosh and colleagues, Campbell and colleagues, and Rojas-Quintero and colleagues (38–40). We evaluated the validation rate of the significant findings from these publications using a P value threshold of <0.05 in our study and a consistent direction of effect.

Cell-Type Specificity Analysis

We reanalyzed data from a previously published lung tissue single-cell RNA-seq experiment to test for differential expression of significant pathways from the bulk lung DGE analyses within individual lung cell types (41, 42). The dataset consists of 312,928 single cells from 78 human samples (28 control, 18 end-stage COPD, 32 IPF). The control samples in this cohort represented donor lungs deemed unsuitable for transplant. Only control and end-stage COPD samples were used for this analysis, keeping the author-provided cell type annotations.

We used the AddModuleScore function from the Seurat package to summarize gene expression signatures for pathway members (43). This function calculates a normalized average expression for a set of pathway genes, subtracting the aggregate expression of control genes. All genes were binned by expression levels, and expression-matched control genes were randomly selected for each pathway. Pathway activity within each cell was computed by subtracting the average control gene expression from the pathway gene expression. We calculated pathway activity scores for the 33 KEGG pathways identified by GSEA across all cells. We compared these scores between controls and end-stage COPD for each cell type using two-sample t tests. Odds ratios for pathway activation were computed for each cell showing activity (score, >0) versus those not showing activity in a given cell type and for the remainder of the data. The odds ratio was calculated as (p × n)/(q × m), representing the fraction of cells from a given cell type showing pathway activity versus the remaining cells. The P values were calculated using Fisher’s exact test. The negative log₁₀ values of the corrected P values were capped at 100 and plotted using the ComplexHeatmap package (44).

Reclustering of the Macrophage Populations

Recognizing that M1/M2 macrophage categorization is overly simplistic (45–49), we reclustered macrophage populations in the single-cell RNA-seq data from Sauler and colleagues (42). We used the sccoda package (50) for differential abundance analysis to examine variations in macrophage subpopulations between COPD cases and control subjects. We then examined the enriched pathways identified through our bulk RNA-seq analysis, generating pathway expression scores for each cell using the AddModuleScore function from the Seurat package (43). These scores were then compared between cells originating from COPD and control subjects using the t test.

Comparison of Lung Tissue and Blood Analyses

To compare DGE between lung tissue and blood, we computed Pearson’s correlation coefficient for the log fold changes of significant DGE in both datasets and determined statistical significance using t tests. Similar analyses were performed for DIUs and DEUs. We also counted overlapping DGEs, DIUs, and DEUs using Fisher’s exact test to determine if the overlap was greater than expected by chance.

We used weighted gene coexpression network analysis (WGCNA) to examine gene expression patterns in LTRC lung tissues (51, 52). Subsequently, using the LTRC dataset as the primary reference, we investigated module preservation in COPDGene to ascertain whether the emphysema-associated modules identified in LTRC lung tissues recapitulate in COPDGene blood samples (53). We computed module eigengenes, the first principal component of the gene expression matrix for the genes (51). Multivariable regression models assessed associations between module eigengenes and emphysema, adjusting for age, race, sex, FEV₁, BMI, current smoking status, pack-years, and CT scanner model. GSEA was performed on each emphysema-associated module using GeneAnswers (54).

To ascertain the preservation of lung coexpression network associations in blood, we projected the LTRC network onto COPDGene blood RNA-seq data by calculating blood module eigengenes using COPDGene expression data for the lung module genes. Following the lung analysis methodology, we then used multivariable regression models to examine associations between these blood eigengenes and emphysema.

Statistical Analyses

Data are reported as means with standard deviations or counts with percents. The Kruskal-Wallis test was used for the continuous variables, and the chi-square test was used for the categorical variables. Upregulated versus downregulated genomic features or pathways were provided with respect to their relationships with emphysema. Because Perc15 decreases with more severe emphysema, log fold changes were multiplied by −100 to make positive changes represent upregulated genes. Thus, the fold change estimates correspond to expression change per 100–Hounsfield unit decrease in lung density (i.e., increasing emphysema). The Benjamini-Hochberg method corrected for multiple comparisons with a significance threshold of a 10% false discovery rate (FDR) (55).

Results

Study Subjects

The study flow diagram is provided in Figure 1, and covariate missingness is provided in Figure E1 in the online supplement. A total of 446 LTRC subjects were included in this analysis. Of the 1,502 LTRC subjects with available RNA-seq data, 499 had comprehensive clinical information and CT-quantified emphysema. We further excluded 53 subjects with IPF. Table 1 shows the demographics and clinical characteristics of the included subjects, who were mostly non-Hispanic White individuals, with a mean age of 64, mean BMI of 28.1 kg/m², and balanced sex representation. Eight percent were current smokers, averaging 35.6 pack-years. Included subjects had more smoking and slightly less spirometric impairment than excluded ones (Table E1). The baseline characteristics of the 446 LTRC and the 3,606 COPDGene individuals were also comparable (Table 1), as were their Perc15 value distributions (Figure E2). The COPDGene study flow diagram is provided in Figure E3.

Figure 1.

Flow diagram illustrating the selection process for the Lung Tissue Research Consortium study subjects. BMI = body mass index; IPF = idiopathic pulmonary fibrosis; LTRC = Lung Tissue Research Consortium; QC = quality control; RNA-seq = RNA sequencing; Perc15 = Hounsfield units at the 15th percentile of computed tomographic density histogram at TLC + 1,000.

Table 1.

Baseline Characteristics of Included Lung Tissue Research Consortium and COPDGene Study Participants

	LTRC	COPDGene	P Value
Sample size	446	3,606
Age, yr	64.0 (11.1)	65.0 (8.8)	0.90
Sex, n (% male)	232 (52.0%)	1,832 (50.8%)	<0.001
Race			<0.001
Non-Hispanic White	390 (87.4%)	2,633 (73.0%)
Asian	0 (0.0%)	0 (0.0%)
African American	43 (9.6%)	973 (27.0%)
Hispanic	11 (2.5%)	0 (0.0%)
Other	2 (0.4%)	0 (0.0%)
BMI, kg/m2	28.1 (6.0)	28.9 (6.2)	0.02
Current smoker	37 (8.3%)	1,294 (35.9%)	<0.001
Smoking pack-years	35.6 (36.5)	41.3 (25.5)	<0.001
FEV1, % predicted	72.4 (25.3)	80.6 (24.3)	<0.001
GOLD grade			<0.001
PRISm	58 (13.0%)	648 (18.0%)
Normal	172 (38.6%)	1,510 (41.9%)
GOLD 1	56 (12.6%)	348 (9.7%)
GOLD 2	87 (19.5%)	652 (18.1%)
GOLD 3	42 (9.4%)	321 (8.9%)
GOLD 4	31 (7.0%)	127 (3.5%)
Perc15	83 (45.3)	84.7 (29.4)	< 0.001

Definition of abbreviations: BMI = body mass index; GOLD = Global Initiative for Chronic Obstructive Lung Disease; GOLD 1 = FEV₁/FVC <0.70 and post-bronchodilator FEV₁ ⩾80% predicted; GOLD 2 = FEV₁/FVC <0.70 and post-bronchodilator FEV₁ 50–79% predicted; GOLD 3 = FEV₁/FVC <0.70 and post-bronchodilator FEV₁ 30–49% predicted; GOLD 4 = FEV₁/FVC <0.70 and post-bronchodilator FEV₁ <30% predicted; LTRC = Lung Tissue Research Consortium; PRISm = preserved ratio impaired spirometry; Perc15 = Hounsfield units at the 15th percentile of computed tomographic density histogram at TLC + 1,000.

Data are reported as means with standard deviations or counts with percents. The Kruskal-Wallis test was used for the continuous variables, and the chi-square test was used for the categorical variables.

DGE in LTRC Lung Tissues

Of 17,348 genes evaluated in lung tissue samples, 1,087 were significant at FDR 10% (Table E2). Among these, 586 genes were upregulated and 501 downregulated with increasing emphysema. The top 20 significant DGEs are listed in Table 2. Among the top genes, TNFRSF6B mediates cell death pathways and impacts cellular survival and homeostasis through interactions with tumor necrosis factor ligands (56). F2RL2 is involved in tissue-specific activation of hemostasis, regulating blood clotting and vascular integrity (57). TNN is associated with integrin binding, crucial for cell adhesion, migration, and tissue organization (58). FOXL1 regulates fibroblast function and activity in various tissue contexts (59).

Table 2.

Top 20 Differentially Expressed or Used Features Associated with Emphysema in Lung Tissue Research Consortium Lung Tissue

Differentially expressed genes
Ensembl Gene ID	Gene	Log Fold Change	Mean Log Expression	FDR
ENSG00000225972	MTND1P23	−1.452	−0.416	0.014
ENSG00000243509	TNFRSF6B	−1.262	−1.402	0.021
ENSG00000164220	F2RL2	−0.809	0.202	0.005
ENSG00000120332	TNN	−0.662	0.875	0.002
ENSG00000176678	FOXL1	−0.471	1.751	0.002
ENSG00000224877	NDUFAF8	−0.297	1.963	0.014
ENSG00000226950	DANCR	−0.198	3.542	0.018
ENSG00000171421	MRPL36	−0.179	2.916	0.019
ENSG00000106628	POLD2	−0.167	5.164	0.022
ENSG00000012061	ERCC1	−0.144	5.513	0.021
ENSG00000102763	VWA8	0.145	4.865	0.018
ENSG00000155313	USP25	0.162	6.019	0.014
ENSG00000129675	ARHGEF6	0.197	6.428	0.023
ENSG00000156052	GNAQ	0.204	7.314	0.018
ENSG00000120162	MOB3B	0.223	5.013	0.018
ENSG00000173706	HEG1	0.346	8.502	0.019
ENSG00000137819	PAQR5	0.389	4.601	0.018
ENSG00000094963	FMO2	0.435	8.377	0.018
ENSG00000138669	PRKG2	0.504	2.877	0.019
ENSG00000196549	MME	0.635	6.516	0.014

Differentially used isoforms
Ensembl Transcript ID	Gene	Log Fold Change	Mean Log Expression	FDR
ENST00000537526	USP22	−2.662	−0.746	1.15 × 10−8
ENST00000589296	CYTH1	−2.378	−3.466	1.05  × 10−17
ENST00000479263	PRKCZ	−2.247	−3.303	1.10 × 10−7
ENST00000546939	CD63	−2.223	−2.768	1.33 × 10−8
ENST00000507699	PALLD	−2.145	−1.913	5.69 × 10−10
ENST00000441627	ZMIZ2	−2.037	−1.417	1.15 × 10−8
ENST00000605543	WDR11	−1.929	−3.751	3.63 × 10−7
ENST00000523282	DNPEP	−1.871	−1.899	1.68 × 10−12
ENST00000428870	KDM4C	−1.809	−4.159	5.91 × 10−8
ENST00000429192	ELN	−1.789	−0.519	7.69 × 10−7
ENST00000373266	KIAA0319L	−1.787	−3.167	1.15 × 10−8
ENST00000571368	MYBBP1A	−1.747	−1.462	5.58 × 10−8
ENST00000566130	ALDOA	−1.655	−1.261	9.71 × 10−8
ENST00000243562	LTBP4	−1.612	−0.862	3.22 × 10−7
ENST00000265637	PPP6R3	−1.380	−0.478	1.98 × 10−6
ENST00000529142	PI4KB	−1.356	−1.440	1.46 × 10−6
ENST00000479806	DYNC1I2	−1.311	0.085	6.63 × 10−8
ENST00000681161	P4HB	−1.248	−0.441	7.69 × 10−7
ENST00000513336	HEXB	−1.073	0.753	1.98 × 10−6
ENST00000528996	SERPING1	−0.957	1.874	2.49 × 10−9

Differentially used exons
Chromosome	Strand	Left	Right	Gene	Log Fold Change	Mean Log Expression	FDR
19	+	1227593	1228012	STK11	−0.393	−0.617	0.003
X	—	154348648	154348690	FLNA	−0.388	2.247	0.001
X	—	154348535	154348647	FLNA	−0.377	1.817	0.002
11	+	67286373	67286392	GRK2	−0.343	−0.035	0.001
7	—	1470277	1470477	INTS1	−0.341	−0.010	0.002
17	+	7514180	7514616	POLR2A	−0.270	2.586	0.002
6	—	31639227	31639499	BAG6	−0.252	−0.066	0.001
16	+	72112640	72112650	DHX38	−0.248	0.608	0.0003
11	—	61299470	61299555	DDB1	−0.227	1.554	0.002
19	+	46390676	46390746	PPP5C	−0.227	−0.045	0.003
3	+	184309001	184309048	PSMD2	−0.221	0.174	0.003
16	+	72112625	72112639	DHX38	−0.221	0.689	0.001
16	+	72112651	72112720	DHX38	−0.219	1.117	0.001
8	—	139730345	139730891	TRAPPC9	−0.217	0.671	0.001
9	—	111361886	111362053	ECPAS	−0.217	1.455	0.002
16	—	27460676	27460980	GTF3C1	−0.198	1.571	0.002
2	—	15166917	15167323	NBAS	−0.197	1.347	0.001
10	—	109864766	109865053	XPNPEP1	−0.186	1.024	0.002
15	+	64138098	64139875	SNX1	0.189	2.662	0.001
5	—	64718148	64724431	SREK1IP1	0.238	3.490	0.001

Definition of abbreviations: FDR = false discovery rate; Log₂ fold change = log₂ expression/usage change per −100 Hounsfield units.

Gene log fold changes are multiplied by −100 so that positive log fold changes represent upregulated genes with emphysema and negative log fold changes represent downregulated genes with emphysema.

Table E3 shows the 34 significantly dysregulated KEGG pathways identified by GSEA on the DGEs. Table 3 lists 12 noteworthy KEGG pathways selected for their biological relevance, including enhanced activity in pluripotency pathways (TGF-β and forkhead box O [FoxO] signaling) and cell barrier function pathways (adherens junction, Rap1 signaling, and gap junction). Figure E4 displays KEGG pathway maps that show gene expression changes for TGF-β, FoxO, adherens junction, and Rap1 signaling pathways, which have previously been found to be dysregulated in COPD or emphysema (60–65). Additional KEGG pathway maps for the remaining eight noteworthy pathways are provided in Figure E5.

Table 3.

Select Significant Emphysema-associated Pathways (False Discovery Rate, 10%) Identified via Gene Set Enrichment Analysis of Differentially Expressed Genes from Lung Tissue Research Consortium Lung Tissue

Term Name	Number of Genes Annotated	Average Log Fold Change	FDR
Oxidative phosphorylation	125	−0.134	1.02 × 10−5
Signaling pathways regulating pluripotency of stem cells	110	0.199	2.15 × 10−2
Adherens junction	70	0.194	4.27 × 10−2
Regulation of actin cytoskeleton	183	0.173	4.75 × 10−2
FoxO signaling pathway	120	0.149	4.87 × 10−2
JAK-STAT signaling pathway	109	0.174	5.62 × 10−2
Rap1 signaling pathway	186	0.242	6.41 × 10−2
Fc epsilon RI signaling pathway	62	0.113	6.84 × 10−2
Ras signaling pathway	195	0.198	9.22 × 10−2
TGF-β signaling pathway	76	0.197	9.22 × 10−2
Prolactin signaling pathway	60	0.109	9.22 × 10−2
Gap junction	75	0.272	9.71 × 10−2

Definition of abbreviations: Average log fold change = mean log₂ fold change of genes in the pathway (gene expression change per 100 Hounsfield units); FDR = false discovery rate; TGF-β = transforming growth factor-β.

Upregulated versus downregulated pathways are provided with respect to their relationships with emphysema. A positive log fold change value indicates a pathway upregulated with emphysema, and a negative log fold change value indicates a pathway downregulated with emphysema. Pathways were reported on the basis of relevance to lung disease.

Differential Isoform and Exon Usage in LTRC Lung Tissues

DIU and DEU analyses were performed to understand emphysema-associated alternative splicing in lung tissue. The top 20 DIUs and DEUs are summarized in Table 2.

Of the 41,853 isoforms evaluated, 802 were significantly associated with emphysema (FDR, 10%) (Table E4). Among these, 291 were upregulated and 511 were downregulated. Mapping these isoforms to their corresponding genes revealed that 5.7% (46 of 656) were significant in the DGE analysis (Figure 2). These genes included MYB binding protein 1a (MYBBP1A) and protein kinase C zeta (PRKCZ), regulators of nuclear factor (NF)-κB (66, 67), and significant isoforms mapping to tetraspanin CD63 and palladin (PALLD), involved in cell adhesion and cell–cell junctions (68, 69).

Figure 2.

Intersection of emphysema-associated gene significance. Unraveling transcriptional signals in lung tissue via differential expression, isoform use, and exon use models in the Lung Tissue Research Consortium study.

Of the 162,723 exons evaluated, 423 were significantly associated with emphysema (FDR, 10%), with 32 upregulated and 391 downregulated (Table E5). Mapping these exons to their respective 312 genes revealed that 4.8% (15 of 312) were differentially expressed (Figure 2). One top exon corresponded to TRAPPC9, a regulator of NF-κB activation (70). Exons were also identified in SNX1, NBAS, and COG2, involved in intracellular transport (71–73).

TopGO Enrichment Analyses

TopGO enrichment analysis identified 97 significant terms (P < 0.05) for DGEs, 175 for DIUs, and 100 for DEUs (Tables E6–E8). No GO pathways were significant in all three analyses. The DIU and DEU analyses shared 12 terms, DGE and DIU shared 5 terms, and DGE and DEU shared 1 term, indicating that DGE and alternative splicing analyses identified largely different processes.

Comparison of Our Lung DGE Results with Previous Literature

We compared our findings with recent COPD and emphysema transcriptomic studies by Ghosh and colleagues, Campbell and colleagues, and Rojas-Quintero and colleagues (38–40). The detailed results are presented in Tables E9–E13. There are reproduced signals, although as is common with transcriptomic studies, there also exists a considerable variability in the significant gene lists across studies, likely because of differences in study populations, disease stage, lung tissue sampling methods, and other confounding factors. Our larger study size makes our findings more reliable, although comparison with previous studies remains important. A detailed discussion of these replication results is included in the data supplement.

Identification of Cell Type–Specific Pathways

To identify lung cell types with the highest expression levels of emphysema-associated pathways, we used pathway activity scores to compute the odds ratio for each pathway in LTRC and cell type in the publicly available lung single-cell RNA-seq data. The odds ratio is calculated by dividing the fraction of cells from a given cell type with pathway activity (score, >0) by the fraction of cells with pathway activity in all the remaining cells in the dataset.

The heatmap of pathway activation and cell types revealed distinct clusters of the most strongly enriched pathways (Figure 3A). The first cluster includes pathways enriched in natural killer (NK) cells, T cells, cytotoxic T cells, and B cells. The second cluster shows gap junction, Rap1, FoxO, Fc-epsilon RI, and other signaling pathways in macrophages, as well as both classical and nonclassical monocytes.

Figure 3.

Heatmaps revealing (A) cell type–specific pathway patterns and (B) chronic obstructive pulmonary disease (COPD)-related pathway alterations in lung tissue samples in the Kaminski single-cell RNA-sequencing dataset. For each emphysema-associated pathway identified in the Lung Tissue Research Consortium, we generated pathway activity scores in each cell in the single-cell dataset. The odds ratio was computed for the cells showing the pathway activity versus not in a given cell type and for the remainder of the data. The P values were calculated using Fisher’s exact test. The two-sample t test was used to compare the pathway activity scores between control subjects and subjects with end-stage COPD for each cell type separately. A positive T statistic (red) indicates that the pathway has higher activity in COPD compared with control subjects. A negative T statistic (blue) indicates that the pathway has lower activity in COPD compared with control subjects. The inclusion of pathways named after various diseases (e.g., non–small cell lung cancer, Huntington’s disease, Alzheimer’s disease, renal cell carcinoma, and Parkinson’s disease) occurred because this is the naming convention used for those pathways from the Kyoto Encyclopedia of Genes and Genomes.

The cell type composition between COPD and control samples in the single-cell dataset is listed in Table E14. We compared pathway activity scores between control subjects and subjects with severe COPD for each lung cell type. Immune cells, including macrophages, monocytes, T cells, B cells, and NK cells, showed the highest number of significantly dysregulated pathways (Figure E6). Alveolar macrophages had the most significant pathway dysregulation.

In agreement with our bulk pathway analysis, Rap1, adherens junction, gap junction, and FoxO signaling pathways were more active in alveolar macrophages from subjects with COPD, whereas oxidative phosphorylation was less active (Figure 3B). Alveolar macrophages, alveolar type II cells, and B cells clustered together, showing similar pathway activity changes in COPD compared with control subjects. Cytotoxic T cells, T cells, and NK cells also clustered together, showing decreased activity in nearly all pathways except ribosome-associated ones. When comparing the activity of pathways in these cells, we observed that M1 and M2 macrophages show opposite activity levels of emphysema-associated pathways (Figure 3B). In addition to the previously observed reciprocal changes in oxidative phosphorylation (74), these pathways include ribosome- and spliceosome-associated pathways, JAK-STAT signaling, FoxO signaling, and regulation of the actin cytoskeleton. Abbreviations of cell types are provided in Table E15.

Given the oversimplified M1/M2 macrophage categorization (45–49), we reclustered macrophage populations in the single-cell RNA-seq data from Sauler and colleagues (42). This yielded eight clusters, with the two largest clusters expressing alveolar macrophage markers (such as FABP4, AMIGO2, and RBP4) or interstitial macrophage markers (such as MARCO, FRP3, APOC1, MRC5, and MSR1) (Figure E7, Table E16). Further subclustering identified four subclusters within each group, resulting in 14 distinct subpopulations (Table 4). Generic M1 and M2 markers indicated poor separation into distinct clusters.

Table 4.

Macrophage Subpopulations Identified by Reclustering of the Macrophages in the Single-Cell RNA-Sequencing Data from Sauler and Colleagues

Macrophage Populations and Subpopulations		Marker Genes
Interstitial macrophages	0_0 (Macrophage_1)	ZNF331, SIPA1L1, TIMP1, AREG, CCL2
	0_1 (Macrophage_2)	SPP1, CHIT1, GPNMB
	0_2 (Macrophage_3)	CD14, PLTP, LYVE1
	0_3 (Macrophage_4)	CD300E, IL1R2, SERPINB9
	2 (Macrophage_5)	CXCL8, EREG, PID1
	4 (Macrophage_6)	SFTPC, SFTPA1, SFTPA2, RALA
	5 (Macrophage_7)	FTH1P2, MYL6B
Alveolar macrophages	1_0 (AlveMacrophage_1)	FABP4, C1QA, C1QB, LSAMP, SERPING1, ALDH2, SERPINA1, CD52, SCD, IFI27
	1_1 (AlveMacrophage_2)	SVL, TANC2, FRMD4A
	1_2 (AlveMacrophage_3)	MSR1, CDC42, ARF6, LYZ
	1_3 (AlveMacrophage_4)	CCDC30, APOO
	3 (AlveMacrophage_5)	CDK1, MKI67, TK1, BIRC5
	6 (AlveMacrophage_6)	COL19A1, NRDE2
	7 (AlveMacrophage_7)	SLC27A4, HLA-DRB6

Differential abundance analysis revealed that only clusters 0_0 and 1_0 demonstrated significant differences in abundance between COPD cases and control subjects (P < 0.05). Cluster 0_0, marked by ZNF331, SIPA1L1, TIMP1, AREG, and CCL2, is associated with inflammation and immune regulation. ZNF331 is linked to transcriptional regulation and immune processes, SIPA1L1 to immune cell migration, TIMP1 to tissue remodeling, AREG to tissue repair and immune regulation, and CCL2 to immune cell recruitment. Cluster 1_0, marked by FABP4, C1QA, C1QB, LSAMP, SERPING1, ALDH2, SERPINA1, CD52, SCD, FN1, and IFI27, likely participates in lipid metabolism, immune modulation, tissue remodeling, and antiviral defense.

The heatmap in Figure E8 and Table E17 shows differential pathway enrichments between COPD cases and control subjects among macrophage subpopulations, with pronounced differences in clusters 1_0 (AlveMacrophage_1), 1_1 (AlveMacrophage_2), 0_0 (Macrophage_1), and 0_1 (Macrophage_2). Pathways at the top of the heatmap, such as T cell receptor signaling and cell–cell interaction pathways, were upregulated in COPD cases. Pathways at the bottom, mainly involving metabolic pathways and oxidative phosphorylation, were downregulated.

Comparison of Lung and Blood Transcriptomics

We performed the same analyses on LTRC and COPDGene whole-blood RNA-seq data (16). We replicated these COPDGene analyses with the same covariates that were used in LTRC, and we compared the results of the blood and lung tissue samples. Among the 1,087 DGEs from the lung and 6,526 DGEs from the blood, only 361 were shared (Figure 4B). Although this degree of overlap was statistically significant (P = 0.02), the log fold changes of the genes that were significant in either blood or lung mainly were different with an overall low correlation (r = −0.033; P = 0.52) (Figure 4A). Between lung and blood samples, 39 DIUs and 4 DEUs were shared (r = 0.28; P = 0.08 for DIUs; r = 0.60; P = 0.40 for DEUs) (Figures 4C and 4D). Complete lists of shared genes, isoforms, and exons are provided in Tables E18–E20. We also ran GSEA and topGO on blood data from COPDGene, identifying shared KEGG and GO pathways related to mitochondrial translational processes, protein localization, sexual reproduction, bone remodeling, and GTPase signal transduction (Table E21).

Figure 4.

Concordance of emphysema-associated features in COPDGene blood and Lung Tissue Research Consortium lung tissues. (A) Log fold change values of differentially expressed genes associated with emphysema in blood and lung tissue. Genes were included if they were significant in blood, lung tissue, or both (false discovery rate, 10%). Emphysema was quantified by Hounsfield units at the 15th percentile of chest computed tomographic density histogram at full inspiration + 1,000 (Perc15). The lower the Perc15 values are, the more computed tomography–quantified emphysema is present. Upregulated versus downregulated pathways are provided with respect to their relationships with Perc15, which is opposite to emphysema. Negative log fold change values represent upregulated genes, and positive log fold change values represent downregulated genes. (B) Differentially expressed genes. (C) Differentially used isoforms. (D) Differentially used exons.

Applying WGCNA to our LTRC lung gene expression dataset revealed 26 coexpression modules (Figure E9, Table E22). Eigengenes for eight modules (orange, light green, salmon, blue, magenta, cyan, gray60, and green-yellow) showed significant associations with emphysema (FDR, 10%) (Table 5). Comparing LTRC and COPDGene datasets, five of these eight emphysema-associated modules maintained their significant associations in COPDGene (FDR, 10%) (Table 5).

Table 5.

Weighted Gene Coexpression Network Analysis Module Associations with Emphysema in Lung Tissue Research Consortium Lung Tissues and Module Preservation in COPDGene Blood

	LTRC		COPDGene Projection
Module	Effect Size (SE)	FDR	Effect Size (SE)	FDR
Orange	−0.022 (5.82 × 10−3)	3.72 × 10−3	−0.002 (9.58 × 10−4)	2.30 × 10−2
Light green	0.021 (5.75 × 10−3)	3.84 × 10−3	0.002 (7.97 × 10−4)	9.90 × 10−3
Salmon	−0.017 (5.72 × 10−3)	2.25 × 10−2	−0.003 (9.92 × 10−4)	2.44 × 10−3
Blue	0.017 (5.92 × 10−3)	2.29 × 10−2	0.002 (7.26 × 10−4)	2.30 × 10−2
Magenta	−0.016 (5.77 × 10−3)	3.11 × 10−2	0.000 (1.07 × 10−3)	9.92 × 10−1
Cyan	−0.014 (5.69 × 10−3)	5.78 × 10−2	0.001 (9.41 × 10−4)	2.48 × 10−1
Grey 60	−0.015 (5.98 × 10−3)	5.78 × 10−2	−0.001 (8.28 × 10−4)	1.29 × 10−1
Green yellow	0.013 (5.89 × 10−3)	8.65 × 10−2	0.003 (7.12 × 10−4)	9.19 × 10−4
Black	−0.011 (5.95 × 10−3)	1.68 × 10−1	0.001 (9.16 × 10−4)	4.74 × 10−1
Dark grey	−0.011 (5.96 × 10−3)	1.68 × 10−1	0.003 (6.61 × 10−4)	4.27 × 10−4
Light cyan	0.011 (5.85 × 10−3)	1.68 × 10−1	−0.001 (9.37 ×  10−4)	4.14 × 10−1
Light yellow	0.011 (5.96 × 10−3)	1.68 × 10−1	0.003 (6.45 × 10−4)	1.04 × 10−5
Tan	0.010 (5.95 × 10−3)	1.88 × 10−1	−0.001 (9.54 × 10−4)	2.48 × 10−1
Purple	0.009 (5.89 × 10−3)	2.57 × 10−1	0.002 (6.05 × 10−4)	2.30 × 10−2
Brown	−0.008 (5.97 × 10−3)	3.13 × 10−1	−0.002 (6.24 × 10−4)	1.24 × 10−3
Pink	−0.007 (5.93 × 10−3)	4.41 × 10−1	−0.001 (6.41 × 10−4)	8.03 × 10−2
Midnight blue	−0.006 (5.94 × 10−3)	4.49 × 10−1	−0.005 (1.10 × 10−3)	3.68 × 10−4
Dark red	−0.005 (5.96 × 10−3)	5.61 × 10−1	0.002 (8.12 × 10−4)	3.14 × 10−2
Gray	−0.005 (5.99 × 10−3)	5.61 × 10−1	0.003 (8.91 × 10−4)	1.23 × 10−2
Yellow	−0.005 (5.87 × 10−3)	5.61 × 10−1	0.001 (7.01 × 10−4)	3.35 × 10−1
Royal blue	0.004 (5.99 × 10−3)	6.95 × 10−1	−0.003 (6.66 × 10−4)	4.27 × 10−4
Dark turquoise	−0.003 (5.82 × 10−3)	7.20 × 10−1	0.001 (7.84 × 10−4)	2.48 × 10−1
Red	−0.001 (5.97 × 10−3)	9.45 × 10−1	0.003 (6.23 × 10−4)	4.27 × 10−4
Dark green	0.0003 (5.95 × 10−3)	9.81 × 10−1	0.004 (1.08 × 10−3)	2.86 × 10−3
Green	0.0001 (5.70 × 10−3)	9.81 × 10−1	0.003 (6.36 × 10−4)	1.75 × 10−4
Turquoise	−0.001 (5.94 × 10−3)	9.81 × 10−1	−0.0004 (7.44 × 10−4)	5.88 × 10−1

Definition of abbreviations: FDR = false discovery rate; LTRC = Lung Tissue Research Consortium.

We used weighted gene coexpression network analysis to examine the gene expression patterns in LTRC lung tissues. We then ascertained the preservation of lung coexpression network associations in COPDGene blood samples. Multivariable regression was used to assess associations between module eigengenes and emphysema, adjusting for age, race, sex, FEV₁, body mass index, current smoking status, smoking pack-years, and computed tomography scanner model. Correction for multiple comparisons was conducted using FDR. Significant results with FDR <10% are bolded.

Tables E23–E30 list significantly enriched pathways (FDR, 10%) for emphysema-associated WGCNA modules identified in LTRC lung tissues and preserved in COPDGene blood. Key findings include the orange module enriched in endopeptidase inhibitor activity and magnesium ion binding; the light green module involving genes in lipid metabolism, energy regulation, nucleotide binding, cytoskeletal organization, and GTPase-mediated signaling; the salmon module enriched for protein synthesis, targeting and localization, mRNA metabolism, ribosome assembly, cytokine activity, and mitochondrial function; the blue module involved in cell signaling, adhesion, growth factor signaling, transcriptional regulation, and metabolic activities; and the green-yellow module suggesting metabolic dysregulation. These findings highlight the complex molecular pathways underlying emphysema, offering insights into potential disease mechanisms and candidates for further investigation.

Discussion

This study identified associations between emphysema and dysregulated molecular pathways in lung tissue, including oxidative phosphorylation, ribosomal function, and pluripotency pathways. Significant associations were found in various cell types with pathways involved in epithelial barrier function. We discovered opposing pathway activities in proinflammatory versus antiinflammatory macrophages. Reclustering macrophage data provided insights into differential pathway activities among macrophage subpopulations in COPD. Although some transcriptomic signals of emphysema were shared between blood and lung tissue, most expression profiles differed, offering novel insights into emphysema pathogenesis.

This project advances our understanding of emphysema-related transcriptomics by 1) conducting the largest differential expression study of emphysema to date and 2) mapping identified pathways to specific cell types. Combining bulk and single-cell RNA-seq data integrates comprehensive gene detection with detailed cellular context, enhancing insights beyond previous studies. Although pathway analyses may not always reflect functional relevance, they highlight broader biological processes and networks potentially dysregulated in emphysema. Incorporating differential splicing data reveals functional associations and molecular pathways, offering a comprehensive understanding of emphysema pathophysiology. Literature shows specific exons or isoforms involved in emphysema pathogenesis through inflammation, extracellular matrix remodeling, and cellular signaling pathways. For instance, alternative splicing of inflammatory cytokines can exacerbate lung inflammation, whereas dysregulated splicing of extracellular matrix genes disrupts lung tissue architecture (75–77). Changes in splicing variants of genes involved in signaling pathways such as mitogen-activated protein kinase and NF-κB can worsen lung injury (78–80). These findings highlight the importance of investigating splicing events to understand emphysema and identify therapeutic targets.

We identified hundreds of isoforms significantly associated with emphysema, with only 6% overlapping with significant genes from the gene-level analysis. This indicates that DIU analysis uncovered novel transcriptional events and emphysema-related pathways. Notably, only the SERPINA1 gene, which encodes the alpha-1 antitrypsin protein, and the p53/hypoxia-related genes NUMB and PGFA have previously been implicated in alternative splicing studies of emphysema (81–84).

Our study identified genes and pathways for emphysema with differential expression in both blood and lung tissues. The low correlation (r = 0.033) between β-coefficients in blood and lung tissue is due to distinct gene expression profiles and the “spillover hypothesis” (85), where only certain immune cells migrate from the lung to the blood. Lung changes are more specific to emphysema, whereas blood changes are subtler. This highlights the need for tissue-specific analysis because biomarkers effective in one tissue may not apply to another. Significant changes were seen in macrophage-associated pathways influenced by neighboring cells and inflammation. We build on the findings of Morrow and colleagues (86), who observed overlapping hemostasis and immune pathways in the large-airway epithelium, alveolar macrophages, and blood. COPD studies showed that macrophages often polarize to an antiinflammatory M2 phenotype (42, 87–90), although M1/M2 distinctions may be oversimplified. M1 macrophages are crucial for early inflammation, whereas M2 macrophages aid later tissue repair. Our finding of both phenotypes in emphysematous lungs suggests cellular microheterogeneity. Comparing our bulk lung analysis with single-cell RNA-seq data (41, 42), we observed differential dysregulation of emphysema-related pathways in macrophage subpopulations. Further targeted studies are needed to define these changes in specific macrophage populations in emphysematous lungs.

Our pathway analysis revealed enrichment for pluripotency FoxO and TGF-β signaling pathways. The FOXO family regulates genes involved in apoptosis, cell cycle control, glucose metabolism, oxidative stress resistance, and longevity (62, 91). FoxO signaling is linked to COPD, inflammation, airspace enlargement, and peripheral muscle atrophy in COPD (62, 92–95). TGF-β regulates cell proliferation, apoptosis, stem cell fates, and other processes, playing a role in respiratory illnesses, including emphysema (38, 96–98). Our analysis supports these findings, showing TGF-β signaling dysregulation in emphysema and confirming our previous studies that identified genetic variants affecting TGFB2 and ACVR1B expression in lung cells (99, 100).

Strengths and Limitations

The present study has several strengths. It represents the largest investigation to date into the connection between emphysema and gene expression/alternative splicing in lung tissue. Pathway enrichment analysis uncovered new biological processes and validated previous research findings. Furthermore, integrating the bulk RNA-seq analysis with another single-cell dataset from subjects with severe COPD allowed the examination of pathways identified in the bulk RNA-seq analysis in relation to specific cell types.

This study has limitations. First, we excluded a sizable number of LTRC participants from our analyses, primarily because of missing emphysema values. In addition, the absence of data on participants’ steroid use underscores the need for its inclusion as a covariate in future studies. The lack of available blood bulk RNA-seq and lung tissue single-cell RNA-seq data in LTRC restricted a comprehensive characterization of emphysema transcriptomics in the same individuals. Instead, existing data from COPDGene and a lung tissue single-cell RNA-seq dataset from Kaminski and colleagues were used. This single-cell dataset comprised patients with end-stage COPD rather than emphysema specifically. However, emphysema is typically present in patients with end-stage COPD (101). Concerns about using donor lungs unsuitable for transplant as control samples in single-cell analyses were acknowledged, although these lungs were not necessarily pathologically compromised. Splicing and isoform composition analysis in the single-cell data was infeasible because of sequencing depth limitations. The absence of extensive proteomic data in LTRC precluded protein-level analysis. Last, further validation in comparable cohorts is necessary to bolster confidence in the identified associations.

Conclusions

CT-quantified emphysema displays unique alternative splicing and transcriptomic patterns, alongside links to various biological processes such as oxidative phosphorylation and pluripotency pathways. Notably, macrophage subpopulations exhibit distinct pathway enrichments, whereas innate and adaptive immune cell types show enrichment for epithelial barrier function. These novel insights into emphysema at the molecular and cellular levels could inform the development of improved screening methods and therapeutic approaches.