Gene Co-expression Networks Reveal Novel Molecular Endotypes in Alpha-1 Antitrypsin Deficiency

Abstract

Background:

Alpha-1 antitrypsin deficiency (AATD) is a genetic condition that causes early onset pulmonary emphysema and airways obstruction. The complete mechanisms via which AATD causes lung disease are not fully understood. To improve our understanding of the pathogenesis of AATD we investigated gene expression profiles of bronchoalveolar (BAL) and peripheral blood mononuclear cells (PBMC) in AATD individuals.

Methods:

We performed RNA-Seq on RNA extracted from matched BAL and PBMC samples isolated from 89 subjects enrolled in the Genomic Research in Alpha-1 Antitrypsin Deficiency and Sarcoidosis (GRADS) study. Subjects were stratified by genotype and augmentation therapy. Supervised differential gene expression analysis and unsupervised, Weighted Gene Co-expression Network Analysis (WGCNA) were performed to identify gene profiles associated with subjects’ clinical variables. The genes in the most significant WGCNA module was used to cluster AATD individuals. Gene validation was performed by NanoString nCounter^® Gene Expression Assay.

Result:

We observed low effect of genotype and augmentation therapy. When WGCNA was applied to BAL transcriptome, one gene module, ME31 (2,312 genes) correlated with the highest number of clinical variables and was functionally enriched with numerous immune T-lymphocyte related pathways. This gene module identified two distinct clusters of AATD individuals with different disease severity and distinct PBMC gene expression patterns.

Conclusions:

We successfully identified novel clusters of AATD individuals where severity correlated with increased immune response independent of individuals’ genotype and augmentation therapy. These findings may suggest the presence of previously unrecognized disease endotypes in AATD that associate with T-lymphocyte immunity and disease severity.

Introduction

Alpha-1 antitrypsin deficiency (AATD) is a genetic condition that causes early onset airway obstruction, emphysema, and liver cirrhosis. It is estimated that 60,000–100,000 individuals in the United States have AATD¹. AATD is caused by a mutation in the gene that encodes the Alpha-1 Antitrypsin protein (AAT), Serpin peptidase inhibitor, clade A, member 1 (SERPINA1). Multiple AATD mutations have been identified that associate with low AAT serum levels, but the Z-allele is the most common point mutation in SERPINA1 (rs28929474) associated with COPD. This mutation results in a substitution of glutamic acid for lysine at position 342 in the AAT protein² and in the homozygous state produces an 85-90% serum and lung reduction in this antiprotease that neutralizes neutrophil elastase³. As a consequence, there is an inability of AAT to neutralize neutrophil elastase, and this contributes to tissue destruction and matrix remodeling that underlie the pathogenesis of COPD⁴.

Currently the only specific treatment for AAT-related lung diseases is augmentation therapy. The rationale for the AAT augmentation therapy is that replenishing AAT in patients deficient for this protein protects the lungs from excessive neutrophil elastase activity². While augmentation therapy has been shown to slow the decline in computed tomography (CT) lung tissue density, it is not a cure for this disease^5–9. Augmentation therapy does not fully reverse accelerated lung function decline and has no proven effect on COPD exacerbations⁵. Therefore identifying other mechanisms by which AATD causes COPD and emphysema is necessary to develop therapies to better treat this disease^10–12 while taking into consideration highly heterogenous AATD clinical presentations^{13 14}.

To determine the extent and diversity of effects that AAT genotype and augmentation therapy have on the gene expression, we analyzed gene expression profiles of paired BAL and PBMC samples based on the Genomic Research in Alpha-1 Antitrypsin Deficiency and Sarcoidosis Study (GRADS). This multicenter cohort study provided clinical data, and RNAseq profiling of bronchoalveolar lavage (BAL) cells and peripheral blood mononuclear cell (PBMC) on well characterized cohort of AATD individuals¹³.

Methods

Study Participants

The GRADS Alpha-1 Study was a prospective, multicenter study of adults older than age 35 years with PiZZ or PiMZ alpha-1 antitrypsin genotypes¹³. The study protocol, including study design, recruitment, and measurements of clinical data, has been previously described¹³. Briefly, 130 individuals with AATD were recruited through the Alpha-1 Foundation Research Registry at the Medical University of South Carolina (MUSC) and directly through physician offices. The recruitment and consenting procedures were approved by the local Institutional Review Board (IRB) and written informed consent was obtained from all participants. Figure 1 shows the overview of the study design, enrollment process and the computational analysis flow. The final population consists of 89 individuals with matched BAL and PBMC samples¹³, including individuals with PiZZ not receiving augmentation therapy (n=29); individuals with PiZZ receiving augmentation therapy (n=22) and individuals with PiMZ not receiving augmentation therapy (n=38, see Table 1).

Figure 1: Overview of the study design and computational analsyis.

(A) The CONSORT Diagram of the GRADS Alpha-1 study. (B) Diagram of an overview of the computation analysis flow.

Table 1:

GRADS A1AT cohort characteristics

	PiZZ off therapy Group 1	PiZZ on therapy Group 2	PiMZ Group 3	p-values
				ANOVA	1_vs_2	1_vs_3	2_vs_3
N	29	22	38
Age, mean ± sd	50.0 ± 9.8	61.8 ± 10.3	52.8 ± 10.0	0.0002	0.0002	0.2773	0.0019
Female (N, %)	16 (55)	16 (73)	27 (71)	0.3184	0.2498	0.2066	1.0000
White (N, %)	28 (97)	22 (100)	38 (100)	0.5730	1.0000	0.4328	1.0000
Ever smoking (N, %)	6 (21)	11 (50)	19 (50)	0.0260	0.0382	0.0213	1.0000
FVC (liters), mean ± sd	4.13 ± 1.09	2.69 ± 2.20	3.79 ± 1.00	0.0018	0.0089	0.1924	0.0370
FVC % predicted, mean ± sd	95 ± 23	91 ± 17	93 ± 14	0.8086	0.5570	0.8221	0.6136
FEV1 (liters), mean ± sd	2.93 ± 1.39	2.01 ± 0.82	2.92 ± 0.82	0.0028	0.0052	0.9744	0.0002
FEV1% predicted, mean ± sd	86 ± 26	67 ± 20	91 ± 16	0.0003	0.0056	0.4432	0.0000
DLCO, mean ± sd	24.1 ± 10.4	18.3 ± 7.4	25.5 ± 8.7	0.0121	0.0235	0.5705	0.0013
DLCO % predicted, mean ± sd	81 ± 23	67 ± 22	89 ± 19	0.0020	0.0450	0.1565	0.0007
Emphysema presence (N.%)	14 (50)	14 (67)	4 (11)	1.711e-05	0.3819	0.0007	0.0000
PD15, mean ± sd	−931 ± 17	−943 ± 19	−917 ± 20	1.139e-05	0.0347	0.0031	0.0000
Bronchiectasis (N, %)	21 (75)	15 (71)	36 (97)	0.0093	0.8409	0.0164	0.0071
Airway Internal Perimeter, mean ± sd	1.15 ± 0.24	1.09 ± 0.22	1.13 ± 0.35	0.7786	0.3789	0.7494	0.6379
Airway Wall Area, mean ± sd	40.50 ± 3.36	41.70 ± 4.09	40.76 ± 3.51	0.4976	0.2841	0.7679	0.3845
Alveolar Macrophage %, mean ± sd	73 ± 36	81 ± 22	80 ± 29	0.9695	0.2980	0.3714	0.8640
BAL Eosinophil %, mean ± sd	−0.10 ± 1.37	1.07 ± 2.92	0.42 ± 2.0	0.3834	0.0922	0.2074	0.3627
BAL Lymphocyte %, mean ± sd	6.3 ± 7.0	9.6 ± 8.4	7.2 ± 7.9	0.3318	0.1530	0.6462	0.2857
BAL Neutrophil %, mean ± sd	0.64 ± 1.48	2.75 ± 3.44	0.62 ± 1.15	0.0007	0.0119	0.9535	0.0096

RNAseq

Detailed RNA isolation, library construction, sequencing and normalization and analysis methods are described in online supplement. After sequencing reads were mapped to human genome (hg38) as described^{15 16} and normalized to address systematic variation that resulted from the sequencing process^{17 18}. Differential gene expression analyses were performed to evaluate the association of PiZ genotype and augmentation therapy status on BAL and PBMC gene expression patterns. For the test of augmentation therapy effect, due to possible confounding between treatment and disease severity that requires treatment we adjusted for age, sex and disease severity as measured by the forced expiratory volume in 1 second (FEV1) % predicted. See the online data supplements for additional details on differential expression analyses.

WGCNA and clustering analysis

We performed Weighted Gene Co-expression Network Analysis (WGCNA)¹⁹ on BAL to identify gene modules and gene networks significantly correlated with clinical phenotypes (demographics, pulmonary function tests, CT chest, augmentation therapy effect, etc). Genes expressed in over 5% of total BAL samples, and with coefficients of variation greater than 1 (n=10,718) were analyzed in WGCNA¹⁹. The default WGCNA setting was applied for data cleaning, outlier detection, network construction and module detection. Module is defined as a group of co-expressed genes across the samples²⁰. Module eigengenes (first principle component of the gene expressions within the module) were taken as the representative of the module and were correlated with AATD health data. Table 2 shows the variables used in WGCNA. Gene module network functional annotation and gene set enrichment analysis were performed using GeneMANIA²¹. GeneMANIA searches large publicly available data sets for connectivity patterns within a module based on co-expression, protein-protein interaction, pathways, and other connection that might suggest any functional relations. Gene Ontology (GO) pathway enrichment analysis was performed based on FDR-corrected hypergeometric tests. K-means clustering analysis was performed using genes from the chosen WGCNA gene module to create a heatmap and reveal visually distinct clusters. Additional details for our clustering analysis are provided in the online data supplement.

Table 2.

Clinical traits used in WGCNA

Genotype	PiMZ vs PiZZ off therapy
Augmentation Therapy	PiZZ on therapy vs PiZZ off therapy
Baseline Variable	Gender	0 female; 1 male
	Age	(Date enrolled – Birthday)/365
Pulmonary Function Test	FVC	Pre-bronchodilator Forced Vital Capacity
	FVC % PRED	Forced Vital Capacity Percent Predicted
	FEV1	Post-bronchodilator Forced Expiratory Volume in the First Second
	FEV1 % PRED	Post-bronchodilator Forced Expiratory Volume in the First Second Percent Predicted
	DLCO	Diffusing Capacity of the Lungs for Carbon Monoxide
	DLCO % PRED	Diffusing Capacity of the Lungs for Carbon Monoxide Percent Predicted
CT Variables	Emphysema presence	Fraction of lung voxels less than −950HU. Dichotomized as 0 when FRAC950 ≤ 0.05 and 1 when FRAC950 > 0.05.
	PD15	Pixel value (HU) at the 15th percentile of HU value histogram. Dichotomized based on the median value.
	Bronchiectasis	Visual scoring for presence as a categorical variable
	Airway Wall Area	Mean airway wall area (mm2) across all airways automatically detected. Dichotomized based on the median value.
	Airway Internal Perimeter	Mean airway lumen perimeter (mm) across all airways automatically detected as a continuous variable.
BAL Cell Differential	Alveolar Macrophage
	BAL Eosinophil
	BAL Lymphocyte
	BAL Neutrophil

NanoString validation

To validate the results from WGCNA analysis, we measured the gene expression of 14 genes in BAL by NanoString nCounter^® Gene Expression Assay, which is generally considered a more accurate technology, in particular for low quality RNA samples²². Among the genes selected were 14 genes with the highest expression level and fold change (FC) from WGCNA clustering analysis (CCL5, IL32, LY9, IFITM1, CD3E, PDCD1, ZAP70, LCK, TGFBR3, FOXP3, IL12RB2, IL18RAP, PRF1, GZMA) and genes related to genotype (EGR3) and therapy effect (CCDC40, MORN2 and SPA17) in BAL and PBMC samples. Analysis was performed on 77 available samples following manufacturer instructions. nSolver 3.0 digital analyzer software was used to analyze data (see online supplementary material).

Results

Supervised analysis for genotypes and augmentation therapy

The subject demographics are provided in table 1. There were no significant differences between the three groups (PiZZ on therapy, PiZZ off therapy and PiMZ) in basic demographic characteristics other than age (Table 1, also see Table 2) but individuals not on augmentation therapy (group 1, PiZZ off) were younger, healthier and had better lung functions than individuals on AAT augmentation therapy (group 2, PiZZ on). BAL cell counts were similar amongst three groups, with exception of increase in neutrophils in PiZZ individuals on augmentation therapy.

We examined the changes in gene expression profiles in BAL and PBMC isolated from individuals with PiMZ and PiZZ genotypes (group 3 vs 1, Table 1). Overall we did not observe significant changes in gene expression profiles. A total of 113 genes in BAL and 181 in PBMC were differentially expressed between PiZZ and PiMZ genotypes (FDR<0.05, see Supplementary Table 1 and 2). The 113 BAL genes were enriched for inflammatory, cellular chemotaxis and antiviral responses (Supplementary Table 3).

We also compared gene profiles of individuals with PiZZ not receiving augmentation therapy (group 1) with PiZZ receiving augmentation therapy (group 2, Table 1) the majority of differentially expressed genes were low expressed genes and with limited statistical significance suggesting a relatively low effect of the therapy on gene expression. They were enriched for cilium morphogenesis, chemotaxis and inflammatory response (Supplementary Table 4 and 5).

To investigate commonality between gene expression profiles in lung and blood within genotypes and therapy effect, we performed comparisons of all expressed genes in BAL and PBMC samples (Supplementary Figure 1). While we observed high overall correlation in gene expression between BAL and PBMC (r=0.91, see Supplementary Figure 2), the overlap of differentially expressed genes across the two tissues was minimal, with only 5 and 6 genes (FPKM>1) significantly differentially expressed in both BAL and PBMC for genotype effect and therapy effect, respectively (Supplementary Figure 1 and Supplementary Table 6 and 7).

Weighted Gene Co-Expression Network Analysis

Since we observed rather low effect of genotype and therapy on BAL and PBMC gene expression in clinically defined groups, we applied an unsupervised approach to analyze gene expression patterns from BAL, the more disease-relevant tissue compartment in the study of AATD lung diseases. We performed WGCNA analysis on 89 BAL samples to identify gene networks and modules associated with AATD genotypes and augmentation therapy samples. 10,718 genes (CV>1%) were included WGCNA. This analysis aimed at identifying modules that are significantly correlated with the measured clinical traits for AATD individuals (Figure 2 and Supplementary Figure 4). 31 modules were identified, of which 10 were significantly correlated with several clinical variables (Figure 2). The strongest correlations were identified for the module ME 31(2,312 genes) that correlated with emphysema presence (dichotomized variable; emphysema present defined as greater than 5% of lung voxels less than a −950HU threshold, r=0.24, p=0.03), bronchiectasis presence as measured by a visual scoring system, r=0.2, p=0.07), alveolar macrophage %, r=−0.63, p=7E-11) and alveolar lymphocyte % (r=0.51, p=6E-07).

Figure 2: Module-trait relationship from WGCNA for 10 selected modules with the strongest correlations.

The numbers in each cell represent the correlation coefficients and p-values between each clinical trait and module eigengenes. The module-trait relationship for all 31 modules are presented in Supplementary Figure 4.

Clustering analysis

To examine if the genes in gene module ME31 (2,312 genes) could cluster AATD individuals into clinically meaning groups, we performed K-means clustering (Figure 3) that distinguished three clusters of AATD individuals. Analysis of individuals’ demographics, clinical and CT chest data (Table 3) among the 3 clusters revealed a small group of 7 individuals in Cluster 1 with the slightly increased emphysema. Cluster 2 had younger predominantly PiMZ individuals (p= 0.059) compared to cluster 3, as well as slightly higher FEV1 and DLCO. Cluster 3 had more subjects with a PD15 value above the median and emphysema presence than observed in cluster 2. The BAL lymphocyte % was higher in cluster 3 than in cluster 2 (9.7± 8.1 vs 2.4± 2.4) as well as neutrophil % (1.4± 2.7 vs 0.6 ± 0.9), while macrophage % were not different. There was no significant difference in therapy, gender or smoking status between clusters 2 and 3 (Table 3). These findings all together suggest the gene expression signature from module ME31 identified 2 well separated groups of individuals in Clusters 2 and 3 with distinct clinical emphysema characteristics that are not primarily driven by either AATD genotype or therapy.

Figure 3: Heatmap for Module 31 expression based on k-mean clustering of samples.

The color bar (k-means) indicates the three clusters (1,2,3 from left to right). The color bar (phenotype) indicates the three groups from the original design (PiZZ on therapy, PiZZ off therapy, PiMZ off therapy). The two subgroups of genes (red and blue bars from the left) are from hierarchical clustering, with the red group visually differentiating clusters 2 and 3.

Table 3:

Clinical Phenotypes and Module 31 Sample Clustering

		Cluster 1	Cluster 2	Cluster 3	p-value
					ANOVA	1_vs_2	1_vs_3	2_vs_3
Prevalence, n		7	27	48
Age, yr, mean ± sd		51.89 ± 7.50	48.43 ± 9.97	57.03 ± 10.40	0.0026	0.3316	0.1408	0.0008
Female sex, n (%)		5 (71)	16 (59)	32 (67)	0.7747	0.6818	1.0000	0.6182
Ever Smoker, n (%)		2 (29)	13 (48)	18 (38)	0.5799	0.4263	1	0.4651
Genotype	PiZZ off, n (%)	3 (60)	6 (25)	18 (51)	0.0851	0.2872	1.0000	0.0599
	PiMZ, n (%)	2 (40)	18 (75)	17 (49)
Therapy	PiZZ on, n (%)	2 (40)	3 (33)	13 (42)	0.8964	1.0000	1.0000	0.7171
	PiZZ off, n (%)	3 (60)	6 (67)	18 (58)
FVC (liters), mean ± sd		3.62 ± 1.23	4.06 ± 0.98	3.48 ± 1.79	0.2920	0.4052	0.7969	0.0742
FVC % predicted, mean ± sd		89.71 ± 17.49	95.15 ± 13.29	94.50 ± 21.53	0.7903	0.4656	0.5289	0.8725
FEV1 (liters) mean ± sd		2.37 ± 1.10	3.21 ± 0.89	2.55 ± 1.11	0.0236	0.0997	0.6982	0.0068
FEV1% predicted , mean ± sd		71.86 ± 28.70	91.30 ± 16.16	83.69 ± 24.59	0.1048	0.1285	0.3328	0.1113
DLCO, mean ± sd		19.60 ± 4.23	27.06 ± 7.41	22.11 ± 9.62	0.0308	0.0029	0.2520	0.0154
DLCO % predicted, mean ± sd		74.71 ± 11.57	87.81 ± 17.41	79.71 ± 25.65	0.2223	0.0322	0.3959	0.1089
Emphysema presence, n (%)		4 (57)	2 (7)	22 (48)	0.0003	0.0096	0.7040	0.0003
PD15, mean ± sd		−939.00 ± 24.39	−915.89 ± 20.21	−932.11 ± 19.69	0.0020	0.0487	0.4982	0.0015
Bronchiectasis, n (%)		7 (100)	27 (100)	44 (96)	0.4896	0.1801	0.7083	0.4786
Airway Internal Perimeter, mean ± sd		1.07 ± 0.20	1.16 ± 0.36	1.14 ± 0.26	0.7423	0.3639	0.4230	0.7589
Airway Wall Area, mean ± sd		41.40 ± 1.27	41.99 ± 3.99	40.35 ± 3.58	0.1723	0.5201	0.1551	0.0854
Alveolar Macrophage %, mean ± sd		90.64 ± 5.10	85.20 ± 31.02	75.61 ± 28.33	0.2214	0.3927	0.0017	0.1911
BAL Eosinophil %, mean ± sd		0.21 ± 0.57	0.22 ± 1.22	0.51 ± 2.61	0.8335	0.9803	0.4985	0.5187
BAL Lymphocyte %, mean ± sd		7.79 ± 5.60	2.41 ± 2.43	9.72 ± 8.13	8.20E-05	0.0441	0.4425	2.70E-07
BAL Neutrophil %, mean ± sd		1.07 ± 1.17	0.59 ± 0.86	1.44 ± 2.65	0.2607	0.3409	0.5395	0.0467

Network Analysis and Functional Enrichment Analysis

As the WGCNA module ME 31 (2,312 genes) showed significant correlation with key AATD clinical variables and identified 2 well-defined clusters of AATD individuals, we aimed to further investigate the biological function of these genes. We focused on a subset of 666 genes that were downregulated in cluster 2 and visually differentiated clusters 2 and 3 (marked in red in the lower half of Figure 3, Supplementary Tables 8 and 9). Figure 4 shows 135 genes in the module that belong to the multiple T cell and immune response enriched pathways (exp: PDCD1,CD3, LCK, ZAP70, CCL5, BTLA, GRAP2, CD247). Functional enrichment analysis revealed that several immune system pathways were overrepresented in this module, including T-cell activation, regulation of immune response, and antigen receptor-mediated and plasma membrane signaling pathways (See Figure 4 and Supplementary Table 10). Together these findings suggest an active immune T cell response related to presence of emphysema in AATD individuals.

Figure 4: The gene network and functional enrichment analysis of the module 31.

Heatmap for the most enriched pathways (FDR<0.05) and their corresponding 135 genes in module 31, and the GO Term significance score (−log₁₀(FDR)) for each pathway.

Differential gene expression for PBMC between newly identified AATD clusters

To better understand biological differences between the individuals in clusters 2 and 3 we performed differential gene expression analysis on PBMC expression profiles using edgeR. We found 125 differentially expressed genes between cluster 2 and 3 (see Figure 5, Supplementary Table 11 and 12). These genes were enriched for plasma lipoproteins, lipid transport, and inflammatory responses (Supplementary Table 13). Most notably, some of these genes, including CCL18, FABP4, FN1 and miR-9, have been implicated in COPD and were indeed highest in cluster 3 characterized by more severe disease.

Figure 5: Box plots for selected differentially expressed genes in PBMC samples across the clusters characterized by module ME31.

(A) MIR9-1 (B) CCL18 (C) FN1 (D) FABP4

NanoString Validation

Validation by Nanostring confirmed that EGR3, a key negative regulator of T cell activation²³ was upregulated in PiZZ vs PiMZ in both BAL and PBMC (Figure 6) and CCDC40, MORN2 and SPA17 genes related to cilia dysfunction in lung disease²⁴ were upregulated in PiZZ on therapy in BAL (Figure 6). In addition, validation of genes associated with emphysema (CCL5²⁵, IL32²⁶), IFN pathway (IFITM1²⁷ and T cell activation and PD1- immune signaling (LY9²⁸, CD3E, PDCD1, ZAP70, LCK, TGFBR3, FOXP3, IL12RB2, IL18RAP, PRF1, GZMA²⁹) revealed complete concordance with RNA-Seq data (Figure 6).

Figure 6: Validation of selected differentially expressed genes using NanoString nCounter^®.

NanoString Log₂(FC) between cluster 2 and cluster 3 on x axis and RNA-seq Log₂(FC) between cluster 2 and cluster 3 on y axis. Gene names are labeled.

Discussion

In this study, we applied supervised and unsupervised methods on gene expression profiling of BAL isolated from AATD individuals, and we identified genes associated with genotype and augmentation therapy as well as a gene module enriched for T cell pathways and immune response that correlated with severity of disease independent of genotype and therapy in AATD affected individuals.

Our most impressive finding was a previously unrecognized endotype of individuals with AATD that regardless of genotype or augmentation therapy had more severe respiratory disease and an altered pulmonary inflammatory and immune response driven by lymphocytic influx (cluster 3, Table 3). Although this AATD cluster had individuals with increased alveolar lymphocyte %, the cluster was not significantly correlated with AATD therapy, or changes in lungs such as: FEV1% Predicted, DLCO, bronchiectasis, or airway wall thickness. Instead, the chest CT analysis using PD15 and emphysema presence were the most informative^30–32 of disease severity. We show that lymphocyte % was elevated in individuals with PD15 lower than the medium (p=0.05) and in individuals with emphysema presence (p=0.009, see Supplementary Table 8) suggesting that increase in lymphocyte % is a biological signal of AATD disease severity. We found slightly increased number of neutrophils in cluster 3 of individuals with emphysema predominant disease. Neutrophils have traditionally been found in AATD in bronchitis³³, exacerbation prone^{34 35}, and emphysema predominant disease^{36 37}.

A potential mechanistic clue of AATD lies in the genes that characterized this cluster. Amongst the most expressed and upregulated BAL genes from cluster 3 were genes implicated in immune responses via T cell activation and PD-1 signaling suggesting the presence of a PD1 immunosuppressive and proinflammatory endotype²⁹. Recently it has been suggested that A1AT has roles that extend beyond its antiproteolytic effects. For instance it can regulate inflammatory milieu by inhibiting proliferation of T helper cells and by controlling antigen presentation^38–40. While our study did not study T-cell functions, the identification of an AATD individual cluster characterized by emphysema and T-cells suggest the need for detailed immunophenotyping of AATD patients and exploration of specific therapies. This is also supported by our PBMC gene expression analysis. The PBMC of the individuals in cluster 3 exhibited distinct gene expression patterns with increased expression of genes such as CCL18, FABP4, FN1 and miR-9 which were previously associated with COPD. CCL18 was shown to be up regulated in COPD and predict risk for exacerbations^{41 42}. FABP4 was shown increased in COPD and was negatively correlated with lung function⁴³. FN1 was also show to be a gene correlated with progression of COPD⁴⁴. Mir-9 was shown to be associated with the loss of muscle force in patients during an acute exacerbation of COPD⁴⁵. Our unsupervised analysis may have revealed a subgroup of AATD individuals that cluster together based on gene expression and segregates a group of individuals with biological changes in lungs and lung function during development of emphysema that goes beyond clinical phenotyping described previously⁴⁶. The challenge will be to develop and validate diagnostic parameters that characterize these individuals that could be used in the clinic. There also will need to be much work done to exclude environmental stimuli that correlate with these genomic signatures of AATD.

Our study has several limitations that should be considered when interpreting the results. The first is that it contains a relatively small cohort of individuals with AATD, with no independent replication. While this is true, our study in fact represents the first and largest study that describes the BAL and PBMC transcriptome in a carefully phenotyped cohort of AATD individuals. We believe that our findings will encourage others to follow up on our results and potentially focus on the subgroup we identified to get detailed mechanistic understanding of the mechanisms regulating emphysema in A1AT. Another limitation is the lack of PiMM individuals in the cohort. While we did not find significant difference between gene expression profiles of PiMZ and PiZZ individuals, more difference might have been detected had PiMM individuals been included. Finally, the obvious limitation of bulk RNAseq of BAL cells and PBMC is that we do not necessarily know whether the changes we observed are derived from changes in cell content, transcriptional regulation or both. While gene expression values could have been normalized for cell proportions, doing so would have hidden potentially relevant signals. Therefore, we decided to consider the cell counts as clinical phenotypic attributes. The fact that we found very few differential expression genes for genotype and therapy effects suggests that the changes in gene expression here were not primary driven by differences in cell counts. As detailed immunophenotyping was beyond the scope of our study, we believe that we have shown convincingly that gene expression associated with T-cell inflammation is common in patients with emphysema. We hope that this work will motivate other more detailed studies as well as the development of interventions that target the immune aberrations that we observed.

In conclusion, using unsupervised data analysis methods we identified a subgroup of AATD individuals characterized by more severe disease, and increased T cell inflammation, that is independent of genotype or augmentation therapy. This finding may represent a novel endotype in AATD. Further studies that apply advanced immunophenotyping approaches, include longer follow up, and focus on emphysema may help to validate this endotype and potentially identify specific interventions.

Key Messages.

What is the key question?

What are the effects of AAT genotype and augmentation therapy on bronchoalveolar lavage (BAL) and peripheral blood mononuclear cells (PBMC) gene expression profiles?

What is the bottom line?

While this effects of genotype and augmentation therapy are not strong, we identified a signature of genes that distinguish two clusters of patients that differ in extent of emphysema, and lymphocytic infiltration, but not in therapy or genotype, potentially reflecting novel endotypes of disease driven by inflammation.

Why read on?

This is the largest study that describes the transcriptome of BAL or PBMC samples from well characterized individuals with AATD, and the first study that aims to use PBMC and BAL gene expression to understand the effects of the causal genetic variant and augmentation therapy and their connection to patient clinical characteristics.

Abbreviation List

AAT

Alpha-1 antitrypsin protein

AATD

Alpha-1 antitrypsin deficiency

BAL

bronchoalveolar lavage

COPD

Chronic Obstructive Pulmonary Disease

CT

computed tomography

DLCO

Diffusing capacity of the lungs for carbon monoxide

FC

fold change

FDR

false discovery rate

FEV1

forced expiratory volume in 1 second

FPKM

Fragments Per Kilobase of transcript per Million mapped reads

GO

Gene Ontology

GRADS

Genomic Research in Alpha-1 Antitrypsin Deficiency and Sarcoidosis

PBMC

peripheral blood mononuclear cells

PD15

Pixel value (HU) at the 15th percentile of HU value histogram

RNAseq

RNA sequencing

TMM

trimmed mean of M-values

WGCNA

Weighted Gene Co-expression Network Analysis

Availability of data and materials

RNAseq data described in this paper has been deposited in the NCBI Gene Expression Omnibus (GEO) under accession code GSE109515.