Characterization of a COPD-Associated Nephronectin Functional Splicing Genetic Variant in Human Lung Tissue via Long-Read Sequencing

Abstract

Background:

Identification of COPD disease-causing genes is an important tool for understanding why COPD develops, who is at highest COPD risk, and how new COPD treatments can be developed. Previous COPD genetic studies have identified a highly significant genetic association near nephronectin (NPNT), a gene involved in tissue repair, but the biological mechanisms underlying this association are unknown.

Methods:

Splicing quantitative trait locus analysis (sQTL) was performed to identify common genetic variants that alter RNA splicing in lung tissues. These lung sQTL signals were compared to COPD genetic association results near the NPNT gene using colocalization analysis to determine whether genetic risk for COPD in this region may act through altered splicing. Long read sequencing characterized COPD-associated splicing events at isoform-level resolution, and in silico protein structural analysis identified likely functional effects of this alternative splicing.

Results:

An established COPD genetic risk variant, rs34712979_A, creates a cryptic splice acceptor site that causes four separate splicing changes im NPNT. The only of these splicing changes that was associated with COPD phenotypes involved a cassette exon (exon 3). Long read RNA sequencing demonstrated that the COPD risk allele causes a shift in isoform usage away from the dominant NPNT Isoform B precursor, which excludes exon 3, to the Isoform A precursor which splices-in exon 3. Alpha-fold protein structural analysis reveals that inclusion of this exon disrupts an EGF-like functional domain in NPNT.

Conclusion:

Genetic variants in the nephronectin (NPNT) gene increase COPD risk by changing RNA splicing of NPNT in the lung.

Shareable abstract:

Common genetic differences play an important role in the development of COPD, and identification of COPD disease-causing genes is an important tool for understanding COPD pathogenesis, identifying individuals at highest risk for COPD, and the development of more effective treatments. This paper demonstrates that genetic variants in the nephronectin (NPNT) gene increase COPD risk by changing RNA splicing of NPNT in the lung.

Graphical Abstract

Introduction

The development of chronic obstructive pulmonary disease (COPD) is influenced by genetic susceptibility factors, and the identification of COPD disease-causing genes is an important tool for understanding why COPD develops, who is at highest risk for COPD, and how new COPD treatments can be developed. Recent genome-wide association studies (GWAS) have identified over 80 distinct genetic loci that influence susceptibility to COPD[1]; however, the responsible genetic variants and effector genes are unknown for most of these associations.

These genetic effects can be discovered by combining GWAS with functional characterization of genetic effects on gene expression (eQTL analysis) or RNA splicing (sQTL analysis), which can identify the specific causal genetic variant and the mechanism by which it alters gene function. In COPD, this approach has led to the identification of HHIP[2], FAM13A[3], TGFB2[4], and ACVR1B[5] as COPD-causing genes. However, eQTL studies do not capture all potentially relevant functional mechanisms, and it has been shown that the use of sQTLs to capture the effects of genetic variants on splicing identified additional disease-causing mechanisms[6, 7], and recent studies have shown that genetic effects on splicing are an important cause of COPD[8, 9].

In this study, we hypothesized that a COPD GWAS signal near nephronectin (NPNT) alters COPD risk through effects on splicing of NPNT. NPNT participates in tissue injury, inflammation and repair[10–13], and an Npnt knockout mouse model shows impaired resolution of inflammation and injury[14]. NPNT splicing has also been associated with COVID-19 disease severity[15, 16], and a specific genetic variant in NPNT (rs34712979) has been identified as the likely COPD causal variant in this genomic region[1, 17–19]. Using colocalization analysis, we confirmed that the COPD GWAS and sQTL signals near NPNT show almost perfect overlap. Analysis of short and long-read RNA sequencing (RNA-seq) from human lung tissues confirmed the effect of rs34712979 on splicing, indicating that the disease-related event is likely related to the inclusion of an exon that disrupts an EGF-like functional domain in NPNT. This study demonstrates the functional splicing mechanism underlying a genetic risk factor for COPD, providing further evidence of a link between individual differences in the wound healing response and risk for COPD.

Methods

Genetic association analysis, splicing QTL, and colocalization results

Summary GWAS statistics were used from our previous study of COPD and two lung function phenotypes, FEV₁ and FEV₁/FVC (http://ldsc.broadinstitute.org/ldhub/)[1, 18]. Generation of splicing QTLs (sQTLs) in LTRC is previously described[9]. GTEx (sQTL) significant results[20] were obtained for all tissues from the GTEx Portal (https://www.gtexportal.org/home/datasets), and complete lung sQTL results were obtained from the Anvil GTEx Terra workspace. Multiple colocalization for GWAS and sQTL results was performed using the moloc R package (https://github.com/clagiamba/moloc), details are available in the supplemental methods. See Supplementary Figure 1 for details of subjects selection.

Human tracheobronchial epithelial cell (HTBE) air:liquid interface (ALI) samples and RNA-seq.

Human lung specimens were obtained under the auspices of the University of North Carolina (UNC) Institutional Review Board protocol # 03–1396. All donors or their authorized representatives provided informed consent for research use of organs. These specimens were from previously normal (no history of prior chronic lung disease) organ donors whose lungs were not used for transplantation. Tracheobronchial epithelial cells were obtained and cultured as described in detail previously[21]. Details of genotyping and RNA-seq are provided in the supplemental methods.

Long read RNA-seq analysis in human lung samples from the LTRC

We conducted targeted Oxford Nanopore Technologies (ONT) long read sequencing on RNA from 10 human lung samples from the LTRC which were selected to include five samples from each homozygous class (i.e., GG genotype, AA genotype) of rs34712979. For each of these ten samples, 100–200ng of total RNA was used to generate full-length cDNA using a modified Smart-seq2 protocol[22]. The enrichment and library generation procedures are described in detail in the Supplemental Methods.

For NPNT isoform quantification (i.e., usage analysis), the proportion of isoform usage for each sample was calculated by dividing the number of reads for each isoform by the total number of isoform reads aligning to the NPNT locus. Differences in isoform usage between genotype classes were identified using the Mann-Whitney test.

NPNT protein sequence analysis

Protein sequence analysis was performed using Uniprot [23] to identify NPNT protein domains. The Chou and Fasman Secondary Structure Prediction server [24] was used to characterize the impact of sequence changes to NPNT structure. Alpha-fold protein structures with exon 3 (UniProt ID: D6RH31) and without exon 3 (UniProt ID: Q6UX19) were downloaded from UniProt and visualized with PyMOL. The structure of the first EGF-like domain in Q6UX19 has a high per-residue model confidence score, as deemed by Alpha-fold. The structure of the disrupted EGF-like domain in D6RH31 has a medium per-residue model confidence score, indicating moderately accurate structure prediction.

Plasmid generation, transfection and NPNT protein detection

Plasmids containing encoding the NPNT sequence with and without the serine insertion were transfected into 16HBE14o- cells (Sigma). NPNT levels were measured in whole cell lysates and supernatants by Western blotting with anti-flag antibodies. Full details are available in the Supplemental Methods.

Results

rs34712979 is associated with respiratory phenotypes and NPNT splicing

Genome-wide significant association signals near NPNT have been identified for COPD and various measures of lung function, and rs34712979 is the most likely causal variant[1, 17–19]. The minor A allele of rs34712979 is associated with lower measures of lung function and increased risk of COPD (odds ratio=1.18). Previous colocalization studies showed strong overlap between COPD GWAS and sQTL signals for NPNT, while there is no colocalization with NPNT eQTL results[9]. To provide further evidence of the link between rs34712979, altered splicing of NPNT, and lung disease, we performed colocalization analysis with the moloc method [25] to compare the genetic association results in the region near NPNT from GWAS studies of COPD, FEV1, and FEV1/FVC as well as lung sQTL results from two large studies (Figure 1, Supplementary Figure 1). This analysis confirmed that the genetic signals for COPD, lung function, and NPNT sQTLs are nearly identical (Figure 2), that rs34712979 is very likely (94% estimated probability) to be the causal genetic variant responsible for these associations to splicing and lung disease. Notably, the eQTL signal for NPNT is not very similar to the GWAS signal, suggesting that the COPD-related genetic effect is likely due to altered splicing rather than to effects on the overall NPNT expression level.

Figure 1: Subjects and datasets included in the study.

GTEx = The Genotype-Tissue Expression (GTEx) Project. LTRC = the Lung Tissue Research Consortium.

Figure 2: rs34712979 is associated with COPD, lung function and NPNT splicing.

Local association plots for genetic associations near NPNT. The lead variant, rs34712979, is highlighted in purple, and is used as the reference SNP for calculating linkage disequilibrium. The color of the other points represents the correlation (r²) between the variant shown and the lead variant. The X axis shows the position of the variant in relation to the NPNT gene, and Y axis shows the -log₁₀ p-value. Plots are shown for a) COPD (Sakornsakolpat 2019), b) FEV₁/FVC, c) FEV₁ (Shrine 2019), d) LTRC sQTL, e) GTEX sQTL, and f) LTRC eQTL.

rs34712979 is associated with a new splice acceptor and alternative inclusion of multiple exons in NPNT

To better characterize the link between rs34712979 and RNA splicing, we examined sQTL results from the Lung Tissue Research Consortium (LTRC). We observed that rs34712979 was associated with multiple splice sites in NPNT (Table 1) that would result in inclusion or exclusion of exons 3, 5, and 11. This observation was confirmed in a separate analysis of sQTL data from GTEx lung tissue (Supplementary Table 2). To better visualize the splicing of exons 3, 5, and 11, we identified all subjects with the less common rs34712979 AA genotype in LTRC and GTEx (n=86 and 33, respectively), and we compared the splicing of these exons to the same number of randomly selected subjects with the more common AG and GG genotype. This confirmed that subjects with the AA genotype have higher rates of inclusion of the 3^rd, 5^th and 11^th exons (Figure 3). We also observed that the A allele creates a novel splice acceptor for exon 2, which results in a 3-bp extension of that exon that results in the insertion of a TAG sequence that codes for a novel serine residue in the NPNT protein (see Supplementary Results).

Table 1:

Association between NPNT splicing in LTRC lung tissue and clinical variables

	Cluster	Position	Event	1Clinical Variables
				COPD	FEV1pp	FEV1/FVC	Perc15
4:105898001:105912188	1	Exon 2–3	+ exon 3	5.33E-03	0.023	1.39E-04 (β= −0.026)	2.75E-04 (β= −6.9)
4:105898001:105927336	1	Exon 2–4	− exon 3	8.66E-03	0.127	1.69E-03 (β= 0.020)	0.125
4:105912238:105927336	1	Exon 3–4	+ exon 3	0.065	0.371	0.089	0.378
4:105927428:105931515	2	Unannotated		0.330	0.272	0.822	8.40E-03
4:105927428:105932614	2	Exon 4–5	+ exon 5	0.049	0.021	0.046	0.897
4:105927428:105937009	2	Exon 4–6	− exon 5	0.013	0.013	0.058	0.758
4:105932703:105937009	2	Exon 5–6	+ exon 5	3.33E-03	0.029	0.080	0.514
4:105942702:105959028	3	Exon 10–12	− exon 11	0.0149	0.132	0.183	0.983
4:105895723:105897898	4*	Exon 1–2 with serine	+ serine	0.176	0.016	0.032	0.933
4:105895723:105897901	4*	Exon 1–2 Without serine	− serine	0.183	0.018	0.027	0.825

Eight exon-exon junctions were found to be associated with the COPD risk variant rs34712979. P-values are shown for association testing between splice ratios for the NPNT junctions of interest and COPD related phenotypes. The threshold for significance is based on correction for multiple comparisons (4 variables tested x 4 clusters for significant threshold p<0.004). Effect sizes (beta coefficients from linear regression) are shown for significant associations.

¹Clinical variables: COPD = GOLD 2–4, FEV₁pp = forced expiratory volume in 1 minute percent of predicted, FEV₁/FVC = ratio of FEV₁ over forced vital capacity, Perc15 = 15th percentile point of Hounsfield Units, i.e., severity of CT-quantified emphysema calculated using the 15th percentile of the lung density histogram technique (lower Perc15 values indicated more CT-quantified emphysema).

^*Cluster 4 was not tested in the sQTL analysis, this splice site was only discovered after detailed investigation into NPNT splicing.

Figure 3. rs34712979 is associated with increased inclusion of NPNT exon 3.

a) Gene model showing all annotated exons of NPNT. Exons 3, 5, and 11 are known to be alternatively spliced. b) Violin plots showing exon inclusion by rs34712979 genotype in LTRC and GTEx. Subjects were selected such that all available subjects with the rs34712979 AA genotype were included (n=86 and 33 for LTRC and GTEx, respectively) and compared to an equal number of randomly selected subjects with the AG and GG genotype. The X axis shows the genotype, and the Y axis shows normalized exon inclusion proportions. c) Sashimi plot showing the mean number of reads spanning each exon-exon junction in NPNT. Each arched line indicates a junction, and the number indicates the number of spliced reads. Colored peaks indicate RNA-seq read coverage for each exon.

NPNT splicing ratios are associated with pulmonary phenotypes in LTRC

To better understand which alternative splicing events in NPNT contribute to pulmonary phenotypes, we tested for a relationship between alternative splicing and clinical variables in LTRC. We found that the inclusion of exon 3 was significantly associated with reduced FEV₁/FVC and reduced perc15 (15th percentile point of Hounsfield Units, indicating increased emphysema). There was no significant association to pulmonary phenotypes observed for exon 5, exon 11, or the novel TAG/serine insertion (Table 2). We also tested for association between NPNT splicing and demographic variables and found that exons 3 and 11 were not associated with any demographic variable, while exon 5 inclusion was associated with sex and current smoking (Supplementary Table 3).

Table 2:

NPNT isoforms identified through long read sequencing analysis

RNA	Exons*	Associated Transcript	Protein	Amino acids**	Isoform Proportion (%)		p-value
					AA	GG
Isoform 1	11,13	NM_001033047 ENST00000379987.7	Isoform B precursor	565 aa	58.6	66.8	0.06
Isoform 2	11	Unannotated		536 aa	0.5	1.1	0.11
Isoform 3	13	NM_001184692.2 ENST00000514622.5	Isoform D precursor		3.4	5.7	0.30
Isoform 4	3,11,13	NM_001184690.2 ENST00000453617.6	Isoform A precursor	582 aa	15.0	8.1	0.037
Isoform 5	5,11,13	NM_001184691.2 ENST00000427316.6	Isoform C precursor	596 aa	16.8	14.9	0.037
Isoform 6	3,5,11,13	XM_047449987.1 ENST00000514837.1		453 aa	5.7	3.5	0.25

^*All observed isoforms include exons 1,2,4,6–10,12 and 14.

^**The number of amino acids excludes the Serine residue which is only present with the A allele. We collapsed isoforms regardless of 5’UTR length, transcriptional start site, and novel serine inclusion.

Long read sequencing demonstrates that rs34712979 is associated with increased usage of isoforms containing exons 3, 5 and 11

To determine the effect of rs34712979 on full-length NPNT isoforms, we selected 10 LTRC lung RNA samples with RIN > 7 and the rs34712979 AA or GG genotype (5 each), and we enriched the RNA for NPNT transcripts and performed ONT long-read sequencing. The experiment yielded 24,747 reads mapping to NPNT. The most highly expressed isoform in both genotype classes was the full-length isoform (Refseq NM_001033047.3, Isoform B precursor) that does not include exons 3 and 5 but does include exon 11. Focusing on exons 3, 5, and 11, we found that each exon’s usage (i.e. inclusion in full length isoforms) was higher in AA vs GG subjects (p=5.68 × 10-¹¹, 0.0058 and 0.007, respectively - Figure 4). To analyze these data at the isoform level, we collapsed the isoforms according to their unique exon inclusion pattern, which considered only the four exons that varied across NPNT isoforms (exons 3, 5, 11, and 13). Since the serine insertion was not associated to any COPD-related phenotypes in LTRC, we also collapsed isoforms without regard to this variation. We found that the AA subjects had a higher proportion of the NPNT Isoform A precursor containing exons 3, 11, and 13 (NM_001184690.2/transcript isoform 4, p=0.037; shown in Table 2) and the Isoform C Precursor containing exons 5, 11, and 13 (NM_001184691.2/transcript isoforms 5, p=0.037).

Figure 4. Long read RNA-seq in 10 LTRC lung tissues confirms increased usage of exon 3-containing isoforms in rs34712979-A.

a) Isoforms detected by long read sequencing in 10 human lung tissues from LTRC (5 GG and 5 AA). b) Isoforms containing exons 3, 5 and 11 are more highly represented in the AA vs GG genotype (using Chi squared test). c) The isoform usage profile differs between rs34712979 GG and AA homozygotes. Of the 6 identified isoforms, 4 have corresponding curated Refseq IDs which are as follows – Isoform 1 = NM_001033047/Isoform B precursor, Isoform 3 = NM_001184692.2/Isoform D precursor, Isoform 4 = NM_001184690.2/Isoform A precursor, Isoform 5 = NM_001184691.2 /Isoform C precursor).

Alternative isoforms of NPNT differ in functional domains

We then performed in silico modeling of the NPNT transcript isoforms associated with the rs34712979 COPD risk allele to identify any predicted differences between these isoforms in known NPNT functional domains. NPNT is composed of three primary components, including five epidermal growth factor-like (EGF-like) domains, a linker region containing an integrin binding motif and integrin-binding enhancer, and a C-terminal MAM domain. Alpha-fold protein structure analysis demonstrated that the inclusion of exon 3 in the NPNT Isoform A precursor disrupts the first EGF-like domain in NPNT (Figure 5), likely leading to functional consequences related to this domain, such as epidermal growth factor signaling or binding affinity to ECM components.

Figure 5: Inclusion of exon 3 results in alteration of an EGF-like domain.

Alphafold structural analysis of the exon 3 alternative splicing event associated with the rs34712979-A COPD risk allele. A.) The predominant transcript expressed in lungs (NM0010330479) contains 5 EGF domains near the N-terminal. The first EGF domain spans exons 2 and 4. B.) In addition to including a serine residue, the alternative A allele increases the insertion of exon 3 which disrupts the first EGF domain. C.) AlphaFold model of N-terminal of NPNT for exon 3 exclusion and intact EGF domain colored in green. D.) AlphaFold model of NPNT N-terminal with exon 3 spliced in (outlined in orange), which is predicted to alter folding of the first EGF domain.

Discussion

Our results confirm that rs34712979 is a functional GWAS variant that acts through alternative splicing of NPNT. Our analyses led to three main findings: 1) genetic association and sQTL studies link NPNT to increased risk of COPD and decreased lung function through alternative splicing, 2) rs34712979 is the functional genetic variant responsible for COPD-associated splicing changes in NPNT, and 3) the splicing change most likely to increase risk for COPD is the increased inclusion of exon 3, which is predicted to disrupt the first EGF-like domain in NPNT. Together these results provide the strong evidence linking a COPD GWAS signal to a specific gene and functional mechanism, and the involvement of NPNT supports an emerging trend from COPD GWAS that links altered wound healing and epithelial-ECM interactions to COPD pathogenesis[26]. The genetic association patterns reflect the cumulative data of hundreds of thousands of subjects, and the splicing effects were demonstrated in lung RNA from over 1,000 subjects in two different human cohorts.

NPNT is an extracellular protein involved in tissue development, remodeling, and repair[27]. Full-length NPNT protein includes five epidermal growth factor (EGF)-like functional domains followed by an Arg-Gly-Asp (RGD) integrin-binding domain, a synergy site for high-affinity binding to ⍺8β1, and a meprin A-5 protein-receptor tyrosine phosphatase mu (MAM) domain[27]. It was first identified as the high affinity ligand for integrin α8β1in kidney[28, 29]. Subsequent studies showed that NPNT-α8β1 interactions were essential for normal kidney development in the mouse[30]. NPNT is secreted into basement membranes, where it is presumed to play a role in ECM-mediated cell-signaling that may involve binding to ECM constituents, activation of cell-surface receptors such as the EGF receptor, or integrin-mediated signaling activity through the RGD domain[31]. Although NPNT is highly expressed in the lung[32], its role in pulmonary biology is not fully understood. Global deletion of NPNT in the mouse results in lung lobar fusion[33], and recent evidence from an Npnt conditional knockout mouse model suggests that NPNT plays a role in resolution of inflammation and injury[14]. NPNT has also been linked to osteoblast differentiation and bone remodeling[34], invasiveness of breast cancer[35], and pulmonary silicosis[36].

The NPNT genetic association lies in a region first described in association with lung function[37, 38]. Subsequent studies using eQTL signals confirmed an association with COPD and identified two independent signals at this locus[39]. Several genes in this region have previously been hypothesized to be the effector genes, including the nearby genes GSTCD and INTS12 [40] in addition to NPNT[1, 41], indicating that this genomic region may contain multiple independent COPD genetic risk variants. Our study elucidates the effect of a causal genetic variant in this region on alternative splicing of NPNT, providing new mechanistic detail that elucidates the COPD-related biology in this genomic region. Interestingly, a recent Mendelian randomization and colocalization analysis discovered that the NPNT sQTL signal also colocalizes with the GWAS signal for COVID-19 severity, and the rs34712979 A allele that increases COPD risk is associated with decreased severity of COVID-19 infection, suggesting that NPNT plays a nuanced role in acute and chronic lung injury. It has also been shown that COVID-19 GWAS data does not colocalize with NPNT eQTLs, further supporting our findings that the GWAS association of rs34712979 is driven by splicing and not total gene expression level. Yoshi et al., 2023 observed an effect of rs34712979 on total NPNT protein levels, but they speculated that this may be caused by unrecognized isoform-specific binding of the SomaScan aptamers used to quantify NPNT in this study[16].

We identified four distinct splicing changes in NPNT that were significantly associated with rs34712979, including increased inclusion of the third exon of NPNT and the creation of an alternate splice acceptor by the COPD risk allele rs34712979-A, which results in the inclusion of a novel serine residue near the signal peptide sequence of NPNT. However, when we tested each of these splicing events for association to COPD and related phenotypes, the serine insertion showed no association to pulmonary phenotypes while the exon 3 inclusion event was significantly associated to decreased FEV1/FVC and the amount of pulmonary emphysema in the LTRC. In silico functional analysis also supported a larger biological effect of the exon 3 event, which is predicted to disrupt the first EGF-like binding domain, providing compelling evidence for future studies of the functional effects of this isoform.

Long read sequencing revealed for the first time the full length NPNT isoforms present in lung, indicating that the most common isoform (Isoform Precursor B) does not include exon 3 and has an intact stretch of EGF-like domains. However, subjects with the AA genotype had significantly more expression of a minor isoform (Isoform Precursor A) that includes exon 3, leading to the hypothesis that this isoform results in a functionally altered version of NPNT that may have measurable effects on lung biology, including wound healing and repair. Validation and quantitative assessment of these NPNT proteoforms in future work will be necessary to confirm this hypothesis, as will functional studies of the role of various NPNT isoforms on lung matrix composition, wound healing, inflammatory responses, and lung development.

These observations for NPNT contribute to an emerging literature on the role of COPD-causing genes that are involved in ECM-epithelial interactions[26]. NPNT is also one of a growing number of genes that are implicated in COPD pathogenesis via alternative splicing[8, 9]. Since COPD is a polygenic disorder, NPNT is one of dozens or hundreds of genes that influence COPD susceptibility, but as the functional picture around each of these genes emerges, the variety of therapeutic possibilities for this debilitating and morbid condition will be greatly expanded.

The strengths of this study are that the GWAS and RNA-seq findings are based on the analysis of a very large number of human samples, and the integrated analysis of short and long-read RNA-seq data provides a high level of resolution to observe changes at the isoform level. To our knowledge, this is one of the first applications of long-read sequencing to human tissue samples to demonstrate splicing-related effects of a GWAS-identified genetic variant. While there is strong statistical support for our findings, these are nonetheless correlative findings from large human cohorts. A limitation of this study is that we do not yet know the functional role of NPNT in the context of COPD and lung function, therefore further work is required to define the underlying molecular mechanisms and provide further experimental evidence to demonstrate how these mechanisms may alter COPD risk. An additional limitation is that the gene expression data is from bulk sequencing of tissues, and therefore we cannot determine the cell specificity of the splicing events observed.

In summary, these analyses demonstrate that the pulmonary disease GWAS association near NPNT is very likely to be mediated by a common genetic variant that alters transcript isoform ratios. Given the known function of NPNT in extracellular matrix biology and the high expression of NPNT in lung tissue and pneumocytes[41], further investigation of the functional consequences of this splicing variant, including its effects on NPNT protein structure and function, are likely to further elucidate causal mechanisms of COPD pathogenesis.

Data and Code Availability

LTRC genotyping data are available on dbGaP with accession number phs001662.v1.p1