Article Text
PDF
Abstract
Introduction Fibroblastic foci represent the cardinal pathogenic lesion in idiopathic pulmonary fibrosis (IPF) and comprise activated fibroblasts and myofibroblasts, the key effector cells responsible for dysregulated extracellular matrix deposition in multiple fibrotic conditions. The aim of this study was to define the major transcriptional programmes involved in fibrogenesis in IPF by profiling unmanipulated myofibroblasts within fibrotic foci in situ by laser capture microdissection.
Methods The challenges associated with deriving gene calls from low amounts of RNA and the absence of a meaningful comparator cell type were overcome by adopting novel data mining strategies and by using weighted gene co-expression network analysis (WGCNA), as well as an eigengene-based approach to identify transcriptional signatures, which correlate with fibrillar collagen gene expression.
Results WGCNA identified prominent clusters of genes associated with cell cycle, inflammation/differentiation, translation and cytoskeleton/cell adhesion. Collagen eigengene analysis revealed that transforming growth factor β1 (TGF-β1), RhoA kinase and the TSC2/RHEB axis formed major signalling clusters associated with collagen gene expression. Functional studies using CRISPR-Cas9 gene-edited cells demonstrated a key role for the TSC2/RHEB axis in regulating TGF-β1-induced mechanistic target of rapamycin complex 1 activation and collagen I deposition in mesenchymal cells reflecting IPF and other disease settings, including cancer-associated fibroblasts.
Conclusion These data provide strong support for the human tissue-based and bioinformatics approaches adopted to identify critical transcriptional nodes associated with the key pathogenic cell responsible for fibrogenesis in situ and further identify the TSC2/RHEB axis as a potential novel target for interfering with excessive matrix deposition in IPF and other fibrotic conditions.
- idiopathic pulmonary fibrosis
Statistics from Altmetric.com
Key messages
What is the key question?
Can we identify a transcriptional signature associated with collagen gene expression in the fibrotic focus, the cardinal fibrotic lesion in idiopathic pulmonary fibrosis (IPF)?
What is the bottom line?
We herein define the major transcriptional programmes involved in fibrogenesis in IPF by profiling myofibroblasts within fibrotic foci in situ by laser capture microdissection.
Why read on?
The data provide strong support for a human tissue-based approach to identify critical transcriptional nodes associated with fibrogenesis in situ and further identify the TSC2/RHEB axis as a potential novel target for interfering with excessive matrix deposition in IPF and other fibrotic conditions.
Introduction
Fibrosis, defined as the abnormal accumulation of extracellular matrix (ECM), is a pathological feature of several major chronic inflammatory and metabolic diseases, including idiopathic pulmonary fibrosis (IPF), the most rapidly progressive and fatal fibrotic condition with a median survival of 3.5 years from diagnosis.1 Fibrosis can also influence cancer progression as part of the stromal response to the tumour.2 Current antifibrotic drugs, pirfenidone and nintedanib, approved for the treatment of IPF, slow rather than halt disease progression and have significant side-effect profiles.3 4 There therefore remains an urgent need to develop novel therapeutic strategies for IPF and other fibrotic conditions.
The aetiology of IPF remains poorly defined but the current favoured hypothesis proposes that IPF arises as a result of a highly dysregulated wound healing response following repeated epithelial injury in response to an environmental inciting agent in genetically susceptible and aged individuals. IPF has a distinct histopathological pattern of usual interstitial pneumonia (UIP) in which fibroblastic foci (FF), comprising aggregates of activated fibroblasts and myofibroblasts embedded in a collagen-rich ECM, represent discrete sites of lung injury, repair and active fibrogenesis.5 These lesions represent a key histological diagnostic feature of UIP and their abundance is related to disease progression in IPF.6 In order to shed novel light on the mechanisms leading to pathogenic fibrogenesis in IPF, we sought to define the transcriptional signature of these cardinal fibrotic lesions.6 While single-cell RNA sequencing (scRNAseq) has been reported in IPF,7 this approach does not discriminate between profiles obtained from cells within FF or from other non-involved locations within the lung. Laser capture microdissection (LCM) of IPF lung tissue provides an ideal approach to capture and profile unperturbed myofibroblasts exclusively from within FF. Comparator cells are challenging to define given the potential diverse origins and phenotypes of myofibroblasts in IPF and importantly the paucity of interstitial fibroblasts, which can be captured from healthy control lung. Moreover, recent scRNAseq data8 suggest that ‘pathological ACTA2-expressing IPF myofibroblasts represent an extreme pole of a continuum connected to a quiescent ACTA2-negative stromal population rather than resident or activated fibroblasts in control lung’. We herein report on the generation of a composite gene expression profile from more than 60 individual FF derived from 13 patients. We adopt a novel informatics approach that obviates the need for a comparator group, based on the principles of WGCNA. We further employ a collagen eigengene approach9 to define clusters of genes, which functionally correlate with collagen I and III gene expression and therefore inform on the critical transcriptional programme underpinning pathogenic fibrogenesis. Key signalling nodes identified within the collagen eigengene cluster comprised gene signatures associated with the master regulator of fibrogenesis, transforming growth factor β1 (TGF-β1) as well as the TSC2/RHEB axis. This axis is a well-characterised molecular switch for the activation of mechanistic target of rapamycin complex 1 (mTORC1), a key molecular node that integrates metabolic, energy, hormonal and nutritional signals to regulate downstream cellular responses, including proliferation, growth, metabolism and protein synthesis.10 Functional studies using CRISPR-Cas9 gene-edited primary human lung fibroblasts (pHLFs) confirmed the involvement of RHEB in TGF-β1-mediated mTORC1 activation and collagen I deposition. This observation was generalisable to other mesenchymal cells, including hepatic stellate cells (HSCs) and cancer-associated fibroblasts (CAFs). Taken together, the findings reported provide strong support for the transcriptomic approach adopted to identify functionally relevant pathways involved in pathogenic fibrogenesis in situ. Our data further suggest that the TSC2/RHEB axis might represent a potential therapeutic target for interfering with fibrogenesis in the setting of fibrosis and the stromal reaction in cancer.
Methods
Patient material
Frozen IPF lung tissue was obtained from patients either undergoing lung transplantation for end-stage disease (n=10) or surgical lung biopsy for diagnostic purposes (n=3). Patients with IPF were diagnosed in accordance with current international guidelines.11 The human biological samples were sourced ethically and their research use was in accord with the terms of the informed consents (11/NE/0291; 12/EM/0058; 10/H0504/9).
LCM and microarray analysis
Frozen optimal cutting temperature compound (OCT) embedded samples were cryosectioned into 8 µm sections and processed with a modified H&E stain.12 FF were identified by immunostaining a serial guide slide for α-smooth muscle actin (αSMA) and captured using a PALM MicroBeam 4 Laser Microdissection microscope (Zeiss). Care was taken to avoid capturing overlying epithelium (visualised by immunostaining for CK7 on a second serial guide slide). Three to six captures (~0.015 mm2) per sample were collected. RNA was extracted using the PicoPure RNA Isolation kit (Life Technologies) and processed for hybridisation to Affymetrix HG-U133_Plus_2.0 microarrays. Five out of 65 cDNA samples did not meet the QC requirement and were therefore excluded for the analysis.
RNAscope in situ hybridisation
All RNAscope analyses were performed on independent IPF lung tissue (Advanced Cell Diagnostics) using a standard RNAscope 2.5 HD Red protocol according to the manufacturer’s instructions.
WGCNA and collagen eigengene analysis
Weighted gene co-expression network analysis (WGCNA) was performed on the 9035 probe sets defined as the composite IPF FF signature in R V.3.3.3.9 Briefly, we selected a power (β) of 4 (the minimum power with R2 >0.9) to construct the co-expression network from which the topological overlap matrix was calculated. To generate gene modules, we used minModuleSize=200 and default cutHeight, which yielded 16 modules. Modules with eigengene correlations R >|0.7| were merged using the mergeCloseModules function leaving 14 independent modules. Biological function to the modules was assigned using MetaCore (Thomson Reuter) pathway enrichment. Sankey plots were generated to describe the biological function of modules.
Statistical analysis
Statistical analysis of gene expression data was performed in R V.3.3.3. Correlation coefficients were calculated using the Pearson method (base). WGCNA was performed using the wgcna package. P values for correlations were calculated using corPvalueStudent function in the wgcna package. Enrichment analysis and network construction were performed in R V.3.3.3 using the metabaseR package (Thompson Reuters) database V.6.36.69400. The Girvan-Newman algorithm13 was used to identify subnetworks in the collagen module.
In vitro experiments
Control and IPF pHLFs (REC reference 12/EM/0058) and CAFs from lung adenocarcinoma (via the Tracking Cancer Evolution through Therapy (TRACERx) clinical study, REC reference 13/LO/1546) were grown from explant cultures as previously described.14 Primary adult HSCs were obtained from Zen-Bio (#HP-F-S). A representative primary line for each disease state was used. RNAseq was performed on control pHLFs as recently described by our laboratory . 15 (GSE102674). CRISPR-Cas9 gene editing followed the protocol described in our previous article 14 using a guide RNA sequence targeting RHEB (AGATGCCGCAGTCCAAGTCCCGG). Protein phosphorylation and expression were assessed by western blotting as previously described . 14 using p-4EBP1 (#13443), p-P70S6K (#9234), 4EBP1 (#9644), P70S6K (#9202) and RHEB (#13879) antibodies (all from Cell Signalling). α-tubulin (#9099) was used as a loading control. Type I collagen deposition under macromolecular crowding conditions was assessed by high-content imaging as described in the study by Chen et al.16 Procollagen production was based on levels of hydroxyproline quantified by reverse phase high-performance liquid chromatography as described in the study by Chambers et al.17 The distribution of the in vitro data was tested for normality using D'Agostino-Pearson test. All data presented were normally distributed and were therefore analysed by two-way analysis of variance with Tukey multiple comparison testing (GraphPad Prism). Data were considered statistically significant at p<0.05.
Results
Identification of an in situ gene expression signature for IPF FF
In order to generate an in situ gene expression signature for IPF FF, lesions from 13 flash-frozen IPF tissue samples (table 1; figure 1A) were captured by LCM (please see Methods section). RNA was isolated from 61 captures, amplified and cDNA hybridised to Affymetrix microarrays. To interrogate the cellular purity of the captures, we examined the normalised microarray intensity values of myofibroblast (ACTA2) and epithelial (CDH1, CK7, EpCAM) marker genes (figure 1B). The large difference in median log2 intensity values obtained for myofibroblast and epithelial cell markers provided assurance regarding the quality of the captures. We next validated and examined the spatial localisation of selected matrisomal mRNAs detected within the FF transcriptome in additional patient samples by selecting genes representing a range of microarray expression intensities (SPARC, ELN, POSTN) using RNAScope in situ hybridisation. An RNAScope signal was detectable for all three genes within FF but not in the overlying epithelium (figure 1C). Finally, we took advantage of a recently available scRNAseq dataset to interrogate enrichment for IPF fibroblast and myofibroblast marker genes.8 The fibroblast marker genes, HAS1 and HAS2, were detected in less than seven patients and were therefore excluded from our FF signature. The myofibroblast marker genes, COL8A1 and ACTA2, were detected in 13/13 and in 12/13 patients, respectively. These data suggest that myofibroblasts are enriched in IPF FF.
Baseline characteristics of IPF patients for tissues used for laser capture microdissection
Transcriptomic profiling of laser-captured fibroblastic foci from IPF patients. (A) Captures were obtained from fresh frozen IPF lung tissue obtained from 13 patients. αSMA-rich FF were identified by αSMA immunostaining and captured from serially stained H&E sections, taking care to avoid the overlying CK7-positive epithelium. Representative sections of αSMA (i) and CK7 (ii) staining are shown, as well as sections before (iii) and after (iv) laser capture of a typical fibroblastic focus. (B) Median log2 intensity and IQR values for myofibroblast (ACTA2) and epithelial (CDH1, CK7, EpCAM) marker genes. Data beyond the end of the whiskers are plotted individually. (C) Confirmation of spatial expression of selected genes within FF in independent patient samples by RNAScope in situ hybridisation (SPARC, POSTN, ELN). (D) Generation of the fibroblastic focus gene expression signature. For each of the 13 patient lung samples, fibroblastic foci were captured, the RNA was extracted and amplified and cDNA was arrayed on an affymetrix HG-U133_plus_2.0microarray. Using the presence call information on the array, we excluded probes that were present in less than two captures for the same patient (approximately 5% error rate). The final probeset intensity was averaged per patient. Probesets detected in the majority (≥7) of the patient signatures were included to generate a final IPF FF gene expression signature/transcriptome. FF, fibroblastic foci; IPF, idiopathic pulmonary fibrosis; αSMA, α-smooth muscle actin.
To generate an in situ IPF FF transcriptome, we first generated independent gene signatures for each of the 13 IPF patients using an informatic pooling strategy (described in figure 1D). We then derived a composite IPF FF transcriptomic signature by selecting probesets expressed in the majority (≥7) of individual IPF patient signatures. This cut-off identified 9035 probesets corresponding to 6667 individual genes. The complete dataset is available on Gene Expression Omnibus (GSE98925) (online supplemental table S1).
Supplemental material
WGCNA of the in situ FF transcriptome
In order to identify transcriptional programmes and biological pathways active in myofibroblasts within FF, we applied WGCNA and Metacore analyses of the composite FF transcriptome. WGCNA identified 14 independent co-expression modules reflecting sets of highly correlated gene transcripts (figure 2A, online supplemental table S1). We next aimed to identify modules with the highest proportion of genes upregulated in IPF compared with control lungs by cross-referencing published whole lung microarray expression data (GSE10667).18 The WGCNA turquoise module comprised genes with the most significant and highest average fold-change in IPF versus control (figure 2B,C) and was therefore considered to represent the most enriched module with potential disease-relevant biology in the laser-captured FF.
Weighted gene co-expression analysis (WGCNA) of the composite fibroblastic focus transcriptome. (A) The composite FF transcriptome was subjected to WGCNA and the gene dendrogram is shown identifying 14 independent co-expression modules. Modules can appear in multiple regions of the dendrogram as highly correlated, smaller modules are collapsed. (B) The gene expression in IPF versus healthy lung from published data (GSE10667) for all genes of each module was interrogated. The log2 ((diseased)/(healthy)) ratios are shown with >2-fold difference shown in black. (C) -log10 p values from a two-sided t test performed on each module log2 ratios, to determine which sets were significantly different from zero. FF, fibroblastic foci; IPF, idiopathic pulmonary fibrosis
Within the 2607 genes represented in the WGCNA turquoise module, we mapped 2051 Metacore network objects (representing genes, proteins, complexes and metabolites) and constructed a single biological network comprising 1021 IPF-enriched genes. In order to deconvolute this network further, we identified subclusters (ie, local neighbourhoods) and investigated the seven local neighbourhoods containing more than 20 genes (figure 3A). Pathway analysis of these local neighbourhoods identified key genes that were most influential in the pathway mapping and may therefore represent important regulatory nodes in FF (figure 3B, online supplementary figure S1). This included genes representing known cardinal features of the activated myofibroblast, such as cytoskeletal remodelling (RHOA, RAC1, ROCK, ARHGEF1), as well as highlighting JAK/STAT and NFκB signalling pathways and a small number of transcription factors.
Supplemental material
Network and pathway analysis of the weighted gene co-expression analysis turquoise module. (A) A biological network was constructed from the 2607 genes in the turquoise WGCNA module. After removal of the edges supported by less than three citations, 7 local neighbourhoods containing at least 20 genes were investigated. (B) Pathway analysis of the seven subclusters is shown displaying the genes that appear most frequently in pathways identified to represent key modulators of neighbourhood biology.
Collagen eigengene analysis of the in situ IPF FF transcriptome identifies key transcriptional programmes associated with fibrogenesis
We next sought to define the transcriptional programmes and therefore the key pathways associated with pathogenic collagen production in IPF. To this end, we adapted the core principle of WGCNA to create a bespoke collagen eigengene from the first principal component of COL1A1 and COL3A1 probesets and then correlated the rest of the microarray probesets making up the FF signature. Three hundred and sixty-two probesets, representing 300 individual genes, correlated either positively or negatively with the collagen eigengene (figure 4A, online supplemental table S1). Fifteen genes correlated with an R>|0.9|, including four genes that were highly inversely correlated with collagen gene expression.
Collagen eigengene analysis. (A) The expression of COL1A1 and COL3A1 mRNA probes across the 13 patient profiles was used to create a collagen eigengene (red line) and all other probesets correlating to this (p<0.01) assigned as collagen correlating genes (black lines). Array expression data is shown as Z-transformed log2 intensity. Z-transformed log2 intensities of negatively correlated genes were inverted for illustration. (B) A single interaction network, mapping genes to network objects, which can include genes, proteins or protein complexes, was derived from the 300 genes forming the collagen eigengene as represented by three or more citations and excluding implied interactions. Genes are represented in the network in uppercase and proteins/protein complexes are represented in lower case. The network includes a TGF-β1/ECM module (green), a Rho kinase module (red), a transcriptional module (yellow), a metabolism module (blue) and a TSC/RHEB module (purple). The proteins are linked with a red arrow to represent activation or a blue line to represent an inhibition with the thickness of the lines reflecting the existing body of evidence for the interaction. Gene names illustrated in black font were positively correlated with the collagen eigengene, whereas genes in red were negatively correlated. (C) Table reflects correlation coefficients for COL1A1, COL3A1, TSC2 and RHEB with comparable eigengenes (COL1A1 and COL3A1) calculated independently in the captured IPF FF (IPF FF LCM) and in previously published datasets for TGF-β1-stimulated human primary lung fibroblasts, IPF whole lung and in the rat bleomycin model. (D) Figure illustrates a proposed model for mTORC1 activation in response to a profibrotic stimulus (eg, TGF-β1) or a fibrotic microenvironment (eg, FF). TSC2 (−ve) and Rheb (+ve) expression correlations relative to the collagen eigengene would favour persistent mTORC1 activation and collagen deposition via GTP-bound Rheb. ECM, extracellular matrix; FF,fibroblastic foci; IPF, idiopathic pulmonary fibrosis; LCM,laser capture microdissection; mTORC1, mechanistic target of rapamycin complex 1; TGF-β1,transforming growth factor β1.
To further characterise this collagen eigengene module, we constructed an interaction network using MetaCore which we refined to a network of 66 genes following removal of predicted rather than experimentally validated interactions and those without directionality (figure 4B). This network featured genes that represent TGF-β1/ECM biology (green) and RhoA-associated cytoskeletal remodelling (red), supporting this network as a core regulatory feature of the FF myofibroblast phenotype. Additional modules included transcriptional regulation (yellow) and cell metabolism (blue), as well as a small module featuring TSC2 and RHEB, a well-characterised molecular switch for the activation of mTORC1 signalling. RHEB, which encodes the GTPase and proximal activator of mTORC1, RHEB, correlated positively with the collagen eigengene (R +0.71). In contrast, TSC2, which encodes tuberin within the TSC complex, a major allosteric inhibitor of RHEB, was inversely correlated (R −0.73). We next determined whether a similar TSC2/RHEB motif was represented in vitro in our published RNAseq dataset of TGF-β1-stimulated fibroblasts (GSE102674) and in published whole lung IPF and rat bleomycin model gene expression datasets (GSE10667 and GSE48455).15 18 A comparable eigengene was calculated using COLA1 and COLA3 expression levels and correlation coefficients for COL1A1, COL3A1, TSC2 and RHEB were generated. As expected, COLA1 and COLA3 expression correlated across all datasets; whereas the specific RHEB and TSC2 correlations were only present in TGF-β1-stimulated fibroblasts but not in datasets from whole fibrotic human and murine lung, where transcriptional profiles represent a composite derived from multiple cell types (figure 4C). The LCM FF and TGF-β1-stimulated fibroblast correlations would therefore support a potential model favouring mTORC1 activation (figure 4D).
Functional validation of the TSC2/RHEB node during fibrogenesis in vitro
Given the robust correlation of the TSC2/RHEB node with the collagen eigengene in IPF FF and in TGF-β1-activated myofibroblasts in vitro, we next sought to determine whether TSC2/RHEB represents a potential molecular switch involved in regulating collagen synthesis in TGF-β1-stimulated pHLFs. To this end, we generated RHEB-deficient pHLFs using CRISPR-Cas9 gene editing. Monitoring the phosphorylation of the mTORC1 downstream substrates, P70S6K and 4E-BP1 (Serine 65) at 3 hours following TGF-β1 stimulation in control and RHEB-deficient fibroblasts revealed that RHEB was critical for TGF-β1-induced mTORC1 activation (figure 5A). Analysis of procollagen synthesis, based on the quantitation of hydroxyproline, further demonstrated that RHEB was necessary for TGF-β1-induced procollagen synthesis (figure 5B). This observation was validated by specifically assessing collagen I deposition in response to TGF-β1 stimulation under macromolecular crowding conditions by high-content imaging (figure 5C,D). Finally, we also measured procollagen synthesis following TGF-β1 stimulation in IPF pHLFs, HSCs and CAFs. In each case, there was an almost complete inhibition of TGF-β1-induced collagen production in RHEB-edited cells (figure 6).
RHEB is essential for TGF-β1-stimulated mTORC1 activation and collagen production in primary human lung fibroblasts (pHLFs). (A) CRISPR-Cas9 gene editing was used to disrupt RHEB in pHLFs. A predesigned control gRNA sequence was used to generate matched wild-type pHLFs (crCTRL). RHEB-disrupted (crRHEB) and crCTRL pHLFs were stimulated with TGF-β1 (1 ng/mL) or media alone for 3 hours. Successful gene editing of RHEB was confirmed by immunoblotting. mTORC1 signalling was evaluated by monitoring the phosphorylation of the mTORC1 downstream substrates, p70S6K and 4E-BP1 by immunoblotting. (B) Procollagen production was measured by quantifying hydroxyproline levels in ethanol-insoluble proteins in cell supernatants after 48 hours stimulation with TGF-β1 (1 ng/mL) by reverse-phase high-performance liquid chromatography and normalised to the cell count. (C) Collagen I deposition after 48 hours stimulation with TGF-β1 (1 ng/mL) was measured by high-content imaging of collagen I immunofluorescence in cells grown under macromolecular crowding conditions. Data are represented as collagen I signal normalised to cell count (n=4 fields per well). (D) Representative images of collagen I deposition are shown. Data in graphs are presented as mean (±) SD of three replicate wells per condition. Differences between groups were evaluated by two-way analysis of variance with Tukey multiple comparison testing, *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001. mTORC1, mechanistic target of rapamycin complex 1; TGF-β1,transforming growth factor β1.
RHEB is essential for TGF-β1-stimulated collagen deposition in IPF pHLFs, CAFs and HSCs. CRISPR-Cas9 gene editing was used to disrupt RHEB (crRHEB) and a crCTRL was used as a control in (A) IPF pHLF; (B) primary CAFs derived from lung adenocarcinoma and (C) primary HSCs. (i) Successful RHEB gene editing was confirmed by assessing RHEB levels by immunoblotting and (ii) procollagen production was measured by quantifying hydroxyproline levels in ethanol-insoluble proteins in cell supernatants after 48 hours following stimulation with TGF-β1 (1 ng/mL) by reverse-phase high-performance liquid chromatography and normalised to the cell count. Data are presented as mean (±) SD of three replicate wells per condition. Differences between groups were evaluated with two-way analysis of variance with Tukey multiple comparison testing, *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001. CAF, cancer-associated fibroblast; HSCs, hepatic stellate cells; IPF, idiopathic pulmonary fibrosis; pHLFs, primary human lung fibroblasts; TGF-β1, transforming growth factor β1.
Discussion
The aim of this study was to identify key transcriptional pathways driving fibrogenesis in IPF by fingerprinting myofibroblasts located within FF, the defining pathological lesion of IPF/UIP. The use of informatics-based analysis of gene expression data to identify genes, pathways and mechanisms of relevance in organs and tissues of interest has become an established path to uncovering important pathobiology.19 scRNAseq methodologies are rapidly advancing, and recent studies have provided important insights into the heterogeneity and diversity of mesenchymal, epithelial and alveolar macrophage populations in the setting of pulmonary fibrosis.20–22 However, spatial information, and therefore the precise anatomical location from which the cells are derived, is lost during single-cell dissociation. It is also worth commenting that the cell dissociation process can also potentially influence gene expression profiles.21 In order to generate an unperturbed gene expression signature for IPF myofibroblasts and identify pathways associated with pathological collagen expression within the cardinal lesion in IPF,6 we chose to adopt an LCM-based approach that was optimised to overcome the challenges associated with isolating high-quality RNA from a small number of cells embedded within a collagen-rich fibrotic lesion. Given the small number of frozen tissue blocks available for this study, five (IQR 1) FF were captured from 13 patient samples to yield a total of 61 FF captured. This allowed us to identify a unique set of 6667 genes representing a generic transcriptome of the FF.
Our bioinformatics approach obviated the need for a comparator cell type, in light of recent findings indicating that pathological ACTA2-expressing IPF myofibroblasts represent an extreme pole of a continuum connected to a quiescent ACTA2-negative stromal population rather than resident or activated fibroblasts in control lung. It is not possible to capture sufficient number of myofibroblasts from the interstitium of uninjured normal adult lung. The lack of a suitable comparator cell type to identify differentially expressed genes in FF was obviated by using contemporary bioinformatic approaches. WGCNA has become widely adopted to identify groups of genes with similar expression profiles (modules) in large high-dimensional genomic datasets. The functional correlation of genes aligned by expression within an eigengene underlies the principle of the WGCNA methodology where individual modules represent discrete functions.9 These modules highlight related biological mechanisms and can provide insights into disease pathomechanisms. The largest WGCNA turquoise module identified within our dataset comprised 2607 genes and was the most significantly enriched module in terms of genes previously identified to be upregulated in whole IPF lungs compared with healthy controls.18 RhoA/Rac1/ROCK signalling was prominent within the WGCNA turquoise module, and this feature is consistent with the coordinated cytoskeletal rearrangements associated with highly contractile cells, such as myofibroblasts.23 Strong representation of NFκB signalling in FF was not as predictable but adds to previous evidence implicating this signalling hub in the context of liver fibrosis24 25 and further provides support for a role for FF myofibroblasts as potential proinflammatory modulators in IPF.26 JAK1, STAT1 and STAT3 signalling were represented in another node. Of interest, TGF-β1, IL-6 and Oncostatin M have all been reported to engage JAK1, STAT1 and STAT3 signalling,27 and genetic modulation of STAT3 is protective in the bleomycin-induced lung fibrosis model.28 Moreover, therapeutic approaches targeting these mechanisms are currently being pursued in the context of IPF and other fibrotic indications.29 30 Finally, WGCNA analysis also identified translation as a prominent feature of the transcriptional signature of FF and is consistent with the highly ECM synthetic nature of myofibroblasts within FF.
The second bioinformatics approach aimed to identify key programmes associated with the generation of pathogenic ECM and was based on the core principle of WGCNA analysis that was adapted in order to use the first principal component of the expression of fibrillar collagen genes as a measure of underlying collagen production. We followed convention by naming this a collagen eigengene.31 To the best of our knowledge, this is the first report describing the implementation of this approach to define pathways underlying fibrogenesis in IPF or indeed any other human fibrotic condition. This analysis identified around 300 genes with expression patterns that were either positively or negatively correlated with the collagen eigengene. It is worth mentioning that we found no significant overlap between these collagen-related genes and the list of genes differentially expressed in myofibroblasts in IPF compared with control lung reported in a recent scRNAseq study.8 This suggest that scRNAseq and our LCM approach could be considered complementary strategies to interrogate different aspects of IPF pathomechanisms.
Interestingly, there was only one interaction network within the 300 genes comprising known protein–protein interactions validated by multiple publications, resulting in a composite module. It is worth commenting that 4 of the top 15 correlated genes in the dataset were inversely correlated with the collagen eigengene (ATG4B, PLEC, SELM and WIZ) and may reflect mechanisms that are downregulated or inhibited during the course of collagen production. Of interest, Atgb4-deficient mice exhibit more extensive and severe fibrosis with increased accumulation of collagen relative to control mice in the bleomycin model of lung fibrosis.32
Cytoskeletal reorganisation, TGF-β1 signalling, ECM production and cell–matrix interactions are fundamental properties expected of myofibroblasts. These mechanisms mapped strongly to the collagen eigengene signature and therefore provided important validation of the utility of the LCM and collagen eigengene approach to identify biology of relevance to the primary functions of these cells. It is worth highlighting the presence of a TGF-β1 signalling submodule, which includes both the TGF-β1 ligand and the αvβ5 integrin, which has been implicated in the activation of latent TGF-β1 in vitro.33 34 To the best of our knowledge, this represents the first human tissue-based evidence demonstrating a link for this TGF-β1-integrin axis and fibrogenesis in IPF.33 35
RhoA signalling again featured strongly within this network and was positively correlated with the collagen eigengene. This network also comprised two guanine nucleotide exchange factors involved in the regulation of the activation of RhoA signalling, ARHGEF2 and AKAP13/LBC, which were also positively correlated with the collagen eigengene. Interestingly, AKAP13/LBC has recently been identified as a susceptibility gene in IPF.36 While expression of AKAP13 was reported to be highest in damaged epithelium and lymphoid cells, expression was also noted in cells within FF. Moreover, ablation of AKAP13 expression in cardiac fibroblasts strongly impinges on collagen production in response to a profibrotic challenge37 so that AKAP13/LBC may represent both a susceptibility gene and a potential drug target for inhibiting profibrotic signalling.
Hand curation of this composite module also identified less predictable elements, such as fatty acid (FA) metabolism, including the gateway enzyme for FA synthesis, FASN, which was inversely correlated with the collagen eigengene. The current literature indicates both profibrotic and antifibrotic roles for altered FA metabolism in the context of pulmonary fibrosis.38 39 Whereas, endogenous-nitrated FAs exert antifibrotic effects by upregulating peroxisome proliferator-activated receptor γ in human lung fibroblasts, another study showed that FASN is required for profibrotic TGF-β signalling.40
The observation that the collagen eigengene network also comprised a small TSC2/RHEB module, a prominent intrinsic regulator of mTORC1 activity, was of particular interest in light of the recent evidence highlighting a key role for the mTORC1/4E-BP1 axis as a core fibrogenic pathway downstream of TGF-β1.14 The current LCM FF transcriptome and our previously published RNAseq dataset of TGF-β1-stimulated pHLFs further revealed that RHEB was positively correlated with the collagen eigengene, whereas TSC2 was inversely correlated, supporting a potential model of persistent mTORC1 activation. The specific mechanism by which TGF-β1 promotes the activation of mTORC1 remains poorly understood, and it is also recognised that TSC2/RHEB signalling can act independently of mTORC1.41 Using RHEB CRISPR-Cas9 gene editing, we now provide strong evidence for a key role for RHEB in mediating both TGF-β1-induced mTORC1 activation and downstream collagen deposition in mesenchymal cells reflecting IPF and other disease settings, including the stromal reaction in cancer. It is also worth commenting that the lack of any correlations for TSC2 and RHEB with the collagen eigengene in previously published datasets based on RNA extracted from whole IPF lungs, the bleomycin model and recent scRNAseq studies of IPF lung7 18 emphasises the power of the LCM approach adopted in the present study.
In summary, the findings reported in this study provide strong support for the experimental and bioinformatics approaches adopted to identify key transcriptional programmes associated with FF. As well as confirming expected pathways associated with myofibroblast biology, this analysis provides unprecedented biological insights into some of the key signalling networks associated with fibrogenesis in situ and place TSC2/RHEB downstream of TGF-β1 signalling as being central to the fibrogenic response in multiple mesenchymal cells. Targeting this axis may hold therapeutic potential in the context of IPF and other conditions associated with pathogenic fibrosis, including the stromal reaction in cancer.
Acknowledgments
The authors acknowledge several institutions for the provision of human IPF lung tissue, including the Institute of Transplantation, Freeman Hospital, Newcastle UponOn Tyne (n=9 patients undergoing lung transplantation for end–stage disease); University College London Hospitals (n=3 patients undergoing surgical lung biopsy for diagnostic purposes) and the Brompton and Harefield NHS Trust (n=1 patient undergoing lung transplantation for end–stage disease). The authors thank members of the lung TRACERx consortium whose study enabled the derivation of the CAFs used in this study. The authors thank Dr Pascal Durrenberger (Centre for Inflammation and Tissue Repair, UCL Respiratory, University College London, London, WC1E 6JJ, UK) for the images provided in figure 1A.
References
Supplementary materials
Supplementary Data
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
Footnotes
DG, ART and MP are joint first authors.
ADB and RCC are joint senior authors.
Twitter @DelphineGuillo6, @Manup253, @robhynds
ADB and RCC contributed equally.
DG, ART and MP contributed equally.
Contributors DG performed in vitro experiments, interpreted data, prepared figures and participated in drafting of the manuscript and reviewed and approved the final version. AT designed the bioinformatics analysis workflow, participated in drafting the manuscript and reviewed and approved the final version. MP performed in vitro experiments and analysed data and reviewed and approved the final manuscript. PFM supervised the study and identified IPF FF and reviewed and approved the final manuscript. LME designed the bioinformatics analysis workflow and approved the final manuscript. RH and GM performed laser capture microdissection (LCM) of IPF FF and microarray analysis and approved the final manuscript. RJM obtained ethical approvals, provided patient tissue, participated in the identification of IPF FF and reviewed and approved the final manuscript. AJF and TMM obtained ethical approvals, consented patients and provided IPF patient tissue, collected clinical data and reviewed and approved the final manuscript. REH isolated and characterised cancer-associated fibroblasts (CAFs) from patients with lung adenocarcinoma (TRACERx clinical study) and approved the final manuscript. MJ-H on behalf of the TRACERx consortium obtained ethics approval, provided lung adenocarcinoma patient tissue, collected clinical data and approved the final manuscript. RM conceived the study and reviewed and approved the final manuscript. RCC and AB conceived and designed the study, interpreted data, drafted the manuscript and reviewed and approved the final version.
Funding RCC acknowledges funding support via a collaborative framework agreement between University College London (UCL) and GSK; and from the National Institute for Health Research (NIHR) University College London Hospitals Biomedical Research Centre (UCLH BRC). AJF acknowledges funding support via a collaborative framework agreement between Newcastle University and GSK; and from the NIHR Newcastle Biomedical Research Centre (Newcastle BRC). TMM is supported by an NIHR Clinician Scientist Fellowship (NIHR Ref: CS-2013-13-017), holds a British Lung Foundation Chair in Respiratory Research (C17-3) and acknowledges funding support from the NIHR Royal Brompton Hospital Biomedical Research Unit (BRU). The TRACERx study is funded by Cancer Research UK (CRUK; C11496/A17786) and the derivation of TRACERx patient models was supported by the CRUK Lung Cancer Centre of Excellence. MP acknowledges funds from the British Lung Foundation.
Competing interests RCC received grants from GSK via her institution during the conduct of the study; consultancy fees from Pieris Pharmaceuticals, consultancy fees from Chiesi, consultancy fees from Theravance Biopharma outside the submitted work; and RCC spouse was an employee of GSK during the time this work was conducted. This potential COI was managed by a research framework agreement between UCL and GSK. DG and PFM received salary from this research framework agreement. RM, AT, LME and AB were employees of GlaxoSmithKline during the time this work was performed and hold GSK shares. AJF has received a research grant from GSK to support work contained in this manuscript. RH and GM were employees of Epistem with a commercial interest in the provision of services in the area of investigation. TMM has, via his institution, received industry-academic funding from GSK, R&D and UCB and has received consultancy or speakers fees from Apellis, AstraZeneca, Bayer, Blade Therapeutics, Boehringer Ingelheim, Bristol-Myers Squibb, Galapagos, GlaxoSmithKline R&D, Indalo, Novartis, Pliant, ProMetic, Respivnat, Roche, Samumed and UCB. MJH reports non-financial support from Achilles Therapeutics outside the submitted work. MP, RJM and REH have nothing to disclose.
Patient consent for publication Not required.
Provenance and peer review Not commissioned; externally peer reviewed.
Data availability statement All the relevant data are included in the article, uploaded as supplementary information, or available on the Gene Expression Omnibus (GSE98925).