Early events marking lung fibroblast transition to profibrotic state in idiopathic pulmonary fibrosis

Raw scRNA-seq data were derived from our previous publication [8]. Briefly, normal and IPF lung tissue (Additional file 1: Table S1, Additional file 2: Fig. S1) was obtained and processed to obtain single cell suspension. Single cell libraries were prepared using the 10X Genomics Chromium instrument and V2 chemistry. Sequencing was performed on an Illumina NextSeq-500 instrument.

scRNA-seq raw count and cell-UMI (Unique Molecular Identifier) count matrix were generated by Cell Ranger [8]. Single-cell Analysis Via Expression Recovery (Saver) was used to impute dropout events in gene expression [18]. Seurat (version 2.3.4) was used to normalize gene expression, perform differentially expressed gene analysis, identify distinct cell populations and visualize clusters graphically [19, 20]. The cell-UMI matrix was filtered and only cells expressing at least 200 genes were further analyzed. Cells containing greater than 35% of mitochondrial genes were also excluded from the analysis. Highly variable genes were identified, based on their average expression and dispersion, and would be used in the downstream analysis. Data were scaled and the number of UMIs per cell as well as the percentage of mitochondrial gene content were regressed out. In this study, we further removed the effect from technical or biological confounders using Harmony, which integrates multiple scRNA seq datasets by projecting cells into a shared embedding in which cells are grouped by cell type, not the specific conditions related to the datasets [21]. t-Distributed stochastic neighbor embedding (t-SNE) plots based on Harmony embeddings were generated to assign clusters and each cluster was identified by differentially expressed genomic signatures.

A mixed population of cells (48,023 cells) from the lungs of three healthy control (21,485 cells) and six IPF (26,538 cells) samples (from three donors) was sequenced using next generation single cell sequencing. An important study design includes tissue from three upper lobes (IPF_UP; 12,577 cells) and fibrotic tissue from three lower lobes (IPF_LOW; 13,961 cells) (Fig. 1A). In general, samples from the upper lobes presented a relatively unaffected lung histology, while all samples from the lower lobes showed advanced fibrosis [8]. Here, we performed imputation of dropouts and gene expression profiles were clustered and visualized (t-SNE). Our analysis resulted in 20 clusters (Fig. 1B) that were identified as distinct cell types (Fig. 1C), using previously described cell markers [8]. Each cluster had variable cell counts coming from healthy and IPF upper and lower lung lobes (Fig. 1D, E and Additional file 2: Fig. S1).

Fibroblasts (Fig. 1C, green box) were re-clustered and visualized in two dimensions using t-SNE (Fig. 2A). Overall, the clustering of fibroblasts from 449 control, 404 IPF_UP and 1085 IPF_LOW samples revealed four different groups. A fifth cluster, neighboring B-cells, was found to contain cells expressing HLA type II family genes and was excluded from this analysis (Additional file 2: Fig. S2). Despite our best efforts, one of the IPF upper lung lobe samples (SC154) showed extensive fibrosis compared to normal controls (Additional file 2: Fig. S1A), posing a risk of masking any differences between the IPF_UP and IPF_LOW fibroblasts. We decided to include this sample in our analysis since it clustered similarly to the other fibroblasts in IPF_UP samples (Additional file 2: Fig. S3A). Similarly, we checked the age distribution of the fibroblast cells. Despite the fact that IPF samples generally come from older people (since IPF usually presents itself > 55 yrs), we did not see any other age bias (Additional file 2: Fig. S3B).

Cluster-1 (608 cells, 31% of all FB) contained mainly cells from the healthy control donors (80% of the control FBs were in Cluster-1) as well as a smaller percentage of the IPF_UP and IPF_LOW cells (24% and 14% respectively). Cluster-2, a much smaller cluster (197 cells, 10% of all FB), included cells from each sample origin. Cluster-3 (920 cells, 48% of all FB) contained mainly IPF_UP and IPF_LOW (56% and 64% respectively). Only 1% of the healthy control FBs classified as Cluster-3. Cluster-4 (213 cells, 11% of all FB) contained a slightly higher percentage of the IPF_UP and IPF_LOW (12% and 13% respectively) than the healthy control fibroblasts (6%). The breakdown of each cluster to sample types is summarized in Fig. 2B, C while the breakdown of each sample to different clusters is summarized in Fig. 2D.

Since fibroblasts in Cluster-1 came predominantly from control samples we consider this to represent the “normal state”; while those in Cluster-3, which came almost exclusively form IPF_UP and IPF_LOW, represent the “disease state”. Cluster-1, Cluster-2 and Cluster-4 had distinct expression profiles, but they were all present in healthy and IPF lungs (Fig. 2C). Overall, IPF_UP and IPF_LOW fibroblasts showed similar clustering and compositional patterns (Fig. 2B, D). These results showed that fibroblasts from the relatively non-fibrotic upper lobes had similar molecular signatures to those from the highly fibrotic lower lobes, suggesting that molecular changes in lung fibroblasts precede the morphological changes identified by histological examination.

We also noticed that both Cluster-1 and CLuster-3 contained some alveolar fibroblasts (SPINT2+FGFR4+GPC3+), while cluster 2 was predominately composed of adventitial fibroblasts (SFRP2+PI16+SERPINF1+) (Additional file 2: Fig. S4), which were previously reported [25]. Additionally, some fibroblast cells in Cluster-3 highly expressed LGR5, and LGR5+ fibroblasts serve as a mesenchymal niche for distal airway epithelial cells, as it was recently reported [26]. From the UMAP plots, LGR5+ fibroblasts are next adventitial fibroblasts (Additional file 2: Fig. S4A). Furthermore, part of Cluster-3 and Cluster-4 are CTHRC1+ fibroblasts, which are uniquely found in fibrotic lungs (Additional file 2: Fig. S4B). In a previous RNA velocity analysis in mouse fibroblasts, vectors from alveolar fibroblasts and adventitial fibroblasts pointed to CTHRC1+ fibroblasts [27]. Consistently, our velocity analysis showed that vectors from myofibroblast Cluster-4, alveolar fibroblasts, and LGR5+ fibroblasts, point towards CTHRC1+ fibroblasts.

Differentially expressed genes in each cluster are listed in Additional file 1: Table S3 and expression of the top 16 genes in these lists is further described per cluster and sample type (Figs. 3, 4, 5, 6). Briefly, fibroblasts in Cluster-1 had higher expression in 263 genes (FDR < 0.05) with the highest being: ICAM1, CXCL1, CXCL2, CXCL3, CXCL8, CCL2, IL6, PTX3, IER3, GADD45B, THBS1, SOD2 and NFKBIA which are associated with immune response and regulation of inflammation (Fig. 3 and Additional file 1: Table S4).

Among the 118 genes that were significantly upregulated in Cluster-2, CXCL14, SFRP2 and SFRP4 had higher expression levels in IPF samples. Interestingly, the highest levels of CXCL14 were detected in IPF_LOW fibroblasts, being one of the most significant differences observed overall between IPF_UP and IPF_LOW samples. Profibrotic gene CXCL14 is shown to play a regulatory role in immune response and inflammation while SFRP2 and SFRP4 are regulators (inhibitors) of Wnt signaling (Fig. 4 and Additional file 1: Table S4).

The majority of the top 16 upregulated genes in Cluster-3 (out of the total 219 differentially expressed) have been previously shown to increase in lung fibrosis (Fig. 5 and Additional file 1: Table S4). Our results corfirmed that cells expressing myofibroblasts markers like POSTN and ASPN also expressed LTBP2, LTBP1, BGN, DPT and the highest levels of COL3A1 and COL8A1. Interestingly, 68–93% of the cells on this cluster that had low expression of myofibroblast-associated genes also had high expression levels of IPF-associated profibrotic markers such as MFAP4, LTBP1, BGN, COMP, MMP2, COL3A1 and COL8A1. All 16 genes had similar expression in IPF_UP and IPF_LOW samples. CXCL14, which was upregulated in Cluster-2, was also upregulated in the Cluster-3 IPF_LOW compared to IPF_UP.

Cells in Cluster-4 expressed profibrotic and myofibroblast markers like TNC, collagens, SPARC, POSTN, FN1 and TPM2 in higher levels than cells in Cluster-3. All 16 top upregulated genes in this group had similar expression levels in IPF_UP and IPF_LOW samples (Fig. 6 and Additional file 1: Table S4).

Full names and relevant references for all genes mentioned above are listed in Additional file 1: Table S4.

Single cell trajectory inference and pseudotime analysis are often used to study cellular dynamics or transitional states and special organization of cells in tissues [28]. Each cell is assigned a numeric value (pseudotime) which indicates where in the underlying dynamic biological process that cell falls into. These methods allow the visualization of intermediate stages that connect distinct cell states, which are often overlooked during cell clustering. We used diffusion map (destiny) to perform pseudotime analysis on the four fibroblast subclusters (Fig. 7A). The diffusion components, as measures of pseudotime [29], revealed temporal ordering and cellular decision on the single-cell transcriptome level. Interestingly, cells in Cluster-1 had two distinct subpopulations (Fig. 7A, ovals) which corresponded to cells from Control and IPF samples with the latter being closer to Cluster-2 and 3 (Fig. 7B). This suggests that IPF fibroblasts in Cluster-1, despite their overall “healthy” phenotype, have been influenced by the disease environment.

To further our understanding of the mechanisms driving disease onset, we calculated RNA velocity [22], which is based on the balance of unspliced (nascent) and spliced (mature) mRNA. This high-dimensional vector can act as an indicator of the future state of mature mRNA, driving cell state. The directionality of the plot arrows depicts the direction of cell state progression. RNA velocity analysis of all fibroblasts (Fig. 7C) showed that the majority of the fibroblasts in Cluster-1 (composed mainly of non-IPF fibroblasts) trended towards a steady state (SS1). Even fibroblasts originating from IPF samples in this cluster appeared to be able to express genes driving them towards SS1. Similarly, cluster-3 fibroblasts are attracted to another steady state (SS2) which is likely disease related since almost all of these cells originated from IPF lungs. Although not as clear, Cluster-4 also showed a directional flow towards SS2 suggesting that these cells may transition to a profibrotic phenotype but have a cellular transcriptome that differentiates them from fibroblasts in Cluster-3 (Fig. 7C). TIMP1, COL1A1, CTHRC1, TUBA1B, SH3BGRL3 and S100A16 were the top six differentially expressed genes between Cluster-3 and Cluster-4 (upregulated in Cluster-4).

Cluster-2 fibroblasts showed a more complex pattern (Fig. 7C) with some cells going towards Cluster-1 (subgroup C2G1; Fig. 7D, gold) and others towards Cluster-3 (subgroup C2G2; Fig. 7D, purple) suggesting that cells in Cluster-2 were captured in a transition state that could either be resolved (SS1) or progress to a profibrotic state (SS2). We found 74 of the 103 genes coding for ribosomal proteins to be significantly downregulated in C2G2 vs C2G1 (Additional file 1: Table S5). The remaining 29 ribosomal protein genes were unaffected. Based on these 74 genes, C2G1 and C2G2 cells were assigned a ribosome score which was significantly decreased in C2G2 (p-value ≪ 0.001). In addition, we calculated a ribosome biogenesis score based on 267 other genes associated with this process (Gene Ontology Browser). Again, C2G2 showed a significant decrease for this score (p-value ≪ 0.001). Both these observations (Fig. 7E) are consistent with the hypothesis that there is a dysregulation of ribosomal function at the early stages of IPF.

To validate this finding, we used a second publicly available scRNA-seq dataset [6], that had similar tissue collection and processing procedures as ours. This dataset consisted of eight transplant donors and four IPF ex-plants, but it did not have distinct upper/lower lobe samples. We processed and reanalyzed the raw scRNA-seq data in the same way as our dataset. Due to the low number of IPF fibroblasts (see Additional file 2: Fig. S5B) we couldn’t confidently identify a subcluster similar to our Cluster-2 in the fibroblast cluster, but we observed three areas containing distinct cell populations originating from Controls, IPF and a mixture of the two, respectively (Additional file 2: Fig. S5A, green). Analysis of those cells showed 32 ribosomal genes to be differentially expressed between control and IPF with 29 of them downregulated. Notably, all 29 were also included in the previous list of 74 differentially expressed genes in our Cluster-2 (p-value < 0.01).

To identify early IPF biomarkers specific to fibroblasts, we examined the Cluster-3 top upregulated genes that also showed low expression in the rest of the cell types (Fig. 5). The five most significant of those included known IPF players (metalloprotease, collagen and two -LTBP1 LTBP2) and MOXD1 (Additional file 1: Table S3). MOXD1 was not previously associated with IPF, but it is expressed significantly higher in Cluster 3 and it is significantly downregulated in controls (Fig. 8A–C). Furthermore, it had been described in replicative senescent fibroblasts [30]. Given that IPF fibroblasts can be resistant to apoptosis [31] and have a senescent phenotype [32], we selected MOXD1 for validation as a new possible fibroblast biomarker in early and advanced IPF.

Immunofluorescence and immunohistochemistry staining of human lung tissues (Fig. 8D, Additional file 2: Figs. S6 and S7) validated the mRNA findings. Cells from healthy donors’ tissues expressed minimal MOXD1 protein. IPF lung upper lobes, showing less advanced fibrotic disease, also showed increased MOXD1 compared to control lungs. IPF lower lobes, showing advanced fibrotic disease, expressed even higher levels of MOXD1. We then proceed to characterize the fibroblast cells expressing MOXD1 by co-staining the sections with PDGFRa (marker of fibroblasts) and alpha smooth muscle actin (α-SMA, a marker of myofibroblasts, which also marks pericytes and smooth muscle cells). As expected, α-SMA staining was increased in upper lobes and even more highly increased in lower lobes of IPF lungs compared to control lungs. Most MOXD1 staining in upper lobes was found in cells not staining with α-SMA, while these markers co-stained many of the same cells in the lower lobe suggesting that MOXD1 is a marker of cells that are maturing into myofibroblasts. This is consistent with increased levels of myofibroblasts in IPF. PDGFRa staining showed that fibroblasts in control lungs are mostly MOXD1/SMA negative, while fibroblasts in lower lobes of IPF lungs are mostly MOXD1/SMA positive, co-staining with these markers indicated that most of the MOXD1 staining cells in IPF are myofibroblasts. To assess location of MOXD1 fibroblasts, we performed immunostaining of whole lung tissues. Lung sections from age matched donor control were negative for MOXD1, whereas sections from IPF lungs showed numerous positive cells (Fig. 8D). Immunohistochemistry staining also showed increased MOXD1 expression in fibroblast foci (Additional file 2: Fig. S7).

Velocity analysis of fibroblasts using their spliced (Fig. 8E, left) and unspliced (Fig. 8E, center) MOXD1 mRNA context indicated transcriptional kinetics towards the intersection between Cluster-3 and Cluster-4. A closer inspection of this region showed high expression of the myofibroblast associated genes POSTN, ASPN, ACTA2 and TAGLN (Fig. 8G) [11, 33]. It was clear from the UMAP plots that the cells carrying these myofibroblast markers did not segregate in one distinct cluster (Fig. 7C) suggesting that myofibroblast specific genes are having similar signatures to the original cells that transitioned to a myofibroblast.

MOXD1 is a copper-binding protein and copper itself has been implicated in pulmonary fibrosis [34]. Blockhuys et al. [23] have identified 54 human Cu-binding proteins, divided in 9 groups according to their cellular localization. We found 39 of them to be significantly up- and 10 down-regulated in Cluster-3 (“fibrotic”) vs Cluster-1 (“control”) (FDR < 0.05; Additional file 1: Table S6A). Furthermore, 20 of them were already significantly upregulated in C2G2 compared to C2G1 subcluster (Additional file 1: Table S6B). Among the top differentially expressed genes we found SPARC, LOXL1, and APP which have been implicated in lung fibrosis [35‐37]. In similar analysis to ribosome genes, we found 29 genes to be significantly upregulated in the validation cohort; and 24 of them were included in the 39 of Cluster-3. Next, we assigned a copper-binding score to all fibroblasts based on the expression of these proteins in our dataset. IPF samples, regardless of lung topology, showed higher scores compared to the control samples (Fig. 8F, left). This difference was also reflected in the fibroblast clusters where Cluster-1 cells had the lowest Cu-binding score and Cluster-3 and 4 had the highest. Cluster-2, consistent with an intermediate cell state, showed an intermediate score (Fig. 8F, right).