Recent single-cell RNA-sequencing (scRNA-seq) techniques have corroborated a previously unappreciated cancer-associated fibroblast (CAF) heterogeneity in multiple cancer types.¹ Among several CAF subtypes, myofibroblastic CAFs (myCAFs) and inflammatory CAFs (iCAFs) are consistently observed across various cancer types and constitute major CAF populations.² To date, it has been regarded that certain CAF subtypes are shaped in response to specific cancer-driven signals and CAFs can transform from one state to another.^{3, 4} However, the mechanisms governing the phenotypic shift from myCAFs to iCAFs (or vice versa) remain largely unknown.

To uncover the hidden mechanism responsible for CAF plasticity, first, we compared the differentially expressed transcription factors (TFs) in CAFs with normal tissue fibroblasts (NFs) using seven large-scale scRNA-seq datasets (Figure 1A and Figure S1A). We found that only HIF1A (hypoxia-inducible factor 1-alpha) and PRRX1 were commonly activated in CAFs (Figure 1B-C). Our recent discovery of PRRX1 as a master TF that determines myCAF led us to think that HIF1A could be a key factor in establishing iCAF phenotype.⁵ Higher HIF1A expression was common in CAFs than in NFs (Figure 1D and Figure S1B). Trajectory analysis indicated the progressive differentiation of NFs to CAFs, and HIF1A expression gradually elevated with increasing pseudotime (Figure 1E and Figure S1C-D). The proportion of HIF1A-expressing CAFs was greatly increased compared to that of NFs in each organ, but not all CAFs expressed HIF1A (Figure 1F), suggesting the heterogeneous nature of CAFs.

From scRNA-seq data, we found two distinct CAF clusters: one is myCAF while the other is iCAF (Figure 2A). As HIF1A is a master transcriptional regulator of the cellular response to hypoxia, we explored the hypoxia pathway signatures in these clusters. Gene set enrichment analysis (GSEA) demonstrated hypoxia pathway enrichment in iCAFs from pancreatic cancer (Figure 2B), as well as in iCAFs from multiple cancer types (Figure 2C and Figure S2A-B). Subsequent gene set variation analysis (GSVA) highlighted the striking contrast between the iCAF and myCAF signatures (Figure 2D). Representatively, iCAFs were enriched in pathways including TNFα signaling, hypoxia, apoptosis, and inflammatory response, as previously indicated.² Through extensive literature review, we found that most of these pathways were related to hypoxia, mechanistically (Figure S2C).

Next, we determined whether hypoxia shapes iCAFs in vitro. Using five lines of pancreatic CAFs cultured under normoxia or hypoxia (Figure 2E and Figure S3A), we measured the expression of iCAF- or myCAF-related genes (Table S1). In general, the level of cytokines was elevated whereas myCAF-related genes were commonly repressed in hypoxic CAFs (Figure 2F and Figure S3B). Thus, hypoxic CAFs had significantly higher iCAF-associated but lower myCAF-associated GSEA scores (Figure 2G). Immunofluorescence (IF) study revealed that CAFs cultured under normoxia exhibited myCAF features expressing high αSMA but no cytoplastic IL6. Whereas, in hypoxic CAF, the expression of αSMA was markedly reduced, but the expression of IL6 was remarkably increased (Figure 2H). This phenotypic alteration was dependent on HIF1A, as HIF1A inhibitor (KC7F2) reversed the increased expression of iCAF-related genes under hypoxia (Figure 2I and Figure S3C). Furthermore, myCAF expresses high levels of contractile cytoskeletons resulting in extracellular matrix (ECM) remodeling,⁵ and we observed significant gel contraction in CAFs cultured under normoxia. In contrast, hypoxic CAFs lost the ECM-contracting capability (Figure 2J). Collectively, these observations indicate that CAFs lose myCAF features but gain iCAF features under hypoxia.

Then, we performed scRNA-seq using two CAF lines cultured under normoxia or hypoxia (Figure 3A), which demonstrated the separation of hypoxic CAF clusters (Figure 3B) with enrichment of hypoxia pathway (Figure 3C). For further characterization, we performed GSVA on those CAFs, separately (Figure 3D and Figure S4-5). This almost perfectly mirrored the up- or down-regulated signaling pathways identified in iCAFs of human tissue (Figure 3E). Significantly higher iCAF but lower myCAF signature scores were also noted in hypoxic CAFs (Figure 3F).

For human tissue validation, we examined whether iCAFs preferentially reside in hypoxic regions. To this end, we performed IF on pancreatic cancer tissue and defined iCAFs as IL8-αSMA double-positive spindle-shaped cells. Since the relative αSMA expression in iCAFs is low but still well detected in tissue, we used αSMA to detect both CAF subtypes. As hypoxia typically arises in solid tumours at a distance of approximately 100 μm from a functional blood vessel (Figure S6),⁶ we counted the number of iCAFs in 50-μm incremental regions from the vessel, with the aid of StrataQuest software (TissueGnostics). As depicted in Figure 4A, iCAFs were scarce in the vicinity of a vessel whereas they were frequently identified in regions distant from the vessel. In the region of interest #3 (ROI#3), the number of iCAFs gradually increased according to the distance from the vessel (Figure 4B), and the same trends were noted in other ROIs (Figure 4C). Finally, we confirmed that the number of iCAFs was significantly increased in distant regions in comparison to the region ≤100 μm from the vessel (Figure 4D), indicating that iCAFs are mainly shaped by the hypoxic TME (Figure 4E).

To date, it has been regarded that cancer-secreted IL1 is a key factor in generating iCAF.⁴ However, this specific ligand-receptor interaction cannot explain the presence of iCAFs in IL1-negative tumours and the various characteristic signaling pathways of iCAFs. Rather, these are easily explained by ‘hypoxia’ because IL1 can be transcriptionally up-regulated in hypoxic cells⁷ and HIF1A is a master regulator in establishing iCAF signature as we demonstrated. Moreover, our concept supports the previously suggested premise that iCAFs and myCAFs can reverse from one cell state to another,³ as hypoxic regions may dynamically change within tumours according to tumour growth and angiogenesis. In line with our discovery, it was quite recently demonstrated that hypoxia promotes iCAF phenotype in a HIF1α-dependent manner in a mouse model.⁸ Our results provide additional evidence that this proof of concept is also valid in various human cancers including pancreatic cancer.

In conclusion, our comprehensive analyses indicate that the hypoxic TME is the main driver of the transformation of CAFs into an inflammatory phenotype. Our discovery provides novel insights into CAF biology and strategies for CAF-directed therapies.