Historically, IHC was the mainstay of CAF functional analyses. However, more recently studies have used transcriptomic analysis to elucidate the heterogeneity of the CAF TME compartment (10, 48, 53, 58, 63–69). Lambrechts and colleagues used single cell RNA (scRNA) sequencing (scRNA-seq) to characterize the stromal compartment in the lung cancer TME. The authors described 5 distinct types of fibroblasts, with a unique collagen and ECM profile for each subclass. Interestingly, they observed that fibroblast cluster 6, described as ‘normal fibroblasts’ and characterized by high elastin levels and low levels of collagens I, III, V, and VI, was significantly correlated to favorable outcome in patients with lung adenocarcinoma (63). Furthermore, gene enrichment analysis revealed that cluster 6 fibroblasts showed upregulation of inflammatory response pathways, as opposed to other clusters such as cluster 7, a subpopulation shown to be substantially associated with poor outcome. Yet, presence of cluster 6 fibroblasts was also significantly associated with poor outcome in patients with lung SCC (63). As a result, the authors suggested that the abundance and unique functions of a given CAF cluster could differ between tumor types, with consequently different associated prognosis and therapeutic response (20, 63). In another study by Neuzillet and colleagues, patient-derived PDAC CAFs were profiled for 770 genes of the PanCancer Progression panel, which revealed 4 distinct CAF clusters that overlapped with some of the CAF clusters defined by Lambrechts and colleagues (Fig. 3). A POSTN-positive subpopulation, overlapping with clusters 1 and 5 described by Lambrechts and colleagues, had the least proliferative effect on cancer cells and showed lower chemoprotection to cancer cells as compared with other CAF subpopulations in an in vitro coculture set up. However, the same POSTN-positive subpopulation was associated with poor prognosis in patients (58). In addition, as observed by the authors, POSTN is strongly expressed at the invasive front of the tumors while being involved in tumor capsule formation and metastatic niche preparation (70). Considering the low protumoral effect and association with poor prognosis, it is likely that POSTN+ CAFs arise as a host response mechanism to restrain tumor growth (5, 71).

Using a melanoma mouse model, Davidson and colleagues also performed scRNA-seq and identified three subpopulations, namely S1 or ‘immune’, S2 or ‘desmoplastic’, and S3 or ‘contractile’ stromal cells (Fig. 3). The S1 subpopulation was shown to overexpress PDPN, PDGF-Rα, and CD34 while S3 showed high levels of α-SMA. S2 was identified as having intermediate expression of PDPN and PDGF-Rα but low expression of α-SMA and CD34. Further analysis revealed that each subpopulation had its specific functional purpose with S1 ‘immune’ subpopulation involved in the recruitment of immune cells through proinflammatory cytokines like SDF-1 or CSF-1 and receptors such as IL6ra and IL6st. S2 ‘desmoplastic’ subpopulation was thought to promote the desmoplastic reaction via overexpression of genes coding for ECM components like collagens (Col1a1, Col1a2, Col6a2), Postn, and Tnc. The S3 ‘contractile’ subpopulation overexpressed genes involved in the regulation and rearrangement of the actin cytoskeleton. S3 stromal cells were reported to be the most proliferative among all subpopulations (59). Similarly, through scRNA-seq, Sebastian and colleagues described three main subpopulations of CAFs in both breast and pancreatic cancers, namely ‘myofibroblastic’ CAFs enriched for α-SMA and other contractile proteins; ‘inflammatory’ CAFs showing overexpression of cytokines involved in inflammation and an MHC class II–expressing CAF subpopulation (7).

scRNA-seq by Bartoschek and colleagues in a mouse model of breast cancer identified ‘matrix CAF’ and ‘cycling CAF’ populations (66). Analysis of single cell transcriptomes in human colorectal tumors by Li and colleagues identified two CAF subpopulations, ‘CAF A’ cells expressing genes related to ECM remodeling while ‘CAF B’ showed ‘myofibroblastic’ features (65). As an attempt to decipher the stromal microenvironment in cutaneous melanoma spheroids through scRNA-seq, Novotny and colleagues recently described an ‘ECM-’ CAF cluster with a proinflammatory profile, an ‘ECM+’ cluster, enriched with ECM markers such as COL1A1, and a ‘ID+’ cluster, which was characterized by overexpression of factors involved in the TGF-β pathway (72).

Despite the expected differences between the different transcriptomic analyses, a common pattern has emerged from the data published describing the presence of different CAF subpopulations (Fig. 3). For instance, both the ‘inflammatory’ and the ‘Cd74-high’ subpopulations characterized by Sebastian and colleagues respectively resemble, at the molecular level, the ‘desmoplastic’ and ‘immune’ subpopulations reported by Davidson and colleagues. In the same way, the ‘CAF A’ subpopulation described by Bartoschek and colleagues and the ‘ECM+’ subpopulation reported by Novotny and colleagues also showed ‘desmoplastic’ features. Also characterized by Novotny and colleagues, the ‘ECM−’ subpopulation showed ‘immune’ characteristics. In addition, a CAF cluster (Cluster 2) defined as ‘dividing/cycling’ in triple-negative breast cancer (TNBC) seems to mirror the ‘contractile’ subpopulation of CAFs linked to melanoma (7, 59).

Accordingly, we suggest that there are 4 broad categories of CAFs described in the literature: ‘immune’, ‘desmoplastic’, ‘contractile’, and ‘aggressive’ (Fig. 3). We believe this nomenclature can help clarify and simplify comparisons between studies until a more comprehensive meta-analysis report is available, which will allow the clustering of CAFs across different cancer types to identify subpopulations and components within each subpopulation. Potential markers for the ‘immune’ subpopulation are C3, PDPN, and ENG. The ‘desmoplastic’ subpopulation can be discriminated from other CAFs through expression of markers like POSTN as well as ECM components, such as elastin and collagen (type 1 and 4). The ‘contractile’ subpopulation is defined by expression of factors involved in actin cytoskeleton rearrangement and/or cell cycle regulation. High expression of markers associated with EMT, such as vimentin or VEGF-A, or the TGF-β pathways characterizes ‘aggressive’ CAFs. Both ‘contractile’ and ‘aggressive’ subpopulations show the highest expression of α-SMA and are most often associated with poor patient survival/outcome.

Interestingly, through coculturing fibroblasts with cancer cells, Neuzillet and colleagues noticed that pancreatic stellate cells could evolve from the least protumoral POSTN+ CAF subtype (subtype A) to fully supportive CAF phenotypes [subtypes B and C, characterized by myosin heavy chain 11 (MYH11) and PDPN expression, respectively; ref. 58]. Similarly, by performing scRNA-seq at various points during tumor progression in a mouse model of PDAC, Dominguez and colleagues identified how two ‘normal fibroblast’ clusters gave rise to two distinct CAF lineages in a mouse model of PDAC (Fig. 3; ref. 10). Most importantly, IL1 and TGF-β–dependent pathways were identified to underlie the reprogramming of each lineage which eventually were observed to take over the PDAC stromal compartment. Similar IL1– and TGF-β–driven CAF subpopulations were described in patients with PDAC, although compared with the mouse model both of these lineages appeared to arise from a single early CAF subpopulation (10).

Early ‘CAF states’ in different cancer types appear to share immune/inflammatory features. These characteristics could be linked with tumor restraining properties of some CAF subpopulations during the early stages of cancer development. In a recent study, Chen and colleagues reported that a CAF subpopulation defined as ‘complement-secreting CAFs’ (csCAF) showed high expression of complement system components, such as C3 or C7, involved in regulating immune/inflammatory response. These csCAFs, which could be described as ‘immune’ CAFs, were present exclusively in early-stage PDAC and were described as potential tumor-inhibitory CAFs (73). In pancreatic tumors, transcriptional profiling of serum amyloid a3 (Saa3)-KO CAFs with tumor-inhibitory capabilities revealed an overall downregulation of cytokine expression compared with Saa3+ CAFs. These (Saa3)-KO CAFs did, however, express proinflammatory markers such as TNFα and genes in the IL6 pathway (40). Similarly, prostate cancer–associated CAFs with tumor-inhibitory capabilities have been shown to upregulate proinflammatory genes such as IL6 and CXCL2 (39).

Similar to the data reported by Dominguez and colleagues in PDAC, the distribution of stromal cell subpopulations in the murine melanoma TME described by Davidson and colleagues was observed to change upon tumor progression. This evolution progressively leads to the prevalence of the highly tumor-supportive ‘contractile’ CAF subpopulation at late stages of tumor development, as opposed to the ‘immune’ and ‘desmoplastic’ subpopulations present at earlier stages (Fig. 3; refs. 7, 59). Sebastian and colleagues also suggested that the ‘desmoplastic’ CAFs they described in murine models of triple-negative breast and pancreatic cancers may evolve and differentiate into ‘aggressive’ CAFs as the tumor progresses (Fig. 3; ref. 7). Taken together, these data suggest that CAF evolution, towards an increasingly tumor-supportive phenotype, is a common feature of many cancers during tumor progression and warrants further investigation (Fig. 4).

While the CAF compartment becomes more supportive as the tumor progresses, CAF heterogeneity appears to decrease. Support for this comes from Venning and colleagues who recently followed the evolution of CAF subpopulations in murine models of TNBC (74). Tumors were collected at day 7, 14, and 21 post injection, from which CAFs were isolated. Based on the expression of 6 markers (α-SMA, FAP, PDGF-Rα and β, CD26, and PDPN), authors defined up to 63 CAF subpopulations present in the TME of both TNBC models at day 7. Of these, 5 became more prevalent by day 21 while other CAF subpopulations decreased in abundance. Eventually, the 5 main subpopulations contributed to more than 60% of the CAF compartment in the TME of both models (74). These data suggest that the CAF compartment becomes more homogenous in its cellular composition over time, albeit the time frame studied was fairly short (74). Similar observations were made by Chen and colleagues in PDAC patient samples where late-stage tumors show only one subpopulation of CAFs, lacking other subpopulations described in early-stage tumors, including ‘immune’ CAFs (Fig. 3; ref. 73).

Overall, among the 4 subpopulations of CAFs we defined, we believe that the ‘immune’ and ‘desmoplastic’ subpopulations are the first ones to emerge as a host response to tumor growth and seem to be the most likely to have tumor inhibitory functions (73). On the other hand, the ‘contractile’ and ‘aggressive’ subpopulations show stronger tumor supporting capabilities (7, 59). In addition, the ‘immune’ and ‘desmoplastic’ subpopulations slowly disappear over time while the ‘contractile’ and ‘aggressive’ subtypes become more prevalent (7, 59, 66, 73, 75). Taken together, it can be stipulated that the tumor-inhibitory CAF/tumor-supporting CAF ratio decreases over time as the CAF compartment eventually evolves into a more homogenous and tumor-supportive environment (Fig. 4; refs. 10, 43, 58, 65, 66, 68, 74).