As an important component of tumor microenvironment, CAFs are described as activated fibroblasts located in the vicinity of cancer cells without the phenotype of epithelial, cancerous, endothelial, and immune cells (1). They are elongated and spindle-shaped in morphology and have some positive markers, such as alpha-smooth muscle actin (α-SMA), fibroblast activation protein (FAP) and fibroblast specific protein 1 (FSP-1) (2). CAFs have merged as the hot-spot of cancer study; however, their phenotypic and functional heterogeneity hinders the clinical application (3). Studies have shown that CAFs could secrete a variety of chemokines, cytokines, and growth factors to facilitate tumor growth, chemotherapy resistance and immunosuppression (4). On the contrary, some studied have reported the tumor-suppressive function of CAFs in certain tumor models (5). This review summarized the heterogeneity of biological origins, phenotypic markers, and biological functions of CAFs, as well as uncovered how their heterogeneity made identification, subtypes classification and clinical therapy challenging. Our review provided a new perspective for CAF research and personalized therapy.

Increasing evidence suggest that CAFs have different cellular origins. Though precise lineage tracing study has shown the origin of fibroblasts in healthy or injured tissues, the origins and specific activation processes of CAFs are still lacking (6, 7). Several cells may be predecessors of CAFs, such as normal fibroblasts (8), mesenchymal stem cells (MSCs) (9), pancreatic stellate cells (PSCs) (10), epithelial cells (11), endothelial cells (12), adipocytes (13), pericytes (14), hematopoietic stem cells (15) and cancer stem cells (CSCs) (16). The changes in the microenvironment where these precursor cells exist in may be a primary inducer of CAF transition (3).

As the major source of CAFs, normal fibroblasts can transform to CAFs by cytokines secreted by stromal or tumor cells. Transforming growth factor-β (TGF-β) can induce the CAF phenotype through SMAD-dependent or independent pathway (17). For example, bladder cancer cells released exosomes contain TGF-β, leading to the activation of SMAD-dependent signaling and the stimulation of normal fibroblasts to CAFs (8). Platelet-derived growth factor-D (PDGF-D) secreted by cholangiocarcinoma cells could stimulate surrounding fibroblasts to produce VEGF-C and VEGF-A, resulting in the expansion of lymphatic vasculature and tumor cell intravasation (18). In addition to cytokines, non-coding RNAs from cancer cells can also induce the conversion of resident fibroblasts to CAFs. Exosomes derived from hepatocellular carcinoma cells were rich in miR-1247-3p, which activated β1-integrin-NF-κB signaling through targeting B4GALT3 in fibroblasts (19). In lung adenocarcinoma, miR-200 deficiency in cancer cells promoted the expression of Jagged1/2 and the activation of Notch in adjacent CAFs, which reprogrammed CAFs from a quiescent state into an active pro-tumorigenic state (20). Additionally, the hypoxia microenvironment also contributes to the activation of resident fibroblasts. Hypoxia was related to the accumulation of ROS, the activation of the HIF-1α signaling pathway in hepatocellular carcinoma cells, and the enhanced expression of FAP in surrounding fibroblasts (21).

MSCs are another important source of CAFs. The transformational potential of MSCs into CAFs was first proved in breast cancer (9). TGF-β secreted by cancer cells recruited MSCs and maintained the differentiation of MSCs into CAFs (22). In colorectal cancer, the high level of stromal cell-derived factor-1 (SDF-1) upregulated the expression of chemokine receptor 4 (CXCR4) and TGF-β in MSCs, leading to the transformation of MSCs (23). In epithelial ovarian cancer, the elevated expression of STAT4 in epithelial cells induced MSCs derived from adipose and bone marrow to obtain CAF-like features, which in turn promoted EMT and peritoneal metastasis of ovarian cancer by secreting CXCL12, IL-6 and VEGF-A (24). In addition to the stimulation of cancer cells, changes in tumor microenvironment like pH can also stimulate the transformation of MSCs. PH induced activation of MSCs to CAFs was decreased by upregulating the expression of proton-sensing G-protein-coupled receptor68 (GPCR68) and activating downstream effector-Yes-associated protein (YAP) in MSCs (25).

The other cellular origins of CAFs have been reported. For example, PSCs could transform to CAFs in pancreatic cancer (10). In pancreatic ductal adenocarcinoma, the IL-1 signaling cascade led to JAK/STAT activation and induced an inflammatory CAF state (26). Epithelial or endothelial cells are found to be the probable origins of CAFs through epithelial-to-mesenchymal transition (EMT) or endothelial-to-mesenchymal transition (EndMT). The human nasal epithelial cells were activated and displayed CAF phenotypes such as FSP or FAP through EMT when they were exposed to matrix metalloproteinase (MMP)-9 (11). TGF-β could induce proliferating endothelial cells into fibroblast-like cells (12). In addition, a recent study reported that tumor cells induced adipocytes to CAFs by activating Wnt/β-catenin signaling in ovarian cancer (13). Cancer cells, especially cancer stem cells, have also been demonstrated to be a source of CAFs through the action of TGF-β (15). Besides these sources of CAFs mentioned above, there also exist some uncommon origins, such as pericytes, hematopoietic stem cells, which needs further exploration (14, 16).

In brief, the activation of CAFs is mainly regulated by different cytokines and signaling pathways of cancer niche ( Figure 1 ). Although the origins of CAFs in solid tumors were not fully elucidated, using lineage tracing technologies to track CAF transition may provide a solution in the future.

The altered protein profiles can be used to identify or isolate CAFs. According to the distinct phenotypic markers, CAFs can be divided into several subpopulations and some of them partially overlap. In this part, we will present the phenotypic differences and subtype classification of CAFs, and provide some suggestions for identifying different CAF populations.

There are several typical CAF markers, such as FAP, α-SMA, FSP-1, PDGFR-α, PDGFR-β, and Thy-1 (27). Despite the diversity of biomarkers, the isolation of CAFs from cells remains a challenge due to low specificity. For example, α-SMA and FAP were highly presented in pericytes, lymphatic endothelial cells and fibroblast reticular cells. Similarly, vimentin was present in endothelial cells, smooth cells and tumor cells (28). Additionally, with the continuous optimization of detection technology, the researchers identified uncommon PSC-derived CAF subsets in pancreatic ductal adenocarcinoma tissues. These CAFs located away from cancer cells, lacked elevated α-SMA expression, and secreted IL-6 and other inflammatory mediators (10). The results highlighted the importance of considering multiple indicators in CAF identification. In addition to classical phenotypic markers, some new ones are studied in recent years. In pancreatic cancer, the high expression of caveolin-1 (Cav-1) in CAFs was associated with the invasiveness of cancer cells and poor prognosis of patients (29). The same results were further proved in lung adenocarcinoma (30). Similarly, a recent study reported that the melanoma cell adhesion molecule+ (MCAM+) CAFs induced by TGF-β in colorectal cancer patients were associated with poor prognosis (31). Another study concluded that focal adhesion kinase (FAK) activity in CAFs was increased in PDAC tissues compared with healthy ones and the FAK+ CAFs could be an independent prognostic marker (32).

Based on surface markers, CAFs are classified into different subtypes that display distinctive secretory phenotypes and perform specific biological functions in dynamic tumor environment, as summarized in Table 1 (33). In a mouse model of pancreatic ductal carcinoma, the researchers demonstrated the existence of myofibroblastic CAFs (myCAFs), inflammatory CAFs (iCAFs) and antigen-presenting CAFs (apCAFs) by single-cell RNA sequencing. MyCAFs were characterized by the expression of α-SMA, TAGLN, MYL9, TPM1, TPM2, MMP11, POSTN and HOPX, which could promote the proliferation, invasion and metastasis of tumor cells. ICAFs could promote metastasis and angiogenesis by producing inflammatory cytokines and chemokines such as IL-6, IL-8, CXCL1, CXCL2, CCL2, CXCL12 and Ly6c. ApCAFs had immunomodulatory capacity in pancreatic ductal adenocarcinoma. They expressed MHC II, Saa3, Slp and could activate CD4+ T cells in an antigen-specific manner in the model system (34). Another study reported four CAF subtypes in pancreatic ductal adenocarcinoma based on transcriptomic analysis. These four subgroups, named A-D, could be distinguished by differential expression of three markers, periostin (POSTN), myosin‐11 (MYH11) and podoplanin (PDPN). Patients with the dominant subtype-C had prolonged survival, whereas those with the dominant subtype D had the worst prognosis, suggesting that specific tumor-stromal interactions are associated with adverse outcomes (35). Furthermore, a novel subtype of CAFs with a highly activated metabolic state (meCAFs) was identified in PDAC. MeCAFs had highly activated glycolysis, and patients with abundant meCAFs had a higher risk of metastasis and poor prognosis, but showed a dramatically better response to immunotherapy (36).

In human breast cancer, four CAF subgroups, known as S1-S4, have been identified by flow cytometry, immunohistochemistry, and RNA sequencing. They can be distinguished according to the expression of FAP, CD29, αSMA, PDPN and PDGFRβ. CAF-S1 stimulated cancer cell migration and mediated EMT transition through the activation of CXCL12 and TGF-β. CAF-S4 induced cancer invasion through NOTCH signaling. The study also found that patients with high levels of CAF-S4 in lymph nodes were prone to late distant metastases, which could be a potential prognostic marker for breast cancer (37). Furthermore, two new cell surface molecules, CD10 and GPR77, can define a CAF subset associated with chemoresistance and low survival in patients with breast cancer and lung cancer. CD10+ GPR77+ CAFs accelerated cancer progression by providing a survival niche for cancer stem cells, and the functional CAF subset could be specifically recognized and isolated, suggesting an effective therapeutic strategy for CSC-driven solid tumors (38). Bartoschek and colleagues defined four spatially and functionally distinct CAF subpopulations through single-cell RNA sequencing in breast cancer. According to different functions, these subgroups were named as vascular CAFs (vCAFs), matrix CAFs (mCAFs), cycling CAFs (cCAFs) and developmental CAFs (dCAFs). VCAFs originated from perivascular location, expressed genes controlling angiogenesis, and invaded tumor stroma during tumor progression. MCAFs were offspring of resident fibroblasts and regulated the tumor immune response. CCAFs were proliferative fragment of vCAFs and had different expression of cell cycle genes. DCAFs underwent EMT and shared expression patterns with tumor epithelium. Thus, the phenotypic and functional heterogeneity of CAFs can be attributed to their different origins (39).

In oral carcinoma, CAFs were grouped into two distinct clusters based on the expression difference of α-SMA. CAF-C1 had low α-SMA-scores and was more supportive for cell proliferation but suppressive for the growth of stem-like cancer cells (SLCCs). BMP4 played a determinant role in C1-type CAF-mediated suppression of SLCCs. However, CAF-C2 had the opposite effects on tumor cells (40). In gastric cancer, the researchers identified a new CAF subset defined as extracellular matrix CAFs (eCAFs). The subset had high expression of POSTN, which could support the adhesion and migration of epithelial cells, as well as be a prognostic marker for gastric cancer (41). In colorectal cancer, two distinct CAF subtypes, named CAF-A and CAF-B, were identified depending on their differential expressions. CAF-A expressed markers related to extracellular matrix remodeling, such as Matrix metalloproteinase-2 (MMP2), decorin (DCN) and collagen 1A2 (COL1A2). CAF-B cells expressed markers of myofibroblasts such as actin alpha 2 (ACTA2), transgelin (TAGLN) and platelet-derived growth factor A (PDGFA) (42). Hu and colleagues identified three subtypes of CAFs in non-small lung cancer. Subtype I highly expressed hepatocyte growth factor (HGF) and fibroblast growth factor 7 (FGF7), and had strong protective effects against cancer. Subtype II expressed FGF7 and had moderate protection against cancer. Subtype III had minimal protection (43). In hepatocellular carcinoma, CAFs isolated from fresh tumor tissues could be divided into activated myofibroblast phenotype and a mesenchymal stromal cell phenotype. They could enhance the stemness of cancer cells and modulate immunosuppression, respectively (44). In high-grade serous ovarian cancer, the CD49e+ CAF population was divided into two subgroups, FAP-high and FAP-low group. The FAP-high subgroup could regulate cancer invasion and immunomodulation, whereas the FAP-low group could regulate glucose homeostasis and lipid metabolism (45).

In summary, CAFs can be divided into several specific subpopulations in different tumor models based on surface markers and protein profiles ( Table 1 ). These studies suggest that CAFs are a cell state rather than end-point of differentiation. Because the subtypes are dynamic, they can be mutually transformed under the influence of cancer status and drug treatment. For example, when CAFs are isolated from cancer tissues and cultured in vitro, CAF subpopulations may change their phenotype. Furthermore, the transition could also occur in different tumor types, even in different parts of the same tissue, so more advanced detection techniques and strategies are needed to further identification. Single-cell RNA sequencing is a cutting-edge technology that can investigate the transcriptome and related markers of individual cells, which may help to more accurately classify CAF subtypes in further studies (46). Additionally, new technologies such as mass spectrometry-based time-of-flight flow cytometry (CyTOF) (47), multiple flow cytometry (48) and multiple immunostaining (49) are helpful to identify CAF subtypes. During the detection of CAF subtypes, it is necessary to guarantee the number of patients to ensure the production of several cell subsets. Second, fresh samples are crucial in the current single-cell RNA sequencing strategy. Finally, batch effects may be involved between batch loaded samples (50). In addition to the inclusion of more molecular markers, the different functions, different positions in cancer tissues, and even different tumor stages of CAFs should also be considered to achieve a more detailed classification of CAFs.

CAFs play a vital role in cancer occurrence and development by regulating the proliferation, invasion and chemoresistance of tumor cells. The abundance in tumor microenvironment and the diverse tumor-supportive roles of CAFs make them an ideal therapeutic target (77). The recent advances in cancer therapy by targeting CAFs were summarized in Table 2 and Figure 3 .

Since the concept of CAFs was proposed in the early 1990s, CAFs have attracted extensive attention in cancer biology. Previous studies have led to a better understanding of the heterogeneity of CAF origins, phenotypes and functions. CAFs are the main cell types in tumor microenvironment which affect the occurrence, and development of cancer cells. They have rich cellular sources and precursor cells such as normal fibroblasts and mesenchymal stem cells have been shown to be the major sources. CAFs are not a cell type but heterogeneous functional subpopulations. Based on the surface markers, CAFs are divided into several subtypes, which have different biological functions. CAF subtypes identified in different cancer types may play opposite roles in cancer progression, such as tumor-promoting and tumor-suppressive functions. CAFs have great potential in clinical applications. Several preclinical studies and ongoing clinical trials have shown that strategies targeting CAFs are possible in cancer therapy. However, there are still some challenges in translating CAF research into clinical benefit. First, the concrete origins of CAFs in specific cancer types remains elusive. In addition, most studies on the origin of CAFs have been performed in vitro and lack appropriate clinical validation. The use of lineage tracing methods will greatly solve these problems in future studies. Second, the lack of uniform nomenclature for CAF subpopulations in different cancer types makes it difficult to compare CAF subgroups in distinct tumors. It would be useful to name them by combining analysis of cell lineage, surface markers, functions and clinical relevance. Additionally, there is still a lack of curate classification of CAF subtypes. Advanced strategies, such as single-cell RNA sequencing, mass spectrometry-based time-of-flight flow cytometry (CyTOF), multiple flow cytometry and multiple immunostaining, may be helpful to accurately classify CAF subtypes. Finally, although many experiments targeting CAFs to improve cancer therapy have been conducted in preclinical models and clinical trials, most of them have failed to pass phase II clinical trials. It has not yet reached practical application. To overcome this limitation, more detailed experimental designs and more clinical samples are needed, and the combination of these CAF-targeting approaches with existing therapies may be beneficial. Overall, it is critical to accurately understand the underlying mechanisms of action between CAFs and tumor cells. It is also important to understand CAF-targeting therapies at the molecular, cellular, and systemic levels based on the interactions between CAFs and tumor cells, to find the most appropriate strategies and avoid adverse effects. In addition, tracing the origins of CAFs may be a key factor in achieving the clinical application of CAF-targeting strategies and avoiding side effects. With the resolution of these problems, CAF-derived therapies are expected to provide new support for clinical cancer therapy in the near future.

CW wrote the manuscript and designed the figures. JG, HG, and XXZ assisted in the manuscript writing and figures drawing. XZ and RJ revised the manuscript. All authors contributed to the article and approved the submitted version.

The present study was supported by Distinguished Young Scholar Project of Jiangsu Natural Science Foundation (BK20200043), the National Natural Science Foundation of China (Grant no. 81702429, 81672416, 81972310), Natural Science Foundation of the Jiangsu Province (Grant No. BK20170561), Zhenjiang Science & Technology Program (Grant No. SH2019051).

The authors would like to thank literature support from Jiangsu University library and the support from fundings.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.