Diversity and Biology of Cancer-Associated Fibroblasts

Abstract

Efforts to develop anti-cancer therapies have largely focused on targeting the epithelial compartment, despite the presence of non-neoplastic stromal components that substantially contribute to the progression of the tumor. Indeed, cancer cell survival, growth, migration, and even dormancy are influenced by the surrounding tumor microenvironment (TME). Within the TME, cancer-associated fibroblasts (CAFs) have been shown to play several roles in the development of a tumor. They secrete growth factors, inflammatory ligands, and extracellular matrix proteins that promote cancer cell proliferation, therapy resistance, and immune exclusion. However, recent work indicates that CAFs may also restrain tumor progression in some circumstances. In this review, we summarize the body of work on CAFs, with a particular focus on the most recent discoveries about fibroblast heterogeneity, plasticity, and functions. We also highlight the commonalities of fibroblasts present across different cancer types, and in normal and inflammatory states. Finally, we present the latest advances regarding therapeutic strategies targeting CAFs that are undergoing preclinical and clinical evaluation.

This review summarizes the current knowledge on cancer-associated fibroblasts (CAFs) and focuses on the recent discoveries of CAF molecular and functional heterogeneity in several malignancies. The discovery of CAF heterogeneity provides a potential explanation for previous seeming conflicting studies that targeted CAFs. Indeed, CAFs have been demonstrated to play multiple tumor-promoting roles, but also to potentially have tumor-restraining functions, which highlights the need to design subtype-specific therapies. The characterization of CAF heterogeneity will be pivotal for the development of effective combinatorial approaches for cancer treatment.

I. INTRODUCTION

Fibroblasts play key roles in disease, tissue homeostasis, cancer progression, inflammatory and fibrotic conditions, and wound healing processes. In cancer, a better understanding of the complex nature of cancer-associated fibroblasts (CAFs) could have prognostic and therapeutic value, especially in the instance that a specific CAF population is associated with a particular cancer subtype, which could help stratify patients and tailor therapies.

CAFs are a major component of the stroma and secrete growth factors, inflammatory ligands, and extracellular matrix (ECM) proteins that promote tumor proliferation, therapy resistance, and immune exclusion (129). For these reasons, CAFs have been historically considered tumor-promoting components. However, studies mainly focused on targeting the Hedgehog (Hh) signaling pathway, which is activated in CAFs, indicate that, in some circumstances, CAFs could also have tumor-restraining functions (221, 248). Previous work has revealed the presence of various stromal transcriptional signatures in human cancer specimens (17, 193, 200, 292), and preinvasive or invasive malignancies have been associated with different stromal signatures (257). However, lack of single-cell resolution or limited cell numbers precluded the identification of distinct CAF subtypes. The development of new co-culture models to study CAF biology and the implementation of single-cell RNA-sequencing (scRNA-seq) techniques have recently revealed a previously unappreciated level of CAF heterogeneity in a number of cancer types. Importantly, the discovery of CAF heterogeneity offers a potential explanation for the seeming controversy of CAFs playing both tumor-restraining and tumor-promoting functions. Although these recent studies provide insights into the nature of CAF heterogeneity, the extent of this heterogeneity, the roles of distinct CAF subtypes, and how to selectively target these subtypes remain unclear. As distinct CAF populations could play different roles in cancer progression, targeting them individually may lead to disparate outcomes. A deep characterization of different CAF subtypes will therefore be pivotal for the design of effective combinatorial approaches. Importantly, the emerging understanding that CAF heterogeneity is common in solid tumors indicates that discoveries made in one cancer type may more broadly impact the field of oncology.

This review aims to summarize historic milestones of the research on CAF biology, highlighting the most recent discoveries on CAF molecular and functional heterogeneity and new potential therapeutic strategies.

II. FIBROBLAST HETEROGENEITY IN NORMAL TISSUES

Fibroblast activation protein (FAP) is expressed in most CAFs (146) and normal fibroblasts from different sites (252), which indicates a common cell lineage. With the increasing understanding that CAFs are heterogeneous, an immediate question to address is whether this fibroblast heterogeneity is restricted to the tumor microenvironment (TME) or whether precursors to distinct CAF subtypes are present in nonmalignant states. Although several studies have demonstrated the molecular and functional heterogeneity of fibroblasts in normal tissues (137, 175, 232) and during development (33), few comparisons have been made between these populations and CAF subtypes (64, 111). Highlighting commonalities between fibroblast phenotypes across normal and malignant states will advance our knowledge of CAF biology.

Recently, two large scRNA-seq data sets of normal cells from several murine tissues have been published (104, 287). These resources begin to define transcriptional signatures of known cell types, identify new cell subsets, and enable the comparison of the same cell population across different normal tissues and malignancies. Importantly, the analysis of normal and developmental states may help identify the cell of origin of distinct CAF populations and reconstruct their lineage relationships and dynamics. In support of this, recent studies indicate the presence of distinct normal fibroblast precursors that likely contribute to the different CAF subtypes in pancreatic ductal adenocarcinoma (PDAC) (64, 111).

III. FIBROBLAST HETEROGENEITY IN WOUND HEALING AND INFLAMMATORY STATES

Studying fibroblast heterogeneity in the context of inflammation and wound healing is highly relevant, as cancers share many features with these conditions, and local wound healing and inflammation have been shown to promote tumor growth (77, 96, 97, 145). Therefore, fibroblasts in different conditions could share common phenotypes, signaling pathways and cell of origin. For example, a number of studies have demonstrated key roles of nuclear factor (NF)-κB and transforming growth factor (TGF)-β signaling pathways in fibroblasts in both inflammatory and malignant states (24, 79, 116, 142, 222). Additionally, ECM deposition occurs both during wound healing and in cancer (27, 74), highlighting the presence of functions common to normal activated fibroblasts and CAFs. However, a rigorous comparison of fibroblast phenotypes in inflammation and cancer is lacking.

Flow cytometry and single-cell resolution studies have started to define the phenotypic and functional heterogeneity of fibroblasts during skin tissue repair (98, 271) and in inflammatory conditions, such as pulmonary fibrosis and colitis (21, 71, 137, 253, 315, 316) (TABLE 1). In rheumatoid arthritis, different fibroblast subtypes have nonoverlapping functions and distinct markers, and mediate either tissue inflammation through chemokine/cytokine secretion or tissue damage through elevated expression of metalloproteases (54, 199). Additionally, studies comparing fibroblasts in normal and inflammatory or wound healing states have shown similarities and differences between these populations (137, 271, 316).

Table 1.

Fibroblast heterogeneity in wound healing and inflammatory states

State	Models/Methods	Fibroblast Subtypes	Fibroblast Subtype Defining Markers	Subtype Putative Functions	Additional Notes
Rheumatoid arthritis (54)	•Murine samples	•Immune effector fibroblasts (F1–F4)	•PDPN+ FAPα+ THY+ IL6+ LIF+	•Inflammatory	•Located in the synovial sub-lining
	•scRNA-seq by 10× genomics	•Destructive fibroblasts (F5)	•PDPN+ FAPα+ THY- MMP9+ RANKL+	•Tissue (bone and cartilage) damaging	•Located in the synovial lining
Rheumatoid arthritis (RA) and osteoarthritis (OA) (199)	•Human samples	•3 Subsets	•PDPN+ CDH11+ CD34- THY+ RANKL+	•Inflammatory, migratory, promote osteoclast differentiation	•Located in the synovial perivascular zone, proliferative, more abundant in RA
	•scRNA-seq by SMART-seq2		•PDPN+ CDH11+ CD34- THY-	•Promote osteoclast differentiation	•More abundant in OA
	•FACS followed by RNA-seq		•PDPN+ CDH11+ CD34+ IL6+ CXCL12+ CCL2+	•Migratory, promote monocyte recruitment	•Located in the synovial sub-lining, proliferative
Pulmonary fibrosis (316)	•Murine samples	•Myofibroblasts	•αSMA+ TAGLN+ MYH11+ HHIPHigh PDGFRαLow	•Contractile signature	•HH activation?
	•scRNA-seq by 10× genomics	•Col13a1 matrix fibroblasts	•αSMALow COL1A1High COL13A1High ITGA8+	•ECM producing
		>•Col14a1 matrix fibroblasts	•αSMALow COL1A1High COL14A1High Pi16+ MMP3+	•ECM producing
		•Lipofibroblasts	•ADRPHigh PPARGHigh FABP4+ CD9+ SLPI+	•Inflammatory	•M2 macrophage-like signature
		•Mesenchymal progenitors	•CD52+		•Proliferative (Mki67High Ccnb2High Cks2High)
		•Mesothelial cells	•UPK3B+ MSLN+ LRRN4+ NKAIN4+
		•PDGFRβHigh fibroblasts	•αSMA+ PDGFRβHigh NOTCH3+		•Pericytes?
Colitis (137)	•Human and murine samples	•Myofibroblasts	•αSMA+ MYH11+	•Contractile signature	•No increase in inflammation
	•scRNA-seq with Fluidgm C2 platform and by 10× genomics	•S1	•COL14A1High FN1+ CXCL12+ ZEB2+	•ECM producing	•High in nonfibrillar collagens, located in lamina propria
		•S2	•COL4A5High WNT5AHigh POSTNHigh CXCL12+ SOX6+	•Support epithelial stem cell maintenance	•High in sheet collagens, close to the epithelial monolayer
		•S3	•SMAD7+	•ECM organization	•Murine S3: mesothelial cell origin? (WT1+)
		•S4	•PDPN+ TNFSF14+ IL33+ CD74+ FDCSPHigh CD24High	•Inflammatory, collagen cross-linking	•Prevalent in ulcerative colitis compared with healthy colon, source of oxidative stress
Skin wound (271)	•Murine samples	•3 Myofibroblastic subtypes	•CD34+ CD29+ SCA1+ PDGFRα+ CD26High CD9+	•ECM producing, inflammatory, collagen cross-linking	•Activated by CD301b+ macrophage-derived PDGFC and IGF1, reduced in advanced age, adipocyte precursors (APs)
	•FACS followed by RNA-seq		•CD29High SCA1- PDGFRα- CD26Low		•Increase in wound beds, in the most superficial outer wound bed edge
			•CD29Low αSMALow SCA1- PDGFRα-

List of recent scRNA-seq and multi-color flow cytometric studies that show differences and commonalities of fibroblast populations in inflammatory and fibrotic conditions, such as colitis, arthritis, and pulmonary fibrosis. ECM, extracellular matrix; IGF1, insulin-like growth factor 1; PDGF, platelet-derived growth factor; αSMA, α-smooth muscle actin.

The development and analysis of in vivo models of inflammatory and fibrotic conditions, such as rheumatoid arthritis, inflammatory bowel disease (IBD), hepatitis, and pancreatitis (77, 93, 99, 115, 143, 156, 170, 263), have given insights into the heterogeneity and biology of fibroblasts in these contexts. The historical model to study pancreatitis employs treatment with the cholecystokinin analogue cerulein (311, 323), although dosage regimens vary substantially across laboratories depending on the desired effect. More recently, genetically engineered mouse models (GEMMs) have been established to induce pancreatitis by expressing the glycan carbohydrate antigen 19–9 (CA19–9) (77) or activating mutations in trypsinogen (93, 99, 115). For inducing IBD as well as liver or pulmonary fibrosis, administration of dextran sodium sulfate (DSS) (156), carbon tetrachloride (CCl₄) (263), or bleomycin (170) is the most commonly used mouse model, respectively.

IV. FIBROBLAST HETEROGENEITY IN CANCER

A. Models to Study CAFs

With the emerging understanding that CAFs are comprised of multiple subtypes, in vitro and in vivo models will facilitate the systematic characterization of CAF heterogeneity and biology (FIGURE 1).

FIGURE 1.

Models to study cancer-associated fibroblast (CAF) biology. Current models include two-dimensional and three-dimensional cultures. Potential complementary models that have not yet been applied to the study of CAF biology include air-liquid interface cultures, organs-on-chips, three-dimensional bio-printed tissues, and new genetically engineered mouse models (GEMMs) for lineage tracing of CAFs.

1. In vitro models

In vitro models enable to mechanistically investigate the crosstalk between different cell populations. Although monolayer two-dimensional cultures of fibroblasts can be useful to dissect some aspects of CAF biology, it has been shown that the transcriptome of CAFs cultured in this fashion does not recapitulate the heterogeneity of CAFs in vivo (240, 292, 305). Three-dimensional matrices, such as Matrigel and hydrogels, and tumor organoid/fibroblast co-cultures have significantly advanced our knowledge of fibroblast biology, as they more faithfully mirror the transcriptional profiles and phenotypes of CAF populations found in vivo (15, 24, 213). Nonetheless, there are limitations to using Matrigel or hydrogel scaffolds, as they are distinct from the composition of the tumor ECM, and co-cultures with these matrices may not entirely recapitulate the in vivo repertoire of fibroblasts (305). Implementation of models that use the ECM produced by the fibroblasts (87), short-term co-cultures that retain all cell populations found in tumors, such as liquid-air interface models (204, 217), multi-cell-type three-dimensional bio-printed tissues (155, 178) and microfluidic culture technologies, such as organs-on-chips (8, 78, 160, 275), are being implemented and optimized and could complement or validate other models (FIGURE 1).

2. In vivo models

GEMMs have been developed for lineage tracing of fibroblasts in wound healing and normal and inflammatory tissues (71, 196, 252–254, 271, 315) and have significantly advanced our knowledge of fibroblast origin, heterogeneity, plasticity, and roles. On the contrary, lineage tracing models of CAFs are lacking. These GEMMs are needed in particular as CAFs can originate from multiple cell populations, and this may partially determine their function. Cre recombinase-based models are present (333) and could be useful to dissect the role of fibroblasts in the TME in vivo. However, recent studies warn about the use of tamoxifen when studying the microenvironment (51, 52), highlighting the need for the generation and analysis of alternative fibroblast-specific mouse models to study CAF biology. Finally, intravital microscopy imaging techniques (80) applied to GEMMs for lineage tracing of CAFs could evaluate the origin and plasticity of dynamic CAF subtypes (24).

B. Molecular Heterogeneity: Markers of CAF Subtypes

A number of intracellular, extracellular, and cell surface proteins have been used to isolate or identify CAFs (FIGURE 2). However, no ubiquitous marker to study CAFs exists. CAF heterogeneity in primary tumors first emerged in immunofluorescence experiments, which showed that, among various fibroblast markers, none was all-inclusive and many were present in different combinations (206, 285). The development of new in vitro models (213), the optimization of multicolor flow cytometry and immunohistochemical methodologies (53, 123, 284), and the extensive use of scRNA-seq techniques enabled the redefinition of some classical markers as subtype-specific. For example, elevated expression of α-smooth muscle actin (αSMA) has been historically considered a distinctive marker in activated CAFs compared with normal fibroblasts (158). Nonetheless, we and others have recently demonstrated the presence of low-αSMA CAFs (19, 20, 24, 53, 73, 111, 166, 213), highlighting the need to consider multiple parameters for the analysis of CAF activation.

FIGURE 2.

Common cancer-associated fibroblast (CAF) markers. Examples of extracellular, intracellular, and surface protein markers of CAFs. LUM, lumican; DCN, decorin; COL, collagen; FAP, fibroblast activation protein; PDPN, podoplanin; PDGFR, platelet-derived growth factor receptor; αSMA, α-smooth muscle actin; VIM, vimentin; FSP-1, fibroblast specific protein 1.

An additional technical challenge for the study of CAFs is the absence of a specific marker for their isolation. This is due to the fact that most markers are shared with at least one other cell population. For example, proteins broadly expressed across CAFs, such as αSMA, podoplanin (PDPN), and FAP, are highly expressed in pericytes, lymphatic endothelial cells, and fibroblastic reticular cells, respectively. Therefore, strategies that rely on the negative selection of other cell populations are preferable in contexts where fibroblast heterogeneity has yet to be dissected. However, when subtype-specific markers are known, a combination of these with broadly expressed fibroblast markers could be used to isolate specific CAF populations.

Whereas some markers are simply identifiers of specific CAF subtypes, some, such as caveolin 1 in breast cancer (94), have been shown to also have a functional role. Additionally, a number of transcription factors, including HSF1 (262), STAT3 (24), MYC (264), and YAP (36), are involved in driving specific CAF signatures and reprogramming. Adding to this complexity, pathways that define a CAF state, such as NF-κB signaling, can be either tumor-promoting or tumor-suppressive, depending on the cancer type and context (24, 79, 142, 222).

Importantly, emerging evidence suggests the presence of similar CAF populations across different cancer types. For example, whereas a tumor-promoting CD10+ GPR77+ CAF subtype has been recently characterized in breast cancer (284), CD10+ CAFs have been previously shown to promote PDAC progression (118). Additionally, scRNA-seq and multicolor flow cytometric data sets of several cancer types demonstrate the presence of transcriptionally distinct CAF populations, but also highlight commonalities of fibroblast signatures across different malignancies (19, 20, 24, 53, 57, 73, 111, 154, 166, 240, 284, 298) (FIGURE 3 and TABLE 2). Analyzing the similarities in fibroblast composition across various cancer types could reveal new therapeutic opportunities with a broad impact.

FIGURE 3.

Cancer-associated fibroblast (CAF) heterogeneity across malignancies. Schematic illustration summarizing the distinct CAF populations identified in different cancer types by single-cell RNA-sequencing (scRNA-seq). αSMA, α-smooth muscle actin.

Table 2.

CAF heterogeneity across malignancies

Cancer Type	Models/Methods	CAF Subtypes	CAF Subtype Defining Markers	Subtype Putative Functions	Additional Notes
PDAC (24, 73)	•KPC tumors and patient samples (PDAC and adjacent normal)	•Inflammatory CAFs (overlaps with iCAFs)	•Ly6CHigh αSMALow CXCL12+ PDGFRαHigh C3+ IL6+	•Immunosuppressive/tumor promoting	•IL-1 and JAK/STAT signatures; distally located; less proliferative; can convert into myCAFs in vivo
	•scRNA-seq by 10× genomics	•Myofibroblastic CAFs (overlaps with myCAFs)	•αSMAHigh CTGF+ TNC+ TAGLN+	•ECM producing	•TGF-β signature; tumor adjacent
		•Antigen-presenting CAFs (apCAFs)	•MHCII+	•Immunomodulatory	•Mesothelial cell origin?
PDAC (20)	•Patient samples (pre-neoplastic IPMNs and PDAC)	•Inflammatory CAFs (iCAFs)	•αSMALow CXCL12+ C3+ IL6+	•Immunosuppressive/tumor promoting	•Absent in IPMNs
	•scRNA-seq by Drop-seq	•Myofibroblasts (myCAFs)	•αSMAHigh	•ECM producing	•Prevalent at later stage
PDAC (111)	•KPC, KPfC and KIC tumors (early and late stage)	•FB1 (overlaps with iCAFs)	•αSMALow CXCL12+ PDGFRαHigh IL6+	•Immunosuppressive/tumor promoting	•Express some apCAF/mesothelial markers (e.g., MSLN, CD74, H2-AA); prevalent at later stage
	•scRNA-seq by 10× genomics	•FB3 (overlaps with myCAFs)	•αSMAHigh TAGLN+ CTGF+
PDAC (64)	•KPP tumors (early and late stage)	•C8 (overlaps with iCAFs)	•αSMALow Ly6CHigh IL6+	•Immunosuppressive	•IL-1 and NF-κB signature; has normal precursor (C3)
	•scRNA-seq by 10× genomics	•C2 (overlaps with myCAFs)	•αSMAHigh TAGLN+ LRCC15+	•ECM producing/immunosuppressive	•TGF-β signature; has normal precursor (C4); prevalent at later stage
Lung cancer (154)	•Patient samples (tumors and matched nonmalignant lung)	•5 Clusters	•#2: αSMAHigh	•Angiogenesis; myogenesis	•Co-clusters with pericytes
	•scRNA-seq by 10× genomics		•#1: EMT signature	•ECM producing	•Enriched in tumors compared with normal tissues; TGF-β signature
Breast/lung cancer (284)	•Patient samples	•CD10+ GPR77+ CAFs	•CD10+ GPR77+ IL6+	•Promote cancer stemness and chemotherapy resistance	•NF-κB signature
	•mRNA microarray analysis and flow cytometry
Breast cancer (53)	•Patient samples (LumA, HER2, and TNBC subtypes)	•CAF-S1	•FAPHigh αSMA+ CXCL12+ IL6+	•Immunosuppressive/ECM producing	•Peri-tumoral location/mostly in TNBC
	•Multicolor flow cytometry and immunohistochemistry	•CAF-S2	•Negative for all markers	•Contractile signature	•Potentially not fibroblasts
		•CAF-S3	•αSMALow PDGFRβ+		•Prevalent in adjacent normal/mostly in TNBC
		•CAF-S4	•αSMA+ CD29High		•Pericytes?
Breast cancer (19)	•MMTV-PyMT tumors	•Vascular CAFs (vCAFs)	•αSMAHigh PDGFRβHigh	•Angiogenesis	•Proximal to the vasculature; Prevalent at later stage
	•scRNA-seq by SMART-seq2	•Matrix CAFs (mCAFs)	•αSMALow PDGFRαHigh	•ECM producing	•Prevalent at earlier stage
		•Cycling CAFs (cCAFs)	•PDGFRβHigh	•Angiogenesis	•Cycling vCAFs
		•Developmental CAFs (dCAFs)	•PDGFRβ- SOX9+ SCRG1+	•Cell differentiation	•EMT cells?
Breast cancer (298)	•MMTV-PyMT tumors	•ECM-CAFs	•TNC+	•ECM producing	•Desmoplastic signature
	•scRNA-seq by Drop-seq	•Inflammatory CAFs (iCAFs)	•Ly6CHigh, C3+, CXCL12+ PDGFRαHigh	•Immunomodulatory	•JAK/STAT signature
		•Myofibroblastic CAFs	•αSMAHigh, MYLK+	•Contractile signature
Melanoma (57)	•Tumors from B16-F10 cell transplantation model	•Immune CAF1	•CD34High CXCL12+ C3+	•Immunosuppressive	•Prevalent at early stage; less proliferative
	•scRNA-seq by 10× genomics	•Desmoplastic CAF2	•CD34Low CTGF+ TNC+ PDGFRα+	•ECM producing	•Prevalent at early stage; intermediate CAFs
		•Contractile CAF3	•αSMAHigh RGS5+	•Contractile signature	•Prevalent at late stage; pericytes?
Head and neck cancer (240)	•Patient samples (tumors and lymph node metastases)	•Myofibroblasts	•αSMAHigh, MYL9+, MYLK+	•Contractile signature	•The prevalent subtype in lymph node metastases
	•scRNA-seq by SMART-seq2	•Activated CAFs	•PDGFRαHigh	•ECM producing	•Non-myofibroblastic; 2 subclusters
Colon cancer (166)	•Patient samples (tumors and matched normal mucosa)	•CAF-A	•αSMALow TAGLNLow FAP+	•ECM remodeling	•Intermediate state?
	•scRNA-seq with Fluidigm C1 platform	•CAF-B	•αSMAHigh TAGLNHigh FAP-		•Activated myofibroblasts

List of recent scRNA-seq and multi-color flow cytometric studies that reveal differences and commonalities in cancer-associated fibroblast (CAF) subtypes across pancreatic ductal adenocarcinoma (PDAC), breast cancer, lung cancer, melanoma, colon cancer, and head and neck cancer. ECM, extracellular matrix; FAP, fibroblast activation protein; IL, interleukin; PDGF, platelet-derived growth factor; αSMA, α-smooth muscle actin; TGF, transforming growth factor.

Overall, the existence of both myofibroblastic and non-myofibroblastic CAF populations appears to be the most consistent observation across different cancer types (19, 20, 24, 53, 73, 89, 111, 166, 231). We first described these two populations in a tumor organoid/fibroblast co-culture model of PDAC that revealed the presence of αSMA-high myofibroblastic CAFs (myCAFs) and αSMA-low inflammatory CAFs (iCAFs) (213). We and others have confirmed these findings in vivo by scRNA-seq and immunochemical analysis of murine and human PDAC specimens (20, 24, 73, 111, 231). Across various cancer types, myofibroblastic CAFs are associated with an ECM signature, whereas non-myofibroblastic CAFs are generally characterized by a secretory, inflammatory phenotype (TABLE 3). Notably, inflammatory CAFs share similar transcriptional profiles and signaling pathway activation with senescent fibroblasts, which are characterized by the senescence-associated secretory phenotype (SASP) (48, 109, 148, 149). Additionally, inflammatory CAFs have been shown to proliferate at a lower rate than myofibroblasts or to not proliferate at all (24, 57). Regardless, we have shown that a substantial fraction of iCAFs actively proliferates in PDAC (24, 213), indicating that iCAFs should not be equated with fibroblasts undergoing SASP. However, these similarities may indicate overlapping functions of iCAFs and senescent fibroblasts. In cancer, the SASP of stromal cells has been reported to play various roles (147, 256), highlighting the need to identify the effects of inflammatory CAF-secreted ligands on cancer progression.

Table 3.

CAF subtypes in PDAC

PDAC CAF Subtype	Other Nomenclature	Subtype Defining Markers (in combination with pan-CAF markers, observed across data sets)	Potential Functions	Potential Cell of Origin	Potential Targeting Agents (nonselective)
myCAF (20, 24, 64, 73, 111, 273)	•Myofibroblastic CAFs	•αSMAHigh	•ECM producing: loss of αSMA+ CAFs has been associated with ECM depletion (24, 161, 215, 221, 248, 330)	•Can originate from PSCs (24, 213, 273)	•TGFBR inhibitors, TGF-β antibodies, losartan
	•Myofibroblasts	•CTGF	•Immunosuppressive: express TGF-β (24, 64, 73)	•Pericytes (similar transcriptional profile)	•SMO inhibitors (IPI-925, Vismodegib, LDE225)
	•FB3	•TAGLN	•Tumor restraining: loss of αSMA+ CAFs has been associated with reduced survival and increased vascularization and/or recruitment of immunosuppressive cell populations (161, 221, 248)	•Col4a1+ ENG+ C4 fibroblasts (64)
	•C2	•THY1
		•LRRC15 (subset)
iCAF (20, 24, 64, 73, 111, 273)	•Inflammatory CAFs	•αSMALow	•Immunosuppressive: express ligands (e.g., CXCL12, CXCL1, IL6, G-CSF, …) involved in T cell exclusion and neutrophil recruitment (24, 73, 213, 273)	•Can originate from PSCs (24, 213, 273)	•JAK/STAT inhibitors
	•FB1	•CXCL12	•Tumor promoting: induce phospho-STAT3 in PDAC organoids (213); loss of iCAFs leads to smaller tumors (24)	•Fbn1+ LY6C+ DPP4+ C3 fibroblasts (64)	•IL1R antagonist
	•C8	•C3	•Overlapping functions with senescent fibroblasts: iCAF secretome ~ SASP (24, 73, 213, 273)		•NF-κB inhibitors
		•IL6
		•Ly6CHigh
		•PDGFRαHigh
apCAF (73)		•MHCII (Cd74, H2-Aa, H2-Ab)	•Immunomodulatory (do not express costimulatory molecules, may act as decoy receptor to inhibit optimal T cell response)	•Cannot originate from PSCs?
			•Tumor promoting: express SAA3, SLPI	•Mesothelial cells (64, 73)
				•MHCII expression induced by IFN-γ

Summary of actual and potential features of inflammatory cancer-associated fibroblasts (iCAFs), myofibroblastic CAFs (myCAFs), and antigen-presenting CAFs (apCAFs) in pancreatic ductal adenocarcinoma (PDAC). ECM, extracellular matrix; G-CSF, granulocyte colony stimulating factor; IFN, interferon; IL, interleukin; PDGF, platelet-derived growth factor; PSCs, pluripotent stem cells; SASP, senescence-associated secretory phenotype; αSMA, α-smooth muscle actin; SMO, Smoothened; TGF, transforming growth factor.

New CAF markers are constantly found, and the implementation of scRNA-seq methods has been significantly contributing to this. CAF subtype-specific markers could guide the development of novel genetic and pharmacological approaches to target specific CAF populations.

C. Specification of CAF Subtypes

Although the genetic and epigenetic drivers of cancer cells have been extensively investigated, the mechanisms governing the recruitment and activation of CAFs are largely unknown. Whereas genetic alterations in CAFs are rare (4, 11, 112, 241), with only a few exceptions (201, 227, 297), we and others have shown that the genetic profile of cancer cells can affect the surrounding stroma (209, 237, 273, 302, 312). Therefore, CAF signatures defined across diverse cancer genetic profiles could be used to stratify patients, provide prognostic information, and tailor therapies.

Several factors have been demonstrated to contribute to the reprogramming of CAFs or the activation of fibroblasts in inflammatory and fibrotic conditions (FIGURE 4), including 1) epithelial cues, such as interleukin-1 (IL-1), platelet-derived growth factor (PDGF), and TGF-β (6, 24, 35, 79, 162, 236, 242, 272, 317); 2) metabolic reprogramming (328); 3) oxidative stress (189, 294); 4) stromal cues (207, 271, 279); 5) microRNAs (83, 198, 319); 6) epigenetic changes (2, 22, 113, 230, 235, 314); and 7) other CAF-secreted ligands (141). In PDAC, we and others found that cancer-secreted IL-1 and TGF-β antagonize each other and define inflammatory iCAF and myofibroblastic myCAF formation, respectively (24, 64, 273, 330).

FIGURE 4.

Factors involved in fibroblast reprogramming in inflammation and cancer. Schematic illustration of multiple stimuli that have been shown to determine fibroblast activation. These factors include epithelial and stromal cues, metabolic reprogramming, epigenetic changes, microRNAs, and oxidative stress. CAF, cancer-associated fibroblast; IGF-1, insulin-like growth factor 1; IL-1, interleukin-1; PDGF, platelet-derived growth factor; TGF-β, transforming growth factor-β.

D. Lineage-Dependent Heterogeneity: Cell of Origin of CAF Subtypes

We and others have shown that CAF activation and reprogramming can occur within the TME. Nonetheless, the identification of the cells of origin of CAF subtypes is a central question, as it may partially determine the functions of distinct CAF populations and could indicate new therapeutic strategies.

Whereas elegant lineage tracing studies have revealed the origin of fibroblasts in normal or injured fibrotic tissues (65, 68, 251, 271), little information exists about the origin of CAFs. Several cell types have been suggested as precursors of CAFs (FIGURE 5), including 1) pancreatic and hepatic stellate cells (10, 16, 322), 2) resident fibroblasts (12, 141, 218, 255, 303), 3) mesenchymal stem cells (MSCs) (128, 132, 135, 159, 197, 242), 4) adipocytes (26, 92, 135), 5) adipose-derived MSCs (282, 288), 6) mesothelial cells (261), 7) endothelial cells (326, 327), 8) myeloid cells (211), 9) pericytes (110), 10) epithelial cells (122), 11) hematopoietic stem cells (195), and 12) circulating bone marrow cells known as fibrocytes (62). However, the majority of the evidence that identifies these cell types as CAF precursors comes from in vitro experiments and bone marrow transplantation studies and has been reported only for one or a few cancer types. Even less information exists about the distinct cells of origin of specific CAF subtypes. We have shown that, at least in vitro, pancreatic stellate cells can differentiate into both inflammatory iCAFs and myofibroblastic myCAFs (213). On the contrary, the antigen-presenting CAF (apCAF) subtype, which we recently identified in PDAC by scRNA-seq and immunohistochemical analysis (73), shares an overlapping transcriptional signature with mesothelial cells (31, 64, 316) (TABLE 3). This observation suggests that apCAFs could derive from mesothelial cells, following their recruitment into the tumor, similarly to what observed in inflammatory conditions (168, 174, 250).

FIGURE 5.

Cell of origin of cancer-associated fibroblasts (CAFs). Schematic illustration of the potential cells of origin of CAFs that have been reported, including epithelial cells, mesothelial cells, resident fibroblasts, pancreatic and hepatic stellate cells, pericytes, adipocytes, mesenchymal stem cells, myeloid cells, fibrocytes, and endothelial cells.

CAF subtype-specific lineage tracing models coupled with in vivo imaging methods could reveal the cells of origin of distinct CAF populations.

E. Functional Heterogeneity of CAF Subtypes

Whereas normal fibroblasts have been shown to inhibit cancer progression (3, 280, 296), several studies have demonstrated that CAFs can promote tumor growth in multiple ways (FIGURE 6), including 1) secreting ECM proteins (124, 215, 239); 2) inducing inflammation and neovascularization (79, 218); 3) increasing angiogenesis (44, 134); 4) increasing the incidence of tumor-initiating cells (157); 5) affecting the signaling of cancer cells (289); 6) changing the metabolism and epigenome of cancer cells (13, 70, 267, 332); 7) establishing an immunosuppressive TME (84, 146, 152); 8) conferring therapy resistance (106, 121, 134, 281, 302) and radioprotection (183); 9) promoting tumor invasion, metastasis formation, and epithelial-to-mesenchymal transition (9, 35, 63, 67, 91, 94, 95, 134, 151, 207, 304, 313, 329); 10) secreting pro-tumorigenic ligands (218, 269); 11) promoting the stemness of cancer cells (40, 45, 66, 163, 182, 247, 284, 286, 325); 12) contributing to systemic effects, such as cachexia, anemia, and immunosuppression (86, 252); and 13) fueling cancer growth by providing metabolites, such as amino acids (214, 276), fatty acids (13), and lactate (229).

FIGURE 6.

Tumor-promoting cancer-associated fibroblast (CAF) functions. Schematic illustration of the functional heterogeneity of CAFs in the tumor microenvironment. CAFs have been demonstrated to play several roles, including promoting tumor growth and metastasis formation, depositing extracellular matrix (ECM), and establishing an immunosuppressive microenvironment. HA, hyaluronan; HGF, hepatic growth factor; IGF1, insulin growth factor 1; IL-6, interleukin-6; LIF, leukemia inhibitory factor.

1. ECM deposition

High mammographic density, which indicates greater abundance of connective tissue compared with fat, is an important risk factor in breast cancer (32). Far from simply being a read-out of the increased incidence in breast cancer, it has been shown that gene expression programs associated with this condition, such as downregulation of the membrane protein CD36, correlate with clinical outcomes (58). Importantly, high mammographic density shares features with the TME, in particular relative to the elevated ECM deposition and stromal content. Indeed, ECM remodeling and deposition are not restricted to cancer, but are also present during wound healing and in fibrotic and inflammatory states (27). However, although following tissue repair and inflammation the formation of desmoplasia is typically reversible (139), ECM levels continue to increase with cancer progression.

ECM deposition plays several roles within the TME, including 1) acting as a barrier to drug delivery (124, 215, 239); 2) leading to hypoperfusion and elevated interstitial fluid pressure by compressing blood vessels and lymphatic vessels, respectively (42, 188, 283); 3) providing nutrients (214); 4) contributing to the establishment of an immunosuppressive TME (105, 126, 259); and 5) supporting tumor growth with biochemical and mechanical cues (55, 233) (FIGURE 6). Indeed, the high tissue tension and stiffness caused by the ECM has been shown to promote cancer progression by increasing cell invasion and cancer spreading (126, 153, 165, 208).

More than any other cancer type, PDAC is characterized by an extensive ECM deposition, thus representing a potential paradigm for its composition and function. A recent study that looked at murine and human normal pancreas, pancreatitis, and PDAC samples found that the matrisome (i.e., ECM proteins, growth factors, and ECM-associated proteins) of pancreatitis almost entirely contributes to that of PDAC (291). This big overlap of ECM components shared by both the inflammatory and malignant states suggests that similar fibroblast populations could also be conserved. Importantly, several ECM components shared between the premalignant and malignant states are absent in normal tissue, suggesting that they could represent biomarkers and therapeutic targets (291). This study also confirmed that fibroblasts are the major producers of the ECM, in particular of fibrillar collagens (e.g., COL1A1, COL1A2), glycoproteins (e.g., fibronectin), and proteoglycans (e.g., perlecan). However, whereas the small percentage of cancer-derived ECM proteins was only found associated with worst prognosis (291), stroma-derived ECM components correlated with both poor and good survival. These results indicate that the ECM is functionally heterogeneous, and scRNA-seq studies could assess whether different ECM components are secreted by distinct CAF subtypes. Alterations of the ECM composition during cancer treatment (155, 293) could, thus, indicate additional changes in CAF subtypes and function.

Several ECM components have been therapeutically targeted to increase the efficacy of immunotherapy and chemotherapy (43, 61, 124, 205, 215, 239) or have been leveraged for noninvasive imaging of fibrotic conditions, premalignant lesions, primary tumors, and metastases (125). Indeed, the ECM is also present in metastatic tumors (182, 219, 220), although differences between the ECM composition and response to chemotherapy at primary and secondary sites have been reported (1, 310). Whether these differences are dependent on the presence of distinct CAF subtypes between these sites remains to be assessed.

2. Immunosuppression

Several lines of evidence have identified a major role of CAFs in the establishment of an immunosuppressive TME (FIGURE 6). CAFs have been shown to prevent cytotoxic T cell activity and recruitment within tumors (84, 152), in part by secreting immunosuppressive ligands, such as TGF-β (85, 101, 184, 290) and CXCL12 (75, 84, 218). Additionally, CAFs can recruit immunosuppressive populations, such as myeloid-derived suppressor cells (MDSCs) and neutrophils (150, 273, 321), which in turn can further activate CAFs (279). Furthermore, CAFs have been involved in monocyte recruitment and in macrophage differentiation and polarization (14, 49, 190). Although these studies highlight the several ways through which CAFs establish an immunosuppressive TME and limit the efficacy of immunotherapy strategies, it remains to be assessed which CAF subtypes are responsible for this. In PDAC, we have identified inflammatory iCAFs as the major producers of CXCL12 (73) and other immunosuppressive ligands (24), such as IL-6 (86), CXCL1 (167), and granulocyte colony stimulating factor (G-CSF) (234). These observations suggest that iCAFs may play a central immunosuppressive role in the PDAC TME (TABLE 3). Additionally, apCAFs, which express MHC class II (MHCII) proteins, may play an immunomodulatory role in PDAC and breast cancer (73, 89). Whereas MHCII molecules are constitutively present in professional antigen presenting cells (APCs), in other cell types, including fibroblasts, the expression of MHCII can be induced by stimuli such as interferon-γ (28, 119, 306). Notably, MHCII-expressing CAFs do not express co-stimulatory molecules (73, 89), which are typically present in professional APCs and are necessary for the induction of CD4+ T cell clonal proliferation (320). This observation indicates that apCAFs may act as a decoy receptor to inhibit optimal T cell response (TABLE 3). Although this is an intriguing hypothesis, a role of MHCII-expressing CAFs in cancer progression remains to be demonstrated.

3. Potential tumor-restraining roles of CAFs

For the reasons outlined above, CAFs have been historically considered tumor-promoting components. However, recent work mostly carried out in PDAC indicates that CAFs may also have tumor-restraining functions (5, 136, 161, 221, 228, 248, 330). These studies largely focused on the genetic depletion of αSMA-positive cells and on the genetic deletion of sonic hedgehog (SHH) or the pharmacological inhibition of smoothened (SMO), which are key components of the Hedgehog (Hh) pathway (5, 136, 161, 221, 228, 248, 270) (FIGURE 7). Targeting CAFs with these approaches led to reduced survival in both preclinical and clinical studies, indicating a potential tumor-restraining role of these cells. Of note, these strategies were thought to be designed to target CAFs as a whole rather than specific CAF subtypes, as CAF heterogeneity had not yet been described and αSMA was still considered a ubiquitous marker of activated CAFs. However, with the current knowledge, we can speculate that selective targeting of αSMA-expressing cells preferentially depleted the myofibroblastic myCAF population (221). Moreover, this genetic approach would also affect other cell populations that express αSMA, such as pericytes, and therefore, it does not necessarily indicate a role of CAFs in restraining tumor progression. It also remains to be determined which CAF subtypes are affected following abrogation of Hh signaling, although the observation that inhibition of this pathway led to a reduction in ECM deposition and αSMA+ cells (215, 248) suggests that ECM-producing myCAFs are targeted.

FIGURE 7.

Potential tumor-restraining cancer-associated fibroblast (CAF) functions. Schematic illustration of the Hedgehog (Hh) signaling pathway (top) and of the genetic and therapeutic approaches that indicated the presence of tumor-restraining functions of pancreatic ductal adenocarcinoma (PDAC) CAFs (bottom). A: genetic deletion of the Hh ligand sonic hedgehog (SHH) in a mouse model of PDAC led to a reduction in survival and tumor differentiation and to an increase in angiogenesis and cachexia, a highly debilitating muscle-wasting condition (248). GLI1, GLI family zinc finger 1; PTCH1, protein patched homolog 1. αSMA, α-smooth muscle actin. B: genetic depletion of αSMA+ cells in a mouse model of PDAC led to a reduction in survival and tumor differentiation and to an increase in the infiltration of immunosuppressive CD4+ Foxp3+ regulatory T cells (221). C: prolonged pharmacological inhibition of the Smoothened (SMO) receptor in a mouse model of PDAC led to reduced survival and tumor differentiation, and to an increase in angiogenesis and cachexia (248).

Overall, although some progress has been made to determine the functions of phenotypically distinct CAF subtypes (24, 73, 284), more work needs to be undertaken to elucidate whether CAF heterogeneity is clinically relevant, rather than simply descriptive.

F. Stage-Dependent Fibroblast Heterogeneity

1. During cancer progression

Although we now have a more complete picture of the CAF composition at primary tumors in various cancer types, it has to be contextualized during cancer progression. Indeed, the identity and proportion of distinct CAF subtypes could differ across normal, inflammatory, premalignant, and malignant states (FIGURE 8). A better understanding of the populations present during cancer progression is crucial to decipher the evolution of CAF heterogeneity and its effects on other cell populations of the TME.

FIGURE 8.

Fibroblast activation in inflammation and cancer. Schematic illustration showing the phenotype changes associated with fibroblast activation during inflammation and cancer progression. Compared with normal quiescent fibroblasts, fibroblasts in inflammatory and malignant states have various degrees of increased α-smooth muscle actin (αSMA) expression, secretory phenotype, extracellular matrix (ECM) deposition, proliferation, contractility, and morphological activation. Cancer-associated fibroblasts can also be immunosuppressive.

An increasing number of scRNA-seq studies currently provide some information about the fibroblast composition in early- or late-stage primary tumors, compared with adjacent normal or metastatic tissues (20, 57, 64, 73, 89, 111, 154, 166, 240). Single-cell pseudo-time trajectory analysis (295) of these and other data sets could predict the evolution of fibroblasts during malignant progression (64). Additionally, longitudinal studies are needed to dissect fibroblast heterogeneity over time and at distinct sites. Since for certain cancer types, such as PDAC, patient tissues will largely represent early cancer stages due to tissue sampling limitations, the use of mouse models will complement the analysis of human fibroblasts during cancer progression (298).

We currently have very little information about the CAF composition in metastases. scRNA-seq studies of metastatic cancers showed that neoplastic cells from different patients clustered separately both at the primary and secondary site, indicating the high degree of inter-individual heterogeneity in the epithelial cancer compartment (176, 226, 240, 292). On the contrary, although different in proportions, fibroblasts and other stromal cells clustered by cell type independent of patients, stages, and tissues of origin, indicating the presence of conserved transcriptional programs across different tumors and during cancer progression (176, 226, 240, 292). Nonetheless, diverse immune microenvironments, which respond differently to the same therapeutic regime, have been identified in distinct metastatic sites of the same ovarian cancer patient (127). A similar scenario could occur for CAFs and may influence the immune representation and therapy response at different sites as well. Accordingly, differences in fibroblast signatures have been identified between micrometastases and macrometastases (264). Furthermore, fibroblasts in metastatic sites can be activated differently than CAFs in primary tumors (207). Additionally, the ECM composition and response to chemotherapy can differ between primary and secondary sites (1, 310), further supporting the presence of functionally and phenotypically distinct CAF populations. Moreover, although it has been shown that CAFs can migrate with cancer cells from the primary site (67), resident fibroblasts have also been involved in the formation of a premetastatic niche (131, 210). Additional comparisons of primary and secondary sites at single cell resolution will help assess the contribution of both resident and recruited cell populations in defining the CAF composition.

2. During aging and senescence

Fibroblast abundance and ECM integrity have been shown to decrease with age (7, 100, 186, 224, 300). Previous reports have highlighted that aging in dermal fibroblasts is associated with accumulated genetic damage (169, 202), telomere attrition (181), increased secretion of inflammatory cytokines (180, 335), loss of identity, and gain of adipogenic traits (260). Transcriptomic, epigenomic, and metabolomic analyses of fibroblasts from young and old mice have revealed the presence of age-dependent heterogeneity in the rate of wound healing and in the efficiency of reprogramming to induced pluripotent stem cells (iPSCs) (180). In particular, the presence of a fibroblast activated signature typical of myofibroblasts involved in tissue repair (74) was associated with efficient reprogramming of aged fibroblasts to iPSCs (180).

Importantly, in melanoma, aged fibroblasts have an altered ECM-deposition compared with young fibroblasts, which can increase tumor motility, metastasis formation, and immunosuppression (133). However, despite the fact that cancer incidence increases with age, our knowledge of how distinct CAF subtypes of an aged microenvironment contribute to progression and therapy response in other cancer types remains limited. Indeed, most in vivo preclinical studies employ mice that have not yet reached the geriatric phenotype (∼18–24 mo of age, corresponding to ∼60 human years) (69). Therefore, such models may not fully recapitulate the stromal composition and molecular interactions between CAFs and cancer cells. New in vitro and in vivo models are needed to understand the impact of an aged TME in cancer progression.

Additionally, as tissues age, they accumulate cells that undergo a persistent arrest of the cell cycle, a process known as senescence, in response to multiple stimuli, such as chromatin remodeling, telomere attrition, and environmental stress (37, 76, 82, 88, 117, 164, 173, 216, 230, 324). Importantly, the accumulation of senescent cells has been associated with inflammation and tissue dysfunction (34, 88). Although senescence has been used as a model to study fibroblasts in aging, aged and senescent fibroblasts only have a few overlapping features (38, 133, 307, 308). In particular, senescent cells are characterized by the SASP, which includes cytokines, chemokines, and ECM-remodeling enzymes (50), and has been shown to play various roles in cancer progression (81, 147, 256). For example, senescent fibroblasts have been associated with inflammatory phenotypes that promoted pancreatic cancer progression (265, 309).

V. FIBROBLAST PLASTICITY

Recent evidence indicates that CAF subtypes are dynamic and able to interconvert depending on tumor cues, culture conditions, and therapeutic regimens (24, 73, 89, 213). This observation suggests that CAF subtypes could be cell states rather than end-points in differentiation. This plasticity represents a technical challenge as CAF populations may lose their subtype-defining phenotype when isolated from tumors, but it also represents a therapeutic opportunity. Indeed, rather than ablating a tumor-promoting CAF subtype, it could be more beneficial shifting it towards a tumor-restraining or quiescent population.

We recently showed that CAF subtype reprogramming in vivo is possible with the JAK inhibitor AZD1480, which shifted iCAFs to myCAFs and led to increased ECM deposition in a mouse model of PDAC (24) (FIGURE 9). These results raise the possibility that, in some circumstances, drug delivery could be decreased following CAF subtype switching. These observations thus indicate that the efficacy of standard of care drugs could be improved by understanding their effects on the stroma. Similarly, drugs that failed in clinical trials could be re-evaluated in combinatorial strategies with treatments that would mitigate their negative effect on the TME.

FIGURE 9.

Cancer-associated fibroblast (CAF) plasticity. Schematic illustration showing the effects of the JAK inhibitor AZD1480 on reprogramming inflammatory CAFs (iCAFs) in a mouse model of pancreatic ductal adenocarcinoma (PDAC) (24). Treatment with the JAK inhibitor shifts the iCAF subtype towards an extracellular matrix (ECM)-producing myofibroblastic CAF (myCAF) population, leading to an increase in the myCAF/iCAF ratio and ECM deposition and to a reduction in tumor cell proliferation and tumor growth.

Fibroblast plasticity is not unique to malignant states, and scRNA-seq techniques have determined the molecular basis of fibroblast conversion to various cell states and types (171). For example, behavior switching between different states has been observed in the skin, where fibroblast populations fluctuate between ECM deposition and cell proliferation (254) and in pulmonary fibrosis (71). This plasticity also occurs during aging, as dermal fibroblasts acquire adipogenic markers (260).

VI. CAFs AND THERAPEUTICS

Although epithelial cells have been at the center of the efforts for the development of new anti-cancer therapies, a number of CAF-targeting strategies have been tested in preclinical and clinical trials (46) (FIGURE 10). The historic rationale for therapeutically targeting CAFs has been to decrease the ECM and the production of tumor-promoting and immunosuppressive ligands, to increase the efficacy of chemotherapies and immunotherapies. Moreover, long-range effects of CAFs in promoting metastasis formation and systemic conditions, such as cachexia and anemia, recently emerged as additional reasons to target CAFs (129, 212).

FIGURE 10.

Stroma-targeting therapies. Schematic illustration of the stroma-targeting strategies that have been tested in preclinical and clinical studies, including agents targeting components of the ECM (e.g., PEGPH20 and FG-3019), anti-immunosuppressive strategies, pathway-specific inhibitors (e.g., SMO inhibitors) and drugs that reprogram CAFs to a quiescent state (e.g., calcipotriol). ATRA, all-trans-retinoic acid; FAK, focal adhesion kinase; FAP, fibroblast activation protein; IL6, interleukin-6; LIF, leukemia inhibitory factor; SMO, Smoothened.

CAFs promote chemotherapy resistance in part through the secretion of ECM proteins, which limit access to drug delivery by forming a thick barrier (61, 124, 178, 215, 239) and compressing blood vessels and lymphatic vessels, leading to hypoperfusion and elevated interstitial fluid pressure (42, 188, 283). Among strategies that deplete the ECM, PEGPH20 targets the ECM component hyaluronan (HA) (124, 239). Treatment with hyaluronidase PEGPH20 and chemotherapy initially appeared promising in clinical trials for PDAC (107, 108), with the exception of a combinatorial approach with mFOLFIRINOX (243). However, a phase III clinical trial targeted at HA-high patients (HALO 301) recently failed, and PEGPH20 development has been discontinued. Similarly, SMO inhibitors, which lead to ECM depletion by blocking the Hh signaling pathway and targeting ECM-producing αSMA+ CAFs (40, 161, 215, 248), have shown adverse or no effects in clinical trials for PDAC (39, 136, 140, 161, 248). These results could be due to the role of Hh dosage and the diverse consequences of Hh inhibition on PDAC progression and pancreatic tissue remodeling (191, 192, 215, 248). Other strategies to target the ECM include the use of Rho-associated protein kinase inhibitors, such as Fasudil and AT13148 (244, 245, 301), and antibodies against ECM proteins, such as connective tissue growth factor (CTGF), fibronectin, and tenascin C (205, 210, 246). Interestingly, blocking CTGF in PDAC led to enhanced chemotherapy response by modulating survival cues in cancer cells without affecting drug delivery (205), indicating that targeting of ECM components can be therapeutically effective independently from the reduction of desmoplasia.

CAF deactivation with loss of myofibroblastic features has been observed during chronic hypoxia (179), arguing that conversion of CAFs to a less activated state is possible. Therapeutic strategies that reverse activated CAFs to quiescence include treatments with all-trans-retinoic acid (ATRA) (75, 90, 103), minnelide (18, 56), and the vitamin D receptor agonist calcipotriol, which ameliorated liver and pancreas fibrosis and enhanced PDAC therapy (60, 266). Moreover, treatment with the angiotensin receptor II antagonist losartan has been shown to decrease TGF-β activation in αSMA+ CAFs, leading to a reduction in desmoplasia and an increase in drug delivery (42, 61) and immunotherapy efficacy (41). Clinical trials with losartan in combination with standard of care regimens for PDAC are ongoing and appear promising for the treatment of local-advanced disease (185, 203).

Some markers, such as FAP (84, 146, 172), fibroblast specific protein (FSP-1) (23, 47), and αSMA (221), have also been used to genetically or pharmacologically deplete CAFs. In particular, FAP has been targeted in preclinical therapeutic strategies with vaccines, antibodies, and chimeric antigen receptor (CAR) T cells. Similar to the genetic depletion of FAP+ CAFs, inhibition of CXCR4 by AMD3100 has been shown to increase T cell infiltration and ameliorate the efficacy of checkpoint inhibitors in PDAC (84). The CXCL12/CXCR4 axis has been also demonstrated to promote disease progression and immunosuppression in breast cancer (53, 218). In this context, abrogation of CXCR4 signaling in the CAFs has been shown to reduce the levels of fibrosis and αSMA+ cells, leading to vasculature normalization, a decrease in immunosuppressive cell populations, and an increase in immunotherapy efficacy (43).

Alternative anti-fibrotic strategies include the anti-histamine drug tranilast (223, 225), the glucocorticoid steroid dexamethasone (187, 268), the hormone relaxin (25, 29, 130), halofuginone (30, 72, 277, 278, 336), and metformin, which has been shown to reduce the levels of fibrosis in both aging and cancer (120, 194, 318). Additional clinically approved drugs for fibrotic and inflammatory conditions, such as idiopathic pulmonary fibrosis (138, 144, 238, 249), could also be repurposed for targeting CAFs and reducing the ECM.

The recent advances on CAF molecular and functional heterogeneity open new avenues for cancer treatment. Potential strategies to design new therapies would be either by targeting CAF-secreted tumor-promoting and immunosuppressive ligands, such as IL-6 (177, 331), LIF (269), and TGF-β (184, 290), or by inhibiting subtype-specific signaling pathways that would ablate a particular CAF population. Another strategy would be to design therapies that leverage the plasticity of CAFs and shift tumor-promoting CAF subtypes towards a quiescent or tumor-restraining phenotype. Finally, only combinatorial approaches that target CAFs, cancer cells, immune cells, and drug delivery may be successful as effective treatments.

The emerging evidence that cancer cells and CAFs share certain signaling pathways will guide the design of therapeutic approaches that target cancer cells without affecting the stroma in ways that could reduce drug efficacy. Moreover, new combinatorial therapies could be designed to target specific pro-tumorigenic functions that are common to both the epithelial and CAF compartment. For example, the IL-1 receptor antagonist anakinra, which has been shown to target PDAC cancer cells (334), could also target potentially tumor-promoting, immunosuppressive iCAFs, whose phenotype is driven by IL-1 signaling (24, 64, 273). However, the presence of overlapping signaling pathways across different cell types also represents a potential issue that should be considered in the design of combinatorial therapeutic approaches. For example, the use of JAK inhibitors, although effective in targeting cancer cells and inflammatory CAFs (24), could also target the proliferation and/or activity of cytotoxic T cells, which would be problematic in combination with immunotherapy strategies. Therefore, imaging, genetic, and immunohistochemical analyses will need to be incorporated in the design of clinical trials to determine the effects of chemotherapy and targeted drugs on the CAF subtypes and on the immune tumor microenvironment. Likely, only combinatorial therapeutic strategies addressing these issues and designed to target multiple tumor-promoting components of the TME will be successful.

VII. CONCLUSIONS: PRESENT CHALLENGES AND FUTURE DIRECTIONS

Throughout this review, we have highlighted the multiple levels of CAF molecular and functional heterogeneity and the outstanding questions that remain to be addressed (FIGURE 11). Valuable information will come from studies that focus on 1) comparing fibroblasts across different tissues of origin and normal, inflammatory, and malignant states; 2) optimizing in vitro co-culture models and generating lineage-tracing in vivo models to study the cell of origin and functions of CAFs; and 3) going beyond the description of CAF heterogeneity and identifying the dependencies of each CAF population for the development of subtype-specific therapies. Additionally, the refinement of CAF subtype-specific transcriptional signatures at single-cell resolution and the identification of these profiles in bulk RNA-seq analyses could represent a powerful tool to predict the course of the disease and instruct patient treatment.

FIGURE 11.

Unanswered questions in cancer-associated fibroblast (CAF) biology. Schematic illustration of key areas in CAF biology that remain underexplored.

A more methodical standardization of the study of CAFs is also needed. One of the challenges in defining CAF heterogeneity by scRNA-seq analysis is that population subclustering can be arbitrarily defined and is limited by the number of samples and cells analyzed. An additional issue is the fact that CAFs are difficult to isolate, as they are embedded in the ECM. Therefore, CAFs are usually under-represented in scRNA-seq data sets, and tissue-specific protocols for CAF isolation should be developed (305). Although many dissociation protocols have been described, little justification has been given for the choice of enzymes and conditions employed. This variety of protocols further complicates the analysis of scRNA-seq data sets, since artefactual disaggregation-associated signatures (i.e., transcriptomic changes associated with a specific dissociation protocol) have been identified and can further confound the data interpretation (299). More gentle protocols may limit this problem, but likely to the detriment of the number of cells retrieved. The comparison of multiple dissociation protocols and sequencing platforms would allow the identification of transcriptional changes associated with the duration, temperature, and digestion conditions of a specific method (59). As an alternative approach, single-nucleus sequencing strategies (102, 114) could assist in the analysis of CAFs.

Additionally, the presence of similar CAF populations across different cancer types (FIGURE 3 and TABLE 2) suggests the need to define a common nomenclature of CAFs. Reaching this consensus is particularly important when multiple scRNA-seq data sets exist for the same cancer type (20, 24, 64, 73, 111, 258). Moreover, the presence of similarities between fibroblasts in normal, inflammatory, and malignant states highlights the need to distinguish between transcriptional signatures that are unique to CAFs or common to fibroblast populations in multiple contexts.

A paradigm-shifting approach that expands our therapeutic target selection is necessary to tackle malignancies with dismal prognosis, such as PDAC. Accordingly, gaining a comprehensive view of the dynamics and heterogeneity of CAFs could identify new therapeutic vulnerabilities. The progress that will be made in the next few years towards the functional characterization of distinct CAF subtypes will determine whether combinatorial strategies targeting cancer cells and tumor-promoting fibroblast populations could provide a clinically actionable approach for cancer treatment.

GRANTS

The authors are supported by National Institutes of Health (NIH) Cancer Center Support Grant 5P30CA045508 and the Lustgarten Foundation, where D. A. Tuveson is a distinguished scholar and Director of the Lustgarten Foundation–designated Laboratory of Pancreatic Cancer Research. D. A. Tuveson is also supported by the Cold Spring Harbor Laboratory and Northwell Health Affiliation, the Cold Spring Harbor Laboratory Association, and NIH Grants 5P30CA45508, U01CA210240, R01CA229699, U01CA224013, 1R01CA188134, and 1R01CA190092. G. Biffi was a fellow of the Human Frontiers Science Program (LT000195/2015-L) and EMBO (ALTF 1203-2014) and is supported by Cancer Research UK core funding (A27463).

DISCLOSURES

D. A. Tuveson reports receiving commercial research grants from Fibrogen and ONO, has ownership interest (including stock, patents, etc.) in Leap Therapeutics and Surface Oncology, and is a consultant/advisory board member for Leap Oncology and Surface Oncology, Cygnal, and Merck. G. Biffi has no conflicts of interest, financial or otherwise.