TGF-β Signaling in Health and Disease

Role of integrins.

The possibility that members of the integrin family of cell adhesion receptors activate latent TGF-β was first suggested by the phenotype of mice lacking the β6 subunit of the integrin αvβ6. αvβ6 is highly induced on epithelial cells in several organs by tissue injury and inflammation.³² Integrin β6 (Itgb6) knockout mice develop exaggerated inflammatory responses in the lungs and skin but are protected from tissue fibrosis in multiple tissues.^33-35 Both phenotypic features are consistent with a deficit in TGF-β. Cells expressing αvβ6 can activate TGF-β1 and TGF-β3 by binding the sequence RGD that is present in an exposed loop of their LAP.^34,36 Integrin αvβ8, which is expressed in neuroepithelial cells, astrocytes, and in subsets of myeloid cells, T cells, epithelial cells, and fibroblasts, binds to the same RGD site and can also activate TGF-β1.²⁷ Mice lacking Itgb8 die during embryonic development or immediately after birth from intracerebral hemorrhage caused by defective vascular development in the central nervous system.³⁷ Intracerebral hemorrhage in these mice is caused by the absence of αvβ8 on neuroepithelial cells which activates TGF-β for presentation to endothelial cells.³⁸

The integrins αvβ6 and αvβ8 are essential for many of the developmental and homeostatic roles of TGF-β1, as shown by knock-in of a point mutation in TGF-β1 which prevents integrin binding.³⁹ Administration of αvβ6 blocking antibody to Itgb8 knockout mice bred to bypass perinatal mortality, or crossing these mice to mice lacking Itgb6 recapitulates most of the developmental phenotypes of mice lacking TGF-β1 and TGF-β3.⁴⁰ These phenotypes include severe multiorgan inflammation (a central feature of TGF-β1 knockout mice) and cleft palate (seen in TGF-β3 knockout mice). These observations indicate that integrins αvβ6 and αvβ8 are crucial for TGF-β activity during development and immune homeostasis. Equivalent protection from hepatic and pulmonary fibrosis by deletion of the integrin αv subunit from fibroblasts or treating mice with a small molecule inhibitor to the αvβ1 integrin support the idea that αvβ1 is the main TGF-β activating integrin in fibroblasts.⁴¹ In contrast to the TGF-β1 and TGF-β3 LAPs, the TGF-β2 LAP lacks an RGD sequence but contains an alternate sequence in an exposed loop that can bind to αvβ6 for activation.⁴² TGF-β2 might also be spontaneously active after secretion.⁴³

Receptors.

TGF-β ligands bind to pairs of transmembrane serine/threonine protein kinase subunits known as receptors type I and type II. Mammalian genomes include 7 type I receptors and 5 type II receptors⁹ which are bound in various pairwise combinations by specific ligands (Table 1). In the case of TGF-β1, each monomer contacts one TGF-β type II receptor (TGFBR2) molecule forming a composite ligand-receptor protein surface that is then recognized by one type I receptor molecule (TGFBR1).⁴⁸ In the case of BMPs and Activins, each monomer contacts independent surfaces of the corresponding type I and type II receptors.^49,50 Thus, ligand binding results in the assembly of a hetero-tetrameric receptor complex bound by the dimeric ligand (Figure 2C). The TGFBR2 subunits then phosphorylate a Gly/Ser-rich region (GS region) situated near the kinase domain of the TGFBR1 subunits. Binding of the small protein FKBP12 to the GS region in the unliganded TGFBR1 locks the kinase activity in an inactive state.⁵¹ Once phosphorylated by TGFBR2, the GS region is thought to release FKBP12 and serve as a docking site for SMAD proteins as substrates of the TGFBR1 kinase.⁵² Numerous small-molecule kinase inhibitors have been developed against TGFBR1 and TGFBR2 that block all TGF-β responses.¹⁴

Co-receptors.

Co-receptors are crucial for binding of TGF-β and several other family members to the signaling receptors (Table 1). The core protein of the membrane-anchored proteoglycan betaglycan (also known as the type III TGF-β receptor) binds TGF-β for presentation to TGFBR2. ⁹ This step is particularly important for TGF-β2, which has low intrinsic affinity for the signaling receptors compared to TGF-β1 and TGF-β3.^53,54 The transmembrane protein endoglin is an essential co-receptor for BMP9 and BMP10. Other co-receptors are anchored to the cell surface via glycosylphosphatidylinositol tails, including the essential Nodal co-receptors Crypto and Cryptic, and Repulsion Guidance Molecules (RGM) as co-receptors for certain BMPs.

SMAD transcription factors.

SMAD transcription factors are direct substrates of type I receptor kinases (Figure 2C). SMAD proteins consist of N-terminal (or MH1) and C-terminal (or MH2) globular domains connected by a flexible linker region.^9,10 The N-terminal domain binds to DNA whereas the C-terminal domain includes sites for SMAD interaction with type I receptors, receptor adaptor proteins, other SMADs, nucleocytoplasmic shuttling factors, DNA binding cofactors, histone acetylases such as p300 and CBP, and chromatin remodeling proteins. The type I receptors for the TGF-β subfamily primarily phosphorylate SMAD2 and SMAD3, whereas those for the BMP subfamily phosphorylate SMADs 1, 5 and 8, with crossover SMAD signaling occurring in certain contexts. These five SMAD proteins are called “receptor-regulated SMADs (R-SMADs).

In the basal state, R-SMAD proteins shuttle between the cytoplasm and the nucleus. Receptor-mediated phosphorylation targets two serine residues at the C-terminus. The resulting pSer-X-pSer-carboxyl group mediates SMAD-SMAD binding for the assembly of trimeric complexes with SMAD4 (R-SMAD–R-SMAD–SMAD4 complexes). SMAD4 is not a receptor substrate nor is it required for R-SMAD nuclear translocation, but it is an essential participant in most SMAD-mediated transcriptional responses. The specific function served by SMAD4 remains unknown. SMAD6 and SMAD7 are inhibitory SMADs which antagonize SMAD4 and the type I receptors, respectively. TGF-β, BMP, interferon-γ, and other signals induce the expression of SMAD7 for negative feedback and antagonistic crosstalk in the pathway.⁵⁵

Receptor-activated SMAD complexes bind to hundreds of genomic loci in the nucleus, where SMADs undergo phosphorylation at the linker region by the RNA polymerase II (RNAPII) kinases CDK8 and CDK9.⁵⁶ This stimulates the transcriptional activity of SMAD complexes while leading to further phosphorylation of the linker by glycogen synthase kinase 3β. GSK3β creates binding sites for the HECT domain ubiquitin ligases SMURF1/2 (in SMAD1 and 5) and NEDD4L (in SMAD2 and 3) to mark the activated SMADs for degradation.^56-58 Alternatively, SMADs undergo dephosphorylation by SCP1/2 phosphatases⁵⁹ and dissociation by poly(ADP-ribose) polymerase-1 (PARP-1) mediated ribosylation, to be recycled for new rounds of signaling.⁶⁰ The regulation of SMADs and RNAPII by related kinases and phosphatases suggests a close coordination in SMAD-dependent activation of RNAPII transcription. The linker region of SMADs is also phosphorylated in the cytoplasm by mitogen-activated protein kinases (MAPKs) and cell cycle CDKs.^61,62 Additional regulators of the TGF-β pathway include decoy receptors, mediators of SMAD nucleocytoplasmic shuttling, transcriptional co-activators and co-repressors, non-coding RNAs, and ubiquitination-based receptor and SMAD turnover¹² (Figure 2C).

Pathway conservation and mutation.

X-ray crystal structures for ligands, receptors, and SMAD proteins have provided a wealth of insights into the function and specificity of the TGF-β pathway, including the steps of latent TGF-β activation, ligand interactions with traps and receptors, and the interaction of SMAD with receptors, regulators, DNA, and DNA-binding cofactors.^9,10,50,63

TGF-β and BMP ligands, receptors, co-receptors, and SMAD proteins, as well as the dichotomy of these two subfamilies are highly conserved across metazoans. The functional complementarity between the two subfamilies is manifest in many contexts, for example, in primordial germ cell development,⁶⁴ hair follicle progenitor differentiation,⁶⁵ and epithelial-mesenchymal transitions (EMT).⁶⁶ Although TGF-β gave name to the entire gene family, TGF-β is restricted to deuterostomes (vertebrates, crinoids, and sea stars) whereas BMPs and Activins are present across all metazoan phyla.⁶³ TGF-β pathway agonists have also emerged by convergent evolution. The rodent intestinal parasitic helminth Heligmosomoides polygyrus encodes a structurally unrelated TGF-β mimic (Hp-TGM) which binds to TGF-β receptors in host T cells to suppress immune attack.⁶⁷

The central components of the TGF-β system are essential for mammalian development, as shown by gene knockouts in mice. There is no gastrulation without Nodal, and deletion of TGF-β receptors, SMAD2, or SMAD4 is embryonic lethal. As we discuss below, loss-of-function somatic mutations in TGFBR1, TGFBR2, SMAD2, SMAD3 and SMAD4 are frequent in certain types of cancer, and inherited SMAD4 mutations cause a juvenile polyposis and hemorrhagic telangiectasia syndrome with propensity to intestinal cancer. Immune dysregulation, fibrosis, and cancer are the most common diseases involving somatically mutated or otherwise altered TGF-β signaling in the adult, and hence are the focus of this review. Notably, inherited mutations in the TGF-β system in human are the cause of rare if devastating diseases of skeletal, connective, and cardiovascular tissues in human (Table 2). Inherited mutations in TGFB1 encoding a hyperactive TGF-β1 variant cause Camurati-Engelmann disease, a debilitating syndrome characterized by abnormally thick skull and limb bones, joint deformities, and spine curvature.⁶⁸ Inherited and occasionally spontaneous mutations in TGFB2, TGFB3, TGFBR1, TGFBR2, and SMAD3 cause the five known types of Loeys-Dietz aortic aneurysm syndrome, which is characterized by multiple connective tissue alterations and is highly variable in penetrance, age of manifestation, and severity of the symptoms. These alterations include craniofacial anomalies (e.g. premature skull bone fusion, hypertelorism, bifid uvula), deformities in spine, chest and foot bones, osteoarthritis, a brittle skin prone to bruising, and, most ominously, an enlarged aorta prone to bulging (aneurysm) and rupture.⁶⁹ Paradoxically, while the mutant alleles causing Loeys-Dietz syndrome encode functionally weakened protein products, the affected tissues show heightened TGF-β signaling activity perhaps resulting from an imbalance in the regulation of other branches of the TGF-β family in these tissues.

Dendritic cells.

Various subsets of dendritic cells (DCs) play central roles in antigen presentation to CD4⁺ T cells (by DC2 dendritic cells) and CD8⁺ T cells (by DC1 dendritic cells) for priming of cytotoxic effector functions as well as regulation of the balance between T helper (T_H) and regulatory T cells (T_reg)¹¹² (Figure 3). TGF-β regulates the function of these DC subsets. Deletion of Tgfbr2 from DCs in mice leads to multiorgan inflammation and death by 15 weeks of age.¹¹³ DCs lacking TGFBR2 express normal levels of MHC class II and costimulatory molecules but produce more interferon γ (IFN-γ), which reduces their ability to induce T_reg cells. Adoptive transfer of wild type T_reg cells, or inhibition of IFN-γ each only partially rescue this phenotype, suggesting that additional mechanisms are also at play. TGF-β is also important for the development of a subset of skin DCs called Langerhans cells. Targeted deletion of Tgfbr2 of Tgfb1 by langerin-Cre prevents the development of Langerhans cells.¹¹⁴ In contrast to the role of DCs as T cell activators, a subtype of RORγt ⁺ antigen-presenting cells, called thetis cells (TC), induce pT_reg differentiation in intestinal lymph nodes during early life and require TGF-β-activating integrin αvβ8 for intestinal pT_reg differentiation. Loss of Itgb8 in these cells causes colitis, suggesting that this population plays an essential role in tolerogenic antigen presentation.¹¹⁵

T helper cells.

TGF-β plays fundamental roles in regulating and balancing the differentiation of naïve T cells into specific effector subsets (Figure 3). CD4⁺ T helper (T_H) cells support the development and function of CD8⁺ effector T cells and fall into two subtypes distinguished by their driving transcription factors and secreted cytokines. T-bet and STAT4 drive the differentiation of naïve CD4⁺ into T_H1 cells, which produce IFN-γ and IL-2, and support CD8⁺ cytotoxic T cells and macrophages. TGF-β-SMAD signaling potently inhibits T_H1 differentiation through coordinated effects on at least three levels: inhibition of IL-12 receptors required for T_H1 differentiation, inhibition of the expression of T-bet and STAT4; and, inhibition of IFN-γ production by natural killer (NK) cells, thereby interfering with a positive feedback loop through which NK cell-derived IFN-γ amplifies T_H1 differentiation.¹¹⁶ T_H2 cells are driven by GATA3 and STAT6, and produce IL-4, IL-5 and IL-13 to support B cells and other effector cells. TGF-β inhibits differentiation of T_H2 cells by down-regulating GATA3,¹¹⁷ and inhibiting GATA3 function indirectly by inducing the expression of SOX4.¹¹⁸ Balanced inhibition of both T_H1 and T_H2 cells by TGF-β is important. Although tissue inflammation and damage due to loss of TGF-β signaling in T cells is predominantly mediated by T_H1 cells, a loss of the T_H1-inducing T-bet led to multiorgan inflammation associated with enhanced T_H2 cell differentiation.¹⁰⁷ Disabling TGF-β signaling in CD4⁺ cells in mammary tumor-bearing mice augmented IL-4 production in T_H2 cells (but not IFN-γ production in T_H1 cells) leading to tumor regression.¹¹⁹

One exception to the general rule of TGF-β as an inhibitor of adaptive immune responses is that TGF-β-activated SMADs cooperate with RORγt to induce the T helper 17 (T_H17) phenotype.¹²⁰ T_H17 cells, are important in immune responses to bacteria and fungi and in the development of autoimmunity. Mice lacking TGFBR2 on T cells and mice lacking the TGF-β activating integrin αvβ8 on DCs have reduced numbers of T_H17 cells and are protected from developing Experimental Autoimmune Encephalomyelitis (EAE), a disease model that depends on T_H17 cells.¹²¹ TGF-β-blocking antibody also protects against EAE while overexpression of active TGF-β by T cells exacerbates CNS inflammation and EAE.¹²²

Regulatory T cells.

T_reg cells are a subset of T cells that suppress immune responses to enforce tolerance.¹²³ T_reg cells perform this role through multiple specialized effects on T helper and effector cells. The transcription factor Foxp3 drives differentiation of naive CD4⁺ T cells into T_reg cells. There are two major subtypes of T_reg cells: natural T_reg (nT_reg), produced in the thymus in early life, and T_reg derived from naïve CD4⁺ T cells in the periphery (pT_reg).

TGF-β positively regulates T_reg differentiation and activity. TGF-β enhances survival of nT_reg cells by suppression of proapoptotic proteins and upregulation of the antiapoptotic protein Bcl2.¹²⁴ TGF-β-activated SMADs cooperate with STAT5 and nuclear factor of activated T cells (NFAT) to induce expression of FOXP3 in naive CD4⁺ T cells¹²⁵ (Figure 3). Mice lacking TGF-β1 or TGFBR2 in T cells display marked reductions in FOXP3⁺ T_reg cell numbers in the periphery, consistent with a role for TGF-β in maintenance of these cells.^106,107,126 TGF-β can also promote retention of pT_reg cells in specific peripheral tissues such as the large intestine.¹²⁷ Mice lacking a particular Foxp3 enhancer that is required for TGF-β-mediated pT_reg induction do not develop the severe, early-onset multiorgan inflammation seen in mice lacking all T_reg cells, suggesting that nT_reg cells are sufficient to prevent this phenotype. However, older mice lacking pT_reg cells develop T_H2-mediated pathology in lung and intestine,¹²⁸ indicating that pT_reg cells do play important roles in controlling immune responses in some peripheral tissues.

Differentiation of naïve CD4⁺ T cells into pT_reg cells or T_H17 cells leads to drastically different effects on tissue inflammation. Upon initial sensing of TGF-β, naïve CD4⁺ T cells upregulate both Foxp3 (critical for T_reg differentiation) and RORγt (critical for T_H17 cell differentiation). Both the local concentration of active TGF-β and the presence of additional extracellular factors are important determinants of the commitment to either pT_reg or T_H17 cell fate. Low TGF-β concentrations inhibit the expression of the IL-23 receptor and favor Foxp3 expression, while high concentrations in conjunction with IL-6 upregulate the IL-23 receptor and favor RORγt expression and T_H17 cell induction.¹²⁰ Foxp3 inhibits RORγt function and prevents IL-17 induction, and this is counterbalanced by IL-6, IL-21, and IL-23 to facilitate the formation of T_H17 cells. Thus, the T_reg, T_H17, T_H1 and T_H2 states are tied to each other through a mutual regulation of their generation and function with TGF-β as a central balancing signal.

Cytotoxic T lymphocytes.

CD8⁺ T cells mature into effector T cells, also called cytotoxic T lymphocytes (CTLs), which eliminate cancer cells and pathogen-infected cells through the release of cytolytic mediators. TGF-β dampens the proliferation and cytolytic functions of CTLs (Figure 3).⁸ In vivo, expression of a CD2-driven or CD4-driven dominant-negative TGFBR2 construct, which reduces TGF-β signaling, leads to expansion of CD8⁺ T cells. However, complete loss of TGF-β signaling in T cells inhibits CD8⁺ T cell development,¹⁰⁷ likely due to the requirement for TGF-β to induce the IL-7 receptor on CD8⁺ T cells.¹²⁹ Thus, distinct effects of TGF-β signaling on CD8⁺ T cell induction, expansion and activation appear to be quantitatively regulated, with low levels of TGF-β signaling acting as an initiator of development in the thymus, but higher levels acting as a brake to inhibit inappropriate expansion and activation in the periphery.

Consistent with a role of TGF-β as a brake on excessive CD8⁺ T cell function, TGF-β suppresses multiple effector functions of cytotoxic T cells, inhibiting expression of perforin, IFN-γ and granzymes A and B through SMADs in partnership with ATF1.¹³⁰ Impairment of TGF-β signaling in T cells, or in vivo treatment with inhibitors of TGF-β activation have been consistently shown to enhance CD8⁺ T cell killing of tumor cells and to reduce tumor growth in multiple in vivo models. TGF-β signaling in CD8⁺ T cells is important in promoting apoptosis in short-lived effector cells.¹²² TGF-β also induces a specialized subset of CD8⁺ T cells residing within the single cell epithelial layer of the intestine that is important for the integrity of mucosal immune responses.¹³¹

Natural killer cells.

Natural killer (NK) cells are cytotoxic cells of the innate immune system. NK cells recognize target cells expressing NK cell chemotactic signals and NK receptor ligands upon viral infection or cancer-related genomic alterations. TGF-β blunts innate responses to viral infection and tumors through suppression of NK cell functions.^8,111 In addition to inhibiting IFN-γ production by NK cells, TGF-β inhibits the expression of cell-surface receptors NKG2D and NKp30, which NK cells use to recognize and kill stressed and malignant cells.^132,133 TGF-β also induces the expression mir-183 in NK cells, which reduces expression of the adaptor protein, DAP12, thus inhibiting responses to cytotoxic NK receptors including NKG2D.¹³⁴ TGF-β further suppresses NK cell activity by inhibiting activating responses to IL-15.¹³⁵ Moreover, TGF-β can contribute to immune evasion by inducing trans-differentiation of NK cells into type 1 innate lymphoid cells, which are not cytotoxic.^136,137

Fibroblast activation.

In response to tissue injury, fibroblast subsets undergo profound changes in gene expression that either enhance or inhibit tissue inflammation, regeneration, and scarring.^145,146 Normally, fibroblast activation results in short term accumulation of fibrillar collagens and other ECM components, together with a coordinated regulation of epithelial, immune, and endothelial cells through immunomodulatory and angiogenic signals.¹⁴⁴ This initial response is followed by fibroblast apoptosis and removal of excess collagen to restore normal tissue architecture.¹⁴⁷

TGF-β is a potent activator of different fibroblast subsets (Figure 4A). In cell culture, TGF-β induces a highly contractile phenotype associated with the expression of α-smooth muscle actin (αSMA, also known as ACTA2), multiple ECM components, and the enzymes and chaperones required for ECM assembly.¹³ In this state, fibroblasts are often called “myofibroblasts”, although the expression of ECM proteins and contractile proteins is not highly correlated in vivo. Besides producing and assembling ECM, activated fibroblasts establish paracrine communication with epithelial cells, promote angiogenesis through production of vascular endothelial growth factor A (VEGFA), and mobilize local innate and adaptive immune functions through the secretion of chemokines.^144,148 Thus, TGF-β-activated fibroblasts are hubs of ECM production and remodeling and of regulatory signals for epithelial, immune, and endothelial cells.

Fibrotic effects through fibroblasts.

Tissue fibrosis results from exaggerated production of collagens and other components of the ECM by tissue resident fibroblasts, often coupled with a reduction in ECM degradation and recycling by these cells. TGF-β promotes fibronectin and collagen production by both mesenchymal and epithelial cells.¹⁵³ Injection of TGF-β1 into the skin or transgenic or adenovirus-mediated overexpression of TGF-β in the lung cause extensive tissue fibrosis.^154,155 TGF-β-blocking antibodies prevent fibrosis in the skin, liver, lung, and kidney.¹³ TGF-β additionally inhibits multiple secreted proteases that contribute to ECM protein degradation. Recent data from single cell RNA sequencing has identified a subset of fibroblasts that emerge in many tissues in the setting of pathologic fibrosis and are characterized by the highest levels of expression of genes encoding fibrillar collagens and other components of the pathologic ECM.^145,146 TGF-β signaling is a major upstream regulator of the gene expression signature that characterizes these cells.

Feed-forward loops involving TGF-β often contribute to fibrosis pathogenesis by exaggerating normal physiologic responses, driving their chronic persistence, and triggering inflammation (Figure 4A). By increasing both ECM production and collagen cross-linking, TGF-β increases tissue stiffness which in turn favors increased collagen production and expression of contractile proteins.¹⁵⁶ Activated TGF-β can drive further expression of TGF-β in both autocrine and paracrine fashions. TGF-β is also a potent inducer of the TGF-β activating integrin, αvβ6.¹⁵¹ Furthermore, fibroblasts migrate toward and accumulate at regions of increased stiffness, a process that has been termed durotaxis.¹⁵⁷ Increased stiffness facilitates TGF-β activation through αvβ6 on epithelial cells and αvβ1 on fibroblasts, since both activate TGF-β through contraction-dependent effects on the conformation of the latent complex and this process is facilitated when cells are tethered to a stiff substrate.⁴⁴

Regulation of phenotypic plasticity.

Phenotypic plasticity refers to the ability of biological systems to change morphology and function in response to environmental and developmental cues. Progenitor cells are adept at responding to such cues during development and injury. TGF-β profoundly influences the phenotypic plasticity of epithelial progenitors, regulating their differentiation and phenotypic transitions during tissue development, morphogenesis, and repair. TGF-β frequently exerts these effects in counterbalance with other inputs, principally from the WNT, BMP, and RAS pathways (Figure 5A). For example, during early development, Nodal-activated SMAD transcriptional complexes and WNT-activated TCF complexes bind to shared target enhancers, activating mesendoderm specification transcription factors.^161,162 Postnatal development and adult homeostasis of epithelial tissues provide numerous examples of progenitor differentiation under the control of TGF-β in combination with WNT, BMP, and RAS signaling, such as in mammary ductal differentiation and branching morphogenesis during puberty, pregnancy and lactation; and in lung and kidney morphogenesis, liver regeneration; and intestinal epithelium homeostasis.^163-166

Epithelial-mesenchymal transitions.

Another manifestation of phenotypic plasticity is the ability of epithelial progenitors to undergo EMTs. EMTs play critical roles during development, injury repair, and disease.^167,168 In an EMT, epithelial cells lose apicobasal polarity and adhesive contacts while gaining actin stress fibers, anteroposterior polarity, motility, and remodeled contacts with neighboring cells and the ECM. EMTs are driven by transcription factors (EMT-TFs) including the zinc-finger proteins Snail (encoded by SNAI1), Slug (SNAI2), ZEB1 and ZEB2, the basic helix-loop-helix proteins Twist1 and Twist2, among others. EMT-TFs cooperatively repress epithelial genes and induce mesenchymal markers. The extent of the mesenchymal traits gained by a cell during an EMT – that is, the “completeness” of an EMT – depends on the range of phenotypic states that a particular epithelial progenitor is programed to access. After undergoing an EMT, cells can revert to an epithelial state through a mesenchymal-to-epithelial transition (MET). However, epithelial progenitors may undergo differentiation during an EMT–MET cycle, emerging from it in a distinct developmental stage.

EMTs are triggered by cell-extrinsic signals, TGF-β being the most widespread and potent of these.¹⁶⁷ TGF-β induces EMTs in epithelial cells in mammary, pulmonary, renal, hepatic, and other tissues during development, injury repair, fibrosis, and cancer, whereas Nodal drives EMT in epiblast cells during gastrulation.¹⁴ To trigger an EMT, signal-activated SMADs induce the expression of SNAI1/2 and ZEB1/2 to repress epithelial junction proteins such as E-cadherin, occludin and claudin-3, and of epithelial transcription factors such as KLF5.

Developmental and regenerative EMT programs.

TGF-β triggers EMTs as part of broad programs that include coordinated changes in cell proliferation, differentiation, and survival.^101,169 In mouse epiblast progenitors, Nodal induces the expression of EMT-TFs and mesendoderm specification transcription factors coordinating EMT and differentiation during gastrulation.¹⁷⁰ In adult mammary cells and in lung, breast, and pancreatic carcinoma cells, TGF-β induces the expression of EMT-TFs (e.g. Snail) and fibroblast-activating cytokines (e.g. IL-11, PDGFB), thereby coupling EMT and fibrogenesis.¹⁰¹ Thus, EMTs induced by TGF-β are associated with multiple programs and outcomes in different contexts: mesendodermal differentiation in epiblast progenitors and fibrogenesis in adult epithelial cells and carcinoma cells.

In all these cases, the effects of TGF-β depend on RAS-MAPK activity.^82,171-174 The RAS effector RREB1 (RAS-responsive element binding protein 1) plays a central role in this process.¹⁰¹ MAPK-phosphorylates RREB1 in the N-terminal domain to enable its binding to cognate DNA sites in target loci. RREB1 target loci include EMT-TF genes and either mesendoderm specification genes in epiblast cells or fibrogenic genes in adult epithelial and adenocarcinoma cells, depending on cell-specific chromatin accessibility patterns. The MAPK-activated, pre-bound RREB1 then enables TGF-β-activated SMADs to drive expression of these target genes (Figure 5B). RREB1 functions both as a nexus between the TGF-β-SMAD and RAS-MAPK pathways and as a link between EMT and developmental or fibrogenic gene expression programs depending on the cell context. Why these SMAD target genes and not others require RREB1, and what function RREB1 provides to enable transcription of these genes remain open questions.

Tumor suppressive apoptosis.

Conditional ablation of Tgfbr2 or Smad4 in pancreatic, intestinal, oral mucosa, and skin epithelia in mice does not interfere with the development of these tissues and causes only a mild hyperplasia. However, loss of Tgfbr2 accelerates tumor progression when these tissues harbor KRAS or HRAS oncogenes.^{83,185,188,189} Genetically engineered mouse models of pancreatic ductal adenocarcinoma (PDAC)^83,188 and CRC^190-192 showed that the TGF-β pathway interferes with the transition of pre-malignant cells to the carcinoma stage during malignant progression. In line with these observations, pre-malignant pancreatic, intestinal, and skin epithelial progenitors with KRAS or HRAS mutations undergo apoptosis in response to TGF-β, accounting for the tumor suppressive role in these cancer models.^82,83,185

TGF-β receptors are not directly coupled to apoptosis effector molecules. How TGF-β becomes a potent inducer of apoptosis was illuminated by studies in normal and malignant pancreatic epithelial progenitors.^82,101 In normal progenitors, TGF-β induces expression of SOX4, which pairs with KLF5 as a transcriptional partner to specify a pancreatic epithelial progenitor state. In pre-malignant pancreatic cells with mutant Kras, the dysregulated MAPKs strongly activate RREB1, which pairs with TGF-β-activated SMADs to trigger an intense induction of SNAIL expression and an EMT (Figure 5B). As a transcriptional repressor of epithelial genes, SNAIL inhibits KLF5 expression, driving SOX4 to activate the expression of Bim and other pro-apoptotic genes. Here, an otherwise normal TGF-β/RAS dependent EMT becomes pro-apoptotic owing to a conflict of cell fate signals: a pro-epithelial SOX4 and an overexpressed pro-mesenchymal SNAIL. Of note, RREB1 is frequently downregulated or genetically inactivated in PDAC.^193,194

Prevention of tumorigenic inflammation.

In addition to these direct tumor suppressive effects on pre-malignant cells, TGF-β acts as an indirect tumor suppressor by preventing inflammatory responses that cause cancer predisposition. TGF-β is a key suppressor of inflammatory responses through coordinated effects on different immune cell types.¹⁵ Disruption of this function can lead to chronic inflammation, which predisposes pre-malignant cells to progress to tumor formation. This is particularly evident in the gastrointestinal tract. TGF-β prevents intestinal inflammatory reactions to the commensal microbiota by exerting tolerogenic effects on immune and epithelial cells.¹⁹⁵ Dysregulation of TGF-β signaling is linked to the pathogenesis of ulcerative colitis, a cancer predisposition condition.¹⁹⁶ Tgfb1 mutant mice develop a lethal multifocal inflammatory disease as a prominent phenotype and are highly susceptible to developing colitis.^104,197 Mice with TGF-β pathway mutations in T cells or dendritic cells also show inflammatory phenotypes^8,111 and a higher incidence of gastrointestinal carcinomas.^198,199

Is growth inhibition tumor suppressive?

The cytostatic effects of TGF-β through the expression of CDK inhibitors and other factors in epithelial cells have been viewed as tumor suppressive effects. However, this notion is challenged by different lines of evidence. In mouse models of PDAC, the tumor suppressive effect of TGF-β is based on apoptosis, not growth arrest of KRAS-mutant progenitors.⁸³ In mouse models of HER2+ breast cancer, expression of a constitutively active Tgfbr1 transgene in mammary epithelial cells decreased the growth rate of mammary tumors but accelerated lung metastasis from these tumors.²⁰⁰ The growth inhibitory effects of TGF-β on epithelial cells are reversible and unlikely to provide an effective mechanism for the sustained suppression of tumor growth. In fact, disseminated breast cancer cells²⁰¹ and lung cancer cells²⁰² enter a growth arrested state in response to TGF-β in models of metastatic dormancy. As we discuss below, metastatic dormancy protects stem-like progenitor cells from immune surveillance, preserving these cells for eventual relapse. In effect, TGF-β-mediated immune evasive growth arrest is a strategy for metastatic progression in these models. Although the antiproliferative effects of TGF-β may dampen the growth of some tumors, there is no compelling evidence that these effects suppress tumor development or progression.

Immunosuppressive tumor microenvironments.

TGF-β has a multitude of effects on virtually all the adaptive and innate immune cell types and most of these effects enforce tolerance and prevent autoimmune responses. Not surprisingly, one after the other, these effects have been implicated in the generation of an immunosuppressive tumor microenvironment (TME) that favors tumor progression and metastasis and limits the effectiveness of immunotherapy.^8,15,111 Immune checkpoint inhibitors, which unleash endogenous anti-tumor immunity by inhibiting pathways that normally restrain CD8⁺ T cell-mediated tumor cell killing, have revolutionized treatment of a wide variety of cancers, but this approach remains ineffective for most patients. Approximately half of solid tumors can be characterized histologically as “immune excluded”, in which CD8⁺ T cells accumulate around the periphery of the tumor in a dense collagen-containing band but fail to infiltrate the tumor itself. These tumors are generally resistant to checkpoint inhibitors and exhibit a gene expression pattern suggestive of increased TGF-β signaling. Recent evidence from syngeneic murine models suggests that immune checkpoint sensitivity can be induced in some immune-excluded tumors by treatment with TGF-β inhibitors. This treatment promotes CD8⁺ T cell infiltration into the core of the tumor, increases granzyme B and interferon production by infiltrating T cells and in many cases leads to induction of long-term anti-tumor immunity.^192,214,215 Similar effects can be caused by inhibition of the TGF-β-activating αvβ8 integrin,^216-218 which can be expressed on tumor cells themselves or on CD4⁺ T cells in the tumors.^216-218 The mechanisms underlying synergy between checkpoint inhibitors and blockade of TGF-β signaling or activation are the focus of current investigation, and these pre-clinical results have led to active clinical trials evaluating TGF-β or αvβ8 integrin inhibitors in patients with cancer.

Tumorigenic effects through CAFs.

Fibroblasts present in tumors are called cancer-associated fibroblasts (CAFs). CAFs are an important component of the TME and influence tumor progression through ECM remodeling and paracrine signaling which are characteristic of activated fibroblasts. There are no specific markers distinguishing CAFs from the activated fibroblasts participating in wound healing or fibrosis. The consensus is that most CAFs result from the activation, likely dysfunctional, of local fibroblasts.¹⁴⁴ Distinct CAF populations distinguished by means of single-cell analysis exhibit either a TGF-β-driven matrix-producing contractile phenotype or an immunomodulating secretome.^219-221 Similar to TGF-β activated fibroblasts in wound healing and fibrotic diseases, CAFs are highly effective at producing and remodeling ECM within the TME. This process generates intra-tumoral fibrosis and ECM stiffness, known enhancers of carcinoma cell growth and invasion.^222-224 Additionally, CAFs are a source of IL-6 and TGF-β itself, which are immunosuppressive in the TME, and VEGF which drives tumor angiogenesis.¹⁴⁴ Beyond these effects on CAFs in carcinomas, TGF-β has been shown to promote growth of other tumor types such as gliomas through the induction of autocrine and paracrine secretomes.²²⁵

Metastatic dissemination.

In cancer cells that retain a functional TGF-β pathway, TGF-β can induce stimulate migration into, and out from blood capillaries for metastatic dissemination (Figure 6). Induction of an EMT by TGF-β-SMAD signaling accompanied with increased motility, invasiveness, and metastasis has been noted in HRAS-driven cutaneous squamous cell carcinomas induced by carcinogens²²⁶ or by genetic engineering in mice.²²⁷ TGF-β signaling in cancer cells in an orthotopic mammary tumor model promoted local invasion and hematogenous dissemination by inducing EMT and a switch from cohesive cell motility to single-cell motility.²²⁸ TGF-β can also function as a promoter of extravasation of circulating tumor cells. The activation of TGF-β stored in blood platelets coating colon cancer cells induced EMT and extravasation of cancer cell in the lungs of mice.²²⁹ To be determined is Whether TGF-β-induced EMTs promote extravasation by maintaining circulating carcinoma cells in an EMT state or by inducing an EMT in circulating epithelioid carcinoma cell clusters that lodge in capillaries remains to be determined. Additionally, TGF-β-rich breast primary tumors release cancer cells expressing angiopoietin-like 4, a TGF-β-inducible mediator of extravasation and lung metastasis.^230,231

Immune evasive dormancy.

Disseminated cancer cells suffer extensive attrition due to immune attack, physical barriers, and metabolic stresses.²³² Metastasis typically develops after a dormancy period lasting from months to decades, implying that cancer cells that survive the stress of dissemination enter a period of dormancy.²³³ In mouse models of metastatic dormancy, disseminated tumor cells localize to perivascular regions where they are able to remain dormant for many months.^201,234 Developmentally, these cells correspond to a stem-like early progenitor stage (e.g., SOX2+ stage in the case of LUAD) and are primed to enter quiescence in response to TGF-β and autocrine WNT inhibition.^202,234,235 During the dormant phase, these metastasis-initiating progenitors are in equilibrium between an immune-privileged quiescent state and a proliferative state liable to immune-mediated clearance (Figure 6). Quiescent cells downregulate MHC class I molecules,^236,237 NK receptor ligands²³⁴ and STING (stimulator of interferon genes)²⁰² to evade elimination by the immune system. Dormant cancer cells reentering the cycle re-express these mediators of immune recognition and consequently are cleared by the combined action of T cells and NK cells. Metastatic outbreaks succeed when proliferative clones elude immunity or when immune surveillance subsides.²⁰² Entry into proliferative quiescence in response to TGF-β appears to be a specific property of early-stage malignant progenitors and not of their developmentally more advanced progeny that constitute the bulk of the tumor mass. Thus, TGF-β-induced dormancy may protect disseminated metastatic progenitors from immune surveillance, preserving these cells for long-term relapse. Interestingly, an immune evasive state is also manifest in normal stem cells in hair follicles and muscle, which remain quiescent, but not in intestinal, mammary, or ovarian proliferative stem cells.²³⁸ Evolutionarily, immune evasive quiescence of adult stem cells may serve to protect longevity of these cells as they accumulate neoantigens during the organism’s lifespan.

Targeting TGF-β ligands and receptors.

Many of the molecules targeting TGF-β or its receptors are highly effective at blocking TGF-β signaling. However, obstacles need to be overcome to realize the therapeutic potential of these strategies. Some obstacles relate to specificity. The small-molecule inhibitors of TGF-β receptors primarily target the ATP-binding pocket of the receptor protein kinase domain. As a result, these inhibitors present the challenge of achieving specificity for TGF-β receptors versus other members of this receptor kinase family. Other obstacles relate to pleiotropy. Because of the many critical roles TGF-β family members play in normal development and tissue homeostasis, potent inhibition of all TGF-β signaling would likely lead to unacceptable toxicity. This possibility is underscored by the embryonic or early post-natal lethal phenotypes of mice with inactivating mutations in each of the three mammalian TGF-β isoforms, and by the severe auto-immune phenotypes of mice with severely impaired TGF-β signaling in T cells or dendritic cells. These dramatic effects might all be consequences of critical developmental roles for TGF-β signaling and do not necessarily preclude treatment of adults with TGF-β inhibitors. Indeed, some of the TGF-β inhibitors currently under development have been given to hundreds of patients without signs of severe toxicity.^15,225

Nevertheless, monkeys, rats and mice have developed thickening of cardiac valves and some patients treated with TGF-β inhibitors developed low grade skin cancers, consistent with the known effects of TGF-β in valve development and as a brake on epithelial carcinogenesis, raising concerns about the safety of this approach. Furthermore, because latent forms of each TGF-β isoform are expressed at high concentrations in many tissues of healthy adults, antibodies or other biologics targeting TGF-β isoforms could be limited in effectiveness because of the likelihood they would be sequestered by irrelevant TGF-β tissue stores. This might be one reason for the apparent lack of serious toxicity for several anti-TGF-β biologics that entered clinical trials thus far and for their surprisingly limited efficacy. Cell-permeable small-molecule inhibitors of TGF-β receptor kinases could overcome the challenges of large extracellular stores of latent TGF-β, but those developed so far have generally had quite short in vivo half-lives and would not, in any case, overcome the challenges of the many roles TGF-β signaling plays in maintaining normal tissue homeostasis.

Targeting TGF-β activation.

Renewed confidence on the viability of therapeutically targeting TGF-β comes from the increasing knowledge about the specific mechanisms of TGF-β activation, and the ability to target TGF-β inhibition to a specific cellular context or specific downstream responses. Several strategies to get around these problems are in various stages of development, most aimed at increasing the precision for inhibiting specific pathologic functions of TGF-β without targeting beneficial homeostatic effects. One such strategy has focused on the pathways for integrin-mediated TGF-β activation. As noted above, three integrins, αvβ1, αvβ6 and αvβ8 are involved in the activation of latent TGF-β stored in the ECM or on cell surface.^27,34,41 TGF-β can also be activated by integrin-independent pathways, such as binding to thrombospondin, narrowing the scope, and thus the potential toxicity of therapeutic targeting. Importantly, each of these integrins is expressed in a distinct, limited number of cells and at low copy number, further overcoming the challenges of indiscriminately targeting all TGF-β isoforms in every tissue. For example, αvβ6 is restricted in its expression to epithelial cell,³² and is generally expressed at very low levels in healthy epithelia but dramatically upregulated in response to injury and at sites of fibrosis.²⁴⁸ Currently, there are at least three drugs targeting TGF-β activating integrins in active clinical trials: a dual small molecule inhibitor of αvβ1 and αvβ6 in phase 2 clinical trials for treatment of pulmonary fibrosis and primary sclerosing cholangitis, a small molecule inhibitor of avβ1 in a phase 1 trial for liver fibrosis in the setting of non-alcoholic steatohepatitis, and a humanized monoclonal antibody targeting the αvβ8 integrin in a phase 1 study for enhancing responses to immune checkpoint inhibition in cancer. All these studies are in early stages, but recently released data for a 12-week phase 2 study of the αvβ1/αvβ6 inhibitor in 67 patients with idiopathic pulmonary fibrosis showed no significant on-target toxicity and an apparent dose dependent efficacy in slowing the rate of loss of lung function and progression of radiographic evidence of fibrosis. However, other drugs targeting TGF-β activating integrins have been withdrawn because of pre-clinical or clinical adverse events, so conclusions about the safety and efficacy of this approach will need to await the results of longer and larger clinical trials.

Targeting downstream mediators.

Other strategies are based on more precisely targeting inhibition of TGF-β signaling to the cells that drive specific disease pathology. One encouraging example of this approach was the identification of epigallocatechin gallate (EGCG) as a natural product found in several fruits and green teas which protects mice from bleomycin-induced pulmonary fibrosis.²⁴⁹ Efforts to identify its mechanism of action showed that when EGCG covalently inhibits lysyl oxidase homolog 2 (LOXL2), the compound itself is converted into a potent, irreversible inhibitor of TGFBR2. Importantly, LOXL2 is not broadly expressed, with substantial expression in normal tissues restricted to fibroblasts. EGCG is thus effectively a specific inhibitor of TGF-β signaling in fibroblasts. Biopsies from patients who received EGCG showed significant reductions in extractable collagen and other indicators of active fibrosis compared to patients treated with placebo.²⁵⁰

Several efforts are underway to treat TGF-β-driven diseases more precisely by targeting key steps downstream of TGF-β signaling that might contribute more to pathology than to normal homeostatic functions. One example is a monoclonal antibody that blocks connective tissue growth factor, a TGF-β-induced protein that has been proposed to mediate some of the pro-fibrotic effects of TGF-β.²⁵¹ Another is the development of drugs targeting nicotinamide adenine dinucleotide phosphate oxidase 4 (NOX4), a TGF-β-induced protein that catalyzes the reduction of molecular oxygen to hydrogen peroxide. NOX4 expression is increased in activated fibroblasts and inhibition of NOX4 reduces collagen production from these cells in vitro and inhibits pulmonary fibrosis in vivo.²⁵² Pulmonary fibrosis is a disease associated with aging and in vivo studies in mice suggest that NOX4 upregulation persists longer in fibroblasts from bleomycin-treated aged mice. Aged mice also exhibit more prolonged fibrosis after a single dose of bleomycin than young mice do, and inhibition of NOX4 inhibits fibrosis persistence in aged mice.²⁵³ Based on these findings, NOX4 inhibitors are currently under development for treatment of pulmonary fibrosis.

The authors thank J.H. Lee for assistance with formatting the manuscript. The primary research work by the authors on the topic of this review is supported by NIH grants R35-CA252978 (JM), P01-CA129243 (JM), R01-HL145037 (DS) and R01-HL142568 (DS), P30-CA008748 (MSKCC), and by the Alan and Sandra Gerry Metastasis and Tumor Ecosystems Center at MSKCC