Overcoming TGFβ-mediated immune evasion in cancer

Transforming growth factor-β (TGFβ) signalling controls multiple cell fate decisions during development and tissue homeostasis; hence, dysregulation of this pathway can drive several diseases, including cancer. Here we discuss the influence that TGFβ exerts on the composition and behaviour of different cell populations present in the tumour immune microenvironment, and the context-dependent functions of this cytokine in suppressing or promoting cancer. During homeostasis, TGFβ controls inflammatory responses triggered by exposure to the outside milieu in barrier tissues. Lack of TGFβ exacerbates inflammation, leading to tissue damage and cellular transformation. In contrast, as tumours progress, they leverage TGFβ to drive an unrestrained wound-healing programme in cancer-associated fibroblasts, as well as to suppress the adaptive immune system and the innate immune system. In consonance with this key role in reprogramming the tumour microenvironment, emerging data demonstrate that TGFβ-inhibitory therapies can restore cancer immunity. Indeed, this approach can synergize with other immunotherapies — including immune checkpoint blockade — to unleash robust antitumour immune responses in preclinical cancer models. Despite initial challenges in clinical translation, these findings have sparked the development of multiple therapeutic strategies that inhibit the TGFβ pathway, many of which are currently in clinical evaluation. This Review discusses the context-dependent functions of transforming growth factor-β (TGFβ) with regard to the composition and behaviour of different cell populations in the tumour immune microenvironment, as well as emerging data that demonstrate that TGFβ inhibition can restore cancer immunity.

The role of transforming growth factor-β (TGFβ) signalling during cancer progression is complex as it can have both tumour-suppressive and tumour-promoting functions [1][2][3][4] . Virtually all cell types are responsive to TGFβ, but its role has been particularly well characterized in epithelial cells. In organs such as skin, colon, breast or pancreas, TGFβ signalling regulates homeostatic growth, inhibiting cell proliferation and transformation during the early stages of tumorigenesis (Fig. 1). Cancers arising in these tissues can avoid the tumour-suppressive effects of TGFβ by acquiring inactivating mutations in pathway components. In other cases, tumour cells remain responsive to TGFβ during disease progression but, in crosstalk with several oncogenic alterations such as KRAS mutations, rewire the signalling pathway's outcome to promote epithelial-to-mesenchymal transition, dissemination, dormancy and metastasis (Fig. 1). The context-dependent roles of TGFβ signalling in healthy and tumorigenic epithelial cells have been reviewed elsewhere [2][3][4] .
Whereas research on TGFβ signalling in cancer has been predominantly tumour cell centric, pioneering studies on TGFβ signalling in the 1980s and 1990s did address the profound effects that this cytokine exerts on the tumour microenvironment (TME) 5 . These early studies showed that inoculation of mice with TGFβ accelerated wound healing by stimulating both the recruitment of immune cells and the production of multiple extracellular matrix (ECM) components by fibroblasts [6][7][8] . These findings were linked to a pivotal role for TGFβ in the differentiation of cancer-associated fibroblasts (CAFs), as well as to the generation of the desmoplastic reaction that characterizes many prevalent tumour types 9 , fuelling the notion that tumours are wounds that do not heal 10 . In parallel, TGFβ signalling was discovered to suppress the function of adaptive and innate immune cells [11][12][13][14] , a mechanism that a decade later was associated with cancer immune evasion 1,[15][16][17] .
We now know that TGFβ controls immune homeostasis in several tissues, and genetic defects in pathway components are linked to loss of immune tolerance and autoimmunity 18,19 . Moreover, in mouse models, exacerbated inflammation associated with the loss of TGFβ signalling in several immune cell types leads to enhanced cancer formation (Fig. 1). In contrast, as tumours progress, the levels of TGFβ increase, concurrent with marked remodelling of the TME (Fig. 1). Combined with the well-documented cancer cell-intrinsic effects of TGFβ on invasion and metastasis [2][3][4] , the net result is a systematic disposition

Desmoplastic reaction
The growth of fibrous tissue around the tumour.
Overcoming TGFβ-mediated immune evasion in cancer to tumour progression, immune evasion and therapy resistance (Fig. 1).
Here we describe how the complex cellular ecosystem of the TME responds to TGFβ throughout the evolution of the disease. We first summarize the basics of the TGFβ signal transduction pathway, emphasizing the mechanisms of TGFβ production, storage and release within the TME. We then review current knowledge of the role of TGFβ signalling in immune homeostasis and its link to tumour initiation in pathogen-exposed organs such as the gut. Subsequently, and forming the main focus of this Review, we discuss how TGFβ signals facilitate malignant tumour growth, dissemination and immune evasion by instructing gene programmes in different TME cell types. We conclude with the current translational and clinical efforts to block the TGFβ signalling pathway, recognizing a promising role of this strategy in immuno-oncology.

Regulation of TGFβ bioavailability
The three TGFβ isoforms, TGFβ1, TGFβ2 and TGFβ3, belong to a 33-member family of structurally related cyto kines known as the TGFβ superfamily 20,21 . These cytokines share many features, including structurally related receptors and downstream signalling effectors, yet they often play functionally distinct roles in physiology and disease 20,21 (Box 1). The TGFβ pathway has been extensively investigated, and several excellent reviews cover its mole cular biology 2,21,22 . As a reference, we summarize here the essential components and critical regulatory steps (Fig. 2). In essence, TGFβ triggers a classical membrane to nucleus signal transduction pathway whereby, upon binding to type I and type II TGFβ receptors (TGFBR1 and TGFBR2) at the cell surface, intracellular SMAD effector proteins translocate into the nucleus and activate transcriptional programmes. The specificity of SMAD-DNA binding and transcriptional regulation is a b  These tumour cells are blind to the action of the cytokine and can expand in a TGFβ-rich tumour environment (TME). Furthermore, during cancer dissemination, TGFβ can impose cell cycle arrest on tumour cells, which results in a dormant state, a phenomenon associated with metastatic latency, chemotherapy resistance and disease relapse. b | In the immune environment of healthy tissues, particularly in the gastrointestinal tract and the skin, TGFβ is necessary to induce tolerance and regulate responses to antigens of commensal bacteria. In this context, TGFβ is a potent suppressor of inflammation, and its lack triggers an excessive inflammatory response that predisposes to tumour formation. TGFβ is also necessary to regulate wound-healing responses. As cancer progresses, tumours hijack these TGFβ functions to promote immune evasion and a continuous wound-healing response. T reg cell, regulatory T cell; T RM cell, tissue-resident memory T cell.
www.nature.com/nrc achieved through the interaction of SMADs with both lineage-determining and signal-driven transcription factors 2 . As a result, TGFβ regulates specific transcriptional programmes depending on the cell type and context, which explains its diversity of roles in physiological and pathological processes 2 . This is particularly relevant in the TME, where TGFβ can instruct disparate gene programmes in each of the different cell types present. It is also important to note that whereas most TGFβ responses involve SMAD-driven transcription, several alternative (non-canonical) pathways can transduce TGFBR signals 21,22 (Fig. 2).

Production and storage of latent TGFβ
The TGFB1, TGFB2 and TGFB3 genes encode prohormones that include a large amino-terminal domain called the 'latency-associated peptide' (LAP) and a short carboxy-terminal domain that corresponds to the mature, bioactive cytokine 19 . In the Golgi complex, the TGFβ prohormone dimerizes through the formation of disulfide bonds and is subsequently cleaved by the protease furin. However, the bioactive and LAP portions that result from this cleavage remain non-covalently linked (Fig. 2). This conformation, known as latent TGFβ (L-TGFβ), impedes signal transduction because the LAP domain obstructs binding of the active portion of TGFβ to the receptors 23 . The LAP dimer is often crosslinked to L-TGFβ-binding proteins (LTBPs), which results in the formation of the large L-TGFβ complex 19 .
TGFβ can be found in the plasma of patients with cancer with poor prognosis 24-26 , suggesting that it can freely diffuse. However, large L-TGFβ complexes are generally retained by the ECM through interaction of LTBPs with, or crosslinking to, several glycoproteins, such as fibrillins 19,27 (Fig. 2). These act as reservoirs from which the active cytokine can be released in a tightly regulated manner. The relevance of these interactions is exemplified by the effect of germ line mutations in FBN1 (which encodes fibrillin 1), which are present in patients with Marfan syndrome. These mutations interfere with the retention of L-TGFβ in the ECM, and the resulting elevated levels of TGFβ signalling cause hypermobile joints, skeletal deformities and aortic aneurysms 28 . In particular cell types, newly synthesized L-TGFβ is not crosslinked to LTBPs but forms disulfide bonds with leucine-rich repeat-containing protein 32 (LRRC32; also known as GARP) 29,30 or with the related LRRC33 (reF. 31 ). After furin cleavage, GARP-bound or LRRC33-bound L-TGFβ is loaded onto the cell membrane, enabling spatially controlled TGFβ1 release and signalling (Fig. 2). GARP tethers L-TGFβ onto the surface of regulatory T cells (T reg cells), tumour cells, endothelial cells and platelets 29 , whereas LRRC33 plays an equivalent function in macrophages and microglia 31 .

Release of active TGFβ in the TME
Like TGFβ production and storage, its release is conducted by a variety of tightly regulated processes. Active TGFβ can be liberated from latent ECM complexes by proteolytic cleavage mediated by various serine proteases, such as plasmin or cathepsin D 19 and, particularly, by matrix metalloproteinases present in the TME 32,33 . The protease thrombin can also cleave GARP on the surface of platelets, releasing active TGFβ and contributing to tumour immune evasion 34 . However, mounting evidence suggests that the main mechanism of TGFβ release from latent deposits depends on integrin activity. In particular, the αVβ6 and αVβ8 integrins bind to an Arg-Gly-Asp (RGD) motif present in the LAP portion of L-TGFβ1 and L-TGFβ3 with very high affinity, which may reflect a specialized function of these integrin isoforms in TGFβ activation rather than cell adhesion or migration 35 . In this context, αVβ6 integrin translates tension resulting from actomyosin-mediated cell contraction on the L-TGFβ molecule, which results in the unfolding of the LAP domain and the release of the active hormone 23,36 (Fig. 2). This mechanical process is performed mainly by highly contractile cells such as cancer cells, myeloid cells and myofibroblasts, and it is facilitated by the tethering of L-TGFβ to stiff substrates through LTBPs 37-39 . In the context of cell surface-bound TGFβ, GARP operates as a chaperone that orients L-TGFβ for binding to αVβ8 integrin 36 . Of note, the cytoplasmic tail of αVβ8 integrin does not interact with the actin cytoskeleton and cannot transmit cell contraction forces onto the L-TGFβ molecule. Instead, αVβ8 integrin enforces a change of L-TGFβ conformation that enables activation of the TGFBRs while the ligand is still bound to GARP 40 . The pivotal role of αVβ8 integrin in the regulation of TGFβ availability is further supported by the analyses of mice with conditional deletion of the

Box 1 | TGFβ isoform-specific functions and therapies
Despite binding the same receptors, the transforming growth factor-β1 (TGFβ1), TGFβ2 and TGFβ3 isoforms exhibit distinct expression patterns and their bioavailability is differentially regulated. In particular, the TGFβ2 latency-associated peptide (laP) domain lacks the arg-Gly-asp (rGD) motif present in TGFβ1 and TGFβ3. as a consequence, latent TGFβ2 is not activated by αvβ6 or αvβ8 integrins 292 , implying the existence of specific mechanisms to release this isoform. Furthermore, Tgfb1-knockout, Tgfb2-knockout and Tgfb3-knockout mice show non-overlapping developmental defects. Global knockout of the Tgfb1 gene results in multifocal inflammatory disease owing to an important role of the corresponding isoform in setting immune tolerance 293,294 . Tgfb2-null mice exhibit a range of developmental abnormalities, including heart, lung, craniofacial, limb, spinal column, eye, inner ear and urogenital defects 295 . Tgfb3-knockout mice develop cleft palate 296,297 . The role of each isoform in the adult organism has been less well characterized as most pharmacological and genetic approaches used to investigate TGFβ functions during tissue homeostasis and disease disrupt signals from all three isoforms. Correlative analyses indicate that TGFβ1 shows higher and more widespread upregulation in the tumour microenvironment than the other two isoforms and is more robustly associated with failure of immune checkpoint inhibitor responses in patients with cancer 50 . In turn, TGFβ3 plays specific roles during wound healing and fibrosis 298 and is highly upregulated in cancerassociated fibroblasts 299 . a role for TGFβ2 in breast cancer cell dormancy has been demonstrated 300 . In addition, TGFβ2 has been associated with neutrophil recruitment in models of metastatic colorectal cancer 86 . of particular interest is the finding that adult Tgfb2-haploinsufficient mice phenocopy patients with germ line loss-of-function mutations in the TGFB2 gene, who develop thoracic aortic aneurysm dissections and other cardiovascular abnormalities 301 . This pathology is similar to that observed in preclinical models treated with pan-TGFβ inhibitors 50,211-213 , suggesting that adverse effects associated with these drugs are due to TGFβ2 inhibition in the cardiovascular system. These findings inspired the development of therapeutic antibodies that target specifically the TGFβ1 isoform and that avoid the cardiotoxicity associated with pan-TGFβ approaches in experimental models 50 . TGFβ ligand traps able to preferentially block TGFβ1 and TGFβ3 have also been engineered 275,276 , and integrin-targeted strategies are also selective for these two isoforms (TaBle 1). naTure revIeWS | CANCeR gene encoding the β8 subunit in dendritic cells (DCs), monocytes and macrophages, all of which develop loss of TGFβ-mediated immune tolerance and inflammatory pathology in barrier tissues (reviewed in reFs 19,41 ).
Both GARP-bound and LTBP-bound TGFβ are main sources of the cytokine in the TME, and cancer cells leverage integrin activity to regulate its bioavailability. Expression of αVβ6 integrin predicts poor prognosis in colorectal cancer (CRC), and its activity mobilizes TGFβ, inducing epithelial-to-mesenchymal transition in cell line models 42 . TGFβ released by tumour cells through αVβ8 integrins also facilitates immune evasion 43,44 . GARP is upregulated in breast, colon and lung cancers, and enforced expression of GARP in breast cancer cells increases TGFβ bioactivity and blocks antitumour responses through T reg cells 45 . In other cases, cells of the TME operate as TGFβ suppliers. As mentioned earlier, platelets carry L-TGFβ bound to GARP on the cell   58 . This study showed that TGFβ signalling negatively regulates the production of the proinflammatory cytokine interleukin-6 (IL-6) by lamina propria-resident CD4 + T cells. IL-6 signals to colonic epithelial cells and promotes their survival and proliferation in an inflamed environment, eventually resulting in dysplasia 58 . Similarly, deletion of Smad4 in mouse T cells also increases the production of several proinflammatory cytokines by T cells, including IL-6 and IL-11, and predisposes mice to spontaneous epithelial neoplasia throughout the gastrointestinal tract 59,60 .
In addition to its role in suppressing T cell responses, TGFβ signalling in stromal cells also contributes to limiting chronic epithelial inflammation. Mice with fibroblast-specific Tgfbr2 knockout develop prostatic intraepithelial neoplasia and invasive squamous cell carcinomas in the forestomach 61 . It was initially proposed that Tgfbr2-null fibroblasts increase the production of hepatocyte growth factor (HGF), which acts as a mitogen for adjacent mucosa cells 61 . However, an alternative mechanism linking TGFβ signalling deficiency in fibroblasts and the formation of epithelial neoplasia was later identified. It was found that tissues surrounding Tgfbr2-null fibroblasts are inflamed and show signs of DNA damage -possibly caused by reactive oxygen species and nitrogen radicals that occur during persistent inflammation. Indeed, the forestomach mucosa exhibits loss of genomic regions encoding the tumour suppressor genes Cdkn2b and Cdkn2a 62 . Moreover, this phenotype is delayed by treatment with anti-inflammatory drugs and is aggravated by Helicobacter pylori infection 62 . Overall, these observations imply that during barrier tissue homeostasis, TGFβ signalling in both T cells and fibroblasts is necessary to control inflammatory responses triggered by exposure to harmful antigens (Fig. 1). The lack of TGFβ results in exacerbated inflammation, leading to tissue damage and cellular transformation.

Innate immune evasion by TGFβ
During advanced stages of cancer, TGFβ plays a central role in the coordination of immune evasion (Fig. 1). In addition to fighting infectious diseases, the innate immune system possesses mechanisms to identify transformed cells. This is partly based on molecular recognition patterns. T he se danger-associated molecular patterns or pathogen-associated molecular patterns include many molecules that are released from damaged or dying cells, and activate inflammatory responses in a number of stromal cell types 63 . These processes occur in cancer and, in principle, can alert the immune system 64 . As discussed later, TGFβ signalling broadly attenuates this vigilance, generally skewing innate immunity towards tolerance or dysfunction (Fig. 3).

Macrophages and monocytes
Macrophages are abundant and highly plastic, phagocytic cells that can polarize into phenotypes that range across the inflammatory-anti-inflammatory spectrum, (2) When the TGFβ molecule is either covalently linked to LTBPs and tethered to the ECM or localized to the cell surface bound to GARP, αVβ6 integrin-transmitted tension generated by cell contraction releases active TGFβ. (3) αVβ8 integrin induces a conformational change that exposes active TGFβ without releasing it from the LAP domain. b | The active TGFβ dimer triggers type I TGFβ receptor (TGFBR1) and TGFBR2 dimerization. In this ligand-induced receptor complex, TGFBR2 phosphorylates TGFBR1, which in turn recognizes and phosphorylates SMAD2 and SMAD3 proteins, the cytoplasmatic mediators of TGFβ signalling. Phosphorylated SMAD2 and SMAD3 interact with SMAD4 to form a trimeric complex that travels to the nucleus. Dimers of phosphorylated SMAD2 and SMAD3 together with SMAD4 form complexes with different signal-driven transcription factors (SDTFs) and lineage-determining transcription factors (LDTFs) to regulate transcription of target genes. Several cofactors (CoFs) are also recruited to these transcriptional complexes. The stability of the nuclear SMAD complex is negatively controlled by poly(ADP-ribose) polymerase (PARP)-mediated PARylation, which causes dissociation of SMAD from DNA. c | The main non-canonical TGFβ pathway involves TRAF6 and TAK1 signals in combination with SMAD7 , and activates downstream kinases in the JNK, p38 and NF-κB pathways independently of SMAD-driven transcription. Other non-canonical pathways that activate mTOR, RHOA or KRAS signalling downstream of TGFBRs are also indicated. CAF, cancer-associated fibroblast; RGD, Arg-Gly-Asp.

Thymocyte selection
Thymocytes (differentiating T cells in the thymus) are positively selected for weak binding to MHC molecules, and negatively selected (killed) if they bind MHC or self antigens too strongly. These processes are influenced by TgFβ.
Danger-associated molecular patterns or pathogen-associated molecular patterns Molecules, molecular motifs or epitopes that are upregulated or exposed in the presence of pathogens or on damaged or dying cells. specialized receptors on innate immune cells recognize these signals and trigger an inflammatory response.
Genetic experiments in mice revealed that TGFβ signalling in myeloid cells, which include macrophages and monocytes, can promote tumour growth 72 , progression 73 and metastasis 74 . More recently, TGFβ was linked to increase programmed death ligand 1 (PDL1) expression on lung adenocarcinoma-associated macrophages 75 , which is consistent with our own finding of myeloid PDL1 involvement in a TGFβ-mediated immune evasion mechanism in CRC liver metastasis 76 . Furthermore, TGFβ1 induces the expression of miR-494 in myeloidderived suppressor cells, mediating their accumulation a TAM differentiation b Innate antitumour cytoxicity c Tumour antigen presentation d T cell polarization e CD8 + antitumour cytoxicity f CAF specification CXCL9, CXCL10 and CXCL11 • ↓ EOMES, ↓ TBET • ↑ Tissue residence • Tolerance or enhanced immunity?
• ↑ ECM deposition and remodelling • ↑ αSMA + • Immunosuppresive factors (e.g. TGFβ1 and TGFβ3) • T cell exclusion   in the TME and exacerbating their immunosuppressive functions 77 . TGFβ-induced microRNAs with effects on immunosuppression in cancer have been observed in multiple immune cell types 78 .

Granulocytes
Of all types of granulocytes, neutrophils have been studied most in the context of cancer, although intratumoural eosinophils have also been identified to produce TGFβ 79 . Neutrophils are highly prevalent, and upon infection are one of the first cell types to be recruited to eliminate pathogens and cause acute inflammation 80 . Given the overlap of some of the danger-associated molecular patterns and pathogen-associated molecular patterns involved in infection and oncogenic transformation, neutrophils can be recruited to, recognize and eliminate cancer cells 81 . However, increased numbers of infiltrating neutrophils as well as circulating neutrophils have been associated with a worse prognosis for most patients with cancer 81 , indicating that these cells commonly fail in their role in immunosurveillance. Accordingly, neutrophils have been ascribed tumour-supportive functions, mediated by exposure to signals in the TME [81][82][83] . Indeed, tumour-associated neutrophils can adopt a markedly protumoural polarization, sometimes called N2, mediated by TGFβ signalling 84 (Fig. 3b). Blockade of this pathway in mice induced the influx of proinflammatory, cytotoxic N1-like neutrophils, impinging on tumour growth 84,85 . Moreover, a recent study with a mouse model for serrated CrC with poor prognosis found that liver metastasis was driven by NOTCH1 through TGFβ2-mediated recruitment of neutrophils 86 .

Natural killer cells
Natural killer (NK) cells play a role in immunosurveillance 87,88 . The cytotoxic powers of NK cells are not indiscriminate, and are controlled by an array of cell surface receptors and regulatory pathways, by which NK cells can adapt to their environment 89 . This intricate regulatory balance can be exploited by the TME, leading to NK cell exhaustion, desensitization or exclusion. Stromal TGFβ can increase the expression levels of inhibitory cues on cancer cells such as non-classical major histocompatibility complex (MHC) molecules and immune checkpoint molecules [90][91][92][93] . Furthermore, TGFβ plays multiple roles in shaping NK cell anergy: it inhibits TBET (also known as TBX21), a transcription factor that drives IFNγ expression 94,95 , it regulates activating 96,97 or inhibitory 98 surface receptors and it represses NK cell metabolism and effector function 99,100 (Fig. 3b). Additionally, TGFβ constrains CD16-mediated antibody-dependent cellular cytotoxicity, one of the functions of NK cells 101 . Apart from as a soluble ligand, membrane-bound TGFβ on myeloid-derived suppressor cells 102 , T reg cells 103 or exosomes 104,105 can also abrogate NK cell function. NK cells can be grouped among a growing family of innate lymphoid cells (ILCs), which functionally and phenotypically mirror several T cell subtypes, except for their antigen specificity 106 . Interestingly, TGFβ can convert NK cells into type 1 ILCs, which, especially under the control of the immunosuppressive cytokine, fail to control local tumour progression [107][108][109] (Fig. 3b).
Furthermore, TGFβ was reported to change the phenotype of type 2 ILCs into an IL-17-producing type 3 ILC phenotype 110 , analogous to a shift from a T helper 2 (T H 2)-type response to a T H 17-type response.

TGFβ and adaptive cancer immunity
The functions of TGFβ signalling in reducing protumorigenic inflammation during early-stage cancer are deflected into creating a permissive TME during disease progression. In the following subsections, we describe how tumours that have co-opted a TGFβ-rich, anti-inflammatory TME evade antitumour T cell responses (Fig. 3c-e).

Suppression of DC function
One of the critical roles in orchestrating immunity is antigen presentation: DCs are professional antigenpresenting cells, known for their ability to mature in inflammatory conditions and phagocytose tumour cells. They can then migrate to lymphoid structures and present tumour antigens on the two types of MHC, interacting with both CD8 + cytotoxic T lymphocytes (CTLs) and CD4 + T H cells 111 .
Active TGFβ can prevent immature myeloid cells from undergoing DC differentiation, a process driven by the SMAD-regulated transcription factor ID1 (reF. 112 ). In the case of monocytes, this process may additionally involve an autocrine TGFβ-mediated feedback loop 113 (Fig. 3c). After DC differentiation, immature DCs promote tolerance and mediate the generation of T reg cells during homeostasis 114,115 . Elevated TGFβ levels can impede DC maturation and lower the expression levels of MHC molecules and inflammatory cytokines, reducing the ability of DCs to activate T cells 116,117 . This regulatory function of TGFβ is critical in preventing autoimmunity 118 , but can limit immunity in the TME. Furthermore, tumour-associated DCs produce TGFβ1, which primes the differentiation of T reg cells 119,120 . In addition, DCs express αVβ8 integrin, which enables the release of active TGFβ from the ECM. In mice, DCs lacking αVβ8 integrin fail to induce T reg cells and cause autoimmunity [121][122][123] .
DC migration is another critical function in steering immune responses, and TGFβ has been reported to restrict DC chemotaxis by regulating chemokine receptor expression 124,125 (Fig. 3c). In in vivo cancer models, DC trafficking to lymph nodes was reduced by TGFβ1 (reF. 126 ), and blockade of TGFβ signalling increased the antitumour efficacy of DC vaccines 127 . Finally, a recent study found that TGFβ can also inhibit the function of plasmacytoid DCs, which includes secretion of type I inferferon and the activation of NK cells 128,129 . This corroborates findings in breast and head and neck cancer, where TGFβ plays a role in suppressing plasmacytoid DC-derived IFNα and IFNβ 130,131 .

Regulation of TCR signalling
TCRs can recognize a large variety of epitopes, including tumour neoantigens, cancer germ line antigens and viral oncoproteins, bound to MHC at the surface of antigen-presenting cells. The strength of the TCR-MHCbound antigen interaction determines whether the Granulocytes also known as polymorphonuclear cells, a group of myeloid cells comprising neutrophils, basophils, eosinophils and mast cells.

Serrated CRC
a non-classical type of colorectal cancer (CrC) that derives from an alternative carcinogenesis pathway and has a sawtooth-like histological appearance.
Natural killer (NK) cells innate cytotoxic immune cells that can kill tumour cells (or pathogen-infected cells) without any priming or prior sensitization.

Antibody-dependent cellular cytotoxicity
Cell killing by virtue of target cell-specific antibodies and effector cells, such as natural killer cells, that express antibody receptors. naTure revIeWS | CANCeR downstream signal is sufficiently robust to activate the T cell. In vitro experiments showed that the earliest biochemical events detectable upon TCR triggering, such as tyrosine phosphorylation and calcium ion influx, are suppressed by TGFβ signalling 132 . Indeed, Tgfbr1-knockout mouse T cells can be activated by weaker TCR stimuli compared with their wild-type counterparts 133 . These observations are further supported by the finding that CD4 + T cells isolated from conditional Tgfbr2-mutant mice display accelerated calcium influx and TCR activation upon suboptimal stimulation 55,56 . Indeed, blockade of proximal TCR signalling by TGFβ has been observed in cancer 134,135 , and genetic inhibition of the TGFβ pathway in CD8 + T cells potentiates antitumour adaptive immune responses by lowering the TCR activation threshold 136 (Fig. 3c).

T H cell proliferation and differentiation
CD4 + T cells are able to redirect their differentiation programme in response to different threats and acquire distinct functions to combat specific pathogens. Extracellular signals from the environment control this process. Among them, TGFβ signalling exerts a powerful influence in the polarization of the four major CD4 + T cell subsets; it prevents T H 1 cell and T H 2 cell differentiation, while promoting T H 17 cell and T reg cell programmes. This role is also evident in the TME and is an important mechanism of immune evasion (Fig. 3d). Although the role of T H 2-type immunity in cancer is still debated, the TME of several tumour types, including subsets of CRC, squamous lung cancer, and luminal A breast cancer, exhibit upregulation of T H 2 cell gene signatures 137 . Early studies showed that combined blockade of IL-10 and TGFβ signalling in tumour-bearing mice elicited T H 2-type responses 15 . Further evidence of TGFβ-mediated suppression of T H 2-type immunity in tumours came from analysis of the MMTV-PyMT breast cancer mouse model. In this strain, genetic or pharmacological blockade of TGFBR2 in CD4 + cells (but not in CD8 + cells) promotes a T H 2-type response that depends on IL-4 and that results in blood vessel reorganization in the TME, leading to tumour hypoxia and death 150,151 . The acquisition of the T reg cell phenotype is, however, counterbalanced by T H 1-and T H 2-polarizing cytokines such as IFNγ and IL-4 (reF. 159 ). In addition, low TGFβ concentrations synergize with IL-6 to promote T H 17 cell differentiation instead of T reg cell differentiation 160 . In turn, T H 17 cells can transdifferentiate into T reg cells by the action of TGFβ and aryl hydrocarbon receptor (AHR) during the inflammation resolution phase 161 .
The TGFβ-rich TME can promote CD4 + T cell polarization to a T reg cell phenotype as a mechanism to enforce tumour antigen tolerance, as seen in pancreatic cancer   162 . Indeed, FOXP3 gene expression correlates with TGFB1 mRNA levels in patient cohorts with several tumour types 148 . Furthermore, pharmacological inhibition of TGFβ signalling results in decreased T reg cell numbers in the TME of tumour models 148,163 . However, the relative contribution of these effects to the outcome of anti-TGFβ therapy remains to be established. For example, conditional deletion of Tgfbr1 in T reg cells does not influence CRC growth or radiotherapy response in syngeneic tumour cell implantation in mice, whereas Tgfbr1 knockout in CD8 + T cells results in potent antitumour immune responses, implying a minor role for TGFβ-induced T reg cell polarization in this model 136 .
T reg cells produce and carry GARP-bound L-TGFβ1 at the cell surface, which can be activated by αVβ8 integrins. However, the relevance of this mechanism is controversial, as Tfgb1-knockout T reg cells can still enforce tolerance 164,165 . Consistent with this finding, in a mouse model of prostate cancer, CD4 + T cell-specific ablation -but not T reg cell-specific ablation -of Tgfb1 enhanced immune responses against the tumour 166 . On the other hand, activated T reg cells upregulate αVβ8 integrin expression, and β8 integrin-deficient T reg cells cannot suppress active T cell-mediated inflammation in an experimental model of colitis 167 . As discussed already, antibodies that target β8 integrin prevent TGFβ mobilization from latent deposits and potentiate antitumour cytotoxic T cell responses 43 . Similarly, antibodies that prevent TGFβ release by targeting GARP inhibit the immunosuppressive capacity of T reg cells in a graftversus-host disease model 168 . Anti-GARP antibodies also promote tumour immunity and synergize with ICI therapy 45, 168,169 . However, it remains to be proven that these effects occur due to inhibition of active TGFβ derived from T reg cells, as multiple other cell typesincluding platelets -carry GARP-L-TGFβ1 on their cell surface 45,46 . Indeed, deletion of Garp (also known as Lrrc32) in T reg cells does not trigger overt immune responses against tumour cells in mice 170 .

Inhibition of CTL activity
CTLs are central players in adaptive immune responses and play a critical role in antitumour immunity. They release cytolytic granules in response to binding specific antigen peptides presented on MHC class I by target cancer cells. Conditional deletion of Tgfbr2 in T cells exacerbates the effector phenotype of CD8 + T cells, which encompasses increased production of granzyme B and IFNγ 55,56 . A pioneering study demonstrated that transgenic mice that express a truncated, defective TGFBR2 in CD4 + T cells and CD8 + T cells mount potent immune responses against tumour cells, characterized by expansion of tumour-specific CD8 + cells 17 . Furthermore, CD8 + T cell-specific Tgfbr1-knockout mice reject tumour cells efficiently, whereas Tgfbr1 deletion in T reg cells or macrophages does not modify the antitumour immune response 136 . Immunotherapy based on the adoptive transfer of autologous tumour-reactive CTLs is improved if transferred T cells are rendered insensitive to TGFβ by expressing a dominant-negative TGFBR2 (reFs [171][172][173]. Taken together, these observations imply that in several tumour types the immunosuppressive function of TGFβ is exerted, to a large extent, by direct inhibition of CD8 + T cell function (Fig. 3e). Besides lowering the TCR activation threshold (discussed earlier), TGFβ suppresses CTL activity through several mechanisms. First, TGFβ downregulates transcription of genes encoding critical elements of the lytic machinery, such as granzyme A, granzyme B, perforin, FAS ligand and IFNγ, by directly repressing their promoters 174 . Also proliferation is inhibited through TGFβ-mediated silencing of Myc and Jun gene expression 175 . SMADs drive these effects in complex with the transcription factors ATF1 (reF. 174 ) and FOXP1 (reF. 175 ). Observations in melanoma mouse models and in T cells isolated from patients with melanoma also indicate that the genes encoding the transcription factors TBET and EOMES, two enforcers of the CD8 + effector programme 176,177 , are downregulated by TGFβ 178,179 . Another mechanism involves the inhibition of CD8 + T cell migration to tumour beds by TGFβ-mediated silencing of the gene encoding C-X-C chemokine receptor 3 (CXCR3), a receptor for the chemoattractant C-X-C motif chemokine 10 (CXCL10) 136 .

Promoting CD8 + T RM cells
Besides suppressing the cytotoxic effector programme of CTLs, TGFβ signalling can also stimulate their conversion to a tissue-resident memory T cell (T RM cell) pheno type (Fig. 3e). CD8 + T RM cells are important mediators of adaptive immunity in peripheral tissues and provide long-lived protection against reinfection. TGFβ downregulates the transcription factors TBET and EOMES during the maturation of T RM cells, initia ting a departure from the T H 1 cell programme 180 . In addition, TGFβ signalling promotes T RM cell residence in epithelial tissues such as skin, intestine or lungs by increasing the levels of αE (also known as CD103) and β7 integrin subunits in T RM cells, which interact with the epithelial adhesion molecule E-cadherin [181][182][183][184][185] . In mice, the induction of lung CD8 + T RM cells by TGFβ does not require SMAD4, suggesting that this subset is specified by non-canonical signalling 186 . It has been observed that TGFβ increases the abundance of CD8 + CD103 + T cells in the TME 187,188 . These findings are at odds with the immunosuppressive role of TGFβ in the TME, as the presence of CD8 + T RM cells in tumours is associated with antitumour immune responses and predicts good prognosis [187][188][189][190][191] . However, it has also been described that in mice TGFβ induces a tolerogenic CD8 + CD103 + T cell subset that expresses immunosuppressive molecules such as CTLA4 and IL-10 and helps tumours evade immunity 49 .

TGFβ-activated CAFs and immune evasion
In healthy tissues, fibroblasts remain largely quiescent but become activated in the event of tissue damage to help wound healing by depositing ECM and contracting the wound. The role of TGFβ in these processes has been extensively investigated 39,192,193 . In cancer, persistent inflammation and other signals sustain continuous fibroblast activation and exacerbate TGFβ production, resulting in a permanent and pathogenic wound-healing programme (Fig. 3f). Furthermore, CAF generation is also affected by cancer-derived exosomes, naTure revIeWS | CANCeR carrying nucleic acids (including mRNAs, microRNAs or other non-coding RNAs) 194 or proteins such as surface-bound TGFβ1. The latter was shown to induce tumour-promoting CAFs in vitro in a manner distinct from that of soluble TGFβ1 (reF. 195 ). Solid tumours recruit fibroblasts without exception, but the microenvironment of some subtypes is particularly rich in CAFs, exhibiting widespread TGFβ signalling in stromal cells as well as prominent ECM deposition. This phenomenon has been associated with poor prognosis and lack of immunotherapy responses in multiple studies (Box 2).
The mechanisms behind the role of TGFβ-activated CAFs in immune evasion remain, however, partially understood. TGFβ produced by CAFs, either through direct secretion or by release from latent deposits stored in the ECM, can directly suppress tumour immunity through signalling in cells of the innate and adaptive immune systems. Evidence also suggests that the composition, extent of crosslinking, and stiffness of the ECM -all of which are the consequence of the fibrogenic programme controlled by TGFβ -regulate T cell infiltration in tumours [196][197][198][199][200] . In addition, TGFβ signalling stimulates the production of a plethora of cytokines and growth factors by fibroblasts 9,201 , including IL-6 (reF. 202 ), leukaemia inhibitory factor (LIF) 203,204 , CXCL12 (reF. 205 ) and prostaglandin E 2 (reFs [206][207][208] ), which impact the immune environment and contribute to immune evasion. Of note, these molecules are produced not only by CAFs but also by other TME cell types or even by cancer cells and, therefore, the relative contribution of CAFs to their expression differs from tumour to tumour depending on the TME composition.

TGFβ inhibition-based immunotherapies
In light of the determining effect of TGFβ signalling in the TME on cancer progression, immune evasion and therapy resistance, a wide range of therapeutic modalities have been developed. These include TGFβ mRNAdirected agents, ligand traps, antibodies, fusion proteins and small-molecule kinase inhibitors of TGFBRs (TaBle 1). However, progress to bring these drugs to the clinic has faced important challenges. There are three main reasons for hesitation. First, TGFβ is a tumour suppressor for early neoplastic lesions. Indeed, a common side effect observed in patients treated with the anti-TGFβ-blocking antibody fresolimumab is the development of acanthomas, benign neoplastic skin lesions that regress after treatment cessation 209 . Second, genetic loss-of-function studies in mice suggest the possibility of serious inflammatory disease in gastrointestinal tissues caused by global TGFβ inhibition 58-61 . Third, and more critical, animal studies with small-molecule TGFBR1 inhibitors such as AZ12601011 and AZ12799734 (reF. 210 ) and pan-TGFβ antibodies have confirmed a risk for overt cardiovascular adverse effects characterized by heart valve thickening, haemorrhage, inflammation, and endothelial and stromal hyperplasia 211-213 (Box 1). Mice with a conditional Tgfbr2 knockout in postnatal smooth muscle cells develop similar cardiovascular disease, implying that the deleterious effects triggered by TGFβ inhibitors are to a large extent due to alterations in vascular smooth muscle cells 214,215 .
Selected for its relatively safe toxicology profile 212 , the TGFBR1 kinase inhibitor galunisertib entered clinical investigation more than a decade ago. In phase I trials, an intermittent dosing schedule was found to be well tolerated, demonstrating a therapeutic window [216][217][218] (TaBle 2). Since then, many other clinical trials have assessed galunisertib alone or in combination with other chemotherapies with manageable safety (reviewed in reFs [219][220][221] ). However, this drug achieved only modest responses in phase II trials, including as monotherapy for patients with refractory hepatocellular carcinoma 222 . The reasons are unclear but may partly be due to suboptimal patient stratification and insufficient inhibitory potency of this molecule with the intermittent dosing strategy used. New TGFBR1 inhibitors more potent and specific than galunisertib have been developed and are currently being tested in patients (TaBles 1,3).
Besides small-molecule TGFBR1 inhibitors, other early agents include a phosphorothioate antisense oligodeoxynucleotide specific for TGFB2 mRNA (trabedersen 223 ), a vaccine derived from an irradiated and TGFB2-antisense transfected non-small-cell lung cancer cell line (belagenpumatucel-L 224,225 ) and a monoclonal antibody to all three TGFβ ligands (fresolimumab). Clinical development of trabedersen has slowed, but second-generation antisense molecules targeting TGFB1, TGFB2 or TGFB3 are still in development 226,227 .

Box 2 | Linking TGFβ signalling in CAFs to immunotherapy responses
In colorectal cancer (CrC), cancer-associated fibroblast (CaF) abundance and elevated expression of fibroblast transforming growth factorβ (TGFβ) response signature (F-TBrS), which includes primarily genes encoding extracellular matrix (eCm) proteins and cytokines induced by TGFβ, predict the risk of relapse after therapy and metastasis robustly 299,302 . It was also found that upregulation of a similar F-TBrS identifies patients with urothelial cancer exhibiting poor therapeutic responses to anti-programmed death ligand 1 (PDl1) therapy in a clinical trial 236 . eCm-encoding genes induced by TGFβ also predict lack of responses to immune checkpoint inhibitors (ICIs) 238 . Subsequent studies have characterized CaF heterogeneity, and its association with response to ICIs [303][304][305][306] . These studies revealed the existence of two major CaF subsets: one exhibits an eCm-producing and contractile (α-smooth muscle actin-positive) phenotype enforced by TGFβ, whereas the other upregulates proinflammatory mediators such as interleukin-6 (Il-6) (reFs 304-306 ) (Fig. 3f). The abundance of the fibrogenic TGFβ-activated CaF subset is associated with a poor response to anti-PDl1 therapy in clinical trials and experimental models 76,236,304,305 . emerging evidence also suggests an essential role for TGFβ signalling in shaping CaF heterogeneity (Fig. 3f). Whereas Il-1 promotes the acquisition of an inflammatory programme in CaFs of pancreatic cancer, TGFβ suppresses Il-1 receptor expression in this population and impedes their specification 306 . Furthermore, an interferonγ (IFnγ)-licensed CaF population emerges upon TGFβ blockade in mouse tumour models 304 (Fig. 3f). This subset expresses major histocompatibility complex molecules and other factors involved in antigen processing and presentation, implying an immunomodulatory role 304 .
CrCs, urothelial carcinomas and possibly other tumour types that exhibit elevated levels of the TGFβ-driven CaF gene expression programme are immune excluded and insensitive to ICI immunotherapy 76,236,238,239 . using human-like mouse models of CrC, we showed that treatment with a type I TGFβ receptor (TGFBr1) inhibitor enables T cell infiltration and renders metastases susceptible to anti-PDL1 therapy 76 . another study demonstrated that treatment with a pan-TGFβ antibody prevents T cell exclusion and enhances responses to anti-PDl1 treatment in tumour models characterized by TGFβ activity in CaFs 236 . The synergism between TGFβ inhibition and ICIs was subsequently corroborated in multiple mouse cancer models and experimental settings 148,155,229,230,233,235,237,304 . However, it remains unclear to what extent CaFs are the culprits of immune evasion and anti-programmed cell death protein 1 (PD1) or anti-PDl1 therapy failure in these models. At present, of these first-generation agents only galunisertib and fresolimumab remain in active trials; however, they have not shown sufficient clinical activity and, as we discuss herein, several second-generation TGFβ pathway inhibitors have already reached clinical trials.
Despite the complexity and risks of clinically targeting the TGFβ pathway, an enduring interest is demonstrated by the long list of recent agents and active trials (TaBles 1,3). A number of these strategies target one or two specific TGFβ isoforms in an attempt to avoid naTure revIeWS | CANCeR adverse effects seen with pan-inhibitory antibodies used in the past (Box 1). Preclinical evidence suggests that individual TGFβ ligands may be safe to target 50, 228 and -in combination with ICIs -could be sufficient in some cancer types 50,229 .
Other ligand sequestering approaches have been taken using TGFBR2 ectodomain fusion proteins, engineered into bispecific drugs. Of these, the most advanced is bintrafusp alfa (also known as M7824), which has an ecto-TGFBR2-derived ligand trap fused to a human monoclonal antibody to PDL1 (reF. 230 ). This agent, as well as the similar molecule SHR-1701, is currently being evaluated in the clinic 231 . Similarly, a ligand trap fused to anti-CD73 (GS-1423) has entered clinical trials (TaBle 2), and a ligand trap fused to an antibody targeting the immune checkpoint molecule CTLA4 has shown promising results in preclinical studies 148 . Furthermore, the aforementioned preclinical CD4 + T H cell-specific TGFBR2 blockade strategy also involves a fusion protein, consisting of the TGFBR2 ectodomain attached to ibalizumab -a non-immunosuppressive CD4 antibody that was previously used to block HIV infection 150,232 .
Most ongoing strategies to block TGFβ signalling involve combination therapies, either together with standard-of-care agents or, increasingly, with immunotherapeutic regimens such as ICIs (Box 2; TaBle 3). The latter is supported by a growing number of promising results in preclinical studies pointing to synergistic immunomodulatory actions of TGFβ blockade 76,148,155,229,230,[233][234][235][236][237] . The prevailing rationale is that TGFβ pathway inhibition can overcome immunosuppressive signalling in the TME, and facilitate T cell tumour infiltration and cytotoxic activity, among a number of other relevant factors that are, unsurprisingly, implicated in failure of ICIs. Indeed, several studies have found elevated TGFβ programmes in ICI-non-responding cancers 50,76,229,236,238-244 (Box 2).  Furthermore, TGFβ is increasingly recognized as a key immunosuppressor that can diminish tumour infiltration and the efficacy of adoptive immune cell transfer therapy, especially for solid cancers. In that field, chimeric antigen receptor (CAR) T cell approaches are being actively investigated 245 . There have been a number of approaches to make CAR T cell products resistant to TGFβ. These include the overexpression of a dominant-negative TGFBR2 (reF. 246 ) or of a constitutively active AKT 247 , or use of CRISPR-Cas9 to knock out the endogenous TGFBR2 (reF. 248 ). Alternatively, the lymphocyte-inhibitory TGFβ ligand has been rewired into a stimulatory signal via a chimeric switch receptor 249 . One such approach combines the extracellular ligandbinding parts of TGFBR1 and TGFBR2 with intracellular IL-12Rβ1 and IL-12Rβ2 signalling domains, expressed on a CAR T cell 250 . A second study used a pooled CRISPR knock-in screen to evaluate a panel of transgenic immunomodulatory constructs, among them an engineered TGFBR2-4-1BB switch receptor 251 . Furthermore, a TGFβ CAR has been reported that switches T cells from immunosuppressed to proliferating, T H 1-type cytokine-producing T cells that can activate neighbouring CTLs 252,253 . In these studies, the transferred CAR T cells show both greater activity and fitness over TGFβ pathway wild-type CAR T cells. Together, these developments demonstrate a broad investment in combining immunotherapeutic strategies with targeted TGFβ inhibition.
Parallel efforts to induce tumoural T cell infiltration and subsequent immunotherapeutic efficacy led to the auspicious combination of ICIs with radiotherapy 254,255 . Interestingly, TGFβ plays a key role in limiting the effect of in situ vaccination, a key therapeutic benefit of radiotherapy, advancing the rationale for a triple combination of an ICI, radiotherapy and TGFβ blockade in a preclinical breast cancer model 256 . Similarly, such a strategy was reported for mouse models of CRC and melanoma 257 . Furthermore, a feasibility trial of the combination of fresolimumab with focal irradiation in patients with metastatic breast cancer was successful 258 (TaBle 2). A similar clinical trial is ongoing for early-stage non-small cell lung cancer (TaBle 3). Other potential TGFβ inhibition-based combinatorial immunotherapies may include oncolytic viruses 259 , NK cell therapy 260 , DC vaccination 127,261 , vaccine-based approaches such as gemogenovatucel-T 262 or blockade of monocyte recruitment 263 .

Discussion and future perspectives
TGFβ is a powerful cytokine capable of dominating the behaviour of most cells present in the TME. Generally, TGFβ enforces immune tolerance, suppresses inflammation and regulates wound healing during homeostasis. These mechanisms are often co-opted during tumour evolution to evade the immune system. However, as we have described herein, the effect of TGFβ signalling can differ substantially depending on the tumour type, organ affected and disease stage. Beyond the findings that genetic or pharmacological TGFβ blockade triggers potent antitumour responses in several preclinical model systems, it is becoming increasingly clear that the type and extent of this response are largely context dependent and the sum of disparate processes. Therefore, how TGFβ remodels different cancer ecosystems remains an important question for the coming years: which cell types are essential in each context, and how are distinct responses coordinated in space and time? It is also worth

Chimeric switch receptor
Fusion proteins that link the binding of (immuno)inhibitory ligands to the activation of intracellular stimulatory signal elements, or vice versa.

In situ vaccination
The effect of therapeutically increasing the release of tumour-associated antigens, combined with innate immune cell activation, which results in (more) effective antigen presentation and T cell or B cell priming. Triggers include immunogenic cell death, radiotherapy and oncolytic viruses. bearing in mind that our current understanding of the roles of TGFβ in cancer emerges from decades of studies of this cytokine in tissue development and organ homeostasis. However, chronic or elevated TGFβ signalling may affect the TME beyond the range of functions identified in homeostatic conditions. Research on all these topics is key to identifying which tumour types or subtypes can benefit from TGFβ-inhibitory therapies, interpreting the results of upcoming clinical trials and optimizing the use of TGFβ inhibitors in combination with other therapeutic modalities. These efforts should include the application of TGFβ-related predictive biomarkers, such as the fibroblast TGFβ response gene signature (Box 2). In our view, progress in translational research also demands a shift from the simplistic subcutaneous tumours commonly used in immunological studies to cancer models that more faithfully reproduce key aspects of TGFβ signalling in human disease.
Despite the impressive effects of TGFβ-inhibitory therapies in preclinical cancer models, the benefits of this therapy have not yet been translated to patients. Research in mouse models has revealed a strong synergism between TGFβ pathway inhibitors and ICIs. To date, combinatorial TGFβ blockade and ICI strategies have not yet been extensively tested in patients, in part due to the scarcity of TGFβ inhibitors in advanced clinical stages. As this situation is rapidly changing (TaBle 3), the field eagerly anticipates the results of these studies, keenly aware of the pending safety concerns. In this regard, a better understanding of the biological basis for the cardiovascular adverse effects shown by many TGFβ inhibitors is crucial for their systematic implementation in the clinical setting. Are the TGFβ isoforms that are important in shaping the TME the same as those that regulate the cardiovascular system? What is the relative contribution of canonical versus non-canonical signalling in the cardiovascular defects triggered by TGFβ inhibition? New strategies, including TGFβ isoform-specific blocking antibodies, some of which already under clinical investigation, antibodies capable of inhibiting the TGFβ pathway in specific TME cell types, and tumour-specific delivery of TGFβ inhibitors may also help reduce adverse effects. In addition, novel small-molecule TGFBR1 inhibitors are advancing with apparently manageable side effects. Finally, a growing group of agents aim at preventing TGFβ activation. Although our knowledge of this area is relatively limited, TME-specific upstream mechanisms have an unmistakable therapeutic potential.
As the number of possible combinations of (immuno) therapies grow exponentially, one inevitable challenge of near-future clinical practice concerns the choice for the best suited targets and therapies on a per-patient level. This requires a much better understanding of the most relevant tumour-specific mechanisms in the TME, and their relation to the individual immunological status 264