Dynamics of transforming growth factor β signaling and therapeutic efficacy

Abstract

Transforming growth factor β (TGFβ) is a multifunctional cytokine, and its signalling responses are exerted via integrated intracellular pathways and complex regulatory mechanisms. Due to its high potency, TGFβ signalling is tightly controlled under normal circumstances, while its dysregulation in cancer favours metastasis. The recognised potential of TGFβ as a therapeutic target led to emerging development of anti-TGFβ reagents with preclinical success, yet these therapeutics failed to recapitulate their efficacy in experimental settings. In this review, possible reasons for this inconsistency are discussed, addressing the knowledge gap between theoretical and actual behaviours of TGFβ signalling. Previous studies on oncogenic cells have demonstrated the spatiotemporal heterogeneity of TGFβ signalling intensity. Under feedback mechanisms and exosomal ligand recycling, cancer cells may achieve cyclic TGFβ signalling to facilitate dissemination and colonisation. This challenges the current presumption of persistently high TGFβ signalling in cancer, pointing to a new direction of research on TGFβ-targeted therapeutics.

Keywords:

• TGFβ signaling
• signaling dynamics
• therapeutic efficacy
• cancer treatment

1. Overview of the TGFβ signalling pathway in normal and disease conditions

Transforming Growth Factor β (TGFβ) signalling has been extensively studied in the past, attributed to its complex roles in multiple physiological and pathological contexts, including embryonic development, wound healing and cancer (Amento and Beck 2007; Massagué 1998; Bierie and Moses 2006). As a pleiotropic cytokine, TGFβ is secreted into the extracellular space as a latent complex, consisting of the dimeric ligand, latency-associated proteins (LAP) and large TGFβ binding proteins (LTBPs) (Taipale et al. 1994; Zhu and Burgess 2001; Annes, Munger, and Rifkin 2003; Shi et al. 2011; Horiguchi, Ota, and Rifkin 2012). LAP clamps TGFβ dimer via non-covalent binding while covalently linked to LTBP, with the latter responsible for sequestering latent TGFβ ligands to the extracellular matrix (ECM) (Taipale et al. 1994; Shi et al. 2011; Horiguchi, Ota, and Rifkin 2012). The activation of TGFβ ligands occurs when the tissue microenvironment encounters certain stimuli, such as during gastrulation, inflammation, hypoxia and other stressful situations (Annes, Munger, and Rifkin 2003; Horiguchi, Ota, and Rifkin 2012). Under these situations, the mature TGFβ dimer is released from LAP and LTBP via enzymatic processes or mechanical forces, mediated by factors including a number of extracellular proteases and structural proteins such as matrix metalloproteinase-9 (MMP-9), integrin αvβ5 and αvβ6 (Taipale et al. 1994; Munger et al. 1999; Annes et al. 2004; Tatler et al. 2011; Horiguchi, Ota, and Rifkin 2012; Sarrazy et al. 2014; Kobayashi et al. 2014). In the absence of active TGFβ, the extracellular and transmembrane domains of the TGFβ receptors (TβR) are oriented in a way that they cannot initiate intracellular signalling, despite TGFβ receptor type II (TβRII) being constitutively phosphorylated with the freedom to oligomerize with TGFβ receptor type I (TβRI) (Zhu and Sizeland 1999a, 1999b). In fact, TβRII is negatively regulated in the cells by SPSB1 through direct interactions at the membrane level, which increases its ubiquitination level and degradation rate (Liu et al. 2015). When an active TGFβ ligand binds to the extracellular domain of TβRII on the target cell surface, it induces a conformational change that results in the phosphorylation and activation of TβRI (Wrana et al. 1994; Zhu and Burgess 2001; Zhu and Sizeland 1999a, 1999b; Huang and Chen 2012). After activation, TβRI is then capable of activating multiple downstream signal transduction proteins on the intracellular side of the plasma membrane with its receptor serine/threonine kinase function, diverging TGFβ signalling into canonical and non-canonical pathways (Figure 1) (Shi and Massagué 2003; Huang and Chen 2012).

Figure 1. TGFβ signalling mechanisms and cross-talks with other pathways. Red dots with “P” represent phosphorylation. Under appropriate stimulation, latent TGFβ complex dissociates into active TGFβ ligands, which bind to TβRII receptors and phosphorylate the TβRI receptors to form an active TGFβ-receptor complex. TβRI activates downstream effectors such as Smad2/3, Ras and PI3K, and consequently conducts intracellular signalling via Smad, MAPK, PI3K and other pathways. ERK from the MAPK pathway is able to transiently support the Smad-dependent pathway by enhancing Smad2/3 signalling and phosphorylating transcription co-factors of Smad, while one of the gene products of Smad-dependent signalling, Smad7, has an inhibitory effect on Smad2/3 functions. Activated TGFβ signalling results in integrated cellular effects including cell survival and migration in normal cells and metastasis in cancer. Internalisation of the TGFβ-receptor complex to endosomes can lead to its degradation or recycling.

In the canonical pathway, regulatory Smads (R-Smads), Smad2 and Smad3 (termed as Smad2/3 thereafter), and the common Smad (Co-Smad), Smad4, are the intracellular signalling mediators for TGFβ (Massagué 1998; Macias, Martin-Malpartida and Massagué 2015). R-Smads contain a TGFβ receptor-binding site at their C-terminal MAD Homology 2 (MH2) domain, which brings them to close proximity of the receptor complex for phosphorylation and promotes their interaction with other Smad proteins to form oligomers (Macias, Martin-Malpartida, and Massagué 2015). At the N-terminus, the MH1 domain contains a nuclear localisation signal and a DNA-binding site for their transcription factor activity. The PPXY motif in the linker region of R-Smads between MH1 and MH2 domain can interact with E3 ubiquitin ligase Smurf2 and become ubiquitinated for degradation (Zhang et al. 2001; Macias, Martin-Malpartida, and Massagué 2015). After being recruited by the TGFβ ligand-receptor complex, phosphorylated R-Smads form a heterotrimeric complex with Smad4, and the complex is transported into the nucleus to carry out its gene regulatory functions by either directly binding to the Smad-specific SBE or CAGA-box elements upstream of TGFβ-related genes, or associating with other transcriptional co-factors in a cell-type specific context (Nakao et al. 1997; Dennler et al. 1998; Itoh et al. 2019). Notably, the inhibitory Smad (I-Smad) Smad7 can regulate the activation and function of Smad2/3 at multiple levels, by forming stable complexes with TβRI to prevent Smad2/3 phosphorylation (Hayashi et al. 1997), interfering with R-Smad/Smad4 complex formation through its interaction with R-Smads and outcompeting the R-Smad/Smad4 complex for DNA binding (Zhang et al. 2007; Yan et al. 2016). Thus, the canonical pathway of TGFβ experiences tight control from multiple mechanisms at different levels of signalling to ensure the cellular response is only activated when required (Zhu and Burgess 2001).

The non-canonical pathway involves crosstalk with the Ras-MAPK, PI3K/AKT and Rho-like GTPase pathways, leading to more complicated cellular responses (Figure 1) (Zhang 2009). One of the cellular events induced by TGFβ is the epithelial-mesenchymal transition (EMT), and there is evidence that in addition to the stimulatory effect of the canonical TGFβ pathway on EMT, non-Smad pathways of TGFβ can also positively or negatively regulate EMT to create a highly controlled response (Moustakas and Heldin 2016). For example, TGFβ is capable of activating MAPK, which in turn induces the phosphorylation of Smads or transcription co-factors that interact with Smad complex, thereby influencing Smad-dependent gene transcription (Mulder 2000; Chapnick et al. 2011). Studies showed that this interaction between the TGFβ and the MAPK pathways leads to the conversion of epithelial cells towards a fibroblast phenotype (Oft et al. 1996; Meyer-ter-Vehn et al. 2006). Additionally, TGFβ leads to the phosphorylation of Akt and promotes its downstream intracellular signalling events, mainly through the mTOR pathway, which functionally regulates EMT and cell motility (Zhang 2009; Hamidi et al. 2017; Lamouille and Derynck 2007; Gulhati et al. 2011). To a similar extent, TGFβ-activated RhoA GTPase and its effector ROCK are responsible for cytoskeleton rearrangement and the downregulation of adherens junction, which contribute to the EMT process (Korol, Taiyab, and West-Mays 2016). The extensive cooperation between these pathways with TGFβ signalling is the key to a higher level of control to cellular processes and more complicated cellular decisions.

Previous reports using genomic studies have revealed large-scale alterations on the RNA and protein levels downstream of the TGFβ signalling cascade under different conditions (Matsumura et al. 2011; Chen, Zaidi, et al. 2018; Guerrero-Martínez et al. 2020). Microarray and qRT-PCR analyses on cDNA derived from lung epithelial cells HPL1D and lung adenocarcinoma cells A549 discovered more than a thousand TGFβ-regulated genes in either cell line, with most genes associated with cytoskeleton, cellular junctions and Wnt signalling pathway (Ranganathan et al. 2007). Similar experiments on human kidney epithelial cells HK-2 with RNA sequencing also showed global changes in gene expression level after TGFβ stimulation (Brennan et al. 2012). Most importantly, as part of the feedback regulation mechanisms, TGFβ promotes the expression of proteins which can either activate or inhibit its own signalling (Miyazono 2000; Yan, Xiong, and Chen 2018). It has been shown that Smad7 and transcriptional co-repressors SnoN and Ski are TGFβ-inducible proteins, which in turn negatively regulate the TGFβ pathway (Stroschein et al. 1999). On the other hand, the expression of matrix metalloproteinase-9 (MMP-9), a proteinase which activates latent TGFβ ligands, was shown to be induced when the cells are treated with TGFβ (Kobayashi et al. 2014). These overarching gene expression profiles and mechanisms suggest that the cellular response to TGFβ is likely a dynamic event that is built on multiple layers of molecular regulations and interactions with other signalling pathways in a context-dependent manner.

TGFβ acts as a main driver of EMT, which is a key event in all TGFβ-involved processes when epithelial cells acquire mesenchymal traits and greater migratory ability (Xu, Lamouille, and Derynck 2009). Prevoius studies provide evidence that TGFβ promotes endocardial- and epicardial-mesenchymal transition during embryogenesis, and mice with homozygous TGFβ depletion showed prenatal or perinatal death (Kulkarni and Karlsson 1993; Sanford et al. 1997). In wound healing, TGFβ-driven EMT allows local epithelial cells to migrate into the wounded area under inflammatory conditions, however prolonged TGFβ signalling is also a causative factor to tissue fibrosis (Kalluri and Neilson 2003). Most interestingly, TGFβ signalling has contrary influences on cancer development at different stages of malignancy (Derynck, Akhurst, and Balmain 2001; Roberts and Wakefield 2003; Kubiczkova et al. 2012). In early stages, TGFβ exhibits anti-tumorigenic effects by inhibiting the cell cycle while promoting apoptosis of tumour cells (Brodin et al. 1999; Shima et al. 1999). On the other hand, in the later stages, TGFβ not only triggers EMT in cancer cells and allows them to become migratory and invasive, but also promote expression of components within the tumour microenvironment that favour metastasis (Padua and Massagué 2009). Thus, TGFβ is seen as a potential therapeutic target for metastatic cancers and an enormous number of studies have been performed on TGFβ signalling mechanisms to understand this pathway.

2. Current knowledge on TGFβ in tumorigenesis and metastasis

2.1. Identifying the TGFβ pathway as a tumour suppressor and a promoter of metastasis

As outlined above, studies of TGFβ signalling in malignancy revealed its diversified roles at different stages of cancer. In the early stages of tumorigenesis, it was found that TGFβ effectively suppresses the growth of tumour cells. For example, treating FET human colon carcinoma cells with TGFβ reduced cell growth by more than 60% (Hoosein et al. 1989). Similarly, suppressing TGFβ signalling with neutralising antibodies stimulated the growth of HS578T and MDA-231 human breast cancer cell colonies on soft agar (Arteaga et al. 1990). Further experiments identified the mechanisms of TGFβ-induced growth inhibition, including the interference with the cell cycle and promoting apoptosis (Brodin et al. 1999; Donovan and Slingerland 2000; Pardali and Moustakas 2007). TGFβ initiates cell cycle arrest in G1 phase and prevents tumour cells from entering S phase for division and proliferation (Donovan and Slingerland 2000; Pardali and Moustakas 2007). Mechanistically, TGFβ induces the expression of p15 for the inhibition of cyclin-dependent kinase 4 (cdk4) and cdk6, which are important mediators for G1/S phase transition (Warner et al. 1999). In particular for cancer, TGFβ acts as a suppressor for the oncogenic c-myc protein that transcriptionally activates multiple cyclins and cdks including cdk4/6 (Feng et al. 2002). The activation of c-Jun N-terminal kinase (JNK) and p38 from the MAPK pathway and inhibition of NF-κB are the main events that allow pro-apoptotic proteins to initiate the cell death program (Schuster and Krieglstein 2002; Pardali and Moustakas 2007). Conjointly, the cooperation between the canonical and non-canonical pathways of TGFβ signalling leads to the apoptosis of tumour cells, preventing their immortality and proliferation.

However, other studies with similar experimental strategies demonstrated opposing results on tumour development under the presence of TGFβ, revealing the pro-tumorigenic effect of TGFβ in malignant cancer (Tang et al. 2003; Meyer-ter-Vehn et al. 2006; Pardali and Moustakas 2007; Padua and Massagué 2009). When implanting TGFβ-overproducing Meth A sarcoma cells in CB6F1 mice, the proportion of cells displaying anaplasia and the tumour volume was significantly increased compared to the control cells with normal TGFβ production (Chang et al. 1993). Likewise, FET cells transfected with antisense expression vectors for TGFβ showed shorter or no lag phase before they began to replicate exponentially, and they were also more efficient in colony formation in vitro and tumour formation in vivo (Wu et al. 1992). As discussed in the previous section, TGFβ is a major driver of EMT under different contexts, particularly in promoting metastasis of cancer cells. TGFβ modulates the change in cell junctions, polarity and shape towards a mesenchymal phenotype, through the regulation of multiple intracellular mediators (Oft et al. 1996; Ozdamar et al. 2005; Padua and Massagué 2009). For instance, E-cadherin, an epithelial marker that plays important role in maintaining epithelial cell junction and intracellular signalling, is transcriptionally repressed by TGFβ during EMT, while the expression of mesenchymal markers such as N-cadherin and vimentin are upregulated (Wendt et al. 2011; Araki et al. 2011; Yang et al. 2015). Not only does TGFβ trigger downstream signalling and altered gene expression in cancer cells, but it also has an influence on stromal cells, immune cells and endothelial cells that reside in or around the tumour microenvironment to establish a pro-metastatic niche that facilitates the survival and invasion of malignant cells (Pickup, Novitskiy, and Moses 2013).

The correlations between TGFβ signalling level and the degree of metastasis, disease advancement or patient outcome have been established to validate the importance of TGFβ in promoting cancer progression (de Kruijf et al. 2013; Javle et al. 2014). Analysis of samples from non-small cell lung carcinoma (NSCLC) patients has found that TGFβ protein level is positively correlated with angiogenesis, lymph node metastasis and advanced stages of cancer (Hasegawa et al. 2001). For studies in other types of malignancies, such as colorectal cancer and breast cancer, high levels of TGFβ were also shown to have implications on tumour progression and poor prognosis (Robson et al. 1996; Ivanović et al. 2006). Immunohistochemical staining of breast tumour samples showed that although TGFβ was present in the stroma throughout the tumour mass, the region of cells with the highest level of TGFβ uptake was at the invasive front, where EMT markers were most detected (Lv et al. 2013). Tumour samples from gastric cancer patients also identified that a high level of active TGFβ is associated with the transformation of tissue-resident fibroblasts into cancer-associated myofibroblasts and poor patient outcomes (Hawinkels et al. 2007). These results provided preliminary evidence for the role of TGFβ in advanced cancer, by that they draw attention to the studies of TGFβ using in vivo and in vitro models, in the direction of developing new therapeutics to treat cancer in advanced stages (Padua and Massagué 2009).

2.2. Establishing experimental models to examine TGFβ’s effects on cancer cells and tumour microenvironment

To understand the metastatic-promoting mechanisms of TGFβ in different types of malignancy, a variety of tumour cell lines and in vitro experimental procedures have been adopted to examine the effect of TGFβ signalling (Feng, Xu, and Lin 2016; Jurukovski et al. 2005). As suggested in the book “TGF-β Signalling: Methods and Protocols” written by Feng, Xu, and Lin (2016), common ways to manipulate the TGFβ signalling pathway in mammalian cell cultures include ligand overexpression and pathway inhibition. Particularly for oncologic studies, stimulating cells with human recombinant TGFβ can trigger extensive signalling response in cancer cells that are highly sensitive to TGFβ, enabling the evaluation of this signalling pathway by measuring the concentration of phosphorylated Smad proteins, the localisation of Smad complexes or the expression level of Smad-induced gene products (Tesseur et al. 2006). To maximise the observation differences between TGFβ-stimulated and control cells, “optimal” conditions are often determined, for which the concentration of ligands and duration of treatment are optimised for the cells to exert the highest level of response to TGFβ. For example, in order to correlate TGFβ/Smad signalling with EMT in squamous cell carcinoma of the head and neck (SCCHN), a titration of 0.05 to 10 ng/ml of TGFβ ligands was performed to treat SCCHN cells Tu686. It was found that treating Tu686 cells at a TGFβ concentration of ≥ 5 ng/ml for 48 hours triggers the greatest change in vimentin and E-cadherin expression, and therefore this condition was used throughout the study for subsequent experiments (Yu et al. 2011). Similarly, Zhang et al. (2017) treated endometrial adenocarcinoma Ishikawa cells that have mesenchymal stem cell (MSC) properties with a titration of 1 to 50 ng/ml of TGFβ ligands to examine the effect of TGFβ on their differentiation. FACS analysis demonstrated that the increase of initial concentrations of TGFβ is correlated with decreased proportion of Ishikawa cells (Zhang et al. 2017). At 10 ng/ml or above, the effect of TGFβ on Ishikawa cells plateaued, and the authors decided to use this concentration as the condition for subsequent experiments.

When inhibiting TGFβ signalling in cancer cell cultures, the experimental strategies can be broadly divided into two categories: genetically knocking down/out TGFβ pathway component proteins, or using inhibitors of the TGFβ pathway (Nagaraj and Datta 2010). By treating the cells with antisense oligonucleotides, the expression and secretion of TGFβ ligands in cancer cells can be suppressed (Marzo et al. 1997; Schlingensiepen et al. 2011; Huber-Ruano et al. 2017). Through sequence-specific hybridisation, the antisense oligonucleotide AP-12009 binds complementary to TGFβ mRNA and restrains its translation in human pancreatic cancer cells Hup-T3, thereby significantly reducing the concentration of TGFβ in the medium (Schlingensiepen et al. 2011). With the treatment, metastasis of Hup-T3 cells was repressed, characterised by a decreased degree of proliferation, impaired migration, enhanced immune evasion and inhibited angiogenesis (Schlingensiepen et al. 2011). Intracellularly, small molecule inhibitors robustly suppress the kinase activity of TGFβ receptors (Wang et al. 2020). As an example, LY2109761 is a kinase inhibitor for TβRI/II, and its addition to human pancreatic carcinoma cells L3.6pl/GLT diminished Smad2 phosphorylation to a level below the baseline (Melisi et al. 2008). LY2109761 also significantly suppressed tumour growth and metastasis in mice injected with L3.6pl/GLT cells (Melisi et al. 2008). SB431542, a kinase inhibitor of TβRI, represses Smad2 phosphorylation and blocked the activation of TGFβ-responsive genes in a dose-dependent manner in various cell lines, including human kidney cells 293 T, FET cells and A549 cells (Halder, Beauchamp, and Datta 2005). From a functional perspective, SB431542 also hinders the EMT process of cancer cells driven by TGFβ, and prevents angiogenesis by blocking vascular endothelial growth factor (VEGF) secretion from cancer cells (Halder, Beauchamp, and Datta 2005). Alternatively, others have used extracellular targeting mechanisms including anti-TGFβ antibodies or ligand traps using TGFβ receptor moieties (Nagaraj and Datta 2010; Liu et al. 2012; Tabe et al. 2013; de Gramont, Faivre, and Raymond 2017). These molecules are not required to be imported through the plasma membrane, and therefore are potentially easier to design and more stable in vivo. 1D11 is a monoclonal antibody against TGFβ, and it has been demonstrated in acute myeloid leukaemia cell cultures that when 1D11 was added to the medium containing TGFβ, it caused reduced Smad2 phosphorylation, inhibition of cancer cell migration and increased apoptosis (Tabe et al. 2013). Another study utilised soluble TGFβ receptor type II (sTβRII) as an extracellular ligand trap against TGFβ (Liu et al. 2012). sTβRII showed similar effects on the growth and metastasis of murine breast cancer cells 4T1 as 1D11, but showed superior ability to suppress the growth and metastasis of human breast carcinoma cells MDA-MB-231 when compared to 1D11 (Liu et al. 2012). This implies a difference between mice cell lines and human cell lines in response to TGFβ inhibitors, which should be taken into account when designing experiments in mouse models.

In effect, in vitro models have shortcomings when mimicking human cancer, as cells in culture often experience artificial conditions which lack tissue factors that exert chemical and mechanical stimuli, do not encounter interactions with surrounding cells such as fibroblasts and immune cells, and have nutrients and antibiotics in excess. Therefore, the cellular response to TGFβ in vitro may not be true or representative of the response in the human body. To compensate for this, animal models are important to examine the effect of TGFβ on cancer cells in a live biological system (Böttinger, Letterio, and Roberts 1997; Serra and Crowley 2005). These include genetically modified mice that either have over-activated or impaired TGFβ signalling that can be induced under specified conditions (Muraoka-Cook et al. 2004; Azhar et al. 2009; XiYang et al. 2013). Other models include immune-compromised mice with implanted tumour cells that are treated with anti-TGFβ reagents, such as those described above. When determining the dose and interval of treatments in mice, the decision is mostly based on the tissue clearance and lethality, by which the efficacy of target inhibition follows. However, mouse models are different from humans in many ways, including cellular response to TGFβ on a time-dependent and cell type-specific basis (de Jong and Maina 2010). Abnaof and his group found that, although similar biological processes were triggered by TGFβ in murine and human cells, they concluded that there are much fewer protein interactions involved in the TGFβ signal transduction pathway in murine cell lines than in human cell lines, suggesting that murine cells or mouse models are not perfect representatives of the response to TGFβ seen in cells of human origin (Abnaof et al. 2014). Thus, the prediction power of mouse models on anti-TGFβ treatment response, such as toxicity and efficacy, remains questionable.

2.3. Translating preclinical findings to clinical use of cancer therapeutics

Developing cancer therapeutics with high cancer-killing efficacy and high specificity to malignant cells has always been the central focus of translational research. In recent years, several targeted therapies have established prominent achievements on a number of cancer types, such as Imatinib and bevacizumab (Iqbal and Iqbal 2014; Kim, Choi, and Lee 2022). TGFβ, as a molecular driver of EMT and metastasis, has been identified as a potential therapeutic target for high-grade cancer that is TGFβ-positive (Haque and Morris 2017; Huang et al. 2021). In fact, a large number of TGFβ-targeted therapies have entered clinical trials. These include monotherapies such as antisense oligonucleotides, small molecule inhibitors, soluble receptors and antibodies, and recently developed dual-targeting therapies using techniques such as bi-specific antibodies (Lind et al. 2020; Huang et al. 2021).

However, although many TGFβ-targeted therapies have shown efficacy in vitro and in mouse models in pre-clinical testing and showed low toxicity in human patients in Phase I clinical trials, rarely do they proceed to Phase III trials and clinical use. Details of completed or terminated clinical trials on TGFβ-targeting therapies with possible explanations for the obstacles have been recently reviewed (Teixeira, ten Dijke, and Zhu 2020). For instance, trabedersen (AP-12009) underwent a Phase IIb clinical trial in patients with high-grade glioma, categorised into glioblastoma multiforme (GBM) and anaplastic astrocytoma (AA) (Bogdahn et al. 2011). Despite the proof of safety, the median survival for both GBM and AA patients in the trabedersen-treated group was not significantly different from the chemotherapy-treated control group (Bogdahn et al. 2011). Another example is the small molecule kinase inhibitor of TβRI, galunisertib, which has gone through Phase II clinical trials for combination therapies with other treatments, such as nivolumab or durvalumab (Zhang et al. 2018; Wick et al. 2020). When testing the efficacy of galunisertib in combination with standard temozolomide-based radiochemotherapy in metastatic pancreatic cancer patients, it was found that the combination therapy has a similar median overall survival time with radiochemotherapy only (18.2 vs 17.9 months), and the median patient free survival time under the treatment of the combination therapy is 3.9 months less than the radiochemotherapy only treatment (7.6 vs 11.5 months) (Zhang et al. 2018; Wick et al. 2020). This shows that the inhibition of TGFβ with galunisertib did not benefit cancer retardation or patient survival, despite its initial success in in vitro and animal experiments. After the clinical study of galunisertib was discontinued, the Phase III clinical trial on bintrafusp alfa, an anti-PD-L1/TGFβ bifunctional fusion protein, was terminated in January 2021 due to the lack of efficacy compared to anti-PD-L1 therapy alone, after its initial success in demonstrating clinical safety and anti-cancer activity in Phase II clinical trial in October 2019 (Strauss et al. 2020). While treating cells in culture and mouse models with anti-PD-L1/TGFβ showed promising effects on limiting cancer progression, the outcome of Phase III clinical trial was disappointing, yet a hypothesis may be drawn from this failure: in vitro and animal models do not recapitulate the behaviour of cancer cells or their resident environment in human. Reasons that may lead to this failure in therapeutic development will be discussed next when we compare pre-clinical and clinical models used for these studies and analyse the most obvious “blind-spots” in the research of TGFβ-targeted therapy: treatment dose, interval and duration.

3. Potential reasons hindering the success of anti-TGFβ therapies

3.1. Understanding the dynamics of TGFβ signalling during metastasis

Tracing back to the building of concept, although high levels of TGFβ in patient samples is correlated with high-grade cancer, and inhibiting TGFβ signalling in vitro or in animal models showed encouraging effects on suppressing metastases, the question of "Do these experimental models factually represent human cancer?” has not been completely answered and needs to be considered when designing translational studies. In vitro studies require artificial induction of TGFβ signalling using exogenous ligands, while the culturing and ligand exposure conditions often lack careful consideration. Many studies regarding TGFβ signalling administer TGFβ ligands at nanograms per microliter of culture medium, as treatment at this level gives the highest desired signal and the lowest side effects on cell behaviour and survival. However, under these circumstances, the effects caused by the interaction of TGFβ signalling with other pathways and the spatiotemporal information of TGFβ signalling on single cells or the cell population are usually ignored. The suppression of TGFβ signalling in most in vitro experiments, therefore, is only seeking the total inhibition of positive TGFβ signals, but it may not achieve the total inhibition of the cellular responses and functional consequences of TGFβ signalling. In fact, under clinically relevant situations, the response of cancer cells to TGFβ is not only dose-dependent but also time-dependent. In in vivo tissue conditions, cells experience stimuli that can be variable in several dimensions, including the frequency, intensity, direction and duration of exposure, leading to dynamic responses. This is first demonstrated through embryonic studies, where the migration and differentiation of embryonic cells are dependent on the concentration gradient and the duration of exposure of TGFβ family cytokines such as bone morphogenetic protein (BMP) (Wu and Hill 2009). Other than the effect of external stimulation, cellular regulation of the pathway is also important to determine the intensity and duration of signalling response. Due to the complex nature of the interaction and regulation of TGFβ signalling in cells, this aspect should be critically considered when examining the functional consequences of TGFβ, rather than simply correlating ligand concentration with the cellular response.

Other than the dynamics of signalling intensity within a single cell according to TGFβ dose and treatment duration, spatial factors of a cell population also need to be considered. Different levels of response to TGFβ may determine the degree of EMT progression at different regions of the tumour mass, thereby controlling the behaviour of cells compartmentally to facilitate metastasis, such as by increasing the migrative rate of cells on the edge of the tumour while suppressing apoptosis of cells at the centre. Also, in collective cell migration, where cancer cells need to maintain cell-to-cell association while having the ability to invade the surrounding ECM, it is likely that the response of cells to TGFβ at the invasive front and the lagging side of the migrating cluster is different. However, using animal models to analyse the behaviour of cancer cells under TGFβ stimulation or with anti-TGFβ treatment, the interrogation of the tumour tissue is often performed via global quantitative methods such as luciferase assay and ELISA, or by the evaluation of tumour volume and cell number. In these situations, the information on the response of individual cells to TGFβ from a temporal or spatial point of view is lost, and only an averaged result from the overall tumour mass is obtained.

Due to the fast-evolving nature of cancer cells, dynamic signalling has previously been seen in other pathways, such as in MAPK and PI3K signalling (Kholodenko 2006; Hu et al. 2013; Valls and Esposito 2022). By stimulating human breast adenocarcinoma cells MCF-7 with epithelial growth factor (EGF) and heregulin-β1 (HRG), it was demonstrated by western blotting that the levels of phosphorylated ERK (pERK) and phosphorylated AKT (pAKT) oscillates every 20 to 30 minutes (Hu et al. 2013). Further, the crosstalk between MAPK and PI3K pathways led to sustained pERK and pAKT oscillations at low amplitude over a long period of time (Hu et al. 2013). As these pathways both interact with TGFβ signalling in malignant cells, the possibility of cross-regulation cannot be excluded, and thus signalling oscillation may also be seen in the TGFβ pathway with potential biochemical and functional consequences.

3.2. Explanations to the mechanisms causing dynamic signalling

Based on current studies of TGFβ signalling in cancer, a few hypothetic phenomena may give rise to dynamic TGFβ signalling, and the following will be discussed in this section: altered intracellular regulations to the signal transduction pathway and the presence of TGFβ-containing tumour-derived extracellular vesicles (TEVs). As mentioned above, due to its dictating roles in cell fate and survival, TGFβ is tightly regulated under normal circumstances with limited cellular functions (Zhu and Burgess 2001). Through negative feedback mechanisms, the signalling intensity and duration rapidly achieve equilibrium and return to the baseline level soon after ligand exposure (Figure 2A). For example, TGFβ induces the expression of SnoN and Ski proteins in normal cells, and these proteins act as transcriptional co-repressors to antagonise Smad-dependent gene transcription (Deheuninck and Luo 2009). Smad7 is also a TGFβ-inducible protein that counteracts with Smad2/3-mediated signal transduction, and it partners with other intracellular inhibitors of the TGFβ pathway to attenuate the signalling at multiple levels (Zhu, Iaria, and Sizeland 1999; Yan, Liu, and Chen 2009). Additionally, Smad ubiquitination regulatory factor (Smurf) proteins are E3 ubiquitin ligases that specifically target Smad for degradation (Inoue and Imamura 2008). Dysregulation of these proteins is commonly seen in malignancies, either through gene upregulation or downregulation. As summarised by Stolfi et al. (2013), an increased level of Smad7 was found to be associated with poorer prognosis in some cancer types while improving patient outcomes in others, and therefore it was predicted to have both pro- and anti-tumour effects. SnoN and Ski were also shown to be either upregulated or downregulated in different types of cancer, which was claimed to be due to their polarised roles as oncoproteins and tumour suppressors (Tecalco-Cruz et al. 2018). Likewise, Smurfs are addressed as either tumour-promoters or tumour-suppressors in various cancer types, and they showed contrasting effects in different situations (Fu et al. 2020). In addition to the direct regulators that primarily target the TGFβ pathway, the cross-regulation between TGFβ and other oncogenic or tumour-suppressive pathways in cancer cells also contributes to the modulation of TGFβ signalling. For instance, one would think that increased expression of Smad7 in cancer cells will lower the intensity of TGFβ signalling and therefore reduce cell metastasis. Yet, it was found that Smad7 overexpression mediated by the epidermal growth factor (EGF)/Stat3 pathway in cancer cells causes the dysregulation of TGFβ signalling by desensitising the cells to the TGFβ pathway, and thus unleashing the TGFβ pathway from its cytostatic control (Luwor et al. 2013). These phenomena raise a question to how exactly TGFβ is regulated in cancer, and whether these regulatory mechanisms are actually unidirectional in malignancy as previously described. More likely is that the alterations in these regulatory compartments of TGFβ signalling result in the loosening of its tight control, thereby causing a unique signalling pattern that is not seen in healthy cells, similar to a “swing” effect (Figure 2B). This less-controlled situation in cancer cells endued TGFβ signalling with more downstream cellular functions that are restricted in normal cells, including the stimulation of EMT and cell invasion.

Figure 2. Schematics of dynamic TGFβ signalling and potential effect on the metastatic process. A. TGFβ signalling pattern in healthy vs metastatic cells. Metastatic cancer cells may have sustained TGFβ signalling of higher magnitude and cycles under the effect of intracellular regulation mechanisms or exosomes, while healthy cells achieve equilibrium in a short duration after the stimulation. B. Illustration of the effect of inhibitor treatment intervals on a metastatic cell population. Green circles outline the cells with low signalling level at the time of treatment, which escapes the effect of treatment. At each treatment time point, the cell population shows heterogeneity towards the magnitude of signalling, and cells with low signalling intensity at the time of treatment are less impacted by the treatment and will account for the continuous signalling immediately after treatment.

Consequences of altered signalling patterns may be evidenced by the change in signalling duration and intensity, which are important parameters to measure downstream cellular response. In normal cells, intracellular regulation of TGFβ signalling has already been shown to be capable of diverging the effect of ligand stimulation into dynamic cellular responses by controlling the duration and intensity of the signalling level. It was found using mouse myoblast C2C12 cells and human keratinocyte HaCaT cells that although the level of R-Smad nuclear translocation is positively correlated with the concentration of TGFβ ligands at all times, the transcription level of TGFβ-regulated genes was only affected by TGFβ exposure within the first 4 hours before it reverted back to baseline levels, and this coincided with the level of Smad4 (Warmflash et al. 2012). This implies that the TGFβ ligand concentration may not directly represent the level of TGFβ-induced cellular responses in the long term, and that these cellular responses are regulated by intracellular feedback. Further in the same study, it was shown that BMP and activin/nodal signalling was able to induce pulses of Smad4 in the absence of TGFβ stimulation, indicating the important role of crosstalks between TGFβ signalling and other cellular pathways. Mathematical models based on in vitro experiments with HaCaT cells also illustrated two different signalling response patterns under transient or sustained TGFβ stimulation (Zi et al. 2011). With a single 30-second pulse of TGFβ, the concentration of phosphorylated Smad2 dropped within 4 hours and maintained the signal only at a very low level after 8 hours. On the other hand, continuous pulses at an interval of 30 minutes led to sustained phosphorylation of Smad2 that mimics the response of cells under prolonged TGFβ exposure, whereas pulses at an interval of 3 hours resulted in a cycling event, highlighting the important role of TGFβ stimulating duration and interval (Figure 2A). Other than stimulation time, dosage is also a major contributor to TGFβ signalling response in cells. Another mathematical model built by Khatibi et al. (2017) illustrated that the higher the TGFβ ligand concentration, the sooner the level of phosphorylated Smad2 saturates. It should be noted, in this model, the authors stated that Smad2 phosphorylation level in tumour cells of different stages will be variable under the same TGFβ stimulus, and this is represented by a fast, short-durational peak of Smad2 phosphorylation in early-stage tumour cells that returns to almost baseline within 100 hours, and prolonged, sustained high level Smad2 phosphorylation in late-stage tumour cells that lasts for over 500 hours. The combination of these data demonstrates the complexity of TGFβ signalling response under the influence of different ligand concentrations, treatment time and cell types (Figure 2B). Therefore, researchers need to be highly mindful when drawing conclusions on TGFβ signalling behaviour with experiments that merely look at single cell lines under one specific condition, and the reference value of such conclusions on other cellular or in vivo models need to be carefully considered.

The role of TEVs in modulating the cancer environment through autocrine and paracrine signalling is increasingly recognised by researchers. TEVs are shown to be secreted by cancer cells and stromal cells for the communication, cross-stimulation or -suppression and self-enhancement of signalling amongst cells in the tumour microenvironment. In particular, TGFβ ligands were found to be contained in these vesicles, with the ability to stimulate surrounding cells and facilitate cancer progression as its free ligand form (Shelke et al. 2019). For instance, a study on cancer-associated fibroblasts-derived exosomes in epithelial ovarian cancer has identified upregulated levels of TGFβ1 and its correlation with the EMT of ovarian cancer cells that are treated by the exosomes (Li et al. 2017). As extracellular vesicles carry intracellular molecules and shuttle to neighbouring cells continuously, the presence of ligands in the ECM may not truly represent the level of active signalling in cells. Experiments with human mesenchymal stem cells (hMSCs) have illustrated that TGFβ-containing exosomes derived from human mast cells HMC-1 are able to stimulate sustained signalling in hMSCs when compared with free TGFβ ligands (Shelke et al. 2019). Interestingly, the authors have also found that exosomes can facilitate the uptake of intracellular TGFβ into endosomes while avoiding lysosome fusion. These events indicate that TGFβ signal transduction through exosome trafficking can bypass extracellular communications and result in prolonged signalling across the cell population. Thus, current experimental techniques of treating cancer cells with a single dose or pulses of TGFβ ligands may not reflect the actual stimulation received by the cells. Combined with intracellular regulations of the pathway, TEVs may sustain a long-term cyclic signalling pattern within cancer cells as they are periodically secreted and uptaken by the cells (Figure 2A). There are yet to be other factors that may contribute to the pattern of TGFβ signalling, however the complexity of TGFβ signalling dynamics has already been demonstrated, with the functional implications to be discussed in the next section.

3.3. Subsequent functional importance of TGFβ signalling dynamics

Despite the fact that more experimental results are required to obtain a better understanding of TGFβ signalling dynamics, logical hypotheses can be formed based on the known functions of TGFβ and its association with different stages of malignancy. Figure 3 illustrates the hypothetical explanations for suspected dynamic TGFβ signalling and the potential reasons for the lack of success of anti-TGFβ-targeted therapies. Cancer cells are constantly experiencing stress and changes in their surrounding environment, and the adaptation to new situations requires rapid adjustment of cellular responses, which are unlikely to be controlled at the genetic level by random mutations. TGFβ is a universal regulator of cell adhesion, motility and survival, and it mediates other signalling pathways and subsequent cellular processes promptly after ligand-receptor binding, which makes it a suitable converging point for controlling cell responses and cell behavioural changes. It has already been shown that early and late-stage cancer cells respond differently to TGFβ, and on average, the increased level of TGFβ signalling is correlated with metastasis and advanced cancer (Robson et al. 1996; Ivanović et al. 2006). By dissecting each aspect of TGFβ signalling dynamics, the pro-metastatic role of TGFβ may be better explained.

Figure 3. Opinions on the current lack of understanding of TGFβ signalling. The differential behaviours of cancer cells spatially and timewise during metastasis have been previously recognised (right). This implies the underlying mechanisms of intensity and durational dynamics of TGFβ signalling during cancer progression and metastasis (middle), which is potentially the result of intracellular regulation and exosomal signalling (left).

Previous studies have shown that as a tumour suppressor, TGFβ signalling in the early stages of cancer is maintained at a low level due to the presence of functional regulators that limits its function (Pardali and Moustakas 2007). Thus, the effect of signalling dynamics may be negligible as the response to TGFβ is low and transient. By analysing TGFβ signalling under different cell cycle stages in tumour cells, it was found that TGFβ induces cell growth inhibition in G1/S phase of the cell cycle, and only triggers apoptosis when the cells were in G2/M phase (Song 2007). A possible conclusion drawn by the authors is that TGFβ-induced growth arrest does not interfere with the survival of tumour cells, but in contrary, the pause of cell cycle progression as a result of TGFβ signalling may prepare the cells for EMT, which only occurs in G1/S phase (Song 2007). This coincides with the role of TGFβ in suppressing tumour cell growth and promoting cancer cell EMT and migration. With the evolution of tumour cells and the change in extracellular environment, high levels of TGFβ signalling are often seen in advanced tumours to promote cell migration through EMT and modulate the tumour stroma for metastasis. During dissemination, however, the mesenchymal phenotype of cancer cells is not in favour of their attachment to the stroma and extravasation (Yang et al. 2018; Ribatti, Tamma, and Annese 2020). Important mediators of epithelial traits, such as E-cadherin for cell junctions, are transcriptionally inhibited by TGFβ signalling, and therefore high TGFβ signalling level may hinder the ability of cancer cells to make primary contact with the endothelium and subsequently colonise the new environment (Wendt et al. 2011; Gunasinghe et al. 2012). A previous study using a breast cancer mouse model has shown that in cancer cells constitutively expressing TGFβ, intravasation and single cell dissemination were enhanced, while lung colonisation and metastatic growth wwere reduced, highlighting the interference of strong TGFβ signalling in cancer cell seeding (Giampieri et al. 2009). Signalling oscillation of a single cell and signalling heterogeneity within a collective migrating cluster are potential ways that cancer cells adapt to preserve the flexibility of fast adaptation to the constantly-changing conditions during dissemination. Cyclic TGFβ signalling may allow the cells to retain a partial-mesenchymal/partial-epithelial phenotype, therefore when encountering a suitable homing site, circulating cancer cells can quickly convert into a full-epithelial phenotype and become tumorigenic again (Figure 2A).

Under the control of intracellular regulation mechanisms developed uniquely in cancer cells, the oscillatory signalling pattern benefits metastasis by enabling the cells to be constantly adapting to the external environment. This achieves a balance between the need of cancer cells to gain invasive ability and the advantage of being opportunistic during extravasation and colonisation. It also gives a potential explanation to why ligand traps and small molecule kinase inhibitors on TGFβ receptor activity failed to perform high efficacy on suppressing cancer development, as the regulation of TGFβ signalling is predominantly controlled intracellularly.

4. Developing advanced methods for research and better TGFβ-targeted cancer therapies

At the fundamental level, experimental studies need to form a strong theoretical basis for the development of anti-TGFβ therapies. To understand the dynamics of TGFβ signalling in cancer, more focus should be given to the change of signalling intensity over time and across a cell population, alterations in the regulatory mechanisms and EV-induced signalling. Currently in the literature, although high intensity TGFβ signalling has been observed in cancer, there remains to be a lack of knowledge on how the regulation of TGFβ signalling is altered in cancer to cause this increase in signalling intensity. The dynamic signalling pattern may also be ignored, as the signalling levels within cells are often averaged due to technical limitations. Further, additional factors such as the involvement of immune cells, stromal cells and exosomes in the tumour microenvironment should be taken into consideration when analysing the effect of TGFβ on cancer. Most importantly, research on TGFβ signalling mechanisms should not be limited by currently established experimental conditions.

The current improvement in technologies that allow close monitoring of signalling at single cell levels is particularly important to allow for a deeper understanding of TGFβ signalling dynamics (Goldman, Swedlow, and Spector 2010; Swedlow 2010). The use of 3D models to recapitulate cancer cell behaviour, drug response and stroma composition is emerging, as the limitations of monolayer cell cultures have been more and more recognised (Fang et al. 2016; Pape, Emberton, and Cheema 2021). Further, advanced live microscopy on 3D structures is able to provide temporal and spatial information on cell signalling, as demonstrated in Drosophila embryonic cell tracking (Tosi and Campbell 2019). Not limited to enhanced equipment, new computational software has been developed to enable precise cell tracking and automatic analysis of the signalling pattern. Some available analytic tools and applications of such digital techniques have been summarised in the paper from Emami and group (Emami, Sedaei, and Ferdousi 2021).

Several methods utilised in the previous studies of TGFβ signalling have already shown the capacity to demonstrate cellular heterogeneity and dynamic signalling patterns. These include the use of adenoviral reagents containing CAGA-luciferase or CAGA-Td-Tomato reporter constructs to track TGFβ/Smad3 signalling activity in cancer cells both in vitro and in vivo (Luwor et al. 2011; Chen, Ware, et al. 2018; Fonseca Teixeira, Iaria, and Zhu 2022). Specifically, single-cell tracking with Ad-CAGA-Td-Tomato in MDA-MB-231 cells showed that individual cancer cells have different levels of TGFβ signalling that are correlated with their migration distance, which is reflective of their EMT status and motility (Luwor et al. 2015). Interestingly, cells with high intensity of TGFβ signalling response showed a medium level of motility, whereas those with low intensity of TGFβ signalling response demonstrated the most migration (Luwor et al. 2015). It may be a consequence of the long half-life of reporter fluorescence protein, thus when the signalling response has already been attenuated, it requires time for the level of detectable signal in the cells to drop. This is implicative of a time-delay between the real-time response and the detected response in the cell tracking method, which has to be taken into account when analysing the results to avoid misinterpretations. Therefore, although emerging methods and techniques have shown promising directions on the evaluation of TGFβ signalling dynamics in cancer, conclusion of the results still needs to be carefully considered based on the actual experiment conditions.

Treatments with the same dosing and timing to all selected patients are generally applied in previous clinical trials for TGFβ-targeted therapies. However, as TGFβ signalling exerts contrasting and integrated effects on cancer progression at different stages of malignancy, patients are expected to have different responses to treatment, unlike in animal studies where all animals are implanted with tumour cells at the same time and will be in similar cancer stages at the time of treatment. Uniformed treatments may result in no or adverse effects in a proportion of recipients, which is a possible explanation for the unsuccessful outcomes of many clinical trials on anti-TGFβ therapies. For example, treatment for patients that have circulating tumour cells but yet to establish a secondary metastatic site may increase the chance of cancer cell seeding, due to the effect of TGFβ inhibition on reversing the mesenchymal phenotype and promoting cell adhesion. Also, as individual cancer cells within the same population may have different TGFβ signalling levels at the time of treatment, the treatment may only impact a sub-group of cells that are positive for TGFβ signalling, while other cells with low or no active TGFβ signalling at that time escape the effect of treatment and re-establish full signalling after the treatment effect diminishes within the body (Figure 2B). Additionally, an inappropriate treatment interval may artificially create cyclic signalling behaviour by forcing the cells to silence TGFβ signalling for a period of time and allowing the signal to rebound potentially greater than the initial peak post-treatment. For small molecule inhibitors or antibodies, the duration of TGFβ signalling inhibition is dependent on the in vivo half-life of the drug and is associated with the tissue clearance rate, which is expected to be different between mouse models and humans. These factors need to be taken into account when designing TGFβ-targeted therapies.

5. Conclusion

By studying the dynamic pattern of TGFβ signalling in cancer cells and the mechanisms behind it, the failure of current therapeutic development may be explained and a better model for therapeutic design could be established. This will also raise the possibility that dynamic signalling may be observed in other oncogenic pathways, and further studies are required to review the current understandings on the biochemistry of cancer for precision medicine.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

Data sharing is not applicable to this article as no new data were created or analysed in this study.

Additional information

Funding

The author(s) reported there is no funding associated with the work featured in this article.