Making sense of latent TGFβ activation

TGFβ is secreted as part of a latent complex that is targeted to the extracellular matrix. A variety of molecules, `TGFβ activators,' release TGFβ from its latent state. The unusual temporal discontinuity of TGFβ synthesis and action and the panoply of TGFβ effects contribute to the interest in TGF-β. However, the logical connections between TGFβ synthesis, storage and action are obscure. We consider the latent TGFβ complex as an extracellular sensor in which the TGFβ propeptide functions as the detector, latent-TGFβ-binding protein (LTBP) functions as the localizer, and TGF-β functions as the effector. Such a view provides a logical continuity for various aspects of TGFβ biology and allows us to appreciate TGFβ biology from a new perspective.

been unambiguously identified. (Although our discussion is based primarily upon LTBP-1, the similar sequences and domain structures of the LTBPs suggest that most of our statements are generally applicable.) LTBP-1 exists in a range of sizes (125-210 kDa) owing to the use of two independent promoters as well as differences in splicing and glycosylation (Koski et al., 1999). Most forms of LTBP-1 have two proteasesensitive regions; proteolysis at the more N-terminal site can release a truncated form of LTBP-1 (or LLC) from the ECM (Fig. 2, step 4) (Taipale et al., 1994). The functions of LTBP-1 may vary depending on its size. For example, LTBP-1 that contains an N-terminal extension (LTBP-1L) generated by use of the upstream promoter associates more readily with the ECM than does LTBP-1 (LTBP-1S) formed by use of the downstream promoter (Olofsson et al., 1995).
In our model the three components of the LLC -TGFβ, LAP and LTBP-constitute a sensor (Fig. 1). This sensor consists of an effector (TGFβ), a localizer (LTBP) and a detector (LAP). We consider TGFβ to be the effector because it is the output of the sensor, LTBP to be the localizer because it interacts with the ECM, and LAP to be a detector because any activation mechanism must act on LAP, since LAP is sufficient to inhibit TGFβ bioactivity (Gentry and Nash, 1990). The characterization of the mechanisms controlling the liberation of TGFβ from the latent complex is central to the consideration of TGFβ action because the release of TGFβ determines the free TGFβ levels. Several mechanisms for the activation of latent TGFβ complexes are known (Munger et al., 1997;Koli et al., 2001), and a diverse group of activators, including proteases, TSP-1, the integrin αvβ6, reactive oxygen species (ROS) and low pH, can activate TGFβ. However, the biological advantage of releasing TGFβ as a latent complex and the relationships between the various activators are obscure. By considering the LLC as a sensor, we think that the role of the latent complex and its activators is clarified.

The latent TGFβ complex as a sensor
What general properties do sensors have and how do these properties relate to latent TGFβ? Consider, as an example, a smoke detector. Before it is used, it must be assembled correctly, placed in an appropriate location and put into a competent state (turned on). The sensor can then change in response to a stimulus (smoke) above a certain threshold, and this change relays information about the environment in the form of an effector (an alarm). Modification of the assembly or the location of the device can alter its effectiveness to respond to smoke.
These features of a smoke detector have analogies in the structure/function of the LLC. The latent TGFβ complex is a sensor that responds to extracellular perturbations and couples these events with the activation of latent TGFβ. As in the case of a smoke detector, the LLC must be appropriately assembled to function properly. The latent TGFβ complex is formed intracellularly and proTGFβ that fails to complex with LTBP is inefficiently secreted (Miyazono et al., 1991). Furthermore, failure to localize appropriately the latent TGFβ complex in the extracellular milieu alters the effectiveness of activation of latent TGFβ. Evidence to support this supposition derives from the ability of both inhibiters of tTGase (Kojima and Rifkin, 1993) and antibodies raised against LTBP-1 to block the activation of latent TGFβ (Flaumenhaft et al., 1993;Dallas et al., 1995;Nakajima et al., 1997;Gualandris et al., 2000). In addition, mice that are null for LTBP-3 or LTBP-4 demonstrate phenotypes consistent with altered TGFβ signaling (Dabovic et al., 2002;Sterner-Kock et al., 2002). Specific LTBP isoforms may differentially localize the latent complex, and different LTBP isoforms may preferentially associate with specific TGFβ isoforms. In fact, the third 8-cys domain of LTBP-4 is reported to bind only to TGFβ1 (Saharinen and Keski-Oja, 2000). As with many sensing devices, the TGFβ complex must be made competent to signal (i.e. turned on). Competence requires proteolytic separation of LAP from TGFβ (i.e. processing of proLLC into LLC; Fig. 2, step 2). ProLLC cannot be activated by any known mechanism, including heat (85°C for 10 min) or pH (1.5). Although proteolytic cleavage of proTGFβ may occur in the Golgi, this is not always the case. For example, multiple glioblastoma cell lines primarily secrete unprocessed proTGFβ as part of proLLC (Leitlein et al., 2001). To be a substrate for TGFβ activation, this proTGFβ must be processed at the furin protease site by a plasma-membrane-bound furin or another extracellular protease, such as plasmin [ (Lyons et al., 1988) our own observation]. Indeed, the addition of furin inhibitors to glioma cultures blocks proTGFβ processing. Once pro-TGFβ is processed, the complex is 'on' (competent), and it can be activated. In our model, we distinguish between the processing of proTGFβ (turning the sensor on or making it competent) and activating TGFβ. Thus, processing of proTGFβ is a regulated step affecting TGFβ bioavailability. Furthermore, it is interesting to speculate that proTGFβ performs a distinct signaling function from TGFβ (perhaps through integrin ligation) similar to the separate signaling capacities of proNGF and NGF (Lee et al., 2001).
We propose that the sensing function of the latent TGFβ complex resides mainly within LAP. This conclusion is supported by several facts: (1) the known TGFβ activators (e.g. plasmin, TSP-1 and αvβ6 integrin) interact directly with LAP (Lyons et al., 1988;Ribeiro et al., 1999;Munger et al., 1999); (2) the physical conditions that release active TGFβ (e.g. heat and pH extremes) denature LAP but not TGFβ (Lawrence et al., 1985); and (3) LAP adopts different conformations in unbound and TGFβ1-bound states (McMahon et al., 1996). Moreover, the relative lack of amino acid sequence conservation among LAP isoforms compared with TGFβ isoforms may provide a mechanism for diversification of TGFβ activation. For example, latent TGFβ1 and TGFβ3 can be activated by αvβ6, whereas TGFβ2 cannot (Annes et al., 2002;Munger et al., 1999). This is due to the presence of the integrinbinding sequence RGD in TGFβ1 and three LAPs but not TGFβ2 LAP. Sequence analysis reveals only 34-38% amino acid sequence identity among LAP isoforms (LAPβ1, β2, β3) compared with 75% identity among TGFβ isoforms (TGFβ1, 2, 3). However, there is considerable conservation of LAP isoform sequences across species (Table 1). The amino acid sequence identity shared by human TGFβ1 LAP and chicken TGFβ1 LAP is 90% (Table 1). We suggest that the relative lack of conservation between LAP isoforms allows LAPs to act as isoform-specific detectors. The divergence between LAP amino acid sequences may explain, in part, the isoform-specific functions of TGFβ in vivo, despite the overlapping expression patterns of the isoforms in vivo and their virtually identical functions in vitro. For example, TGFβ1 and TGFβ2 mRNAs are the predominant isoforms observed in the mouse heart during endocardial cushion and valvular genesis (Akhurst et The latent TGFβ sensor model. The figure depicts the sequential events in the bioavailability of TGFβ, from synthesis to signaling consequences, according to consideration of the TGFβ LLC as a sensor. Sensor assembly (1) occurs cotranslationally when the localizer (LTBP; L) is covalently linked to pro-TGFβ (D-E). As shown, the next step (step 2) is the proteolytic cleavage of the bonds between the detector (LAP; D) and the effector (TGFβ; E). This step turns the sensor 'on' or, in other words, makes the sensor competent (note that the timing of this step is variable and may occur after secretion). Once secreted, the sensor is stored in the ECM (step 3). Subsequently, the complex may be solubilized from the matrix (step 4) by cleavage of LTBP in the hinge region. This soluble form of TGFβ is still latent and may be activated (step 5). Under other conditions, activation of the matrix-bound sensor occurs (step 5′). Binding of liberated TGFβ (E) to its receptors (step 6) with subsequent signal transduction has multiple results (green arrows; A), including induction of TGFβ expression (B), enhanced expression of transcripts encoding TGFβ activators (C), and increased synthesis of ECM components (D). Millan et al., 1991), and both recombinant TGFβ1 and TGFβ2 function in in vitro assays of endocardial cell transformation (Nakajima et al., 1997). However, TGFβ2 -/but not TGFβ1 -/mice have defects in endocardial and valvular genesis (Sanford et al., 1997). Structural differences in LAP may provide a mechanistic basis for activation of TGFβ2 and not TGFβ1 in this setting.
The paradigm of latent TGFβ as a sensor also suggests that the response threshold of the latent TGFβ complex might be modulated. Although no examples have been reported, the existence of molecules that either bind LAP and prevent an activator from binding or, conversely, alter the conformation of LAP to facilitate recognition by an activating molecule is a likely possibility.

TGFβ activation, or tripping the TGFβ sensor
A variety of molecules, from protons to proteases, have been described as latent TGFβ activators (Fig. 2, steps 5,5′). A commonality among these activators is that they are all indicative of ECM perturbations. Indeed, given the profound effects of TGFβ on matrix homeostasis, the primary change that the TGFβ sensor detects may be alterations in the matrix. In this section we discuss some of the known TGFβ activators.

Proteolytic activation of latent TGFβ
A number of proteases including plasmin, MMP-2 and MMP-9 have been identified in vitro as latent TGFβ activators (Sato and Rifkin, 1989;Yu and Stamenkovic, 2000). Plasmin and MMP2/9 belong to the serine protease and metalloprotease families, respectively. These protease families, along with the adamalysin-related membrane proteinases, are the primary enzymes involved in ECM degradation (Werb, 1997). The ability of these enzymes to activate the latent TGFβ complex couples matrix turnover with the production of a molecule, TGFβ, that has a primary role in maintaining matrix integrity and stability (Ignotz and Massague, 1986;Verrecchia et al., 2001). There are three ways in which proteases might facilitate the activation of latent TGFβ. First, the protease-sensitive hinge region in LTBP is a potential target for the liberation of a still-latent remnant of the LLC, which would have to be further processed for activation (Taipale et al., 1994). Second, as discussed above, proteases can act in the extracellular environment to convert proLLC to LLC and thereby render the latent complex activation competent. Third, proteolytic cleavage of LAP, resulting in destabilization of LAP-TGFβ interactions, might release active TGFβ from its latent complex (Lyons et al., 1988). Degradation of LAP is an attractive mechanism for sensor activation because heightened levels of proteases are associated with several processes that involve increased TGFβ activation. However, thus far, mice that have null mutations in the genes that encode the known activating proteases do not demonstrate any phenotype consistent with TGFβ deficiency. This may reflect redundancy among the activating enzymes or the fact that these mice have not been studied in the correct context.

Activation by thrombospondin-1
The matricellular protein TSP-1 activates latent TGFβ (Schultz-Cherry and Murphy-Ullrich, 1993). The mechanism involves a direct interaction between TSP-1 and LAP (reviewed by Murphy-Ullrich and Poczatek, 2000). A short amino acid sequence (RFK) located between the first and second type 1 properdin-like repeats is believed to be responsible for latent TGFβ activation. Surprisingly, a tetrapeptide (KRFK) also functions as a TGFβ activator in vitro and in vivo (Crawford et al., 1998). This peptide probably acts by disrupting the non-covalent interactions between LAP and TGFβ. Interestingly, TSP-1 null mice demonstrate a partial phenotypic overlap with TGFβ1-null animals, thereby supporting the contention that TSP-1 is an in vivo activator of latent TGFβ (Crawford et al., 1998). TSP-1 facilitates wound repair in several ways: modulation of cell adhesion, promotion of angiogenesis, and reconstruction of the matrix (Frazier, 1991). The correlation between wounding and enhanced TSP-1 expression suggests that TSP-1 is an appropriate molecule for activation of the latent complex, since TGFβ plays a prominent role in wound healing (Border and Ruoslahti, 1992). TSP-1 is also expressed throughout development in a number of tissues, where it may function as a TGFβ activator (Iruela-Arispe et al., 1993;Majack et al., 1987).

Journal of Cell Science 116 (2)
The first integrin to be identified as a TGFβ activator was αvβ6 (Munger et al., 1999). The mechanism of activation depends upon a direct interaction between αvβ6 and the RGD amino acid sequence present in LAP β1 and LAP β3 (Fig. 1). The expression of αvβ6 is restricted to epithelia, and in most epithelia the integrin is normally expressed at low levels (Breuss et al., 1993). In response to wounding or inflammation, the expression of αvβ6 increases (Breuss et al., 1995;Miller et al., 2001). Therefore, epithelial cell upregulation of αvβ6 and subsequent TGFβ activation is a situation in which the cellular response to a process (inflammation) produces a potent suppressor of that process. Consistent with both the ability of β6 integrin to activate latent TGFβ and the pro-fibrotic effects of TGFβ (Border and Ruoslahti, 1992) is the observation that wild-type mice develop pulmonary inflammation followed by fibrosis in response to the inflammatory and profibrotic drug bleomycin, but integrin β6 -/mice have only a minor fibrotic response (Munger et al., 1999). In addition, global analysis of gene expression in the lungs of integrin β6 -/mice treated with bleomycin compared with similarly treated wild-type mice demonstrates a pronounced failure to induce expression of TGFβ-regulated genes in the mutant mice. These results indicate that fibrosis is the result of excess TGFβ produced by heightened expression of ανβ6 in response to the inflammatory stimulus. Since TGFβ dramatically increases the generation of αvβ6 by primary airway epithelial cells in vitro (Wang et al., 1996), it is likely that bleomycin triggers a feed-forward mechanism for coordinately up-regulating integrin expression and TGFβ generation. We suggest that fibrosis is the result of a failure to interrupt this feed-forward loop that is perpetuated by persistent ECM perturbation after wounding or inflammation.
Recently, Mu et al., reported that the integrin αvβ8 can activate latent TGFβ1 (Mu et al., 2002). It is interesting that activation by αvβ8 requires protease (MT1-MMP) activity in addition to the integrin. Although the exact roles of MT1-MMP and αvβ8 in this activation mechanism remain to be elucidated, the authors suggest that the integrin concentrates latent TGFβ on the cell surface, where it is subsequently activated by MT1-MMP. A cooperative interaction between different classes of latent TGFβ activator has been suggested previously (Yehualaeshet et al., 1999): the cell-surface-associated proteins (CD36 and TSP-1) concentrate latent TGFβ on the membrane where it is subsequently activated by plasmin.

Activation by reactive oxygen species (ROS)
Barcellos-Hoff and her co-workers showed that when ROS are produced in vitro (either by ionizing radiation or a metalcatalyzed ascorbate system) or in vivo after irradiation, latent TGFβ1 is activated (Barcellos-Hoff et al., 1994;Barcellos-Hoff and Dix, 1996). This is probably a result of scissions and side group modifications caused by hydroxyl radicals that disable LAP. The response of the TGFβ sensor to certain types of oxidative stress may reflect a need to produce TGFβ during processes such as inflammation and apoptosis that can cause ECM damage through the production of ROS.
Activation by pH Latent TGFβ present in conditioned medium is activated by mild acid treatment (pH 4.5) (Lyons et al., 1988), which probably denatures LAP, thereby disturbing the interaction between LAP and TGFβ. In vivo, a similar pH is generated by osteoclasts during bone resorption when an integrin-dependent sealing zone is generated between the bone and the cell (Teitelbaum, 2000). Since the bone matrix deposited by osteoblasts is rich in latent TGFβ, the acidic environment created by osteoclasts in vitro might result in latent TGFβ activation (Oreffo et al., 1989;Oursler, 1994).

TGFβ biology and the role of the sensor
The evidence that TGFβ is released in a latent form and must be activated is derived primarily from in vitro studies. There is little in vivo evidence demonstrating a requirement for latent TGFβ activation for several reasons, including the fact that measurement of changes in active TGFβ levels in tissues or animals is extremely difficult. In this section we discuss the in vivo evidence supporting the importance of extracellular TGFβ activation by examining the phenotypes of animals or people in whom specific steps in the post-translational assembly and/or processing of latent TGFβ are defective (Fig. 2). By incorporating the sensor model into our analysis, we have arrived at new interpretations of these phenotypically complex situations.
The effect of improper LLC assembly is illustrated in the phenotypes of mice that have null mutations in the LTBP-3 or LTBP-4 genes. LTBP-3 -/mice display bone phenotypes including osteoarthritis and osteopetrosis (Dabovic et al., 2002), which also occur in mice that have defective TGFβ signaling pathways resulting from either mutations in Smad3 (osteoarthritis) (Yang et al., 2001) or the expression of a dominant negative type II TGFβ receptor in osteoblasts (osteopetrosis) (Filvaroff et al., 1999). LTBP-4 -/mice develop pulmonary emphysema, cardiac myopathy and colorectal cancer (Sterner-Kock et al., 2002). It is interesting that the defects in LTBP-4 -/animals are consistent with both increased and decreased TGFβ activity: (1) emphysema has been associated with both increased and decreased TGFβ activity (Kaartinen et al., 1995;Zhou et al., 1996); (2) cardiac myopathy is associated with increased TGFβ activity (Schultz Jel et al., 2002); and (3) colorectal cancer is associated with a lack of TGFβ activity (reviewed by Gold, 1999). Thus, the phenotypes displayed by the LTBP-mutant mice are not necessarily described by a simple deficit in TGFβ.
Does consideration of TGFβ biology in terms of the sensor model clarify aspects of this situation? In the absence of a specific LTBP, TGFβ may be (a) inefficiently secreted and unable to localize to the ECM or (b) secreted in a complex with a different LTBP, presuming the cell expresses more than one LTBP isoform. According to the sensor model, these scenarios have varying effects on TGFβ activity. Whereas decreasing TGFβ secretion results in less TGFβ activity, eliminating or changing the isoform of LTBP is predicted to modulate the localization and/or activation pattern of the complex in a context-dependent manner. Therefore, it is not accurate to say that there is more or less TGFβ in these LTBP-null mice; rather, the distribution and timing of TGFβ activities may be modified. For instance, LTBP-3-null mice have increased bone density, which is similar to transgenic mice expressing a dominant negative type II TGFβ receptor under control of the osteocalcin promoter, but TGFβ1-null mice become osteoporotic rather than osteopetrotic as they age (Geiser et al., 1998). It is likely that the LTBP-3 -/phenotype emphasizes the effect of altered local distribution of a TGFβ in a cell or tissue type, whereas the TGFβ1-null phenotype illustrates the result of a global loss of the cytokine.
A localization defect can occur not only when there is a defect in LLC assembly but also if there is an alteration in ECM binding. This might occur if the binding partner for LTBP is missing or defective or if tTGase, which cross-links LLC to the matrix, is absent. However, mice with a null mutation in the TGase2 gene do not display a phenotype consistent with a global deficit in TGFβ (Nanda et al., 2001;De Laurenzi and Melino, 2001). This may indicate the existence of redundant TGases. We suggest that closer examination will reveal TGFβ-related changes in those tissues or cells that depend exclusively upon TGase2 for fixing of the sensor into the ECM.
An example of a human pathology related to altered TGFβ latency is Camurati-Engelmann disease (CED). This autosomal dominant disease results from mutations in the TGFβ1 LAP sequence and is characterized by hyperostosis and sclerosis of the base of the skull and long bones, respectively (Janssens et al., 2000;Kinoshita et al., 2000;Nishimura et al., 2002). Most of the mutations in CED occur at or close to the cysteine residues involved in the interchain bonds of the LAP dimer. Earlier work with mutated TGFβ cDNAs indicated that proper disulfide bond formation is required to produce latent TGFβ, because mutation of C223 and C225 yields constitutively active TGFβ (Brunner et al., 1989). Studies with fibroblasts from three patients with CED mutations at or close to C225 indicate that the mutant cells produce substantially more active TGFβ1 than do wild-type cells (Saito et al., 2001). Why the CED cells generate enhanced levels of active TGFβ is not clear, since disulfide bonds between the appropriate cysteine residues do form; however, the answer to this question may be clarified by consideration of the available data on CED in terms of the sensor model of latent TGFβ.
There are curious differences between the TGFβ produced by wild-type and CED fibroblasts. First, CED and normal cells produce similar amounts of total TGFβ1 as judged by TGFβ1 LAP immunoblotting; however, after acid activation of the latent TGFβ, medium conditioned by CED cells contains five times the amount of active TGFβ1 compared with medium conditioned by normal cells (Saito et al., 2001). Thus, there is a discrepancy between the amounts of immunoreactive and biologically active TGFβ1 produced by the two cell types. Second, there is a difference in the degree of proteolytic processing of proTGFβ1 by CED fibroblasts compared with wild-type cells (Saito et al., 2001). Whereas wild-type cells produce substantial amounts of unprocessed proTGFβ1, CED cells process all of the proTGFβ1 to LAP and TGFβ1. According to the sensor model, all of the latent complex produced by CED, but not wild-type, cells is in an activationcompetent state (i.e. the CED LLC is 'on' because it has been proteolytically cleaved) (Fig. 2, step 2). This is in contrast to the primarily proTGFβ1 produced by wild-type cells. This form of TGFβ is considered to be 'off' and cannot be activated by any known mechanism. Our definition of 'on' or competent latent TGFβ clarifies why there is significantly more TGFβ activity in CED, compared with wild-type, conditioned medium following acid activation, despite the fact that the cells secrete equal amounts of the TGFβ propeptide. Apparently, the CED mutation alters the susceptibility of LAP to proteolysis by furin and or other processing proteases. It is interesting to speculate that this same conformational change might make LAP more sensitive to activating proteolytic events. Therefore, we suggest that the latent TGFβ complex of CED individuals is assembled and localized normally but is hyper-responsive.
Two reports indicate that altered expression of molecules that activate the latent complex result in pathologies. The first report describes lung fibrosis after bleomycin treatment (Munger et al., 1999). In this example, fibrosis is impaired in mice missing β6 integrin, an activator important for generating TGFβ during inflammatory states (Munger et al., 1999). A second example is the developmental pulmonary emphysema observed in fibrillin-1-hypomorphic mice (E. R. Neptune, P. A. Frischmeyer, D. A. Arking et al., personal communication). These animals have a defect in the terminal septation of the alveoli that correlates with excess of both TGFβ and TGFβ signaling. It is likely that the defect in terminal alveolar septation in these mice is due to excess TGFβ, because higher levels of TGFβ activity were detected in the lungs of mutant animals, and the administration of TGFβ-neutralizing antibodies reverses the pathology. The lack of fibrillin might result in defective localization of LLC and subsequent TGFβ activation, because the LLC normally localizes with fibrillin-1 . Thus, the abnormal distribution of LLC results in inappropriate activation. An additional explanation as to why fibrillin-1 -/mice have altered TGFβ levels is revealed through consideration of latent TGFβ as a sensor. We propose that the altered ECM of the mutant mice cues cells to remodel the matrix and that this remodeling is associated with the inappropriate and persistent expression of a TGFβ activator.

Conclusion
We have conceptualized latent TGFβ as an ECM-localized sensor in order to unify our current understanding of TGFβ biology. In our model, the sensor comprises a localizer (LTBP), a detector (LAP) and an effector (TGFβ). Failure to localize latent TGFβ appropriately results in altered TGFβ activity. The role of the latent TGFβ complex in coordinating ECM perturbation with ECM reorganization is emphasized if one considers latent TGFβ as a sensor that responds to ECM damage or other extracellular perturbations. The storage of latent TGFβ in the ECM provides a mechanism for spatially and temporally linking perturbation with restructuring. The sensor model provides a framework for understanding the complex and varied nature of TGFβ activity: the primary role of TGFβ is to 'report' an alteration of the extracellular milieu and initiate a response.
The sensor model clearly separates two aspects of TGFβ biology that are often misunderstood: the processing of the proTGFβ (turning the sensor 'on') and the liberation of TGFβ from the latent complex. Visualizing the latent TGFβ complex as a sensor has offered insight into the somewhat confusing results reported for Camurati-Engelmann syndrome. Moreover, the consideration of active TGFβ formation in terms of a matrix-localized sensor makes it easier to imagine the existence of accessory molecules that interact with the sensor Latent TGFβ activation and either potentiate or dampen activation as well as the context-specific use of or localization by specific LTBP forms. In addition, a commonality of TGFβ activators is made apparent by representing TGFβ activation as a process involving sensor detection: all identified TGFβ activators are associated with ECM perturbation. Finally, the latent TGFβ sensor could allow the activities of the three nearly identical TGFβ cytokines to be distinguished, in part, through a diversity in LAP sequences that permits differential response to individual activators. By viewing latent TGFβ as a matrixlocalized sensor, we can understand TGFβ assembly, latency, activation and activity as coordinated events rather than as disparate aspects of TGFβ biology.