Structural basis of latent TGF-β1 presentation and activation by GARP on human regulatory T cells

See allHide authors and affiliations

Science  23 Nov 2018:
Vol. 362, Issue 6417, pp. 952-956
DOI: 10.1126/science.aau2909

Visualizing TGF-β1 regulation by GARP

Regulatory T cells (Tregs) can suppress immune responses through a variety of mechanisms. One such mechanism involves the activation of a surface-bound latent form of the cytokine transforming growth factor–β1 (TGF-β1). Within the cell, newly synthesized pro-TGF-β1 homodimers form disulfide bonds with the transmembrane protein GARP, which acts to chaperone and orient the cytokine for activation at the cell surface. Liénart et al. reveal how GARP interacts with TGF-β1, using a crystal structure in which the complex was stabilized using a Fab fragment from a monoclonal antibody (MHG-8) that binds to the complex. In so doing, they also demonstrate how MHG-8 prevents membrane-associated TGF-β1 release. These structural and mechanistic insights may inform treatments of diseases with altered TGF-β1 functionality and dysfunctional Treg activity, including cancer immunotherapy.

Science, this issue p. 952


Transforming growth factor–β1 (TGF-β1) is one of very few cytokines produced in a latent form, requiring activation to exert any of its vastly diverse effects on development, immunity, and cancer. Regulatory T cells (Tregs) suppress immune cells within close proximity by activating latent TGF-β1 presented by GARP (glycoprotein A repetitions predominant) to integrin αVβ8 on their surface. We solved the crystal structure of GARP:latent TGF-β1 bound to an antibody that stabilizes the complex and blocks release of active TGF-β1. This finding reveals how GARP exploits an unusual medley of interactions, including fold complementation by the amino terminus of TGF-β1, to chaperone and orient the cytokine for binding and activation by αVβ8. Thus, this work further elucidates the mechanism of antibody-mediated blockade of TGF-β1 activation and immunosuppression by Tregs.

Cytokines of the transforming growth factor–β (TGF-β) family exert widespread and diverse effects on metazoan cells (1). Owing to their potency and pleiotropy, family members such as the closely related TGF-β1, -β2, and -β3 isoforms are produced as latent, inactive cytokines, which require a tightly regulated step of extracellular activation to acquire the ability to bind their cognate heterotetrameric receptor (2, 3). The mechanisms of TGF-β activation appear to be isoform- and cell type–specific, but none have been fully elucidated on a structural basis.

TGF-β1 is the predominant isoform expressed by immune cells, including the immunosuppressive FOXP3+ CD4+ regulatory T cells known as Tregs (4). Immunosuppression by Tregs prevents autoimmunity but is detrimental in cancer or chronic infections (5). Tregs suppress other immune cells within close proximity by activating latent TGF-β1 on their surface (6). This results from the following multistep process. Newly synthesized pro-TGF-β1 homodimers form disulfide bonds with GARP (also known as LRRC32), a transmembrane protein retained at the surface of Tregs and a few other cell types (711). Pro-TGF-β1 is then cleaved by furin to produce latent TGF-β1, in which the C-terminal dimer, or mature TGF-β1 (mTGF-β1), remains noncovalently associated to the N-terminal dimer known as LAP (latency associated peptide). Latent TGF-β1 is inactive because the two LAP monomers, by means of their arm and “straightjacket” domains, form a ring around mTGF-β1, masking the interaction sites with the TβRI and TβRII receptor chains (12). TGF-β1 activation entails the release of mTGF-β1 from LAP. Tensile force exerted by the cytoskeleton and transmitted by integrin αVβ6 to LAP is hypothesized to unfasten and elongate the straightjacket, leading to the release and activation of mTGF-β1 (13). However, this mechanism of activation is restricted to cells expressing αVβ6. Tregs lack αVβ6. Instead, latent TGF-β1 presented by GARP on the surface of Tregs is activated by integrin αVβ8. This was shown using antibodies against either the GARP:(latent)TGF-β1 complex or the β8 integrin subunit, which block TGF-β1 activation and immunosuppression by Tregs (6, 14). Despite these advances in our understanding of TGF-β1 activation, the mechanism by which GARP structurally contributes to activation via integrin αVβ8 has eluded the field. Thus, we sought to determine the crystal structure of GARP bound to latent TGF-β1 by using a specific helper antibody to stabilize the complex.

We first coexpressed the extracellular domain of human GARP (GARPECD) and latent TGF-β1 and purified a soluble form of the GARPECD:TGF-β1 binary complex for x-ray crystallography. As this complex proved recalcitrant to crystallization, we reasoned that a ternary assembly with a Fab fragment might be more amenable to crystallization. We chose to use the Fab derived from the monoclonal antibody MHG-8 because it binds GARP:TGF-β1 complexes but not free GARP and blocks TGF-β1 activation by Tregs (6). Resolving such a complex would also provide key insights into therapeutically relevant epitopes.

Preparative reconstitution of GARPECD:TGF-β1:MHG-8Fab complexes yielded crystals that enabled determination of a structure to 3.1-Å resolution (fig. S1 and table S1). The structure features a tripartite assembly obeying a 1:1:1 stoichiometry, comprising one molecule of GARPECD, one complex of latent TGF-β1 (i.e., one LAPA/LAPB and one mTGF-β1A/mTGF-β1B dimer), and one copy of MHG-8Fab (Fig. 1A and table S2). The structure provided a snapshot of how latent TGF-β1 is sequestered by GARP via an intricate combination of covalent and noncovalent interactions. Tethering of latent TGF-β1 by GARP occurred on the opposite side from the RGD integrin-binding motifs in LAP (9, 13), in an orientation compatible with integrin αVβ8 binding (Fig. 1A).

Fig. 1 Structure of the GARPECD:TGF-β1:MHG-8Fab complex.

(A) (Left) Cartoon representations of the GARPECD:TGF-β1:MHG-8Fab structure as determined by x-ray crystallography. Latent TGF-β1 is composed of two LAP molecules (LAPA and LAPB) and two mTGF-β1 molecules (mTGF-β1A and mTGF-β1B). The red dashed line denotes the linker between the extracellular domain and the transmembrane domain of GARP. (Top right) Cartoon representation of a different view of the GARP:TGF-β1 complex, with MHG-8Fab not illustrated for clarity. (Bottom right) Schematic representation of protein components participating in the GARPECD:TGF-β1 complex. Yellow circles represent interchain disulfide bonds. (B) Cartoon representation and close-up view of the GARPECD. Secondary structure elements, modeled N-glycans, and intramolecular disulfide bonds are depicted. (C) Cartoon representation of a typical LRR fragment (R6) in GARP. By convention, the loop fragments connecting the concave to the convex side of the repeats form the lateral ascending face, whereas the loops connecting the convex to the concave side form the lateral descending face. (D) Cartoon representation and β-strand complementation observed in the GARP structure. The hydrogen bonding network is represented with black dashed lines. IC, intracellular; EC, extracellular; HC, heavy chain; LC, light chain; N-ter, N terminus; C-ter, C terminus; R, LRR (leucine-rich repeat); RNT, LRR N terminus; RCT, LRR C terminus; RGD, RGDLXX(I/L) binding motif for RGD-binding integrins.

GARPECD adopted a solenoid horseshoe structure with a nearly perfect circular curvature and featured 20 leucine-rich repeats (LRRs, R1 to R20) capped by N- and C-terminal motifs (Fig. 1B). Each repeat, comprising a consensus sequence of 21 to 29 residues, manifested a hydrophobic core running through the horseshoe solenoid (Fig. 1C). The concave side of the horseshoe was lined by an extended β sheet of parallel β strands contributed by each of the 20 repeats and the two capping motifs (Fig. 1, B to D). This side also displayed N-linked glycosylation at four of the five predicted sites (Fig. 1B). The convex side displayed an irregular mix of secondary structures connected by loops of various lengths (Fig. 1B and fig. S2). The GARP scaffold was further stabilized by two intrachain disulfide bonds: one (Cys26-Cys35) anchoring a β hairpin in the N-terminal cap and a second (Cys425-Cys436) anchoring the ends of an otherwise disordered loop protruding from the convex side of R15 (Fig. 1B). This structure of GARP may serve as a structural prototype for LRRC33, the only other LRR protein known to bind and present latent TGF-β1 on cell surfaces and that shares high levels of sequence identity with GARP (fig. S2) (15, 16).

The most mechanistically relevant insights from our structure center on the GARPECD:TGF-β1 interface (Fig. 2 and figs. S2 to S4). Half (28 of 56) of the GARP residues interacting with latent TGF-β1 were conserved in LRRC33 (fig. S2). Thus, LRRC33 may interact with latent TGF-β1 in a manner similar to GARP. Most residues (22 of 33) of latent TGF-β1 involved in the interaction with GARP were conserved in TGF-β2 and -β3, which also associate with GARP (figs. S4 and S5, A and B) (17). Thus, GARP is poised to actively regulate the bioavailability of all three TGF-β isoforms.

Fig. 2 GARP:TGF-β1 interfaces and plasticity of latent TGF-β1.

(A) Cartoon representation and close-up views on the three main GARP:TGF-β1 interfaces, outlined by dashed boxes. Details of the GARP:mTGF-β1B, GARP:LAPA, and GARP:LAPB interfaces are depicted in panels (i), (ii), and (iii), respectively. Proteins are shown in cartoon representation, and secondary structure elements are indicated. The main interfacing residues are shown as sticks, and polar interactions are depicted with black dashed lines. Only contacts observed in both copies of the complex in the asymmetric unit are shown. Disulfide bonds are represented by sticks and spheres, and sulfur atoms are colored yellow. (B) Close-up view on the GARP:mTGF-β1B interface, with structural superposition of the porcine mTGF-β1 from porcine latent TGF-β1 (chain D, PDB code 3RJR) showing the main conformational change observed in our structure. (C) Close-up view on the GARP:LAPA interface, with structural superposition of LAPB in our structure and the two LAP in porcine latent TGF-β1 (chains C and D, PDB code 3RJR) showing a second main conformational change in our structure. This is observed as an unwinding of the LAP helix α1. The disulfide bond between LAPA and GARP is depicted.

The GARPECD:TGF-β1 complex unmasked rarely seen combinations of structural features, not only due to the diversity and extent of its interaction landscape but also because it involved a twofold-symmetric growth factor (latent TGF-β1), bound by a molecular partner (GARP), that lacks any element of symmetry. Our structure revealed that latent TGF-β1 uses both copies of LAP and only one of the two mTGF-β1 modules to saddle the side of the GARP horseshoe. This results in three main interfaces, namely GARP:mTGF-β1B, GARP:LAPA, and GARP:LAPB (Fig. 2A). The binding footprint of latent TGF-β1 to GARP comprised the ascending lateral and convex surfaces roughly halfway into the horseshoe (R4 to R11) and buried a total of ~1750 Å2 of molecular surface area. Latent TGF-β1 interacted with a predominantly hydrophobic cushion in GARP via van der Waals contacts mediated by the N-terminal loop of LAPA and LAPB, by the loop comprising residues 323 to 337 in mTGF-β1B, and by helix α3 in mTGF-β1B (Fig. 2A, i).

Two structural features were particularly notable in the interfaces of GARP with the LAP modules. First, latent TGF-β1 was covalently tethered to GARP via two disulfide bonds: one between Cys33 in LAPA and Cys211 in GARP, the other between Cys33 in LAPB and Cys350 in GARP (Fig. 2A, ii and iii) (9). Covalent linkage between distinct proteins is an unusual mode of interaction when redox regulation is not at play. The LAPA–GARP disulfide was nested in a hydrophobic pocket capped by Trp297 of GARP (Fig. 2A, ii). Second, the N-terminal loop of LAPA inserted into a hydrophobic pocket formed by strands β12 in R6 and β14 in R7 of GARP (Fig. 2A, ii). This unanticipated structural feature led to complementation of the GARP horseshoe fold by a β-strand addition contributed by LAPA (strand β1A), supporting a role of GARP as a molecular chaperone for latent TGF-β1 biogenesis. GARP sealed this unusual interaction by inserting a β hairpin formed by residues 290 to 299. Notably, this hairpin was the only ordered part within a long segment (residues 280 to 317) that originates at R10 and was predicted to be intrinsically disordered (Fig. 1B and fig. S2). A binding mode of this nature was proposed to partially compensate for the high entropic cost associated with the binding of intrinsically disordered segments to partner molecules. However, structural snapshots of this phenomenon have only rarely been experimentally observed (18).

Our structure unveiled the structural plasticity of latent TGF-β1 and conformational changes that underlie its sequestration by GARP. The structure of latent TGF-β1 bound to GARP was globally similar to that of unbound porcine latent TGF-β1 (12), with structural disorder in the bowtie region that can associate with αVβ6 integrin (13). However, it displayed three drastic local conformational changes compared with all TGF-β1 structures known to date. First, helix α3 in mTGF-β1B became a 310 helix element (η1) and was rotated upward by 45° (Fig. 2B). Second, the preceding loop with residues 323 to 337 was substantially restructured toward the GARP concavity (Fig. 2B). Such extensive conformational switching resolved otherwise severe steric clashes and introduced Tyr336 of mTGF-β1B into a new structural context stabilized by a stacking interaction with Tyr137 of GARP and by hydrogen bonding between its main-chain atoms and Ser160 of GARP (Fig. 2A). Third, helix α1 in LAPB, which mediates interactions between LAPB and GARP, underwent marked unwinding while remaining physically tethered to GARP via the LAPB-GARP disulfide (Fig. 2, A and C). Unwinding of helix α1 was also seen in LAPA. Collectively, such extensive structural plasticity in latent TGF-β1 in a region far from the integrin binding site may explain why it can associate with two structurally very distinct classes of proteins. Namely, GARP or LRRC33 on one hand and LTBPs (latent TGF-β binding proteins) on the other (15).

The MHG-8Fab used to generate the crystals blocked TGF-β1 activation by human Tregs as efficiently as the full MHG-8 antibody (Fig. 3A). Thus, our crystal structure also revealed how an antibody can block TGF-β1 activation from GARP:TGF-β1 complexes and antagonize Treg suppression (6). MHG-8 bound a mixed conformational epitope, concomitantly contacting residues from GARP and latent TGF-β1 (Figs. 1A and 3, B to D). Specifically, CDR1 and CDR3 of the MHG-8 light chain and CDR3 of the heavy chain contacted the convex side of GARP, whereas the heavy chain also contacted the C-terminal arm domain of LAPB and the 310 helix η1 of mTGF-β1B via its CDR2, CDR3, and framework regions (Fig. 3, C and D, and figs. S6 to S8). We confirmed involvement of certain residues by measuring MHG-8 binding to GARP:TGF-β1 complexes including hemagglutinin (HA)–tagged mutant forms of GARP or latent TGF-β1 expressed in human embryonic kidney (HEK) 293T cells. Mutations at three sites in GARP (Leu140→Lys and Glu142→Leu, Arg143→Ala, or Arg163→Glu) and one site in mTGF-β1 (Lys338→Glu) substantially decreased binding of MHG-8 (Fig. 3B). A majority (9 of 17) of LAP or mTGF-β1 residues in contact with MHG-8 were not conserved in the corresponding positions of TGF-β2 and -β3. This explains why MHG-8 does not bind GARP:TGF-β3 complexes and only binds GARP:TGF-β2 with very low affinity (figs. S4 and S5C).

Fig. 3 Structural description of the MHG-8Fab:GARP and MHG-8:TGF-β1 interfaces.

(A) Immunoblot analysis of SMAD2 phosphorylation (pSMAD2) as a readout for TGF-β1 activation in human Tregs stimulated in the presence or absence of the indicated antibodies (20 μg/ml) or Fab (13 μg/ml). STIM, stimulation with anti-CD3– and anti-CD28–coated beads; mAb, monoclonal antibody; IgG1, immunoglobulin G1. (B) Binding of MHG-8 to transfected HEK293T cells expressing GARP:TGF-β1 complexes comprising the indicated N-terminally HA-tagged mutant forms of GARP or TGF-β1. Bar graphs indicate residual binding to a given mutant complex by comparison with the wild-type (WT) complex taken as 100% (mean of three to five experiments). Error bars represent the 99% confidence interval around the mean. Mutations in red are associated with significantly reduced binding compared with the WT, as their confidence interval does not contain the 100% binding threshold indicated by the dashed horizontal line. (C) (Left) Close-up view on the MHG-8Fab:GARP and MHG-8:TGF-β1 interfaces. GARP and TGF-β1 are shown in surface representation, and the variable region of MHG-8Fab is depicted as a cartoon. CDR1, -2, and -3 loops are displayed in orange, yellow, and black, respectively. (Right) Schematic representation of the overall structure and contact area by MHG-8VH (dark blue) and MHG-8VL (light blue). (D) Details of the MHG-8VL:GARP, MHG-8VH:LAPA, and MHG-8VH:LAPB:mTGF-β1B interfaces are depicted in panels (i), (ii), and (iii), respectively. Proteins are represented as cartoons, and secondary structure elements are indicated. Main interfacing residues are shown as sticks, and polar interactions are indicated by black dashed lines. Only contacts observed in both copies of the complex in the asymmetric unit are shown. CDR, complementarity determining region; VH, variable domain of the heavy chain; VL, variable domain of the light chain. Single-letter abbreviations for amino acid residues are as follows: A, Ala; C, Cys; D, Asp; E, Glu; F, Phe; G, Gly; H, His; I, Ile; K, Lys; L, Leu; M, Met; N, Asn; P, Pro; Q, Gln; R, Arg; S, Ser; T, Thr; V, Val; W, Trp; and Y, Tyr.

The antagonistic activity of MHG-8 on Tregs may result from its ability to impair the release of mTGF-β1 induced by combined pulling by integrin αVβ8 on one side and resistance or synergistic pulling by GARP on the other side of LAP (Figs. 3C and 4). By simultaneously contacting amino acids in the arm of LAPB and in GARP, MHG-8 may reinforce the fastening of the straightjacket, preventing the elongation required to release mTGF-β1 (Figs. 3C and 4). Alternatively, by cross-linking amino acids in GARP with the 310 helix η1 of mTGF-β1B, MHG-8 may prevent the release of mTGF-β1 and immobilize its 310 helix η1 in a position that is incompatible with binding to the TβRI chain of its receptor, even if the straightjacket is elongated and relaxed. In either scenario, MHG-8 sterically prevents mTGF-β1 release by engaging both GARP and latent TGF-β1 while still allowing binding of αVβ8, as demonstrated previously (14).

Fig. 4 Model for the integrin- and GARP-dependent mechanical activation of TGF-β1 and its antagonism by monoclonal antibody MHG-8.

A schematic representation of the GARPECD:TGF-β1:MHG-8Fab complex, with the MHG-8 binding footprint depicted in blue, is shown. Visualization of the covalent sequestration of latent TGF-β1 by GARP suggests how GARP may facilitate TGF-β1 activation by integrin αVβ8 on the Treg surface. Structural analyses of a pro-TGF-β1 monomer bound by the integrin αVβ6 head show how cytoskeletal forces elicit reshaping of latent TGF-β1 to release mTGF-β1 from LAP (13). Maximal harnessing of such forces depends on the direction of the force vector (20). Thus, GARP may provide strong anchoring to the cell surface, with its juxtamembrane region providing the necessary flexibility that allows for the optimal alignment of pulling-force vectors. Our crystal structure also suggests how an antibody can block TGF-β1 activation from GARP:TGF-β1 complexes on Tregs by cross-linking GARP to one LAP and one mTGF-β1 monomer, which sterically prevents mTGF-β1 release while still allowing integrin αVβ8 binding.

Thus, we elucidate the role of GARP in TGF-β1 biogenesis and activation, as well as the mechanistic principles underlying antibody-mediated blockade of TGF-β1 activation and immunosuppression by human Tregs. Visualization of this large molecular assembly illustrates the feasibility of blocking TGF-β1 activity emanating from a precisely defined and restricted cellular source, such as the surface of Tregs. This work also suggests how other TGF-β1–presenting proteins, such as LTBPs deposited in the extracellular matrix (2, 19), may operate as alternative sources of active TGF-β1 and may also be specifically targeted by antibody- or small-molecule–based reagents. Such structural and mechanistic insights may facilitate the design of particularly specific approaches to treat various diseases associated with altered TGF-β1 activity and/or dysfunctional Treg responses, most notably for cancer immunotherapy.

Supplementary Materials

Materials and Methods

Figs. S1 to S8

Tables S1 to S3

References (2124)

References and Notes

Acknowledgments: We thank S. Depelchin for editorial assistance and J. Cuende for early technical assistance. Proteros biostructures GmbH performed the crystallization and synchrotron radiation experiments. Funding: This work was supported by grants from WELBIO (grant BF-2014-01) and the ERC (Consolidator grant TARG-SUP 682818) to S.Lu., and from the VIB to S.N.S. S.Li. was supported by a Télévie fellowship (F.R.S.-FNRS). Author contributions: S.Li., R.M., B.v.d.W., H.D.H., M.S., P.G.C., S.N.S., and S.Lu. analyzed the data; S.Li., R.M., S.N.S., and S.Lu. wrote the manuscript; S.Lu. conceived of the study; S.Li., C.V., F.L., D.C., and J.S. performed the mutagenesis experiments, binding experiments and experiments with Tregs and produced and purified the mAb; B.v.d.W., H.D.H., and M.S. produced the GARP:latent TGF-β1 complexes; and R.M. performed crystallographic refinement and structure analysis with contributions from S.N.S. Competing interests: The authors declare the following financial competing interests: B.v.d.W., H.D.H., and M.S. are full-time employees of argenx BVBA. Patents pertaining to the results presented in the paper have been filed under the Patent Cooperation Treaty (International Publication Number WO 2016/125017 A1), with S.Lu., S.Li., B.v.d.W., H.D.H., P.G.C., and M.S. as inventors and UCLouvain and argenx as applicants. S.Lu. has received research support from argenx and owns stock options in argenx. Data and materials availability: Atomic coordinates and structure factor files have been deposited in the Protein Data Bank (PDB) with accession code 6GFF. The MHG-8 antibody is available from S.Lu. under a material transfer agreement with the UCLouvain. All other data needed to evaluate the conclusions of this study are present in the paper or the supplementary materials.

Stay Connected to Science

Navigate This Article