Structural determinants of integrin β-subunit specificity for latent TGF-β

Abstract

Eight integrin α-β heterodimers recognize ligands with an Arg-Gly-Asp (RGD) motif. However, the structural mechanism by which integrins differentiate among extracellular proteins with RGD motifs is not understood. Here, crystal structures, mutations and peptide-affinity measurements show that α_Vβ₆ binds with high affinity to a RGDLXXL/I motif within the prodomains of TGF-β1 and TGF-β3. The LXXL/I motif forms an amphipathic α-helix that binds in a hydrophobic pocket in the β₆ subunit. Elucidation of the basis for ligand binding specificity by the integrin β subunit reveals contributions by three different βI-domain loops, which we designate specificity-determining loops (SDLs) 1, 2 and 3. Variation in a pair of single key residues in SDL1 and SDL3 correlates with the variation of the entire β subunit in integrin evolution, thus suggesting a paradigmatic role in overall β-subunit function.

Integrins are α-β heterodimers that connect diverse extracellular ligands to the cytoskeleton and regulate cell growth and differentiation¹. The primary function of most of the 24 vertebrate integrins is to mediate cell adhesion and migration; in contrast, integrins α_Vβ₆ and α_Vβ₈ are specialized to activate TGF-β1 and TGF-β3 (refs. ^2,3). The similarity in phenotypes of mice deficient in TGF-β1 (ref. ⁴) to those of mice deficient in integrin α_Vβ₆ (ref. ²) or α_Vβ₈ (ref. ³) and to those of mice in which RGE in latent TGF-β1 (pro-TGF-β1) has replaced RGD⁵ demonstrates the importance of the RGD motif and integrins α_Vβ₆ and α_Vβ₈ in TGF-β1 activation in vivo. How integrins α_Vβ₆ and α_Vβ₈ achieve specificity, and how integrin β subunits in general contribute to ligand specificity, remain unclear. Little is known beyond mutational evidence for the importance of a disulfide-bonded loop (the β2-β3 loop) in the βI domain⁶ and the invariant binding of the metal ion–dependent adhesion site (MIDAS) to an acidic residue present in all integrin ligands^7–10. The issue of how the β subunit contributes specificity is particularly acute for the five RGD-recognizing integrins that contain the α_V subunit and differ only in having the β₁, β₃, β₅, β₆ or β₈ subunit.

Here, we report the molecular mechanism by which α_Vβ₆ achieves high specificity for the RGD peptide motif present in the prodomains of TGF-β1 and TGF-β3 and the determinants of specificity for integrin β subunits in general.

RESULTS

Pro-TGF-β1 activation correlates with high integrin affinity

Transfectants expressing α_Vβ₆ and α_Vβ₈ but not α_V, cotransfected with the β₁, β₃ and β₅ subunits, can activate pro-TGF-β (Fig. 1a), results in agreement with previous studies¹¹. In correlation with activation, α_Vβ₆ and α_Vβ₈, but not other α_V integrin transfectants, strongly bound 50 nM fluorescein isothiocyanate (FITC)-labeled pro-TGF-β1 (Fig. 1b).

Figure 1.

Activation and binding of pro-TGF-β1 by wild-type and mutant α_V integrins. (a) Indicated HEK293T transfectants assayed for TGF-β1 activation with mink lung luciferase reporter cells, measured as relative light units (RLU). Mock, control mock transfection. (b) Saturation binding of FITC–pro-TGF-β1 to HEK293T transfectants, shown as percentage mean fluorescence intensity (% MFI) of α_V P2W7 monoclonal antibody binding. Slashes denote ‘and.’ (c,d) Binding of peptides to α_Vβ₆ (c) or α_Vβ₃ (d) headpieces, measured by fluorescence anisotropy. Anisotropy is measured as millianisotropy units (mA), as (F_‖ – F_⊥)/(F_‖ + 2F_⊥) × 1,000, where F_‖ is the fluorescence intensity parallel to the excitation plane, and F_⊥ is the fluorescence intensity perpendicular to the excitation plane. Data show mean ± s.e.m. of technical triplicate samples. Peptides at the indicated concentrations were used with 200 nM α_Vβ₆ or 4 µM α_Vβ₃ headpiece and 5 nM of fluorescent peptide probe. K_d was calculated from IC₅₀ as described³⁰.

Ligands bind to the integrin headpiece, which contains the α-subunit β-propeller domain and thigh domain as well as the β subunit βI, hybrid, PSI and EGF1 domains. There are no previous measurements of α_Vβ₆ affinity for ligand despite the extensive characterization of specificity and comparison among TGF-β1, TGF-β2 and TGF-β3 and among integrins in adhesion and binding assays. α_Vβ₆ can be affinity purified with both the TGF-β1 prodomain and fibronectin². Adhesion assays and enzyme-linked immunosorbent assays have indicated stronger binding of α_Vβ₆ than α_Vβ₃ to pro-TGF-β1 and to pro-TGF-β3 and a lack of binding in the same assays to pro-TGF-β2 (refs. ^2,12). We accurately measured the affinity of monomeric pro-TGF-β3 peptide GRGDLGRL for the α_Vβ₆ and α_Vβ₃ headpieces with fluorescence anisotropy, using either direct binding of FITC-labeled peptide or competition with unlabeled peptide. All measurements were with the physiologic cations Mg²⁺ and Ca²⁺. Nonapeptides containing RGD from pro-TGF-β1 and pro-TGF-β3 bound to α_Vβ₆ with remarkably high affinity (10.3 and 8.5 nM, respectively; Fig. 1c). In contrast, the same peptides bound to α_Vβ₃ with 1,000-fold-lower affinity (Fig. 1d). Interestingly, the homologous peptide from pro-TGF-β2, which has SGD in place of RGD, also bound to α_Vβ₆ but with a 1,000-fold-lower affinity (8.5 µM) comparable to that of the GRGDSP peptide of fibronectin (2.5 µM; Fig. 1c). It is quite interesting that α_Vβ₆ binds to pro-TGF-β2 peptide with an affinity that is in a range typically found for integrin binding to biological ligands. These results suggest that further investigation is warranted of a role for integrins, possibly distinct from that of α_Vβ₆, in the activation of pro-TGF-β2.

α_Vβ₆ crystal structures

We turned to crystal structures to determine the basis for the unprecedented high affinity of α_Vβ₆ for pro-TGF-β and its peptides. Crystals of the α_Vβ₆ headpiece, with or without a pro-TGF-β3 undecapeptide soaked in, diffracted to 2.6 and 2.85 Å (Table 1), respectively, and contained two molecules per asymmetric unit, with almost identical structures. The headpiece adopts the closed headpiece conformation in the absence and presence of soaked ligand (Fig. 2a), i.e., with the hybrid domain swung in toward the α subunit and with the β1-α1 and β6-α7 loops and α7 helix in the βI domain in the closed conformation^8,13–15. Compared to those of β₃, the βI and PSI domains of β₆ are similar in structure, with greater differences in the hybrid domain (Supplementary Table 1).

Table 1.

Data collection and refinement statistics

	αVβ6	αVβ6 + TGF-β3 peptide
Data collection
Space group	C2	C2
Cell dimensions
a, b, c (Å)	184.5, 168.3, 101.8	184.4, 170.0, 102.4
α, β, γ (°)	90, 98.2, 90	90, 98.7, 90
Resolution (Å)	50.0–2.85 (2.95–2.85)a	50.0–2.60 (2.69–2.60)
Rmerge	17.5 (349)	12.9 (198)
I / σI	4.8 (0.4)	4.9 (0.5)
CC½ (%)b	98.2 (10.0)	98.8 (18.3)
Completeness (%)	97.4 (83.0)	96.4 (96.6)
Redundancy	2.4 (2.3)	2.4 (2.4)
Refinement
Resolution (Å)	50.0–2.85	50.0–2.60
No. reflections	69,928 (4,837)	91,038 (9,101)
Rwork / Rfree	23.6 (38.0) / 27.9 (41.7)	21.4 (37.1) / 25.9 (37.3)
CCwork / CCfree	93.0 (29.5) / 91.0 (19.7)	94.9 (39.1) / 92.6 (27.4)
No. atoms
Protein	16,485	16,626
Ligand/ion	− /12	166/14
Water	161	230
B factors
Protein	96.0	76.0
Ligand/ion	− /81.5	84.7/71.9
Water	56.3	51.1
r.m.s. deviations
Bond lengths (Å)	0.005	0.009
Bond angles (°)	0.76	1.3

^aValues in parentheses are for highest-resolution shell. The resolution for each crystal was determined at which CC_free of the highest-resolution shell for the finial model is about 20%.

^bPearson’s correlation coefficient between average intensities of random half data sets for each unique reflection³¹.

Figure 2.

Crystal structures and comparisons of the α_Vβ₆ headpiece. (a) Overall ribbon diagram of the α_Vβ₆ headpiece (with each domain in a different color) with pro-TGF-β3 peptide (magenta). (b) Conformational change of the βI α2-α3 loop in the absence and presence of pro-TGF-β3. Carbon color code: green, in absence of peptide α_V; cyan, β₆; yellow, in presence of peptide β₆; magenta, peptide aspartate. Metals are white or gray spheres. (c) pH dependence of binding affinity. Binding of FITC–pro-TGF-β3 peptide measured with fluorescence anisotropy is shown. Data show mean ± s.e.m. of technical triplicate samples. (d,e) Ligand binding of α_Vβ₆ to pro-TGF-β3 peptide (d) and α_Vβ₃ to cilengitide (e)⁷. Carbon color code: green, α_V; cyan, β₃ or β₆, with different shades for SDLs 1, 2 and 3; magenta, ligands. The MIDAS metal ion is a silver sphere. (f,g) Key residues that contribute to packing between SDLs 1, 2 and 3 in β₆ (f) and β₃ (g). SDL color code is as in d and e. Van der Waals surfaces around interacting side chains are shown as dots. (h) Phylogenetic tree for integrin β-subunit SDL sequences²⁹. Ligand-contacting residues in SDL1 and SDL3 in the X₁ positions are highlighted in pink. Residues that form packing interactions of SDL1 and SDL3 with SDL2 in the X₂ position are highlighted in orange. Cysteines forming disulfides are highlighted in yellow. Residues that coordinate metals are asterisked in orange (MIDAS), green (ADMIDAS) and cyan (SyMBS).

Three closely spaced metal ion–binding sites are present in the integrin βI domain: the synergistic metal ion–binding site (SyMBS), the MIDAS and the site adjacent to the MIDAS (ADMIDAS). α_Vβ₆ crystallized at pH 6.5 loses its SyMBS metal ion; furthermore, the SyMBS-coordinating α2-α3 loop also remodels and invades the ligand-binding pocket (Fig. 2b). Remodeling enables SyMBS residues Asn218 and Asp220 to point outward and to form three strong, 2.4- to 2.7-Å hydrogen bonds in place of Ca²⁺ coordination (Fig. 2b). Similar remodeling of the β₃-subunit α2-α3 loop in the absence of a SyMBS Ca²⁺ (refs. ^13,15) is blocked by the large side chains of residues that characterize its ligand-binding pocket, especially β₃ Arg214 and Tyr166 in place of β₆ Ala217 and Lys170 (Fig. 2f–h).

We hypothesized that crystallization at pH 4.5–6.5 might be responsible for variable loss of the SyMBS, MIDAS and/or ADMIDAS metal ions from α_Vβ₃ (refs. ^13,15) and α_Vβ₆, in contrast to occupation of all three sites in α_IIbβ₃ crystalized at higher pH^14,16. To test this hypothesis, we examined the effect of pH on affinity of α_Vβ₆ for the TGF-β3 nonapeptide. Indeed, fluorescence anisotropy demonstrated strong pH dependence with a particularly sharp decrease in affinity between pH 7 and pH 6 (Fig. 2c). Because many cells coexpress integrins with their ligands, including epithelial cells that coexpress α_Vβ₆ and pro-TGF-β1, it is possible that this pH dependence may contribute to the inhibition of ligand binding during biosynthesis in the Golgi (pH 6.0–6.7) and transport in endosomes (pH 6.3–6.5)¹⁷.

Ligand binding by α_Vβ₆

Soaking ligand into crystals restored a Ca²⁺-bound conformation of the SyMBS α2-α3 loop (Fig. 2b) and revealed how the TGF-β3 peptide binds with high affinity (Fig. 2d). Simulated annealing composite omit maps show excellent ligand density (Supplementary Fig. 1). Ligand binding induced a local ~1.5-Å displacement of the β1-α1 loop toward the aspartate of RGD and the MIDAS Mg²⁺ (Fig. 2b), as seen in intermediate states of other integrins with soaked-in RGD^7,9,13,16. Comparing the structure of ligand-bound α_Vβ₆ with six intermediate states of integrin α_IIbβ₃ between closed (state 1) and open (state 8), we found that the ligand-bound β₆ βI domain is similar to the intermediate state 2. In contrast, the ligand-free α_Vβ₆ structure is clearly closed (state 1)^14,16. The aspartate of RGD coordinated the MIDAS Mg²⁺ ion through one side chain oxygen and formed hydrogen bonds to NH groups of Asn218 and Ala126 through the other side chain oxygen (Fig. 2d). The arginine of RGD formed bidentate hydrogen bonds through its guanido group to the side chain of Asp218 in the α_V β-propeller domain (Fig. 2d), as in binding to α_Vβ₃ (Fig. 2e)⁷. Furthermore, as the ligand spanned the α_V-β₆ interface, the backbone of the RGD arginine formed a hydrogen bond to the side chain of Thr221 in the β₆ α2-α3 loop (Fig. 2d). A similar hydrogen bond to the ligand backbone can form with β₈ but not with β₃ or β₅, which have alanine in the position of β₆ Thr221 (Fig. 2e,h).

The largest conformational difference in the ligand-binding region between α_Vβ₆ and α_Vβ₃ is in the β2-β3 loop. This loop is displaced in β₆ relative to β₃ as a consequence of sequence differences in both the β2-β3 loop itself and in the β1-α1 and α2-α3 loops with which it interacts (Fig. 2f–h). The path of the β2-β3 loop is altered in β3 by the insertion of cis-Pro169 (Fig. 2g,h) as well as by π-cation bonds between β2-β3 residue Tyr166 and α2-α3 residues Arg214 and Arg216 (Fig. 2g). The three residues forming π-cation bonds are replaced in β₆ by Lys170, Ala217 and Ile219 (Fig. 2f). Furthermore, a hydrogen bond between Tyr185 in the β2-β3 loop and Asp129 in the β1-α1 loop constrains the conformation at the C-terminal portion of the β2-β3 loop in β₆ (Fig. 2f). Thus, backbone differences in the β2-β3 loop derive not only from the difference in this loop’s own sequence but also from differences in the sequences of loops that interact with the β2-β3 loop.

Strikingly, the TGF-β3 peptide forms an α-helix that extensively interfaces with the β₆ subunit (Fig. 2a,d). Immediately following the aspartate of RGD, the sequence 244-LGRLK-248 forms an amphipathic α-helix. TGF-β Leu244 binds in a β₆-subunit hydrophobic pocket formed by the side chain of Ala217 and the backbone of Asn218 in the α2-α3 loop; the backbone of Pro179 and the side chains of Cys180 and Ile183 in the β2-β3 loop; and the side chain of Ala126 in the β1-α1 loop (Fig. 2d). The aliphatic portion of the ligand Lys248 side chain contributes to burying Leu244. Ligand residue Leu247 further buries Leu244 and binds in the same hydrophobic pocket by interacting with the backbone and side chain of Ala126 in the β1-α1 loop and with the side chain of Ile183, the disulfide bond of Cys177 and Cys184, and the aromatic ring of Tyr185 in the β2-β3 loop. Thus, three different loops in the βI domain make contacts with the TGF-β ligand (Fig. 2d).

Integrin β subunits vary markedly at the positions in the β1-α1 and α2-α3 loops where Ala126 and Ala217 contact the amphipathic TGF-β α-helix (Fig. 2h). Ala126, in the β₆ 123-DLSAS-127 MIDAS motif, is conserved in β₈ but is a tyrosine in β₃ (Tyr122; Fig. 2g) and in the β₁, β₂ and β₇ subunits (Fig. 2h). Introduction of the tyrosine residue with the A126Y mutation substantially decreased both binding of pro-TGF-β1 and activation of TGF-β1 (Fig. 1a,b). Ala217 in the α2-α3 loop is a small residue (glycine or alanine) in most integrin β subunits but is a large arginine in the β₃ and β₅ subunits (Fig. 2h). The A217R and double A126Y A217R mutations completely abolished pro-TGF-β1 binding and activation (Fig. 1a,b).

Between the two disulfide-bonded cysteines in the β2-β3 loop, in which Ile183 and Tyr185 contact the pro-TGF-β α-helix, integrins are highly diverse⁶ (Fig. 2h). However, individual β₆ I183Y and Y185S mutations had no effect, and the double mutation only slightly affected α_Vβ₆ binding and activation of pro-TGF-β1 (Fig. 1a,b).

The integrin-binding loop in pro-TGF-β

In the ligand, the glycine residue preceding RGD extends back toward the amphipathic α-helix (Fig. 2d). Thus, Gly240 and Lys248 in 240-GRGDLGRLK-248 are only 8 Å apart (Fig. 2d). The sequence in between has an overall loop-like conformation, with Asp243 and Leu244 most buried in the binding pocket, which is centered on the β₆ subunit rather than the α_Vβ₆ interface (Fig. 2d). Because the two ends of the pro-TGF-β3 peptide are near each another and orient away from the integrin, the peptide complex is highly compatible with integrin binding to a pro-TGF-β3 macromolecule.

The integrin-binding residues identified here lie near the middle of an 18-residue segment that is disordered or only weakly ordered in the structure of pro-TGF-β1 (ref. ¹⁸) and can easily protrude from the shoulder region of pro-TGF-β to bind α_Vβ₆ in the helical conformation identified here. Interestingly, foot-and-mouth disease virus (FMDV) uses an RGD motif followed by an amphipathic sequence very similar to those in pro-TGF-β1 and pro-TGF-β3 (Fig. 3a)¹⁹ to bind integrin α_Vβ₆ and infect epithelial cells.

Figure 3.

Ligands of α_Vβ₆. (a) RGD sequences from pro-TGF-β and VP1 protein from FMDV. (b) Competitive binding affinities of TGF-β3 peptide truncations. Fluorescence anisotropy data are mean ± s.e.m. of technical triplicate samples scaled logarithmically. (c) Western blot of pro-TGF-β1 secreted by the indicated HEK293T transfectants, with antibody to the prodomain as described previously¹¹. (d) TGF-β bioassay of pro-TGF-β1 and its double-proline mutant. Data show mean ± s.e.m. of technical triplicate samples.

We tested the importance of the amphipathic α-helix for binding to α_Vβ₆ by helix truncation and mutation, truncating the C terminus of the TGF-β3 undecapeptide one residue at a time (Fig. 3b). The largest stepwise drops in affinity, about ten-fold each, occurred when Leu244 and then Leu247 were removed (Fig. 3b), thus verifying the importance of these residues. To disrupt the α-helix conformation upon binding of pro-TGF-β1 to α_Vβ₆, we mutated residues Ala220 and Thr221, which correspond to α-helix residues Gly245 and Arg246 in the pro-TGF-β3 peptide and have no contact with α_Vβ₆, to proline. These mutations had no effect on pro-TGF-β1 expression (Fig. 3c), results consistent with the postulated lack of importance of this region for structural stability. However, the double-proline mutant was deficient in its ability to be activated by α_Vβ₆ (Fig. 3d), thus supporting the importance of the α-helical conformation for interaction with pro-TGF-β1.

DISCUSSION

Specificity determinants of integrin β subunits

We have revealed specializations in an integrin β subunit that enable ligand recognition with high affinity and specificity. α_Vβ₆ recognizes not only RGD but also an LXXL/I motif that folds into an amphipathic α-helix fitting into a hydrophobic pocket composed solely of residues from the β₆ subunit. We observed these interactions with a TGF-β3 peptide bound to α_Vβ₆ in an intermediate state. With mutations, we tested the importance of the observed interactions in binding and activation of macromolecular pro-TGF-β1 by α_Vβ₆ on cell surfaces. These assays interrogate α_Vβ₆ in the open conformation, because in the absence of crystal-lattice contacts, pro-TGF-β induces α_Vβ₆ headpiece opening in the presence of Mg²⁺ and Ca²⁺ (ref. ¹⁸). Furthermore, we also measured binding in the presence of Mn²⁺, an activator of integrin headpiece opening²⁰. These results confirmed the importance of contact residues in the β1-α1 loop and α2-α3 loops in pro-TGF-β1 binding to open α_Vβ₆. Moreover, in the βI domain’s shifting to the open conformation, the α2-α3 and β2-β3 loops do not shift, and shifting in the β1-α1 loop (1.4 Å at Tyr122 in β3, equivalent to β₆ A126) is insufficient to change contact residues^8,16. Therefore, we conclude that the observations here on β₆ residues that determine specificity for pro-TGF-β1 and pro-TGF-β3 are independent of integrin conformational state.

In contrast to our α_Vβ₆ complex, complexes of α_Vβ₃, α_IIbβ₃ and α₅β₁ have revealed little interaction beyond that with RGD itself^7–9,16,21. The β₆ hydrophobic-pocket interaction with the amphipathic α-helix revealed here enables α_Vβ₆ to achieve ~1,000-fold selectivity for pro-TGF-β over the RGD motif present in fibronectin and ~1,000-fold selectivity over α_Vβ₃ for recognition of pro-TGF-β.

The nanomolar affinity measured here is unprecedented for a monomeric integrin ligand and for a measurement in Mg²⁺ and Ca²⁺. By comparison, RGD peptides show, in Mg²⁺ and Ca²⁺, half-maximal effective concentration (EC₅₀) values of 7 µM to 1 mM for α_IIbβ₃ (refs. ^22,23) and half-maximal inhibitory concentration (IC₅₀) values of 10–200 µM for α_IIbβ₃ (ref. ²⁴) and 1–20 µM for α_Vβ₃ (ref. ²⁵). The resting affinity of integrin α_Lβ₂ for its biological ligand is 700 µM (ref. ²⁶). The 10-nM affinity of α_Vβ₆ for ligand is thus unprecedented. Low integrin affinity is important to reverse adhesion in retracting regions of cells during cell migration²⁷. The extraordinarily high affinity of α_Vβ₆ may reflect a specialization to support activation of TGF-β rather than cell migration.

Previously, little had been known about the contribution of integrin β subunits to ligand recognition beyond that of the MIDAS. Our crystal structure shows that contacts with the amphipathic α-helix in the ligand, which confer high affinity for α_Vβ₆, are mediated by the βI-domain β1-α1, β2-β3 and α2-α3 loops. Mutations demonstrated that β1-α1 and α2-α3 are important both for binding to pro-TGF-β1 and for its activation. The β2-β3 loop has previously been shown by mutation to be important for ligand selectivity by α_Vβ₃ and α_Vβ₁ (ref. ⁶) and has been designated a specificity-determining loop (SDL)²⁸.

We propose to designate the β1-α1, β2-β3 and α2-α3 loops SDLs 1–3, according to their order in the amino acid sequence (Fig. 2d–h). In addition to contacting the amphipathic helix, the β1-α1 (SDL1) and α2-α3 (SDL3) loops contact the glycine and aspartate of RGD. The SDL designation for the three loops in the βI domain that bind ligand is analogous to the complementarity-determining region (CDR) designation for antibody and T cell–receptor chains, which also use three surface-exposed loops to contact ligand.

In the eight human integrin β subunits, SDL1 and SDL3 have the sequences D(V/L/F)SX₁SMX₂(D/N)(D/N) and (V/I)SX₁NX₂D(A/S/T)PE, respectively (Fig. 2h). The two alanines in β₆ that contact pro-TGF-β are in the X₁ position in each loop. Variation in the X₁ positions of SDL1 and SDL3 correlates with variation in the entire β subunit through evolution (Fig. (2h and ref. 29), thus suggesting a paradigmatic role in overall β-subunit function. Thus, residues in the X₁ positions in SDL1 and SDL3 (respectively alanine and alanine/glycine in the β₆ and β₈ subfamily; tyrosine and glycine in the β₁, β₂ and β₇ subfamily; tyrosine/leucine and arginine in the β₃ and β₅ subfamily; and asparagine and glycine in β₄) have a unique pattern in each subfamily, and within a subfamily variation is confined to chemically similar residues (alanine/glycine and tyrosine/leucine). Residues in the X₂ position in SDL1 and SDL3 pack against SDL2 and contribute to its conformational variation (Fig. 2f–h). SDL1 and SDL3 also bind metal ions (Fig. 2h), and thus their backbone conformations would not be free to vary unless metal ions were lost, but this loss would not be consistent with ligand binding. Switching SDL3 to a β₃-like sequence with an alanine-to-arginine substitution abolished pro-TGF-β1 binding and activation, whereas the alanine-to-tyrosine substitution in SDL1 led to partial loss of binding and activation (Fig. 1a,b).

As the outermost and only non-metal-binding SDL, SDL2 has a conformation that has been free to vary in evolution, as shown here in a comparison between integrins with identical α subunits, α_Vβ₃ and α_Vβ₆. Differences stem from the presence of a cis-proline in SDL2 of β₃ and packing interactions with X₂ residues in SDL1 and SDL3. In agreement with the lack of effect seen here of two SDL2 mutations, only one of the six residues between the two cysteines in SDL2 is identical in β₆ and β₈ (Fig. 2h), yet α_Vβ₆ and α_Vβ₈ bind and activate pro-TGF-β comparably well (Fig. 1a,b). This supports backbone-dependent contributions of SDL2 to integrin specificity.

Because integrins are important therapeutic targets, the identification of the three SDLs of integrin β subunits not only advances understanding of how β subunits contribute to integrin-ligand specificity but also advances the ability to rationally design antagonists.

ONLINE METHODS

Pro-TGF-β cell-surface binding and activation

α_V in a modified pEF1 vector and β₁, β₃, β₅, β₆ or β₈ mutants in pcDNA3.1 (−) vector were transiently transfected into HEK293T cells¹¹. Purified pro-TGF-β1 (ref. ¹⁸) was fluorescently labeled with fluorescein isothiocyanate (FITC) with the Pierce (Thermo Fisher Scientific) FITC labeling kit according to the manufacturer’s instructions. Cells were resuspended in HBS buffer (20 mM HEPES, pH 7.4, 137 mM NaCl, 5 mM KCl, 5.5 mM glucose and 1% bovine serum albumin) and incubated at room temperature for 30 min with 50 nM FITC–pro-TGF-β1 in the presence of 5 mM EDTA, 1 mM Mg²⁺/Ca²⁺ or 1 mM Mn²⁺/0.2 mM Ca²⁺ and subjected to flow cytometry without washing. To test the expression of different α_V integrins, cells were resuspended and incubated at room temperature for 30 min with 2 µg/ml P2W7 antibody (anti-α_V, Sigma-Aldrich, cat. no. I6778, monoclonal antibody specific for human integrin α_V (160 kDa); validation provided on the manufacturer’s website) and then stained on ice for 30 min with FITC–anti-mouse IgG (1:500) (Sigma-Aldrich). Cells were washed once and subjected to fluorescence flow cytometry. Ligand binding was measured as the mean fluorescence intensity (MFI) of pro-TGF-β1 after subtraction of the MFI in the presence of EDTA.

TGF-β assays used HEK293T cells transiently transfected with the α_V and β subunits as above along with wild-type or mutant human pro-TGF-β1 in pcDNA3.1 (−) and then cocultured with transformed mink lung cells expressing a luciferase gene under the control of a TGF-β1–inducible promoter^18,32.

α_Vβ₆ and α_Vβ₃ headpiece expression and purification

Soluble α_Vβ₆ headpiece was prepared similarly as in ref. 33. In brief, the α_V headpiece (residues 1–594) with the M400C mutation was followed by a 3C protease site, the ACID coiled coil, a strep II tag and a histidine tag. β₆ headpiece residues 1–474 with I270C or β₃ headpiece residues 1–472 with Q267C were followed by the 3C site, the BASE coiled coil, and a histidine tag. The cysteine mutations generated a disulfide bond that prevented α/β subunit dissociation. Proteins expressed in HEK293S GnTI⁻ cells with Ex-Cell 293 serum-free medium (Sigma) were purified with Ni-NTA affinity columns (Qiagen). Protein was cleaved by 3C protease at 4 °C overnight, passed through Ni-NTA resin and further purified with an ion-exchange gradient from 50 mM to 1 M NaCl, 20 mM Tris-HCl, pH 8.0 (Q fast-flow Sepharose, GE Healthcare) and gel filtration (Superdex 200, GE Healthcare).

Fluorescence anisotropy

Fluorescence anisotropy was in 150 mM NaCl, 1 mM Mg²⁺/Ca²⁺, 20 mM HEPES, pH 7.4 or buffer at the indicated pH with 5 nM fluorescence probe (FITC–pro-TGF-β3 peptide, FITC-aminohexanoic-GRGDLGRL). Binding affinities were calculated as described³⁰. In saturation binding assays, the anisotropy of the fluorescence probe was measured while α_Vβ₆ headpiece (starting at 2.67 µM) or α_Vβ₃ headpiece (starting at 75 µM) was serially diluted in three-fold decrements. Competition binding assays used 200 nM α_Vβ₆ or 4 µM α_Vβ₃ headpiece, 5 nM of fluorescent probe, and competing peptide serially diluted in three-fold decrement from 500 µM to 0.5 nM.

Crystallization, data collection and structure determination

Crystals in hanging drops were formed with 3 mg/ml α_Vβ₆ headpiece in 20 mM Tris-HCl, pH 8.0, 150 mM NaCl, 1 mM CaCl₂ and 1 mM MgCl₂ buffer (1 µl) and 1 µl reservoir solution of 20% PEG 4000, 0.1 M sodium cacodylate, pH 6.0 and 0.2 M ammonium sulfate. Identical v/v mixture of the protein and crystallization buffers yielded a pH of 6.5. Pro-TGF-β3 peptide (1 mM each Ac-HGRGDLGRLKK-NH₂, MgCl₂, and CaCl₂, 0.2 µl) was added to drops of ~1.5 µl (~130 µM final concentration) for 4 h before crystals were harvested. Immediately before flash freezing in liquid N₂, crystals were dipped in reservoir solution containing 25% glycerol with or without peptide, MgCl₂, and CaCl₂, each at 0.25 mM.

Diffraction data from GM/CA-CAT beamline 23-ID of the Advanced Photon Source (APS) at the Argonne National Laboratory were processed with XDS³⁴ with cross-correlation to determine the diffraction limit³¹. Structures were solved with molecular replacement by PHASER³⁵ with the α_Vβ₃ headpiece from PDB 4G1E as the search model¹⁵. The structure was refined with PHENIX³⁶, manually built with Coot, and validated with MolProbity³⁷. I/σI and CC_1/2 in the highest-resolution shell increase as a function of the number of diffraction images in plots generated with XDS. The resolution limit was first chosen generously on the basis of CC_1/2 of ~10% and after refinement was truncated to the resolution at which CC_free of the highest-resolution shell was ~20% before one final round of refinement was done at the final resolution cutoffs. Furthermore, R_work/R_free of the outer shell were 38.0%/41.7% and 36.5%/37.3% for the apo and peptide-soaked models, respectively (Table 1). These results show that the weak diffraction data in the outer shell contribute to structure determination. In the α_Vβ₆-headpiece model (PDB 4UM8), 95.7%, 4.1% and 0.2% of residues have backbone dihedral angles in the favored, allowed, and outlier regions of the Ramachandran plot³⁷, respectively. The MolProbity³⁷ percentile scores are 100 and 100 for clash and geometry, respectively. In the α_Vβ₆ headpiece TGF-β3 peptide complex model (PDB 4UM9), 95.8%, 4.1% and 0.1% of residues are in the favored, allowed and outlier Ramachandran regions, respectively and the MolProbity scores are each in the 98 percentile.