Atypical interactions of integrin α_Vβ₈ with pro-TGF-β1

Significance

Integrins are cell-surface molecules that link extracellular ligands to the cytoskeleton. Force exerted by the cytoskeleton that is resisted by the ligand is thought to be important in activating the integrin by stabilizing an extended-open conformational state with high affinity for ligand. However, integrin α_Vβ₈ does not interact with the cytoskeleton in the same way. Here, we show that, although the closely related integrins α_Vβ₈ and α_Vβ₆ bind the same ligand, pro-TGF-β1, their conformational responses to ligand binding and regulation by metal ions are quite different. These differences correlate with their distinct linkage to the cytoskeleton.

Abstract

Integrins α_Vβ₆ and α_Vβ₈ are specialized for recognizing pro-TGF-β and activating its growth factor by releasing it from the latency imposed by its surrounding prodomain. The integrin α_Vβ₈ is atypical among integrins in lacking sites in its cytoplasmic domain for binding to actin cytoskeleton adaptors. Here, we examine α_Vβ₈ for atypical binding to pro-TGF-β1. In contrast to α_Vβ₆, α_Vβ₈ has a constitutive extended-closed conformation, and binding to pro-TGF-β1 does not stabilize the open conformation of its headpiece. Although Mn²⁺ potently activates other integrins and increases affinity of α_Vβ₆ for pro-TGF-β1 25- to 55-fold, it increases α_Vβ₈ affinity only 2- to 3-fold. This minimal effect correlates with the inability of Mn²⁺ and pro-TGF-β1 to stabilize the open conformation of the α_Vβ₈ headpiece. Moreover, α_Vβ₈ was inhibited by high concentrations of Mn²⁺ and was stimulated and inhibited at markedly different Ca²⁺ concentrations than α_Vβ₆. These unusual characteristics are likely to be important in the still incompletely understood physiologic mechanisms that regulate α_Vβ₈ binding to and activation of pro-TGF-β.

Integrins are cell-surface adhesion molecules that mediate cell–cell, cell–extracellular matrix, and cell–pathogen interactions. In mammals, 24 different integrins are formed by specific, noncovalent association of 18 α-subunits with 8 β-subunits (1–5). Almost all integrins link to the actin cytoskeleton through talin and kindlin-binding motifs in their β-subunit cytoplasmic domains and thus provide traction for cell migration. The only exceptions are integrin α_Vβ₈, which binds through its β₈-subunit to differentially express in adenocarcinoma of the lung (DAL-1, also known as Band 4.1B), and integrin α₆β₄, which links through its β₄-subunit to intermediate filaments (6).

Here, we focus on the atypical integrin α_Vβ₈ and compare it to integrin α_Vβ₆. Integrins α_Vβ₆ and α_Vβ₈ are specialized for binding to and activation of pro-TGF-β1 and -β3. The pro-TGF-βs are biosynthesized and stored in tissues as latent forms. The dimeric TGF-β growth factor is kept latent by noncovalent association with its dimeric prodomain, which in turn is linked noncovalently and through disulfide bonds to a “milieu molecule” that stores the latent complex in the extracellular matrix or on the cell surface for subsequent, integrin-dependent activation. Integrins bind to an RGD motif that is present in the prodomains of pro-TGF-β1 and -β3. However, integrin binding alone is not sufficient for latent TGF-β activation by α_Vβ₆; force is also required. Experiments suggest that tensile force is transmitted from the cytoskeleton to the integrin, is resisted by anchorage of TGF-β1 in the extracellular matrix or on cell surfaces, and results in distortion of prodomain straitjacket elements that loosely surround the growth factor followed by release of the growth factor (7, 8). In some systems, activation of TGF-β1 by α_Vβ₈ is dependent on matrix metalloprotease (9) whereas in others activation requires linkage of pro-TGF-β1 to a surface, suggesting a role for force exertion (10).

We examine here whether association of integrins α_Vβ₆ and α_Vβ₈ with different types of cytoskeletal adaptor proteins correlates with differences in regulation of the conformation and ligand-binding affinity of their ectodomains when they interact with pro-TGF-β1. Classical integrins that associate with talins and kindlins, including β₁, β₂, β₃, and β₇ integrins, as well as α_Vβ₆, exhibit three conformational states, and conformational change is proposed to be regulated by the adaptor proteins and tensile forces that the cytoskeleton exerts when integrins bind to immobilized ligands (Fig. 1) (2, 4). Such integrins exhibit a bent-closed conformation in which the α- and β-leg domains are bent at their knees and the ligand-binding headpiece associates with the lower legs. Knee extension gives an extended-closed conformation in which the legs straighten and the head moves much farther from the ectodomain C-termini that connect to the transmembrane domains. Finally, in headpiece opening, conformational change occurs in the ligand-binding domain in the β-subunit, the βI domain. The βI domain is inserted into the hybrid domain. Rearrangements that increase affinity at the ligand-binding interface are relayed by βI domain α-helix pistoning to the hybrid domain interface, resulting in pivoting of the hybrid domain away from the α-subunit (Fig. 1). The extended-open integrin conformation has much higher ligand binding affinity than the bent-closed or extended-closed conformations (11, 12). Previously, it has been reported that integrin α_Vβ₈ is constitutively extended and does not undergo headpiece opening in the presence of an RGD peptide (13). However, the affinity of the RGD peptide for α_Vβ₈ was not known, and RGD peptides that lack an LXX(I/L) motif present in pro-TGF-β1 and -β3 have substantially lower affinity for α_Vβ₆ than intact pro-TGF-β1 (14). RGD peptides are too small to be visualized in negative stain EM, and thus it was unclear whether RGD had remained bound, or had dissociated, before visualization of α_Vβ₈ in EM. Furthermore, even if RGD had bound, it was unclear whether intact pro-TGF-β1 would be capable of stabilizing the open headpiece conformation of α_Vβ₈. To stabilize the open conformation, ligands must bind with sufficiently higher affinity to the open than the closed conformation to drive conformational change to the higher energy open headpiece conformation. Because such stabilization has been seen with β₁, β₂, β₃, and β₆ integrins (2, 15), it was important to test whether α_Vβ₈ was truly resistant to the ligand-induced headpiece opening with a high-affinity, biological ligand. Moreover, affinity measurements are of fundamental importance for understanding integrin interactions with their biological ligands. Therefore, we have measured the affinity for pro-TGF-β1 of α_Vβ₈ and compared it to the affinity of α_Vβ₆.

Fig. 1.

Schematics showing the three major integrin conformational states.

The metal ion dependence of the ligand-binding affinity of α_Vβ₈ is also of interest. Integrin βI domains contain a Mg²⁺ ion and two Ca²⁺ ions. The Mg²⁺ ion directly coordinates ligand at the metal ion-dependent adhesion site (MIDAS). The two Ca²⁺ ions bind nearby and stabilize the conformation of ligand-binding loops. The adjacent-to-MIDAS (ADMIDAS) Ca²⁺ ion moves ∼6 Å between the open and closed conformations. Mn²⁺ is a commonly used activator of integrins and has been shown to work for integrin α₄β₇ by replacing Ca²⁺ at the ADMIDAS, which differs in coordination geometry in the closed and open conformations (16, 17). For several integrins, Mn²⁺ can induce headpiece opening even in the absence of ligand binding (2). Thus, it is interesting to compare effects of Mn²⁺ on α_Vβ₆ and α_Vβ₈ and explore correlations between headpiece opening and the ability of Mn²⁺ to increase ligand-binding affinity. Furthermore, because Ca²⁺, Mg²⁺, and Mn²⁺ may all contribute to regulating the integrin headpiece opening, we wondered if ligand-binding affinity would be regulated differently by metal ions in α_Vβ₈ than in α_Vβ₆. Here, we report that integrin α_Vβ₈ is indeed atypical in conformation, in responsiveness to Mn²⁺ of its affinity for pro-TGF-β1, and in regulation of its ligand-binding affinity by Ca²⁺ and Mn²⁺.

Results

Negative Stain EM Comparisons of Integrin α_Vβ₈ and α_Vβ₆ Ectodomain and Headpiece Fragments and Their Complexes with pro-TGF-β1.

We carried out negative stain EM on six distinct preparations of α_Vβ₈ including complexes with pro-TGF-β1. For comparison, we also carried out negative-stain EM on two α_Vβ₆ ectodomain preparations. Integrin fragments and their complexes with pro-TGF-β1 were subjected to gel filtration (Fig. 2). Peak fractions were subjected to negative-stain EM, and >5,000 particles were subjected to multivariant grouping and averaging in 50 classes (Fig. 3 and Figs. S1–S8). The α_Vβ₈ ectodomain was extended with its headpiece in the closed conformation, i.e., with the upper β-leg including the hybrid domain swung in, toward the α_V subunit (Fig. 3 A and B). The α_V- and β₈-subunits are readily distinguished by the larger size of the β-propeller than the βI domain in the head. Discrete densities were clear in the α_V-subunit for the β-propeller, thigh, calf-1, and calf-2 domains. Similar orientations between the thigh and calf-1 domains in class averages suggested that thigh and calf-1 domains adopted a uniform orientation after extension at the α-subunit knee. The β₈-subunit densities were clear for the βI and the hybrid domain. The length of the hybrid domain density suggested that it might also include density for the PSI and I-EGF1 domains. No density was present for the β-subunit lower leg I-EGF2-4 and β-tail domains. Essentially, identical class averages were obtained for ectodomain preparations with and without a C-terminal coiled-coil clasp (Fig. 3 A and B). No density was evident for the lower β-leg in either preparation, suggesting that it is highly flexible. In contrast, α_Vβ₆ ectodomain preparations adopted both bent-closed and extended-closed conformations (Fig. 3 C and D). The proportions of bent-closed and extended-closed particles were ∼3:1 in clasped and ∼1:3 in unclasped particles, suggesting that the clasp stabilizes the bent conformation.

Fig. 2.

Gel filtration of integrin complexes with pro-TGF-β1. Superdex 200 gel-filtration profiles of clasped α_vβ₈ ectodomain (A), unclasped α_vβ₈ ectodomain (B), and α_vβ₈ headpiece (C) complex with pro-TGF-β1 in the presence of Mg²⁺ or Mn²⁺ in HBS buffer. To form a protein complex, 10 μg α_vβ₈ ectodomain or headpiece was mixed with 5 μg pro-TGF-β1 in 50 μL and incubated on ice for 30 min before injection.

Fig. 3.

Representative class averages of negatively stained α_vβ₈ and α_vβ₆ and their complexes with pro-TGF-β1. (A and B) The α_vβ₈ ectodomain (ecto) adopts an extended-closed conformation when either clasped (A) or unclasped (B). (C and D) The clasped and unclasped α_vβ₆ ectodomain exhibits both bent-closed and extended-closed conformations. The percentage of particles with each conformation is shown. (E–H) Clasped and unclasped α_vβ₈ ectodomain complexes with pro-TGF-β1 showed both 2:2 and 1:2 complexes. (I and J) The α_vβ₆ ectodomain 2:2 and 1:2 complexes with pro-TGF-β1. (K and L) The α_vβ₈ and α_vβ₆ headpieces adopt the closed conformation in the absence of bound ligand. (M and N) The 2:2 and 1:2 complexes of the α_vβ₈ headpiece and pro-TGF-β1. The headpiece remains closed. (O and P) The 2:2 and 1:2 complexes of the α_vβ₆ headpiece and pro-TGF-β1. Ligand binding induces headpiece opening. All samples were prepared in Mg²⁺/Ca²⁺, except the α_vβ₈ headpiece complex with pro-TGF-β1 was prepared in Mn²⁺/Ca²⁺. The α_vβ₆ class averages in I, J, L, O, and P were previously published (15, 18) and are shown for comparison. (Scale bar: 10 nm.)

Fig. S1.

Class averages of clasped α_vβ₈ ectodomain (5,445 particles) by multireference alignment and K-means classification.

Fig. S8.

Class averages of α_vβ₈ headpiece complexed with pro-TGF-β1 (4,401 particles) by multireference alignment and K-means classification.

Fig. S3.

Class averages of clasped α_vβ₆ ectodomain (4,481 particles) by multireference alignment and K-means classification.

Fig. S4.

Class averages of unclasped α_vβ₆ ectodomain (5,135 particles) by multireference alignment and K-means classification.

Fig. S7.

Class averages of α_vβ₈ headpiece (4,886 particles) by multireference alignment and K-means classification.

The α_Vβ₈ ectodomain formed a complex with pro-TGF-β1 that was stable to gel filtration in buffer containing 1 mM Mg²⁺ and 1 mM Ca²⁺ (Fig. 2 A and B). Negative-stain EM on the complex peak showed class averages representing both 2:2 and 1:2 α_Vβ₈:TGF-β1 complexes (Fig. 3 E–H). The integrins bound at the interface between their β-propeller and βI domains to the ring-shaped pro-TGF-β1. In both 2:2 and 1:2 α_Vβ₈ complexes, the headpiece was closed, with the β-subunit hybrid domain pointing toward the interface between the α_V thigh and calf-1 domains, as in the uncomplexed ectodomain. In contrast, in α_Vβ₆:TGF-β1 complexes the headpiece was open with the hybrid domain swung away from the α-subunit (18) (Fig. 3 I and J).

Because Mn²⁺ potently activates integrins, we examined whether Mn²⁺ combined with pro-TGF-β1 could induce headpiece opening of α_Vβ₈ and extended our studies to headpiece fragments. On its own, the α_Vβ₈ headpiece was closed (Fig. 3K), as was the α_Vβ₆ headpiece (Fig. 3L). We next examined particles from a 2:2 α_Vβ₈:TGF-β1 complex peak from gel filtration in 1 mM Mn²⁺ and 0.2 mM Ca²⁺ (Fig. 2C). Most class averages showed 2:2 complexes, and a minority showed 1:2 complexes (Fig. 3 M and N). In 2:2 complexes, the better-resolved α_Vβ₈ integrin headpiece, which was more coplanar with the grid, was clearly closed, whereas the hybrid domain was out of the plane for the other monomer (Fig. 3M). In 1:2 α_Vβ₈:TGF-β1 headpiece complexes, the headpiece was clearly closed with an acute bend of the hybrid domain with respect to the head (Fig. 3N). In contrast, in α_Vβ₆:TGF-β1 headpiece complexes, the headpiece was open, with an obtuse bend of the hybrid domain with respect to the head (Fig. 3 O and P).

Affinity of α_Vβ₈ for Its Biological Ligand pro-TGF-β1 and Regulation by Metal Ions.

We measured affinities of our integrin preparations for pro-TGF-β1 using surface plasmon resonance (SPR). We used amine coupling to immobilize pro-TGF-β1 on the sensor chip and integrin preparations as analytes (Fig. 4A). Global fits to single on- and off-rates with a 1:1 Langmuir binding model were excellent, as shown by the black sensorgram and gray fit curves in Fig. 4A. Affinities (expressed here as K_d values) for pro-TGF-β1 of the α_Vβ₈ clasped and unclasped ectodomains and headpiece ranged from 400 nM to 100 nM in 1 mM Mg²⁺ and 1 mM Ca²⁺ (Fig. 4 A and B). In contrast, the affinity for pro-TGF-β1 of the α_Vβ₆ clasped ectodomain in 1 mM Mg²⁺ and 1 mM Ca²⁺ was much higher with a K_d of 15 nM (Fig. 4 A and B). In 1 mM Mn²⁺ and 0.2 mM Ca²⁺, affinity of the α_Vβ₈ preparations increased only marginally by two- to threefold. In contrast, the affinity of α_Vβ₆ increased much more, by 55-fold to 0.3 nM (Fig. 4 A and B). The on-rate in 1 mM Mg²⁺ of the α_Vβ₈ clasped ectodomain was 10-fold slower than the on-rate of the α_Vβ₆ clasped ectodomain (Fig. 4 A and B), whereas the off-rate for α_Vβ₈ was faster than the off-rate for α_Vβ₆. Mn²⁺ had little effect on the on-rate compared with Mg²⁺ for either integrin. In contrast, the off-rate was only marginally decreased in Mn²⁺ for α_Vβ₈ whereas it was dramatically decreased by 50-fold for α_Vβ₆. The effects on the off-rate are consistent with conformational change to the open headpiece after ligand binding that is stabilized by Mn²⁺ in α_Vβ₆ and not in α_Vβ₈.

Fig. 4.

Kinetics measurements of α_vβ₈ and α_vβ₆ binding to pro-TGF-β1 by SPR. (A) Sensorgrams show the overlay of fitting curve (thicker gray) and experimental curve (thinner black) of SPR measurement. Constructs and metal ions used are indicated. Concentrations were 800, 400, 200, 100, 50, 25, and 0 nM for the α_vβ₈ ectodomain; 400, 200, 100, 50, 25, 10, and 0 nM in Mg²⁺ and 200, 100, 50, 25, 10, and 0 nM in Mn²⁺ for the α_vβ₈ headpiece; and 200, 100, 50, 20, and 0 nM in Mg²⁺ and 100, 50, 20, 10, 5, and 0 nM in Mn²⁺ for the α_vβ₆ ectodomain. (B) Affinities and kinetic rates. Values are mean ± difference from mean of two independent experiments.

We used fluorescence polarization (FP) to extend measurements from the heterogeneous phase of SPR to the solution phase. A small fluorescent pro-TGF-β3 peptide tumbled rapidly; binding to the much larger integrin slowed tumbling and was measured as an increase in FP. We first measured affinity for fluorescent peptide using saturation binding with integrin. Affinities for the pro-TGF-β3 peptide of the α_Vβ₈ preparations ranged from 42 nM to 86 nM, and affinity of the headpiece was roughly twofold higher than the ectodomain in the presence of 1 mM Mg²⁺ and 1 mM Ca²⁺ (Fig. 5 A, C, and E). In 1 mM Mn²⁺ and 0.2 mM Ca²⁺, affinity of the α_Vβ₈ preparations increased about twofold (Fig. 5 B, D, and F). These results are consistent with SPR measurement in showing that the α_Vβ₈ headpiece has slightly higher affinity to the ligand than the ectodomain and that Mn²⁺ only marginally enhances α_Vβ₈ affinity. For comparison with α_Vβ₈, we made similar measurements on α_Vβ₆. In Mg²⁺, α_Vβ₆ bound the TGF-β3 peptide with fourfold higher affinity than α_Vβ₈ (Fig. 5 G and I). Mn²⁺ increased affinity of α_Vβ₆ for the pro-TGF-β3 peptide by a further 22- to 30-fold (Fig. 5 H and J). Finally, inhibition of the FP assay with peptides from pro-TGF-β1 and -β3 that contained the RGDLXXI/L motifs showed threefold higher affinity to α_Vβ₈ for pro-TGF-β3 than for pro-TGF-β1 (Fig. 5 K and L).

Fig. 5.

The α_vβ₆ and α_vβ₈ affinity for peptide ligands using fluorescence polarization. (A–J) Saturation binding of α_vβ₆ and α_vβ₈ ectodomain (ecto) and headpiece (hp) preparations to FITC-labeled pro-TGF-β3 RGD peptide (FITC-GRGDLGRLKK). Fluorescent FITC-labeled peptide was used at 10 nM and in H and J also at 5 nM and 20 nM. Data in H and J were fit globally to different probe concentrations to account for the effect of high-affinity binding on ligand depletion. (K and L) Affinity of the α_vβ₈ headpiece for pro-TGF-β1 and pro-TGF-β3 peptides measured by competition with FITC-labeled pro-TGF-β3 peptide in the presence of 1 mM Mg²⁺/Ca²⁺. The error bars in each plot represent the mean ± SD of triplicates. Errors in K_d values represent the fitting error.

We next used FP to measure the influence of metal ions on TGF-β3 peptide ligand-binding affinity. Titration of Mg²⁺ in the presence of 1 mM Ca²⁺ showed that lower concentrations of Mg²⁺ were more effective in supporting ligand binding by α_Vβ₈ than by α_Vβ₆ (Fig. 6 A and B). Titration of Ca²⁺ in the presence of 1 mM Mg²⁺ showed that Ca²⁺ was required for ligand binding of α_Vβ₈ with an EC₅₀ of 19 μM, whereas a much lower concentration of Ca²⁺ was sufficient for ligand binding of α_Vβ₆ with an EC₅₀ of 0.8 μM. This value was lower than background Ca²⁺ levels present in laboratory solutions and required use of Ca²⁺–EGTA buffers for measurement. At higher concentrations, Ca²⁺ inhibited both integrins, with higher concentrations required to inhibit α_Vβ₈ than α_Vβ₆ (Fig. 6 C, D, and G). Mn²⁺ also showed distinctive effects on the two integrins. Mn²⁺ activated both integrins in the presence of 1 mM Ca²⁺ at lower effective concentrations than seen with activation by Mg²⁺. However, at higher concentrations Mn²⁺ inhibited ligand binding by α_Vβ₈, an effect that was not seen with α_Vβ₆ (Fig. 6 E and F). These results reveal several atypical features of α_Vβ₈ in the regulation by metal ions of ligand binding.

Fig. 6.

Divalent cation dependence of α_vβ₈ and α_vβ₆ binding to ligand. (A–F) Fluorescence polarization was used to measure integrin headpiece binding to FITC-labeled pro-TGF-β3 peptide in varying concentrations of Mg²⁺ or Mn²⁺ with 1 mM Ca²⁺ and in varying concentrations of Ca²⁺ with 1 mM Mg²⁺. Ca²⁺/EGTA buffer was used for Ca²⁺ concentrations lower than 10 μM. Lines show nonlinear least square fitting of dose–response curves to the mean of triplicates; in C–E, two separate lines show fits to EC₅₀ and IC₅₀ values. Plotted points show mean ± SD of triplicates. All curves show fits to a 1:1 Langmuir model to [S3] or [S4] (Fig. S9); i.e., the Hill slope is fixed at 1. All plots fit this model well, except for Mg²⁺ and Mn²⁺ dependence of α_vβ₆, which may suggest two enhancing effects at different concentrations. (G) Summary of results showing mean ± fitting error.

Discussion

Here, we demonstrate that integrin α_Vβ₈ is atypical in multiple respects. It is constitutively extended. Binding to its biological ligand, pro-TGF-β1, does not induce the open conformation of the α_Vβ₈ headpiece, in agreement with previous findings using a low-affinity RGD peptide (13). The ability of pro-TGF-β1 to induce the open headpiece conformation of α_Vβ₆ and not α_Vβ₈ correlates with higher affinity binding with α_Vβ₆ in Mg²⁺ and greater augmentation by Mn²⁺ of affinity with α_Vβ₆ than α_Vβ₈. Finally, the metal ion Mn²⁺ has a more complex effect on ligand binding by α_Vβ₈ than by α_Vβ₆.

We show that α_Vβ₈ is constitutively extended and that its headpiece does not open when it complexes with pro-TGF-β1 either in the presence of Mn²⁺ or Mg²⁺. Previously, in the absence of ligand, α_Vβ₈ was found to be largely extended, with ∼5% of unclasped but not clasped particles showing a bent-closed conformation, and Mn²⁺ was found to give little augmentation relative to Mg²⁺ of binding to the TGF-β1 prodomain (13). We did not find any unclasped or clasped class averages that clearly corresponded to bent-closed α_Vβ₈, although some class averages appeared to correspond to the head region only, and we cannot rule out a small fraction of the bent-closed conformation (Figs. S1 and S2). Although complexes between pro-TGF-β1 and α_Vβ₆ have previously been characterized in EM, the conformation in EM of the α_Vβ₆ ectodomain on its own was not included (18). We examined here the clasped and unclasped α_Vβ₆ ectodomain and found a ratio of extended to bent particles of about 1:3 and 3:1, respectively. This contrasts with the low frequency or lack of observation of bent α_Vβ₈ reported here and previously (13). Our class averages for bent α_Vβ₆ resemble the class average reported for bent α_Vβ₈ (13). Based on these overall results it appears that α_Vβ₈ can be present in a bent-closed conformation, but the bent conformation is less stable than the extended-closed conformation, whereas α_Vβ₆ is similarly stable in these two conformations. Previously, RGD peptide and Mn²⁺ were found not to open the α_Vβ₈ headpiece (13); however, we found it important to confirm these observations with pro-TGF-β1 for several reasons. First, the affinity of RGD for α_Vβ₈ is unknown, and ELISAs that showed binding of α_Vβ₈ to the TGF-β1 prodomain failed to show binding to the RGD-bearing ligand fibronectin (13). Second, it is common for ligands to dissociate from integrins during gel filtration or grid preparation as previously suggested for integrin α_Vβ₃ and RGD peptide (19) and observed here for 2:2 integrin:pro-TGF-β1 complexes that yielded a number of class averages with 1:2 integrin complexes or integrins alone (Figs. S5 and S6). Visualization of the ligand itself, which is not possible with small peptides in EM, is the best confirmation that a ligand is bound.

Fig. S2.

Class averages of unclasped α_vβ₈ ectodomain (6,306 particles) by multireference alignment and K-means classification.

Fig. S5.

Class averages of clasped α_vβ₈ ectodomain complexed with pro-TGF-β1 (4,065 particles) by multireference alignment and K-means classification.

Fig. S6.

Class averages of unclasped α_vβ₈ ectodomain complexed with pro-TGF-β1 (5,235 particles) by multireference alignment and K-means classification.

By covisualizing α_Vβ₈ and its biological ligand pro-TGF-β1 in complexes here in EM, we have definitively established that ligand binding does not stabilize the open headpiece conformation of α_Vβ₈. Integrin α_Vβ₆ served as a positive control in our study. The finding was verified with both integrin ectodomain and headpiece fragments in complexes of 2:2 and 1:2 stoichiometry and in Mg²⁺ and Mn²⁺. Previously, ligand binding has been shown to result in headpiece opening for integrins α_Vβ₃, α_IIbβ₃, α₅β₁, α_Xβ₂, and α_Vβ₆ (2, 15). For integrin α₄β₇, binding to ligand in Mg²⁺ induced an intermediate headpiece conformation that is hypothesized to mediate transient, rolling adhesion, whereas binding to ligand in Mn²⁺ induced an open conformation that is hypothesized to mediate firm adhesion (20). Thus, maintenance of a ligand-bound integrin headpiece in the closed conformation in Mg²⁺, and lack of headpiece opening for a ligand-bound integrin in Mn²⁺, are unprecedented.

We have measured the affinity of α_Vβ₈ for pro-TGF-β1, which is fundamental to understanding how α_Vβ₈ binds and activates pro-TGF-β1 in vivo. Because other integrins exist in an ensemble of conformations, measurements of their affinities represent an average of the affinity of each conformation weighted by the population of that conformation in the ensemble. By using Fabs of the allostery conformation-specific antibodies to stabilize integrin α₅β₁ in open, closed, and extended conformations, we have been able to measure the affinity intrinsic to each integrin conformation (12). The affinity of fibronectin for the extended-open conformation of α₅β₁ is 1.4 nM and for the bent-closed and extended-closed conformations is ∼9,000 nM. Affinity measurements here of α_Vβ₈ are unique for an integrin because, rather than measuring an average affinity for an ensemble of conformations, they measure the affinity of a single conformation, the extended-closed conformation.

Previous studies have suggested that α_Vβ₈ has high affinity; however, rather than measuring affinity, the relative amount of binding in flow cytometry or ELISAs of α_Vβ₈ was compared with another integrin such as α_Vβ₃ (13, 21). The 100-nM affinity of α_Vβ₈ for pro-TGF-β1 is indeed ∼100-fold higher than the affinity for biological ligand of the closed conformations of α₅β₁ (12). However, the fairest comparison is between α_Vβ₈ and α_Vβ₆ binding to the same ligand, pro-TGF-β1. This comparison suggests that we cannot consider the closed conformation of α_Vβ₈ to be a high-affinity conformation because α_Vβ₆, which can access an open conformation, binds to pro-TGF-β1 with 30-fold higher affinity in Mg²⁺ and 500-fold higher affinity in Mn²⁺.

Thus, although the 100-nM affinity of α_Vβ₈ for pro-TGF-β1 appears high for a closed integrin conformation, it is far lower than the affinity for pro-TGF-β1 that α_Vβ₆ achieves with headpiece opening. The affinity of α_Vβ₈ for pro-TGF-β1 is also ∼100-fold lower than achieved with binding of the open headpiece conformation of α₅β₁ to fibronectin. Studies with α₅β₁ suggest that Mn²⁺ both shifts the conformational equilibrium toward the open headpiece conformation and raises its intrinsic affinity (12). An increase in intrinsic affinity of the closed conformation must account for the two- to threefold higher affinity in Mn²⁺ compared with Mg²⁺ of α_Vβ₈ for pro-TGF-β1 found here because EM showed that α_Vβ₈ had a closed conformation when bound to pro-TGF-β1 in both Mn²⁺ and Mg²⁺. On the other hand, the more profound increase in affinity of 55-fold in Mn²⁺ for α_Vβ₆ suggests that Mn²⁺ additionally shifts the equilibrium so that the α_Vβ₆ conformational ensemble contains a higher proportion of the open headpiece in Mn²⁺ than in Mg²⁺. Additionally, the presence of a small proportion of the open headpiece in the α_Vβ₆ conformational ensemble in Mg²⁺ is likely to account for the 30-fold higher affinity for pro-TGF-β1 in Mg²⁺ of α_Vβ₆ compared with α_Vβ₈. Given the ∼1,000-fold higher affinity of integrin open than closed states, a small percentage of the open conformation can make a major contribution to ensemble affinity (12).

We found unusual effects of metal ions on α_Vβ₈ affinity for ligand. In integrin βI domains, the MIDAS metal ion forms a direct coordination to an acidic residue in the ligand. The MIDAS is flanked on opposite sides by the synergistic metal ion-binding site (SyMBS) and the ADMIDAS. Physiologically, it appears that Mg²⁺ is bound to the MIDAS and that Ca²⁺ is bound to the SyMBS and ADMIDAS. Mn²⁺ can replace the metals at all three sites (22, 23). In conformational change from closed to open, the ADMIDAS metal ion moves ∼6 Å toward the MIDAS, and its coordination sphere alters from pentagonal bipyramidal that favors Ca²⁺ to octahedral that favors Mn²⁺ and Mg²⁺ (23, 24). Mutations of metal ion-coordinating residues in integrin α₄β₇ showed that the SyMBS was required for synergistic enhancement of ligand binding by Ca²⁺ and Mg²⁺ (16). In contrast, in integrins α₄β₇ and α₅β₁, the ADMIDAS was required for inhibition at higher concentrations by Ca²⁺ and appeared to be the site at which Mn²⁺ competed with Ca²⁺ to enhance ligand binding (16, 25). Integrin α_Vβ₈ resembled other integrins in showing synergism between Ca²⁺ and Mg²⁺ (16, 26, 27). Thus, ligand binding by α_Vβ₈ in 1 mM Ca²⁺ was dependent on Mg²⁺ with an EC₅₀ of 75 μM, and binding in 1 mM Mg²⁺ was dependent on Ca²⁺ with an EC₅₀ of 19 μM. Similarly, binding in 0.2 mM Ca²⁺ was dependent on Mn²⁺ with an EC₅₀ of 1.6 μM. These results may reflect synergism between Ca²⁺ at the ADMIDAS and Mg²⁺ or Mn²⁺ at the MIDAS. Most unusual was inhibition of ligand binding to α_Vβ₈ by Mn²⁺ at higher concentrations with an IC₅₀ of 13 mM. Such inhibition was not seen with α_vβ₆ or reported to our knowledge for any other integrin. Inhibition by Mn²⁺ may reflect substitution for Ca²⁺ at the SyMBS or ADMIDAS metal ion site in the closed conformation. In other integrins, Mn²⁺ is thought to selectively replace the ADMIDAS metal ion in the open conformation, which is distinctive in both metal ion location and coordination geometry (23, 24).

Our characterization here of α_Vβ₈ bound to its biological ligand pro-TGF-β1 has definitively shown that ligand binding does not stabilize an open conformation of α_Vβ₈ (13). Our measurements here of the affinity of α_Vβ₈ for ligand and comparative measurements of α_Vβ₆ provide unique information. Although previous studies have attempted to address whether α_Vβ₈ has high or low affinity for ligand, ligand-binding affinity has not been measured. Quantitation shows that the closed conformation of α_Vβ₈ has higher affinity than the closed conformation of some integrins, but is certainly lower in affinity than the open conformation of integrin α₅β₁ for fibronectin (12) or the affinity of the basal ensemble of conformations of α_Vβ₆ for pro-TGF-β1. Currently, we can only measure a population-weighted average affinity for pro-TGF-β1 of the α_Vβ₆ conformational ensemble; its closed conformation might well have a similar affinity to that of α_Vβ₈. The pro-TGF-β1–binding integrins α_Vβ₆ and α_Vβ₈ do appear under basal conditions to have unusually high affinity for ligand compared with other integrins, including three other α_V integrins. High affinity for ligand may reflect the requirement that these integrins not only bind but also transmit force from the cytoskeleton to the prodomain to release TGF-β, as suggested in at least some cellular systems by the requirement for a covalent linkage between pro-TGF-β1 and a milieu molecule for activation by α_Vβ₆ and α_Vβ₈ (10; but see ref. 9). Evidence is mounting that activation of most integrins requires force exertion by the actin cytoskeleton and resistance by the ligand to stabilize the extended-open integrin conformation compared with the bent-closed and extended-closed conformations (11, 28, 29). As α_Vβ₈ couples to a distinct type of cytoskeletal network through the DAL-1/Band 4.1B adaptor (6), it may either not be subjected to and regulated by cytoskeletal force or be subjected to force of a different magnitude than applied by the actin cytoskeleton. Our studies provide a quantitative framework for comparisons among integrins that are activated by distinctive cellular mechanisms.

Materials and Methods

The cDNA encoding the signal sequence and mature residues of the α_v ectodomain (1–960) or headpiece (1–594) with the M400C mutation and Gly inserted prior to residue 400 (15) was cloned into pcDNA3.1-Hygromycin^(-) vector. The cDNA encoding the mature residues of the β₈ ectodomain (1–639) or headpiece (1–456) with the V259C mutation was cloned into the ET10 vector (14). The expression constructs were cotransfected in HEK293S Gnt1^−/− cells. Clonal cell lines were selected and protein was purified as described (14). Human pro-TGF-β1 with a R249A cleavage site mutation was prepared as described (15).

Negative-stain EM was as described (19, 30, 31). Purified fresh protein from gel-filtration peaks was loaded on glow-discharged carbon grids and fixed with uranyl formate. Low-dose images were acquired with an FEI Tecnai-12 transmission electron microscope at 120 kV and a nominal magnification of 52,000×. Image processing was performed with SPIDER (32) and EMAN (33) as described (30). About 5,000 particles were picked manually and subjected to multireference alignment and K-means classification.

SPR studies were performed using a Biacore3000 instrument (GE Healthcare). The pro-TGF-β1 with a R249A cleavage site mutation was immobilized on a CM5 chip through amine coupling. Purified α_vβ₈ ectodomain or headpiece was injected at 20 μL/min in HBS buffer (20 mM HEPEs, pH 7.5, 150 mM NaCl) containing either 1 mM Mg²⁺ and 1 mM Ca²⁺ or 1 mM Mn²⁺ and 0.2 mM Ca²⁺. The surface was regenerated with a pulse (50 μL/min, 30 s) of 15 mM HCl at the end of each cycle. Kinetics were analyzed with Biacore evaluation software. A 1:1 Langmuir binding model was applied for experimental data fitting, and kinetic parameters were fit globally to sensorgrams at different analyte concentrations. Low χ² values (0.6–1.7) indicated good fits.

Fluorescence polarization was in HBS buffer with FITC-labeled pro-TGF-β3 RGD peptide (FITC-Aminocaproic acid-GRGDLGRLKK). In saturation binding assay, α_vβ₈ and α_vβ₆ were serially diluted in 1.4-fold decrements and mixed with 10 nM of probe (for α_vβ₆ ectodomain and headpiece in the presence of Mn²⁺, 5, 10, and 20 nM of probe were used for global fitting to account for the ligand depletion effect for high-affinity binding) in the presence of 1 mM Mg²⁺ and 1 mM Ca²⁺ or 1 mM Mn²⁺ and 0.2 mM Ca²⁺ at 20 °C for 30 min. Fitting FP as a function of integrin concentration to [S1] (Fig. S9) at fixed probe concentrations yielded the K_d values for fluorescent pro-TGF-β3 peptide. When multiple probe concentrations were used, FP was globally fit to both probe and integrin concentrations using [S1] (Fig. S9) to yield K_d values. In competitive binding assays, unlabeled pro-TGF-β1 and pro-TGF-β3 peptides were serially diluted in 1.4-fold decrements and mixed with 10 nM of probe and 100 nM of α_vβ₈ headpiece in HBS buffer containing 1 mM Mg²⁺ and 1 mM Ca²⁺ and incubated at 20 °C for 30 min. FP and competitor concentrations at fixed integrin and probe concentrations were fit to [S2] (Fig. S9) using the known integrin affinity for the probe to yield the competitor peptide K_d values. To test the effects of cations, 100 mM of Mg²⁺, Ca²⁺, or Mn²⁺ were serially diluted in twofold decrements and mixed with 100 nM α_vβ₈ or 20 nM α_vβ₆ and 10 nM of probe in HBS buffer with other indicated cations. To avoid interference by background Ca²⁺ present in laboratory water, concentrations of free Ca²⁺ of 100 μM and below were achieved by including 1 mM of EGTA using a total Ca²⁺ concentration calculated as described (34). The mixture was equilibrated at 20 °C for 30 min, and FP data were recorded on a Synergy NEO HTS plate reader. FP and cation concentrations were fit to [S3] (Fig. S9) to yield the EC₅₀ values for cations that only activated binding. For cations that both activated and inhibited, FP and cation concentrations were fit to [S4] (Fig. S9) to obtain EC₅₀ and IC₅₀ values.

Fig. S9.

Fitting functions. [S1] Fitting function for FP saturation binding assay. [S2] Fitting function for FP competitive binding assay. [S3] Fitting function for cation activating effect. [S4] Ftting function for cation activating and inhibiting effect at different concentrations. FP_obs is the measured FP. FP_L and FP_Integrin·L are FP of free FITC-labeled pro-TGF-β3 peptide and integrin-bound complex, respectively. [Integrin]_tot, [L]_tot, and [C]_tot are total concentrations of integrin, FITC-labeled pro-TGF-β3 peptide, and competitor, respectively. [Integrin]′ is the concentration of competitor-free integrin, either with or without bound FITC-labeled pro-TGF-β3 peptide. $K_{\text{d}}^{\text{L}}$ and $K_{\text{d}}^{\text{C}}$ are affinities of integrins for pro-TGF-β3 peptide and competitors, respectively. In [S3] and [S4], FP₀ is the FP without added cation; FP_max is highest value of FP reached at any cation concentration; [Cation]_tot is total concentration of cation or total concentration of Ca²⁺ not bound to EGTA; EC₅₀ and IC₅₀ are the cation concentrations at the inflection point where half-maximum change in FP was observed. Flags are assigned to data points depending on whether they are used to fit EC₅₀ (Flag = 1) or IC₅₀ (Flag = −1).