A systematic survey of conformational states in β1 and β4 integrins using negative-stain electron microscopy

Integrins are a family of cell adhesion receptors that mediate cell-cell and cell-extracellular matrix interactions and govern migration and anchorage for almost all kinds of cells. Mammalian genomes contain up to 18 α- and 8 β-subunits that combine to form 24 different heterodimers, each of which has an apparently unique ligand-binding profile and biological function (Humphries, 2000; Hynes, 2002). Crystal structures of the full-length extracellular domain have been determined for three integrins (αVβ3, Xiong et al., 2001, 2002; αIIbβ3, Zhu et al., 2008; and αXβ2, Xie et al., 2010; Sen et al., 2013) among the 24 dimers to date. Although these integrins were all crystallized in a highly compact and counterintuitive ‘bent’ conformation, subsequent electron microscopy (EM) studies identified very different and much more intuitive ‘extended’ conformations in addition to the bent conformation, especially when integrins were activated (Nishida et al., 2006; Takagi et al., 2002). As this large conformational change coincided with the change in affinity and/or ligand binding states of integrins, a hypothesis called the ‘switchblade’ (or jack-knife) model was proposed to describe the mechanism for the bidirectional signal transduction (outside-in and inside-out signaling) across a cytoplasmic membrane (Takagi et al., 2002; Takagi and Springer, 2002). The description of the conformational states has been refined by numerous structural studies and it is now generally accepted that integrin can assume three distinct conformations: a bent integrin with a closed headpiece, an extended integrin with a closed headpiece and an extended integrin with an open headpiece, where the former two represent integrins with low ligand affinity (Luo and Springer, 2006).

In β2 and β3 integrins, experimental data from a range of sources including EM, biophysical, immunochemical and computational studies overwhelmingly support the close linkage between affinity state modulation and the structural rearrangements in overall ectodomain and headpiece conformations (Xie et al., 2004; Rocco et al., 2008; Chen et al., 2010, 2011; Nishida et al., 2006; Takagi et al., 2002). Furthermore, in addition to the studies using soluble ectodomain truncations of integrins, detergent-solubilized intact integrins or intact integrins embedded in phospholipid nanodiscs support the structural rearrangement model during activation (Iwasaki et al., 2005; Eng et al., 2011; Xu et al., 2016; Ye et al., 2010). Considering the high degree of sequence conservation among 8 integrin β-subunits and the fact that many β-subunits share identical α-subunits, it was reasonable to expect that the switchblade model applies to all integrin subfamily members. However, recent EM-based studies suggested that this simple assumption may not hold true, at least for some integrins. For example, it was shown that αVβ8 integrin assumed a constitutively extended conformation, regardless of the affinity states toward its physiological ligand, latent TGF-β (Wang et al., 2017; Minagawa et al., 2014). Furthermore, Springer and colleagues reported that the α5β1 integrin ectodomain rarely assumed the acutely bent conformation even in its resting (i.e. low-affinity) condition, and its affinity state was strongly correlated with head-opening but not with global extension (Su et al., 2016). Therefore, it is becoming clear that the link between the global conformational change and the ligand affinity state for a given integrin subtype must be more carefully examined for individual integrins. In particular, elucidation of the regulatory mechanisms for laminin-binding integrins that have evaded structural scrutiny to date is urgently needed, because they comprise ancient and fundamental integrin classes responsible for cell attachment to the basement membrane, which is crucial for all multicellular animals (Hynes and Zhao, 2000; Hutter et al., 2000).

In the current work, we obtained negative-stain EM images of the ectodomain fragment of β1 and β4 integrins, including fibronectin-binding (α5β1), laminin-binding (α3β1, α6β1 and α6β4) and collagen-binding (α2β1) integrins, in both resting (i.e. in 5 mM Ca²⁺) and activating (i.e. in 1 mM Mn²⁺) conditions, and made side-by-side comparisons with images of the well-studied αVβ3 integrin. In contrast to the αVβ3 integrin, which primarily assumes a bent conformation in the resting condition, all β1 and β4 integrins exhibited a range of conformations with the majority of the particles showing overall extended conformation. Surprisingly, they do not undergo either local (i.e. head opening) or global (i.e. extension) conformational change upon the shift to a more activating condition, suggesting that the affinity upregulation of these integrins can occur in the absence of obvious conformational change that can be detected by low-resolution EM analysis.

Soluble extracellular fragments of integrins each containing a releasable C-terminal clasp (Fig. 1A) were purified as described in the Materials and Methods. The α- and β-subunits were linked by a disulfide-bonded clasp and migrated as a single band in the SDS-PAGE under nonreducing conditions (TEV− lanes in Fig. 1B). By contrast, they were separated into two bands after TEV protease treatment (TEV+ lanes in Fig. 1B). Thus, it was confirmed that all purified integrin heterodimers were linked via a disulfide-bonded clasp at the C-terminal, which can be released (or ‘unclasped’) by TEV protease treatment. These integrins were subjected to the following experiments (ligand binding assay and structural analysis by electron microscopy). As reported previously (Takagi et al., 2002), binding of αVβ3 integrin to its ligand was negligible in the presence of 5 mM Ca²⁺ but was upregulated 4- to 5-fold over the background (i.e. BSA control) level in the presence of 1 mM Mn²⁺ in both clasped and unclasped conditions (Fig. 1C, far left panel), indicating that the 5 mM Ca²⁺ and the 1 mM Mn²⁺ conditions correspond to the low- and high-affinity states, respectively. We deliberately employed a non-physiologically high concentration of Ca²⁺ (5 mM) to push the equilibrium toward the low-affinity state by saturating the βI domain ADMIDAS with Ca²⁺, which has been reported to have negative regulatory function (Mould et al., 2003). The ADMIDAS-bound Ca²⁺ is believed to stabilize the low-affinity conformation of the ligand-engaging MIDAS metal through its preference toward pentagonal bipyramidal over octahedral coordination geometry (Xia and Springer, 2014). When another RGD-dependent integrin α5β1 was subjected to the same assay using immobilized fibronectin, the results were essentially the same, although the overall binding signal was much higher (Fig. 1C, second panel). We then examined the ligand binding ability of all other non-RGD integrins toward laminin-10 (for integrins α3β1, α6β1 and α6β4) or type I collagen (for α2β1) using the same assay conditions. As shown in Fig. 1C, they all showed similar ligand binding to αVβ3 integrin; binding was negligible in 5 mM Ca²⁺ but high ligand binding activities were observed in the presence of Mn²⁺. Importantly, the extent of ligand binding under Mn²⁺ conditions for each integrin was almost the same irrespective of the state of the C-terminal clasp (clasped or unclasped), except for α6β4 integrin. In the case of the α6β4 integrin, unclasping resulted in diminished binding to laminin-10, although there was still clear binding above background levels and above the level seen in Ca²⁺ conditions. From these results, we conclude that 5 mM Ca²⁺ and 1 mM Mn²⁺ can be used as representative conditions to induce low- and high-affinity states, respectively, for all integrin ectodomain fragments used here.

Having established the experimental conditions to maintain various integrins in the low- and high-affinity states, we turned to negative-stain EM imaging to visualize the conformation of each integrin under both conditions. To this end, integrin samples were loaded onto a gel filtration column equilibrated with a buffer containing 5 mM Ca²⁺ (low-affinity state) or 1 mM Mn²⁺ (high-affinity state) and the monodisperse peak fraction containing the heterodimeric integrin ectodomain was used to make EM grids. We also included one more condition, where 1 mM Mg²⁺ was added to 5 mM Ca²⁺ buffer. Data were collected for all six integrins under four different conditions (clasped/5 mM Ca²⁺, clasped/5 mM Ca²⁺+1 mM Mg²⁺, clasped/1 mM Mn²⁺ and unclasped/1 mM Mn²⁺). As can be seen in the representative raw EM image (Fig. S1), all samples showed well-dispersed individual particles that allowed efficient image analysis. From the EM images obtained, ∼1000 particles in each condition were boxed out, and they were averaged after classification into 20 classes. We used αVβ3 integrin as a reference, because this is the most extensively studied integrin by multiple groups using EM analysis. As expected, nearly all αVβ3 particles exhibited a highly bent conformation in 5 mM Ca²⁺ regardless of the additional presence of Mg²⁺. In the Mn²⁺ condition, a large majority changed their shape and showed a completely extended and open conformation consistent with the ‘switchblade model’ (Fig. S2A). The behavior of α5β1 integrin was quite different from that of αVβ3, however, because it rarely assumed the acutely bent conformation in the low-affinity condition but rather exhibited varying shapes, with partly to fully extended conformations being predominant (Fig. 2A and Fig. S2B). In fact, this result is essentially the same as that reported by Springer and colleagues (Su et al., 2016). The authors showed that only a fraction of EM class averages of ligand-unbound and clasped α5β1 ectodomain fragment appeared as the acutely bent form in a buffer containing 1 mM Ca²⁺ and 1 mM Mg²⁺. More importantly, the shape distribution of α5β1 particles did not change when the high-affinity condition (i.e. 1 mM Mn²⁺) was employed (Fig. 2B and Fig. S2B), suggesting the lack of a strong correlation between the affinity state and the global conformation in α5β1. Despite high variability of the particle shapes of α5β1 in the EM images, most 2D class averages showed clear and distinctive features that helped us to assign each chain or domain present in the construct (Fig. 1A) into the densities, enabling us to interpret the 3D structure (Fig. 2C) and to classify particles according to their global conformation (see below).

We next performed similar EM imaging on laminin- and collagen-binding integrins that had escaped structural scrutiny until now, namely α3β1, α6β1, α6β4 and α2β1. In general, the results were almost the same as that of α5β1 integrin (Fig. 3 and Fig. S2C-F). No severely bent conformation was observed for any of these integrins, even in the low-affinity state, and the majority of the class averages exhibited an overall extended conformation. Furthermore, the conformational distribution seen in the Mn²⁺-activated condition for the three laminin-binding integrins was indistinguishable from that in the Ca²⁺ or Ca²⁺/Mg²⁺ condition, regardless of the presence of the C-terminal clasp (Fig. 3 and Fig. S2C-E). In particular, α6β4 integrin almost always assumed the completely extended structure regardless of the condition (Fig. S2E), suggesting that it is mostly refractory to conformational regulation. In the case of α2β1 integrin, there is an additional I (A) domain in the α-subunit, which was clearly visible at the top of the molecule in the class-averaged images (Fig. S2F). Although it shared the same trend with laminin-binding integrins of having the overall extended conformation in both low- and high-affinity conditions, there was a clear local conformational change upon addition of Mn²⁺, corresponding to head opening (i.e. the swing-out of the hybrid domain of the β-subunit). As the αI domain functions as an internal ligand to the β-subunit (Sen et al., 2013), Mn²⁺-treated α2β1 would represent the ‘active and liganded’ state rather than the ‘active and non-liganded’ state, as in the case of all other integrins lacking domain I. Therefore, these results are highly consistent with the notion that the β-hybrid swing is coupled to ligand binding rather than the affinity state of integrin (Su et al., 2016).

The individual class average was categorized into five groups based on their shape (Fig. S2), and the prevalence of each group was calculated from the numerical values obtained during single-particle analysis. As none of the integrins showed a significant difference in the conformational distribution between Ca²⁺ and Ca²⁺/Mg²⁺ conditions, as well as between clasped/Mn²⁺ and unclasped/Mn²⁺ samples, we will focus on the comparison between the clasped/Ca²⁺ and clasped/Mn²⁺ conditions. The population prevalence determined as above can be regarded as the approximate ‘conformational distribution spectra’ for each integrin or condition (Fig. 3). It is evident from this analysis that all examined β1 and β4 integrins prefer the extended conformation (colored in yellow, orange or magenta) over bent conformation (blue and green) under the Ca²⁺-induced low-affinity condition, and rarely assumed the severely bent conformation (blue) that is strongly favored by the well-studied β2 and β3 integrins. Furthermore, there seems to be no or very limited coupling between the affinity state and the global conformation of these integrins, because nearly identical conformational distributions were obtained for the Ca²⁺ and Mn²⁺ conditions. We are thus inclined to think that the switchblade model is not applicable to most if not all β1 and β4 integrins, at least as an affinity regulation mechanism. However, data must be interpreted in a careful manner to avoid overgeneralization, as the current study used only the soluble integrin ectodomain fragments. In addition, as the direct structural analysis is only possible with isolated integrin preparations, it is difficult to know the true conformational spectrum on the cell surface. For example, Springer and colleagues estimate that more than 98% of the α5β1 molecule on K562 cells are in the bent form (Li et al., 2017). It is possible that certain mechanisms exist on the cell surface to stabilize a bent conformation, at least for α5β1. Nevertheless, our data indicate that the ectodomain portions of β1 and β4 integrins lack the intrinsic property of rearrangement upon affinity manipulation by divalent cations.

Unlike the integrins present on circulating blood cells, including αIIbβ3, αVβ3, αXβ2 and αLβ2, which have to bind ligands during transient encounters, laminin-binding integrins (α3β1, α6β1, α7β1 and α6β4) and collagen-binding integrins (α1β1, α2β1, α10β1 and α11β1) on stationary cells generally have ample time to establish firm adhesion due to continuous contact with the extracellular components, and may not need mechanisms of rapid affinity upregulation. Thus the rapid ‘switchblade’ activation, in a strict sense, may only be applicable to those integrins on blood cells that emerged at a relatively late evolutionary point after the acquisition of the vascular system. Although there is strong coupling between ligand binding and global and local conformation of a wide variety of integrins (Takagi et al., 2002; Nishida et al., 2006; Chen et al., 2010, 2011; Rocco et al., 2008; Xie et al., 2004; Su et al., 2016), we do not have direct evidence that such coupling is applicable to laminin-binding integrins. This is mainly due to the limited structural information available for the laminin-integrin interaction (Takizawa et al., 2017). We predict that further structural analysis as well as development of biochemical tools, such as high-affinity laminin mimetic ligands to probe the interaction between laminin and integrin, will be essential to increase our understanding about this fundamental and ancient cell-matrix interaction event.

Soluble integrin heterodimers were constructed using a strategy described previously (Takagi et al., 2001). Briefly, the expression constructs for the α-subunits contained an extracellular portion of each α-chain (residues 1-1103 for α2, residues 1-957 for α3, residues 1-950 for α5, residues 1-988 for α6, and residues 1-960 for αV) followed by a 30-residue ACID-Cys peptide. Constructs for β-subunits contained an extracellular portion of each β-chain (residues 1-708 for β1, residues 1-691 for β3 and residues 1-683 for β4) followed by a TEV protease recognition sequence, a 30-residue BASE-Cys peptide and a hexahistidine tag. When combined, the C-terminal ACID-Cys and BASE-Cys segments form the inter-subunit disulfide-bridged α-helical coiled-coil (‘clasp’) that can be released by a treatment with TEV protease (Takagi et al., 2002). The general architecture of recombinant soluble integrin heterodimers is shown in Fig. 1A. Appropriate combinations of α- and β-constructs were co-transfected into either CHO lec 3.2.8.1 cells (for α3β1, α5β1, α6β1, α6β4 and αVβ3) (a gift from Pamela Stanley, Albert Einstein College of Medicine, New York, USA) or HEK293-EBNA cells (for α2β1) (Thermo Fisher Scientific) to establish stable cell lines. Recombinant integrins were purified from the culture supernatants by an immunoaffinity chromatography using anti-coiled-coil antibody 2H11 (a gift from Ellis Reinherz, Dana Farber Cancer Institute, Boston, USA; Chang et al., 1994), followed by a gel filtration on a Superdex 200 HR column (1.6×60 cm, GE Healthcare) equilibrated with 20 mM Tris-HCl, 150 mM NaCl, pH 7.5 (TBS) containing 1 mM CaCl₂, 1 mM MgCl₂. The peak fraction was concentrated to ∼1 mg/ml and stored at −80°C until used. To obtain unclasped integrins, purified clasped integrins were treated with recombinant TEV protease at 20°C for 16 h.

Ligand binding assays were performed as described previously (Nishiuchi et al., 2006). Briefly, solutions of recombinant laminin-10 (20 µg/ml, a gift from Kiyotoshi Sekiguchi, Institute for Protein Research, Osaka Japan), bovine fibronectin (10 µg/ml, Sigma), or rat type I collagen (10 µg/ml, Sigma) in TBS were used to coat 96-well polyvinylchloride microtiter plates (Nunc, Maxisorp) by an overnight incubation at 4°C. Coating with bovine serum albumin (BSA) was used to determine the background values of unspecific binding. After a 1 h blocking step (1% BSA in TBS), various integrins were added to the plates at 25 µg/ml (α6β4) or 1 µg/ml (all other integrins) and allowed to bind the absorbed ligand for 4 h at room temperature. The reaction mixture contained either 5 mM CaCl₂ (low-affinity condition) or 1 mM MnCl₂ (high-affinity condition). The plates were washed with TBS containing 1 mM MnCl₂ and the bound integrins were quantified by an enzyme-linked immunosorbent assay using biotinylated rabbit anti-clasp (ACID/BASE coiled-coil) antibody (made in-house; Nishiuchi et al., 2006) and HRP-conjugated streptavidin (SA-5004, VECTOR Laboratories).

Approximately 10 μg of each purified integrin was subjected to an additional gel filtration on a Superdex 200 HR column equibrated with 50 mM Tris-HCl, 150 mM NaCl, pH 7.5 containing 5 mM CaCl₂ or 1 mM MnCl₂. The samples after the gel filtration were immediately absorbed to glow-discharged carbon-coated copper grids. Samples were negatively stained with 2.5% (w/v) uranyl acetate and examined under an electron microscope (H9500SD; Hitachi, Japan) operated at 200 kV and a nominal magnification of ×80,000. Tobacco mosaic virus was added to specimens to control the depth of staining, which was also used for calibrating magnification (large rod-like objects in Fig. S1). Images were recorded on a 2048×2048 CCD camera (TVIPS, Gauting, Germany). Single-particle analysis, including particle selection and 2D classification and averaging, was performed using the EMAN software suite (Ludtke et al., 1999) and IMAGIC (van Heel et al., 1996). Particles were selected from individual frames (with an effective pixel size of 0.21 nm) using the program Boxer in the EMAN software suite. The particle images were rotationally and translationally aligned by a multi-reference alignment procedure and subjected to multivariate statistical analysis specifying 20 classes using the IMAGIC program.

We thank Zuben P. Brown and Kevin Stapleton for critically reading the manuscript. Part of this work was carried out at the Research Center for Ultra-high Voltage Electron Microscopy, Osaka University.