A distinct talin2 structure directs isoform specificity in cell adhesion

Integrin receptors regulate normal cellular processes such as signaling, cell migration, adhesion to the extracellular matrix, and leukocyte function. Talin recruitment to the membrane is necessary for its binding to and activation of integrin. Vertebrates have two highly conserved talin homologs that differ in their expression patterns. The F1 – F3 FERM subdomains of cytoskeletal proteins resemble a cloverleaf, but in talin1, its F1 subdomain and additional F0 subdomain align more linearly with its F2 and F3 subdomains. Here, we present the talin2 crystal structure, revealing that its F0-F1 di-subdomain displays another unprecedented constellation, whereby the F0-F1-F2 adopts a new cloverleaf-like arrangement. Using multiangle light scattering (MALS), fluorescence lifetime imaging (FLIM), and FRET analyses, we found that substituting the corresponding residues in talin2 that abolish lipid binding in talin1 disrupt the binding of talin to the membrane, focal adhesion formation, and cell spreading. Our results provide the molecular details of the functions of specific talin isoforms in cell adhesion.

Integrin receptors regulate normal cellular processes such as signaling, cell migration, adhesion to the extracellular matrix, and leukocyte function. Talin recruitment to the membrane is necessary for its binding to and activation of integrin. Vertebrates have two highly conserved talin homologs that differ in their expression patterns. The F1-F3 FERM subdomains of cytoskeletal proteins resemble a cloverleaf, but in talin1, its F1 subdomain and additional F0 subdomain align more linearly with its F2 and F3 subdomains. Here, we present the talin2 crystal structure, revealing that its F0-F1 di-subdomain displays another unprecedented constellation, whereby the F0-F1-F2 adopts a new cloverleaf-like arrangement. Using multiangle light scattering (MALS), fluorescence lifetime imaging (FLIM), and FRET analyses, we found that substituting the corresponding residues in talin2 that abolish lipid binding in talin1 disrupt the binding of talin to the membrane, focal adhesion formation, and cell spreading. Our results provide the molecular details of the functions of specific talin isoforms in cell adhesion.
Vertebrates have two highly conserved isoforms with ;76% sequence identity, but they are not functionally redundant. We know a lot about the talin1 structure and function, whereas much less has been published on talin2 both in terms of structure and function. Talin1 is ubiquitously expressed, whereas talin2 is primarily found in the heart, brain, and skeletal muscle (38,39). Talin1 is also expressed at smaller focal adhesions in the peripheral region, whereas talin2 is mainly found at large focal adhesions and fibrillar adhesions (40,41). Upon phosphorylation by Cdk5, talin1 controls focal adhesion dynamics, integrin activation, cell migration, invasion, and metastasis (2,3), whereas talin2 controls focal adhesion assembly and focal adhesion kinase signaling in cells that are depleted of talin1 (14).
Talin1 knockout mice are embryonic lethal by E8.5-E9.0 due to gastrulation defects (42). Talin2 knockout mice display mild skeletal myopathy at 3 months of age resulting from defects in the myotendinous junction (43). Talin2 can fully rescue the profound defects in focal adhesions and the organization of the cytoskeleton that are manifest in talin1-null cells (44). Thus, talin1 and talin2 seemed at first glance to be functionally redundant. However, the two isoforms transduce mechanical force differently. Under tension, talin1 recruits vinculin, but with talin2, vinculin recruitment occurs even in the absence of its F-actin-binding domain (45). Further, cardiac myocyte talin2 is necessary for proper integrin b1D expression, whereas talin1 can preserve heart function in the absence of talin2. Loss of both talin isoforms from the heart muscle results in myocyte instability and dilated cardiomyopathy (46).
Cell migration and invasion is inhibited by an antibody therapeutic that down-regulates talin2 by targeting human epidermal growth factor 2 for cancer treatment (47). However, depletion of talin2 has no effects on integrin b1 activation (3). A recent talin2-null study on T-cell exosomes showed reduced binding of integrins aLb2 and a4b7 to the intercellular adhesion molecule-1 (ICAM-1) and to the mucosal vascular addressin cell adhesion molecule-1 (MAdCAM-1) (48). Collectively, the talin isoforms do not compensate for the loss of the other, and their functions are distinct.
To determine the molecular basis for the difference between the talin isoform functions, we solved the crystal structure of the talin2 head domain (residues 1-403) to 2.56 Å resolution and provide structural and functional insights into the talin isoforms. Despite their ;76% sequence identity and ;88% sequence similarity for mammalian isoforms, in talin2, whereas the F0-F1 and F2-F3 domains resemble those seen in talin1, F1 engages in distinct interdomain interactions with F2 to result in a completely distinct head domain constellation that has not been observed for any other FERM domain-containing protein. Our solution studies obtained from size-exclusion chromatography support these findings.
Despite their distinct head domain constellation, the mammalian talin isoforms share the same membrane-binding sites. Both full-length proteins are monomeric as determined by multiangle light scattering. This previously disputed observation is now also confirmed by the cryo-EM structure (49) that was published while this manuscript was in revision.
Our confocal microscopy and in vivo fluorescence lifetime imaging microscopy (FLIM) with FRET analyses revealed the functional relevance of the talin interaction with the cell membrane. Collectively, we provide significant new insights into talin structure and function that pave the way toward a better understanding of the regulation of integrin function and of fundamental processes in cell biology including the cytoskeletal organization.

Talin2 has an unprecedented FERM constellation
To provide structural insights into the functional differences between the two talin isoforms, we determined the crystal structure of human talin2 (residues 1-403) to 2.56 Å resolution (Tables 1 and 2). Whereas the individual talin2 di-subdomains F0-F1 and F2-F3 are almost identical as seen in the talin1 structure, the F0-F1 motif is rotated by ;140°in talin2 relative to its orientation in talin1 ( Fig. 1A and Fig. S1). This is accomplished by isoform-specific F1-F2 interdomain interactions.
We assessed the contribution of crystal contacts in the distinct constellations of the F0-F1 di-domain relative to the F2-F3 di-domains in the two talin isoforms (Figs. S2 and S3 and  Tables 3 and 4). Murine talin1 crystallized in P 22 1 2 1 as a chimera of residues 1-138 fused to 169-400, whereas our human talin2 (residues 1-403) structure crystallized in space group P 6 5 . The volume/mass ratios are comparable, with 2.27 and 2.18 Å 3 /Da corresponding to a solvent content of 0.49 and 0.436 for talin1 and talin2, respectively. The crystal contacts in the talin1 structure ( Fig. S2 and Tables 3 and 4) were already analyzed by the University of Leicester team (50) and Campbell (51) and were not thought to contribute to the talin1 conformation. In our talin2 crystal, the screw axis packs the talin2 F1 subdomain against F0' from a symmetry-related molecule (Fig. S3A) and allows for F1-F1' (Fig. S3C) or F3-F0' (Fig. S3D) intramolecular interactions, whereas the crystal is held together along the y axis through F0-F2' intramolecular interactions (Fig. S3B).
Regions that engage in talin2 intersubdomain-specific interactions involve residues 198-211 that connect F1 and F2 and those that are in proximity of F2 (residues 120-129), whereas the last F2 a-helix (residues 294-309) is also engaging in intradomain F1-F2 contacts (Fig. 1B). These unique interactions provide the novel talin2 cloverleaf-like constellation that is distinct not only from the talin1 linear arrangement but also from all other known FERM domain protein structures. For example, whereas kindlin (PDB entry 5XQ1; Fig. S4A) also has an additional F0 subdomain (52), the F2-F3 di-domains are oriented distinctly in talin when aligning their respective F0 and F1 subdomains (Fig. S4, A-C). For comparison, the constellation of a canonical FERM architecture as, for example, seen in merlin (PDB entry 6CDS) (53) has a kindlin-like architecture (Fig.  S4D), which again is distinct from that seen in talin.
Interestingly, the pivot point that drives the conversion of the talin1 linear arrangement versus the semi-cloverleaf  Talin2 has a novel FERM configuration type arrangement in talin2 stems from the positioning of crucial talin2 residues (e.g. . In talin2, these residues assume a cis conformation leading to stabilized hydrophobic stacking interactions, formed by Ile-132, Trp-175, and Tyr-201, on either side of the already stacked Lys-198 and Phe-199 residues (Fig. 1B). However, in talin1, the corresponding Lys-196 and Phe-197 (the numbering differs) are arranged in a trans configuration leading to the linear arrangement of the loop preceding the terminal b-strand of the F1 domain (Fig. 1C), an interaction that is probably responsible in maintaining the distinct F0-F1 orientation with respect to F2-F3. Further isoform-specific interactions might result from the long F1 loop (residues 139-170) that is missing in the talin1 structure due to deletion to aid crystallization (50). The F1 loop (residues 144-167) was thought to form an a-helix (54), and this loop is visible for the first time in talin2 (with electron density only missing for residues 143-153) where residues 156-166 form an a-helix (Figs. 1A and 2).
Superposition of the F0-F1 domains of the two isoforms reveals that the talin2 F1 loop resides in the space occupied by F2 in the talin1 structure (Fig. S1C). The same steric hindrance is seen when superimposing the talin1 F1 solution structure (PDB entry 2KC2) (54) onto the talin1 head structures that have the 139-168 deletion (PDB entries 6MFS and 3IVF) (50, 55) (Fig. S1D). Thus, the unknown native full-length talin1 head domain structure must either have its F1 loop distinct from the F1 solution structure, or the talin1 F1-F2 interdomain interactions are distinct from the linear arrangement found in the D139-168 talin1 head domain structure.

The inactive mammalian talin isoforms are monomeric
Turkey gizzard talin1 was reported to be a dimer (20). To determine the oligomeric state of mammalian talin, we analyzed full-length talin by multiangle light scattering coupled with size-exclusion chromatography, which also confirmed that our proteins are not misfolded aggregates. We found that both full-length murine talin1 and human talin2 are monomeric in solution ( Fig. 3A) even though full-length talin2 showed some aggregates. Full-length talin1 exhibited a molar mass of 245.2 g/mol 6 0.6%, which is close to the monomeric estimated mass of 270 kDa with a hydrodynamic radius of 6.8 nm 6 0.2%. Talin2 exhibited a molar mass of 280.4 g/mol 6 1.9% g/mol, which corresponds to the calculated mass of 274 kDa with a hydrodynamic radius of 8.7 nm 6 0.6%, suggesting an overall extended conformation. Likewise, the individual talin A, superposition of the F0-F1 di-domains of murine talin1 onto human talin2 (F0, residues 4-83, orange; F1, residues 87-199, yellow). The talin1 F2 and F3 domains are shown as a ribbon, and the talin2 F2-F3 (F2, residues 211-308, green; F3, residues 314-402, blue) are shown as a cartoon. Murine talin1 has its F0-F1 di-subdomain unit aligned linearly with respect to its F2-F3 di-subdomain motif. The F1 loop that is truncated in the talin1 structures (PDB entries 6MFS and 3IVF) (51, 56) is indicated as residue 169. Notably, the talin2 F0-F1 di-subdomains are making talin2-specific interactions with F2-F3. Most of the F1 loop is visible in our talin2 structure (shown in red) and harbors an a-helix. Whereas the F2-F3 di-subdomains are rotated ;140°relative to their position seen in talin1 (double arrow), the F0-F1 di-subdomains align with root mean square deviation of 0.841 Å for 1,256 atoms. B, talin2-specific F1 (yellow) interdomain interactions with F2 (green). Thr-124 and Glu-128 along with Gly-122 interact extensively with the residues from the F1- Talin2 has a novel FERM configuration head domains were predominantly monomeric (Fig. 3B) and exhibited molar masses of 42.4 g/mol 6 0.6% and 44.9 g/mol 6 0.7% for the talin1 (residues 1-400) and talin2 (residues 1-403), respectively. Interestingly, despite the differences in the talin1 and talin2 respective subdomain arrangements, the hydrodynamic radii were similar, 3.3 nm 6 0.3% and 3.2 nm 6 0.4%, respectively. While our manuscript was under review, the full-length human talin1 structure was also shown to be monomeric (49).

The inactive mammalian talin isoforms bind lipid vesicles
To confirm that our recombinant talin proteins are functional, we tested their binding to the membrane. The talin rod domain was thought to hinder binding of full-length inactive talin to the membrane (56). Superposition of the F3 subdomain of our talin2 structure onto the F3 subdomain of the talin1 head/tail complex structure (PDB entry 4F7G) (57) (Fig. 4A) shows that talin1 Gly-376 is replaced by talin2 Ser-379 and too close to Asp-1676 from the talin1 rod domain. Talin1 Asp-1676 corresponds to talin2 Glu-1678, and the larger side chains on both head and tail domains (Ser/Glu versus Gly/Asp) make this talin1 interaction thus unlikely in talin2. Further, talin1 Met-1802 corresponds to talin2 Leu-1804, which again is too close to conserved Gly-321 (corresponding to Gly-324 in talin2). It remains to be seen whether the full-length talin2 structure has the same head-tail interface as seen in talin1.
Because most membrane-binding experiments were performed with individual talin domains (e.g. the F3 FERM subdomain and residues 1,654-2,344, termed talin-R M for talin rod middle segment, from the talin1 rod domain) (50,56,58,59), we performed lipid co-sedimentation assays with full-length proteins. In agreement with the earlier experiments from individual domains (58), our full-length talin1 did not bind phosphatidylcholine (PC). However, we found that full-length talin1 bound PIP 2 (Fig. 4B). Because the F3 interaction with talin-R M is disrupted by PIP 2 (56), our data suggest that PIP 2 binding to talin is the first step in integrin inside-out activation. Further, the N-terminally truncated talin1 constructs with (residues 437-2,541) and without (residues 437-2,493) the "dimerization domain" did not bind PIP 2 or PC (Fig. 4C). Thus, our recombinant proteins are functional with respect to membrane localization, and full-length inactive talin binds the plasma membrane. The PIP 2 -binding site resides on the talin head domains.
The talin R9 domain (residues 1,655-1,822) is part of the above mentioned talin-R M domain (residues 1,654-2,344) and in talin1 was initially implicated in autoinhibitory regulation through its interaction with the F3 subdomain (residues 311-398) (60). The recently determined talin1 cryo-EM structure showed that in addition to the head-tail F3-R9 interaction, another head-tail interaction between F2 (residues 209-304) and R12 (residues 2,131-2,291) further prevents integrin and phospholipid binding to talin and confirmed that the lipidand tail-binding sites are overlapping on the talin head domain and the integrin-binding site is obstructed by the tail domain (49). Collectively, our data provide mechanistic insights into how talin binding to the cell membrane relieves the talin autoinhibition.

The talin isoforms share the same lipid-binding site
We identified the bona fide talin1 lipid-binding site (that is strictly conserved in the two mammalian talin isoforms, residues K272Q, K316Q, K324Q, E342Q, and K343Q) structurally, biochemically, and functionally (Fig. 5). The talin1 PIP 2 -bound and -unbound structures are very similar (Fig. 5B) (55). However, when superimposing the F2 subdomains of our PIP 2bound talin1 and apo-talin2 structures, residue Lys-324 of the talin1 lipid-binding loop is moved by about 7 Å in the unbound talin2 structure (Fig. 5C). Despite significant attempts, we were unable to capture the talin2/PIP 2 complex crystal structure. Thus, we mutated the equivalent talin2 residues that were identified as bona fide lipid binders in talin1 and cross-verified the structural integrity through thermal shift analyses, which exhibited a melting temperature (T m ) comparable with the WT, representing a well-folded protein (Fig. 5D). We performed lipid co-sedimentation assays and found that the talin2 equivalent mutant (K274Q, K319Q, K327Q, E345Q, and K346Q) did not bind to vesicles at all (Fig. 5E). Thus, the Table 3 Analyses of the crystal contacts of the two talin isoforms by the PDBePISA server (RRID:SCR_015749) The last column in the table refers to the view of said crystal contact in the indicated figure A. The murine talin1 head domain structure, PDB entry 3IVF, residues 1-400 D139-168, has a surface area of 20,705 Å 2 xyz i N at xyz i N res Symmetry operator Sym i N at Sym i N res Interface area (Å 2 ) D i G (kcal/mol) D i G p value N hb N sb N ds Figure   72 23 x, Talin2 has a novel FERM configuration talin2 head domain has a single membrane attachment site that is conserved in talin1.
To measure the association of the talin isoforms with the cell membrane in living cells, we developed a FRET-based sensor between the monomeric enhanced GFP (mEGFP)tagged talin isoforms and a plasma membrane-targeted mCherry (mCherry-CAAX). WT and lipid binding-deficient mutant talin proteins were expressed in papillary collecting duct cells in a talin 2/2 knockout background, together with the membrane-targeted mCherry fluorophore (Fig. 6A). When talin-GFP-and the CAAX-mediated membrane-localized mCherry are in proximity (,10 nm), FRET between them occurs and was measured through FLIM-FRET (61). Calculation of the fraction of overexpressed talin bound to the membrane (binding fraction) was quantified based on lifetime measurements and compared with (i) control cells expressing only GFP, (ii) GFP-and membrane-targeted mCherry with and without the unlabeled talin isoforms, and (iii) GFP-tagged talin isoforms without mCherry (Fig. S5A). Importantly, expressing either unlabeled or GFPtagged talin proteins in the talin 2/2 knockout background induced a similar robust spreading morphology as well as formation of focal adhesion and stress fiber formation.  Talin2 has a novel FERM configuration Thus, the GFP-tagged talin isoforms were functional and localized as seen for WT talin. Immunoblotting showed robust expression of WT and mutant proteins, suggesting that the impaired cell spreading in cells expressing the lipid binding-deficient mutants is indeed due to defective lipid binding (Fig. 6B).
Both talin isoforms showed significant binding to the cell membrane (p , 0.0001; Sidak's multiple-comparison test after one-way ANOVA), and mutations in talin1 and talin2 that disrupt PIP 2 binding in vitro significantly impaired this association in live cells. In addition, our data suggested that talin2 binds the membrane more robustly than talin1 (Fig. 6C), which was supported by PIP 2 vesicle pulldowns that showed a higher maximum binding for talin2 compared with talin1 (55% versus 37%, respectively) (Fig. 6D). Their binding constants were similar (3.4 and 4.6 mM for talin2 and talin1, respectively) and also comparable with binding affinities determined for a different FERM domain protein (focal adhesion kinase; 6 mM) (62). These findings were significant irrespective of relative expression levels of the talin constructs or the relative ratio of GFP-talin relative to membrane-localized mCherry (Fig. S5).
Confocal microscopy studies support the observation that overexpression of WT talin in papillary collecting duct cells on   (for 544 atoms). Notably, talin1 Gly-376 is replaced by talin2 Ser-379 and too close to Ala-1676 from the talin1 rod domain. Talin1 Ala-1676 corresponds to talin2 Glu-1678, and the larger side chains on both head and tail domains (Ser/Glu versus Gly/Asp) make this talin1 interaction thus unlikely in talin2. Talin1 Met-1802 corresponds to talin2 Leu-1804, which again is too close to conserved Gly-321 (corresponding to Gly-324 in talin2). B, lipid co-sedimentation assays of fulllength murine (Mm) talin1 (residues 1-2,541) and human (Hs) talin2 (residues 1-2,542). S, supernatant; P, pellet. C, lipid co-sedimentation assays of the talin1 rod domain with (residues 437-2,541) and without (residues 437-2,493) the "dimerization domain." S, supernatant; P, pellet. Compared with full-length talin proteins (B), there is some aggregation seen with the N-terminally truncated talin proteins in the absence of lipids. a talin 2/2 knockout background showed robust spreading, focal adhesions, and actin stress fiber formations (Fig. 7). We used band pass filtering and threshold application to calculate the total cell areas and the focal adhesion areas (Fig. 8). The morphology of cells was clearly reflected in the total cell area analyses (Fig. 8A). Cells expressing the lipid binding-deficient talin mutants had a mean total area of ;250 mm 2 , whereas those expressing WT talin showed a much larger cell spread (;500 mm 2 for talin1 and ;350 mm 2 for talin2) (Fig. 8B). The effect of expressing the lipid binding-deficient mutant on the cell spread was also demonstrated by Sidak's multiple-comparison tests, within ordinary one-way ANOVA, with an adjusted p value of ,0.0001. Whereas the focal adhesion sizes in the cells expressing WT or mutant proteins were similar (Fig. 8C), the focal adhesion area was on average about 0.17 mm 2 for talin1 and about 0.13 mm 2 for talin2. The value observed for the WT is comparable with the size reported previously for such focal adhesions in papillary collecting duct talin knockout cells (33). Sidak's comparison test did not reveal any significant difference between cells expressing WT and mutant proteins. However, the focal adhesion size appeared to be larger for talin1 compared with talin2 (Fig. 8C). Collectively, binding of talin2 to the cell membrane is crucial for talin2 in regulating various cell functions, including migration, adhesion, and maintenance of cellular integrity.

Isoform-specific talin head domain structures and functions
Talin is the key cytoplasmic protein that mediates integrin adhesion to the extracellular matrix (63) and acts as a bridge between the cytoskeleton and the extracellular matrix. Talin is of great interest to the field of mechano-biology as it unfolds and exposes cryptic vinculin-binding sites under force stimulation. Significantly, when talin activators are dysregulated, abnormal integrin activation and mechano-transduction lead to changes in cell migration and spreading and cell survival that result in various diseases (64).
Talin has two main functional units, the N-terminal head (residues 1-400) and the C-terminal tail (residues 437-2,541), that are connected by a calpain-susceptible linker region. The two most important functions of the head domain are its binding to and activation of integrin and its binding to the plasma membrane (4,5). These interactions are regulated by the head domain binding to the autoinhibition site on the talin tail domain. Our structural and biochemical data support a mechanism whereby the plasma membrane severs the talin autoinhibitory head-tail interactions.
Vertebrae have two talin isoforms with ;76% sequence identity (;88% sequence similarity) throughout the polypeptide chain. Talin1 was discovered 35 years ago (65) in chicken gizzard smooth muscle and was found to target to focal adhesions and membrane ruffles. Talin2 was discovered more recently via the sequencing of the human genome. Whereas talin1 knockout is embryonic lethal (42), talin2 knockout mice are viable and fertile, although they display a mild dystrophic phenotype (66).
Talin2 has a novel FERM configuration that are stabilized in a globular cloverleaf arrangement as its F1 domain aligns linearly with the F2-F3 di-domains.
We determined the crystal structure of the entire talin2 head domain without an engineered truncation that deletes residues 139-168 in the structure of the talin1 head domain (50,55). The full-length talin1 F1 subdomain structure (54) suggested that these residues could be deleted without affecting the structure of F1. Superposition of our full-length talin2 head domain structure onto the truncated talin1 head domain structure (50) shows that in talin2, residues 139-168 are occupying the corresponding space where F2 interacts with F1 in talin1. This steric hindrance is also evident when superimposing the talin1 F1 solution structure (PDB entry 2KC2) (54) onto the truncated talin1 head domain structures (PDB entries 6MFS and 3IVF) (50,55) where the talin1 F1 loop prevents the F1-F2 interaction seen in the truncated talin1 head domain structure. Although residues 171-174 at the C terminus of the F1 loop in talin1 fold back toward F0-F1 but away from F0-F1 in talin2, these loops protrude similarly from the ubiquitin-like F1 domain into the space currently occupied by F2 in the talin1 structure. It remains to be seen how the F1 loop is positioned in the talin1 head domain.
Because the double subdomain units of F0-F1 or F2-F3 are similar in both talin isoforms, and the latter is as seen in all other FERM domain structures, the subdomain interactions between the F1 and F2 subdomains are unique and isoformspecific. Indeed, no other FERM domain has been determined that had the arrangement seen in our talin2 structure. Our talin2 structure shows that the F2-F3 di-domains and the F0-F1 di-domains seem to be the structural building blocks of FERM domains that can arrange in distinct ways by varying the F0-F1 interaction with F2-F3 while the interfaces between F0 and F1 or F2 and F3 are conserved. It is intriguing how proteins use the same F0-F1 and F2-F3 di-subdomain building blocks and arrange these differently in the tertiary structure.
The talin1 head domain showed an angle of ;125°between the F0-F1 and F2-F3 di-domains as determined by SAXS (50) compared with 155°angle observed in the linear arrangement of the crystal structures (PDB entries 6MFS and 3IVF) (50,55). Such flexibility between F1 and F2 was also noted by SAXS data from the human talin1 head domain (50). Indeed, in the fulllength talin1 cryo-EM structure, F0-F1 were not visible, again suggesting flexibility between the F0-F1 and F2-F3 di-domains. Whereas the effects of crystal contacts cannot be completely ruled out in a potential or no role of a certain F0-F1 orientation relative to the F2-F3 double subdomains, the collective structural and biochemical data suggest that the isoform-specific amino acid sequence dictates its configuration. It is thus likely that we will uncover many other variations of this tertiary arrangement.

The talin isoforms are monomers
Originally, talin1 was isolated from chicken gizzard and was reported to be in an equilibrium between a monomer and a Talin2 has a novel FERM configuration dimer (72). We overcame the tendency of talin to aggregate and eliminated contributions from impurities of proteins that are enriched when isolating talin from gizzards by using highly purified recombinant proteins. We determined the absolute mass of mammalian full-length talin1 and talin2. Our multiangle light-scattering experiments clearly show that murine talin1 and human talin2 are monomeric as had also been determined for human talin1 (49) while our manuscript was in review. Similarly, the isolated head domains (residues 1-400 for talin1 and 1-403 for talin2) were monomeric. The different elution times of the full-length proteins further suggest that the full-length proteins have an isoform-specific tertiary structure. Thus, the functional differences of the talin isoforms are evident on a primary, secondary, and tertiary structural level.
In the crystal structure, the talin1 F0 subdomain engages in extensive 2-fold related contacts with a symmetry-related F29 that creates an antiparallel dimer. Residues in this F0-F29 interface are conserved between the two isoforms, so these could occur in talin2, but given that talin2 is not linear, it would explain the aggregation observed in our size-exclusion chromatography (SEC)-multiangle light scattering (MALS) as F0 from one talin2 polypeptide chain could engage in interactions with F29 from a second talin2 polypeptide chain, whereas its F2 subdomain could simultaneously interact with F099 from a third talin2 polypeptide chain. In the talin2 crystal, the F1 loop of one talin2 molecule stretches over F29 from a second talin2 molecule to interact with F099 of a third talin2 molecule. Talin2 crystal contacts also involve an extensive interface comprising F0 with all three FERM subdomains (F19, F29, and F39) of a symmetry-related molecule.
The talin isoforms share the membrane attachment site PIP 2 is one established talin activator (73) and contributes to integrin activation by talin recruitment to the plasma membrane, Figure 7. PIP 2 confers membrane-mediated activation of talin. Papillary collecting duct cells with a talin 2/2 background, transiently co-expressing full-length WT or LBD mutant (K274Q, K319Q, K327Q, E345Q, and K346Q) murine GFP-talin1 or human GFP-talin2 with mCherry-CAAX, were fixed and immunostained for F-actin with Alexa Fluor 633 phalloidin. High-resolution confocal images were obtained to show the expression of GFP-talin (green), mcherry-CAAX (red), and F-actin (magenta). Significant differences were observed due to the overexpression of WT and mutant talin, highlighting the requirement of the talin-PIP 2 association for talin attachment to the cell membrane. Scale bar, 10 mm. Figure 8. Membrane attachment of talin is necessary for cell adhesion. A, the focal adhesion area, seen as green specks on the GFP channel is extracted through band pass filtering, threshold, and particle analysis in Fiji is as shown. Representative images for the GFP-talin2 WT (top) and LBD mutant (bottom) are shown with overlay of the outline of focal adhesions over the image of GFP-talin2-expressing cells. B, the total cell areas, measured after applying band pass filtering, threshold, and particle analysis, are shown for the nontransfected control papillary collecting duct cells (n = 61), WT GFP-talin1 (n = 73), mutant GFP-talin1 (n = 22), WT GFP-talin2 (n = 87), and mutant GFP-talin2 (n = 63). The comparison shows mutant constructs exhibiting significant reduction in cell spreading for both isoforms. C, the focal adhesion areas for WT GFP-talin1 (n = 300), mutant GFP-talin1 (n = 188), WT GFP-talin2 (n = 463), and mutant GFP-talin2 (n = 475) were determined and plotted for comparison. The focal adhesion area analyses show significant differences as observed by Sidak's multiple-comparison test p , 0.0001 between the isoforms. However, the mutants seem unaffected by the size of their focal adhesions. The p test values derived from one-way ANOVA analyses were provided for the pairs analyzed in B and C. D, co-expression of talin proteins as GFP-and Neptune2.5-tagged constructs shows similar expression of both constructs within the same cells. Scale bars, 5 mm.

Talin2 has a novel FERM configuration
where talin then activates integrin (54,67,74). The amount of active talin that has its head-tail interaction severed determines adhesion formation. It remains to be seen where the equilibrium between open and closed talin conformer lies.
Whereas charge reversal mutant studies of the isolated F1 loop peptide (residues 139-168, which is ;3% of the entire talin polypeptide chain) alone or of the isolated F1 domain suggested its binding to small unilamellar vesicles and thus suggested that the F1 loop is involved in binding to the plasma membrane (54), such mutagenesis of full-length talin is missing. Notably, superposition of the unbound F2-F3 subdomains of either talin isoform onto the integrin b1D-bound talin2 F2-F3 structure (PDB entry 3G9W) (67) shows that for talin2, the F1 loop is intracellular, whereas in talin1, it is oriented toward the membrane (Fig. 9). This is still the case when taking into consideration that the loop might be very flexible in solution and all sterically possible configurations. Thus, at least for talin2, the F1 loop is not involved in binding to the membrane as previously proposed (54).
Immunostaining studies and mutagenesis strongly support our observation that binding of talin to the membrane is critical to efficiently carry out its cellular functions related to adhesion and maintaining cellular integrity, as evidenced by the lack of actin stress fibers and reduced focal adhesions. Indeed, overexpression of mutant talin proteins in papillary collecting duct talin knockout cells showed limited spreading and impaired focal adhesion formation compared with WT cells. Thus, the downstream talin signaling cascade is severely affected. Clearly, membrane attachment of talin is important to tether integrin to the actin cytoskeleton. The periphery of cells expressing the talin PIP 2 binding-deficient mutants displayed focal adhesionlike features, and about 50% of talin1 or 20% of talin2 still bound to the cell membrane, perhaps through the ;10-fold weaker interactions of talin with other acidic phospholipids, such as phosphatidylserine (58,59). Collectively, our structural, biochemical, and functional data provide significant new molecular insights into the talin isoform-specific cell attachment mechanisms that contribute to the various cell adhesion functions.

DNA constructs
Talin2 head domain residues 1-403 were cloned using fulllength human talin2 as a template into a modified pET-28a vector (EMD Millipore) with an N-terminal octahistidine tag followed by a PreScission protease cleavage site. Mutant talin2 (K274Q, K319Q, E345Q, K346Q, and K327Q) head domain (residues 1-403) was generated by site-directed mutagenesis from the WT talin2 using Agilent Technology's QuikChange site-directed mutagenesis kit.
Full-length human talin2 (residues 1-2,542) and the rod domains of murine talin1 (residues 437-2,541 and 437-2,493) were cloned similarly into modified pET-28a vector to have an N-terminal octahistidine tag with a PreScission protease cleavage site. Full-length murine talin1 was cloned in pET30a vector to have a C-terminal octahistidine tag using standard cloning protocols.
For immunostaining and FLIM studies, full-length talin1 (residues 1-2,541) and its lipid binding-deficient mutant were cloned into a modified pCDNA4 vector with an N-terminal decahistidine tag fused in-frame with EGFP as described previously (53). This His 10 -GFP-talin1 backbone vector was used to subclone full-length talin2 and its mutant to generate His 10 -GFP-talin2 constructs. Similarly, a Neptune2.5 tag was subcloned in place of the GFP tag to produce Neptune2.5-tagged talin2 construct for co-expression experiment. The cytomegalovirus promoter-driven mCherry-CAAX constructs were used directly as described earlier (75). All clones were verified by in-house sequencing.

Protein preparation
Bacterial expression plasmids were transformed in the Escherichia coli strains BL21 (DE3) Rosetta2 (Novagen) or in BL21-CodonPlus (DE3)-RIL (Agilent Technology). Proteins were either expressed by autoinduction (55) or induced with 0.2 mM isopropyl b-D-1-thiogalactopyranoside at 30°C. For Figure 9. Talin isoform-specific membrane anchoring. Superposition of the unbound F2-F3 subdomains of talin1 (A) or talin2 (B) onto the integrin b1D-bound talin2 F2-F3 structure (PDB entry 3G9W) (68). For clarity, only integrin (residues 750-788) is shown (in gray) of the talin/integrin structure. The talin subdomains are colored spectrally (F0 (orange), F1 (yellow), F2 (green), and F3 (blue)). The asterisks highlight the lipid-binding residues shown in stick representation. The talin2 F1 loop, shown in red (residues 130-174), is intracellular, even when taking into account that the loop might be very flexible in solution compared with the crystal structure, while in talin1, it is oriented toward the membrane (drawn in red). A space-filling model of the PIP 2 ligand of our talin1/PIP 2 crystal structure is shown to highlight (i) its location within our full-length talin1 model and (ii) the steric hindrance.
purification of WT or mutant talin2, cell pellets were resuspended in 20 mM Tris-HCl (pH 7.5), 400 mM NaCl, and 10 mM imidazole, lysed by sonication, and clarified by ultracentrifugation (100,000 3 g for 30 min). The clarified cell lysates were loaded onto a nickel affinity column (GE Healthcare), which was equilibrated with 20 mM Tris-HCl (pH 7.5), 400 mM NaCl, and 20 mM imidazole. Proteins were eluted using a 0.5 M imidazole gradient. The histidine tags were removed by overnight cleavage with PreScission protease at 4°C and dialyzed into 20 mM Tris (pH 7.5), 400 mM NaCl, 2 mM DTT, and 1 mM EDTA. Proteins were further purified by size-exclusion chromatography using a Superdex 75 column (GE Healthcare) equilibrated with 20 mM Tris-HCl (pH 7.5), 400 mM NaCl, 2 mM DTT, and 0.1 mM EDTA. All proteins were concentrated, and aliquots were frozen in liquid nitrogen and stored at 220°C. Full-length talin1, talin2, and talin rod domains were purified by nickel-affinity chromatography followed by size-exclusion chromatography on a Superose 6 10/ 300 column (GE Healthcare). For full-length proteins and the talin rod domains (residues 1-2,541, 437-2,541, 437-2,493 for talin1; 1-2,542 for talin2), the salt concentration was maintained at a physiological level of 150 mM.

Talin2 crystallization
Human talin2 (residues 1-403) was concentrated to 5-20 mg/ml. Crystals were grown by hanging drop vapor diffusion at room temperature using a protein/reservoir ratio of 1:1 from 100 mM HEPES (pH 7) and 8% (w/v) PEG 8,000. Hexagonal needle-shaped crystals formed within 2-3 days and continued to grow for a week. Crystals were transferred briefly to a drop containing the reservoir solution that included 25% glycerol as cryoprotectant and mounted with no preferred orientation and flash-frozen in liquid nitrogen. Diffraction-quality crystals were obtained from 100 mM HEPES (pH 7) and 8% (w/v) PEG 8,000.
Talin2 X-ray data collection and processing Human talin2 (residues 1-403) crystals diffracted X-rays at beamline ID22 at the Advanced Photon Source at Argonne National Laboratory to ;2.6-Å Bragg spacings. The beamline was equipped with a Rayonix MX300-HS (MarCCD 300) detector, and a complete data set of 360 frames were collected at 1.0043 Å wavelength with a 0.5°oscillation range. X-ray data were indexed, integrated, and scaled using XDS and AIMLESS as implemented in autoPROC (76). Data reduction statistics are provided in Table 1.

Talin2 structure determination and crystallographic refinement
Phases were obtained by molecular replacement as implemented in the program PHASER (77) using the talin1 F0-F1 subdomain structure (PDB entry 6MFS) (55) and the talin2 F2-F3 subdomain structure (PDB entry 3G9W) (67) as search models. A unique solution was only obtained when both search models were used as ensembles in Phaser with one polypeptide chain in the asymmetric unit in space group P 6 5 , resulting in a calculated volume/mass ratio of 2.18 Å 3 /Da, corresponding to a solvent content of ;0.436. A small fraction of the data (0.06) was identified as merohedrally twinned by Phenix Xtriage (78). Initial refinements without translation, libration, screw rotation (TLS) parameters or twinning operator were performed until convergence using autoBuster (79), and then manually built the model using Coot (80). Further model refinements were carried out by including 12 TLS groups and the twinning operator using Phenix. By taking the twinning operator into account, the R-factor gap was reduced from 10 to 5% using the P 6 5 merohedral twinning operator. Also, part of the F1 insert residues 134-173 were unambiguously resolved after final refinement with Phenix, and the missing region was manually built using omit maps in Coot. However, a small part of the F1 insert, residues 143-153, were not built because of weak electron densities. The quality of the final model was assessed using MolProbity (81), which revealed no outliers, and the final crystallographic refinement statistics are provided in Table 2.
Lipid co-sedimentation assays PIP 2 binding to WT and mutant talin proteins was assayed as described previously (82). Lipid vesicles of PC and PIP 2 were prepared in chloroform to a final composition of 80% PC and 20% PIP 2 . The lipid mixture was dried under an argon stream until the dry lipid film attached to the glass vial bottom and then resuspended in 20 mM Tris (pH 7.5), 400 mM NaCl, and 2 mM 1,4-DTT (buffer A). Small unilamellar vesicles were then produced by sonication after incubating the lipid vesicles at 37°C for 30 min. Samples containing 50 mg of total lipid in a 12ml suspension and 50 mM proteins (WT or mutant) were incubated at 4°C for 1 h followed by centrifugation at 100,000 3 g for 30 min. Supernatants and pellets were separated, and the pellets were carefully washed with buffer A twice and then resuspended in 12 ml of SDS sample buffer. The supernatant and pellet samples were analyzed by SDS-PAGE.

Vesicle pull-down assays
Lipid vesicles were prepared by mixing dissolved phospholipids (Avanti Polar Lipids) to obtain 6% (mol/mol) PIP 2 in PC. The solvent was dried under an argon stream. Lipid films were resuspended in 20 mM Hepes (pH 7.5), 400 mM NaCl, 0.5 mM EDTA, 0.5 mM tris(2-carboxyethyl)phosphine (TCEP) and sonicated until resuspended. Vesicle pulldowns were performed with 2 mM talin1 (residues 1-400) or talin2 (residues 1-403) by serial dilution of the vesicles. After 2 h of incubation, the vesicles were centrifuged for 15 min at 16,100 3 g at 4°C. Supernatants were collected, and the pelleted vesicles were washed and resuspended. Protein pellets and supernatants were quantified by the Bradford method (Bio-Rad) at an absorption of 590 nm. For all samples, the buffer blanks were subtracted.

Multiangle light scattering (MALS)
SEC with MALS analyses were carried out on an Agilent 1260 Infinity HPLC system equipped with a variable wavelength detector for UV absorption monitoring coupled to a DAWN-HELEOS II multiangle light-scattering detector (Wyatt Technology) and OptiLab T-rex differential refractive index detector (Wyatt Technology). SEC was carried out at a Talin2 has a novel FERM configuration flow rate of 0.5 ml/min in 20 mM Tris-HCl, pH 8, 400 mM NaCl, 0.5 mM EDTA, and 1 mM DTT buffer for the talin head domains and 20 mM Tris-HCl, pH 7.5, 150 mM NaCl, and 0.5 mM TCEP buffer for the full-length talin protein. BSA at 2 mg/ml was used as a standard, and ASTRA software version 6.1 was used to control the data acquisition and analysis to determine the molar mass of the protein.

Thermal shift analysis
The structural integrity of the mutant proteins was verified by thermal shift analyses to compare the denaturation profile with the WT protein and was carried out as reported earlier (83). Briefly, proteins were subjected to thermal melt analyses in a standard white opaque quantitative PCR plate in a total volume of 20 ml. The analyses were carried out in a buffer of 20 mM Tris-HCl, pH 7.5, 400 mM NaCl, and 1 mM DTT containing a 10 mM protein sample and 2.5-fold concentration of SYPRO Orange dye (Thermo Fisher Scientific). The Roche 480 thermocycler was used for measuring the denaturation profile using the predefined melting curve program with excitation at 483 nm and emission of 568 nm. Samples were allowed to equilibrate at 25°C for 15 s prior to ramping the temperature to 90°C at 0.1°C/s with continuous monitoring of the change in fluorescence. All experiments were carried out twice. The averages of triplicate sample measurements from each experiment were plotted to obtain denaturation profiles, and the first derivative plot of change in fluorescence versus temperature was used to derive melting temperature, T m .
Immunostaining and FLIM cell preparation talin 2/2 papillary collecting duct cells (kindly provided by Roy Zent, University of Vanderbilt) were used for transient transfection and further analyses by FLIM and immunostaining. Cells were routinely cultured in Dulbecco's modified Eagle's medium F-12 (Thermo Fisher Scientific) supplemented with 10% fetal bovine serum, 100 mg/ml streptomycin, and 100 units/ml penicillin at 37°C with 5% CO 2 in a humidified incubator. For live-cell imaging or confocal microscopy, full-length GFP-talin constructs were transiently co-transfected in papillary collecting duct cells along with mCherry fused at the C terminus by the K-Ras-derived CAAX domain (DGKKKKKKSK-TKCVIM). Transfections were carried out in a 6-well plate using Lipofectamine 3000 (Lifesciences Inc.) as per the manufacturer's suggested protocol. Cells transfected with GFP-talin alone were used as a control. After transfections, cells were used to probe expression of GFP-talin constructs by immunoblotting using the anti-GFP antibody (Millipore Sigma), and the remaining cells were used either for live-cell imaging for FLIM analyses or for immunostaining. For the latter, 24 h posttransfection, cells were trypsinized and seeded on a 2-mg/ml fibronectin-coated coverslip at a cell density of 0.5 3 10 6 cells in a 24-well plate in a total of 0.5 ml of growth media. Cells were allowed to attach for 24 h and washed twice with PBS and subsequently fixed and permeabilized concurrently using 0.5% Triton X-100 and 4% paraformaldehyde for 5 min at 37°C and 5% carbon dioxide (CO 2 ). After washing with PBS, cells were incubated for another 20 min at 37°C and 5% CO 2 to aid proper cell fixation. Cells were washed three times with PBS, blocked with 1% BSA in PBS, and stained for actin with Alex Fluor 633 phalloidin in 1% BSA for 1 h. Subsequently, cells were washed twice with PBS and mounted with or without NucBlue TM . Confocal images of the immunostained cells were recorded on an Abberior Instruments Expert-Line STED (stimulated emission depletion) nanoscope using the pulsed 485-, 561-, and 640-nm excitation laser lines in a line-interleaved fashion with the detection windows set between 500 and 550 nm for the green channel, between 610 and 630 nm for the red, and between 650 and 720 nm for the far-red channels. All images were acquired using a 100-fold, 1.4 numerical aperture oil immersion objective lens (Olympus), the QUAD beam scanner, and the Imspector software package (Max-Planck Innovation).
For the total cell area and focal adhesion measurements, confocal images were carried out on Zeiss LSM880 laser-scanning confocal microscope using either plan-apochromat 320 (1.4 numerical aperture) or plan-apochromat 363 (1.4 numerical aperture) oil differential interference contrast objectives. Excitation of the GFP fluorophore was achieved using the 488-nm laser, and the emission was recorded at 524 nm. Appropriate band pass filtering and threshold were applied to enable proper measurement of the total cell area and focal adhesions using the "Analyze particles" option in the Fiji (ImageJ) software (84). To eliminate the inclusion of measurement outliers (including, for example, multiple fused adhesions, thick cell periphery in minimally spread cells), focal adhesion areas that were more than 3 times the areas represented by the median were excluded from the analyses. For the mutants, this was much more difficult, as only some cells exhibited clear focal adhesions and the remaining cells had connecting features that falsely represented focal adhesions. Here, low band pass filtering was administered.

FLIM with FRET two-photon (2p) microscopy and 2p FLIM
To measure the talin association with the membrane, we performed 2p FLIM imaging using a custom 2p microscope (85). Briefly, GFP was excited with a Ti:sapphire laser (Chameleon, Coherent) at a wavelength of 920 nm and a power of 1.4-1.6 milliwatts measured below the objective. mCherry was excited by a 1,100-nm, 120-fs pulsed, 80-MHz two-photon INSIGHT laser (Spectra-Physics). The fluorescence was collected with an objective (360, 1.0 numerical aperture, Olympus), divided with a dichroic mirror (565 nm), and detected with a lowtransfer time spread photoelectron multiplier tube (H7422-40p from Hamamatsu) placed after a 510/70 2p wavelength filter (Chroma) for green fluorescence and a 620/90 2p wavelength filter (Chroma) to collect red fluorescence. Fluorescence lifetime images were acquired with two PCI-6110 cards (National Instruments) using a time-correlated single-photon counting board (Timeharp 260 P; PicoQuant) controlled with custom software written in C# (source code available at https://github.com/ryoheiyasuda/FLIMage_public).

FLIM analyses
The fluorescence lifetime decay curve F(t) from cells with similar expression levels of the talin sensor (number of photons Talin2 has a novel FERM configuration less than 500,000 and donor/acceptor ratios between 0.5 and 4) was fit with a biexponential function convolved with the Gaussian pulse response function, FðtÞ ¼ F 0 ðP D Hðt; t 0 ; t D ; t G Þ 1 P AD Hðt; t 0 ; t AD ; t G ÞÞ, where t AD is fluorescence lifetime of the donor bound with acceptor, P D and P AD are the fraction of free donor and donor undergoing FRET with the acceptor, respectively, and H(t) is an exponential function convolved with the Gaussian instrument response function Hðt; t 0 ; t D ;t G Þ ¼ where t D is the fluorescence lifetime of the free donor, t G is the width of the Gaussian pulse response function, t 0 is the time offset, and erfc is the complementary error function. We obtained t D from single-exponential fits (P AD = 0) to fluorescence lifetime curves of GFP-talin1 expressed in live cells (2.72 ns 6 0.01, n = 14) and GFP-talin2 (2.75 ns 6 0.008, n = 8) and estimated t AD (1.33 ns). For experimental data, we fixed t D and t AD to these values to obtain stable fitting. To generate a fluorescence lifetime image, we calculated the mean photon arrival time (,t.), in each pixel as ,t. = ð t F t ð Þdt= ð F t ð Þdt. The mean photon arrival time is related to the mean fluorescence lifetime (t m ) by an offset arrival time t 0 , which is obtained by fitting the whole image to t m = ,t. 2 t o. The binding fraction of talin (P AD ) was calculated from fitting.

Data availability
The coordinates were deposited with the Protein Data Bank, accession code 6U4K. All other data are contained within the article and the accompanying supporting information.
Acknowledgments-We are indebted to the staff of the South Eastern Regional Access Team (Argonne National Laboratory) for synchrotron support. talin 2/2 cells were a generous gift from Roy Zent (Vanderbilt School of Medicine). The human talin2 cDNA was a generous gift from Richard O. McCann (University of Kentucky). We thank Charmane Gabriel (Oxbridge Academy) for technical help, Nicolai Urban (Max Planck Florida Institute) for help with microscopes, and Peter Moody (Leicester University) and Katrin Karbstein (The Scripps Research Institute) for helpful structure and biochemistry insights, respectively.
Funding and additional information-T. I. is supported by grants from the National Institutes of Health, the Department of Defense, and the National Science Foundation and by start-up funds provided to The Scripps Research Institute from the State of Florida. This is publication 29850 from The Scripps Research Institute. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.
Conflict of interest-R. Y. is a founder of Florida Lifetime Imaging LLC.