C-terminal COOH of Integrin β1 Is Necessary for β1 Association with the Kindlin-2 Adapter Protein*

Background: Modulation of integrin/kindlin interactions can lead to human disease pathogenesis. Results: Interaction between integrin β1 and the kindlin-2 is mainly governed by the β1 C-terminal carboxylate moiety. Conclusion: The interaction chemistry identified here between integrin β1 and kindlin-2 appears to represent a novel protein-protein interaction mode. Significance: The unusual integrin β1/kindlin-2 association chemistry sheds new light on this aspect of integrin biology. Protein-protein interactions are driving forces in cellular processes. As a prime example, transmembrane integrins link extracellular matrix and intracellular proteins, resulting in bidirectional signaling that regulates cell migration, proliferation, differentiation, and survival. Here we provide the first evidence that interaction between the integrin β1 cytoplasmic tail and kindlin-2, a member of a family of adapters implicated in human disease pathogenesis, is mainly governed by the β1 C-terminal carboxylate moiety and is required for laterality organ development in zebrafish. Affinity measurements indicate that this unusual protein-protein interaction mode is coordinated by a putative carboxylate-binding motif in the kindlin-2 FERM subdomain F3. Contrary to the C terminus of proteins that engage PDZ domains, the C-terminal three residues of β1, per se, do not contribute to kindlin-2 binding or to laterality organ development. Thus, by employing zebrafish as an in situ physiological tool to correlate protein structure and function, we have discovered an unexpected association chemistry between an integrin and a key adapter involved in integrin signaling.

Protein sequence motifs are the basic building blocks of protein-protein interactions, where the identity of amino acid side chains in these motifs and their three-dimensional configurations provide the specificity required for complex cellular signaling networks to function (1)(2)(3). Integrins represent a large family of adhesion and signaling receptors that interact with a multitude of extracellular and intracellular proteins by using various sequence motifs (4). Modulation of these interactions can impact physiological events such as angiogenesis, hemostasis, and immune cell function (5). Recently, the interaction of integrins with members of the kindlin family of adapter proteins has gained considerable attention because kindlins can cooperate with other intracellular proteins to activate integrins and to mediate adhesion-dependent cellular responses, in part through interactions of the kindlin FERM 2 (four point one, ezrin, radixin, moesin) subdomain F3 with integrin ␤ cytoplasmic tails (6 -8). To date, there is no available high-resolution structure of an integrin ␤ tail/kindlin F3 interaction, and consequently, the molecular basis of integrin/kindlin interactions remains incompletely understood (9,10). Therefore, in the present study, we utilized zebrafish as a physiological platform in parallel with in vitro biochemical analyses to provide the first insights into the interaction chemistry between the integrin ␤1 tail and the ubiquitously expressed kindlin family member, kindlin-2.
was chosen because of its cleavage efficiency in zebrafish (16). All mRNAs, including T2A bicistronic constructs, were synthesized with the mMESSAGE mMACHINE T7 Ultra kit (Ambion, Austin, TX). Either mRNAs were injected alone into 1-4 cell stage blastulae or in rescue experiments 250 pg of respective mRNAs were co-injected with either ␤1bEI10 or Kin2b morpholinos to a final concentration of ϳ6 ng/embryo.
For ELISA binding assays, biotinylated recombinant integrin tails (100 ng) were bound at saturating concentrations (20 g/ml) to wells of NeutrAvidin-coated microtiter plates (Thermo Scientific) (17,18). Purified FERM proteins were diluted in PBS with 1% BSA, added to wells in duplicates or triplicates, and incubated overnight at 37°C. After three washes with PBS ϩ 0.2% Tween 20 and subsequent incubations with mouse anti-HA antibody and horseradish peroxidase-conjugated antimouse IgG, bound proteins were quantified in an ELISA reader at 490 nm. Competition assays were performed in the presence of 10 M synthetic Q-peptides with varying concentrations of FERM domains. Binding isotherms were generated by nonlinear regression to a single-site binding model with PRISM (GraphPad, San Diego, CA).
Synthetic Peptides-To assess carboxylate function at the C terminus of the ␤1 cytoplasmic tail, we used synthetic peptides originating from the ␤1 tail membrane-distal 15-amino acid segment (see Fig. 4). These peptides either had free acid at the C terminus (acid peptides) or were amidated (amide peptides). Synthetic Q-acid and Q-amide peptides were obtained from NeoBioLab (Cambridge, MA), and WT sht -amide was from Selleckchem (Houston, TX). All three peptides were biotinylated at the N terminus with a mini PEG linker, and their molecular identities were confirmed by mass spectroscopy.
Live Imaging, Immunohistochemistry, and Image Analysis-General methods were similar to Ref. 12 After live DFC scoring, the embryos were allowed to develop to 6 -9 somite stage at 28.5°C, and some live embryos were scored for the presence/absence of Kupffer's vesicle (KV) with a Zeiss stereo microscope, and/or remaining embryos were fixed in 4% paraformaldehyde and used for immunohistochemistry. Phenotypic classification of KVs was also similar to Ref. 12: WT, under bright field a visible circular, single vesicle, with a diameter larger than notochord, was located at the ventral-caudal tip of notochord, or by confocal imaging, GFP(ϩ) cells were located on the lining of KV, and its inflated lumen was GFP(Ϫ); Mutant, under bright field a visible circular, single vesicle, or occasionally multiple vesicles, with a diameter equal or smaller than notochord, or by confocal imaging inflated GFP(Ϫ) KV lumen was not identifiable and GFP(ϩ) cells at the ventralcaudal tip of notochord were in disarray; Absent, embryos with GFP(ϩ) endoderm with no visible KV, and/or embryos with no axis.

RESULTS
mRNAs That Encode Mutant Integrin ␤1b Cytoplasmic Tails Perturb Laterality Organ Development-In a recent study we showed that antisense MO-mediated depletion of the integrin ␣V and ␤1b subunits leads to defective DFC migration during zebrafish gastrulation (12). DFCs are the precursor cells of the KV, a ciliated organ involved in zebrafish left-right body axis specification (20,21) Disorganized DFC migration in ␤1b morphants impedes KV formation, resulting in left-right defects in zebrafish (12). Because integrin ␤ cytoplasmic tails represent a major signaling hub for integrin interaction with cellular proteins involved in integrin signaling (22), we first explored the signaling potential of the integrin ␤1b tail in the DFC-to-KV organization processes by taking advantage of the dominantnegative effects of integrin ␤1b overexpression (22)(23)(24). We hypothesized that a cytoplasmic tail-less version of ␤1b ("⌬C mutant"), but not wild-type ␤1b (WT), would phenocopy the gastrulation defects of ␤1b morphants (Fig. 1, A and B) (12). When WT ␤1b mRNA was delivered into zebrafish blastulae, over 77% of embryos formed oval-shaped DFC clusters (Fig. 1, C and F), and later the majority developed single, normally inflated KVs (77.4 Ϯ 4.8%) (Fig. 1, D, E, and G). In sharp contrast, only 32.2 Ϯ 4.6% of embryos injected with ␤1b(⌬C) had a normal DFC phenotype. Instead the majority of ␤1b(⌬C) DFCs (65.0 Ϯ 5.4%) were confined to a linear domain that had multiple gaps within this cellular cluster ( Fig. 1, C and F). Later in development, ␤1b(⌬C) mutants also showed a major reduction in inflated KVs (44.8 Ϯ 3.7%) ( Fig. 1, D
Consequently, we knocked down kindlin-2 with MOs to evaluate kindlin-2 involvement in the process of DFC-to-KV organization. DFCs were properly clustered in over 71% of control morphants, which later developed KVs with fully inflated lumens (Fig. 3, G-K). In contrast, 48.8 -51.1% of embryos injected with either of two kindlin-2 MOs (Kin2a; Kin2b) showed DFCs that did not aggregate properly to form inflated KVs. Furthermore, the effects of these MOs could be partially rescued by co-injection of kindlin-2 mRNA (Fig. 3K). When DFC-selective morphants were generated (DFC Kin2a , DFC Kin2b )(12), they exhibited DFC phenotypes similar to that of the Kin2a and Kin2b morphants (Fig. 3L). As the DFC-to-KV organization processes require integrin ␣V␤1b (12), we then evaluated ␣V and kindlin-2 genetic interaction by delivering substantially lower doses of each individual MO with the result that less than 15% of embryos injected with either ␣V1 or Kin2 MOs had mutant phenotypes (Fig. 3M). This profile was significantly increased (Kin2a, 38.2 Ϯ 5.3%; Kin2b, 51.5 Ϯ 8.3%; p Ͻ 0.05) when ␣V1 and Kin2 MOs were co-injected, whereas no such effect was observed when a negative control MO (␣V1ctl) was co-injected with Kin2a or with Kin2b (11.3 Ϯ 2.1%, and 8.8 Ϯ 5.2%, respectively) (Fig. 3M). In addition, the effects of these MO co-injections could be partially rescued by inclusion of ␣V and/or kindlin-2 mRNAs (Fig. 3M). Collectively these in vivo results are consistent with the idea that kindlin-2 can be recruited to the integrin ␤1b tail through the latter's EGK segment, and that this interaction is essential in laterality organ development.
Kindlin Binding-defective Integrin Mutants Show Differences in Their Affinity Profiles, and These Behaviors Can Be Phenocopied in Zebrafish-To further elucidate the biochemical nature of the ␤1b/kindlin-2 interaction, we measured direct binding affinities of recombinant ␤1b cytoplasmic tail model peptides to purified kindlin-2 F3. This subdomain of kindlin-2 was chosen because: 1) it harbors the ␤1 tail interaction interface ( Fig. 4A) (6,8); 2) it enables ease of in vitro production and purification; and 3) F3 subdomains can function autonomously (30,31).
Binding affinities of the human ␤1A tail and the zebrafish ␤1b tail for kindlin-2 F3 were similar (0.23 and 0.24 M, respectively), and they were in agreement with the reported binding affinity of full-length human kindlin-2 with the ␤1A tail (9, 32). However, three ␤1b tail mutants that failed to pull down kindlin-2 ( Fig. 2C) exhibited distinct differences in their affinity profiles (Fig. 4B). For example, although ␤1b(Y795A) tail affinity was reduced 16-fold as compared with wild-type ␤1b, the ␤1b(T788A) mutant showed a 110-fold reduction, and ␤1b(⌬EGK) had no discernable affinity (Fig. 4, B and C). In analogous experiments, when ␤1b mRNAs were delivered over a wide range of concentrations (8 -125 pg/embryo) into zebrafish embryos, all three integrin mutant ␤1b(⌬EGK), ␤1b(T788A), and ␤1b(Y795A) mRNAs caused dose-dependent increases in the malformed DFC phenotype (Fig. 4D). Collec-tively, these results suggest that under physiological conditions, all three kindlin-2 binding-impaired mutants of integrin ␤1b (⌬EGK, T788A, and Y795A) could induce DFC organization defects. The Integrin ␤1 and Kindlin-2 Association Is Mainly Governed by the ␤1 C-terminal Carboxylate Moiety and Is Required for Laterality Organ Development in Zebrafish-Because the ␤1b ⌬EGK mutation had the most potent impact on both in vitro and in vivo assays, we investigated further the EGK-mediated kindlin-2 recruitment to the ␤1b tail by testing an additional series of mutant ␤1b-EGK sequences in pulldown and direct binding ELISA assays (Fig. 5). Surprisingly, none of these modifications eliminated ␤1b/kindlin-2 interaction in pulldowns or impaired the affinity of ␤1b for kindlin-2 F3. Thus, there is no particular amino acid requirement within the EGK segment for ␤1b binding to kindlin-2.
The only major functional chemical moiety that was left unexplored by these binding experiments is the COOH moiety at the ␤1b C terminus. To evaluate this, three independent experimental schemes were employed (Fig. 6). First, the impact of gradual ␤1b C-terminal extensions on kindlin-2 affinity was examined because the effective concentration of peptide tail ends decreases inversely with an increase in the chain length (33), thus making ␤1b C-terminal COOH less available to interact with a hypothetical carboxylate binding-segment in kindlin-2 F3. Therefore, we introduced 1ϫ, 2ϫ, and 3ϫ FLAG tags at the caudal end of ␤1b-Y795, resulting in ϩ6, ϩ18, and ϩ30 residue extensions to ␤1b (Fig. 6A). All three extended peptides pulled down full-length talin-1, and all showed similar affinities to purified talin-1 F2,3 (data not shown). However their affinities for full-length kindlin-2 and its F3 subdomain were reduced as the ␤1b tail length was increased (Fig. 6, B and C).
Second, short peptides derived from the membrane-distal, C-terminal 15 amino acids of the ␤1 tail were tested either in their free acid form (COOH: WT-acid, Q-acid) or in their amide form (CONH 2 : WT-amide, Q-amide) (Fig. 6D). Two such peptides were designed because there is a glutamate residue (Glu-796) near the C terminus of ␤1, and we evaluated its potential side chain carboxylate function with an E796Q mutant peptide (Q-acid). Although both acid peptides were able to pull down full-length kindlin-2 and showed considerable binding affinities for kindlin-2 F3 (0.54 Ϯ 0.09 and 1.4 Ϯ 0.34 M), their amide forms failed to pull down kindlin-2 or to interact with kindlin-2 F3 with discernable affinity (Fig. 6, E and  F). As a control, both amide peptides still retained the ability to pull down another ␤1b-interacting protein, Lyn (Fig. 6F). In addition, although the Q-acid peptide was able to inhibit ␤1b/ kindlin-2 F3 association, its amide form lacked this ability (Fig.  6G), and neither form of the Q peptides blocked ␤1b tail binding to talin-1 F2,3 (data not shown). In analogous experiments, when the Q-acid peptide was delivered to zebrafish blastulae over a wide range of concentrations (4.4 -44 pg/embryo), it was able to induce mutant DFC phenotypes, but its amide form did not seem to have this ability (Fig. 6H).  APRIL 18, 2014 • VOLUME 289 • NUMBER 16

JOURNAL OF BIOLOGICAL CHEMISTRY 11187
Third, we studied the effects of overexpression of an irrelevant, large C-terminal ␤1b fusion construct (␤1b-GATA1a) in zebrafish embryos to ascertain the effects of ␤1b C-terminal blockade in a physiological context (Fig. 6I). Unlike WT ␤1b, injection of mRNA coding for a large fusion construct caused significantly impaired DFC clustering (p Ͻ 0.05) and led to malformed KV development (p Ͻ 0.01) (Fig. 6, J-M). Collectively, these results indicate that the carboxylate moiety at the ␤1b C terminus is essential for ␤1b/kindlin-2 association.  . Kindlin-specific ␤1b tail mutants show differences in their kindlin-2 F3 binding profiles, and these behaviors can be phenocopied in zebrafish. A, the sequence alignments of human and zebrafish kindlin-2 F3 and talin-1 F3 subdomains. Eight residues that have been identified as essential for talin-1 F3 binding and/or activation of ␣IIb␤3 are indicated with a red asterisk. Two out of the eight seem to be conserved in kindlin-2; Ile-654 is not essential for kindlin-2/␤1 interaction (green asterisk) (6). B and C, direct binding of purified zebrafish kindlin-2 F3 (B) and talin-1 F2,3 (C) to wild-type and mutant integrin ␤1b tails assessed by enzyme-linked immunosorbent assay using NeutrAvidin-bound biotinylated ␤1b cytoplasmic tails. WT, black filled circle; Y795A, blue filled circle; T788A, green filled circle; ⌬EGK, red filled circle. Nonspecific binding was assessed by ␣IIb tail binding (gray cross). Binding isotherms were generated by nonlinear regression to a single-site binding model (solid lines). D, WT and mutant ␤1b mRNA titrations on DFC clustering in live zebrafish embryos. Shown is the ␤1b mRNA dose dependence of mutant DFC clustering. Number of embryos used in this study: WT, 1138; T788A, 600; Y795A, 805; ⌬EGK, 329. Error bars indicate mean Ϯ S.E. FIGURE 5. EGK sequence in ␤1b cytoplasmic tail is not essential. A, summary of amino acid sequences alignments of ␤1b tail mutants, their associated pulldown results, and their observed direct binding affinities to purified zebrafish kindlin-2 F3. B and C, pulldown analyses with recombinant ␤1b cytoplasmic tails with HUVEC lysates. Conditions for pulldown analysis were similar to Fig. 2, and analyses of binding isotherms were similar to Fig. 4. N.B., no binding; n.d., not determined.

A Putative Carboxylate-binding Motif in Kindlin-2 F3 May Coordinate Interactions with the ␤1 C-terminal Carboxylate
Moiety-We postulated that kindlin-2 F3 has a corresponding carboxylate-binding segment. To our knowledge, PDZ domain-containing proteins are the only known examples of proteins that possess a C-terminal carboxylate-binding motif (h-Gly-h) (34,35). In addition, a highly conserved, positively charged amino acid located at the N terminus of the h-Gly-h motif is essential for coordination of electrostatic interactions with the terminal carboxylate (34 -36). Indeed, the kindlin-2 F3 subdomain has a potential carboxylate-binding segment located between Lys-581 and Ile-588 (KKDELIGI) (Fig. 7A). Orthologous alignments of kindlin-2 sequences indicated that this sequence is completely conserved between mammalian, FIGURE 6. The C-terminal COOH of the integrin ␤1 cytoplasmic tail mediates ␤1/kindlin-2 association. A, gradual C-terminal extensions of recombinant ␤1b tails were generated with successive FLAG tags. B and C, pulldown assays with HUVEC lysates (B) and direct binding curves for kindlin-2 F3 (C) with C-terminally extended ␤1b cytoplasmic tails. Numbers of FLAG tags are indicated in parentheses: WT, black filled circle; Y(1)FLAG, blue filled triangle; Y(2)FLAG, magenta filled triangle; Y(3)FLAG, purple filled triangle; ⌬EGK, red filled circle; and ␣IIb, gray cross. D, sequence alignments of recombinant and synthetic ␤1b tail peptides used in panels E-G. Either a recombinant 15-amino acid membrane-distal (MD) segment of the WT ␤1 tail C terminus was bacterially expressed as a fusion protein, or its synthetic derivatives, E796Q mutant (Q), were prepared either as free acid (WT sht -acid, Q-acid) or as amides (WT sht -amide, Q-amide). E and F, direct binding profiles of free acid and amide peptides with purified kindlin-2 F3 (E) and pulldown assays with HUVEC lysates (F). WT sht -acid, red triangle; WT sht -amide, cyan triangle; Q-acid, red square; Q-amide, cyan square; and ␣IIb, gray cross. G, inhibition of ␤1b/kindlin-2 F3 association by synthetic Q peptides (10 M). ␤1b, full-length ␤1b tail, black circle; ␤1b ϩ Q-acid, ␤1b tail with Q acid peptide, red filled circle; ␤1b ϩ Q-amide, ␤1b tail with Q-amide peptide, cyan filled circle; and ␣IIb, gray cross. H, effects of Q-acid and Q-amide peptide titrations on DFC clustering in live zebrafish embryos. Shown is the effect of synthetic peptide dose on mutant DFC clustering. Number of embryos used in this study: Q-acid, 376; Q-amide, 380. I, schematic of bicistronic mCherry-T2A-␤1b vectors that were used to co-express mCherry and ␤1b proteins. J and K, confocal images of Tg(sox17:GFP)-expressing embryos (green) injected with mRNAs that code for WT and ␤1b-GATA1a fusion. J and K, dorsal views of 80% E embryos (J) and 6 -8 somite stage (SS) embryos (K) are shown in all panels; anterior is at the top. J, multiple focal planes at the center of the DFC cluster. Embryos were immunolabeled with anti-cherry antibody (red) and nucleus-stained (blue); channels are merged. K, multiple focal planes at the center of KV immunolabeled with anti-acetylated tubulin antibody (white). avian, amphibian, teleost, and insect lineages (Fig. 7A). Therefore, we generated point mutants of kindlin-2 F3 and evaluated their binding affinities to the ␤1b tail (Fig. 7B). Of the mutant kindlin-2 F3 subdomains, K581E, G587L, and I586G/I588G completely abrogated binding affinity to the ␤1b tail (Fig. 7, C  and D). Although a conserved tryptophan-to-alanine (W615A) mutant of kindlin-2 is known to disrupt ␤1 tail/kindlin-2 interaction by a pulldown assay (6,8), the binding profile of this kindlin-2 F3 mutant revealed a considerable affinity for the ␤1 tail (11.15 Ϯ 0.35 M), but less than that of WT kindlin-2 F3 (Fig. 7E). In contrast, conserved mutations in three other positively charged residues (R592E, R595E, and R608E) had no effect on binding affinity (Fig. 7F).
Although these observations support a key role for kindlin-2 in laterality organ development through its interaction with the C-terminal EGK residues of integrin ␤1, to our surprise mutational analyses of the EGK segment revealed that the identity of the amino acid in this segment is not important for ␤1/ kindlin-2 interaction. Rather, the C-terminal carboxylate of ␤1-EGK is the most essential moiety for this interaction. Specifically, 1) pulldown profiles of full-length kindlin-2 and direct binding affinity of kindlin-2 F3 were inversely correlated with C-terminal extensions onto the ␤1b tail; 2) chemical modification of the C-terminal carboxylate of synthetic peptides derived from the ␤1b tail completely abolished affinity for kindlin-2; 3) WT synthetic peptide was able to induce mutant DFC phenotypes, and this ability was lost when its C-terminal amide form was delivered into zebrafish blastulae; and 4) overexpression of a large C-terminal ␤1b fusion protein led to DFC-to-KV organization defects during zebrafish gastrulation (Fig. 6). It is interesting to note that whereas the last residue of the ⌬EGK mutant, Tyr-795, carries two potential binding moieties at the same time (Tyr and backbone COOH), it still cannot interact with kindlin-2. This might arise from a conformational dichotomy caused by a stearic hindrance because F3 domains are known to have a relatively rigid conformation (37), which would not allow tandem attachments to take place at the same time. In addition, lack of ␤1/kindlin-2 association in ␤1b ⌬EGK could be rescued by inclusion of a single residue (YG peptide, Fig. 5) indicating that flexibility resides in this region of the ␤1 tail, which was also seen in ␤3/talin-1 NMR studies (37). Evidence provided here also indicates that this interaction requires a generalized carboxylate-binding motif (h-Gly-h) (34) in the kindlin-2 F3 subdomain, and mutations within this segment both obliterate ␤1/kindlin-2 F3 affinity in vitro and induce DFC-to-KV organization defects in vivo (Fig. 7).
To date, proteins carrying PDZ domains are the only known example where the C-terminal carboxylate moiety of their ligands contributes to the interaction and where the interaction has an absolute sequence requirement for the C-terminal three (or four) residues of the ligand (1,3,36,38). In contrast, these corresponding residues in integrin ␤1 do not contribute to discriminative binding. Therefore, we postulate that the interaction chemistry identified here between integrin ␤1 and kindlin-2 F3 represents a novel protein-protein interaction mode. However, it should be emphasized that such interaction chemistry may not necessarily be a shared feature of other integrin ␤ tail/kindlin complexes, such as that between integrin ␤2 or ␤3 and kindlin-2 (9) or that between integrin ␤6 and kindlin-1 or kindlin-2 (39). Therefore, additional studies with different ␤ integrin/kindlin permutations will be required to fully understand their association chemistries.