The focal adhesion protein kindlin-2 controls mitotic spindle assembly by inhibiting histone deacetylase 6 and maintaining α-tubulin acetylation

Kindlins are focal adhesion proteins that regulate integrin activation and outside-in signaling. The kindlin family consists of three members, kindlin-1, -2, and -3. Kindlin-2 is widely expressed in multiple cell types, except those from the hematopoietic lineage. A previous study has reported that the Drosophila Fit1 protein (an ortholog of kindlin-2) prevents abnormal spindle assembly; however, the mechanism remains unknown. Here, we show that kindlin-2 maintains spindle integrity in mitotic human cells. The human neuroblastoma SH-SY5Y cell line expresses only kindlin-2, and we found that when SH-SY5Y cells are depleted of kindlin-2, they exhibit pronounced spindle abnormalities and delayed mitosis. Of note, acetylation of α-tubulin, which maintains microtubule flexibility and stability, was diminished in the kindlin-2–depleted cells. Mechanistically, we found that kindlin-2 maintains α-tubulin acetylation by inhibiting the microtubule-associated deacetylase histone deacetylase 6 (HDAC6) via a signaling pathway involving AKT Ser/Thr kinase (AKT)/glycogen synthase kinase 3β (GSK3β) or paxillin. We also provide evidence that prolonged hypoxia down-regulates kindlin-2 expression, leading to spindle abnormalities not only in the SH-SY5Y cell line, but also cell lines derived from colon and breast tissues. The findings of our study highlight that kindlin-2 regulates mitotic spindle assembly and that this process is perturbed in cancer cells in a hypoxic environment.

ever, remains unknown. We note that the other kindlin family member, kindlin-1, has been reported to interact directly with histone deacetylase 6 (HDAC6) to regulate ␣-tubulin acetylation affecting mitotic spindle assembly (30,31). However, kindlin-1 is expressed primarily in epithelial cells, thus precluding its role as a regulator of mitotic spindle assembly in other cell types (5). Whether kindlin-2, which is widely expressed, regulates HDAC6 and ␣-tubulin acetylation, thereby affecting mitotic spindle assembly, warrants investigation.
Here, we show for the first time that kindlin-2 regulates mitotic spindle assembly by inhibiting HDAC6 activity via signaling pathways involving AKT, GSK3␤, and paxillin, thereby modulating ␣-tubulin acetylation in mitotic spindles. Further, this regulatory mechanism is perturbed in cancer cells under hypoxic stress as a consequence of kindlin-2 down-regulation.

Kindlin-2-depleted neuroblastoma SH-SY5Y cells exhibit delayed mitosis and spindle abnormalities
To examine the role of kindlin-2 in maintaining mitotic spindle integrity, we used a lentivirus-based shRNA transduction system to generate kindlin-2-depleted neuroblastoma SH-SY5Y stable cell line, which was verified by immunoblotting and qRT-PCR (Fig. 1A). We chose the SH-SY5Y cell line because it only expressed kindlin-2. Depletion of kindlin-2 did not cause any up-regulation of kindlin-1 or kindlin-3 (Fig. 1B). Positive controls DLD-1 (colorectal adenocarcinoma cell line) and K562 (chronic myelogenous leukemia cell line) were included for kindlin-1 and kindlin-3, respectively. Hence, these data exclude functional compensation by other kindlins in kindlin-2-depleted SH-SY5Y cells in subsequent assays. In line with its role as a positive regulator of integrin function, cells depleted of kindlin-2 were defective in adhesion on fibronectin based on an electric cell-substrate impedance-sensing assay (Fig. 1C). We also transfected kindlin-2-depleted cells with a shRNA-resistant HA-tagged kindlin-2 WT plasmid to generate another stable cell line (henceforth referred to as HA-kindlin-2 (WT)-rescued cells). HA-kindlin-2 (WT)-rescued cells did not exhibit adhesion defect compared with WT and control shRNA cells (Fig. 1C). To determine cell proliferation, we performed tumor spheroid assays. Kindlin-2-depleted cells but not HA-kindlin-2 (WT)-rescued cells formed smaller tumor spheroids compared with WT or control shRNA cells (Fig. 1D), suggesting that kindlin-2 is important in cell proliferation.
We then performed cell cycle synchronization and determined the time taken for these cells to transit from G 2 /M phase to cytokinesis by live-cell imaging (Fig. 1E). Kindlin-2-depleted but not HA-kindlin-2 (WT)-rescued cells took a significantly longer time to transit from G 2 /M phase to cytokinesis compared with WT and control shRNA cells, suggesting a delay in mitosis upon kindlin-2 depletion. We ruled out shRNA offtarget effects because the delay in mitosis was also observed in another SH-SY5Y stable cell line expressing a different kindlin-2-targeting shRNA sequence (Fig. S1, A and B).
Mitotic spindles are microtubules (MT) formed by ␣and ␤-tubulins. Hence, we performed DAPI and immunofluorescence (IF) staining of ␣-tubulin to visualize the spindle assem-bly in all groups of mitotic cells. Kindlin-2-depleted but not HA-kindlin-2 (WT)-rescued cells exhibited multipolar spindles, abnormal bipolar spindles, and chromosome misalignment (collectively referred to as spindle abnormalities) when compared with WT and control shRNA cells (Fig. 1, F and G), suggesting that kindlin-2 is important in mitotic spindle assembly. A defect in mitotic spindle assembly was also observed in another stable cell line expressing a different kindlin-2targeting shRNA sequence (Fig. S1C).

Kindlin-2 depletion diminishes ␣-tubulin acetylation in mitotic spindles
Tubulin acetylation provides mechanical stability to MT, which regulates intracellular trafficking, cell motility, and cell cycle (32)(33)(34). In the early stages of mitosis, spindle MT contains a high level of acetylated Lys 40 in ␣-tubulin (Ac-tubulin) (34 -36). Acetylation of ␣-tubulin has been shown to increase the stability of bent microtubules (37,38). Ac-tubulin is a substrate of HDAC6, an enzyme that deacetylates nuclear and cytoplasmic proteins (39). Tubacin is a selective inhibitor of HDAC6 (40). Tubacin has a hydroxamate group that chelates Zn 2ϩ ion of the catalytic centers in HDACs, and it has a 1,3dioxane group that accounts for its selectivity for HDAC6 (41,42). Hence, we performed IF staining of Ac-tubulin and total ␣-tubulin in mitotic cells without or with tubacin treatment ( Fig. 2A). Quantitative analysis of the fluorescence intensity ratio of Ac-tubulin to total ␣-tubulin of the mitotic spindles in each cell was determined (Fig. 2B). In the absence of tubacin treatment, a significantly low level of Ac-tubulin was detected in the mitotic spindles of kindlin-2-depleted but not HA-kindlin-2 (WT)-rescued cells compared with that of WT and shRNA control cells. The diminished level of Ac-tubulin in mitotic spindles was also detected in another stable cell line expressing a different kindlin-2-targeting shRNA sequence (Fig. S1D), thus excluding an shRNA off-target effect. In the presence of tubacin, there was an increase in the level of Actubulin in mitotic spindles of all cells compared with untreated cells (Fig. 2B). Importantly, tubacin treatment restored the level of Ac-tubulin in mitotic spindles of kindlin-2-depleted cells to a level comparable with other groups of treated cells. Together, these data suggest that depleting kindlin-2 in cells leads to an increase in HDAC6 activity. Given that cells depleted of kindlin-2 exhibited delayed transit from G 2 /M to cytokinesis, we also examined the effect of tubacin on these cells. Tubacin treatment effectively recovered the G 2 /M-to-cytokinesis delay in kindlin-2-depleted cells (Fig. 2C). These data provide further evidence supporting a role of kindlin-2 in modulating acetylation of ␣-tubulin in mitotic spindles by inhibiting the activity of HDAC6.
Unattached kinetochore delays anaphase onset (43). The spindle assembly checkpoint (SAC) network is activated even when a single kinetochore is not attached to the spindle. Mitotic arrest deficiency protein 2 (MAD2), a key component of the SAC (44), is recruited to unattached kinetochores, and it inhibits APC/C cdc20 activity that is required for entry into anaphase (45). We therefore performed IF staining of MAD2 together with an anti-centromere antibody (ACA) in all groups of mitotic cells. In line with spindle abnormalities observed in a Kindlin-2 regulates mitotic spindle assembly high percentage of mitotic kindlin-2-depleted cells, MAD2 was readily detected at the centromere of nonaligned chromosomes in these cells but not HA-kindlin-2 (WT)-rescued cells (Fig. 2D).

Kindlin-2 regulates HDAC6 activity via the PI3K/AKT (PKB)/GSK3␤ and paxillin signaling pathways
Kindlin-1 localizes at centrioles in mitotic cells, and it directly interacts with HDAC6 to regulate mitotic spindle MT acetylation (30). Hence, we performed IF staining of kindlin-2 in both interphase and metaphase cells using two antibodies from different sources. Like focal adhesion protein paxillin, kindlin-2 was readily detected at focal adhesions in interphase cells on fibronectin (Fig. S2, A and B). However, its localization at centrosomes containing ␥-tubulin in metaphase cells was not evident (Fig. S2, C and D). This suggests the possibility of a different mechanism by which kindlin-2 regulates mitotic spindle MT acetylation.
It has been shown that spindle abnormalities and mitotic arrest in kindlin-1-depleted cells is due to an increase in HDAC6 activity (30). The PI3K/AKT (PKB)/GSK3␤ signaling axis is important in mitotic spindle assembly in which AKT and GSK3␤ are detected at the centrosome (46). AKT inhibits GSK3␤ and prevents GSK3␤ from activating HDAC6 (47). Given that kindlin-2 is involved in integrin-mediated activation of AKT (48 -50), we conjectured that HDAC6 activity will be elevated in kindlin-2-depleted cells. We first examined active AKT and GSK3␤ levels in G 2 /M synchronized cells on fibronectin without or treated with integrin-activating manganese by immunoblotting (Fig. 3A). Under both conditions, active AKT level was significantly lower in cells depleted of kindlin-2 compared with that in other groups of cells. By contrast, the opposite trend was observed for GSK3␤. We then determined the activity of HDAC6 in these cells using a fluorometric enzyme assay (Fig. 3B). An increase in HDAC6 activity was detected in kindlin-2-depleted cells relative to other groups of cells. Consistent with these observations, Ac-tubulin level was also lower in kindlin-2-depleted cells (Fig. 3A). Kindlin-2 promotes integrin activation and outside-in signaling. Manganese directly activates integrins, bypassing the need for kindlin-2 in integrin activation. The similar activity profiles of AKT and GSK3␤ in both untreated and manganese-treated kindlin-2-depleted cells lend support to the regulation of HDAC6 by integrin outside-in signaling involving kindlin-2/ AKT/GSK3␤ pathway. Because kindlin-2 mediates integrin outside-in signaling that regulates AKT activity, we asked whether inhibiting AKT activity can cause mitotic spindle defects. Indeed, WT SH-SY5Y cells treated with SH6, a phosphatidylinositol analog inhibitor of AKT (51), exhibited a significant increase in spindle abnormalities compared with untreated cells (Fig. 3C).
Interestingly, paxillin has been reported to inhibit HDAC6, thereby regulating MT acetylation, although direct interaction between them remains to be verified (52). Given that kindlin-2 recruits paxillin to nascent adhesion sites (50), we performed co-immunoprecipitation assay using 293T cells overexpressing HA-tagged kindlin-2 and mCherry-paxillin (Fig. 3D). Both mCherry-paxillin and endogenous HDAC6 were detected in the co-precipitate of HA-kindlin-2 (left). Both HDAC6 and HA-kindlin-2 were also detected in the co-precipitate of mCherry-paxillin (right). We extended the study to examine whether depletion of paxillin leads to spindle abnormalities in SH-SY5Y cells. We generated three stable cell lines expressing control shRNA and two paxillin-targeting shRNAs (different sequences) (Fig. 3E). Depletion of paxillin resulted in a significant increase in spindle abnormalities detected. Taken together, these data suggest that an alternative signaling axis that regulates MT acetylation can be one that involves integrin, kindlin-2, paxillin, and HDAC6. Both signaling pathways need not be mutually exclusive.
To validate that the kindlin-2 expression is relatively even across the population of cells analyzed, we performed immunofluorescence staining of metaphase cells (Fig. S3A). We examined the image of each cell using the ImageJ software to determine its kindlin-2 fluorescence intensity and its area. The ratio of kindlin-2 fluorescence intensity over cell area (in arbitrary units) was calculated and plotted. Expression levels of HAkindlin-2 (WT) and mutants in the rescued cells were comparable, and the expression of HA-kindlin-2 (WT), but not its mutants, significantly decreased the percentage of metaphase cells with spindle abnormalities (Fig. S3B). Expression of HA- Figure 1. Depletion of kindlin-2 in SH-SY5Y cells caused adhesion defects, delayed mitosis, and mitotic spindle abnormalities. A (left), immunoblotting (IB) of cell lysate from cells expressing ctrl shRNA or kindlin-2 (K2)-targeting shRNA with antibody against kindlin-2. GAPDH served as loading control. Protein bands were quantified by densitometry. Values below protein bands represent the mean -fold differences in protein expression levels relative to WT (normalized to 1.0) from three independent experiments. Right, qRT-PCR analysis of kindlin-2 transcript level in cells. The expression of kindlin-2 transcript in WT cells is normalized to 1.0. Three independent experiments were performed. Statistical calculation was performed using a two-tailed paired Student's t test. p Ͻ 0.05 is considered significant. n.s., not significant. B, immunoblotting of cell lysate with antibodies against kindlin-1 or kindlin-3. GAPDH served as loading control. C, ECIS measurements of cells on fibronectin. Each data point is the mean Ϯ S.D. (error bars) of technical triplicates. D, 3D culture of cells was performed, and the spheroid image area was measured. Each data point in the graph represents one cell. Two-tailed unpaired t test was performed. Representative images of spheroids are shown. Scale bar, 400 m. E, time taken (min) for cells to transit from G 2 /M to cytokinesis. Statistical calculation was performed using a two-tailed unpaired Student's t test. F, immunofluorescence staining of ␣-tubulin and chromosomes in metaphase WT, ctrl shRNA, K2 shRNA, and HA-K2 (WT) rescued cells. Scale bar, 5 m. G, the percentage of metaphase cells containing normal versus abnormal spindles was determined. The numbers of cells analyzed were as follows: 17 (WT), 19 (ctrl shRNA), 20 (K2 shRNA), and 42 (HA-K2 (WT) rescued) collated from three independent experiments. Statistical calculation was performed using a two-tailed unpaired Student's t test. p Ͻ 0.05 is considered significant. n.s., not significant.

Kindlin-2 regulates mitotic spindle assembly
kindlin-2 (WT), but not its mutants, also rescued mitotic spindle MT acetylation (Fig. 3H), and fully recovered the delayed transit from G 2 /M to cytokinesis (Fig. 3I). Taken together, these data lend additional support to the regulatory role of kindlin-2 in mitotic spindle MT acetylation and the maintenance of spindle integrity.

Hypoxia down-regulates kindlin-2 expression, which affects mitotic spindle assembly
Our analyses thus far showed that kindlin-2 plays a role in mitotic spindle assembly by, in part, regulating mitotic spindle MT acetylation. Our assays were based on comparing cells with or without kindlin-2 depletion using the shRNA method. We then asked whether kindlin-2 expression is down-regulated under specific physiological conditions, leading to mitotic spindle anomalies. Hypoxia alters the cell cycle and perturbs normal mitotic spindle assembly (54 -56). Hypoxia is a hallmark of solid tumors, and it is well-documented to affect gene expression in cancer cells (57). Thus, we examined the effect of hypoxia on SH-SY5Y cells in culture. Hypoxia reduced cell proliferation over time (Fig. 4A), and there was a significantly higher number of mitotic cells with spindle abnormalities compared with cells under normoxia (Fig. 4B). Mitotic spindle MT acetylation was also significantly reduced in cells under hypoxia compared with those under normoxia (Fig. 4C). Notably, the expression level of kindlin-2 transcript and protein decreased over time in cells under hypoxia (Fig. 4, D and E), whereas the opposite trend was detected for HIF-1␣ and VEGFA (Fig. 4, F and G (left)), which are in line with hypoxia-mediated stabilization of HIF-1␣ and induction of VEGFA (57).
Next, we determined the expression of microRNA (miR)-138 in these cells because miR-138 has been demonstrated to target the 3Ј-UTR of kindlin-2 transcript, leading to the down-regulation of kindlin-2 expression and ␤1 integrin signaling in prostrate cancer cells (58). miR-138 expression was significantly upregulated in cells under hypoxia over time (Fig. 4G, right). This is concordant with changes in the expression profile of miRNA, including miR-138, in cells under low oxygen tension (59). To further demonstrate that hypoxia-induced miR-138 down-regulates kindlin-2, which leads to mitotic spindle abnormalities, we first generated a stable SH-SY5Y cell line expressing GFPkindlin-2 that lacks the 3Ј-UTR-targeting site of miR-138. These cells were then subjected to normoxia or hypoxia culture conditions. Overexpression of GFP-kindlin-2 effectively reduced the number of cells with mitotic spindle abnormalities under hypoxia (Fig. 4H), lending support to its role in maintaining normal mitotic spindle assembly. Taken together, these data suggest that hypoxia-induced miR-138 down-regulates kindlin-2, which could in part account for the spindle abnormalities observed in mitotic cells under hypoxia.
To rule out cell type-specific effects, we also tested the effects of hypoxia on four additional cell lines: HCT116 (colorectal cancer), MDA-MB-231 (breast cancer), CCD841 CoN (colon epithelial, nontumorigenic origin), and MCF10A (breast epithelial, nontumorigenic origin). Similar to SH-SY5Y, kindlin-2 expression was reduced in these cell lines under hypoxia together with a concomitant increase in mitotic spindle abnormalities detected ( Fig. 4I and Fig. S4A). To further verify that kindlin-2 plays a role in regulating mitotic spindle assembly in these cells, kindlin-2 depletion was performed by using the lentivirus-based shRNA transduction system described above. For HCT-116 and MDA-MB231, stable cell lines with kindlin-2 depletion were generated. For CCD841 CoN and MCF10A, transiently transduced cells were analyzed instead because of the high rate of cell death under prolonged antibiotic selection for stable lines, and CCD841 CoN is very slow-growing. Nevertheless, in all four cell lines, an increase in mitotic spindle abnormalities was detected as a result of kindlin-2 depletion compared with control shRNA and WT cells ( Fig. 4J and Fig.  S4B).
3D culture of cancer cells mimics better the tumor microenvironment compared with 2D culture (60 -63). Thus, we generated stable SH-SY5Y cells carrying the 5HRE-GFP plasmid in which GFP expression is regulated by five copies of human VEGF-hypoxia-responsive element (64). We used these cells for in vitro culture of tumor spheroids and isolated cells under hypoxic stress by flow cytometry cell sorting based on GFP signal (Fig. 5A). Kindlin-2 expression was significantly lower in GFP hi cells compared with GFP lo cells based on immunoblotting (Fig. 5A) and qRT-PCR (Fig. 5B). By contrast, both miR-138 and VEGFA transcripts were up-regulated in GFP hi cells compared with GFP lo cells (Fig. 5B). To further validate the effect of hypoxia on kindlin-2 expression in vivo, we performed subcutaneous xenograft of the 5HRE-GFP cells in immunecompromised Rag Ϫ/Ϫ /IL2R Ϫ/Ϫ mice. Tumors were excised and cryosectioned, followed by IF staining and imaging (Fig.  5C). Hypoxic regions, as indicated by GFP expression, were detected in the tumor sections, and expression of kindlin-2 in these regions was lower than that in neighboring nonhypoxic regions based on fluorescence intensity analysis.

Discussion
Kindlins are well-established positive regulators of integrinmediated cell adhesion and migration. The kindlin-2 ortholog Fit1 has previously been identified to maintain proper spindle assembly in Drosophila S2 cells, but the mechanism is unknown (29). Here, we show that kindlin-2 mediates integrin outside-in signaling that activates AKT (Fig. 5D). Active AKT then inhibits GSK3␤, preventing GSK3␤ from activating HDAC6. In turn, this prevents deacetylation of ␣-tubulin, leading to stable spin-

Kindlin-2 regulates mitotic spindle assembly
dle MT assembly in mitotic cells. In parallel, kindlin-2 may recruit paxillin that directly inhibits HDAC6. Our findings also indicate that hypoxia up-regulates miR-138 that suppresses kindlin-2 expression, leading to mitotic spindle abnormalities.
Cell rounding is a hallmark of most cell types undergoing mitosis. Thus, it appears intriguing that kindlin-2-mediated integrin signaling regulates spindle assembly in cells undergoing mitosis, which is known to disassemble focal adhesions. However, the contact between cells and ECM before entry into mitosis has been shown to be crucial for spindle positioning (65,66). Growth of MT from centrosome and bipolar spindle formation are also regulated by integrins (67). A recent detailed analysis on cell-substrate remodeling during mitosis revealed the presence of punctate active ␤1 integrin contacts that are maintained between the mitotic cell and its ECM substrate (68). Therefore, the signaling axis involving integrin, kindlin-2, AKT, GSK3␤, paxillin, and HDAC6 remains relevant in cells undergoing mitosis.
Kindlin-2 is not the only kindlin that regulates mitotic spindle assembly. In an early study by Patel et al. (31), it was reported that mitotic spindle formation is regulated by kindlin-1, which is phosphorylated by Polo-like kinase 1 (Plk-1), a serine/threonine kinase that regulates mitotic checkpoints (69).
Another key observation in their study is that kindlin-1 localizes to the centrosome. This may explain how kindlin-1 is phosphorylated by Plk-1, which is also found at the centrosome and mitotic spindle poles. In our analyses, we were unable to detect endogenous kindlin-2 at centrosomes of mitotic cells, although kindlin-2 was readily detected at focal adhesions in interphase cells. Whether this is due to the level of sensitivity of detection remains to be explored. More studies are therefore needed to determine whether Plk-1 interacts and regulates kindlin-2 function. Recently, Patel et al. (30) also reported that kindlin-1 directly interacts with HDAC6 to regulate ␣-tubulin acetylation in mitotic spindles. In line with this, our data suggest that kindlin-2 inhibits HDAC6 activity, which affects mitotic spindle ␣-tubulin acetylation. However, this regulation occurs via signaling pathways involving AKT, GSK3␤, and paxillin.
Finally, another important finding from this study is that kindlin-2 expression is down-regulated in cells under hypoxic stress as a result of elevated level of miR-138. Interestingly, cells under hypoxia exhibited pronounced mitotic spindle abnormalities that were effectively prevented by overexpressing kindlin-2 in 2D culture. The down-regulation of kindlin-2 expression was also observed in 3D tumor spheroid and in vivo tumor, although we are technically limited in terms of specifically isolating G 2 /M phase cells from the tumors to examine mitotic spindle integrity and ␣-tubulin acetylation. Nevertheless, our data collectively provide mechanistic insights into the regulation of mitotic spindle assembly by kindlin-2, which could be perturbed in cancer cells in a hypoxic tumor microenvironment.

Chemicals and reagents
All general chemicals and reagents were purchased from Bio-Rad and Merck (Sigma-Aldrich) (Darmstadt, Germany) unless stated otherwise.

Cells and culture conditions
The human neuroblastoma SH-SY5Y cell line was kindly provided by Dr. Esther Siew-Peng Wong (School of Biological Sciences, Nanyang Technological University, Singapore). The cell lines MDA-MB-231 (human breast cancer), HEK-293T (human embryonic kidney), and HCT 116 (human colorectal cancer) were obtained from the American Type Culture Collection (ATCC) (Manassas, VA). SH-SY5Y, HEK-293T, and MDA-MB-231 cells were maintained in DMEM (Hyclone, Logan, UT), and HCT 116 cells were maintained in McCoy's 5A (modified) medium (Thermo Fisher Scientific) supplemented with 10% (v/v) heat-inactivated fetal bovine serum (Gibco, Thermo Fisher Scientific), penicillin (100 IU/ml), and streptomycin (100 g/ml) (Nacalai Tesque, Kyoto, Japan). The cell line CCD841 CoN (human colon, nontumorigenic origin) (ATCC) was kindly provided by Dr. Karen Crasta (Lee Kong Chian School of Medicine, Nanyang Technological University) and maintained in Eagle's minimal essential medium (Gibco) containing supplements mentioned above. The cell line MCF10A (breast epithelial, nontumorigenic origin) (ATCC) was kindly provided by Dr. Andrew Tan Nguan Soon (Lee Kong Chian School of Medicine, Nanyang Technological University) and maintained in DMEM/F-12 medium (Gibco) supplemented  HDAC6. A, AKT and GSK3␤ activities in G 2 /M synchronized cells were determined by immunoblotting using antibodies against phospho-AKT at Ser 473 and phospho-GSK3␤ at Ser 9 . Protein bands were quantified by densitometry. Values below protein bands represent the mean -fold differences of phosphoprotein/total protein relative to WT cells (normalized to 1.0) from at least three independent experiments. The level of acetylated ␣-tubulin and total ␣-tubulin in cells was also determined by immunoblotting. Statistical calculation was performed using a two-tailed unpaired Student's t test. B, HDAC6 activity in G 2 /M synchronized cells was determined using a fluorometric enzyme assay kit. Eight independent experiments were performed. Statistical calculation was performed using a two-tailed paired Student's t test. C, treatment of G 2 /M synchronized WT SH-SY5Y cells with 10 M AKT inhibitor (SH6) induced a significant increase in spindle abnormalities compared with untreated cells. 43 cells per group from three independent experiments were analyzed. Statistical calculation was performed using a two-tailed unpaired Student's t test. D, immunoprecipitation (IP) assay using 293T cells transfected with HA-kindlin-2 and mCherry-paxillin. Left, immunoprecipitation using anti-HA antibody. Right, immunoprecipitation with anti-paxillin antibody. IB, immunoblotting. E, stable SH-SY5Y cells expressing control shRNA, paxillin-targeting shRNA (A), or paxillin-targeting shRNA (B) were generated. Top, paxillin expression in these cells was determined by immunoblotting. GAPDH served as loading control. Bottom, depletion of paxillin in SH-SY5Y cells resulted in an increased number of cells with spindle abnormalities. The numbers of cells analyzed were as follows: 38 (ctrl shRNA), 42 (pax-shRNA A), and 49 (pax-shRNA B) from three independent experiments. Statistical calculation was performed using a two-tailed unpaired Student's t test. F, immunoblot analysis of stable HA-kindlin-2 (WT)-rescued cells, HA-kindlin-2 (Q614A/W615A)-rescued cells, and HA-kindlin-2 (PH-deleted)-rescued cells. Kindlin-2depleted cells were transfected with shRNA-resistant HA-tagged WT, integrin-binding defective (QW/AA), or paxillin-binding defective (PH⌬) kindlin-2 expression plasmid. Re-expression of kindlin-2 was determined by immunoblotting with anti-kindlin-2 antibody. Actin served as loading control. G, the percentage of metaphase cells containing normal spindles versus abnormal spindles was determined. The numbers of cells analyzed were as follows: 47 (ctrl shRNA), 45 (K2 shRNA), 42 (K2 WT rescued), 30 (K2 QW/AA rescued), and 36 (K2 PH⌬ rescued) from at least three independent experiments. Statistical calculation was performed using a two-tailed unpaired Student's t test. H, fluorescence intensity ratio of acetylated ␣-tubulin to total ␣-tubulin in mitotic spindles was determined. I, time taken for cells to transit from G 2 /M to cytokinesis. In F and G, each data point represents one cell analyzed. Data were collated from at least three independent experiments. Statistical calculation was performed using a two-tailed unpaired Student's t test. Error bars, S.D.

Generation of stable cell lines
Stable expression of shRNA targeting kindlin-2 in SH-SY5Y cells was generated by using a lentivirus-based shRNA transduction system. Five GFP-C-shLenti plasmids were generated in which four plasmids contained different kindlin-2-targeting shRNA sequences (referred to as K2A, K2B, K2C, and K2D) and one plasmid contained a scrambled negative control sequence (referred to as ctrl) (OriGene Technologies, Rockville, MD). Each of these plasmids was transfected together with thirdgeneration lentiviral packaging plasmids (Applied Biological Materials) into HEK-293T cells, and the culture supernatant that contained pseudovirions was harvested. SH-SY5Y cells (2 ϫ 10 5 ) were cultured in 1 ml of fresh medium supplemented with 1 ml of virion-containing supernatant and 6 g/ml Polybrene (Sigma-Aldrich). 24 h post-transduction, cells were sorted based on GFP signal on a FACSAria flow cytometer (BD Biosciences) and maintained in full medium containing 1.5 g/ml puromycin (Thermo Fisher Scientific). The expression of kindlin-2 in these cells was determined by immunoblotting and qRT-PCR. Cells expressing the kindlin-2-targeting sequence (K2B) 5Ј-ACAACAGTGACAAAGAAGTTGATGAA-GTT-3Ј showed the highest efficacy in silencing the expression of kindlin-2. These cells were used in subsequent experiments. Another SH-SY5Y stable cell line expressing K2A shRNA with the targeting sequence 5Ј-GAAGAACTTATTGGAATTGCA-TACAACAG-3Ј was also used in verification studies (Fig. S1) Prior to the rescue experiments, SH-SY5Y K2B cells were subcloned. The expression of kindlin-2 in these clones was screened and verified by immunoblotting. One clone (referred to as K2B clone 4) was selected and expanded in culture. To re-express kindlin-2 in K2B clone 4, transfection was performed by incubating 1.5 ϫ 10 5 cells in 250 l of serum-free DMEM containing 2.5 g of shRNA-resistant HA-kindlin-2 (WT, Q614W615/AA, or PH-deleted) plasmid (zeocin resistance gene) and 7.5 l of TransIT-X2 reagent (Mirus Bio LLC, Madison, Wisconsin) according to the manufacturer's instructions. Cells with stable expression of kindlin-2 were selected in full culture medium containing 700 g/ml Zeocin TM (Thermo Fisher Scientific).
The stable SH-SY5Y cell line that expressed GFP under hypoxia was generated by transfecting 2 ϫ 10 5 WT SH-SY5Y cells with 3 g of 5HRE-GFP plasmid (neomycin resistance gene) (a gift from Dr. Martin Brown and Dr. Thomas Foster, Stanford University School of Medicine, Stanford, CA (64), Addgene plasmid 46926) in 250 l of serum-free DMEM containing 3 l of Lipofectamine 2000 according to the manufacturer's instructions. Cells were maintained in full culture medium containing 900 g/ml G418-sulfate (PAA Laboratories).
To generate stable paxillin-depleted cells, -H1 vector set (eGFP reporter, puromycin selection marker) containing paxillin-targeting shRNA (four different targeting sequences) or control sequence was purchased from Genecopoeia (Rockville, MD). Cells were transfected with 2.5 g of plasmid, each using Lipofectamine 2000 as mentioned above, followed by selection in medium containing 1.5 g/ml puromycin. Paxillin expression was effectively silenced in cells expressing paxillin-shRNA targeting sequence A (5Ј-GGAGAGTCTCTTGGAT-GAACT-3Ј) or targeting sequence B (5Ј-GCAACCTTTCTG-AACTCGACC-3Ј) compared with cells expressing control shRNA, and they were used in subsequent experiments.
Stable kindlin-2-depleted HCT 116 and MDA-MB-231 cells were generated by using the GFP-lentivirus-based shRNA transduction system (as described above for SH-SY5Y cells) followed by selection in full medium containing 1.5 g/ml puromycin. The same lentiviral system was used to transduce CCD841 CoN and MCF-10A, but stable cell lines were not generated due to the high percentage of cell death after prolonged exposure to puromycin selection. Instead, these cells were used immediately for subsequent experiments 2 days after transduction.

Expression plasmids
HA-tagged full-length human kindlin-2 (isoform 2) in pcDNA3.1(zeo) (3,70) was used to generate shRNA-resistant HA-kindlin-2 expression plasmid. Silent mutations were made at the third base of each codon found in the kindlin-2 cDNA sequence that is the target of kindlin-2 shRNA K2B using the QuikChange TM site-directed mutagenesis kit (Agilent Tech- Kindlin-2 regulates mitotic spindle assembly nologies, Santa Clara, CA) and relevant primers. This K2B shRNA-resistant HA-kindlin-2 expression plasmid was then used to generate the two mutants: the integrin-bindingdefective Q614W615/AA HA-kindlin-2 mutant (17, 53) (in this case Q621W622/AA in the kindlin-2 isoform 2) by performing additional site-directed mutagenesis using relevant primers and the paxillin-binding defective PH-deleted HA-kindlin-2 mutant using overlapping PCR with relevant primers to remove the region Pro 350 -Gln 480 .

Tumor spheroid culture
Cells (8 ϫ 10 3 ) in 100 l of full culture medium were seeded into each well of a 96-well culture plate that was precoated with 1% (w/v) autoclaved agarose (Vivantis Technologies, Selangor Darul Ehsan, Malaysia) followed by centrifugation at 500 ϫ g for 5 min. Cells were maintained under standard culture conditions for 7 days. To determine the size of the spheroids, they were visualized under a microscope with a 4ϫ objective and phase contrast (Olympus IX83 microscope). Images were taken and analyzed using the CellSens Dimension software.

Hypoxia treatment
For comparison between the effects of normoxia versus hypoxia on cells after 24 h or 48 h, the seeding density for cells was 2 ϫ 10 6 cells/100-mm cell culture dish. To make a comparison after 96 h, the seeding density for cells was reduced to 4 ϫ 10 5 cells/100-mm cell culture dish to prevent overgrowth of cells. For each time point, cells were seeded into two cell culture dishes. The medium was replaced with fresh medium after 24 h. One dish of cells was maintained under standard culture conditions (normoxia), whereas another dish of cells was placed in a custom-built hypoxia chamber. The chamber was connected to a flow meter (Dwyer Instruments Inc., Michigan City, IN) and filled with a gas mixture of 95% N 2 and 5% CO 2 , and the level of O 2 was measured using a gas detector (Industrial Scientific, Pittsburgh, PA). The chamber was then transferred to a 37°C incubator. The chamber was flushed twice with the gas mixture on the same day.

FACS
SH-SY5Y 5HRE-GFP tumor spheroids were cultured as described above. The spheroids were washed in PBS, trypsinized, and resuspended in PBS containing 20% (v/v) heatinactivated fetal bovine serum. Cells were passed through an appropriate size filter to remove clumps and sorted by FACS based on GFP signal on a FACSAria flow cytometer (BD Biosciences). Cells sorted into GFP lo and GFP hi groups were used for further analyses. Flow cytometry data are presented using the FlowJo software (Tree Star Inc., Ashland, OR).

Cell proliferation assay
Cell proliferation and viability was determined using the ADAM-MC TM automatic cell counter and viability kit (Nano-EnTek, Inc., Seoul, Korea).

Electric cell-substrate impedance sensing (ECIS) measurement
ECIS measurement was performed as described previously (3). In brief, 8 ϫ 10 4 cells were seeded into each well of a 16-well E-plate device (ACEA Biosciences, San Diego, CA) that was precoated with 2.5 g/cm 2 human fibronectin (Merck, Sigma-Aldrich). AC impedance measurements were taken at 5-min intervals on a real-time cell electronic system (ACEA Biosciences) placed in a cell culture incubator. AC impedance is expressed as arbitrary units termed the cell index. This value correlates with the magnitude of cell adhesion and spreading. Each data point represents the mean Ϯ S.D. of technical triplicates.

Immunoblotting
Cells were lysed in lysis buffer (150 mM NaCl, 1% (v/v) Igepal, and 10 mM Tris, pH 7.4) containing inhibitors of proteases (Roche, Basel, Switzerland) and phosphatases (Nacalai Tesque, Nakagyo-ku, Kyoto). Protein concentration was measured using the bicinchoninic acid assay kit (Thermo Fisher Scien-  Table  presentations are shown in expanded images. D, an illustration of kindlin-2 signaling in the regulation of ␣-tubulin acetylation. The interaction between kindlin-2 and integrin promotes phosphorylation and activation of AKT. Activated AKT inhibits GSK3␤ activity, leading to a decrease in HADC6 activity, which in turn allows the acetylation of ␣-tubulin at mitotic spindles. Acetylation of ␣-tubulin has been shown to increase the stability of bent microtubules (37,38). The activity of HDAC6 can be inhibited by paxillin (52). In solid tumors, hypoxia could lead to a decrease in kindlin-2 expression as a consequence of an increased expression of miR138. The result of a diminished level of kindlin-2 expression includes the formation of abnormal spindles in mitotic cells. Error bars, S.D.

Kindlin-2 regulates mitotic spindle assembly
tific). Proteins were resolved by SDS-PAGE under reducing conditions and electrotransferred onto polyvinylidene difluoride membrane (Bio-Rad). Polyvinylidene difluoride membrane was incubated in TBST blocking buffer (150 mM NaCl, 0.1% (v/v) Tween 20, 25 mM Tris, pH 8.0) containing 5% (w/v) nonfat milk at room temperature (RT) for 30 min. For phosphorylated protein detection, TBST containing 5% (w/v) BSA was used instead. The membrane was then incubated in blocking buffer containing the primary antibody at 4°C overnight. After incubation with the primary antibody, the membrane was washed three times in TBST buffer and incubated with appropriate HRP-conjugated secondary antibody at RT for 1 h. The following secondary antibodies were used: HRP-conjugated goat anti-mouse IgG and HRP-conjugated goat anti-rabbit IgG antibodies from Advansta (Menlo Park, CA) and HRPconjugated goat anti-rat IgG antibody from GE Healthcare. The membrane was washed three times in TBST buffer followed by ECL detection using the WesternBright ECL kit (Advansta). Images were captured on a ChemiDoc TM Touch imaging system (Bio-Rad).

Immunoprecipitation assay
The immunoprecipitation assay was performed as described previously with modifications (70). 3 ϫ 10 6 293T cells were seeded into a 10-cm culture dish. The following day, cells were co-transfected with expression plasmids HAkindlin-2 (WT) pcDNA3.1(zeo) (30 g) and pmCherry-paxillin (a gift from Kenneth Yamada; Addgene plasmid 50526; RRID:Addgene_50526) (10 g) using a polyethyleneimine (30 g) (Sigma-Aldrich)-based method. Cells were harvested by trypsinization 24 h post-transfection, resuspended in full culture medium containing 5 mM MnCl 2 , and seeded on a fibronectin-coated 10-cm culture dish followed by a 3-h incubation under culture conditions. Cells were harvested and lysed in lysis buffer (150 mM NaCl, 1% (v/v) Igepal, and 10 mM Tris, pH 7.4) containing protease inhibitor mixture (Roche) on ice for 30 min. The lysate was precleared with 3 g of mouse serum IgG (Sigma-Aldrich) and protein A-Sepharose beads (GE Healthcare) for 30 min at 4°C on a roller. The precleared lysate was separated into two equal portions. For the HA-kindlin-2 immunoprecipitation exper-iment, 3 g of mouse serum IgG was added to one portion and 3 g mouse anti-HA antibody was added to the other portion (Merck). For the paxillin immunoprecipitation experiment, 4 g of mouse serum IgG and 4 g of mouse anti-paxillin antibody (clone 5H11) were added to the two portions of lysate, respectively. Protein A-Sepharose beads were added to each portion, and the samples were incubated for 3 h at 4°C on a roller. Beads were recovered by centrifugation followed by washing three times in lysis buffer. Precipitated proteins were eluted by boiling beads in Laemmli buffer containing DTT followed by SDS-PAGE and immunoblotting.

Synchronization of cells
Cells were seeded in a culture dish precoated with fibronectin (2.5 g/cm 2 ). Thymidine (Sigma-Aldrich) was added to a final concentration of 2 mM, followed by incubation for 14 h under standard culture conditions. The medium was discarded, and cells were washed in PBS. Deoxycytidine (Sigma-Aldrich) was added to the culture at a final concentration of 24 M followed by incubation for 9 h under standard culture conditions. Thymidine was then added to a final concentration of 2 mM followed by another 14 h of incubation. Cells were washed twice in PBS, and the medium was replaced with fresh medium containing 24 M deoxycytidine followed by 2 h of incubation. Cells were then washed twice in PBS before adding fresh medium containing 10 M RO3306 (Sigma-Aldrich) followed by 19 h of incubation. Cells were released from the effect of RO3306 by washing twice in PBS and adding fresh medium to the cells. Cells were immediately used for live-cell imaging or incubated for another 90 min under standard culture conditions before the immunoblotting assay or immunofluorescence staining. In experiments involving tubacin treatment, immediately after RO3306 release, cells were either treated with 2 M tubacin (Sigma-Aldrich) and subjected to live-cell imaging or treated with 2 M tubacin for 90 min followed by fixing, immunostaining, and confocal laser-scanning microscopy analysis.
For AKT inhibition experiment, cells were first treated with thymidine, deoxycytidine, and RO3306 as described above. After 6 h of RO3306 treatment, AKT inhibitor SH6 (Abcam, UK) was added to the cell culture to a final concentration of 10 M followed by overnight incubation. Cells were released from RO3306 treatment by washing, and fresh medium containing 10 M SH6 was added back to the cells. After a 90-min incubation, cells were washed and fixed for IF microscopy.

Live-cell imaging to determine the time taken to transit from G 2 /M to cytokinesis
Aforementioned, cells were released from the effect of RO3306 by washing in PBS and adding fresh medium to the cells. The dish containing the cells was placed immediately on an inverted microscope (Olympus IX83) equipped with a motorized temperature-controlled (37°C) and 5% CO 2 incubator stage and a cooled monochrome digital camera. The capturing field was selected randomly within 5 min after the dish was placed on the stage. Live-cell images under a ϫ10 phasecontrast objective (UPLFLN-PH, Olympus) were captured every 2 min for 6 h using CellSens Dimension software (Olym-Kindlin-2 regulates mitotic spindle assembly pus). The duration of mitosis was calculated based on the time taken for a round cell to divide into two daughter cells.

IF microscopy
Cells (3 ϫ 10 3 ) were seeded into a glass-bottom 14-mm microwell dish (MatTek Corp., Ashland, MA) that was precoated with 2.5 g/cm 2 fibronectin followed by synchronization. Cells were incubated for 90 min after release from RO3306. Cells were fixed in PBS containing 4% (w/v) paraformaldehyde at RT for 10 min. Cells were permeabilized in cytoskeleton stabilization (CSK) buffer (100 mM NaCl, 300 mM sucrose, 3 mM MgCl 2 , 1 mM EGTA, 10 mM PIPES, pH 6.8) containing 0.3% (v/v) Triton X-100 at RT for 1 min. Cells were washed three times in PBS, and nonspecific binding sites were blocked by incubating cells in PBS containing 5% (w/v) BSA at RT for 30 min. Cells were then stained with relevant primary antibody in PBS containing 5% (w/v) BSA at RT for 1 h. The following primary antibodies were used: rabbit anti-␣-tubulin (1:300 dilution, Thermo Fisher Scientific), mouse anti-acetylated tubulin (1:100 dilution; Sigma-Aldrich), rabbit anti-MAD2 (1:300 dilution; Thermo Fisher Scientific), and human anti-centromere antibody (1:300 dilution; Antibodies, Inc., Davis, CA), which was kindly provided by Dr. Soak-Kuan Lai and Dr. Hoi-Yeung Li (Nanyang Technological University). Cells were washed three times before staining with secondary antibody in PBS. The following secondary antibodies were used: Alexa Fluor 647-conjugated goat anti-mouse IgG (1:600 dilution; Invitrogen), Alexa Fluor 594 -conjugated goat antirabbit IgG (1:600 dilution; Invitrogen), and Rhodamine Red-Xconjugated donkey anti-human IgG (1:10,000 dilution, Jackson Immunoresearch Laboratories Inc., West Grove, PA), which was kindly provided by Dr. Soak-Kuan Lai and Dr. Hoi-Yeung Li. DNA was stained with DAPI (0.1 mg/ml). Images of cells were acquired on a confocal laser-scanning microscope (Zeiss LSM710, Carl Zeiss, Oberkochen, Germany) with a ϫ63 oil objective lens. Data were analyzed using the Zen 2011 software (Carl Zeiss). The intensity ratio of acetylated to total ␣-tubulin was determined using ImageJ software. To improve clarity in the figure presentations, the overall brightness and contrast of entire images were adjusted.

HDAC6 activity assay
HDAC6 activity in G 2 /M synchronized cells was measured using a fluorometry-based detection kit according (BioVison, Inc., Milpitas, CA) according to the manufacturer's instructions.

Xenograft tumor formation and cryosectioning
Animal studies were approved by the Institutional Animal Care and Use Committee at Nanyang Technological University. SH-SY5Y 5HRE-GFP cells (5 ϫ 10 6 ) in 100 l of Matrigel (growth factor-reduced and phenol red-free) (Corning) and PBS mixture were subcutaneously injected into the right flank of immune-compromised BALB/c-Rag-IL2R␥ null mice (kindly provided by Dr. Klaus Erik Karjalainen and Dr. Christiane Ruedl, School of Biological Sciences, Nanyang Technological University). After 24 days, mice were sacrificed, and the tumors were excised. Each tumor was rinsed in PBS before immersing into PBS containing 4% (w/v) paraformaldehyde and incubated at 4°C overnight. The tumor was then immersed in PBS containing 15% (w/v) sucrose and incubated at 4°C until the tumor settled at the bottom of the container. The step was repeated in PBS containing 30% (w/v) sucrose. The tumor was embedded in OCT (Baxter, Deerfield, IL) and frozen in liquid nitrogen. The tumor was sectioned at 8 m/slide on a CM1950 cryostat (Leica, Wetzlar, Germany). The tumor tissue slide was then fixed in PBS containing 4% (w/v) paraformaldehyde at RT for 10 min followed by permeabilization in CSK buffer containing 0.3% (v/v) Triton X-100 at RT for 1 min. The tumor tissue slide was washed three times in PBS followed by incubation in PBS containing 1% (w/v) BSA at RT for 30 min. Tumor tissue was then stained with rabbit anti-kindlin-2 (1:100 dilution; Sigma-Aldrich) primary antibody in PBS containing 1% (w/v) BSA at RT for 1 h. Tumor tissue slide was washed three times in PBS before staining with Alexa Fluor 635 conjugated goat antirabbit IgG (1:600 dilution; Invitrogen), Alexa Fluor 594-phalloidin (0.27 ng/ml) and DAPI (0.1 mg/ml) in PBS. Images were acquired on a confocal laser-scanning microscope (Zeiss LSM710) (Carl Zeiss) with a ϫ10 and ϫ63 (oil) objective lens. Data were analyzed using the Zen 2011 software (Carl Zeiss).

Statistical calculations
GraphPad Prism 5 software (GraphPad Software, San Diego, CA) was used to perform statistical calculations. p Ͻ 0.05 is considered significant. The statistical test used in each analysis is described in the figure legends.

Data availability
All data are contained within the article.