Regulation of microtubule dynamics by DIAPH3 influences amoeboid tumor cell mechanics and sensitivity to taxanes

Taxanes are widely employed chemotherapies for patients with metastatic prostate and breast cancer. Here, we show that loss of Diaphanous-related formin-3 (DIAPH3), frequently associated with metastatic breast and prostate cancers, correlates with increased sensitivity to taxanes. DIAPH3 interacted with microtubules (MT), and its loss altered several parameters of MT dynamics as well as decreased polarized force generation, contractility, and response to substrate stiffness. Silencing of DIAPH3 increased the cytotoxic response to taxanes in prostate and breast cancer cell lines. Analysis of drug activity for tubulin-targeted agents in the NCI-60 cell line panel revealed a uniform positive correlation between reduced DIAPH3 expression and drug sensitivity. Low DIAPH3 expression correlated with improved relapse-free survival in breast cancer patients treated with chemotherapeutic regimens containing taxanes. Our results suggest that inhibition of MT stability arising from DIAPH3 downregulation enhances susceptibility to MT poisons, and that the DIAPH3 network potentially reports taxane sensitivity in human tumors.

Metastatic dissemination is a multistep process that involves cell migration, invasion and growth at distant sites. The 'amoeboid' phenotype has emerged as a migratory mechanism that facilitates metastasis 1,2 . Amoeboid behavior is prominent at the invasive front of tumors 3,4 , confers rapid migration rates 1,2,5,6 , and enables survival within the vasculature 7 . Collectively, these malignant features suggest that amoeboid cells are highly aggressive tumor cell variants that potentially evolve subsequent to an epithelial-to-mesenchymal transition (EMT 8 ).

Results
DIAPH3 loss correlates with shorter overall survival. We previously reported an increased frequency of DIAPH3 loss in patients with metastatic disease 5,19 . DIAPH3 silencing in preclinical models promoted amoeboid features, migration and invasion, and experimental metastasis 5,19 . Conversely, enforced expression suppressed amoeboid characteristics and promoted mesenchymal features, including upregulation of N-cadherin, actin stress fibers, and lamellipodia, suggesting that DIAPH3 is a node capable of regulating the transition between amoeboid and mesenchymal phenotypes.
Consistent with this notion, DIAPH3 loss suppressed EMT-like features. DIAPH3 silencing attenuated expression of N-cadherin ( Supplementary Fig. S1A), E-cadherin (Supplementary Fig. S1B-C), and β -catenin ( Supplementary Fig. S1B,D). Loss of E-cadherin is observed during the 'cadherin switch' of EMT 8 . However, loss of this epithelial marker is classically accompanied by upregulation of the ectopic mesenchymal marker N-cadherin 8 . That expression of both N-cadherin and E-cadherin are reduced by DIAPH3 loss implies that transition to an amoeboid phenotype can occur after cells have progressed through an EMT. Consistent with previous reports 5, 19 , these findings suggest that DIAPH3 silencing promotes features of heightened tumor cell aggressiveness. Analysis of DIAPH3 expression in human cohorts supports this hypothesis. PCa patients 36 with 'low' intratumoral DIAPH3 exhibited significantly shorter OS times as compared to those with 'high' expression ( Fig. 1A). Similarly, worsened survival rates were detected in a cohort of glioblastoma patients (Fig. 1B, 37 ) with 'low' DIAPH3 expression. These observations suggest DIAPH3 loss may be of clinical significance and relevant to patient prognosis.

DIAPH3 loss decreases MT stability and alters global MT topology. DIAPH3 loss can perturb
the MT cytoskeleton 5 . To better understand the significance of this association, we assessed the effect of DIAPH3 loss on MT stability, using Ac-tubulin as a marker of stable MT 21,22 . DIAPH3 deficiency induced shortened acetylated MT in DU145 and LNCaP PCa cells, and in HRAS-transformed HMEC cells. This effect was accompanied by reduced Ac-tubulin levels ( Fig. 2A-C, Supplementary Fig. S2A-D, 5 ). Stable MT reformation following cold-induced depolymerization 38 was also attenuated by DIAPH3 silencing (data not shown). Concordantly, enforced expression of GFP-DIAPH3 increased Ac-tubulin levels ( Supplementary Fig. S2E, 5 ). These findings suggest that DIAPH3 loss alters MT architecture and reduces MT stability, thereby implicating DIAPH3 as a MT-stabilizing protein.
Because only a subpopulation of MT are acetylated/stable 21,22 , we also examined the total MT pool. Similarly to observations for Ac-tubulin, MT polymers containing α -tubulin (Fig. 2D), β -tubulin (Fig. 2E), or the isotype tubulin β III (Fig. 2F), were reduced in length in DIAPH3-deficient cells compared to control cells. Quantitation of the longest MT lengths in cells with enforced expression of GFP-tubulin ( Fig. 2G-H) confirmed this reduction. No differences in expression levels of unmodified tubulins were detected by immunoblotting 5 . Taken together, these findings implicate DIAPH3 in regulation of the MT life cycle.

DIAPH3 loss increases MT dynamics.
We next employed live cell imaging of fluorescent tubulin to assess whether reduced MT stability was accompanied by altered MT dynamics. MT length in control DU145 cells was relatively constant over a period of 15 s (Fig. 3A), whereas it changed rapidly in DIAPH3-deficient cells, with faster disappearance (inset), or more rapid elongation, during the same timeframe (Fig. 3A). The maximum change in MT length, quantified over a period of 30 s, was more pronounced in DIAPH3-silenced cells (Fig. 3B). This greater change in MT length was observed when monitoring both emGFP-tubulin (Fig. 3B, Supplementary Movies 1-4) and TagRFP-tubulin. Together, these findings suggest that loss of DIAPH3 increases the fraction of dynamic MT. To our knowledge, this is the first demonstration of increased MT dynamics in cells displaying amoeboid features.
To further characterize amoeboid mechanics, we utilized traction force microscopy (TFM), which monitors the contractile force exerted by a cell on an elastic substrate 39 . Cell traction increased with substrate stiffness in both control and DIAPH3-deficient cells (Fig. 3C). However, these traction forces were reduced, modestly albeit significantly, in DIAPH3-silenced cells (Fig. 3C). DIAPH3 loss also greatly reduced overall contractile strength ( Supplementary Fig. S3A), as detected when computing the cell contractile moment ( Supplementary Fig. S3A), an integrated measure of traction and cell spreading area ( Supplementary Fig. S3B). Following treatment with the MT depolymerizing agent nocodazole, traction forces were enhanced in both control and DIAPH3-silenced cells but to a greater extent in the latter, especially at 11 and 26 kPa stiff substrates. Importantly, tractions were now comparable between control and DIAPH3-silenced cells. Because dynamic MT counterbalance cell contractility 40,41 (Fig. 3D), this lack of difference in the nocodazole condition supports the contribution of dynamic MT in mediating the mechanical effects of DIAPH3 loss.
Next, we characterized traction orientation and polarity. We observed that the traction forces exerted by control cells were polarized and increasingly asymmetric with substrate stiffness (Fig. 3E-G), while they were evenly oriented in cells lacking DIAPH3 (Fig. 3E-F). In the latter, traction polarity was also invariant across substrate stiffness (Fig. 3G). These results suggest that DIAPH3 silencing influences both the extent and polarity of contractile force, as well as the ability of cells to polarize in response to matrix rigidity. These data reveal a novel role for formin proteins in regulating mechanical responsiveness to substrate stiffening. DIAPH3 interacts with MT. Given the findings above implicating DIAPH3 in regulation of MT dynamics, and of the reciprocity between DIAPH3 and MT in regulating cell contractility, we sought to further understand the DIAPH3-MT relationship. We performed an unbiased proteomic analysis to identify DIAPH3-interacting proteins. Initially we used U87 cells, a glioblastoma cell line displaying amoeboid properties ( Supplementary Fig. S4) that can be attenuated by enforced expression of DIAPH3 5 . Lysates from U87 cells stably expressing GFP or GFP-DIAPH3 were immunoprecipitated with antibodies to GFP and subjected to liquid chromatography/tandem mass spectrometry (LC-MS/ MS). Of 417 proteins identified, 130 displayed 2-fold greater binding to GFP-DIAPH3 as compared to GFP alone. Multiple tubulin isotypes and MT-associated proteins preferentially precipitated with DIAPH3 in U87 cells (Fig. 4A). Consistent with this, DIAPH3 bound to α -tubulin during reciprocal co-immunoprecipitation ( Supplementary Fig. S5A), and co-localized with α -tubulin ( Fig. 4B) in DU145 cells. DIAPH3 also co-localized (Fig. 4C) and co-immunoprecipitated ( Supplementary Fig. S5B) with Ac-tubulin. Thus, DIAPH3 appears to associate with MT, including a subpopulation of stable (acetylated) MT.
Because these findings suggest that DIAPH3 promotes MT stability, a process occurring at later stages of the MT lifecycle (after MT polymerization), we next asked whether DIAPH3 preferentially forms complexes with polymerized rather than soluble tubulin. To do this, we assessed the effect of temperature   TagRFP-tubulin visualized by spinning disc confocal microscopy. Note relative persistence of example MT in control cell (inset, top), yet rapid disappearance of example MT in DIAPH3-silenced cell (inset, bottom), indicating increased MT dynamics. B. Quantitation of the maximum MT length change in DU145 cells. MT length was measured over 10 frames (30 s period), and the maximal length change calculated as described in the Methods. The median of the maximum MT length change in control or DIAPH3-silenced cells is shown as a Tukey plot, and was analyzed with a Mann-Whitney test. C. TFM measurements of RMS value of traction in control or DIAPH3-silenced cells, with and without incubation with 2 μ M nocodazole. Note increased contractility in response to nocodazole in both cell lines, but to a greater extent in DIAPH3-deficient cells, especially at 11 and 26 kPa stiff substrates, at which tractions were of comparable magnitude between control and DIAPH3-silenced cells. * = p < 0.0001, Student's t-test. D. Schematic summarizing effects of DIAPH3 silencing or nocodazole on contractile force. Dynamic MT inhibit contractility 40,41 . MT depolymerization with nocodazole disrupts MT, and increases traction (left). DIAPH3 loss, by increasing dynamic MT, is predicted to decrease contractility (right). E. Representative contraction maps, demonstrating the location and magnitude of traction exerted by control and DIAPH3-silenced DU145 cells. Note asymmetry of forces in control cells versus lack of force polarity in DIAPH3-deficient cells. F. Representation of traction orientation and polarity. For each ellipse, representing a single cell, semi-axes are determined by the eigenvalues of the matrix comprised by the first-order moment of the traction (M) and orientation determined by the corresponding eigenvectors of M. Note the greater circular contour (symmetry) of DIAPH3-deficient cells, indicating reduced traction polarity relative to more elliptical (polarized asymmetry of) control cells. G. Traction polarity, obtained from the ratio of the eigenvalues of M in control or DIAPH3-silenced DU145 cells at the indicated substrate stiffnesses. Note increased traction polarity with increasing substrate stiffness in control cells, yet relative insensitivity to substrate stiffness in DIAPH3deficient cells.  heterodimers. This interpretation was confirmed by reciprocal co-immunoprecipitation of DIAPH3 and α -tubulin in DU145 (Fig. 4E) and U87 (Fig. 4F) cells. In both cell types, DIAPH3 also interacted with MT-regulatory and stabilizing proteins (Fig. 4A,D, and Supplementary Fig. 5C-D), some of which were common between the two cell lines. These findings support a preferential association of DIAPH3 with polymerized tubulins.
DIAPH3 silencing increases responsiveness to taxanes and epothilone B. Next we examined how DIAPH3 silencing affects MT responses to MSA. Immunoblotting showed that in the absence of MSA, Ac-tubulin levels were reduced by DIAPH3 silencing (Fig. 5A), consistent with data in Fig. 2. Paclitaxel increased tubulin acetylation in a dose-dependent fashion in both control and DIAPH3-silenced cells. Similar dose-dependent inductions were observed by docetaxel or epothilone B (data not shown). With DIAPH3 knockdown, Ac-tubulin levels at all MSA concentrations remained Bottom, data were quantified after normalization of Ac-tubulin to β -tubulin intensities, followed by ratiometric comparison between MSA-treated and untreated conditions. Note greater fold-change in Ac-tubulin levels in DIAPH3-deficient cells following treatment with MSA relative to baseline (untreated) conditions. Data shown are average ± SD from 3 combined, independent trials. C. Quantitation of Ac-tubulin fluorescence, from at least 25 cells per condition. Note that the reduced Ac-tubulin fluorescence in cells silenced for DIAPH3, relative to controls, persists in the presence of taxol. D. Quantitation of perinuclear Ac-tubulin fluorescence, from at least 25 cells per condition. Note the greater fold-increase in fluorescence by MSA treatment (relative to untreated conditions) in DIAPH3-deficient cells. A (*) indicates p < 0.0001. E. Intracellular accumulation of OG-PTX in cells expressing or silenced for DIAPH3 was monitored spectrophotometrically, and fluorescence normalized to protein concentration in each well. n = 2 independent trials.
Scientific RepoRts | 5:12136 | DOi: 10.1038/srep12136 below those of controls, consistent with higher levels of dynamic MT. Furthermore, the fold-change in Ac-tubulin levels in response to MSA treatment was substantially greater when DIAPH3 was silenced (Fig. 5B). This greater magnitude of change in MT stability is consistent with a greater extent of dynamic MT in DIAPH3-silenced cells. It is also suggestive that cells lacking DIAPH3 are more sensitive to MSA.
Quantifying intracellular Ac-tubulin intensities, using immunofluorescence and 3D computational modeling 42 , yielded analogous results. DIAPH3 silencing reduced basal Ac-tubulin levels relative to those of control cells (Fig. 5C). This reduction persisted during paclitaxel treatment (Fig. 5C). In interphase cells, acetylated MT are predominantly perinuclear 43 (e.g. Fig. 4C). DIAPH3 silencing reduced Ac-tubulin intensity in the perinuclear region (Fig. 5D). Paclitaxel did not appreciably alter this intensity in control cells, but increased perinuclear Ac-tubulin levels in DIAPH3-deficient cells (Fig. 5D). These findings suggest a greater degree of stabilization of MT by MSA when DIAPH3 is silenced.
DIAPH3 depletion also increased intracellular retention of fluorescent Oregon Green 488-paclitaxel (Fig. 5E), suggesting that DIAPH3 loss increases the intracellular concentration of MSA. Consistent with greater activity, survival of DU145 cells in the presence of paclitaxel, docetaxel or epothilone B was significantly reduced by DIAPH3 loss, as evidenced by reduced cell number (Fig. 6A-C). DIAPH3 loss also sensitized androgen-dependent LNCaP PCa cells to taxanes (Fig. 6D), suggesting mechanisms at least partially independent of androgen receptor activity 26,[44][45][46] . Similar results were seen using HMEC-RasV12 breast cancer cells in which DIAPH3 was silenced (Fig. 6E). Low DIAPH3 expression is associated with greater sensitivity to MT-directed chemotherapy drugs. To assess the potential generality of the association between low levels of DIAPH3 and sensitivity to taxanes and other MT-directed drugs, we performed COMPARE analysis using the CellMiner ™ tool 47 . We calculated pair-wise correlations between cytotoxicity Z-scores derived from GI 50 values (the concentration that causes 50% Growth Inhibition) and gene expression Z-Score patterns within the NCI-60 cell line panel. Using NCI Anticancer Drug Screen's Standard Agents Database 47 , we selected agents known to target tubulin, which were then classified as either stabilizing or depolymerizing. Compounds unable to be categorized into these groups were excluded from analysis. As shown in Supplementary Table S1, cells with low DIAPH3 expression uniformly exhibited increased sensitivity to these agents (Pearson's correlation less than − 0.25 in 7 agents). The closely-related genes DIAPH1 and DIAPH2 did not show this relationship; DIAPH1 in fact exhibited an inverse correlation (increased resistance with lower gene expression; Pearson's correlation ≥ 0.25 for 8 MT-targeting agents). DIAPH2 did not show any recognizable sensitivity pattern. When performing the same analysis against 75 known Topoisomerase I-targeting agents, no significant correlations or sensitivity patterns were detected. Taken together, these data suggest that sensitivity to MT-targeted agents from reduced gene expression is unique to DIAPH3 within the Diaphanous formin family.
Low DIAPH3 expression correlates with response to taxane-containing chemotherapy in BCa patients. Given the above findings of in vitro chemosensitivity, we assessed whether DIAPH3 loss was associated with MT-directed chemotherapeutic response in patients. We interrogated gene expression profiles from three randomized clinical trials in BCa patients designed to test taxane-containing regimens in the neoadjuvant setting 29,[33][34][35] . Microarray data were available from 111 BCa patients in trial 1 29 , 91 patients in trial 2 34 , and 310 patients in trial 3 35 , from which outcomes measures including pCR, RFS and OS following taxane-containing chemotherapy were assessed (Supplementary Table S2).
The DLDA-30 signature 33,34 predicted the occurrence of a pCR in patients with low intratumoral DIAPH3 levels (Fig. 7A). Similarly, pCR was predicted to occur with greater frequency in 'DIAPH3 low' cancers ( Fig. 7B). Consistent with this prediction, histologically-confirmed pCR was achieved with greater frequency following neoadjuvant chemotherapy (Fig. 7C, 35 ). Because pCR correlates with RFS 30 , we examined whether DIAPH3 expression correlated with time to progression. Univariate analyses 35 revealed that low levels of DIAPH3 mRNA correlated with extended RFS in cohorts from 2 independent clinical trials (Fig. 7D,E,G). Similarly, in the I-SPY 1 trial 29 , prolonged OS (Fig. 7F,G) was observed in patients with 'DIAPH3 low' cancers treated with T/FAC chemotherapy.
Next we determined whether the negative association between DIAPH3 and clinical outcome following neoadjuvant chemotherapy was greater in triple-negative BCa (TNBC) patients. In two TNBC cohorts 33,34 , pCR predicted by the DLDA-30 signature was greater when DIAPH3 expression was low (Fig. 8A). Similarly, in patients predicted by the DLDA-30 signature to respond to neoadjuvant T/FAC chemotherapy, relative to those predicted not to do so, median DIAPH3 expression was significantly lower (Fig. 8B). When using histologically-confirmed pCR as the outcome, patients with low intratumoral DIAPH3 expression exhibited a greater frequency of response (Fig. 8C,D) following chemotherapy and longer RFS (Fig. 8E). These findings suggest that patients with low intratumor DIAPH3 expression are more responsive to taxane-based chemotherapy regimens.

Discussion
This study has identified a novel link between molecular mechanisms underlying amoeboid behavior, cytoskeletal dynamics and sensitivity to MT-directed chemotherapy. We previously reported that DIAPH3 depletion promotes invasive, amoeboid features in multiple tumor cell backgrounds 5 . Here we present evidence that amoeboid behaviors elicited by DIAPH3 loss likely occur as a result of disruption of the MT cytoskeleton. This conclusion is supported by assessments of MT stability and dynamics; quantitative measurements of polarized force generation, contractility and response to substrate stiffness; and identification of likely members of DIAPH3 protein complexes by LC-MS/MS. DIAPH3 down-regulation also sensitized cancer cells to growth suppression by MT-targeting chemotherapies. To our knowledge, this represents the first therapeutic vulnerability of amoeboid cells to agents routinely used in clinical practice for aggressive disease.
Some tumor cells can interconvert between mesenchymal and amoeboid modes in order to navigate changing environmental contexts. This plasticity provides challenges for therapeutic strategies targeting metastatic tumors. Our study demonstrates that cells with EMT features contain a greater number of stable MT (Figs 2,5), which are less responsive to taxane-induced stabilization, than cell displaying amoeboid characteristics. These cells were also less sensitive to taxane-induced cytotoxicity than amoeboid cells (Fig. 6). Thus, DIAPH3 loss appears not only to induce MAT but also to enhance MT-responsiveness and susceptibility to taxanes, leading to greater drug-induced cell death. These findings are of interest in light of reports that EMT confers taxane resistance 48 . These observations highlight the importance of considering mechanisms of tumor cell plasticity when evaluating chemotherapeutic efficacy in the context of precision medicine.
The androgen receptor (AR) is a transcription factor that plays a key role in PCa pathogenesis, in part through induction of EMT 45 . Several reports provide compelling evidence that MT dynamics control AR intracellular trafficking [44][45][46] . Taxane-induced MT stabilization suppresses AR nuclear translocation, precluding AR promoter occupancy and transcriptional activity, and evoking cell death. In patients with CRPC, reduced AR nuclear localization correlates with patient response to taxane-based chemotherapy 44 . These reports highlight an importance of MT dynamics for AR-dependent prostate tumorigenesis, and may explain the observed clinical benefit of neoadjuvant taxanes prior to radical prostatectomy 49 . These reports also underscore a role for MT stability in taxane responsiveness in cells having undergone EMT; our findings highlight that MT additionally contribute to the responsiveness of amoeboid cells to these agents. We report that both androgen-dependent LNCaP cells and castrate-resistant DU145 cells are sensitized to taxane-induced cytotoxicity by DIAPH3 loss (Fig. 6). Thus, DIAPH3 loss augments the anti-neoplastic effects of taxanes through pleiotropic, and in some cases androgen-independent, mechanisms. These data raise the intriguing possibility that taxanes could be of utility in AR-independent prostate tumors with low DIAPH3. DIAPH3 appears to act as a MT-stabilizing protein, suppressing MT turnover. While the role of the actin cytoskeleton in determining the mechanical properties of cells is widely appreciated, the contribution of MT has been much less studied, especially in amoeboid cells. Using TFM, we observed Fisher's exact test demonstrates increased chemotherapeutic response in patients with low median DIAPH3 expression 34 . C. DIAPH3 levels are lower in cancers with a complete pathologic response (pCR) based on histopathological inspection. Contingency table demonstrating the frequency of clinically-achieved pCRs in patients treated with T/FAC therapy. Cox proportional hazards analysis revealed a significant association of low DIAPH3 expression with therapeutic responsiveness 35 . D-F. Kaplan-Meier curves demonstrating that low DIAPH3 expression is associated with a longer time to recurrence (D, 35 and E, 29 ) and with greater overall survival (F, 29 ) in patients treated with taxanecontaining neoadjuvant chemotherapy. G. Cox proportional hazards regression analyses, derived from datasets used in panels D-F, demonstrating increased hazards ratios (improved response) for recurrence-free and overall survival status in patients with low DIAPH3 expression.
Scientific RepoRts | 5:12136 | DOi: 10.1038/srep12136 that DIAPH3 loss reduced contractile forces generated by DU145 cells, and this reduction of traction was reversed by nocodazole, suggesting that dynamic MT contribute substantially to the magnitude of force generation. This was observed across a range of physiologically relevant stiffnesses, suggesting that reduced traction by amoeboid tumor cells can facilitate movement through diverse tissue spaces. These findings are consistent with reports that fast-migrating keratocytes and neutrophils display lower tractions than those of highly adherent cells 50,51 .
In summary, the results of this study show that DIAPH3 loss can mediate a transition to a more aggressive phenotype, which can arise from cells that have undergone EMT. Ironically, DIAPH3 loss also sensitizes cells to MT-directed chemotherapy. These conclusions are diagramed in Fig. 9. DIAPH3 stabilizes MT, while DIAPH3 downregulation increases MT turnover and dynamics, leading to reduced cell contractility and a more uniform distribution of traction forces. This phenotypic transition makes the tumor cells more sensitive to taxanes and other MT-targeted agents. Johansson et al. suggest DIAPH3 loss to be a candidate driver of BCa 52 . Consistent with this, we find that DIAPH3 loss correlates with worse prognosis in multiple cancer cohorts. The association of low DIAPH3 expression with both metastatic disease and shorter overall survival suggests that amoeboid behavior promotes metastatic dissemination. Our present findings suggest that profiling of tumors for DIAPH3 loss and related network perturbations, including levels of Ac-tubulin, may have predictive value in selecting rational treatment strategies for aggressive disease.

Material and Methods
Cell lines and reagents. Parental and DIAPH3-silenced or over-expressing cell lines, such as DU145, Human Mammary Epithelial (HMEC), or U87, have been described 5 . Paclitaxel (Cayman Chemicals, Sigma), epothilone B (Cayman Chemicals), and docetaxel (LC Laboratories). Cholera toxin B (CTxB, Figure 8. Greater pathologic complete response (pCR) after treatment with neoadjuvant taxanecontaining therapy in TNBC expressing low DIAPH3. A. Prediction of a pCR in 2 cohorts of TNBC patients whose cancers display low DIAPH3 expression 33,34 . B. DIAPH3 expression is lower in TNBC patients predicted to respond to T/FAC therapy than those predicted not to respond 34 . C. Achievement of a histologically-defined pCR in cancers with low DIAPH3 levels 35  Immunofluorescence. Immunostaining was performed as described 5 . Images were acquired on an Identification of DIAPH3 interactomes. U87 cells (1.5 × 10 6 ), stably expressing GFP or GFP-DIAPH3, were lysed in RIPA buffer and immunoprecipitated with anti-GFP antibodies at 4 o C, as above. DU145 cells (1.25 × 10 6 ), stably expressing GFP or GFP-DIAPH3, were immuno-precipitated at 4 o C or 25 o C. Proteins eluted from the beads were separated in a 10% SDS-PAGE gel, and in-gel reduced, alkylated, and tryptically digested 56 . Tryptic peptides were extracted, concentrated, reconstituted in 0.1% formic acid, separated on a 25 cm EASY-Spray C18 column, and analyzed by an LTQ Orbitrap Elite mass spectrometer (Thermo Scientific). After each survey scan, up to 20 collision-induced dissociation (CID) spectra were acquired in the rapid CID scan mode. For protein identification and quantitation, raw mass spectrometric data were searched against the Uniprot_Human database (released on 02/20/14, including 88647 sequences) with MaxQuant (v 1.3.0.5) 57 and Andromeda 58 . False discovery rates for protein and peptide identifications were set at 0.01. Identified proteins were quantitated based on their summed ion intensities. All data have been deposited into ProteomeXchange 59 . The DIAPH3 interactome was analyzed using the DAVID bioinformatics database 60 .
Live cell imaging. DU145 cells were infected with 100 μ l of CellLight emGFP-tubulin or 70 μ l of TagRFP-tubulin (Life Technologies) in DMEM + 10% FBS lacking phenol red. After 1-2 days at 37 o C, Figure 9. Model for the association of DIAPH3 loss with amoeboid motility and taxane sensitivity. Low DIAPH3 expression reduces the extent of MT stability, and in turn increases dynamic MT content. These transitions may lead to either 1) amoeboid behavior and worse prognosis (in untreated patients) or 2) taxane sensitivity and better prognosis (in taxane-treated patients). cells were imaged using a 100 x Apo TIRF NA 1.49 oil objective, on a Nikon Ti inverted confocal microscope coupled to a Yokogawa Spinning Disc head.
Quantitation of MT length. The length of the longest individual MT discernable in a single cell expressing emGFP-tubulin was measured over 10 frames, indicating a 30 s period; 2-3 MT per cell were measured. MT were traced using the freehand line function in ImageJ. The maximum length change was calculated as: [(MT at its longest length -MT at its shortest length) / (MT at its longest length) x 100]. This percentage was determined for 41 MT in control and 35 MT in DIAPH3-silenced DU145 cells.
Tubulin acetylation induced by MSA. Cells were incubated with varying concentrations of paclitaxel for 30 min at 37 o C, or with 500 nM only of paclitaxel, docetaxel, or epothilone B. Where indicated, after immunoblotting proteins were analyzed by densitometry with ImageJ. In brief, Ac-tubulin band intensity was normalized to that of β -tubulin, in both untreated and MSA-treated conditions. The MSA-induced fold-increase in tubulin acetylation was then calculated, as normalized Ac-tubulin intensity in MSA-treated versus untreated conditions. Cytoplasmic and perinuclear tubulin acetylation quantitation. DU145 cells (10 4 ) were incubated with 1 μ M paclitaxel for 8 h, fixed with 0.1 M cacodylate buffer, pH 6.7, containing 3.7% paraformaldehyde, stained with Ac-tubulin and α -tubulin, and counter-stained with DAPI. Serial optical sections were collected using a TCS SP5 X Supercontinuum confocal microscope (Leica Microsystems) and Plan-Apo 63 × 1.4 oil immersion lens. Images from paclitaxel-treated cells were acquired with a gain of 600 V, and untreated cells with 800 V. Ac-tubulin intensities were normalized using a standard curve of intensities acquired at gains of 600, 700, or 800 V and a second order polynomial model y = a2*x^2 + a1*x + a0, in which y represents the output intensity and x the input intensity. To quantify signal intensity in cytoplasmic and perinuclear shells, global cytoplasmic (Ac-tubulin) and nuclear (Dapi image) masks were defined as described 42 . Perinuclear shells were defined as hollow 3-D spherical volumes adjacent to the nucleus, between the nuclear envelope and a fixed distance ~1/4 of the average nuclear radius away. Perinuclear shells and cytoplasm masks were superimposed onto the tubulin image, and for each cell the two quantities averaged.

MSA-induced cytotoxicity.
Cell viability was assessed after treatment with paclitaxel, docetaxel, epothilone B, or DMSO for 4 d by crystal violet. Absorbance was measured at 570 nm.
Patient cohorts and analysis of DIAPH3 in gene expression profiles. The preprocessed and normalized expression profiles for microarray datasets GSE22226 29 , GSE20271 34 , and GSE25055 35 were from the Gene Expression Omnibus (GEO) database, or MD Anderson Cancer Center (http://bioinformatics.mdanderson.org/pubdata.html) 33 . Cohort characteristics are summarized (Supplementary Table S2). Dataset GSE22226 (Agilent G4112A, Trial 1) contained gene expression profiles of a 149-patient cohort, from which 111 patients were selected based on their treatment record of 4 cycles of taxane (paclitaxel or docetaxel) following 4 cycles of doxorubicin/cyclophosphamide (T/AC chemotherapy). Patients treated with Herceptin, or who had not received taxanes after doxorubicin/cyclophosphamide-based therapy, were excluded. Dataset GSE20271 (Affymetrix U133A, Trial 2) contained gene expression profiles of a 178-patient cohort. From this, 91 patients receiving 12 cycles of paclitaxel + 4 cycles of fluorouracil, anthracyclines doxorubicin or epirubicin, and cyclophosphamide (T/FAC chemotherapy), and 87 patients receiving FAC chemotherapy, were selected for analysis. Patients treated with paclitaxel as adjuvant therapy, Xeloda or tamoxifen, or temporal variations in the regimens, were excluded. Dataset GSE25055 (Affymetrix U133A, Trial 3) contained gene expression profiles of a 310-patient cohort, consisting of 227 cases treated with 12 cycles of paclitaxel + 4 cycles of fluorouracil, doxorubicin and cyclophosphamide and 83-cases treated either with 4 cycles of doxorubicin and cyclophosphamide + 4 cycles of paclitaxel (n = 60) or docetaxel (n = 18), or for whom the taxane was not specified (n = 5). Clinical information was extracted from the GEO database.
Additionally, the relationship of DIAPH3 expression to overall survival (OS) was assessed using the Oncomine ® Power Tools Clinical Outcomes Beta v3.1 Library (Life Technologies). Microarray profiling data for the two selected studies 36,37 were exported and segmented into four expression quartiles based on the transcript levels of the DIAPH3 locus. Data were then subdivided into categories of "low" (< 25 th percentile) or "high" (≥ 25 th percentile) DIAPH3 expression, and correlated with clinical outcome as survival curves obtained by Kaplan-Meier analysis. A Log Rank test was used for statistical comparison. Statistical analyses. Normalized DIAPH3 intensities were used to assess the association of DIAPH3 gene expression with OS and/or RFS. Cox proportional hazards regression was utilized when correlating DIAPH3 expression with OS and RFS. Wilcoxon Rank-Sum tests were applied when examining differences in DIAPH3 expression for patients achieving clinically-verified or DLDA30-predicted pCRs, versus those who did/were not. Fisher's exact test was used to evaluate the number of patients predicted to achieve pCR by the DLDA-30 signature. MT length measurements were evaluated with a Mann-Whitney test. For remaining biological and biochemical studies, a two-sided Student's t-test was used.