X-linked inhibitor of apoptosis protein (XIAP) is a client of heat shock protein 70 (Hsp70) and a biomarker of its inhibition

Heat shock protein 70 (Hsp70) and Hsp90 are molecular chaperones that play essential roles in tumor growth by stabilizing pro-survival client proteins. However, although the development of Hsp90 inhibitors has benefited from the identification of clients, such as Raf-1 proto-oncogene, Ser/Thr kinase (RAF1), that are particularly dependent on this chaperone, no equivalent clients for Hsp70 have been reported. Using chemical probes and MDA-MB-231 breast cancer cells, we found here that the inhibitors of apoptosis proteins, including c-IAP1 and X-linked inhibitor of apoptosis protein (XIAP), are obligate Hsp70 clients that are rapidly (within ∼3–12 h) lost after inhibition of Hsp70 but not of Hsp90. Mutagenesis and pulldown experiments revealed multiple Hsp70-binding sites on XIAP, suggesting that it is a direct, physical Hsp70 client. Interestingly, this interaction was unusually tight (∼260 nm) for an Hsp70–client interaction and involved non-canonical regions of the chaperone. Finally, we also found that Hsp70 inhibitor treatments caused loss of c-IAP1 and XIAP in multiple cancer cell lines and in tumor xenografts, but not in healthy cells. These results are expected to significantly accelerate Hsp70 drug discovery by providing XIAP as a pharmacodynamic biomarker. More broadly, our findings further suggest that Hsp70 and Hsp90 have partially non-overlapping sets of obligate protein clients in cancer cells.

In cancer, high levels of the chaperones Hsp70 and Hsp90 are associated with poor prognosis and resistance to chemotherapeutics (1)(2)(3)(4). These chaperones are not oncogenes themselves; rather, they are thought to create a permissive environment by binding to multiple proteins and stabilizing them (5), resulting in "non-oncogene addiction" (6,7). Indeed, a hallmark of Hsp90 inhibition is that a subset of these proteins (sometimes termed "clients") becomes unstable and is degraded after treatment (8,9). This response is so robust that loss of specific clients, such as Akt1 or Raf-1, is routinely used as a surrogate biomarker of Hsp90 target engagement (10). Indeed, medicinal chemistry campaigns often take advantage of this property to help guide the optimization of Hsp90 inhibitors (11,12). In other words, the relative ability of a molecule to reduce the levels of Hsp90 clients is used to guide structure-activity relationships. However, no equivalent client for Hsp70 has yet been reported. Rather, Hsp90 clients are often used to estimate activity on Hsp70, making it unclear whether the compound acts strictly through that chaperone.
Hsp70 is composed of an N-terminal nucleotide-binding domain (NBD), 3 a substrate-binding domain (SBD), and C-terminal disordered region (13). In its ATP-bound state, Hsp70 has a poor affinity for clients, but upon ATP hydrolysis, the SBD adopts a tight-binding conformation (14,15). In mammals, there are two major Hsp70 family members in the cytosol: Hsc70 (HSPA8) and Hsp70 (HSPA1A). For both proteins, cycling between the tight-and weak-binding states is further regulated by co-chaperones, such as J proteins (also called Hsp40s), that increase the rate of nucleotide hydrolysis and nucleotide exchange factors that promote release of ADP (16,17). Clients of Hsp70s contain short stretches of non-polar amino acids within extended polypeptides (18 -20) and hydrophobic "patches" within partially folded conformations (21)(22)(23). These client interactions often involve a hydrophobic cleft in the ␤-basket subdomain of the SBD (24). The physical features of Hsp70 clients are likely to be found in many unfolded and metastable proteins (25)(26)(27), such that the theoretical client pool is large.
Although Hsp70 and Hsp90 often work together to stabilize shared clients (28), such as steroid hormone receptors (29), the two chaperones have no sequence or structural similarity.
Hsp90 is an obligate dimer that uses its N-terminal and middle domains to interact with partially folded and near-native states (30). Hsp90 has been reported to bind a more restricted set of ϳ200 kinases and transcription factors (5), but it also binds to some intrinsically disordered proteins, including Tau (31,32). Much of what we know about how Hsp70 and Hsp90 bind to clients comes from pioneering work on glucocorticoid receptor (29). In that system, Hsp70 uses its SBD to bind to newly synthesized glucocorticoid receptor during primary folding, whereas Hsp90 is recruited to bind the later, partially folded states. Because of these differences in chaperone structure and binding preference, it seems possible that some clients might be more dependent on one chaperone than the other for stability.
Recent work has yielded first-generation Hsp70 inhibitors, which have proven to be useful chemical probes (16). For example, analogs of MKT-077 (including JG-98) stabilize the ADPbound state of Hsc70 and Hsp70 (33) and have anti-proliferative activity in cell-based and animal models of breast cancer (34,35). Likewise, pifithrin-(PES) and its analogs bind to the SBD of Hsp70 and have anti-proliferative activity in multiple cancer cells (36 -38). In both cases, the compounds have been found to be relatively selective for Hsp70 based on pulldown studies and genetics (39,40). Moreover, both JG-98 and PES are less toxic to normal, non-transformed cells, consistent with the "addiction" of cancer cells to Hsp70. Together, these experiments have helped solidify the proposed role of Hsp70 in tumorigenesis (4). However, further development of Hsp70 inhibitors has been limited by a lack of pharmacodynamic biomarkers. In particular, the reported Hsp70 clients, such as Raf-1 (34), are also Hsp90 clients, so it is not clear whether their degradation is due to inhibition of just Hsp70 or a combination of the chaperones.
The availability of Hsp90 inhibitors, such as 17-DMAG, was essential in discovering the clients of Hsp90 (5,41). In that work, the inhibitors were identified first and then used to reveal that many kinases and transcription factors rely on that chaperone. Inspired by this history, we wanted to take advantage of the new Hsp70 inhibitors to learn whether any reported chaperone clients might be relatively more sensitive to Hsp70 versus Hsp90. Based on recent reports (35), we selected a handful of candidate clients and found that the inhibitor of apoptosis proteins (IAPs), such as c-IAP1 and XIAP, appear to be obligate clients of Hsp70 that are less reliant on Hsp90. This was a particularly interesting result because IAPs are important mediators of cell survival signaling and are overexpressed in many cancers (42). The IAP family includes c-IAP1, c-IAP2, and XIAP, and it is defined by the inclusion of one or more baculoviral IAP repeat (BIR) domain (43,44). The BIR domains bind to caspases and prevent apoptotic signaling (45) and are bona fide drug targets in their own right. Here, we show that treatment with Hsp70 inhibitors leads to rapid and dramatic loss of the IAPs in MDA-MB-231 breast cancer cells. To understand this relationship in more detail, we explored the interaction between XIAP and Hsp70 in vitro and found that the chaperone binds to multiple sites within the BIR2 and BIR3 domains. Mutagenesis and NMR studies suggested that the interaction is tight and not canonical; it seemed to involve regions outside of Hsp70's expected binding cleft. Together, these results suggest that IAPs are direct clients of Hsp70, and they might be candidate biomarkers of Hsp70.
Moreover, these results illuminate surprising differences between how cancer cells rely on Hsp70 and Hsp90.

Hsp70 inhibition results in rapid degradation of XIAP
Chemical inhibitors of Hsp70 have been reported to enhance turnover of a number of proteins, including IAP-1, XIAP, Raf-1, tau, androgen receptor and others (35,40,46). However, it is not clear whether any of them might be relatively selective for Hsp70 compared with Hsp90. To address this question, we first examined the levels of XIAP, c-IAP1, and Raf-1 after treatment with either Hsp70 or Hsp90 inhibitors. These studies employed MDA-MB-231 breast cancer cells because growth of these cells had previously been shown to be sensitive to both Hsp70 and Hsp90 inhibitors (34). Moreover, we initially focused on XIAP and c-IAP1, rather than other putative clients, based on serendipitous observations made during recent studies of necroptosis (59). Finally, to provide greater confidence in the results, we used two structurally distinct inhibitors of each chaperone. For Hsp70, we used PES and JG-98, and we used AUY-922 and 17-DMAG as Hsp90 inhibitors (see Fig. 1A). In the first experiments, MTT assays were used to confirm that all of the inhibitors have anti-proliferative activity at the expected EC 50 values (Fig. 1B). These experiments allowed us to use each of the compounds at a concentration that ensured maximal activity (10 M for JG-98, AUY-922, and 17-DMAG and 30 M for PES). Next, we characterized the kinetics of cell death by performing MTT assays at 24, 48, and 72 h after treatment. We found that both JG-98 and PES caused relatively rapid responses, with 50% cell proliferation lost by ϳ10 h and more than 80% by 24 h (Fig.  1C). In contrast, the response to AUY-922 and 17-DMAG took a longer period of time, with 72ϩ h required to reduce growth by 80%. The results of these kinetic experiments were important in our search for Hsp70 clients because we were most interested in those clients that are degraded prior to extensive loss of cell viability. More explicitly, we considered it likely that nonspecific client degradation, triggered by downstream caspase activation and/or other proteolytic events, might be confounding at later times; whereas the bona fide clients should be direct physical interaction partners.
With these criteria in mind, we performed Western blottings of the candidate proteins at 0, 1, 3, 6, 12, and 24 h after treatment. From these experiments, we confirmed (47) that Raf-1 is a selective client of Hsp90 (Fig. 1D). Specifically, the levels of Raf-1 were significantly (Ͼ95%) decreased by 12 h after treatment with either 17-DMAG or AUY-922. In contrast, treatment with these inhibitors tended to have a less pronounced effect on cIAP-1 and XIAP levels. In contrast, treatment with either of the Hsp70 inhibitors JG-98 or PES caused a dramatic loss of cIAP-1 and XIAP at relatively early time points, although Raf-1 was largely spared (Fig. 1D). This effect was particularly strong in the first 6 h after treatment, when c-IAP1 and XIAP levels were reduced ϳ75% and Raf-1 levels were only reduced ϳ20%. Thus, XIAP and c-IAP1 seemed to be relatively selective clients of Hsp70 but not Hsp90.
To better understand the mechanism of IAP loss in response to Hsp70 inhibitors, we first tested whether activation of Hsp70 stabilizes IAPs caspase activity might be involved by blotting the JG-98-treated MDA-MB-231 cell lysates for cleaved caspase-3 and poly(ADPribose) polymerase (PARP). Interestingly, cleaved caspase-3 and PARP only emerged 12-24 h after treatment (Fig. 1D), which is substantially after the levels of XIAP and c-IAP1 were reduced. This finding is consistent with a recent result, in which removal of the RING domain of XIAP was found to block turn-over by JG-98 (59). Thus, inhibition of Hsp70 seems to initiate the normal turnover of IAPs rather than triggering loss through caspase activation.

Hsp70 and XIAP form a complex in vitro and in cells
We next wanted to determine whether Hsp70 physically interacts with the IAPs. In these studies, we focused on XIAP

Hsp70 stabilizes IAPs
( Fig. 2A) as a representative member of the IAP family because of a wealth of available structural data (44). Using co-immunoprecipitations, we found that endogenous XIAP was bound to Hsp70 in MDA-MB-231 cell lysate (Fig. 2B). To understand whether this interaction might be direct, we first used a computational algorithm to predict possible Hsp70-binding sites in full-length human XIAP (18). This approach suggested that XIAP contains nine putative binding sites, all but two of which are located within the BIR2 and BIR3 domains (see Fig. 2A). To test this prediction, the region of XIAP(120 -356) corresponding to the BIR2 and BIR3 domains was purified, and its binding to recombinant human Hsp70 was measured by ELISA. We found that Hsp70 bound XIAP(120 -356) with a K D ϭ 260 Ϯ 20 nM (Fig. 2C). Consistent with binding in the BIR2 and BIR3 domains, addition of a Smac mimetic (48) blocked interactions with Hsp70 (Fig. 2D). Also, addition of JG-98 (10 M) weakened the apparent affinity of Hsp70 for XIAP(120 -356) (see Fig. 2D).
Thus, XIAP appears to be a direct client of Hsp70 in vitro and in cells, and its interaction with Hsp70 is mediated by the BIR2/3 domains.
We next wondered why Hsp90 inhibitors might have a less pronounced effect on XIAP stability. One possibility is that Hsp90 might not bind to this client. Hsp90 and Hsp70 are known to bind similar or even overlapping sites in other clients. To understand whether this might be the case for XIAP, we tested the ability of recombinant human Hsp90␣ to compete
To further explore the interaction of XIAP with Hsp70, we created point mutations to disrupt each of the seven putative binding sites in XIAP(120 -356) (Fig. 3A). Other than the possible exception of L231S, each of the mutants was folded to the same extent as wildtype, as judged by circular dichroism (CD) (Fig. 3B). In the ELISA platform, we found that Y190E, L207S, L307S, and L331S all had weaker affinity for Hsp70, with dissociation constants ranging from 2-fold to greater than 10-fold higher than the wildtype protein (Fig. 3C). To test whether this difference in affinity would translate into resistance to Hsp70 inhibitors, we expressed full-length XIAP (XIAP WT , XIAP Y190E , and XIAP L207S ) in HeLa cells and measured protein stability after treatment with JG-98. Consistent with the model, we found that XIAP Y190E was difficult to overexpress, likely because of reduced Hsp70 binding, and that its levels were not sensitive to JG-98 (Fig. 3D). The other mutant, XIAP L207S was not as dramatically affected, suggesting that some sites might be more functionally important for stability than others.
The ELISA results suggested that Hsp70 may bind to multiple sites on XIAP(120 -356) because none of the individual mutations were sufficient to completely block the interaction. To examine this possibility, we incubated Hsp70 with XIAP(120 -356) and analyzed the complex by size-exclusion chromatography and multiangle light scattering (SEC-MALS). Consistent with the other binding studies, we found that Hsp70 and XIAP(120 -356) formed a stable, multimeric complex (Fig.  3E). Moreover, based on the apparent molecular mass of ϳ240 kDa, we estimate that approximately three Hsp70s may be bound.

Interaction with XIAP involves non-canonical interactions with Hsp70
Next, we asked whether XIAP(120 -356) binds to Hsp70 in the chaperone's canonical SBD. As a first test, we titrated a known SBD ligand, the NRLLLTG peptide, into the ELISA experiment. Surprisingly, we found that NRLLLTG was not able to compete with XIAP(120 -356) for binding to Hsp70 (Fig. 4A), suggesting that the XIAP(120 -356) was not binding the canonical cleft. A growing number of non-canonical interactions with Hsp70 have been discovered (49,50), and these interactions often do not fit the normal profile for nucleotide dependence. In other words, the interaction is not always tighter in the ADP-bound state (20). Similarly, we found that the affinity of XIAP(120 -356) for Hsp70 did not follow the canonical behavior; its affinity for ADP-bound or apo-Hsp70 was 1.5-fold worse than in the ATP-bound state Hsp70 (Fig.

Hsp70 stabilizes IAPs
4B). This result was consistent with our observation that the apparent affinity of Hsp70 for XIAP(120 -356) was slightly weakened, not strengthened, by addition of JG-98 (see Fig. 2D), as this compound is known to stabilize the ADP-bound state (51). Some non-canonical interactions with Hsp70 are reported to involve contacts outside the SBD (49). To probe whether XIAP(120 -356) might share this feature, we purified truncated mutants of Hsc70 that are composed of either the NBD (Hsc70 NBD ) or SBD (Hsc70 SBD ) alone and tested them for the ability to compete with full-length Hsp70 for binding to XIAP(120 - 356). Unlike what would be expected for a strictly canonical interaction, we found that both Hsc70 NBD and Hsc70 SBD competed with full-length Hsp70 for binding to XIAP(120 -356) (Fig. 4C), although the Hsc70 NBD interaction seemed to be significantly weaker. Finally, we titrated NRLLLTG or XIAP(120 -356) into samples of 15 N-labeled Hsc70 SBD (residues 395-508, lacking the C-terminal subdomain). This NMR-based assay has previously (50) been used to measure binding of canonical clients, such as NRLLLTG, to the hydrophobic cleft, and it is based on the appearance of characteristic cross-peaks upon "rigidification" of the ␤-subdomain. Consistent with previous observations, we found that adding NRLLLTG to Hsc70 SBD caused the appearance of ϳ 15 new cross-peaks (Fig. 4D). Because the peptide is not labeled, these cross-peaks are likely due to "rigidification" of the SBD after binding to the canonical hydrophobic cleft. Consistent with the ELISA studies, XIAP(120 -356) failed to produce these characteristic peaks, again suggesting that it is a non-canonical client. Together, these results all suggest that XIAP makes non-canonical contacts with Hsp70 and that the NBD is involved.

Stability of IAPs is responsive to inhibitors of Hsp70 but not Hsp90
Our experiments in MDA-MB-231 cells showed that Hsp70 inhibitors induce degradation of XIAP and c-IAP1. Thus, we envisioned that IAPs might be useful surrogates for Hsp70 target engagement, similar to how Raf-1 is used to monitor pharmacological inhibition of Hsp90 (11). However, we wanted to test this relationship in additional cell lines. Similar to what we found in MDA-MB-231 cells, we found that the IAPs in HeLa and MCF7 cells were selectively degraded in response to JG-98, but not 17-DMAG, at early time points (ϳ6 h) (Fig. 5A). Next, we wanted to test whether IAP stability was sensitive to Hsp70 inhibition in normal healthy cells. This experiment was impor-

Hsp70 stabilizes IAPs
tant because clients often depend on chaperone only in a cancer cell-specific context (4). Consistent with this idea, JG-98 did not reduce levels of XIAP or c-IAP1 in IMR90 normal human lung fibroblasts (Fig. 5B). Finally, we wanted to understand whether IAPs might be surrogate biomarkers of Hsp70 engagement in vivo. Accordingly, we established MCF7 tumor xenografts in mice and treated with JG-98, using a previously reported dosing scheme (35). In this model, daily i.p. injections of JG-98 (7 mg/kg/day) begin to show a therapeutic effect on tumor volume 4 -6 days after the initiation of treatment. When we isolated tumors from these treated animals, we found that the lysates had significantly lower levels of XIAP and c-IAP1 (Fig. 5C) when compared with those from mock-treated animals. Together, these results suggest that XIAP and c-IAP1 are candidate biomarkers of Hsp70 in cells and animals.

Discussion
There is growing interest in Hsp70 inhibitors as potential anti-cancer therapeutics (3,52,53), and the development of such molecules would benefit from the identification and characterization of selective biomarkers. Particularly sensitive Hsp70 clients could be used as surrogates for target engagement in cancer cells, facilitating hit-to-lead medicinal chemistry campaigns. In addition, such clients could potentially be used in animal models or in the clinic to develop relationships between pharmacokinetics and pharmacodynamics. Importantly, the ideal client should be more sensitive to inhibition of Hsp70 than Hsp90 to allow for the rapid discrimination of their relative contributions. Here, we found that degradation of c-IAP1 and XIAP occurs rapidly (ϳ6 h) and dramatically (ϳ50 -80%) after treatment with Hsp70 inhibitors in MDA-MB-231 cells. Furthermore, the levels of these proteins did not seem to be responsive to Hsp90 inhibitors; for example, treatment with AUY-922 only reduced XIAP levels by ϳ25% in 24 h. Despite the fact that Hsp90 inhibitors had little effect on IAP levels, this chaperone did seem to bind relatively tightly to XIAP in vitro and to share Hsp70-binding sites. These chaperones are known to work together in the folding and stability of many

Hsp70 stabilizes IAPs
clients (29), but to our knowledge, no clients have been experimentally shown to be clearly reliant on Hsp70 but not Hsp90. Future work will be directed at better understanding the structural differences between how Hsp90 and Hsp70 interact with XIAP to reveal the molecular mechanisms that make this client different from previously studied ones. This question is also motivated by the fact that IAPs themselves are validated targets for cancer therapy (54), so ways of reducing their levels might be directly beneficial in cancer.
Why does treatment with an Hsp70 inhibitor lead to a reduction in the levels of XIAP? There are still many important mechanistic details to uncover, but it seems that treatment with JG-98 weakens the affinity of Hsp70 for XIAP by more than 5-fold in vitro. Furthermore, it is known that the RING domains are important for normal XIAP turnover and that the RING domain is critical for turnover in response to JG-98. Thus, we speculate that Hsp70 might contribute to keeping XIAP in an inactive state, such that release would favor auto-ubiquitination and degradation, possibly correlating with changes in oligomerization. Because Hsp70 inhibitors of different chemical classes bind different locations and "trap" different nucleotide states, it will be interesting to use a broader panel of probes to understand the relationships between client affinity and turnover in the XIAP system.
A major goal of this work was to characterize the proteinprotein interaction (PPI) between putative clients and Hsp70. We considered it important to understand the physical contacts Hsp70 makes with possible biomarkers, similar to the detailed information available for binding of other obligate clients to Hsp90 (55). This is important because many putative clients are expected to be indirectly degraded during cell death, and such clients might therefore not be the best choice as biomarkers. Accordingly, we focused on the potential interaction between XIAP and Hsp70 and confirmed that these two proteins bind to each other in vitro and in cells. We were initially surprised to find that the measured affinity of Hsp70 for XIAP was relatively tight (ϳ230 nM), compared with the affinity for many other clients (typically low to mid-micromolar). This tight affinity may be due, in part, to avidity effects from binding to multiple sites. Mutagenesis and SEC-MALS studies suggested that as many as three Hsp70s may bind to XIAP(120 -356). Multiple Hsp70 molecules have also been shown to bind other clients, including denatured rhodanese (21,56), so multivalency seems to underlie the interaction with a subset of clients. Second, the relatively tight affinity of the XIAP-Hsp70 complex might be a product of the non-canonical nature of the PPI. We found that binding to XIAP does not follow the normal nucleotide dependence and that it involves at least one additional contact within the Hsp70's NBD. Furthermore, the NMR titrations clearly showed that the NRLLLTG peptide, but not XIAP(120 -356), binds the canonical hydrophobic cleft of the SBD. Currently, the Hsp70 NBD is not fully assigned, so the specific binding site of XIAP(120 -356) is not clear. However, there are intriguing chemical shift perturbations in the XIAP(120 -356) treated samples (see Fig. 4D) that might lead to additional insights. We speculate that XIAP could be a good model for understanding the location of non-canonical contacts. In addition, although the exact site(s) of the non-canoni-cal interaction(s) are not known, recent work has shown that phosphoserines bind to the ␣-helical lid of Hsp70 (49), so we posit that secondary contacts here may contribute to XIAP binding. More generally, our results suggest that IAPs might serve as good models for probing this category of poorly understood Hsp70 interactions. We speculate that non-canonical interactions may make a major, unresolved contribution to Hsp70-mediated quality control and that new model systems will have a substantial impact on our knowledge of the system.

Protein expression and purification
All His-tagged Hsp70 proteins (HSPA1A, HSPA8, Hsc70 NBD (1-383), and Hsc70 SBD (395-509)) were purified as described previously (57) using batch purification with Ni-NTA resin (Novagen) and subsequent cleavage of the His tag with tobacco etch virus protease. Hsp70, Hsc70, and Hsc70 NBD were further purified using an ATP column, whereas Hsc70 SBD underwent gel-filtration chromatography on a Superdex 75 16/60 column (GE Healthcare). Hsc70 (HSPA8) domain truncations were used in Fig. 4, A and C, because we have found that they are more stable than the corresponding Hsp70 (HSPA1A) domains. WT His-tagged XIAP(120 -356) and its mutants were batch-purified with Ni-NTA resin and eluted with 400 mM imidazole. DTT was added to 10 mM, and XIAP(120 -356) was further purified by gel-filtration chromatography on a Superdex 75 16/60 column (GE Healthcare) in 20 mM Tris buffer containing 200 mM NaCl, 50 M zinc acetate, and 1 mM DTT, pH 7.5. Fractions containing XIAP(120 -356), as assessed by SDS-PAGE, were pooled and concentrated, and DTT was Hsp70 stabilizes IAPs added to 10 mM before storing at Ϫ80°C. The bicinchoninic acid (BCA) assay kit (Thermo Fisher Scientific, Inc.) was used to determined protein concentration, and protein purities were estimated at Ͼ90% by SDS-PAGE and Q-TOF LC-MS (Agilent).

Tissue culture, viability assays, and transfections
MCF-7 and HeLa cells (ATCC) were maintained in DMEM (Invitrogen) supplemented with 10% fetal bovine serum and 1% penicillin/streptomycin. MDA-MB-231 cells (ATCC) were maintained in DMEM supplemented with 10% fetal bovine serum, 1% penicillin/streptomycin, and non-essential amino acids. Cells were used at low passage, typically less than 20. No further validation of the cell lines was performed. If indicated, cell viability was determined using the MTT assay as described previously (34). XIAP pCMV6 plasmids were transfected using Lipofectamine 2000 (Invitrogen) according to the manufacturer's instructions.

Western blotting
Cell extracts were prepared in chilled RIPA buffer (50 mM Tris, pH 8, 150 mM NaCl, 1% Triton X-100, 0.5% sodium deoxycholate, 0.1% SDS) unless otherwise indicated. Protein concentration was determined by the BCA assay, and 20 g of total protein was separated by SDS-PAGE on 10% Mini-PROTEAN TGX gel (Bio-Rad) and transferred to PVDF membrane (Thermo Fisher Scientific Inc). Membranes were blocked with 5% milk in TBS, 0.05% Tween for 1 h at room temperature, incubated with primary antibodies overnight at 4°C, washed with TBS, 0.05% Tween, and incubated with the appropriate horseradish peroxidase-conjugated secondary antibody for 1 h at room temperature. Membranes were developed using chemiluminescence (ECL Prime, GE Healthcare). Xenograft experiments were carried out as described (35,58).

Co-immunoprecipitation
Cell extracts were prepared in chilled lysis buffer (50 mM Tris, pH 8, 150 mM NaCl, 1 mM ATP, 10 mM KCl, 5 mM Mg(OAc) 2 , 1% Nonidet P-40) supplemented with protease inhibitor mixture (Roche Applied Science). The total protein concentration was adjusted to 5 mg of protein in 1 ml of cell extract. PureProteome protein G magnetic beads (Millipore) were incubated with 6 g of the appropriate antibody or nonspecific mouse IgG (Santa Cruz Biotechnology) for 30 min at room temperature with mixing, followed by antibody crosslinking with bis(sulfosuccinimidyl) suberate (Thermo Fisher Scientific, Inc.) for 1 h at room temperature with mixing. The cross-linking reaction was quenched with 1 M Tris, pH 7.5, for 1 h at room temperature with mixing. Meanwhile, equal 100-l samples of cell lysate were pre-cleared by incubation with 50 l of protein G beads for 1 h at room temperature with mixing. Protein complexes were immunoprecipitated by incubation of the pre-cleared lysate (1 mg of total protein per immunoprecipitation) with 50 l of antibody-cross-linked protein G beads for 1 h at room temperature with mixing. The immunocomplexes were washed three times with 500 l of wash buffer (PBS, pH 7.4, 0.1% Tween 20) and eluted with 0.1 M glycine, pH 2.6. Proteins were visualized by Western blotting.

ELISA
XIAP(120 -356) was non-covalently immobilized in the wells of a clear, flat-bottom 96-well plate (Thermo Fisher Scientific, Inc.) by incubating 100 l of 100 nM XIAP(120 -356) in immobilization buffer (20 mM MES, pH 5.2) overnight at 37°C. XIAP(120 -356) was removed from the wells, and the wells were washed with 3ϫ 150 l of TBS supplemented with 0.05% Tween (TBS-T). Each wash was incubated, with gentle rocking, for 3 min at room temperature. Following washing, 30 l of Hsp70 was added at the indicated concentrations in binding buffer (25 mM HEPES, pH 7.4, 40 mM KCl, 8 mM MgCl 2 , 100 mM NaCl, 0.01% Tween), supplemented with 1 mM nucleotide and 1 mM DTT. The plates were incubated at room temperature for 24 h with gentle rocking. Solutions of Hsp70 were removed, and each well was washed as before and blocked with 100 l of 5% milk in TBS-T for 5 min at room temperature. The plates were developed using 50 l each of Hsp70 primary antibody (1:5000 in TBS-T) and an HRP-conjugated secondary antibody (1:5000 in TBS-T) and washed with TBS-T between each 1-h incubation at room temperature. Binding was detected using the TMB substrate kit (Cell Signaling Technology), and the absorbance was read at 450 nm. Data were analyzed using GraphPad Prism software and fit to the Langmuir binding isotherm (Y ϭ B max X/ [K D ϩ X]).

Size-exclusion chromatography and multiangle light scattering
XIAP(120 -356) and Hsp70 were subjected to size-exclusion chromatography using a KW-803 silica resin column (Shodex Group) with an AKTA micro-HPLC (GE Healthcare) in 20 mM Tris buffer, pH 7.5, containing 200 mM NaCl, 5 mM MgCl 2 , 10 mM KCl, 50 M zinc acetate, 1 mM DTT. Separated samples were then analyzed for light scattering using a DAWN HELEOS II MALS detector and for protein concentration using the change in refractive index measured by a Optilab rEX. Molecular weights of species contained in the SEC peaks were calculated using ASTRA VI software (Wyatt Technology).

Circular dichroism
CD spectra of XIAP(120 -356) and mutants were acquired on a J-715 spectropolarimeter (Jasco Inc.) using a 1 mM pathlength quartz cuvette, subtracting the CD signal acquired for buffer alone (10 mM sodium phosphate, pH 7.6, 100 mM NaF, 50 M zinc acetate, 0.5 mM DTT). Data were converted to mean residue ellipticity (degrees cm Ϫ1 dmol Ϫ1 ) according to the equation ⌰ ϭ ⌿/(1000 nlc), where ⌿ is the CD signal in degrees; n is the number of amides; l is the path length in centimeters; and c is the concentration in decimoles/ml. Each spectrum reported is the average of six scans.