A single discrete Rab5-binding site in phosphoinositide 3-kinase β is required for tumor cell invasion

Phosphoinositide 3-kinase β (PI3Kβ) is regulated by receptor tyrosine kinases (RTKs), G protein–coupled receptors (GPCRs), and small GTPases such as Rac1 and Rab5. Our lab previously identified two residues (Gln596 and Ile597) in the helical domain of the catalytic subunit (p110β) of PI3Kβ whose mutation disrupts binding to Rab5. To better define the Rab5–p110β interface, we performed alanine-scanning mutagenesis and analyzed Rab5 binding with an in vitro pulldown assay with GST-Rab5GTP. Of the 35 p110β helical domain mutants assayed, 11 disrupted binding to Rab5 without affecting Rac1 binding, basal lipid kinase activity, or Gβγ-stimulated kinase activity. These mutants defined the Rab5-binding interface within p110β as consisting of two perpendicular α-helices in the helical domain that are adjacent to the initially identified Gln596 and Ile597 residues. Analysis of the Rab5–PI3Kβ interaction by hydrogen-deuterium exchange MS identified p110β peptides that overlap with these helices; no interactions were detected between Rab5 and other regions of p110β or p85α. Similarly, the binding of Rab5 to isolated p85α could not be detected, and mutations in the Ras-binding domain (RBD) of p110β had no effect on Rab5 binding. Whereas soluble Rab5 did not affect PI3Kβ activity in vitro, the interaction of these two proteins was critical for chemotaxis, invasion, and gelatin degradation by breast cancer cells. Our results define a single, discrete Rab5-binding site in the p110β helical domain, which may be useful for generating inhibitors to better define the physiological role of Rab5–PI3Kβ coupling in vivo.

Our lab initially identified a region of p110␤ that mediates binding to Rab5 using a conservation-based approach. Two residues were identified (Q596C and I597S) whose mutation disrupts binding to Rab5 (13). The purpose of the present study was to fully define the Rab5-binding interface within p110␤ to understand the physiological role of this interaction. Using a GST-Rab5 in vitro pulldown assay and hydrogen-deuterium exchange MS, we identified a discrete binding site for Rab5 in the helical domain of p110␤. We were unable to replicate previous reports showing direct binding of Rab5 to p85 or to the RBD of p110␤ (14,15). The Rab5-binding interface within p110␤ is restricted to two perpendicular ␣-helices in the helical domain that are located near the G␤␥-binding loop. In vitro kinase assays revealed that soluble Rab5 does not affect PI3K␤ kinase activity. However, replacement of endogenous PI3K␤ with a Rab5 binding-deficient mutant in MDA-MB-231 breast cancer cells inhibited chemotaxis, invasion, and gelatin degradation. Our characterization of the physiologically important Rab5-p110␤ interface will facilitate the development of better tools to study the Rab5-PI3K␤ interaction in cell-based and animal models.

Rab5 binds exclusively to the helical domain of p110␤
To define the Rab5-binding interface within p110␤ (PI3K␤), we first examined whether p110␤ selectively bound to any of the three Rab5 isoforms (A, B, and C), which have been shown to have distinct cellular roles (8,16,17). Using lysates from HEK293T cells expressing wild type (WT) PI3K␤ heterodimer and an in vitro pulldown assay, we were unable to detect any difference in PI3K␤ binding to the three Rab5 isoforms (data not shown). We opted to use Rab5A for the remainder of the study as this isoform was previously used by our lab and by others in studies examining the Rab5-p110␤ interaction (13,15).
HEK293T cells were transfected with p85␣ alone or with either WT p110␤ or the previously reported Rab5-uncoupled p110␤ mutant I597S (13). The lysates from these cells were incubated with nucleotide-loaded Rab5A beads and assessed for binding by immunoblotting. The WT p110␤/p85␣ heterodimer exhibited selective binding to GTP␥S-Rab5A (12fold over GDP-Rab5), whereas the Rab5-uncoupled p110␤ I597S heterodimer failed to bind to either form of Rab5A (Fig.  1B). In contrast to previous reports that Rab5 binds to p85␣ (14), we did not detect binding of Rab5A to p85␣ alone (Fig. 1B). Similarly, we detected no binding to p85␣ in the context of the I597S p110␤/p85␣ heterodimer. These data show that p110␤ is solely responsible for the Rab5A interaction.
We previously identified two residues in the helical domain of p110␤ (Gln 596 and Ile 597 ) whose mutation disrupts binding to GST-Rab5 (13). To more completely map the Rab5-p110␤ interface, we mutated additional residues in the helical domain of p110␤ and evaluated the effect on binding to Rab5A. We chose 35 surface-accessible residues surrounding Gln 596 /Ile 597 . Lysates from transfected cells expressing p85␣ and WT or mutant p110␤ were incubated with nucleotide-loaded GST-Rab5A beads and assessed for binding via immunoblotting (Fig. 1C). For each mutant, we calculated the percentage of the input (lysate) that bound to GTP␥S-Rab5A; values in each experiment were normalized to WT p110␤ binding, which was set to 100%.
The relative binding of the p110␤ mutants to WT p110␤ was stratified into three groups: 0 -33% binding (red), 33-66% binding (blue), and Ͼ66% binding (green). Of the 35 mutated helical domain residues tested, eight showed binding that was less than 33% of WT, and four showed binding that was 33-66% of WT (Figs. 1D and 2). Residues whose mutation significantly inhibited Rab5 binding mapped to two ␣-helices (Asp 509 -Glu 517 and Leu 585 -Ile 597 ), which are located below the G␤␥binding loop (Fig. 1E). The mutagenesis data suggest the primary Rab5 interface localizes to a discrete region within the helical domain of p110␤.

The RBD of p110␤ does not bind Rab5A
A previous study reported that the RBD mediates Rab5 binding to p110␤ (15). However, this study evaluated Rab5 binding using a nonphysiological, truncated iSH2 domain-p110␤ fusion rather than full-length p85␣/p110␤. To reexamine the role of the p110␤ RBD in Rab5 binding in the context of the full-length heterodimer, we mutated the five amino acids in the RBD that were previously assessed for Rab5 binding (15). Using the GST-Rab5A pulldown assay and lysates from cells expressing p85␣ and p110␤ (Fig. 3A), all five p110␤ RBD mutants exhibited binding to Rab5A that was comparable with WT p85␣/p110␤ (Fig. 3, B and C). These data indicate that residues in the p110␤ RBD are not involved in Rab5 binding.

Mutations in p110␤ that disrupt Rab5 binding do not affect binding to Rac1
To ensure that the mutations affecting p110␤ binding to Rab5A did not compromise the overall structure or folding of the p110␤ subunit, we examined the binding of p110␤ mutants to another small GTPase, Rac1, which binds to the p110␤ RBD (6). Previous studies demonstrated that mutation of two residues in the RBD (S211D and K230A) was sufficient to disrupt binding to GTP␥S-Rac1 (6). We compared the binding of p85␣/p110␤ heterodimers to GST-Rac1 and GST-Rab5 in our in vitro pulldown assay (Fig. 4A). Binding of WT p110␤ to GTP␥S-Rab5 was 3.3-fold higher than to GTP␥S-Rac1 (17.7% of input as compared with 5.4% of input) (Fig. 4B). This is consistent with Fritsch et al. (6), who observed in vitro that p110␤ exhibited weaker binding to Rac1 than to Rab5. Also consistent with previous studies (18,19), we could detect binding of p85␣ to GST-Rac1 (data not shown). However, binding was weak compared with p110␤ (1% of the input, even when using 4-fold more p85␣ lysate as compared with p85␣/p110␤ heterodimer lysates). Thus, the binding of Rab5 and Rac1 to p85␣ is negligible as compared with their binding to p110␤.
As expected, the I597S mutant did not bind active Rab5A but did bind to active Rac1. Conversely and consistent with the observations of Fritsch et al. (6), the RBD-DM mutant of p110␤ bound active Rab5A but not Rac1 (Fig. 4C). Using the in vitro pulldown assay with GTP␥S-loaded GST-Rac1, we also tested the newly identified Rab5-uncoupled p110␤ mutants as heterodimers with p85␣. All of the helical domain mutants tested exhibited binding to GST-Rac1 that was comparable with that seen with WT p110␤ (Fig. 4, D and E). In addition, mutations in Rab5 binding to PI3K␤ is required for invasion the RBD that were reported to disrupt Rab5 binding (15) showed no significant difference in binding to GST-Rac1 with the exception of I234A, which is near the previously identified RBD-DM mutations (S211D and K230A) (Fig. 4, E and F). These data demonstrate that p110␤ mutations that disrupt Rab5 binding do not affect the binding of Rac1 to p110␤.

p110␤ mutations that disrupt Rab5 binding do not affect PI3K␤ kinase activity or activation by G␤␥
To verify that the enzymatic activity of the p110␤ helical domain mutants was intact, we performed in vitro kinase assays. For these assays, we selected a subset of p110␤ mutants: F508A and I512A, which showed Ͻ33% binding to GTP-loaded A, space-filling model of the murine p110␤ catalytic subunit with the iSH2 and cSH2 domains of the p85␤ regulatory subunit (green) (Protein Data Bank (PDB) code 2Y3A). Yellow, the N-terminal ABD; gray, the RBD; orange, the C2 domain; cyan, the helical domain; purple, the C-terminal kinase domain. The blue dashed line represents the C2-helical linker, which was not observed in the X-ray structure. The arrows indicate where Rac1, Rab5, and G␤␥ bind to p110␤. Gln 596 /Ile 597 , whose mutation disrupts Rab5 binding, are shown in white. B, representative immunoblots (IB) of the GST-Rab5A pulldown (PD) assay. Human GST-Rab5A was immobilized on GSH-agarose beads and loaded with GDP or GTP␥S. The beads were incubated with whole-cell lysates (Input) from HEK293T cells transfected with p85␣-FLAG without or with WT p110␤-Strep or the Rab5-uncoupled I597S mutant. C, representative immunoblots of the GST-Rab5A pulldown assay with p110␤ mutants. Samples were analyzed by SDS-PAGE and blotted for Strep (p110␤) and FLAG (p85␣). D, table of Rab5A-binding activity for representative p110␤ mutants. Binding to GTP␥S-Rab5A was calculated as a percentage of the input and then normalized to WT p110␤ binding, which was set to 100%. Binding, as compared with WT p110␤, was stratified into three groups: 0 -33% (red), 33-66% (blue), and Ͼ66% (green). Residues in the G␤␥-binding loop are indicated with an asterisk. E, ribbon diagrams of p110␤ (upper panel) and a magnified view of the helical domain (lower panel). Residues are color-coded based on their Rab5A-binding activity (PBD code 2Y3A).

Rab5 binding to PI3K␤ is required for invasion
Rab5 relative to WT, and K510A and E517A, which showed 33-66% binding relative to WT (Fig. 5A). The p85␣/p110␤ heterodimers were expressed in HEK293T cells, isolated on Strep-Tactin beads, and eluted with desthiobiotin. The eluted product was then assayed for activity using vesicles containing 2.9 mol % phosphatidylinositol 4,5-bisphosphate (PIP 2 ) as a substrate. As expected, the negative control kinase-dead (KD; K805R) p110␤ exhibited minimal kinase activity. The Rab5uncoupled mutants of p110␤ exhibited specific activities comparable with WT p110␤ (Fig. 5B). These data demonstrate that the basal kinase activity of p110␤ is unaffected by mutations in the helical domain.
To determine whether the p110␤ mutants responded to a known activator of p110␤, we measured in vitro kinase activity in the presence of purified G␤␥ (5). We chose G␤␥ because the G␤␥-binding loop is in close proximity to the Rab5-binding interface in the p110␤ helical domain. Heterodimers of p85␣ with WT or mutant p110␤ showed an ϳ2-fold activation by G␤␥ (p values Ͻ0.01; Fig. 5, C and D); the I512A mutant, which showed a 2-fold activation, trended toward significance (p ϭ 0.0712). In contrast, a mutation known to disrupt G␤␥ binding to PI3K␤ (K532D/K533D (5)) abolished activation by G␤␥ (Fig.  5E). Taken together, our data show that helical domain muta-tions that disrupt Rab5 binding have no effect on the other biochemical activities of p110␤.

Hydrogen-deuterium exchange MS (HDX-MS) analysis of the Rab5-binding site in p110␤
To examine the Rab5-binding site in p110␤ by an orthogonal approach, we used HDX-MS. Experiments were carried out at five time points (3 s at 1°C as well as 3, 30, 300, and 3000 s at 18°C). Sequence coverages of 86.8 and 90.2% were achieved for p110␤ and p85␣ with 155 and 104 peptides, respectively. Decreases in hydrogen-deuterium exchange rates (Ͼ5%) were observed within the helical domain of p110␤ in peptides spanning residues His 492 -Glu 513 and Arg 566 -Leu 578 (Fig. 6, A and C). Peptide His 492 -Glu 513 (Fig. 6A, right panel) overlaps with ␣-helix Asp 509 -Glu 517 , which was critical for Rab5 binding in the pulldown assay (Fig. 6A, left panel). Peptides corresponding to the other critical ␣-helix, Leu 585 -Ile 597 , were not detected in the MS analysis. However, the Arg 566 -Leu 578 peptide (Fig. 6A, right panel) includes Glu 567 whose mutation caused a greater than 50% decrease in Rab5 binding (Fig. 6A, left panel). An increase in the rate of hydrogen-deuterium exchange was also observed in the kinase domain of p110␤ in peptides spanning residues Asn 729 -Met 742 (Fig. 6, B and C). No significant

Rab5 binding to PI3K␤ is required for invasion
changes were observed in the RBD of p110␤ or in the p85␣ regulatory subunit. The HDX-MS data support our findings that regions Tyr 505 -Glu 517 and Ala 589 -Ile 597 of the helical domain of p110␤ constitute the single Rab5-binding interface.

Rab5-PI3K␤ interactions are required for tumor cell chemotaxis, invasion, and gelatin degradation
Previous work from our lab established a role for GPCRmediated PI3K␤ kinase activity in breast cancer metastasis (20).
To study the role of Rab5 binding to PI3K␤ in metastasis-associated cellular activities, we used stable knockdown/rescue MDA-MB-231 cells that express the murine myc-tagged WT and Rab5-uncoupled p110␤ mutant (I591S) at similar levels ( Fig. 8A). In a transwell migration assay, chemotaxis toward LPA was reduced by ϳ70% in cells expressing the Rab5-uncoupled mutant (Fig. 8B); chemotaxis toward EGF was reduced by ϳ50% (Fig. 8C). Similar results were observed with a transwell invasion assay in which EGF-stimulated invasion through Matrigel was reduced by ϳ50% in cells expressing Rab5-uncoupled PI3K␤ (Fig. 8D).
Metastasis requires that tumor cells invade from the primary tumor into the surrounding extracellular matrix. To investigate the role of the Rab5-PI3K␤ interaction in matrix degradation, we compared the matrix-degrading activity of MDA-MB-231 knockdown/rescue cells expressing murine WT, KD (K799R) or Rab5-uncoupled mutant p110␤. Cells were plated on fluorescently labeled gelatin, and the area of degradation per cell was measured (Fig. 8, E and F). Expression of either p110␤ mutant inhibited gelatin degradation by 80 -90%. Taken together, these data suggest a critical role for Rab5 binding to PI3K␤ in the motility and invasion of breast tumor cells.

Discussion
We previously described a p110␣/p110␤ chimera, which contained the N terminus of p110␣ (ABD and RBD) linked to the C terminus of p110␤ (C2, helical, and kinase domains) (21) and showed that it could bind to GST-Rab5 (13). Subsequent mutagenesis studies identified two residues (Gln 596 and Ile 597 ) in the helical domain whose mutation to the corresponding residues in p110␦ abolished binding to Rab5 (13). The present study used two independent techniques, structure-directed alanine-scanning mutagenesis and HDX-MS, to unambiguously define the full Rab5-binding interface within the catalytic subunit (p110␤) of PI3K␤. Error bars represent S.E. Statistical analyses were performed using one-way ANOVA. No statistical differences were observed between WT and RBD p110␤ mutant proteins. C, ribbon diagram of p110␤ showing residues in the helical domain and RBD that were targeted for mutagenesis. Color coding reflects Rab5-binding activity relative to WT p110␤: red, 0 -33%; blue: 33-66%; green, Ͼ66%. Residues in the RBD and G␤␥-binding loop that are not observed in the X-ray structure are depicted as dashed lines.

Rab5 binding to PI3K␤ is required for invasion
Our screen identified specific residues in the helical domain of p110␤ (Tyr 505 , Phe 508 , Asp 509 , Ile 512 , Glu 513 , Ala 516 , and Ala 593 ) that are critical for Rab5A binding (mutation leads to Ͻ33% of WT binding). These residues are located within two perpendicular ␣-helices (Asp 509 -Glu 517 and Leu 585 -Ile 597 ; human sequence) that are situated just below the G␤␥-binding loop ( Fig. 1A) (3). Residues critical for Rab5 binding in helices Asp 509 -Glu 517 and Leu 585 -Ile 597 have their side chains oriented toward a common surface (Fig. 1E). We believe that these residues make up the main Rab5-binding interface. In contrast, residues Lys 510 and Glu 517 have side chains that are oriented away from the primary Rab5-binding interface, and their mutation caused only a partial reduction in binding (33-66%). Two residues whose mutation showed limited effects on Rab5 binding (Gln 563 and Glu 567 ) are found on a helix that is parallel to and above helix Leu 585 -Ile 597 ; these residues most likely define the distal edge of the Rab5-binding interface. Our mutagenesis studies define a discrete binding site, as residues critical for binding were surrounded by residues whose mutation did not affect the interaction.
HDX-MS experiments identified two peptides, His 492 -Glu 513 and Arg 566 -Leu 578 , which were protected from solvent exchange by p110␤ binding to GTP-Rab5. His 492 -Glu 513 overlaps significantly with the Asp 509 -Glu 517 helix identified in our screen. Arg 566 -Leu 578 includes a residue (Glu 567 ) whose mutation decreases binding by over 50%. Although a decrease in solvent accessibility in these peptides could be due to a secondary conformational change caused by Rab5 binding, their coincidence with the region identified by our mutagenesis studies suggests that the changes are due to direct Rab5 binding. Inter-

Rab5 binding to PI3K␤ is required for invasion
estingly, Asn 729 -Met 742 from the kinase domain showed an increase in solvent exposure upon incubation with Rab5A, presumably due to a secondary conformational change. This region of p110␤ has been reported to interact with membranes (5), suggesting that conformational changes in p110␤ upon Rab5 binding could promote membrane targeting and provide a possible mechanism of activation, similar to that observed with the p110␣ oncogenic H1047R mutation (22).
We verified that mutation of the helical domain did not affect the overall functionality of p110␤ by measuring binding to Rac1. For WT p110␤, we observed 3-fold greater binding to Rab5 as compared with Rac1, but mutation of the Rab5-binding site had no effect on Rac1 binding, and mutation of the RBD did not affect Rab5 binding. Similarly, mutations that disrupt Rab5 binding did not affect basal PI3K␤ activity or its activation by G␤␥. Given the proximity of the Rab5-and G␤␥-binding sites in p110␤, activation by G␤␥ is an important control for the specificity of mutations that disrupt Rab5 binding. We did observe a modest reduction in the activation of the E517A mutant by G␤␥, perhaps because this residue resides at the base of the G␤␥-binding loop (5). Importantly, our biochemical analyses demonstrate that the loss of Rab5 binding in the mutants described here is not due to overall, nonspecific disruptions or indirect conformational changes of the p110␤ structure.
Of note, previous reports about the effect of Rab5 binding on p110␤ kinase activity are conflicting. In in vitro kinase assays performed with nonlipidated Rab5, we could not detect any effect on p110␤ kinase activity. Further studies with prenylated or membrane-targeted Rab5 will be required to fully explore the regulation of PI3K␤ kinase activity by Rab5.
Metastasis is a complex process that involves the migration and invasion of tumor cells though the extracellular matrix, intravasation into blood vessels, and extravasation at distal sites (23). Both invasion and transendothelial migration require the formation of degradative structures called invadopodia, actinrich protrusions that promote the secretion of matrix metalloproteases (20). Work from our lab previously showed that G␤␥ coupling to p110␤ activity in breast cancer cells is critical for macrophage-induced invasion, matrix degradation, and tumor extravasation (20). We now show that the PI3K␤-Rab5 interaction is similarly necessary for EGF-mediated chemotaxis and invasion as well as gelatin degradation. Currently, we do not understand how PI3K␤ binding to Rab5 regulates these processes. However, Rab5 is required for matrix degradation, through its regulation of endocytic trafficking and activation of Rac1 (24). Given that Rac1 and PI3K␤ can form a positive feedback loop in some cell types (25), it is possible that PI3K␤-Rab5 interactions may contribute to the endosomal activation of Rac1.
Our data are not in agreement with earlier studies on the PI3K-Rab5 interaction by the Anderson and co-workers (14), who reported that the p85␣ regulatory subunit binds to Rab5 in a nucleotide-independent manner. We could not detect bind-

Rab5 binding to PI3K␤ is required for invasion
ing of Rab5 to p85␣ alone or to p85␣ when expressed as a heterodimer with the Rab5-uncoupled mutant, I597S p110␤. Our data are consistent with the original study that identified p110␤ as a Rab5-interacting protein (7) and that demonstrated Rab5 binding to in vitro translated p85␣/p110␤ heterodimer and to the p110␤ catalytic subunit, but not to p85␣.
Whitecross and Anderson (15) recently reported that the RBD of p110␤ binds Rab5. However, this study used a chimera in which a fragment of the p85␣ iSH2 domain was linked to the N terminus of p110␤ via a short seven-residue glycine linker (15). The truncated iSH2 (residues 466 -567) has not been biochemically or structurally characterized. Furthermore, the truncated iSH2 domain deletes residues that mediate interactions with the C2 domain of p110␣ (27) and p110␤ (3) and whose deletion or mutation leads to oncogenic activation of PI3K␣ (29 -31). Thus, the chimera is unlikely to accurately reflect the conformation of the full-length p85␣/p110␤ heterodimer. Importantly, our study, which used full-length p85␣ and p110␤, demonstrated that the p110␤ mutants described by Whitecross and Anderson (15) exhibit WT levels of binding to Rab5. Consistent with these findings, HDX-MS analysis failed to detect interactions between Rab5 and either p85 or the RBD of p110␤, despite being present in 7.5-fold molar excess.
In summary, we have defined a single, discrete binding site for Rab5 in PI3K␤, comprising two perpendicular ␣-helices in the helical domain. Individual point mutants in this region abolish PI3K␤-Rab5 binding, but have no effect on PI3K␤ binding to Rac1 or PI3K␤ kinase activity, under basal or G␤␥-stimulated conditions. Using both biochemical and biophysical approaches, we could not detect any contributions from the RBD of p110␤ or from the p85␣ regulatory subunit to Rab5 binding. Mutation of the Rab5-binding site in PI3K␤ in breast cancer cells has significant inhibitory effects on tumor cell motility and invasion and blocks matrix degradation. The unambiguous determination of a single Rab5-binding site will facilitate the development of biochemical tools to further study the functions of this interaction in tumor metastasis and other physiological and pathophysiological cell behaviors.

GST-Rab5/GST-Rac1 binding assays
GSH-agarose beads were incubated with 1.0 nmol of GST, GST-Rab5A, or GST-Rac1 protein overnight at 4°C. The beads were washed three times with nucleotide loading buffer and loaded with either 1 mM GDP (Sigma) or GTP␥S (Sigma) for 15 min at 30°C in loading buffer. MgCl 2 was added to the beads to a final concentration of 20 mM. The beads were incubated at 30°C for 3 min and then transferred to ice for 20 min. After nucleotide loading, the beads were washed with nucleotide stabilization buffer (25 mM Tris-HCl, pH 7.5, 50 mM NaCl, 20 mM MgCl 2 , 0.06% (w/v) CHAPS, 2 mM DTT) containing 10 M nucleotide, GDP, or GTP␥S.

Rab5 binding to PI3K␤ is required for invasion
HEK293T cell lysates (100 g of total protein) were incubated with the GST beads in the presence of 10 M nucleotide (either GDP or GTP␥S) for 1 h on a rotating wheel at 4°C. The beads were washed four times with nucleotide stabilization buffer and 10 M nucleotide and boiled in 2ϫ Laemmli sample buffer.
One-third of the eluted bead samples and 1 ⁄ 20 of the input samples (5 g of total protein) were separated by 7.5% SDS-PAGE and analyzed by immunoblotting. Membranes were incubated with a rabbit StrepII antibody (Abcam, ab76949) and a mouse FLAG antibody (Sigma, F4042) in Odyssey Blocking Buffer (LI-COR Biosciences). Goat anti-rabbit (LI-COR Biosciences, 925-68071) and goat anti-mouse (LI-COR Biosciences, 925-32210) antibodies conjugated to IR dyes were used with a LI-COR Biosciences Odyssey Fc Imaging System and Image Studio Lite software to visualize and quantify the bands.
The input values were multiplied by 20 to determine the total p110␤ incubated with the beads. The sample values for GST-Rab5 binding and GST-Rac1 binding were corrected to account for the fraction of the eluted samples that was loaded on the gel. The percentage of input bound was calculated by dividing the total sample values (bead-bound p110␤) by the total input values and multiplying by 100. To pool data from individual experiments, binding was normalized to Rab5-GTP or Rac1-GTP binding for WT PI3K␤ in each experiment

Rab5 binding to PI3K␤ is required for invasion
without G␤ 1 ␥ 2 for 10 min at room temperature. The reaction volume was brought to 55 l with kinase assay buffer. To start the reaction, 5 l of the ATP mixture (100 M ATP containing 10 Ci of [ 32 P]ATP final) was added and incubated for 10 min at room temperature before spotting duplicate 4-l samples on nitrocellulose (35). Once dry, the nitrocellulose was washed twice with wash buffer (1 M phosphoric acid, 1 M NaCl) for 1 min and then four more times for 5 min. Total counts in each reaction mixture were determined by spotting diluted samples (1:20) onto nitrocellulose without washing. Radioactivity was quantitated using a PhosphorImager (GE Healthcare).
The relative units of p110␤ in each reaction were calculated by loading equal volumes of the p110␤ elutions for 7.5% SDS-PAGE and immunoblotting as described above. To normalize the intensity values across experiments, a standard curve of purified KD p110␤-StrepII(ϫ3) was included on each blot. The slope (intensity/l) of the best-fit line for the standards was used to calculate the relative units of the p110␤ sample by dividing the intensity of the sample by the slope. This was then corrected to represent the total units of p110␤ in the reaction.

Protein expression and purification for HDX-MS
BL21(DE3) cells were transformed with constitutively active Rab5A Q79L (Agilent). Bacterial cultures in 2ϫ YT (Sigma) broth (16 g/liter tryptone, 10 g/liter yeast extract, 5 g/liter NaCl) were induced with 0.5 mM isopropyl 1-thio-␤-D-galactopyranose after growth to an A 600 of 0.6 -0.8 and then grown at 37°C for an additional 4 h. The bacteria were harvested by centrifugation, washed with ice-cold PBS, and resuspended in lysis buffer (20 mM Tris-HCl, pH 7.5, 100 mM NaCl, 2 mM ␤ME, protease inhibitor mixture set III (Sigma)). The cells were sonicated on ice for 5 min (10 s on, 10 s off, level 6.0; Misonix Sonicator 3000). Triton X-100 was added to the lysate to a final concentration of 0.1% (v/v) and centrifuged at 20,000 ϫ g for 45 min.
The supernatant was loaded at a flow rate of 2 ml/min onto two 5-ml GSTrap HP columns (GE Healthcare) in tandem, preequilibrated with 30 ml of H 2 O followed by 30 ml of Hep A buffer (20 mM Tris-HCl, pH 7.5, 100 mM NaCl, 2 mM ␤ME). The columns were washed with 50 ml of Hep A buffer, and the GST tag was cleaved by incubating the column with 10 ml of lipoyl domain-Tev protease solution containing 10 mM ␤ME on ice overnight. The protein was eluted with 20 ml of Hep A. Imidazole was added to a final concentration of 10 mM.
The cleaved Rab5A was loaded onto a 5-ml HisTrap FF column (GE Healthcare) pre-equilibrated with 10 ml Ni-NTA buffer A (20 mM Tris-HCl, pH 7.5, 100 mM NaCl, 10 mM imidazole, pH 8.0, 2 mM ␤ME) to remove the His-tagged lipoyl domain-Tev. The flow-through and a 10-ml Ni-NTA buffer A wash were pooled and concentrated to 1 ml using a 10,000 molecular weight cutoff Amicon concentrator (Millipore). GTP␥S was added in a 2-fold molar excess relative to Rab5A along with 25 mM EDTA. After incubation for 1 h at room temperature, the solution was exchanged with gel filtration buffer (20 mM HEPES, pH 7.0, 150 mM NaCl, 1 mM MgCl 2 ) and concentrated to a final volume of 900 l. MgCl 2 was added to 30 mM, and the solution was incubated for 10 min on ice. Rab5A was purified using a Superdex 75 10/300 GL size exclusion column (GE Healthcare) equilibrated in gel filtration buffer.
The eluate was loaded onto two 1-ml StrepTrap HP columns (GE Healthcare) equilibrated in Hep A buffer (20 mM Tris-HCl, pH 8.0, 100 mM NaCl, 5% (v/v) glycerol, 2 mM ␤ME). To cleave the streptavidin tag, the proteins were loaded with a solution of LipTev protease in 10 mM ␤ME and incubated at 4°C overnight. The protein was eluted by passing Hep A buffer through the columns. The StrepTrap elution was concentrated using a 50,000 molecular weight cutoff Amicon concentrator (Millipore). The concentrated protein was additionally purified using a Superdex 200 Increase 10/300 GL size exclusion column (GE Healthcare) equilibrated with gel filtration buffer (20 mM HEPES, pH 7.5, 150 mM NaCl, 0.5 mM tris(2-carboxyethyl)phosphine).

HDX-MS
HDX experiments were carried out as described previously (28). In brief, experiments were performed in a 5-l reaction volume with final PI3K␤ and Rab5A concentrations of 1.6 and 12 M, respectively. The two conditions tested consisted of PI3K␤ incubated with or without Rab5A for a duration of 10 min. Deuterium exchange was initiated by the addition of 3.35 l of D 2 O buffer (10 mM HEPES, pH 7.5, 100 mM NaCl, 99% (v/v) deuterium oxide). Exchange was evaluated at five time points (3 s at 1°C as well as 3, 30, 300 and 3000 s at 18°C) and was terminated by the addition of 65 l of quench buffer (2 M guanidine HCl, 3% formic acid). All experiments were carried out in triplicate. Samples were immediately frozen in liquid nitrogen and stored at Ϫ80°C.
Samples were quickly thawed and injected onto an ultraperformance LC (UPLC) system at 2°C. Samples were loaded onto two immobilized pepsin columns (Applied Biosystems, Poroszyme) at 10 and 2°C, respectively, at a flow rate of 200 l/min for 3 min. Peptides were collected and desalted on a VanGuard precolumn trap (Waters) and loaded onto an ACQUITY 1.7-m particle, 100 ϫ 1 mm C 18 UPLC column (Waters). Separation and elution of peptides from the analytical column were achieved using a gradient of 5-36% mobile phase B (Buffer A, 0.1% formic acid, LC/MS grade; Buffer B, 100% acetonitrile,

Rab5 binding to PI3K␤ is required for invasion
LC/MS grade) over 16 min. Mass spectrometry experiments were performed using an Impact II TOF (Bruker) with acquisition over a mass range of 150 to 2200 m/z using an electrospray ionization source operated at 200°C and a spray voltage of 4.5 kV. Peptides were identified using a data-dependent acquisition approach following MS/MS experiments (0.5-s precursor scan from 150 to 2000 m/z; 12 0.25-s fragment scans from 150 to 2000 m/z). MS/MS data sets were analyzed using PEAKS7 (PEAKS), and the false discovery rate was set at 1% using a database of purified proteins and known contaminants.
HD-Examiner software (Sierra Analytics) was used to automatically calculate the level of deuterium incorporation into each peptide. All peptides were manually inspected for correct charge state and the presence of overlapping peptides. Deuteration levels were calculated using the centroid of the experimental isotope clusters. The results for these proteins are presented as relative levels of deuterium incorporation, and the only control for back-exchange was the level of deuterium present in the buffer (64.32%). Changes in any peptide at any time point greater than 5% and 0.5 Da between conditions with a unpaired t test value of p Ͻ 0.01 were considered significant. Full deuterium incorporation data for all analyzed peptides are included in Table S1.

Rab5-PI3K␤ kinase assay
Expression constructs (p85␣-FLAG(ϫ3) and p110␤-Stre-pII(ϫ3)) were transfected into a 30-ml culture of Expi293F cells (Gibco) following the manufacturer's protocol. 72 h post-transfection, the cells were harvested by centrifugation and resuspended in 2 ml of lysis buffer (20 mM Tris-HCl, pH 7.4, 150 mM NaCl, 1% Tween 20, 1 mM DTT, 1 mM MgCl 2 , 10% glycerol, protease inhibitor mixture (cOmplete, EDTA-free, Roche Applied Science)). Cells were lysed by rotating at 4°C for 15 min and cleared by centrifugation at 13,000 ϫ g for 10 min. The soluble lysate was incubated with a 100-l slurry of Strep-Tactin Superflow high-capacity beads (IBA Lifesciences) for 60 min at 4°C with rotation. Beads were washed three times in wash buffer (20 mM Tris-HCl, pH 7.4, 150 mM NaCl, 1 mM DTT, 1 mM MgCl 2 , 10% glycerol, protease inhibitors). Protein was eluted in two sequential fractions of 100 l in elution buffer (wash buffer with addition of 5 mM desthiobiotin). The eluted proteins were pooled and dialyzed overnight against kinase assay buffer (as described above). Purity was assessed by SDS-PAGE, and protein concentration was determined by absorbance. Aliquots were flash frozen and stored at Ϫ80°C in assay buffer.
Recombinant GST-Rab5 was produced as described above, immobilized on GSH-agarose, and cleaved by overnight incubation with recombinant tobacco etch virus protease at 4°C. The cleaved Rab5 was eluted and dialyzed into kinase assay buffer. Protein purity was determined by SDS-PAGE, and protein concentration was determined by absorbance. Nucleotide loading was achieved by incubation of Rab5 with 1 mM EDTA and 10 mM nucleotide (GDP or GTP␥S) for 20 min at 30°C followed by the addition of MgCl 2 (3 mM final in Rab5 mixture). The protein was incubated for an additional 3 min at 30°C and then stored on ice.
Recombinant PI3K␤ (17.6 nM final) was incubated with freshly prepared vesicles (37 M PIP 2 final) and GDP-or GTPloaded Rab5 (10 M final) or bistyrosyl phosphopeptide (DDG-pYMPMSPGAGAGAGAGAGNEDpYMPMSPKS; 1 M) for 10 min. Assays were started by the addition of ATP (100 M final) containing 10 Ci of [ 32 P]ATP. 4-l aliquots were spotted onto nitrocellulose filters every 2 min. Filters were washed, and radioactivity was quantitated as described above. Activity is expressed as the slope of the progress curve between 0 and 10 min as determined by nonlinear regression.

Cell-based assays
Stable MDA-MBA231 p110␤ knockdown/rescue cell lines expressing WT and kinase-dead p110␤ have been described previously (20). Knockdown/rescue cells expressing Rab5-uncoupled p110␤ were produced using the same methods. Transwell migration and invasion assays using MDA-MB-231 cells stably expressing WT or mutant p110␤ have been described previously (20). For the chemotaxis assay, breast cancer cells were serum-starved overnight and seeded in starvation medium in the upper chamber of transwell inserts coated with 10 g/ml fibronectin (Sigma). The lower chambers contained starvation medium without or with 10 M LPA or 5 nM EGF (EMD Millipore). Cells were allowed to chemotax for 8 h at 37°C. For the invasion assay, serum-starved cells were seeded onto transwell inserts coated with 300 g/ml growth factorreduced Matrigel (BD Biosciences). The lower chambers contained starvation medium without or with 5 nM EGF. Cells were allowed to invade for 24 h.
Degradation of Oregon Green 488 -conjugated gelatin (Molecular Probes) by MDA-MB-231 cells stably expressing WT or mutant PI3K␤ has been described previously (20). Cells were imaged, and the area of matrix degradation per cell was quantitated using ImageJ software.

Statistical analysis
Statistical analyses were performed using GraphPad Prism version 7.0c (GraphPad Software) to calculate the p values for one-way ANOVA and unpaired Student's t tests.