A PI3K p110β–Rac signalling loop mediates Pten-loss-induced perturbation of haematopoiesis and leukaemogenesis

The tumour suppressor PTEN, which antagonizes PI3K signalling, is frequently inactivated in haematologic malignancies. In mice, deletion of PTEN in haematopoietic stem cells (HSCs) causes perturbed haematopoiesis, myeloproliferative neoplasia (MPN) and leukaemia. Although the roles of the PI3K isoforms have been studied in PTEN-deficient tumours, their individual roles in PTEN-deficient HSCs are unknown. Here we show that when we delete PTEN in HSCs using the Mx1–Cre system, p110β ablation prevents MPN, improves HSC function and suppresses leukaemia initiation. Pharmacologic inhibition of p110β in PTEN-deficient mice recapitulates these genetic findings, but suggests involvement of both Akt-dependent and -independent pathways. Further investigation reveals that a p110β–Rac signalling loop plays a critical role in PTEN-deficient HSCs. Together, these data suggest that myeloid neoplasia driven by PTEN loss is dependent on p110β via p110β–Rac-positive-feedback loop, and that disruption of this loop may offer a new and effective therapeutic strategy for PTEN-deficient leukaemia.

D ysregulation of the molecular pathways involved in the self-renewal, differentiation and proliferation of haematopoietic stem cells (HSCs) can cause leukaemia. Notably, the serine/threonine kinase Akt, which acts downstream of PI3 kinase (PI3K), is hyper-phosphorylated in up to 80% of acute myeloid leukaemia (AML) cases 1 . This is unlikely to be due to mutations in upstream receptor tyrosine kinases alone. In chronic myelogenous leukaemia, PI3K/Akt signalling can also be activated through downregulation of the phosphatase and tensin homologue (PTEN) by BCR-ABL 2 . PTEN is a lipid phosphatase that counteracts PI3K signalling by dephosphorylating phosphatidylinositol-3,4,5-trisphosphate (PIP3). PTEN is frequently inactivated in haematological malignancies 3,4 , including in AML and T cell acute lymphoblastic leukemia (T-ALL) 5 . Notably, PTEN expression is often reduced in the disease through several other modes of PTEN regulation, for example, microRNAs, epigenetic modifications and ubiquitination [6][7][8][9] , which likely contribute to the high frequency of Akt phosphorylation in myeloid leukaemia.
In mice, genetic ablation of PTEN in the haematopoietic system leads to HSC depletion in the bone marrow (BM), myeloproliferative neoplasia (MPN) and transplantable acute leukaemia (myeloid or T-cell leukaemia) [10][11][12] . In patients, MPNs such as chronic myelogenous leukaemia or myelofibrosis can progress to AML 13 .
Class I PI3Ks are heterodimeric lipid kinases that produce the lipid second messenger PIP3 on stimulation of cells by many growth factors. Class I PI3Ks are divided into class IA (p110a, p110b and p110d) and class IB (p110g) enzymes; of these, the p110a and p110b isoforms are ubiquitously expressed, while p110d and p110g are enriched in leukocytes. Work in several different murine models has documented distinct requirements for different PI3K isoforms in particular tumour types 14,15 . For example, p110a is essential in a model of mutant Kras-induced lung adenocarcinoma 16 . Recently, we showed that Ras-mutated myeloid leukaemia is also dependent on the p110a isoform, and combined pharmacologic inhibition of p110a and mitogenactivated protein kinase kinase (MEK) could be an effective therapeutic strategy for Ras-mutated myeloid malignancies 17 . Although p110b plays a less prominent role in receptor tyrosine kinase (RTK) signalling, it mediates G protein-coupled receptor (GPCR) and integrin signalling [18][19][20] , and has been shown to interact specifically with Rho family GTPases Rac1 and CDC42 (ref. 21). Several recent studies demonstrated that p110b is required in many, but not all, PTEN-deficient solid tumours 20,22,23 . However, it is not known which PI3K isoforms are most important for myeloid neoplastic transformation driven by PTEN loss.
A number of pan-class I PI3K and dual class I/mTOR inhibitors are now in clinical trials for cancer, including leukaemia. However, targeting PI3K with these inhibitors could potentially lead to severe toxicity, which could be prevented by targeting single PI3K isoforms. To this end, numerous isoform-selective compounds are currently under development with some already in clinical trials 14 . The p110d-selective inhibitor idelalisib (referred to here as GS1101) has been remarkably effective in treating indolent B-cell malignancies, and is now approved by the FDA for the treatment of chronic lymphocytic leukaemia 24 . In the case of solid tumours, p110a-selective inhibitors have shown great promise in early-phase trials for patients with tumours bearing PIK3CA mutations 14 .
Notably, selective inhibitors of p110b are in clinical trials as anticancer reagents for advanced solid tumours with PTEN deficiency (NCT01458067). Thus, unravelling the role of each PI3K isoform, and its contribution to leukaemic transformation driven by PTEN loss, would inform rational approaches in targeting the PI3K pathway with a better therapeutic window.
In the present study, we used genetically engineered mouse models to determine which of the class IA PI3K isoforms are most important in mediating the effects of Pten loss in HSCs. We show that, in the setting of Pten loss, p110b is the main PI3K isoform responsible for MPN development and HSC depletion in the BM. Furthermore, we show that isoform-selective PI3K inhibitors recapitulate our genetic findings. We also found that a signalling loop featuring p110b and Rac plays an important role in the absence of Pten. Our results suggest that targeting p110b and/or Rac may lead to an effective therapeutic strategy for PTEN-deficient myeloid leukaemia.
Consistent with previous studies 10,11 , all Pten D/D mice developed MPN and reached the survival end point 20-40 days post injection (DPI; Fig. 1a). Pten D/D ;p110a D/D and Pten D/D ; p110d À / À mice also developed MPN with slightly extended survival (Fig. 1a). Notably, Pten D/D ;p110b D/D mice lived the longest, with median survival significantly longer than that of any other group (Fig. 1a). Further observation revealed that, whereas control, Pten D/D , Pten D/D ;p110a D/D and Pten D/D ;p110d À / À animals developed MPN, six out of nine Pten D/D ;p110b D/D mice succumbed to T-ALL (Fig. 1b). BM from the three Pten D/D ; p110b D/D mice that did develop MPN was analysed for excision of the pik3cb allele. Notably, we found that BM of these mice had incomplete deletion of Pik3cb alleles suggesting that p110b is critical for the development of MPN in this model ( Supplementary Fig. 1b). Deletion of p110b in HSCs using Mx1-Cre in animals that are wild-type (WT) for Pten does not significantly affect blood counts ( Supplementary Fig. 1c).
Histopathological analysis of moribund animals of Pten D/D , Pten D/D ;p110a D/D and Pten D/D ;p110d À / À mice at 20-40 days post pIpC showed that they developed massive splenomegaly with a marked increase in cells expressing myeloperoxidase, a marker used to detect leukaemic cells of the myeloid lineage, in both the spleen and liver ( Supplementary Fig. 2a), confirming MPN development in these mice. Notably, thymuses in these moribund mice appeared normal ( Supplementary Fig. 2b). However, most Pten D/D ;p110b D/D animals became moribund at 50-70 days post pIpC with markedly increased thymus weights ( Supplementary  Fig. 2b), infiltration of terminal deoxynucleotidyl transferasepositive T lymphoblasts (CD4 þ or CD4 and CD8 double-positive T-cell blasts) in the thymus and BM and increased white blood cell counts ( Supplementary Fig. 2a,c-e), all of which are manifestations of T-ALL. These results suggest that the p110b isoform of PI3K plays a uniquely important role in driving myeloid neoplastic transformation in mice with Pten-deficient HSCs, but does not contribute to the development of T-ALL in this murine model.
Previous studies have shown that Pten deletion in T-cell progenitors causes malignant transformation in the thymus and leads to T-cell lymphoma/T-ALL within 50-150 days 28,29 . Mice with Mx1-Cre-mediated deletion of Pten D/D , Pten D/D ;p110a D/D or Pten D/D ;p110d À / À showed infiltrating MPN disease and became moribund within 20-40 days after pIpC, earlier than the disease latency for T-cell disease development. Since Pten D/D ;p110b D/D mice did not develop MPN, they survived longer and developed T-cell lymphoma/T-ALL B50-70 days post pIpC injection, a timeline consistent with previous reports on T-cell lymphoma/ T-ALL formation in models of Pten loss in T-cell progenitors 28,29 .
This suggests that p110b ablation does not prevent T-ALL formation driven by Pten loss.
To examine further the PI3K isoform dependence in T-ALL induced by Pten loss, we investigated the roles of p110a and p110b in a different T-ALL model driven by Pten ablation in T-cell progenitors using Lck-Cre 30 . Interestingly, deletion of either p110a or p110b had no effect on T-ALL in this model 3 ( Supplementary Fig. 2f). These results underscore the distinct roles of p110b in myeloid and lymphoid neoplasia induced by Pten deletion in haematopoietic cells.   p110b mediates myeloid expansion induced by Pten loss. To further characterize disease in the Pten D/D ;p110 D/D mice, we killed mice of each genotype at 26 DPI, the time point at which the Pten D/D mice become moribund. Similar to previous reports, all animals in the Pten D/D group displayed massive splenomegaly, increased spleen cellularity and loss of spleen architecture at this time point (Fig. 1c,d). Both Pten D/D ;p110a D/D and Pten D/D ; p110d À / À mice showed evidence of MPN similar to that of Pten D/D mice, suggesting that ablation of p110a or p110d failed to rescue this disease phenotype. Notably, Pten D/D ;p110b D/D mice had significantly reduced spleen cellularity and size, compared with Pten D/D mice (Fig. 1c,d). Consistently, pathological analysis of the spleen and liver revealed infiltration of myeloid cells in Pten D/D , Pten D/D ;p110a D/D and Pten D/D ;p110d À / À animals, but not in Pten D/D ;p110b D/D mice (Fig. 1c). Flow cytometric analysis confirmed an increased population of myeloid cells (Mac1 þ Gr1 þ ) in the BM, spleen and peripheral blood of Pten D/D animals ( Fig. 1e; Supplementary Fig. 3). Again, the numbers of Mac1 þ Gr1 þ cells in these organs in Pten D/D ;p110b D/D but not in Pten D/D ;p110a D/D and Pten D/D ;p110d À / À animals were consistently reduced compared with those of Pten D/D mice ( Fig. 1e; Supplementary Fig. 3), suggesting that ablation of p110b suppressed myeloid cell expansion on Pten loss.
To further validate the role of p110b in the myeloid expansion caused by Pten loss, we performed colony assays in methylcellulose supplemented with myeloid growth factors. Compared with wild-type controls, both BM and spleen cells from Pten D/D animals generated an increased number of colonies, which was significantly reduced in Pten D/D ;p110b D/D mice (Fig. 2a). Together, these results demonstrate that p110b is required for MPN development in the absence of Pten.
To determine whether the contribution of p110b to myeloid neoplasia in the absence of Pten is a cell-autonomous or indirect effect, we transplanted whole BM cells from Pten D/D , Pten D/D ; p110b D/D or control mice into recipient mice (Fig. 2b). Four weeks after transplantation, all groups were treated with pIpC, and the relative frequency of CD45.2 þ donor-derived Mac1 þ Gr1 þ cells was monitored over 16 weeks. The proportion of Pten D/D donor-derived myeloid cells expanded significantly 4 weeks after pIpC and remained elevated during the course of the experiment. In contrast, the Pten D/D ;p110b D/D donor-derived myeloid population remained stable during the entire experiment, with levels much comparable to that of wild-type control mice (Fig. 2c). There were no significant differences in the percentage of donor-derived CD3-positive T cells among any of the groups tested ( Supplementary Fig. 4a). The percentage of donor-derived B220-positive B cells was reduced after Pten deletion, and deletion of p110b did not alter B-cell chimaerism ( Supplementary  Fig. 4b). These data suggest that p110b mediates the expansion of myeloid cells in a cell-autonomous manner.
p110b perturbs HSC homeostasis on loss of Pten. Earlier studies showed that HSC-specific deletion of Pten leads to the exhaustion of HSCs in the BM, their accumulation in the periphery and extramedullary haematopoiesis 10,11 . Hence, we wanted to test whether p110b ablation could rescue HSCs. In Pten D/D mice, the numbers of both the Lin À Sca-1 þ c-kit þ (LSK) cells, containing HSCs and the CD150 þ CD48 À Lin À Sca-1 þ c-kit þ population, which is enriched for long-term HSCs (LT-HSCs) were significantly reduced at 26 DPI, consistent with previous findings 11 ( Fig. 3a; Supplementary Fig. 5a). Ablation of p110b was able to partially rescue LSK cells and restore LT-HSCs in Pten-null BM (Fig. 3a). Loss of Pten did not change the total number of myeloid progenitors, or the frequencies of the common myeloid progenitors and granulocyte macrophage progenitors or megakaryocyte-erythroid progenitors, but led to a significant decrease in the number of common lymphoid progenitors consistent with original reports 11   To determine whether p110b is responsible for the increased extramedullary haematopoiesis in the spleen seen after Pten loss, we measured the frequency and absolute numbers of HSCs and progenitors in the spleens of Pten D/D and Pten D/D ;p110b D/D animals. Pten deficiency led to a significant increase in the number of LSK cells, as well as LT-HSCs, short-term HSCs (ST-HSCs) and progenitors in the spleens of Pten D/D animals ( Fig. 3b; Supplementary Fig. 5c,d) 11,31 , which was partially suppressed in the spleens of Pten D/D ;p110b D/D animals ( Fig. 3b; Supplementary Fig. 5c,d). These results suggest that deletion of p110b normalizes the distribution of phenotypic LSK cells between the BM and extramedullary tissues in Pten D/D ;p110b D/D animals compared with Pten D/D controls. Thus, our findings are consistent with the idea that p110b contributes to the perturbed HSC homeostasis observed in the absence of Pten leading to extramedullary haematopoiesis and the development of MPN.
p110b mediates leukaemia initiation in the absence of Pten. It has been reported that Pten loss in the BM leads to the depletion of HSCs and the generation of leukaemia-initiating cells 10,11 . To determine whether the p110b isoform also uniquely plays critical roles in these processes in the Pten-loss setting, we performed competitive multi-lineage repopulation assays to compare the contribution of marked (CD45.2) Pten D/D , Pten D/D ;p110a D/D , Pten D/D ;p110b D/D donor BM cells to that of wild-type competitor cells (CD45.1) following transplantation into lethally irradiated mice. Consistent with earlier reports, the contribution of donorderived cells to the peripheral blood was progressively decreased and eventually depleted for Pten D/D donors, but not for control donors 10,11 (Fig. 4a). HSCs derived from Pten D/D ;p110a D/D mice also failed to sustain long-term reconstitution. In striking contrast,  ARTICLE HSCs derived from Pten D/D ;p110b D/D mice were able to reconstitute recipient animals for more than 20 weeks (Fig. 4a). In addition, the majority of Pten D/D and Pten D/D ;p110a D/D -recipient mice developed T-ALL as evidenced by the abundance of donor-derived CD45.2 þ CD3 þ ;CD4 þ or CD3 þ CD4 À T lymphoblasts at the experimental end point of 20 weeks in these mice ( Fig. 4b; Supplementary Fig. 6).
In contrast, recipients of BM from wild-type control mice and from the majority of Pten D/D ;p110b D/D animals remained leukaemia free with few CD45.2 þ CD3 þ cells at the experimental end point ( Fig. 4b; Supplementary Fig. 6). Furthermore, analysis of the BM at week 20 showed that the donor chimaerism in the LSK, ST-HSC and LT-HSC compartments was significantly improved in the Pten D/D and Pten D/D ;p110a D/D groups (Fig. 4c). These data suggest that, in the absence of Pten, p110b is the major PI3K isoform critical for the loss of HSCs and for leukaemia initiation.
To determine the cellular mechanism underlying the improved reconstitution of Pten-null BM cells on loss of p110b, we examined the cell cycle status, senescence, apoptosis and homing properties of HSCs. As reported earlier, we also found that loss of Pten led to increased cycling of HSCs and reduced the number of quiescent HSCs (Supplementary Fig. 7a) 11 . Although ablation of p110b resulted in a tendency towards rescuing these effects, the results did not reach statistical significance ( Supplementary  Fig. 7a). Moreover, we did not observe any change in the proportion of whole BM or LSK cells expressing senescenceassociated b-gal activity or undergoing apoptosis in any of the groups tested ( Supplementary Fig. 7b,c). We then performed homing assays, in which fluorescently labelled BM cells were transplanted into irradiated wild-type hosts, and donor-derived cells were quantified after 24 h. We found that Pten deficiency significantly reduced the homing capacity of transplanted cells to the BM, and p110b ablation could partially rescue the homing potential (Fig. 4d). We also found that p110b ablation does not affect homing in Pten-wild-type BM cells (Fig. 4d). Thus, we conclude that p110b is responsible, at least in part, for the reduced homing activity of Pten-deficient HSCs.

Inhibition of p110b suppresses myeloid leukaemogenesis.
To determine whether our findings using a genetic method could be recapitulated by pharmacologic approaches utilizing PI3K isoform-selective inhibitors at effective doses as published in earlier studies, we first examined the effects of PI3K isoform inhibition on myeloid progenitor function. We cultured BM cells and splenocytes from Pten D/D animals in methylcellulose supplemented with myeloid growth factors in the presence of PI3K inhibitors. As reported in previous studies 10 , the pan-PI3K inhibitor GDC0941 and the mTOR inhibitor RAD001 significantly suppressed the increased colony formation arising from Pten deletion in a dosedependent manner (Fig. 5a). Consistent with our genetic findings, inhibition of p110a with BYL719 (a p110a-selective inhibitor) 32 and p110d with GS1101 (ref. 33) had a modest effect on the expansion of myeloid cells in the context of Pten loss (Fig. 5a). Since p110g is also expressed in leukocytes, we tested the p110g inhibitor NVSPI35 (ref. 34). Interestingly, inhibition of p110g with NVSPI35 showed some effect on BM cells, but not on splenocytes (Fig. 5a). Notably, inhibition of p110b with KIN193 (a p110bselective inhibitor, also known as AZD6482) 14,22 significantly reduced formation of both BM-and spleen-derived myeloid colonies in methylcellulose in a dose-dependent manner (Fig. 5a) suggesting that pharmacologic inhibition of p110b is highly effective in suppressing myeloid cell expansion driven by Pten loss.
To further determine the effects of pharmacologic inhibition of PI3K isoforms in vivo, we treated a group of pIpC-induced Pten D/D animals with BYL719, KIN193, IC87114 (a p110dselective inhibitor in the same class as GS1101, but with better bioavailability in mice) 35 , AS605240 (p110g-selective inhibitor suitable for in vivo studies) 36 or a vehicle control 10 days after induction (Fig. 5b). All the inhibitors were used at the effective doses in vivo as published in earlier studies [36][37][38] . Treatment with BYL719 (ref. 37), IC87114 (ref. 38) or AS605240 (ref. 36) resulted in a minimal survival benefit compared with vehicle-treated animals (Fig. 5c). Notably, mice treated with KIN193 had a significantly longer survival as compared with mice in any other group (Fig. 5c). Moreover, KIN193-treated animals appeared healthy and had significantly reduced spleen weights and normallooking spleen architecture compared with vehicle-treated mice ( Fig. 5d; Supplementary Fig. 8), consistent with our genetic data that ablation of p110b largely prevented myeloid leukaemia in Pten-deficient mice.
Next, we examined PI3K/Akt signalling in Pten-null BM cells in response to isoform-selective inhibition. As expected, vehicletreated Pten-null BM cells from Pten D/D animals showed markedly increased Akt phosphorylation compared with wild-type BM cells (Fig. 5e). Interestingly, inhibition of either p110a, p110b or p110d led to a significant reduction of p-Akt, compared with the controls (Fig. 5e). We also examined Akt activation by measuring basal p-Akt levels in LSK cells by intracellular phosopho-flow cytometry after Pten deletion and short-term isoform-selective inhibitor treatment of lineagenegative cells from Pten D/D BM. We detected significantly higher basal levels of p-Akt in Pten D/D LSK cells compared with WT LSKs, and again inhibition of p110a, p110b, p110d or p110g led to a significant reduction of p-Akt compared with the vehicle group (Fig. 5f). Since p110b is not the only isoform responsible for mediating Akt signalling in Pten-deficient BM and HSCs, Akt signalling alone is not sufficient to explain the specific biological effects of p110b ablation or inhibition observed in our study.
Pten-deficient HSCs depend on the p110b-Rac axis. Recent data suggested that p110a, p110d and p110g bind to and are activated by the Ras subfamily of GTPases, while p110b instead binds to and is activated by the Rho subfamily GTPases, Rac1 and CDC42 via its 'Ras-binding domain' (RBD) 17,21 . Previous studies also reported that an intact RBD was required for signalling and oncogenic transformation by wild-type p110b, suggesting a potential role for the interaction of Rho GTPase with p110b in transformation 39,40 . Notably, Rac1 and CDC42 can also be activated downstream of PI3K by PIP3-dependent guaninenucleotide exchange factors 41,42 . It has been previously reported that Rac plays important roles in the homing and survival of HSCs 43,44 . Given the significant rescue of HSCs in Pten D/D ; p110b D/D mice, we hypothesized that a unique positive-feedback signalling loop might exist between p110b and Rac, in which p110b is activated by Rac and Rac could in turn be activated by the phosphoinoside products of p110b in the setting of Pten-null haematopoietic cells.
Notably, we detected higher levels of Rac-GTP in the BM of Pten D/D mice, which could be suppressed by deletion of p110b but not p110a (Fig. 6a). To investigate whether the binding of p110b to Rac is important in mediating p110b activity in Ptendeleted BM cells, we mutated the two highly conserved key residues within the p110b RBD to generate a p110b-S205D/ K224A double mutant lacking the binding activity to Rac1 (ref. 21), and performed an add-back experiment with either wild-type or RBD-mutant p110b in BM cells derived from Pten D/D ;p110b D/ D mice ( Fig. 6b; Supplementary Fig. 10). Colony-forming assays revealed that, while adding back a wild-type p110b in Pten D/ D ;p110b D/D BM cells restored myeloid colony numbers comparable to those of Pten D/D BM cells, the RBD-mutant p110b failed to rescue colony formation (Fig. 6c). To determine whether p110b affects Rac-GTP levels in HSCs/progenitor cells (HSPCs), we performed the Rac-GTP assay either on Lin-negative Pten D/D BM cells or Pten D/D ;p110b D/D cells expressing wild-type p110b and RBD-mutant p110b. We detected higher levels of Rac-GTP in the Pten-deficient cells compared with WT control cells, and these levels were significantly reduced in Pten D/D ;p110b D/D cells. Adding back wild-type p110b to Pten D/D ;p110b D/D HSC/P cells, partially rescued Rac-GTP levels but adding back RBD-mutant p110b failed to rescue Rac activity ( Fig. 6d; Supplementary  Fig. 10) Together, these data suggest that the interaction of p110b with Rac plays an important role in mediating the myeloid clonogenic activity driven by Pten loss.
To further investigate the functional dependency of Pten-deleted leukaemic cells on the p110b-Rac axis, we utilized NSC23766, a potent Rac inhibitor 45 in our Pten-null model (Fig. 7a). We found that treatment of these mice for 10 days led to a reduced disease burden, as demonstrated by reduced spleen size and cellularity ( Fig. 7a; Supplementary Fig. 9). Similarly, treatment of Pten D/D mice with NSC23766 resulted in a significant reduction of HSC and myeloid progenitor numbers in the spleen compared with vehicle controls (Fig. 7b;  Supplementary Fig. 9). Treatment with NSC23766 also led to a significantly prolonged survival of Pten D/D mice (Fig. 7c), recapitulating the findings for genetic ablation or pharmacological inhibition of p110b in Pten-null animals.
Since Rac is required for p110b activation downstream of GPCRs 21 , we assessed the functional importance of the Rac-p110b signalling axis in HSPC function in response to activation of CXCR4, a GPCR important in the regulation of HSPCs. We used Transwell migration assays with CXCL12, a NATURE COMMUNICATIONS | DOI: 10.1038/ncomms9501 ARTICLE potent chemo-attractant of stem cells that signals through CXCR4. Lineage-negative BM cells from Pten D/D mice showed increased migration towards CXCL12 compared with WT control cells (Fig. 7d). This migration was abolished by the GPCR inhibitor pertussis toxin (PTX; Fig. 7d). Interestingly, we observed significantly reduced migration of Pten D/D ;p110b D/D cells, and of Pten D/D cells treated with either KIN193 or NSC23766, compared with Pten D/D and Pten D/D ;p110a D/D cells (Fig. 7d). This suggests that deletion of p110b, or pharmacologic inhibition of either p110b or Rac, partially interferes with the migration of Pten D/D  Freshly isolated lineage-negative BM cells were treated with inhibitors as in (e) and subjected to flow cytometry for LSK staining and intracellular P-Akt staining (n ¼ 3) for each, and median fluorescence intensities were normalized to control. cells towards a CXCL12 gradient, likely through perturbed GPCR signalling.
Because murine models of haematopoietic-specific Rac1 and Rac2 deficiency have revealed differential roles of Rac proteins in terms of HSPC function, we wanted to understand which Rac isoform is more important in the absence of Pten. To this end, we used siRNA to knockdown either Rac1 or Rac2, or both, and performed colony assays on Pten D/D and Pten D/D ;p110b D/D BM cells (Fig. 7e,f; Supplementary Fig. 10). We also tested the Rac inhibitor NSC23766, which targets both Rac1 and Rac2 (ref. 46). Knockdown of either Rac1 or Rac2, or their combined knockdown or pharmacological inhibition significantly reduced colony formation by Pten D/D cells to levels obtained with Pten D/D ; p110b D/D BM cells (Fig. 7f). However, knockdown of Rac1 or Rac2, the combination, or NSC23766 treatment did not further suppress colony formation beyond the effects of p110b deletion (Fig. 7f), suggesting that there is no additive or synergistic effect of Rac inhibition with p110b deficiency in Pten D/D cells. Together, these results suggest that p110b-Rac1/2 work in concert to mediate the effects of Pten loss in promoting myeloid neoplasia.

Discussion
We and others have reported that Pten-deficient solid tumours frequently rely on p110b (refs 20,23,47). In this study, we report for the first time an essential role for p110b in promoting haematologic neoplasia driven by Pten deletion in HSCs despite the expression of four different PI3K isoforms in haematopoietic cells. We have also found that p110b contributes to HSC depletion in the BM after Pten deletion. Interestingly, we found that Mx1-Cre-mediated deletion of p110b in HSPCs of animals that are WT for Pten does not significantly affect blood counts. In fact, these animals appear healthy for many months after excision, suggesting that targeting p110b may lead to an effective therapy for myeloid leukaemia with little toxicity to normal HSCs.
Despite the marked impact of genetic deletion or pharmacologic inhibition of p110b on MPN, and the significantly delayed onset of leukaemia, a fraction of Pten D/D ;p110b D/D animals succumbed to T-ALL at a later time. Berenjeno et al. 48 showed that in Pten þ /animals, inactivation of p110b led to reduced PIP3 generation in lymphoma tissues, but had little impact on lymphoma formation. It is possible that these tumours become p110b independent through the acquisition of secondary alterations. In fact, Yilmaz et al. 10 have documented the presence of cytogenetic alterations in leukaemic blasts from Pten D/D animals. Alternatively, isoform dependency may shift with cell differentiation. For example, the isoform dependency in the skin hamartoma driven by Pten loss changed from a p110b dependency in the basal layer of the epidermis to a p110a dependency in the suprabasal cells as the basal cells underwent stepwise differentiation to become suprabasal cells 37 . In this study, we also provide evidence that neither p110a nor p110b has any effect on T-ALL driven by T lymphocyte-specific deletion of Pten using Lck-Cre. In this system, it has been shown that p110d and p110g contribute to T-ALL induced by Pten loss in T cells 28 . These studies provide additional data that accentuate the distinct roles of p110b in the HSCs and in myeloid and lymphoid tumour initiation in the absence of Pten. Interestingly, we found that inhibition of p110b, p110a or p110d could similarly reduce p-Akt in Pten-deficient BM and HSCs, suggesting an Akt-independent pathway specific to p110b is important in Pten-deficient HSCs in promoting myeloid neoplasia. We report a new mechanistic insight that may explain the unique role of p110b in this setting. Since p110b binds to Rac rather than to Ras via its RBD, unlike the other class I PI3K isoforms 21 , we investigated the role of the p110b-Rac axis in the setting of myeloid neoplasia induced by Pten loss in HSCs. Notably, Rac signalling is not only important for the activation of p110b but it itself is also activated by PIP3 via PIP3-activated guanine-nucleotide exchange factors, forming a potential signalling loop (Fig. 7g) 14,49 . We found strong evidence that this loop is indeed active in our Pten-deficient model. We show that Rac was activated in Pten-null BM cells, and this activation was suppressed in Pten/p110b double KOs, but not in Pten/p110a double KOs. This hypothesis was further supported by our finding that only wild-type p110b, but not the RBD mutant of p110b, rescued colony formation in Pten/p110bdeficient BM cells. Notably, the effect of the RBD-mutant p110b on inhibiting colony formation in Pten-null BM cells is comparable to that of p110b deletion. An intact RBD was reported to be required for membrane localization of p110b for both signalling and oncogenic transformation by wild-type p110b in cultured cells 40 . Our data suggest that the interaction of p110b-Rac may play an important role in mediating p110b activity downstream of GPCRs and tyrosine kinases in the context of Pten deficiency. Moreover, pharmacologic inhibition of p110b or Rac in Pten-deficient mice resulted in a strikingly similar functional rescue in vivo, with a reduction in extramedullary haematopoiesis in the spleen and improved survival.
Of the three isoforms of Rac family GTPases, Rac2 is expressed specifically in haematopoietic cells, while Rac1 and Rac3 are ubiquitously expressed 44,50 . The Rac inhibitor NSC23766 targets all three Rac members: Rac1, 2 and 3 (ref. 46). Both Rac1 and CDC42 have been shown to bind p110b in a recent study 21 . Binding to Rac2/3 was not tested, but might also be expected, based on the homology of their effector domains with that of Rac1. Previous studies have suggested that Rac1 and Rac2 play both distinct and overlapping roles in HSCs and progenitor cells, while the role of Rac3 in haematopoiesis has not been defined 44,51 . It also has been shown that targeting both Rac1 and Rac2 was effective in a mouse model of BCR-ABL-induced MPN, as well as in a mouse model of MLL-AF9 AML 46,52 . Interestingly, our data show that the effect of knockdown of Rac2 is comparable to that of combined knockdown of both Rac1 and Rac2, or a pan-Rac inhibitor NSC23766. Since Rac2 is primarily expressed in haematopoietic cells, our data suggest that Rac2 could potentially be a better pharmacologic target with reduced toxicity.
Yilmaz et al. have shown that the mTOR inhibitor rapamycin can rescue HSC depletion and can suppress the development of leukaemia in vivo. More recently, the Armstrong and Morrison groups reported that mTORC1 and mTORC2 play critical roles in haematopoiesis and Pten-loss-driven leukaemogenesis, respectively 53,54 . By ablation or inhibition of p110b, we obtained similar results suggesting that the activation of mTOR in HSCs by Pten loss may be mediated largely by p110b. A recent study demonstrated that Rac1 regulates the activity of both mTORC1 and mTORC2 (ref. 55), providing a potential Akt-independent link between p110b-Rac and mTOR. Together, these data suggest that p110b-Rac acts upstream of Akt and mTOR. We feel that our data are most consistent with the working model shown in Fig. 7g, in which p110b works in a signalling loop with Rac to generate the key signals arising from Pten loss. Notably, these signals include both Akt-dependent and Akt-independent pathways leading to cell proliferation and migration.
In summary, our results provide the first evidence that PI3K-p110b plays an essential role in controlling HSC function in the setting of Pten loss 56 . We have also uncovered a specific role for p110b in myeloid leukaemia induced by Pten deficiency. In contrast, we found that p110b is dispensable for T-ALL induced by Pten loss. Most importantly, our data show that a p110b-Rac signalling loop is important for the induction of myeloid neoplasia in the absence of Pten. Thus, secondgeneration p110b-or Rac-selective inhibitors may interrupt this loop, thereby providing a promising new therapeutic strategy for Pten-deficient myeloid leukaemias while preserving normal haematopoiesis.
Colony-forming assays. BM and spleen cells were collected, subjected to red-cell lysis and resuspended in Iscove's modified Dulbecco's medium/10% fetal bovine serum/5% penicillin-streptomycin. Cells were plated in the presence of inhibitors in duplicate in M3434 methylcellulose media (Stemcell Technologies) at 1 Â 10 4 cells per dish for BM and 5 Â 10 4 cells per dish for spleen cells. Colonies were scored after 7 days.
Rac activation assay. BM cells from corresponding mice at 7 DPI of pIpC were collected and immediately subjected to Rac1 activation assay with the Rac1 activation assay kit (Millipore) according to the manufacturer's instructions.
Histology. Recipient mice were killed at the indicated time points, or when they began to show signs of disease. Organs were fixed in formalin, and histology slides were prepared and stained at the Brigham and Women's Rodent Histology Core Facility. Digital images were acquired on a Nikon Eclipse E400 microscope equipped with a digital camera and analysed using Spot Advanced software.
Immunohistochemistry. For histological analyses, formalin-fixed tissue sections were embedded in paraffin, sectioned and stained with haematoxylin and eosin by the Dana-Farber/Harvard Cancer Center Rodent Histopathology Core. P-Akt flow cytometry analysis of LSK cells. Phospho-flow cytometry was performed as previously described 56 , with the following modifications: lineagenegative BM cells from mutant animals were isolated using lineage depletion kit (Miltenyi Biotec) and serum starved for 1 h, and treated for 2 h with 1 mM of BYL719, KIN193, GS1101 and NVSPI3. Cells were then fixed with 4% paraformaldehyde, and permeabilized with cold 100% acetone. Cells were than stained simultaneously with c-kit, Sca-1 and anti-mouse P-Akt (Alexa 647; 1:100 dilution, cat. no. 2337, Cell Signaling). All data acquisition was performed on a LSRII (BD) flow cytometer, and results were analysed and basal level of P-Akt was calculated as normalized to WT cells by calculating median fluorescent intensity using FlowJo v.8.8.7 (Tree Star).
Long-term competitive repopulation assays. Recipient mice (4-6 weeks old female mice; B6.SJL strain) received two doses of 540 rads each, delivered 3 h apart. Nucleated BM cells from control and Pten D/D mice (C57 Bl/6) or from compound mutant mice (Pten D/D ; p110a D/D , Pten D/D ;p110b D/D and Pten D/D ;p110d À / À ) were mixed with wild-type competitor BM-nucleated cells (B6.SJL), and were injected into the retro-orbital venous sinus of irradiated recipients. Peripheral blood was obtained retro-orbitally every 4 weeks, subjected to red-cell lysis and analysed by flow cytometry to assess donor cell engraftment for up to 20 weeks after transplantation. Pilot experiments showed that Pten-mutant animals had a greatly reduced repopulation capacity; therefore, an excess of mutant cells over control cells was used. Each recipient mouse received 1 Â 10 6 Pten mutant, ctrl or compound mutant BM-nucleated cells plus 2 Â 10 5 competitor cells. Tissues from recipient mice were collected and stained for pathological examination.
Plasmid constructs. Mutations were generated in the pBabe-HA wild-type P110b by site-directed mutagenesis to obtain p110b-S205D/K224A double mutant to obtain Ras-binding mutant P110b. This plasmid was used to transduce Pten D/D ; p110b D/D BM cells and for colony-forming experiments.