PI3Kγ/δ and NOTCH1 Cross-Regulate Pathways That Define the T-cell Acute Lymphoblastic Leukemia Disease Signature

T-cell acute lymphoblastic leukemia (T-ALL) is an aggressive disease characterized by frequent relapse, resistance to chemotherapy, and reduced survival as compared with B-precursor ALL (1). Consistent with most lymphoid malignancies, tumor development and disease progression depend upon deregulation of an oncogene(s) and/or tumor suppressor genes that permit unchecked proliferation and resistance to programmed cell death. Chromosomal translocations driven by T-cell receptor–regulatory elements occur frequently in T-ALL, resulting in aberrant expression of a select group of transcription factor oncogenes such as TAL1, LMO1, LMO2, TLX1, and TLX3 (2–4). These gene rearrangement events are typically accompanied by constitutive NOTCH1 signaling due to mutations within its heterodimerization (HD) and C-terminal proline-, glutamic acid-, serine-, and threonine-rich (PEST) domains, causing ligand-independent activation and increased stability of the γ-secretase cleaved intracellular domain (ICN1; refs. 5, 6). Subsequent nuclear translocation of ICN1 stimulates transcription of downstream target genes such as HES1, DTX1, and MYC, the latter being an essential mediator of NOTCH1 signaling (7, 8). cMYC is a known master transcription factor that regulates genes essential for cell metabolism, proliferation, and survival (9, 10).

Aberrant activation of the PI3K/AKT pathway has also been implicated in the oncogenic transformation and tumor progression of T-ALL and is associated with aggressive biological features, drug resistance, and poor prognosis (11–13). In the majority of cases, this results from a loss or reduced function of the phosphatase and tensin homolog (PTEN) tumor suppressor, which limits levels of PIP₃ generated by class I PI3Ks (14). Interestingly, NOTCH1 has also been reported to indirectly regulate this signaling pathway due to its ability to alter PTEN expression, which has been implicated in inducing γ-secretase inhibitor (GSI) resistance in T-ALL by switching “oncogene addiction” (15). However, other studies have shown that PTEN status does not always correlate with GSI responsiveness, and that NOTCH1 and PI3K/AKT pathways may work in concert to promote tumor growth and survival (16, 17).

Previously, we have demonstrated the importance of the class I PI3K isoforms p110γ and p110δ in T-cell development and their ability to cooperate as non-classical oncogenes in supporting leukemogenesis in the absence of negative regulation by PTEN (18, 19). This was further evidenced by our observations that a highly selective dual PI3Kγ/δ inhibitor CAL-130 significantly reduced disease burden, prolonged survival of mice with established PTEN-null T-ALL, and induced apoptosis in human T-ALL tumor cells with aberrant PI3K/AKT signaling. Yet, it remains to be determined whether PI3Kγ/δ regulates transcriptional pathways typically associated with activated NOTCH1 (e.g., cMYC). Investigations along these lines are essential for establishing whether p110-selective small-molecule inhibitors could synergize or substitute for GSIs in the treatment of T-ALL. Indeed, the variable antitumor effects of GSIs reported in phase I clinical trials would suggest that such an approach is warranted (20, 21). To this end, we evaluated the molecular and genetic interplay between these pathways using the Lmo2-driven mouse model of T-ALL where both PI3K/AKT and NOTCH1 are activated (22, 23). Additionally, gene set enrichment and pathway analyses were used to determine if these disparate pro-oncogenic pathways controlled similar genes and transcriptional regulators that are determinants of the disease signature associated with T-ALL.

Cohorts of CD2-Lmo2 and Gt(ROSA)26Sor^{tm13(CAG-MYC, -CD2*)Rsky}/J;Lck-cre/Pten^fl/fl, both on C57BL/6J background, were monitored for the onset of leukemia (19, 22, 25). Experiments were performed in accordance with guidelines set forth by the Institutional Animal Care and Use Committee of Columbia University. Animals with established T-ALL received either the dual PI3Kγ/δ inhibitor CAL-130 (10 mg/kg every 8 hours; Calistoga Pharmaceuticals; ref. 19) or the GSI dibenzazepine (DBZ; 10 μmol/kg IP daily; Tocris; ref. 26) for a total of 7 days (27). Kaplan–Meier survival and statistical analyses were performed using GraphPad Prism Version 6.0 software. Values were considered significant at P < 0.05.

Cryopreserved human T-ALL samples were provided by St. Jude Children's Research Hospital and Vanderbilt University Medical Center after appropriate Institutional Review Board (IRB) review. All samples were collected with informed consent. Murine CD2-Lmo2 T-ALL cell lines 03007 and 03027 were generated as previously described (23). In brief, they were generated in the Davé lab (Vanderbilt University) from T-cell leukemia that arose in transgenic B6.CD2-Lmo2 mice. Once established in culture, aliquots of cells were banked in liquid nitrogen; samples for this study were from 2014. Cell lines were validated before use by determining their immunophenotype and by T-cell receptor Jβ rearrangement (23).

Plasmid pMSCV-IRES-mCherry and pMSCV-cMyc-IRES-mCherry were kindly provided by the laboratory of Dr. Riccardo Dalla-Favera (Columbia University, NY, NY). Retroviruses were produced in ecotropic packaging cell line 293T Platinum-E according to the manufacturer's instructions (Cell Biolabs Inc). Viral transduction was performed using the RetroNectin (Takara) and spinoculation method as previously described (28).

Sequencing of the Notch1 and Pten genes was performed on primary mouse T-ALL cells by PCR using Pfu DNA polymerase (Stratagene) with primers specific for exon 34 and exons 3 to 7, respectively.

Preparation, staining, and detection of cell surface and cytoplasmic proteins in primary T-ALL cells and murine T-ALL cell lines were performed as previously described (19).

Cell lysates were prepared on ice in M-PER Mammalian Protein Extraction reagent (Pierce) containing a cocktail of protease and phosphatase inhibitors (19). Lysates were subjected to SDS-PAGE, transferred to PVDF membrane (Immobilon-P, Millipore), and membranes probed by overnight incubation (4°C) with appropriate primary antibodies. Bound antibodies were visualized with HRP-conjugated secondary antibodies and ECL chemistry (SuperPico West).

Murine CD2-Lmo2 T-ALL cell lines 03007 and 03027 were plated at optimal density into 384-well tissue culture plates (Greiner 781080) using the Cell::Explorer automation system (PerkinElmer). CAL-130 or the GSI CompE [ENZO Biochem; was added using HP D300 Digital Dispenser, and drug activity measured at 60 hours using Cell Titer Glo (Promega; ref. 29) ]. The standard reference model of Bliss independence was used for the results analysis (30).

Cell lines were also grown in 6-well plates and individually treated with DMSO, CAL-130, IC87114 (Calistoga; ref. 31), and CompE for 72 hours. Proliferation and cell death were determined by cell counting with trypan blue and by staining with propidium iodide (PI; BD Biosciences) followed by flow cytometric analysis, respectively. For primary murine and human T-ALL samples, cells were harvested from thymi of diseased animals and plated with MS5-DL1 stromal cells in the presence of recombinant IL7 and FLT-3 (mouse) or recombinant IL7, FLT-3, and SCF (human). T-ALL cell viability following 72 hours of drug treatment was assessed using the BD Cell Viability Kit (BD Biosciences) coupled with fluorescent counting beads as previously described (19). After 72 hours of treatment, the absolute number of cells in DMSO control was set to 100%.

Mitochondria were isolated from murine T-ALL cells treated with DMSO or CAL-130 (2.5 μmol/L, 12 hours) and respiration analyzed in TRIS-MOPS buffer using Clark's type electrode as described (32).

Murine CD2-Lmo2 T-ALL cell lines treated with CAL-130 (2.5 μmol/L, 12 hours), CompE (1 μmol/L, 48 hours), or vehicle (DMSO) were centrifuged and directly suspended in XF assay medium (Seahorse Biosciences) supplemented with 10 mmol/L glucose (Sigma-Aldrich) and 10 mmol/L sodium pyruvate (Invitrogen). For primary CD2-Lmo2 T-ALL tumors, cells were grown on MS5-DL1 stromal cells in the presence of recombinant IL7 and FLT-3 (mouse) for 24 hours before the addition of DMSO, CAL-130 (2.5 μmol/L, 12 hours) or CompE (1 μmol/L, 48 hours). Tumor cells were then disrupted by pipetting, filtered through a 40-μm strainer to separate from the stromal cells, and centrifuged and suspended in XF assay medium (Seahorse Biosciences) supplemented with 10 mmol/L glucose (Sigma-Aldrich) and 10 mmol/L sodium pyruvate (Invitrogen). Primary tumors or tumor cell lines were added to a XF24 cell culture microplate coated with poly-l-Lysine (Sigma-Aldrich). After 30 minutes at 37°C, oxygen consumption rate (OCR) was measured using a XFe24 Analyzer (Seahorse Biosciences) after serial injections of oligomycin (1 μmol/L), carbonyl cyanide 4-[(trifluoromethoxy)phenylhydrazone (FCCP; 0.75 μmol/L), rotenone (1 μmol/L) and antimycin A (1 μmol/L)]. Coupling efficiency, as a measure of the fraction of basal mitochondrial OCR used for ATP synthesis, was assessed by the decrease in respiration on inhibiting ATP synthase with oligomycin. Cell respiratory control ratio is the ratio of the uncoupled OCR (rate after FCCP injection) to OCR in the presence of oligomycin. Spare respiratory capacity was measured by the ratio of the uncoupled OCR to the basal rate. Experiments were performed using at least 3 biological replicates and all values were internally normalized (33).

Murine CD2-Lmo2 T-ALL cell lines treated as described were resuspended in glucose-free RPMI-1640 medium containing 1% fetal calf serum (FCS) and 150 μmol/L 2-NBDG (Life Technologies), incubated for 10 minutes at 37°C, washed with ice-cold PBS and analyzed using flow cytometry. Apigenin (100 μmol/L, 30 minutes of treatment), an inhibitor of Glut1 expression, was used as a positive control.

A representative murine CD2-Lmo2–derived T-ALL cell line (03007) was treated with DMSO (0.6%) or CAL-130 (2.5 μmol/L; 10 hours) or CompE (1 μmol/L; 48 hours). Biological triplicates were performed with each experimental condition and gene expression profiling performed by Expression Analysis, Inc. using the Affymetrix Mouse Genome 430 array (#901570). Microarray data were then normalized using the Robust MultiChip Averaging (RMA) algorithm as implemented in the Bioconductor package Affy (34). A modified t test was used in Bioconductor Limma package to identify differentially expressed probe sets between each drug treatment group and control group (35). The P values from Limma analysis were adjusted by the Benjamini and Hochberg method to control false discovery rate (FDR) and genes were ranked to identify top 50 up- and 50 downregulated genes for each treatment condition (36).

Results are expressed as mean ± SEM for a minimum of 3 independent experiments. Statistical analysis was performed using either the Student unpaired t test or two-way ANOVA followed by the Bonferroni posttest. P values < 0.05 were considered significant. Further detailed statistical analysis of the microarray data and information on experimental methods can be found in Supplementary Materials available on the Molecular Cancer Therapeutics Web site.

Disease penetrance was 100% with a median survival rate of 227 days in CD2-Lmo2 transgenic mice (Fig. 1A). Tumor cells displayed high levels of both phosphorylated AKT (Thr308 and Ser473) and intracellular NOTCH1 (ICN) with activation of downstream targets (Fig. 1B); all tumors expressed the p110γ and p110δ catalytic domains of class I PI3Ks (Fig. 1C). Of note, the majority of Lmo2-driven T-ALL lacked expression of PTEN, acquired Notch1 mutations within the PEST domain (Supplementary Table S1) and had evidence of upregulation of the Notch1 target genes such as Deltex1 (∼4-fold; Fig. 1D). Moreover, the majority of circulating blasts had an immunophenotype characteristic of NOTCH1-induced T-ALL in mice: CD8 single positive or CD4/CD8 double positive with high expression of CD25 (Fig. 2A and B). Thus, the CD2-Lmo2 transgenic animal model of T-ALL provided an ideal system to determine whether the combined activities of PI3Kγ/δ regulate similar or distinct oncogenic programs typically associated with NOTCH1 activation.

To determine the reliance of Lmo2-driven T-ALL on the PI3K/AKT versus NOTCH1 pathway, we treated diseased animals with either the dual PI3Kγ/δ inhibitor CAL-130 (10 mg/kg every 8 hours) or the potent GSI DBZ (10 μmol/kg per day) for 7 days. Previously, we have shown that CAL-130 is highly selective for the p110γ and p110δ catalytic domains of class I PI3Ks by virtue of its ability to (i) yield an immunologic phenotype similar to that of mice lacking both p110 isoforms when administered to wild-type (WT) animals, and (ii) significantly reduce tumor burden and prolong survival in mice that develop PTEN-null T-ALL (19). Despite NOTCH1 activation, all mice had a rapid response to the dual PI3Kγ/δ inhibitor as evidenced by a reduction in circulating blasts and presence of apoptotic cells 4 days posttreatment (detected sub-G₀ population after PI staining; Fig. 2A). Administration of DBZ also reduced disease burden (Fig. 2B) and resulted in a similar prolongation in survival: median of 27 days for DBZ-treated versus 29 days for CAL-130–treated animals (Fig. 2C, P = 0.64).

In order to investigate the direct effects of PI3K/AKT and NOTCH1 inhibition on tumor cell proliferation and survival, we performed in vitro coculture of primary leukemic blasts from mice and human T-ALLs using MS5-DL1 stromal cells (19). Consistent with in vivo results, primary T-ALL cell viability was significantly reduced in the presence of either CAL-130 or the GSI CompE (Fig. 3A; Supplementary Fig. S1A and S1B; and Supplementary Table S1) while the viability of stromal cells was not affected by either compound.

We next evaluated the effects of these inhibitors alone or in combination using two representative T-ALL cell lines, designated 03007 and 03027, which were derived from diseased CD2-Lmo2 transgenic animals (23). Consistent with results obtained with primary tumors, treatment of either cell line with CAL-130, but not the p110δ-selective inhibitor IC87114, had a cytostatic effect and induced cell death within 24 hours based on PI staining (Fig. 3B and Supplementary Fig. S1C). The effects of CompE were similar albeit delayed (Fig. 3C and Supplementary Fig. S1D). To determine whether the combination of CAL-130 and CompE would be synergistic in nature, we used a high-throughput cell viability screen based on cellular ATP detection and the Bliss independence model to quantify the degree of synergy between the pathways (30). Excess over Bliss (EoB) was calculated as the difference between the observed effect and the predicted response; a 5% to 10% EoB response was considered as weak, 10% to 20% as moderate, and over 20% was considered as strong synergism. Results indicate that CAL-130 (2.5 μmol/L) in combination with CompE did not yield a synergistic effect in either cell line, suggesting redundancy in the NOTCH1 and PI3K/AKT pathways in Lmo2-driven T-ALL (Supplementary Fig. S2).

NOTCH1 directly regulates cMYC expression, which governs an oncogenic program supporting T-ALL cell growth and survival (7, 8). Accordingly, CompE blocked NOTCH1 activation in Lmo2-driven T-ALL cell lines and also reduced protein levels of cMYC, albeit at distinct time points (24 hours and 48 hours, respectively; Fig. 3D and Supplementary Fig. S1E). In contrast to the delayed effects associated with GSI, dual inhibition of PI3Kγ/δ activity abolished cMYC protein but not transcript expression within 12 hours of treatment (Fig. 3D and E). Importantly, CAL-130 was still able to inhibit activation of AKT (phosphorylation of Thr308 and Ser473) and its downstream targets such as glycogen synthase kinase-3β (GSK3β), the eukaryotic translation initiation factor 4E-binding protein 1 (4E-BP1), and the ribosomal protein p70 S6 kinase (p70 S6K; ref. 19), which play a major role in cell-cycle progression and tumor-cell proliferation (Fig. 3F and Supplementary Fig. S1F). Of note, the blockade of PI3Kγ/δ activity or GSI abrogated the expression of the antiapoptotic effectors MCL-1 and phospho-BAD (S136) and lead to upregulation of proapoptotic BIM (Supplementary Fig. S3A and S3B), with >50% of CAL-130-treated cells becoming apoptotic by 24 hours in contrast to those treated with CompE as determined by flow cytometry (sub-G₁; Supplementary Fig. S3C–S3F). In contrast, the majority of effects of CompE were not observed until 48 hours.

In order to confirm that loss of cMYC protein expression is essential for CAL-130 or CompE-induced tumor cell death, we generated a Lmo2-driven murine T-ALL line with constitutive cMYC expression. Indeed, overexpression of this oncogene permitted continued tumor cell growth despite incubation with either inhibitor. Moreover, cMYC protein levels remained constant in contrast to cells expressing vector alone (Fig. 4A–F). To further demonstrate that downregulation of cMYC is essential for the antitumorigenic effects of CAL-130 and that it may exert this effect independent of NOTCH1 activation, we crossed Gt(ROSA)26Sor^{tm13(CAG-MYC, -CD2*)Rsky}/J mice that overexpress cMyc with Lck-cre/Pten^(fl/fl) mice that lack Pten. Previously, we have shown that T-ALL that develops in PTEN-null mice is exquisitely sensitive to the dual PI3Kγ/δ inhibition (19). In contrast, T-ALL with enforced cMYC expression was resistant to CAL-130 treatment (10 mg/kg every 8 hours for 7 days) and showed continued expansion of circulating blasts (Ki67⁺/Thy1.2⁺ population) and lack of PI-staining (Fig. 4G).

To better understand the genetic basis by which PI3Kγ/δ and NOTCH1 regulate tumor progression in Lmo2-driven T-ALL, we generated gene expression signatures based on disease phenotype and drug treatment. Global gene expression profiles were obtained in triplicate from tumor cells cultured in the presence of CAL-130 (2.5 μmol/L), CompE (1 μmol/L), or DMSO. Drug-treated cells were harvested at time points known to affect cMYC expression (12 hours and 48 hours, respectively). A Venn diagram was created to illustrate the overlap in genes altered by either PI3Kγ/δ blockade (red circle) or GSI (green circle) using an FDR of 0.0005 as cutoff (Fig. 5A). Of note, 10-fold more genes were affected following PI3Kγ/δ blockade and included the majority (∼62%) of genes altered by GSI monotherapy, a highly significant finding (Fisher's exact test, P < 2.2 × 10⁻¹⁶). Heat maps were also generated from complete linkage hierarchical cluster analyses of statistically significant changes in gene expression associated with drug treatment versus vehicle control (Fig. 5B). In the case of PI3Kγ/δ blockade, downregulated genes also included those involved in cMYC-regulated cellular metabolism and/or metabolic reprogramming such as the glycine/serine biosynthetic pathway (e.g., Phgdh, FDR = 6 × 10⁻⁸; Shmt1, FDR = 8.2 × 10⁻⁸; Shmt2, FDR = 6.2 × 10⁻⁸), folate metabolism (e.g., Mthfd1, FDR = 6.04 × 10⁻⁸), glycolysis (e.g., Hk-2, FDR = 5.4 × 10⁻⁸), nucleotide metabolism (Ak4, FDR = 2.3 × 10⁻⁸), and mitochondrial proline metabolism and synthesis (Pycr1, FDR = 3.1 × 10⁻⁷; 37–41). As expected, gene expression profiling of CompE-treated cells demonstrated significant downregulation of canonical NOTCH1 target genes including Hes1 (FDR = 1.1 × 10⁻⁷), Hey1 (FDR = 2.6 × 10⁻⁸), and Deltex1 (FDR = 4.4 × 10⁻⁶), as well as modulating the expression of genes involved in glycolysis (e.g., Slc2a3, FDR = 4.14 × 10⁻⁶; Hk2, FDR = 1.68 × 10⁻⁵) and nucleotide metabolism (e.g., Ak4, FDR = 1.66 × 10⁻³; ref. 42).

We next defined a murine T-ALL disease signature by ranking all genes based on their differential expression in T-ALL samples compared with wild-type murine thymocytes. To determine the contribution of PI3Kγ/δ and NOTCH1 to the overall disease signature, we performed a gene set enrichment analysis (GSEA) of the 200 most transcriptionally activated and repressed genes following treatment with either CAL-130 or CompE in differentially expressed genes. Although both drugs significantly inverted the T-ALL disease signature, underscoring the importance of these signaling pathways in maintaining the leukemic phenotype (Fig. 5C), the enrichment was much more significant with CAL-130 [normalized enrichment score (NES) of −5.8 (P < 0.0001) for CAL-130 vs. NES of −2.6 (P = 0.006) for CompE]. To show that treatment of murine cells is highly consistent with treatment of human cells, we assessed whether the top 200 most differentially expressed genes following CompE treatment of murine T-ALL cells were significantly enriched in human-to-mouse mapped genes differentially expressed following CompE treatment of human cells (Fig. 5D). To accomplish this goal, we integrated data from seven distinct dataset representing human T-ALL cells treatment with CompE (7). Analysis confirms that results from drug treatment of Lmo2-driven murine T-ALL are relevant to human T-ALL.

Previously, we have shown the existence of Master Regulator (MR) proteins responsible for causally (i.e., physically) regulating the transcriptional state of a cancer cell (43). As such, their activated and repressed targets are highly enriched in genes that are over- and underexpressed in tumor cells, respectively, and can thus be efficiently identified by analyzing a regulatory network that represents the transcriptional targets of every protein using the Virtual Proteomics by Enriched Regulon (VIPER) algorithm (44). VIPER is an extension of the Master Regulator Inference algorithm (MARINa; ref. 45), which supports analysis of MR proteins on an individual sample basis. Such an analysis has helped elucidate novel mechanisms of tumorigenesis, progression, and drug sensitivity in glioma, leukemia, lymphoma, prostate, and breast cancer, among others. As MR protein activity efficiently recapitulates overall tumor state and drug response, we first identified key T-ALL–specific MR proteins and then assessed whether compound treatment could effectively reverse their activity. MR proteins were inferred by VIPER analysis of a previously assembled, genome-wide T-ALL–specific regulatory network (interactome) using a signature of a Lmo2-driven murine T-ALL cell line (7) compared with wild-type murine thymocytes. This interactome was previously used to successfully elucidate the Mechanism of Action (MoA) of small molecule compounds targeting human T-ALL pathways (7). The VIPER algorithm was then used to compute the activity of each of the top 40 T-ALL MR proteins following drug perturbation. This was accomplished by computing the enrichment of their regulons (i.e., the set of their transcriptional targets), as represented in the interactome, in genes differentially expressed following drug treatment (drug signature). Although the gene signatures for the two drugs overlapped significantly (P < 6 × 10⁻¹³, Fisher exact test), ∼3-fold more T-ALL MRs were reverted by CAL-130 compared with GSI CompE (Fig. 5E and Supplementary Table S2; NES < −1.96 and P < 0.05). Of note, Lyl1, Tal1, and Hhex were not affected by CAL-130 according to our analysis. We also performed GSEA of ∼4,500 annotated gene sets collected from Molecular Signatures Database (MSigDB) on the CAL-130 and CompE-induced gene signatures to study the perturbed pathways (46). Although CAL-130 altered many of the same genes as compared with CompE, it had stronger effects on pathways regulated by PML (originally identified as a tumor suppressor in acute promyelocytic leukaemia; NES 2.57 and P < 0.01 vs. NES 0.82 and P = 0.41), EDF1 (NES −3.58 and P < 0.005 vs. NES −0.53 and P = 0.6) and IRF9 (NES 3.18 and P < 0.005 vs. NES 0.07 and P = 0.944). Thus, CAL-130 treatment recapitulated majority of mechanisms of action of CompE, while reversing the activity of a large number of additional master regulators and pathways associated with the T-ALL tumor cell state.

To access drug effects on cMYC activity, we analyzed published microarrays of human T-ALL (47) and used GSEA to compute the enrichment of cMYC-regulated genes differentially expressed following CAL-130 or CompE treatment (Fig. 5F and Supplementary Table S3). Although genes regulated by cMYC were observed in both drug treatment signatures (P < 10⁻³), CAL-130 treatment had a greater effect in terms of reversing the global expression of cMYC targets (NES of −6.23), compared with CompE (NES of −5.10). Of note, ∼74% of genes regulated by cMYC were in the top 1,000 genes in the CAL-130-treated gene expression signature as compared with ∼42% in the CompE-treated gene signature (Supplementary Table S3).

To better assess the functional categories represented by the set of genes altered in response to the inhibitors, we performed pathway analysis using DAVID (48). A highly conservative threshold (P < 1 × 10⁻⁵) was used to identify the top downregulated genes in CAL-130 or CompE-treated Lmo2-driven T-ALL cells. In the case of PI3Kγ/δ blockade, there was a significant enrichment in functional categories related to cellular metabolism and, most interestingly, pathways that regulate mitochondrial activity (Fig. 6A and B). CompE treatment not only altered similar metabolic pathways including those related to carbohydrate, nucleotide, pyruvate, and TCA (Fig. 6C and Supplementary Fig. S4A and S4B), but also reduce glucose uptake in T-ALL tumor cells as observed for CAL-130 (Fig. 6D and Supplementary Fig. S5A); it did not, however, appear to affect genes that directly regulate mitochondrial function (7, 42, 49).

As aforementioned, the blockade of PI3Kγ/δ activity preferentially altered genes that affect mitochondrial activity, such as those involved in the electron transport chain and oxidative phosphorylation (Fig. 7A and Supplementary Fig. S4C). Consistent with this observation, CAL-130 but not CompE significantly reduced in situ OCRs, ATP levels and spare respiratory capacity, the latter indicating that PI3Kγ/δ inhibition limits the ability of tumor cell lines and primary tumor cells to respond to stress or metabolic challenge (Fig. 7B–E and Supplementary Fig. S5B–S5H). Further evidence to support these observations was provided by directly studying mitochondria isolated from CAL-130–treated tumor cells. Indeed, a robust inhibition of resting and phosphorylating respiration rates in purified mitochondria were observed suggesting that the alterations in genes encoding the respiratory chain complexes I (e.g., Ndufa12), II (e.g., Sdhb), and III (e.g., Cyc1) were functionally significant (Figs. 6B and 7F). Our experiments also suggested dysregulation of complex IV as evidenced by the marked depression of mitochondrial respiration with TMPD-ascorbate as substrate (50). Of note, CAL-130 treatment affected the expression of mitochondrial protein serine hydroxymethyl transferase (SHMT2; Fig. 7F, inset). SHMT2 is a key enzyme in the serine/glycine synthesis pathway that plays a critical role in the metabolic reprogramming in cancer and is a transcriptional target of cMYC (51). Thus, PI3Kγ/δ inhibition not only affects major metabolic pathways that support tumor homeostasis but also affects the bioenergetics of T-ALL cells.

Cumulative evidence indicates that the PI3K/AKT and NOTCH1 pathways not only contribute to the malignant transformation of T-cell progenitors but also play a major role in regulating the growth, metabolism, and survival of T-ALL. Here, we explored the interdependence of these oncogenic pathways in promoting tumor-cell maintenance, which has profound implications for the development of a rational therapeutic strategy to treat this aggressive hematologic malignancy. It is for this reason that we chose the Lmo2-driven transgenic mouse model of T-ALL that is characterized by the spontaneous accumulation of several oncogenic mutations in T-cell precursors, resulting in the coactivation of the PI3K/AKT and NOTCH pathways. In the context of this model, we now provide evidence that (i) although there is no preferential addiction of tumor cells to either pathway, PI3Kγ/δ blockade appears to exert a more rapid and potent effect on tumor cells than GSI, and (ii) there is a lack of synergy between PI3K/AKT and NOTCH1 in supporting tumor-cell proliferation and survival.

Previous studies have demonstrated that NOTCH1 can indirectly impact on the activity of the PI3K/AKT pathway by controlling a transcriptional network that regulates PTEN expression in leukemic T cells (15). Indeed, this mechanism is believed to be responsible for bypassing the requirement for aberrant NOTCH1 signaling that normally supports tumor-cell growth and metabolism. In contrast, our studies indicate that NOTCH1 does not regulate PI3Kγ/δ activity or PTEN expression, suggesting that these pathways can function independently of one another. This is exemplified by the lack of correlation between activation status of either pathway and the ability of primary mouse or human T-ALL cells to respond to treatment with CompE or CAL-130. That said, they do appear to control similar oncogenic networks that are indispensable for tumor-cell maintenance. Results indicate a genetic underpinning for these observations as evidenced by the considerable overlap in genes regulated by both signaling pathways, including cMYC. The latter was confirmed by the ability of constitutively expressed cMYC to reverse the effects of either inhibitor. Collectively, these data demonstrate that PI3Kγ/δ can support a NOTCH1-independent pathway of cMYC activation in T-ALL, which may account for similar reduction in cell growth and survival but lack of synergistic effect observed in response to CAL-130 and/or CompE (52).

Why then does PI3Kγ/δ blockade appear to exert a more rapid and potent antileukemic effect than inhibition of NOTCH1 by GSI? Gene expression profiling and pathway analysis based on DNA microarrays of drug-treated Lmo2-driven T-ALL provided possible answers. On a global scale, CAL-130 treatment had a greater effect on altering the disease signature compared with CompE, as demonstrated by the reversal of top master regulators (e.g., HNRNPAB, Tp53, PA2G4, ASH2L, and SNAPC5), and affect pathways (e.g., PML, EDF1, and IRF9) associated with the CD2-Lmo2 T-ALL tumor cell state. This was consistent with the ability of CAL-130 to downregulate a 10-fold greater number of genes at an early time point. Interestingly, the majority of altered genes are known to support cellular metabolism and/or the metabolic reprogramming implicated in tumor progression, including those involved in serine/glycine metabolism, polyamine synthesis, and glycolysis. Indeed, the ability of cancer cells to alter their metabolism is essential to support the increased energy demands associated with rapid tumor growth. Germane to this study is the ability of the PI3K/AKT pathway to act as a master regulator of aerobic glycolysis and cellular biosynthesis that critically relies on mitochondrial bioenergetics (53, 54). In fact, the shift from oxidative to more glycolytic metabolism is one of the most common changes found in tumor cells. Although our results clearly demonstrated that PI3Kγ/δ blockade or GSI significantly reduce glucose uptake in leukemic cells, only the former was able to rapidly alter mitochondrial function. Indeed, we noted that genes encoding proteins involved in oxidative phosphorylation were uniquely susceptible to CAL-130, including those essential for the mitochondrial respiratory chain complex formation (FDR < 0.005). Consistent with this observation was the robust inhibition of mitochondrial resting and phosphorylating respirations rates. This finding is of particular interest as enhancement in the bioenergetics properties of tumor cells is thought to be a prerequisite for survival.

Overall, our results suggest the existence of a regulatory circuit linking NOTCH1 and PI3K/AKT signaling, with cMYC serving as a crucial node of convergence that intertwines but does preferentially support the activity of either pathway. Moreover, we provide new insights into the mechanism(s) by which the combined activities of PI3Kγ/δ contribute to the maintenance of T-ALL and how pharmacologic blockade of these two isoforms exerts a more rapid and pronounced effect on tumor-cell proliferation than GSI, yet can yield similar rates of survival of diseased animals. This involves regulating transcriptional pathways and genes typically associated with NOTCH1, as well as distinct metabolic and bioenergetic pathways essential for tumor progression and survival. These results provide a rationale for the combined targeting of p110γ and p110δ catalytic domains of class I PI3Ks in T-ALL where NOTCH and/or PI3K/AKT pathways are dysregulated.

A. Califano is cofounder and chief scientific advisor of, and has ownership interest (including patents) in, DarwinHealth, Inc. No potential conflicts of interest were disclosed by the other authors.

Conception and design: T. Diacovo

Development of methodology: U.P. Davé, Z. Niatsetskaya, V. Ten, A. Califano, T. Diacovo

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): E. Efimenko, I.V. Lebedeva, M.J. Sanchez-Quintero, D. Diolaiti, A.L. Kung, R. Realubit, Z. Niatsetskaya, V. Ten, C. Karan, T. Diacovo

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): E. Efimenko, U.P. Davé, I.V. Lebedeva, Y. Shen, A.L. Kung, J. Chen, R. Realubit, Z. Niatsetskaya, V. Ten, C. Karan, Xi Chen, A. Califano, T. Diacovo

Writing, review, and/or revision of the manuscript: U.P. Davé, B.J. Lannutti, C. Karan, A. Califano, T. Diacovo

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): B.J. Lannutti, R. Realubit, T. Diacovo

Study supervision: T. Diacovo

Other (generated engineered cells): D. Diolaiti

We thank Juan Carlos Zuniga-Pflucker for providing the MS5-DL1 cells, Andrew Nicholson, Anastasiya Egorova, and Anna Binyamin for genotyping and animal husbandry, and St. Jude Children's Research Hospital for providing primary T-ALL cells from patients.

This research is supported by National Cancer Institute grant R01CA169162-01 (T.G. Diacovo), Department of Veterans Affairs (I01BX001799, U.P. Davé), and the Vanderbilt Ingram Cancer Center (P30 CA68485; U.P. Davé).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.