PI3Kα inhibitors have shown clinical activity in PIK3CA-mutated estrogen receptor-positive (ER+) patients with breast cancer. Using whole genome CRISPR/Cas9 sgRNA knockout screens, we identified and validated several negative regulators of mTORC1 whose loss confers resistance to PI3Kα inhibition. Among the top candidates were TSC1, TSC2, TBC1D7, AKT1S1, STK11, MARK2, PDE7A, DEPDC5, NPRL2, NPRL3, C12orf66, SZT2, and ITFG2. Loss of these genes invariably results in sustained mTOR signaling under pharmacologic inhibition of the PI3K–AKT pathway. Moreover, resistance could be prevented or overcome by mTOR inhibition, confirming the causative role of sustained mTOR activity in limiting the sensitivity to PI3Kα inhibition. Cumulatively, genomic alterations affecting these genes are identified in about 15% of PIK3CA-mutated breast tumors and appear to be mutually exclusive. This study improves our understanding of the role of mTOR signaling restoration in leading to resistance to PI3Kα inhibition and proposes therapeutic strategies to prevent or revert this resistance.
These findings show that genetic lesions of multiple negative regulators of mTORC1 could limit the efficacy of PI3Kα inhibitors in breast cancer, which may guide patient selection strategies for future clinical trials.
The PI3K pathway regulates many cellular processes, including cell growth, motility, survival, apoptosis, and metabolism. Class IA PI3Ks consist of heterodimers of the p110 catalytic subunit and the p85 regulatory subunit. The p110 catalytic subunit isoforms are encoded by the PIK3CA, PIK3CB, and PIK3CD genes (p110α, p110β, and p110δ, respectively). Activation of PI3K catalyzes the phosphorylation of PI(4,5)P2 (PIP2) to PI(3,4,5)P3 (PIP3), which triggers downstream signaling and is tightly regulated by PTEN that converts PIP3 back to PIP2. Accumulation of PIP3 at the plasma membrane leads to the recruitment of the pleckstrin homology domain-containing proteins PDK1 and AKT. PDK1 phosphorylates AKT, which further activates mTORC1 (1). Mechanisms that hyperactivate this pathway include activating mutations of PIK3CA and PIK3CB, and AKT1, and loss-of-function alterations of PTEN, PIK3R1 (coding for p85), TSC1, TSC2, STK11, and other modulators of the mTOR complex (2–4).
As the most frequent genomic alterations in estrogen receptor-positive (ER+) breast cancer, inhibitors targeting PI3K p110α (PI3Kα) have shown promising clinical activity in patients with PIK3CA-mutated tumors, particularly when given in combination with endocrine therapy (5). This led to FDA approval of alpelisib in combination with the ER degrader fulvestrant in patients with PIK3CA-mutated ER+ advanced breast cancer. However, clinical response to PI3K inhibitors is variable, with some patients achieving prolonged clinical benefit and others showing intrinsic resistance to therapy. Although PIK3CA mutations are a confirmed biomarker of sensitivity to specific PI3Kα inhibitors (6, 7), it is crucial to understand the molecular mechanisms underlying resistance to these agents and provide valid therapeutic strategies to overcome or delay such resistance.
We have previously shown that sustained activity of mTORC1 upon PI3K inhibition promotes resistance to these therapeutic agents in preclinical models and clinical samples (8). Known mechanisms to date include alternative phosphorylation of TSC2 by the kinase SGK1 (9), upregulation of Proviral Insertion site in Murine leukemia virus (PIM) kinases (10), or re-activation of the PI3K pathway through insulin growth factor receptor (IGF1R)/p110β (11, 12). Other reports indicate that loss of PTEN (13, 14), overexpression of FGFR1 (15) or ER (16, 17) may also limit sensitivity to PI3Kα inhibitors. However, these reported mechanisms only partially explain why some breast cancer patients are or become resistant to PI3Kα inhibitors. In this work, we aimed to identify and characterize potential genetic lesions that limit the sensitivity to PI3Kα inhibitors by means of unbiased genome wide CRISPR screenings.
Materials and Methods
Cell lines, chemical reagents, and plasmids
MCF7 (ATCC catalog no. HTB-22, RRID: CVCL_0031) and T47D (ATCC cat# HTB-133, RRID: CVCL_0553) cells were cultured in DMEM/F-12 (Corning) or RPMI1640 (Corning) respectively, supplemented with 10% FBS and penicillin/streptomycin under normal oxygen conditions (5% CO2, 37°C). 293T cells (ATCC catalog no. CRL-3216, RRID: CVCL_0063) were cultured in DMEM (Corning) supplemented with 10% FBS under normal oxygen conditions (5% CO2, 37°C). Cells within 1 month in culture were used for experiments. T47D alpelisib-tolerant or DMSO-tolerant cells were generated by chronic treatment with 1 μmol/L alpelisib or DMSO for 2 months. All cells were tested for Mycoplasma with Mycoplasma Detection Kit (Lonza).
BYL719, GDC0032, GDC0068, MK2206, AZD5363, and everolimus were purchased from Selleck Chemicals or MedChemExpress (MCE).
Cas9 expression construct (SVC9B-PS) and the whole genome sgRNA library (KOHGW-M123-P) were ordered from Cellecta. Individual sgRNAs (Supplementary Table S1) were either designed using Benchling (http://www.benchling.com) or from the library and were ordered from Cellecta. Constructs expressing RagA (RRID: Addgene_73031), Rag D (RRID: Addgene_19316), RagD S77 L (RRID: Addgene_19317), and RagDQ121L (RRID: Addgene_19318) were ordered from Addgene. RagAT21L and RagAQ66L were generated using Q5 site-directed mutagenesis kit from NEB (E00554S) with primers listed in Supplementary Table S1 following the manufacturer's instruction and verified by DNA Sanger sequencing.
To generate the lentivirus of Cas9, the sgRNA library, individual sgRNAs, or RagA and Rag D mutants, 293T cells were seeded into 15 cm dishes 16 hours before transfection. For each dish, 3.5 μg pMD2.G envelope vector, 7 μg packaging vector pCMV-dR8.2, and 10.5 μg plasmids of interest were added to 3 mL jetPRIME buffer (Polyplus), and then 42 μL jetPRIME transfection reagent was added. The mixture was incubated at room temperature for 10 minutes before being added to the cells. Medium was refreshed 6 hours posttransfection and the supernatant containing lentivirus was collected 48 hours posttransfection.
Stable cell lines generation
To obtain MCF7-Cas9 and T47D-Cas9 cells, MCF7 and T47D were transduced with Cas9 lentivirus produced as above with 8 μg/mL of polybrene for 24 hours and then selected with 10 μg/mL blasticidin (GIBCO) for 3 days.
To obtain MCF7 and T47D cells with the loss of target genes, MCF7-Cas9 or T47D cells were transduced with lentivirus of individual sgRNAs with 8 μg/mL of polybrene for 24 hours and then selected with 2 μg/mL puromycin (GIBCO) for 3 days.
RagA or RagD WT and mutant overexpressing cells were obtained by transducing parental or DEPDC5 deficient MCF7 and T47D cells with the corresponding lentivirus and selected with 2 μg/mL puromycin (GIBCO) for 3 days.
Whole genome CRISPR-Cas9 knockout screen
The pooled sgRNA library comprises 150,000 individual sgRNAs targeting over 19,000 human protein-coding genes with 8 sgRNAs per gene. The sgRNA lentivirus library was generated as described above. Lentivirus titration was determined by a RT-qPCR–based method according the manufacturer's protocol (Lenti-X qRT-PCR Titration Kit, catalog no. 631235; Clontech).
Next, lentivirus of the sgRNA library were added to 500 million MCF7- or T47D- Cas9 cells at a MOI ∼0.3 to reach 1,000× coverage of the library. The library-transduced cells were subjected to either DMSO, alpelisib (1 μmol/L) or taselisib (60 nmol/L) treatment for 20 days. Surviving cells were pooled and genomic DNA was extracted using the Gentra Puregene Kit according to the manufacturer's protocol (Qiagen). Roughly 1,000 μg genomic DNA from each group was amplified by PCR to reach 600× coverage of the library and the sgRNA sequences were retrieved by sequencing the PCR products according to the manufacturer's protocol (Cellecta, catalog no. LNGS-120).
Gene set enrichment analysis
Four gene lists of T47D-alpelisib_enriched, T47D-alpelisib_dropped out, MCF7-taselisib_enriched, and MCF7-taselisib_dropped out were selected based on the following criteria: top 2.5% enriched or dropped-out sgRNAs (3745 sgRNAs) and at least three sgRNAs for each gene shown up. We conducted over-representation analysis for these four gene lists using hallmark gene set via MSigDB website and extracted the significant pathway with FDR <0.05. FDR was calculated in hypergeometric test after the correction for multiple hypothesis testing based on Benjamini and Hochberg.
Briefly, cell lysates were prepared with RIPA lysis buffer supplemented with protease inhibitor (Roche) and phosphatase inhibitor (Thermo Fisher Scientific). Protein concentration was determined by the BCA assay (Pierce) according to the manufacturer's protocol. Total proteins with sample buffer were separated by SDS-PAGE on 4% to 12% Bis-Tris gradient gels (Invitrogen) and transferred to NC membrane (Bio-Rad). SuperSignal west Pico Plus or Femto Maximum chemiluminescent detection reagents were used (Thermo Fisher Scientific).
Primary antibodies used in this study include: Rabbit anti-Vinculin (Cell Signaling Technology catalog no. 13901, RRID: AB_2728768), Rabbit anti-TSC1 (Cell Signaling Technology catalog no. 4906, RRID: AB_220979), Rabbit anti-TSC2 (Cell Signaling Technology catalog no. 4308, RRID: AB_10547134), Rabbit anti-TBC1D7 (Cell Signaling Technology catalog no. 14949, RRID: AB_2749838), Rabbit anti-AKT1S1 (Cell Signaling Technology catalog no. 2691, RRID: AB_2225033), Rabbit anti-DEPDC5 (Sigma-Aldrich catalog no. SAB1302644, RRID: AB_2889910), Rabbit anti-NPRL2 (Cell Signaling Technology catalog no. 37344, RRID: AB_2799113), Rabbit anti-NPRL3 (Sigma-Aldrich catalog no. HPA011741, RRID: AB_1845577), Rabbit anti-MARK2 (Abcam catalog no. ab133724, RRID: AB_2889911), Rabbit anti-PDE7A (Abcam catalog no. ab154857, RRID: AB_2889912), Rabbit anti-STK11 (Abcam catalog no. ab199970, RRID: AB_2889913), anti-pAKT(S473) (Cell Signaling Technology catalog no. 4060, RRID: AB_2315049), anti-pS6K (T389) (Cell Signaling Technology catalog no. 9205, RRID: AB_2734746), anti-pS6(S235/236) (Cell Signaling Technology catalog no. 2211, RRID: AB_331679), anti-p4E-BP1(S65) (Cell Signaling Technology catalog no. 9459, RRID: AB_330985).
MCF7 WT and DEPDC5 knockout (KO) cells or their derivatives expression empty vector, RagAT21L or RagAT21L/RagDQ121L were seeded on coverslips (Thermo Fisher Scientific, 12CIR-1). After 24 hours, cells were treated with different conditions, including amino acid-starved, and followed by 5 minutes amino acid stimulation, 4 hours treatment of DMSO, 1 μmol/L alpelisib, or 20 nmol/L everolimus. Cells were then rinsed once with PBS and fixed in 4% paraformaldehyde for 10 minutes at room temperature and permeabilized with 0.2% Triton X-100 in PBS for 10 minutes. After two washes in PBS, cells were blocked with PBG (0.5% BSA and 0.2% gelatin from cold water fish skin in PBS) blocking solution for 1 hour, incubated with primary antibodies against mTORC1 (1:100, Cell Signaling Technology cat# 2983, RRID: AB_2105622 and Santa Cruz Biotechnology cat# sc-18822, RRID: AB_626858) diluted in PBG for 1 hour and incubated with goat anti-mouse Alexa Fluor-647 or goat anti-rabbit Alexa Fluor-488 secondary antibodies (1:500 in PBG) for 45 minutess in a dark humid chamber. Cells were washed for 5 minutess three times with PBG blocking solution Cells between each incubation. After staining with 0.2 μg/mL DAPI for 1 to 2 minutes, coverslips were washed in PBS twice and rinsed once in deionized water and finally mounted with Mowiol on glass slides and let dry in the dark overnight. Two-μmol/L-thick z-stack images were taken on a Zeiss LSM880 microscope with a 63× oil lens. At least 6 different fields from two independent experiments were scanned for each sample. Z-projections of the stacks were then analyzed using Fiji software, Coloc2 plugin. Approximately 50 or 100 cells per sample were analyzed.
Competitive cellular growth assay
MCF7 and T47D cells stably transduced with sgNT-GFP, sgNT-RFP, or sgRNA-RFP for individual target genes were generated as described above. sgNT-GFP cells were mixed with sgNT-RFP or sgRNA-RFP cells (GFP+ and RFP+) at 1:1 ratio and RFP+/GFP+ ratios of each sample were measured as T0. sgNT-GFP cells and sgNT-RFP cells were used as control. A total of 30,000 mixed MCF7 cells or 40,000 mixed T47D cells per well were seeded in six-well plates. These cells were subjected to DMSO, 1 μmol/L alpelisib (BYL719) or 60 nmol/L taselisib (GDC-0032) for 6 days. RFP+/GFP+ ratios of each sample after treatment were measured by flow cytometry as T7. The change of RFP+/GFP+ ratio was scored as the impact of candidate gene KO on cell growth or drug response compared with sgNT control cells. The FlowJo Software was used for FACS data analysis.
Clonogenic-based cellular growth assay
A total of 40,000 MCF7 cells or 50,000 T47D cells per well were seeded into 12-well plates. Cells were treated with DMSO, 1 μmol/L alpelisib, 60 nmol/L taselisib, 2 μmol/L GDC0068, 2 μmol/L MK2206, 2 μmol/L AZD5363, or 20 nmol/L everolimus for 6 days. On day 7, cells were fixed with 3.7% formaldehyde and stained with 0.1% crystal violet.
CellTiter-Glo-based cell viability assay
A total of 2,000 MCF7 cells or 4,000 T47D cells per well were seeded into 96-well plates. Cells were treated with various doses of DMSO, alpelisib, taselisib, or everolimus for 4 days. Cell proliferation was determined by CellTiter-Glo assay (Promega) following the manufacturer's protocol.
In vivo tumor xenograft
All mouse experiments were approved by Research Animal Resources and Compliance and Institutional Animal Care and Use Committee at MSKCC. 0.72 mg/90d-release oestrogen pellets were transplanted into 6-week-old female NSG mice 3 days prior to the tumor cell transplantation. Ten million T47D cells transduced with sgNT or sgTSC1 per mouse were orthotopically transplanted. When xenografts reach 100 mm3, mice were randomized to the alpelisib (25 mg/kg orally every day, Monday to Friday) or untreated control groups. The tumor sizes were measured twice a week across the experiment.
The Cancer Genome Atlas and MSK datasets analysis
Datasets of The Cancer Genome Atlas (TCGA) breast invasive carcinoma (PanCancer Atlas) and Memorial Sloan Kettering (MSK) breast cancer (18) were available in the cBioPortal (http://cbioportal.org/). Genomic alterations of PIK3CA and the validated candidate genes in ER+ patients with breast cancer within these two datasets were analyzed and visualized by using the cBioPortal following the tutorials (19, 20). A total of 294 PIK3CA-mutated ER+ breast cancer samples from TCGA and 443 PIK3CA-mutated ER+ breast cancer samples from MSK-IMPACT dataset were analyzed.
Mutation VCF files from Bertucci and colleagues (21) were retrieved and annotated using the list of hotspots from Cancer Hotspots (cancerhotspots.org; ref. 22). A total of 129 cases harboring hotspot PIK3CA mutations were selected and mutations of the validated candidate genes were shown. Mutual exclusivity was calculated using WExT, a weighted exact test for mutual exclusivity (23).
Quantification and statistical analysis
All statistical analyses are shown in the appropriate method and figure legend.
Genome-wide CRISPR KO screens identify regulators of PI3Kα inhibitor response
To investigate possible mediators of response to PI3Kα inhibitors, we have conducted genome-wide CRISPR/Cas9 sgRNA KO screens in PIK3CA-mutated ER+ MCF7 and T47D breast cancer cell lines, which are intrinsically sensitive to PI3Kα inhibition. The library-transduced cells were treated with either DMSO or a PI3Kα inhibitor (alpelisib or taselisib) for 20 days. The genomic DNA of each group was extracted, and the sgRNA sequences were amplified by PCR and retrieved by next-generation sequencing (NGS; Supplementary Fig. S1A). To systematically understand possible resistance mechanisms, we performed gene set enrichment analysis of our screen results using hallmark gene set and found that several pathways were significantly enriched in both cell lines (Supplementary Table S2), such as MYC targets, mTORC1 signaling, and PI3K–AKT–MTOR signaling.
To identify top candidate genes, the loss of which can cause resistance to PI3Kα inhibition, we used the following criteria: (i) the top 2.5% sgRNAs (3,745 of 150,000 sgRNAs) enriched in the PI3Kα inhibitor group compared with the DMSO group (Supplementary Tables S3 and S4), (ii) each sgRNA had at least 150 reads in the untreated group (600 reads/sgRNA on average), and (iii) at least four of eight sgRNAs targeting the same gene were enriched. The resulting lists of top candidates are shown in Supplementary Tables S3 and S4.
Consistent with previous observations (12–14), PTEN is one of the top candidates mediating resistance to PI3Kα inhibition in both cell lines. Other candidates shared between MCF7 and T47D cells include TSC1, TSC2, DEPDC5, and NPRL2 (Fig. 1A and B). TSC1 and TSC2 are subunits of the TSC1–TSC2 complex, which integrates environmental signals such as stress, energy status, and growth factors in mammals into TORC1 signaling. This complex is a critical negative regulator of mTORC1 by acting as a GTPase-activating protein (GAP) for the small GTPase RHEB (24, 25). DEPDC5, NPRL2, and NPRL3 are components of the GATOR1 complex, which negatively regulates mTORC1 activity by acting as the GAP for RagA/B GTPases in the amino acids-sensing pathway (26). We further looked at other candidates involved in the mTORC1 pathway in either MCF7 or T47D cells. We found that STK11 (also known as LKB1), MARK2, and PDE7A in MCF7 cells, all potentially involved in the energy-sensing AMPK pathway (27–29). PRAS40, encoded by AKT1S1, an mTORC1 binding partner that mediates AKT signaling to mTOR (30, 31), is another candidate in MCF7 cells. TBC1D7, a candidate found in T47D cells, is a subunit of the TSC1–TSC2 complex and maintains the association of the complex (32). NPRL3, the other component of the GATOR1 complex, and the KICSTOR complex components SZT2 and ITFG2, were also identified in T47D cells. The KICSTOR complex is required for the localization of the GATOR1 complex on the lysosome surface (Fig. 1A and B; refs. 33, 34). The other component of the KICSTOR complex C12orf66 was also included even though only three individual sgRNAs targeting C12orf66 were highly enriched in T47D cells (Supplementary Table S2).
To validate these findings, we generated MCF7 and T47D stable cell lines deficient for each of these candidates. We selected two sgRNAs with the best knockout efficiencies (Supplementary Fig. S1B) and investigated whether knocking out each candidate gene would affect cell proliferation or drug response to PI3Kα inhibitors in competitive growth fitness assays (Supplementary Fig. S1C). Our data show that KO of each candidate gene did not affect per se cell proliferation in MCF7 or T47D cells, with the exceptions of TSC1 and TSC2 KO that significantly decreased cell growth in MCF7 cells (Supplementary Figs. S1D and S1E). However, consistent with our screening data, we observed that loss of the shared candidates, the TSC1/TSC2 complex, the GATOR1 complex and the positive control PTEN, had the most dramatic effects on limiting the sensitivity to alpelisib and taselisib in both MCF7 and T47D cells (Fig. 1C and D; Supplementary Figs. S2A and S2B). Most candidates were validated in both cell lines with significant effects in limiting the sensitivity to alpelisib or taselisib, with the exception of MARK2 in T47D cells or C12orf66 and ITFG2 in MCF7 cells (Fig. 1C and D; Supplementary Figs. S2A and S2B). We then tested the in vivo sensitivity of TSC1-depleted MCF7-derived xenografts to alpelisib and found that loss of TSC1 was sufficient to reduce the antitumor efficacy of alpelisib (Supplementary Fig. S2C).
We also generated alpelisib-tolerant or DMSO-tolerant T47D cells by long-term treatment with 1 μmol/L alpelisib or DMSO for 2 months. Alpelisib-tolerant T47D cells were indeed resistant to increasing doses of alpelisib compared with DMSO-tolerant cells (Supplementary Fig. S2D). More importantly, we found that DEPDC5 was dramatically decreased, and TSC2, STK11, and PDE7A were slightly decreased in alpelisib-tolerant T47D cells (Supplementary Fig. S2E).
Altogether, our data identified and confirmed multiple candidate genes whose loss or reduced expression mediates resistance to PI3Kα inhibitors. These genes encode for negative regulators of mTORC1 activity, acting in different branches of the mTOR network.
Convergent mTORC1 activation dictates PI3Kα inhibitor response
Given that these candidates are known negative regulators of mTORC1, it is conceivable that depletion of each candidate rewires mTORC1 signaling upon PI3Kα inhibition, thereby leading to the reduced antiproliferative effects of these therapeutic agents. In the absence of PI3Kα inhibition, we found that depletion of these candidates had minimal effects on the phosphorylation of the mTORC1 downstream effectors S6 ribosomal protein (S6) and 4E-BP1 compared with the control, whereas AKT phosphorylation [pAKT (S473)] was slightly suppressed upon loss of several candidates (Fig. 2A; Supplementary Fig. S3A). These data suggest that the effects on mTORC1 caused by the loss of the candidate genes are likely compensated by a negative feedback loop that serves to restrain AKT phosphorylation (35–39). Upon alpelisib treatment, pAKT(S473) levels remained suppressed in both control and KO cells, suggesting that PI3Kα inhibition still blocks the PI3K–AKT cascade independent of the status of these genes.
As expected, mTORC1 downstream signaling was effectively inhibited by alpelisib treatment in cells transduced with nontargeting sgRNA (Fig. 2A; Supplementary Fig. S3A). On the other hand, we observed a higher level of pS6 or p4E-BP1 in cells with individual candidate loss, especially in cells deprived of TSC1, TSC2, STK11, or the GATOR1 components, implying convergent activation of mTORC1 caused by loss of each of these candidates (Fig. 2A; Supplementary Fig. S3A). These data indicate that loss of these individual genes upon treatment with PI3Kα inhibitors could result in sustained mTORC1 activity in an AKT-independent manner, which consequently limits the response to these agents. To challenge this hypothesis, we treated cells deprived of STK11 or the GATOR1 components with three different AKT inhibitors: AZD5363, GDC-0068, and MK-2206. Consistently, loss of each gene resulted in resistance to AKT inhibitors in both MCF7 and T47D cells (Fig. 2B; Supplementary Fig. S3B).
We then tested the sensitivity of each KO cell line to the mTORC1 inhibitor everolimus. As expected, everolimus effectively blocks mTORC1 signaling regardless of the status of these candidates (Fig. 3A; Supplementary Fig. S4A) and elicits similar antiproliferative activity in both MCF7 and T47D cells (Fig. 3B; Supplementary Fig. S4B). We found that everolimus resensitizes KO cells to PI3Kα inhibition by colony formation assays, underscoring the causative role of mTORC1 in limiting the effects of PI3Kα blockade (Supplementary Fig. S4C).
In summary, our data revealed that loss of each candidate gene restores mTORC1 signaling independent of the PI3K–AKT axis upon PI3Kα inhibition, limiting the sensitivity to PI3Kα or AKT inhibitors.
The nutrient-sensing mTORC1 pathway affects PI3Kα inhibitor response
The GATOR1 (26) and KICSTOR (33, 34) complexes are negative regulators of mTORC1 activity in the amino acid sensing pathway, which is a key branch within the mTOR network. Mutational inactivation or decreased expression levels of DEPDC5 or NPRL2 have been reported in glioblastoma, ovarian cancer, and gastrointestinal stromal cancer (26, 40, 41). Loss of GATOR1 complex, the GAP for RagA or RagB, via DEPDC5, NPRL2 or NPRL3 KO should increase levels of GTP-bound RagA or RagB and, therefore, lead to constitutive activation of mTOR (26). In line with the literature, DEPDC5 KO cells indeed displayed increased mTORC1 lysosomal localization (Supplementary Fig. S5A) and mTORC1 downstream signaling (Fig. 4A), irrespective of the amino acid status, suggesting that DEPDC5 loss leads to an increase in the active form of RagA or RagB in MCF7 and T47D cells.
We next explored whether dominant negative Rag GTPases could reverse DEPDC5 loss induced drug resistance and mTORC1 restoration. We overexpressed dominant-negative RagA GTPase, RagAT21L, or dominant-negative heterodimer, RagAT21L/RagDQ121L, in DEPDC5-deficient MCF7 and T47D cells. We found that expression of RagAT21L or RagAT21L/RagDQ121L reduced mTORC1 lysosomal localization (Supplementary Fig. S5B) and mTORC1 activity in DEPDC5 depleted cells (Fig. 4B), and reversed DEPDC5 loss-mediated drug resistance to PI3Kα and AKT inhibitors (Fig. 4C; Supplementary Fig. S5C). Overexpression of RagAT21L alone was sufficient to block mTORC1 activation in DEPDC5-deficient cells (Fig. 4C), indicating that the deregulation of RagA or RagB GTPase induced by DEPDC5 depletion could be responsible for these phenotypes.
To further determine the impact of deregulated Rag GTPases on drug sensitivity to PI3Kα, AKT, or mTOR inhibitors, we overexpressed different forms of Rag A or RagA/RagD heterodimer in MCF7 and T47D cells. Overexpression of WT RagA or RagA/D did not change drug sensitivity for any of the inhibitors tested. Conversely, overexpression of active form of RagA, RagAQ66L, or RagA/D heterodimer, RagAQ66L/RagDS77L, mimicked DEPDC5 depletion-induced responses and conferred resistance to PI3Kα and AKT inhibitors but not to mTOR inhibitor (Fig. 5A and B; Supplementary Figs. S6A–S6C). In the absence of inhibitors, changes in mTORC1 signaling were not significant upon overexpression of different forms of Rag GTPases. However, RagAQ66L and RagAQ66L/RagDS77L overexpression allows sustained mTOR signaling upon PI3Kα inhibition (Fig. 5C and D; Supplementary Figs. S6D and S6E), whereas mTOR inhibition effectively blunts the signaling (Fig. 5D and E; Supplementary Figs. S6E and S6F).
Genomic lesions of the candidates were identified in primary and metastatic PIK3CA-mutated ER+ breast tumors
To study the prevalence of genomic lesions affecting the mTORC1 related genes that were found in our CRISPR screen, we first analyzed 294 PIK3CA-mutated luminal breast cancer samples from TCGA breast invasive carcinoma database (PanCancer Atlas). Truncating mutations or homozygous deletions (predicting to lead to loss of function) were observed for DEPDC5, NPRL2, SZT2, and STK11 (Fig. 6A). We then analyzed an additional cohort of 617 metastatic breast cancer samples from Bertucci and colleagues (21), of which, 381 samples were from ER+ tumors. Of these, 129 harbored hotspot mutations in PIK3CA and, therefore, were potentially amenable to pharmacological PI3Kα inhibition. We found mutations of SZT2, TSC2, TSC1, DEPDC5, STK11, MARK2, PDE7A, NPRL3, NPRL2, AKT1S1, and C12orf66 in cumulatively 14.7% of PIK3CA mutant samples (Fig. 6B). Importantly, we observed mutual exclusivity of these genes (P = 0.0429, WexT).
We then analyzed a curated cohort of 1,918 tumors from 1,756 patients with breast cancer, in which 443 samples are ER+/HER2− breast cancers with PIK3CA mutations (18, 42). In this cohort, genomic alterations of more than 400 cancer-associated genes were identified using the FDA-authorized Memorial Sloan Kettering-integrated mutation profiling of actionable cancer targets (MSK-IMPACT) platform (43). TSC1, TSC2, and STK11 are included in the list of sequenced genes. In line with the findings in the metastatic setting, we found that about 3% of PIK3CA-mutated ER+ patients carried genomic alterations of TSC1, TSC2, or STK11 (∼1% for each gene, Fig. 6C).
These results suggest that genomic alterations involving the genes that potentially affect mTOR activity are present in a significant proportion of primary and metastatic PIK3CA-mutant breast tumors and may play a role in limiting the sensitivity to PI3Kα inhibition.
Gain-of-function mutations in the PIK3CA gene are the most frequent genomic alterations in ER+ breast cancer, occurring in over 40% of cases (6, 18, 44). Tumors bearing these mutations are sensitive to specific PI3Kα inhibitors, particularly when given in combination with endocrine therapy (5, 7, 45). However, clinical responses to PI3Kα inhibitors are variable and often short-lived, suggesting that many patients with PIK3CA-mutated ER+ breast cancer harbor or acquire alterations conferring resistance to PI3Kα inhibitors.
In our study, we employed an unbiased approach to identify a list of genes whose loss may affect sensitivity to PI3Kα inhibitors. Most of these genes encode components of different complexes that negatively regulate mTORC1 activity in different signaling branches of the mTOR network. Interestingly, in the absence of PI3Kα inhibitors, the presence of these genomic aberrations was insufficient to change mTORC1 downstream signaling, likely due to the activation of feedback mechanisms on AKT phosphorylation that maintained mTORC1 activity at a normal level (35–39). When the PI3K/AKT pathway was inhibited in the presence of PI3Kα inhibitors, however, loss of these negative regulators activated mTORC1 via the regulation of other parallel and/or interconnected pathways in the mTORC1 network. Our data highlight that this activation of mTORC1 is AKT-independent and suggest that loss-of-function alterations of these genes will allow a constitutively sustained activity of mTOR signaling in the presence of PI3Kα or AKT inhibitors, limiting their antitumor activity.
Genomic lesions of these identified genes were observed in different datasets of PIK3CA-mutated ER+ breast cancers. In one dataset of 129 PIK3CA-mutated ER+ breast cancer (21), 14.7% samples were found to harbor alterations involving SZT2, TSC2, TSC1, DEPDC5, STK11, MARK2, PDE7A, NPRL3, NPRL2, AKT1S1, or C12orf66, suggesting that, collectively, alterations of these genes are relatively frequent.
mTORC1 activity is sophisticatedly regulated and tailored in responding to various upstream stimuli including growth factors, energy, oxygen, and amino acids. Known molecular mechanisms to date that can cause mTORC1 activation and drive resistance to PI3Kα inhibitors are mainly signaling through the tuberous sclerosis complex and the small Ras-related GTPase Rheb (9–13). Although recent studies have greatly progressed our understanding about the activation of mTORC1 by amino acids (26, 33, 34, 40, 41, 46), a correlation between the amino acids-sensing pathway and cancer drug resistance has never been reported. In this work, we have dissected the role of the amino acids-sensing mTORC1 pathway in limiting the sensitivity to pharmacologic PI3Kα or AKT blockade. As a critical regulator of amino acids-sensing pathway in the mTORC1 network, the GATOR1 complex regulates the GTP/GDP-bound status of RagA/B in the Rag heterodimers and determines the localization and activity of mTORC1. Consistent with a previous study (26), we found that loss of GATOR1 complex upon DEPDC5 KO resulted in high level of mTORC1 signaling irrespective of nutrient status in ER+ breast cancer cell lines, suggesting that DEPDC5 loss leads to the increase of active form of RagA/B, thereby leading to constitutive activation of mTORC1. Indeed, overexpression of dominant-negative RagA GTPase or dominant-negative RagA/D heterodimer blunts the mTORC1 activation in DEPDC5-deficient cells and reverses DEPDC5 loss-induced PI3Kα inhibitor resistance, indicating that the deregulation of RagA or RagB GTPase induced by DEPDC5 KO is responsible for these phenotypes.
A significantly higher response rate and a longer median PFS duration have been observed in patients with advanced-stage ER+, HER2− breast cancer treated with everolimus in combination with the aromatase inhibitor exemestane compared with placebo plus exemestane (47). However, the clinical efficacy of this treatment has proven suboptimal, at least in part due to dose-limiting toxicities (4, 48, 49). Thus, there is an unmet clinical need of identifying patients who are more likely to respond to mTOR inhibitors to rationalize the use of these agents.
Pioneering work in this field reported that TSC1 loss-of-function mutations in bladder cancer predicted for everolimus sensitivity (3). More recent work suggested that lack of expression of PTEN and mutations in TSC1/2 and MTOR are enriched in renal cell carcinoma patients responding to everolimus (50). Thus, based on our data, we can hypothesize that patients with breast cancer with tumors bearing one or more genomic lesions affecting mTOR activity may benefit from the treatment of everolimus or other rapalogs.
In summary, this study dissects potential mechanisms for sustained mTOR activity in the presence of PI3Kα inhibitors, leading to resistance to these agents. Our findings suggest that the identification of a specific subpopulation bearing loss of function mutations in negative regulators of mTORC1 may experience a limited response to PI3Kα inhibition and would benefit by concomitant mTOR blockade to prevent or delay drug resistance.
P. Castel reports personal fees and other support from Venthera, Inc. outside the submitted work. K. Jhaveri reports personal fees from Novartis, Genentech, Pfizer, AstraZeneca, Lilly Pharmaceuticals, Seattle Genetics, ADCT Pharmaceuticals, Spectrum Pharmaceuticals, Taiho Oncology, Seattle Genetics, Jounce Therapeutics, and Blueprint Medicines; other from Novartis, Genentech, Pfizer, Lilly Pharmaceuticals, Puma Biotechnology, Immunomedics, AstraZeneca, Clovis Oncology, Zymeworks, ADCT Pharmaceuticals, and Novita Pharmaceuticals outside the submitted work. E. Toska reports personal fees from AstraZeneca outside the submitted work. F. Andre reports grants from Roche, Daiichi, Pfizer, Novartis, AstraZeneca, and Lilly outside the submitted work. S. Chandarlapaty reports grants from Daiichi-Sankyo, Sanofi; grants and personal fees from Novartis, Paige.ai, personal fees from Lilly, Sermonix, Context Therapeutics, and BMS outside the submitted work. J.S. Reis-Filho reports personal fees from Paige, Repare Therapeutics, Goldman Sachs, Grupo Oncoclinicas, Roche Tissue Diagnostics, Novartis, Genentech, Roche, In Vicro, and Eli-Lilly outside the submitted work. P. Razavi reports grants from Grail/Illumina and Archer Dx; grants and personal fees from Novartis, AstraZeneca, and Epic Sciences; personal fees from Foundation Medicine, Inivata, and Natera outside the submitted work. M. Scaltriti reports grants from AstraZeneca, Puma Biotechnology, Daiichi-Sankio, Menarini Ricerche, and Targimmune; other support from AstraZeneca during the conduct of the study; other support from AstraZeneca outside the submitted work. No disclosures were reported by the other authors.
Y. Cai: Conceptualization, data curation, formal analysis, validation, methodology, writing–original draft. G. Xu: Conceptualization, data curation, supervision, investigation, methodology, writing–original draft. F. Wu: Data curation, investigation, methodology. F. Michelini: Conceptualization, data curation, formal analysis, supervision, validation, investigation, methodology. C. Chan: Data curation, software, formal analysis, validation, investigation, methodology. X. Qu: Data curation, software, formal analysis, validation, investigation, methodology. P. Selenica: Data curation, software, formal analysis, investigation, methodology. E. Ladewig: Conceptualization, data curation, software, supervision, validation, investigation, methodology, writing–original draft. P. Castel: Conceptualization, data curation, software, supervision, validation, investigation, visualization, writing–original draft. Y. Cheng: Data curation, software, validation, investigation, visualization, methodology. A. Zhao: Conceptualization, resources, data curation, formal analysis, supervision, investigation, methodology. K. Jhaveri: Conceptualization, resources, data curation, supervision, funding acquisition, investigation. E. Toska: Resources, data curation, software, formal analysis, supervision, validation, investigation, visualization, methodology. M. Jimenez: Conceptualization, data curation, software, formal analysis, supervision, validation, methodology, writing–review and editing. A. Jacquet: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, methodology, writing–original draft. A. Tran-Dien: Conceptualization, data curation, formal analysis, validation, investigation, methodology. F. Andre: Conceptualization, data curation, formal analysis, supervision, validation, investigation. S. Chandarlapaty: Resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation. J.S. Reis-Filho: Resources, data curation, software, formal analysis, validation, investigation, visualization, methodology. P. Razavi: Data curation, software, formal analysis, supervision, investigation, writing–review and editing. M. Scaltriti: Conceptualization, resources, formal analysis, supervision, funding acquisition, validation, investigation, writing–original draft.
We would like to thank all the Scaltriti lab members for their advice and support. This work has been supported by NIH grants P30 CA008748 and R01 CA190642-01A1, the Breast Cancer Research Foundation, the Geoffrey Beene Cancer Research Center (to M. Scaltriti and S. Chandarlapaty). This work was also supported by grants from Stand Up to Cancer (Cancer Drug Combination Convergence Team) grant no. SU2C 2015-004, the V Foundation grant no. D2015-036, and the National Science Foundation grant no. PHY-1545853 (to G. Xu and M. Scaltriti). E. Ladewig was supported by NCI grant no. K00CA212478. F. Andre was supported by BCRF and foundation ARC. M. Scaltriti and S. Chandarlapaty are supported by a kind gift from Mrs. Barbara Smith.
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.