Targeting the Dependence on PIK3C3-mTORC1 Signaling in Dormancy-Prone Breast Cancer Cells Blunts Metastasis Initiation Open Access

Nearly one quarter of patients with breast cancer experience metastatic relapse, months to over two decades after their initial diagnosis due to metastatic dormancy (1–3). The latter phenomenon occurs when tumor cells in secondary organs become quiescent (cellular dormancy) or maintain a balance between proliferation and apoptosis (micrometastatic dormancy), leading to clinically undetectable metastatic disease. The understanding of the molecular and biological dynamics of metastatic dormancy is still emerging (4). Identifying the molecular determinants that mediate the survival mechanisms of dormant disseminated tumor cells (DTC) and their transition to metastatic outgrowth is needed to develop therapies to prevent relapse. These molecular determinants include cancer cell intrinsic factors and microenvironmental cues that can direct the fate and response of DTCs to therapies (5–8).

Different human and murine breast cancer cell line models with shared origins but differential metastatic behaviors have been invaluable in identifying factors that mediate DTCs’ fate (reviewed in ref. 8). One example is the 4T1 and 4T07 murine breast cancer cell lines, originating from the same spontaneous metastatic nodule in BALB/c mice (9). In immunocompetent BALB/c mice, spontaneously disseminated 4T1 cells form overt lung metastases after a short latency period, whereas 4T07 cells remain dormant and show a low frequency of metastasis. In immunocompromised mice, disseminated 4T07 cells escape dormancy and form overt metastases though less aggressively than 4T1 cells (10), indicating intrinsic differences between the two cell lines. Indeed, differential gene expression analysis between 4T1 and 4T07 tumors identified TWIST1 as an upregulated gene in 4T1 cells mediating epithelial–mesenchymal transition and metastasis (11). Moreover, a gain-of-function screen using a cDNA library derived from 4T1 identified Coco as a mediator of the reactivation of disseminated dormant 4T07 cells (12). The parallel study of these two cell lines, therefore, offers an opportunity to define new cancer cell intrinsic factors that contribute to metastatic progression.

In this study, we investigated whether the 4T1 and 4T07 cells exhibit differential activity in signaling pathways that could be exploited for therapeutic intervention in vivo. Using a genome-wide loss-of-function CRISPR screening framework, we systematically identified the fitness genes that are differentially essential for the growth of 4T1 versus 4T07 cells. Identifying fitness genes involved in a specific signaling pathway indicates potential activity of this pathway and sensitivity to its targeted inhibition (13). We discovered that 4T1, but not 4T07, cells rely on and show higher activity of class I PI3K. Interestingly, 4T07 cells demonstrate higher activity of the downstream mechanistic target of rapamycin complex 1 (mTORC1). We show that 4T07, unlike 4T1 cells, depends on class III PI3K to maintain peripheral lysosomal positioning, which is linked to mTORC1 hyperactivity. This lysosome positioning–associated PIK3C3-mTORC1 signaling circuit is present in human breast cancer cell lines, including a patient-derived xenograft (PDX) line originating from a patient with breast cancer metastatic relapse. Notably, selective inhibition of Pik3c3 exclusively kills the disseminated 4T07 cells. Our findings suggest that targeting the PIK3C3-mTORC1 signaling axis could be a promising therapeutic strategy to reduce metastatic burden in some patients with breast cancer by eliminating DTCs.

Parental 4T07 cells (9) were obtained from the Karmanos Cancer Institute, Wayne State University. The 4T07-TGL cells (Memorial Sloan Kettering Cancer Center, New York, NY; ref. 12) were obtained from Dr. Filippo Giancotti (Herbert Irving Comprehensive Cancer Center, New York, NY). 4T1 cells were obtained from ATCC. To generate 4T1-TGL cells, HEK293T cells were seeded in 6 cm plates and transfected with 3 μg of TGL vector (gift from Dr. Vladimir Ponomarev, Memorial Sloan Kettering Cancer Center; ref. 14) and 3 μg of pCL-Ampho using CaCl₂ for retroviral production. Then, parental 4T1 cells were infected, and stable 4T1-TGL cells were generated and established after sorting for GFP-positive cells. To generate 4T07 cells expressing doxycycline-inducible short hairpin RNA (shRNA), we first generated pLKO-Tet-ON plasmids encoding different sequences predicted to either target Pik3c3 or act as a nontargeting control (see Table 1 for sequences). HEK293FT cells were then transfected with the generated plasmids in addition to psPAX2 and pMD2.G to produce lentiviruses. 4T07 cells were then transduced with these lentiviruses and selected with puromycin (3 µg/mL) for 3 to 5 days. 4T07-shRNA cell lines were induced with doxycycline (1 µg/mL). All the 4T1 and 4T07 cell lines, in addition to T47D cells (obtained from Dr. Morag Park’s lab, McGill University), were cultured in DMEM (Wisent, #319-005-CL), supplemented with 10% FBS (Gibco, #12483-020) and 1% penicillin/streptomycin (P/S) antibiotics (Wisent, #450-201-EL). CAMA-1 (obtained from ATCC) and HCC70 (obtained from Dr. Peter Siegel’s lab, McGill University) were cultured in Eagle Minimum Essential Medium (Wisent, #320-026-CL) and RPMI 1640 (Wisent, #350-000-CL), supplemented with 10% FBS and 1% P/S. For the majority of signaling experiments presented in the article, cells were maintained in culture after thawing for a maximum of 3 weeks.

For serum starvation, cells were cultured overnight in serum-free DMEM supplemented with 1% P/S. For amino acid starvation, cells were first serum-starved overnight, then rinsed with PBS, and incubated in Earle balanced salt solution (EBSS; Thermo Fisher Scientific, #24010043) for 2 hours at 37°C. Refeeding with serum was done by replacing the serum-free DMEM with 10% FBS-supplemented DMEM. To refeed with insulin, 100 nmol/L insulin (Humulin R U-100) was added to the serum-free DMEM, and cells were incubated for 30 minutes at 37°C. To refeed with amino acids, Minimum Essential Medium amino acid solution (50×; Sigma-Aldrich, #M5550) was added to the EBSS (1:50), and cells were incubated for 30 minutes at 37°C.

All PDX epithelial cell lines were derived from tissues obtained according to the ethical regulations of McGill University (Research Ethics Board–approved protocols) and after all patients signed a written consent form, in accordance with the Declaration of Helsinki, as stated previously in the original study that characterized them (15). PDX 1915 (15) was cultured in DMEM (Gibco, #11995-065) and F-12 Nutrient Mixture (Ham; Gibco, #11765-054; 1:4), 5% FBS (Life Technologies, #12483012), 0.4 µg/mL hydrocortisone (Sigma-Aldrich, #H0888-1G), 5 µg/mL insulin (Gibco, #12585-014), 8.4 ng/mL cholera toxin (Sigma-Aldrich, #C8052), 10 ng/mL EGF (BPS Bioscience, #90201-1), 10 µmol/L Y-27632 (AbMole BioScience, #M1817), 50 µg/mL gentamicin (Gibco, #15710-072), 1% P/S (Thermo Fisher Scientific, #15140-122), and amphotericin B (1 µg/mL; Thermo Fisher Scientific, #15290026).

Primary 54074 and 99142 cells and the recurrent 42929 and 48316 cells were established and maintained in culture as previously described in detail (16). All cell lines were maintained in DMEM and supplemented with 2 mmol/L L-glutamine, 100 U/mL P/S, 10% bovine calf serum, 10 ng/mL EGF, and 5 μg/mL insulin. Primary cell lines 1 and 2 were additionally supplemented with 2 μg/mL doxycycline, 1 µmol/L progesterone, and 1 μg/mL hydrocortisone. All cells were grown in a humidified incubator at 37°C with 5% CO₂.

The PIK3CA inhibitor BYL719 (ChemieTek, #CT-BYL719), the AKT inhibitor MK-2206 2HCl (Selleck Chemicals, #S1078), and VPS34-IN1 (Cayman Chemical, #17392) were dissolved in DMSO and used in different concentrations as indicated.

Parental 4T07 cells were Mycoplasma-negative as last tested by PCR (see primer sequences below) in October 2022. Given that 4T07-TGL cells were contaminated with Mycoplasma upon receiving them, they were treated with Plasmocin (InvivoGen, #ant-mpt-1) for 2 weeks as per the manufacturer’s protocol before being used in the experiments described in this work. 4T1 cells were Mycoplasma-negative as last tested by PCR in October 2022. Primary 54074 and 99142 cells, in addition to the recurrent 42929 and 48316 cells, tested negative for Mycoplasma via PCR (IMPACT 2 mouse pathogen testing; IDEXX BioAnalytics) in August 2022. PDX 1915 was tested for Mycoplasma before being used for experiments, using the MycoAlert detection kit (Lonza, #LT07-318) as per the manufacturer’s instructions. T47D, CAMA-1, and HCC70 cells were not tested for Mycoplasma contamination by PCR. However, it should be noted that all the described cell lines were frequently used in immunofluorescence experiments in which DAPI staining was performed and verified Mycoplasma-negative status for all the lines included in this work. For the PCR, we used a mixture of six forward primers (CGCCTGAGTAGTACGTTCGC, CGCCTGAGTAGTACGTACGC, TGCCTGAGTAGTACATTCGC, TGCCTGGGTAGTACATTCGC, CGCCTGGGTAGTACATTCGC, and CGCCTGAGTAGTATGCTCGC) and a mixture of three reverse primers (GCGGTGTGTACAAGACCCGA, GCGGTGTGTACAAAACCCGA, and GCGGTGTGTACAAACCCCGA). In all the conducted tests, a positive (Mycoplasma-positive sample) and a negative control were used to verify the test. This protocol was adapted from https://bitesizebio.com/23682/homemade-pcr-test-for-mycoplasma-contamination/ based on refs. 17, 18.

The mouse Genome-Scale CRISPR Knockout (mGeCKO) library A (Addgene, #1000000052), composed of ∼68,000 guide RNAs (gRNA): 3 gRNAs/gene, 4 gRNAs/miRNA, and 1,000 nontargeting sequences, was amplified and prepared for next-generation sequencing (NGS; MiSeq Illumina) to assess the gRNA distribution and library quality, as previously described (19). In reference to the original library, the assessed parameters of the amplified library (percentage of undetected gRNAs and perfectly matching gRNAs, in addition to the skew ratio of the top 10% represented gRNAs to the bottom 10%) were all within the recommended ranges for a reliable library (19).

Lentiviral production was performed as previously described (19). Briefly, low-passage HEK293FT cells were plated in Nunc EasYFlask cell culture flasks (Thermo Fisher Scientific, #156340) and transfected using Lipofectamine 2000 (Thermo Fisher Scientific, #11668027) with psPAX2 (23.4 µg), pMD2.G (15.3 µg), and 30.6 µg of the lentiCRISPR version 2 vector encompassing the Cas9 in addition to the amplified library. Forty-eight hours later, the virus-containing media was collected and filtered using 0.45 μm sterile filters (Sarstedt, #83.1826). A multiplicity of infection (MOI) test was performed to define the optimum volume of the virus-containing media to be used for the screens.

The 4T1 and 4T07 cells were infected independently, as previously described (19). Briefly, the two cell lines were seeded in 12-well plates (1.5–2 × 10⁶ cells/well) and then subjected to spinfection (MOI <0.3) by centrifugation at 2,000 RPM for 2 hours at 33°C in the presence of polybrene (10 µg/mL; Sigma-Aldrich, #H9268-5G). This method allowed for ∼1,300-fold coverage of the mGeCKO library A. After 24 hours, cells were trypsinized, pooled, and seeded in 15 cm plates in media containing puromycin for selection (3 µg/mL for the 4T07 cells and 2 µg/mL for the 4T1 cells; Wisent, #400-160-EM). Cells were propagated under selection for a week while maintaining a library coverage of at least 700-fold (50 × 10⁶ cells). Then, the first time point (T0) was collected (50 × 10⁶ cells/time point to keep the stated fold coverage). The cells were propagated in a similar manner while collecting the subsequent time points (T1, T2, T3). The collected cells were spun down in 15 mL tubes and kept at −80°C until they were processed. Every screen was repeated three independent times following the same procedure, in which replicates were annotated A, B, and C. The collected samples were then processed for genomic DNA (gDNA) extraction according to the manufacturer’s protocol (Quick-gDNA Midiprep Plus Kit, Zymo Research, #D4075) and prepared for NGS (HiSeq 4000, Illumina) as previously described (19).

Cells were plated on fibronectin-coated glass coverslips to the desired confluence for ∼6 hours with the one exception of CAMA-1 cells that were plated directly on glass coverslips 72 hours prior to staining (VWR, #CACB354008). For the 4T07-Pik3c3 shRNA cells, cells were plated on fibronectin-coated glass coverslips and induced with doxycycline (1 µg/mL; Sigma-Aldrich #D9891) for 48 hours before proceeding to staining. For amino acid starvation, cells were serum-starved (overnight, for 16 hours) and then incubated with EBSS for 2 hours. For refeeding, a 50× minimum essential medium amino acids solution was diluted to 1× in EBSS and added to the cells for 30 minutes. Cells were fixed with 3.7% formaldehyde in cytoskeleton buffer (100 mmol/L NaCl, 300 mmol/L sucrose, 3 mmol/L MgCl₂, and 10 mmol/L PIPES, pH 6.8) for 15 minutes. Immunostaining was performed in saponin buffer (TBS supplemented with 1% BSA and 0.01% saponin) using the following antibodies: Rat anti-LAMP1 (Abcam, #ab25245) at a concentration of 1:400, mouse anti-LAMP-2 (Developmental Studies Hybridoma Bank, #H4B4) at a concentration of 1:50, mouse anti-phospho(Ser2448)-mTOR (Cell Signaling Technology, #2971) at a concentration of 1:200, and rabbit anti-Rptor (14C4; Bioss Antibodies, #51285M) at a concentration of 1:100. Images were collected from a Zeiss confocal LSM 700 (Carl Zeiss) microscope with oil immersion objective lenses (Plan-Apochromat, 60×, 1.40 numerical aperture; Carl Zeiss).

Image analysis was performed using the FIJI software version 2.3. For the measurement of pmTOR at the lysosomes, a region of interest corresponding to the lysosomes was created using the auto-threshold of the LAMP1 channel and was transferred to the corresponding background-subtracted pmTOR channel to measure the mean fluorescence intensity at the lysosomes. Data represent the mean with SD of at least 30 images from three independent experiments. Statistical significance was determined using a one-way ANOVA test with a confidence interval of 95%. The lysosome distribution was measured from confocal images of LAMP1 and pmTOR costaining captured as described above. For at least 75 cells per condition, a single line, five pixels in width, was traced manually from the center of the nucleus to the cell edge as delineated by pmTOR staining. The intensity values along this line were obtained using the “plot profile” plugin from FIJI. The mean fluorescence intensity and the positions along the line for each cell were normalized from 0 to 1 using MATLAB as used in ref. 20. The normalized data were transferred to Prism 9 software (GraphPad Prism, GraphPad Software) for further analyses. The percentage of signal at the periphery was calculated by dividing the AUC between positions 0.75 and 1 by the total AUC between positions 0 and 1. Data are representative of three independent experiments and are shown as the mean with SD (n > 75). Statistical significance was determined using the nonparametric Mann–Whitney test with a confidence interval of 95% (P value <0.0001). This test analysis compares the distributions of two unpaired groups. Data are shown as the mean with SD. For tubulation quantification, the percentage of cells presenting at least one tubulated lysosome was determined among at least 220 cells per condition from three independent experiments. The percentages in the DMSO- and VPS34IN1-treated cells (PDX1915) were then compared using a two-tailed t test to calculate the P value as stated in the figure.

All animal experiments were performed in accordance with the Canadian Council of Animal Care guidelines and were approved by the Animal Care Committee of the Montreal Clinical Research Institute.

For the 4T07 and 4T1 experiments,1 × 10⁵ 4T07-TGL or 4T1-TGL cells were, respectively, injected into the fat pad of the fourth mammary gland of 6-week-old NU/J mice (The Jackson Laboratory, #002019). For the 4T07 experiment, 1 week after injection, the mice were randomized into two groups to receive either vehicle (5% DMSO in corn oil) or VPS34IN1 (50 mg/kg/day) by oral gavage daily for either 6 or 12 days. For the 4T1 experiment (in nude mice), 1 week after injection, mice were randomized to receive either vehicle or VPS34IN1 (50 mg/kg/day) for 2 weeks.

For experiments in immunocompetent mice using the 4T07 model, 2 × 10⁵ 4T07 cells were resuspended in 50 µL of sterile PBS and injected into the fat pad of the fourth mammary gland of 8- to 10-week-old BALB/C mice (The Jackson Laboratory, #0651). On day 24 after injection, mice received one daily oral gavage of 50 mg/kg/day of VPS34IN1 for 6 consecutive days. On day 30, mice were euthanized to recover the primary tumor and the lungs. Lungs were placed in 2 mL of 10% FBS-supplemented DMEM, chopped into pieces, and incubated with 10 mg/mL collagenase IV (Sigma-Aldrich, #C5138) for 3 to 4 hours at 37°C. The tissue preparation was vortexed every hour to facilitate tissue digestion. A single-cell suspension was prepared by pipetting up and down 25 times with a P100 and filtered through a 100 µm filter to remove debris. Cells were pelleted to perform red blood cell lysis according to the manufacturer’s instructions (Sigma-Aldrich, #R7757). Half of the lung cell preparation was plated in a 10 cm dish in DMEM supplemented with 10% FBS, 1% P/S, and 25 µmol/L thioguanine (Sigma-Aldrich, #A4660), in duplicate. Colonies were allowed to grow for 7 days before being stained with crystal violet. For the staining with crystal violet, plates were washed twice with sterile PBS, fixed for 20 minutes with methanol, stained with a crystal violet solution (0.5% crystal violet, Sigma-Aldrich, #C6158; 25% methanol; 75% PBS) for 40 minutes, and washed twice with sterile water. Plates were allowed to dry overnight, and colonies were counted using a light microscope. Results were plotted using GraphPad Prism (GraphPad Software). Different conditions were compared using the Mann–Whitney test to calculate statistical differences, and the resulting P value is indicated on the graph.

For experiments in immunocompetent mice using the 4T1 model, 5 × 10⁵ 4T1 cells were injected into the fat pad of the fourth mammary gland of 8- to 10-week-old BALB/c mice (The Jackson Laboratory, #0651). Two weeks later, primary tumors were resected. Mice were left to recover for ∼3 to 5 days and then randomized for treatments with either vehicle or VPS34IN1 for a week as described above. At the experimental endpoint, mice with recurrent primary tumors were excluded from the final presented analyses. Otherwise, mice were sacrificed, and lungs were collected immediately, immersed in PBS, and then fixed in 4% PFA for 24 hours at 4°C. Lungs were then placed in 70% ethanol before being embedded in paraffin. Paraffin blocks were sectioned at 4 μm thickness and processed for hematoxylin and eosin staining, and the 20× objective of the Aperio-XT slide scanner (Leica Biosystems) was used to scan the slides. Fiji (ImageJ) was used to visualize the generated images and measure the size of metastatic lesions. Given the irregularity between different lesions, the length of the longest dimension was considered a measure to categorize micrometastases (<100 μm), macrometastases (>200 μm), and lesions scoring between 100 and 200 μm. The vehicle- and VPS34IN1-treated groups were compared using a two-tailed t test to calculate the P value (n = 6 or 7 mice/group).

At the experimental endpoint, mice were euthanized, and lungs were processed as previously described (6). Briefly, lungs were dissected and immediately immersed in PBS and cut into small pieces. Then, the lung pieces were digested in PBS supplemented with 75 µg/ml TH Liberase (Roche, #05401127001), 75 µg/ml TM Liberase (Roche, #05401151001), and 12.5 µg/mL DNase. Samples were then incubated in the digestion buffer on a rotor at 37°C. Samples were then centrifuged at 1,300 RPM for 5 minutes, and the supernatant was discarded. Pellets were resuspended in PBS, pipetted up and down multiple times, and filtered through a 70 µm strainer to get rid of the undigested lung pieces. One milliliter of red blood cell lysis solution (Sigma-Aldrich, #R7757-100ML) was then added to each sample, and the manufacturer’s protocol was followed. Cells were finally seeded in 15 cm plates in thioguanine-containing DMEM (Wisent) supplemented with 10% FBS and 1% P/S antibiotics. Plates were left for 48 hours, and D-luciferin (Promega, #E1605; dissolved in PBS) was added to the DMEM at a concentration of 150 µg/mL. Plates were then imaged using the Xenogen IVIS 200 machine (PerkinElmer) and analyzed with Living Image 4.2 software.

The difference in the DTC burden between the vehicle- and VPS34IN1-treated groups was quantified as a fold-change difference in bioluminescence emitted signal between the two groups. Fold-change values were represented in log values, and the two groups (control and drug-treated) were compared using a two-tailed t test to calculate the P value. The resulting values are stated on the graph and in the main text.

Mice were injected intraperitoneally with a 150 mg/kg D-luciferin (in PBS) solution and were euthanized 5 to 7 minutes later. Lungs were immediately collected and imaged using the Xenogen IVIS 200 machine (PerkinElmer).

Individual gRNAs targeting genes or the Rosa locus (Table 2) were cloned in either LentiGuide-gRNA-NLS-GFP-2A-PURO or LentiGuide-NLS-mCherry-2A-PURO using the Golden Gate assembly method (19, 21). Producing lentiviruses from the cloned vectors and the lentiCas9-Blast (Addgene, #52962) was performed as detailed earlier but with scaling down the production procedure. The 4T1 or 4T07 cells were first transduced with the lentiCas9-Blast and selected with blasticidin (5 µg/mL; Thermo Fisher Scientific, #R210-01). An MOI test for the produced individual gRNA lentiviruses was performed. The 4T1-Cas9 and 4T07-Cas9 cells were subsequently infected and selected with puromycin as detailed earlier.

Immediately after selection, 5 × 10⁴ of the 4T1-gRNA (in LentiGuide-gRNA-NLS-GFP-2A-PURO) were seeded with 5 × 10⁴ of the 4T1-Rosa gRNA (in LentiGuide-NLS-mCherry-2A-PURO) in a 12-well plate and left overnight. The following day, the plates were imaged using the 4× objective of IncuCyte (Essen Bioscience) to assess the representation of the green and red populations at the first time point of the assay [passage 0 (P0)] by counting the GFP- and mCherry-positive nuclei. Every 2 days, cells were split 1:10 into new 12-well plates, kept overnight, and imaged the following day to assess the representation of the two populations. This pipeline was repeated over 10 passages. The same procedure was repeated for all the investigated gRNAs (targeting Pik3ca, Pik3c3, and Pik3r4) in this study.

In every passage, the percentage of the GFP-positive population from the total population (GFP- and mCherry-positive combined) was calculated. Then, the ratio between the percentage of the GFP-positive population at every passage over P0 (reference time point) was used to assess the dropout of the GFP-positive population. The calculated ratio in the last passage (P10) of each condition (i.e., gRNA) was compared with that of the control condition (i.e., the two red and green cocultured populations expressing gRNA targeting the ROSA locus) by a two-tailed t test to calculate the P value. The resulting values are stated on graphs.

In a 48-well plate, 5 × 10³ 4T1 or 4T07 cells were seeded overnight, and the corresponding drugs (BYL719 or MK-2206) were added in different concentrations (two wells/concentration as technical replicates). Plates were immediately transferred to be imaged with the IncuCyte (Essen Bioscience). After 72 hours, the cells’ confluency was determined for all the drug concentrations. The increase in confluency was determined for every condition in reference to the confluency at the first time point (immediately after adding the drugs). Then, a ratio between the determined change in confluency under different drug doses and that of the control condition (DMSO-treated) was calculated for each cell line. These values for the two cell lines were then plotted to compare the response of the two cell lines to the same drug. The values (response to the same dose in reference to the DMSO-treated condition) of the two cell lines were compared using a two-tailed t test to calculate the P value. The resulting values are stated on the graphs.

This three-dimensional (3D) in vitro culture condition has been used as a surrogate for the in vivo extracellular matrix environment. To mimic matrix-associated growth factors, 96-well cell culture plates were coated with 40 mL of Cultrex Basement Membrane Extract (BME; R&D Systems, #3433-010-01; ref. 22), containing laminin, collagen IV, entactin, and heparan sulfate proteoglycan, and left to polymerize into a hydrogel for 30 minutes at 37°C. Three thousand 4T1 and 4T07 cells suspended in RPMI 1640 media supplemented with 2% FBS, 2% BME, and 1% P/S were seeded onto the BME. Plates were further placed into the IncuCyte imaging system (Essen Bioscience) and imaged once every 12 hours with the 10× objective. Four days after seeding, the VPS34IN1 drug was added at the indicated doses (5 or 10 μmol/L) together with SYTOX Near-IR Dead Cell Stain (Thermo Fisher Scientific, #S11382) and imaged for the following 4 days to quantify the cell death burden. At the experimental endpoint (eighth day after seeding), counts positive for SYTOX near IR were quantified via the IncuCyte software, and data were plotted in GraphPad Prism (GraphPad Software). Different conditions were compared using a two-way ANOVA test, and P values are indicated on the graphs.

Primary and recurrent cells were seeded in a 96-well plate and incubated overnight before being treated with seven increasing doses of VPS34IN1 (10 nmol/L–100 μmol/L) for 72 hours. At the experimental endpoint, cell viability was measured via CellTiter-Glo (Promega, G7570), as per the manufacturer’s instructions. Quantifications were plotted in GraphPad Prism. To compare the VPS34IN1 efficacy or the sensitivity of the different lines to VPS34IN1, the AUC was calculated in GraphPad and compared using ANOVA and Tukey multiple comparisons test. The response of each cell line to VPS34IN1 was statistically different from that of all the other lines.

As represented in the experimental schematic, primary 54074 cells were maintained in vitro for a week in media either supplemented with or deprived of dox. Then, cells were seeded in six-well plates in three replicates overnight before being treated with either DMSO or 10 μmol/L of VPS34IN1 combined with Cytotox (Sartorius #4632). Immediately after treating the cells, plates were placed in the IncuCyte (SX5) for imaging to define the percentage of Cytotox-positive cells in the whole population/well as an indicator of cell death burden under different conditions. The resulting numbers were plotted in GraphPad Prism, in which a two-way ANOVA test was performed to calculate P values as indicated on the graph.

Cells were rinsed with PBS and lysed using ice-cold RIPA buffer [1% Nonidet P-40, 150 mmol/L NaCl, 50 mmol/L Tris (pH 7.5), 5 mmol/L EDTA, 0.5% deoxycholic acid, 0.1% SDS] supplied with 1 mM Na₃VO₄, 5 mmol/L NaF, and 1× complete protease inhibitor (Roche). Samples were then centrifuged (15 minutes, 13,000 g, 4°C) and boiled with 6× Laemmli sample buffer for 5 minutes at 95°C for denaturation.

Tumors excised from mice were placed in a mortar filled with liquid nitrogen and then mechanically disrupted with a pestle. RIPA buffer (prepared as stated earlier) was added to the disrupted tumor pieces. Once the buffer returned to the liquid state, the mixture of the buffer and the tumor pieces was transferred to tubes and pipetted vigorously up and down several times. The tubes were then left on ice for 15 minutes. Samples were centrifuged, and supernatants were transferred to new tubes and boiled with 6× Laemmli sample buffer as detailed earlier.

Protein separation was then performed using the standard SDS-PAGE protocol, and proteins were transferred to nitrocellulose membranes (Bio-Rad, #1620115) via wet transfer for 2.5 hours at 4°C. Membranes were subjected to blocking for 0.5 to 1 hour in TBS solution supplemented with 0.1% Tween 20 (TBST; Bio Basic, #TB0560) and 1% BSA (Wisent, #800-095-CG) before incubation with primary antibodies overnight at 4°C. The following day, membranes were washed (3× for 10 minutes) in TBST and incubated with secondary antibodies for 1 hour at room temperature. Membranes were washed (3× for 5 minutes) in TBST and incubated with enhanced chemiluminescence substrate (Bio-Rad, #1705061) for 1 minute prior to exposure.

All primary antibodies were diluted in TBST supplemented with 1% BSA and used at the following concentrations: rabbit anti-phospho-AKT(S473) (#9271S), rabbit anti-AKT (#9272S), rabbit anti-phospho-S6K1 (Thr389) (#9205S), rabbit anti-S6K1 (#9202S), rabbit anti-phospho-S6 (Ser240/244) (#2215S), and rabbit anti-S6 (#2217S), which were obtained from Cell Signaling Technology and used at a concentration of 1:1,000. Mouse anti-α-tubulin (Sigma-Aldrich, #T5168) was used at a concentration of 1:8,000. The secondary antibody mouse anti-rabbit IgG-HRP (Santa Cruz Biotechnology, #2357) for pAKT, AKT, pS6K, S6K, pS6, and S6 was diluted (1:5,000) in TBST supplemented with 1% BSA. The secondary antibody goat anti-mouse IgG-HRP for α-tubulin (Sigma-Aldrich, #A4416) was used at the same concentration. The antibody anti-FLAG M2-peroxidase (HRP; Sigma-Aldrich, #A8592) was used at a concentration of 1:8,000 in TBST supplemented with 1% BSA to assess the expression of PIK3C3.

The gel analyzer tool of Fiji (ImageJ) was used to quantify the Western blot signals. mTORC1 activity was assessed as a ratio between the pS6K and total S6K signals. The calculated ratio (i.e., mTORC1 activity in arbitrary units) was either used directly to compare the 4T1 and 4T07 cells or normalized to show relativity between the two lines (Supplementary Fig. S5A–S5C). In all cases, a two-tailed t test was used to calculate the P value for the difference between the two tested conditions. The calculated P values were stated in figures and text. To quantify the mTORC1 activity in the analyzed tumors, the phospho-S6K1/total S6K1 ratio was similarly calculated for every sample as a readout. Then, to compare two groups (drug vs. vehicle) while considering the variability in activity even within the same group, the fold change of each sample’s activity was calculated relative to each sample’s activity in the opposing group, and an average was determined as a final value for each sample. The latter calculation was made for every sample in the two groups. Finally, the resulting values were represented in log values, and the two groups (control and drug-treated) were compared using a two-tailed t test to calculate the P value. The resulting values are stated on the graphs.

RNA sequencing (RNA-seq)–related datasets/analyses (e.g., 4T1-4T07 differentially expressed genes and correlation with human breast cancer cell lines) are included in Supplementary Tables S1 and S8. The RNA-seq data generated in this study are publicly available in the Gene Expression Omnibus at GSE203296. 4T1-4T07 CRISPR screens’ data are provided in Supplementary Tables S2–S5. Whole-genome sequencing (WGS) datasets are provided in Supplementary Tables S6 and S7. The clinical outcome data analyzed in this study were obtained from The Cancer Genome Atlas (TCGA) and METABRIC studies included in cBioPortal (https://www.cbioportal.org/). All other raw data generated in this study are available upon request from the corresponding author.

We characterized the molecular subtype of breast cancer modeled by 4T1 and 4T07 cells. We generated and correlated RNA-seq transcriptomic data from these two cell lines with gene expression data from 47 human breast cancer cell lines. We conclude that 4T1 and 4T07 cells are closer to the human basal B molecular breast cancer subtype, predominantly within the triple-negative clinical breast cancer subtype, complementing a recent investigation of 4T1 cells (Supplementary Fig. S1; Supplementary Table S1; ref. 23). The 4T07 cells display features of the basal B subtype, consistent with a common origin of the 4T1 and 4T07 cells (Supplementary Fig. S1; Supplementary Table S1).

Genome-wide loss-of-function CRISPR screens identify core and context-dependent fitness genes, revealing common and specific active signaling nodes (13, 24, 25). Fitness genes are those that, when perturbed, result in a proliferative or survival disadvantage for cells. We hypothesized that fitness genes specific to 4T1 may be involved in survival or metastatic progression, whereas those specific to 4T07 might regulate survival during dormancy or outgrowth when implanted in permissive microenvironments. To explore this, we used a CRISPR screening pipeline on 4T1 and 4T07 cells, employing the mGeCKO library encompassing ∼70,000 gRNAs targeting ∼21,000 murine genes (19). Following selection for cells with gRNA integration (T0), gDNA was extracted at three consecutive time points over 3 weeks (T1, T2, and T3) to map gRNA representation relative to T0 (Fig. 1A). Quality control analyses confirmed adequate library coverage (Supplementary Fig. S2A and S2B) and clustering of biological replicates (Supplementary Fig. S2C and S2D). The two screens showed high sensitivity, with precision–recall effectively distinguishing essential from nonessential genes (Supplementary Fig. S2E and S2F). Finally, fold-change distributions of gRNAs targeting essential genes over the course of the screens were prominently shifted, whereas those targeting nonessential genes showed minimal change (Supplementary Fig. S2G and S2H). These analyses demonstrated the screens’ robustness in identifying fitness genes in both cell lines.

We identified 2,014 fitness genes in 4T1 cells and 2,038 in 4T07 cells, including 1,257 common genes (Supplementary Table S2) that functionally mediate biological processes known for core essential genes in cancer (Fig. 1B; Supplementary Fig. S3; refs. 13, 26). Consequently, 757 and 781 specific fitness genes were identified in 4T1 and 4T07 cells, respectively (Fig. 1B; Supplementary Tables S3 and S4). We focused on identifying differential signaling pathway dependencies. Among the top-ranking 4T1-specific fitness genes were Pik3ca (catalytic subunit of class I PI3K) and Rictor [core component of mTOR complex 2 (mTORC2; Fig. 1C)], both of which are positive regulators of the PI3K pathway (Fig. 1D). Conversely, PI3K-negative regulators, such as Pten, Tsc1, Tsc2, and Pras40, were identified as proliferation suppressor genes, in which gRNAs were enriched at the last time point (T3) in 4T1 cells (Fig. 1D and Supplementary Table S5). These data suggested that 4T1 cells may possess higher PI3K-AKT activity in comparison with 4T07 cells. Indeed, a gene set enrichment analysis of the two cell lines’ global transcriptome identified elevation of AKT and downregulation of PTEN gene signatures to be enriched in 4T1 in comparison with 4T07 cells (Fig. 1E and F). By analyzing the 4T07-specific fitness genes, we identified Pik3c3 (catalytic subunit of class III PI3K) and its regulatory subunit Pik3r4 (Fig. 1C). Given the clinical relevance of the class I PI3K pathway and the less explored class III PI3K pathway in breast cancer, we investigated these differential dependencies.

To confirm the essentiality of Pik3ca in 4T1 cells, we performed a two-color competition assay (21). We generated 4T1 cells expressing Cas9 and then infected them with either a vector to express a Pik3ca-targeting gRNA and GFP or a vector to express mCherry and a gRNA targeting Rosa26 (Fig. 2A). Coculturing these lines in a 1:1 ratio led to the dropout of cells expressing Pik3ca-targeting gRNA(s) as indicated by monitoring the percentage of green cells within the total population over 10 passages (Fig. 2B and C). We validated these findings pharmacologically by treating 4T1 and 4T07 cells with the PIK3CA selective inhibitor BYL719 (27), which significantly inhibited 4T1 cell proliferation but did not affect 4T07 cells (Fig. 2D). Next, we investigated the effect of the AKT inhibitor MK-2206 and found that it inhibited the proliferation of 4T1 cells more prominently than that of 4T07 cells (Fig. 2E).

To validate the essentiality of Pik3c3 in 4T07 cells, we employed the same two-color competition assay (Fig. 2A). We observed a significant drop in the population of 4T07 cells expressing gRNAs targeting Pik3c3 or Pik3r4 over time, as compared with the control (Fig. 2F and G). In contrast, in a similar assay, 4T1 cells showed only a modest reduction (Supplementary Fig. S4A and S4B). Comparing the representation of Pik3c3 gRNA-expressing 4T1 and 4T07 populations after normalization to their respective control conditions confirmed the predicted differential dependence on Pik3c3 (Supplementary Fig. S4C). To validate these findings pharmacologically, we treated 4T07 and 4T1 spheroids with two doses of VPS34IN1, a PIK3C3-specific inhibitor (28), and assessed cell death (Fig. 2H). VPS34IN1 treatments led to a 2- to 3-fold increase in the death rate of 4T07 cells, with no significant effect on the 4T1 cells compared with vehicle-treated controls (Fig. 2I and J). These findings indicate that Pik3c3 is essential for 4T07 but not 4T1 cell survival.

These data demonstrate that the differential genetic dependence on PI3K classes in 4T1 and 4T07 cells is reflected in differential vulnerability to specific inhibitors of these PI3Ks. This prompted us to investigate whether these phenotypes result from differential signaling activity in the PI3K-AKT-mTORC1 pathway.

We assessed the activity of downstream components of the PI3K pathway. 4T1 cells exhibited higher PI3K activity under untreated conditions, indicated by increased phosphorylated AKT levels (Fig. 3A–C). Short-term treatments with BYL719 similarly inhibited the PI3K-AKT-mTORC1 pathway in both cell lines, as demonstrated by reduced levels of phosphorylated AKT, S6K1, and S6 (Fig. 3A). However, short-term AKT inhibition did not mediate the same effects as Pik3ca inhibition (Fig. 3B). Specifically, mTORC1 activity, measured by phospho-S6K1 levels, was reduced only in 4T1 cells upon AKT inhibition. Prominently in the untreated conditions in both experiments (Fig. 3A and B), 4T07 cells demonstrated 2-fold higher mTORC1 activity than 4T1 cells (Fig. 3C) despite the dispensability and low activity of class I PI3K in 4T07 cells (Fig. 3A–C). This finding is intriguing from a signaling and biological perspective. First, sustained or reactivated mTORC1 activity confers resistance to PIK3CA inhibition (29, 30). Second, PIK3C3 mediates mTORC1 activity, especially when class I PI3K is inhibited long-term (31). Third, high mTORC1 activity is known to support the survival of dormant cells in vivo (32). This led us to further explore the regulation of mTORC1 activity in 4T1 and 4T07 cells.

We investigated the reliance of each cell line on Pik3ca for mTORC1 activity. Low-dose BYL719 (1 μmol/L), under serum starvation or subsequent insulin stimulation, reduced mTORC1 activity preferentially in 4T1 over 4T07 cells (Supplementary Fig. S5A–S5C), indicating that 4T07 cells are less dependent on Pik3ca for mTORC1 regulation. To determine whether mTORC1 hyperactivity in 4T07 cells affects their response to BYL719, we treated 4T07 cells with a high dose of BYL719 (10 μmol/L; did not affect proliferation; Fig. 2D) and monitored mTORC1 activity over time. We observed that mTORC1 activity started to increase after 48 hours of treatment (Supplementary Fig. S5D). These findings suggest noncanonical regulation of mTORC1 in 4T07 cells.

To determine whether genomic alterations in PI3K-AKT-mTORC1 pathway genes could explain the differences between 4T1 and 4T07 cells, we performed WGS on both cell lines, in reference to gDNA from a female BALB/c mouse. Despite high sequencing coverage (see Supplementary Methods), we found no significant differential genomic alterations of established functional consequences in the PI3K-AKT-mTORC1 pathway genes (33) to explain the differential dependence and signaling activity (Fig. 3D). We identified two missense mutations in 4T07 cells: Pik3ca (p.A423S) and Rictor (p.R576S; Fig. 3D; Supplementary Table S6). Both mutations occur in conserved residues and are predicted to have a moderate impact, but they have not been found in human tumors (n = 75,661 patients; see Supplementary Methods), suggesting limited functional relevance. We also noted differential copy number alterations in Pten, Inpp4b, Mtor, Prr5, Tsc1, and Akt1s1 (Fig. 3D). Only alterations in Inpp4b, Tsc1, and Akt1s1 correlated with mRNA expression patterns between the two lines (Supplementary Tables S7 and S8). Interestingly, 4T1 cells demonstrated genomic gain, correlated with mRNA overexpression, of Inpp4b, a tumor suppressor and a negative regulator of PI3K activity that is frequently lost in basal breast cancer (Fig. 3D; Supplementary Table S8; refs. 34, 35). Conversely, 4T1 cells showed losses in Tsc1 and Akt1s1, which align with their identification as proliferation suppressors of these cells in our CRISPR screens (Fig. 1D). Although these results suggest that 4T1 cells are wired to restrict the negative regulators of the canonical PI3K pathway, none of these could explain how mTORC1 activity is elevated in 4T07 cells.

mTORC1 activity is regulated by both growth factors and amino acids (30). Growth factors act mainly through PIK3CA-AKT (30), whereas amino acids act through various mediators, including PIK3C3 (36). Hence, we examined mTORC1 activity under different nutritional conditions to identify key regulatory mechanisms in 4T07 cells. Compared with 4T1 cells, 4T07 cells exhibited higher basal mTORC1 activity under serum-fed conditions and maintained residual mTORC1 activity under starvation (Fig. 3E). Amino acid starvation following overnight serum starvation completely abolished mTORC1 activity in 4T07 cells (Fig. 3E), whereas amino acid refeeding in serum-free conditions restored the activity to baseline levels (Fig. 3E). Similarly, amino acid refeeding induced mTORC1 activity in 4T1 cells. These observations suggest differential regulation of mTORC1 in the two cell lines and highlight the potential significance of amino acid signaling in 4T07 cells.

In addition to stimulating mTOR tethering on lysosomal surfaces to allow its consequent activation (30), amino acids activate PIK3C3. The latter mediates mTORC1 activity through different mechanisms, such as lysosomal positioning, in which mTORC1 activity is higher when lysosomes are positioned in the periphery compared with the perinuclear region (36–39). This led us to investigate lysosomal mTORC1 levels and lysosomal positioning in 4T1 and 4T07 cells.

We assessed the colocalization of Rptor (core component of mTORC1) and the phospho-S2448 residue of mTOR (more specific to mTORC1 than mTORC2; ref. 40) with the lysosomal marker Lamp1 (Fig. 4A and B). Under basal conditions, we observed a diffuse staining pattern for pMtor, with colocalization with Lamp1-positive lysosomes in both cell lines (Fig. 4A and B, first row). Amino acid starvation localized mTORC1 to small, diffuse cytoplasmic puncta (Fig. 4A and B, middle; ref. 41) and decreased lysosomal mTORC1 levels in both cell lines (Fig. 4A–C). These effects were rescued to baseline by amino acid refeeding (Fig. 4A–C, bottom). We also found Rptor in the cytoplasm, where it colocalizes with Lamp1, and in the nucleus (Fig. 4A and B). However, the intact mTORC1 was primarily localized in the cytoplasm in both 4T1 and 4T07 cells, with no pMtor signal in the nucleus, consistent with findings in other cells (42, 43). Notably, 4T07 cells exhibited 1.4-fold higher levels of lysosomal mTORC1 than 4T1 cells under fed conditions (Fig. 4C). We also observed distinct lysosomal positioning: in 4T07 cells, Lamp1-positive lysosomes accumulate at the tip of cell extensions, whereas in 4T1 cells, they accumulated in the perinuclear region (Fig. 4A, B, and D; Supplementary Fig. S5E–S5G). This peripheral positioning remained unchanged in 4T07 cells when manipulating the nutritional conditions that affected the mTORC1 lysosomal colocalization (Fig. 4D). These results suggest that lysosomes in 4T07 cells are preferentially positioned at the cellular periphery compared with 4T1 cells, providing mechanistic insights into their differential mTORC1 activity.

We investigated whether Pik3c3 regulates mTORC1 activity and/or lysosomal positioning in 4T1 and 4T07 cells. Treatment of 4T07 cells with 1 μmol/L of VPS34IN1 for 2 hours reduced mTORC1 activity by 50% (Fig. 5A and B). A similar trend was observed in 4T1 cells though with higher variability due to the low basal level of mTORC1 activity (Fig. 5A and B). Notably, VPS34IN1 blocked the response of 4T07 cells to amino acid refeeding after starvation (Supplementary Fig. S6A), aligning with known functions of PIK3C3 (36–38). We then examined the impact of VPS34IN1 on lysosomal positioning and noticed a 10-fold shift of lysosomes from the periphery to the perinuclear area, without affecting mTORC1 tethering to the lysosomes in 4T07 cells (Fig. 5C and D). In 4T1 cells, VPS34IN1 further reduced the basal percentage of peripheral lysosomes by ∼3-fold compared with the control (Fig. 5C and D). To confirm that these effects were specific to Pik3c3 and not due to off-target actions of VPS34IN1, we generated 4T07 cell lines with doxycycline-inducible Pik3c3-targeting or scrambled shRNAs (Supplementary Fig. S6B). Inducing the expression of these shRNAs demonstrated that the knockdown of Pik3c3 also shifts the lysosomes from the periphery to the perinuclear region (Supplementary Fig. S6C and S6D). These results demonstrate that Pik3c3 regulates mTORC1 activity and lysosomal positioning in both 4T1 and 4T07 cells, with varying magnitudes due to differences in basal mTORC1 activity.

Next, we assessed whether VPS34IN1-mediated suppression of mTORC1 activity could sensitize 4T07 cells to BYL719 treatments. 4T07 cells were treated either with BYL719 alone or in combination with VPS34IN1, and their proliferation was monitored over 72 hours. The combination significantly reduced the proliferation rate of 4T07, particularly at high doses of BYL719, resulting in a 30% decrease at 40 μmol/L (Supplementary Fig. S6E). We examined whether VPS34IN1 could enhance the sensitivity of 4T07 cells to Pik3ca inhibition and found that treating serum-starved 4T07 cells with BYL719 and VPS34IN1 significantly decreased mTORC1 activity in contrast to single-drug treatments (Supplementary Fig. S6F and S6G). However, VPS34IN1 did not enhance the BYL719 inhibition of mTORC1 activity after stimulating 4T07 cells with insulin (Supplementary Fig. S6F and S6G), suggesting that Pik3c3 primarily regulates mTORC1 through amino acid signaling (36).

We questioned whether the differential dependence on and activity of Pik3c3 between the 4T1 and 4T07 could be attributed to specific genomic or genetic events. We examined the WGS data for the genomic status of the three PI3K classes. Besides the Pik3ca p.A423S missense mutation, the 4T07 cells harbored a Pik3c3 missense mutation at a conserved residue (p.E562K; Supplementary Fig. S7A). This residue is located in a loop of the kinase domain of PIK3C3 (Supplementary Fig. S7B and S7C; ref. 44) and has been observed to be mutated in two patients (E562A, malignant peripheral nerve sheath tumor; E562K, penile carcinoma; Supplementary Methods). However, the oncogenic potential of these mutations is unknown. In our analysis, the E562K mutation is predicted to have a moderate effect (Supplementary Table S6). To determine whether this mutation increases PIK3C3 activity, we generated HeLa Flp-In T-REx cells expressing either wild-type or mutated PIK3C3 in a tetracycline-inducible manner. Induction of both forms of PIK3C3 did not result in a significant difference in mTORC1 activity (Supplementary Fig. S7D), suggesting that this mutation is unlikely to be responsible for the 4T07 cells’ mTORC1 hyperactivity. Overall, we did not identify any other notable genomic differences between the two lines that could explain the observed molecular and biological phenotypes (Supplementary Fig. S7A).

Next, we examined whether the phenomena observed in the 4T1 and 4T07 models were also present in human breast cancer. To do this, we examined early passages of PDX-derived breast cancer cell lines for mTORC1 activity (Supplementary Fig. S8A and S8B). This identified two potential PDX cell lines (1915 and 1735; ref. 15) with high mTORC1 activity. We focused on the PDX line 1915 because it originated from a patient with basal breast cancer (15), similar to 4T1/4T07 cells, and because the associated patient experienced a metastatic relapse 1.5 years after tumor resection (Supplementary Fig. S8C). WGS of the patient tumor and the PDX indicated the lack of PIK3CA activating hotspot mutations or PTEN deletion (like 4T1/4T07 cells), which are common causes of mTORC1 hyperactivity (Supplementary Fig. S8C). PDX1915 had an amplification in the PIK3CA gene without corresponding overexpression in its transcript level compared with PDXs with wild-type or altered PIK3CA genes (Supplementary Fig. S8D). We tested whether inhibiting PIK3C3 activity could reduce mTORC1 activity in PDX1915 cells and found that VPS34IN1 treatment decreased mTORC1 activity by approximately 50% (Supplementary Fig. S8E and S8F). VPS34IN1 also significantly reduced the percentage of peripheral lysosomes in these cells (Fig. 5E and F). Furthermore, VPS34IN1 increased the percentage of cells with tubulated lysosomes (Fig. 5E and G), consistent with previous studies (45). To extend these findings, we screened three additional human breast cancer cell lines: CAMA-1, T47D, and HCC70. These lines differ in their breast cancer intrinsic subtype and dependency on PIK3C3 (Supplementary Fig. S8G). VPS34IN1 treatment reduced mTORC1 activity by approximately 50% in all three lines (Supplementary Fig. S8H and S8I). Moreover, this treatment reduced the percentage of peripheral lysosomes in T47D and HCC70 but not in CAMA-1 cells (Fig. 5H–M).

In summary, inhibiting PIK3C3 activity in six murine and human breast cancer cell lines reduced mTORC1 activity and shifted peripheral lysosomes to the perinuclear region in five of them. This supports that PIK3C3 is an upstream mediator of mTORC1 activity in breast cancer, partially through lysosomal positioning.

We next investigated whether mTORC1 activity could predict relapse incidence and/or poor clinical outcomes in patients with breast cancer. Previous studies have shown that high expression of RPS6KB1(S6K1) correlates with worse distant metastasis-free survival in patients with early breast cancer (46). However, expression levels alone may not fully reflect high mTORC1 activity. To address this, we used gene set variation analysis to generate an “mTORC1 score” for each tumor in the METABRIC and TCGA-BRCA cohorts, based on a previously annotated and curated mTORC1 signaling-mediated gene signature (Fig. 6A; ref. 47). We observed that basal and HER2-positive breast tumors exhibited higher mTORC1 activity scores compared with the less aggressive luminal subtypes (Fig. 6B; ref. 2). Furthermore, patients with tumors characterized by a high mTORC1 signature score tended to have a poorer prognosis compared with those with low scores in the METABRIC dataset (P < 0.0001) and with a trend toward a similar correlation in the TCGA dataset (P = 0.1; Fig. 6C). These results align with the established impact of mTORC1 activity in human cancer (48). To explore the clinical significance of PIK3C3 activity, we generated a PIK3C3-specific gene signature (see supplementary Methods). We found that basal breast tumors had the highest PIK3C3 score in the METABRIC and TCGA datasets (Fig. 6D). Similar to the mTORC1 signature, patients with high PIK3C3 scores had worse prognosis when compared with those with low PIK3C3 scores in the METABRIC (P < 0.0001) and TCGA (P = 0.022) datasets (Fig. 6E). These observations highlight the potential value of targeting the PIK3C3-mTORC1 signaling axis in breast cancer.

We explored whether targeting the PIK3C3-mTORC1 pathway in animal models could have a beneficial outcome on breast cancer progression and metastasis burden. We performed spontaneous metastasis assays using the 4T07 and 4T1 models with or without treatment with VPS34IN1. 4T07 cells were implanted in the mammary fat pad of immunocompetent BALB/c mice, and primary tumors were allowed to grow for 3 weeks before the mice were randomly assigned to receive either vehicle or VPS34IN1 (Fig. 7A). After 6 days of treatment, the mice were sacrificed for analysis. There was no significant difference in primary tumor weight between the two groups, suggesting no effect on primary tumor burden (Fig. 7B). We then asked whether VPS34IN1 affected the DTC burden. Therefore, we dissociated the lungs of these mice and cultured the isolated cells in vitro in thioguanine-supplemented media to select for 4T07 cells (resistant to thioguanine) for a week, toward performing a colony-formation assay as an indicator of the DTC burden in the 4T07 model, where overt metastases are not observed, as previously established (9, 10). Indeed, VPS34IN1-treated mice had approximately a 3-fold reduction in DTC burden compared with vehicle-treated mice (Fig. 7C). To further investigate whether Pik3c3 inhibition could suppress the metastatic progression of the dormancy-prone cells when implanted under metastasis-permissive conditions, we implanted 4T07 cells expressing GFP and luciferase (4T07-TGL; ref. 12) in the mammary fat pad of nude mice (Supplementary Fig. S9A). One week after implantation, the mice were randomized to receive either the vehicle or VPS34IN1 (Supplementary Fig. S9A). Six days after treatment initiation, half of the mice were sacrificed for early time point analysis (Supplementary Fig. S9A). There was no difference in tumor weight between the two groups (Supplementary Fig. S9B). However, VPS34IN1-treated mice had a significantly lower DTC burden compared with the control group as assessed by bioluminescence imaging (Supplementary Fig. S9C and S9D). Further analysis was performed on the second group of mice after 12 days of treatment (3 weeks after implantation) with either VPS34IN1 or vehicle. Although there was no statistically significant difference in tumor weight between the two experimental groups (Supplementary Fig. S9B), the incidence of visible metastases in the VPS34IN1-treated group (1/5 mice) was 40% of that of the control group (3/6 mice; Supplementary Fig. S9E). Analyzing the digested lungs of mice that showed no visible lung metastases suggested that the VPS34IN1-treated mice demonstrated a trend toward lower DTC burden compared with controls (P = 0.06; Supplementary Fig. S9F). Next, we asked whether VPS34IN1 affected mTORC1 activity in vivo. We found lower mTORC1 activity in the VPS34IN1-treated mice versus the control group (Supplementary Fig. S9G and S9H). These data suggest that pharmacologic inhibition of the Pik3c3-mTORC1 pathway decreases the DTC burden, hence reducing the incidence of metastasis.

To investigate the role of Pik3c3 activity in metastatic progression in the 4T1 model, we implanted 4T1 cells in the mammary fat pads of immunocompetent BALB/c mice. Two weeks later, the primary tumors were surgically resected (Fig. 7D). After recovery, the mice were randomized to receive either VPS34IN1 or vehicle treatment for 1 week, followed by an assessment of lung metastasis burden via immunohistochemistry. There was no difference in the incidence of metastasis, as both experimental groups showed a 100% incidence rate (Fig. 7E). Moreover, there was no significant difference in the number or size of metastatic lesions (Fig. 7F–H). To maintain a consistent comparison between the 4T1 and 4T07 models, we implanted 4T1-TGL cells into the mammary fat pads of nude mice. After 1 week, the mice were treated with either VPS34IN1 or vehicle for 2 weeks before analyzing the primary tumor and metastasis burdens (Supplementary Fig. S10A). There were no significant differences between the two groups in terms of primary tumor burden (Supplementary Fig. S10B) or metastasis incidence (Supplementary Fig. S10C) despite a trend toward reduced mTORC1 activity in the 4T1 primary tumors (Supplementary Fig. S10D and S10E). Collectively, these results demonstrate that pharmacologic inhibition of the Pik3c3-mTORC1 pathway can reduce breast cancer metastasis, particularly in models with peripheral lysosomal positioning and high mTORC1 activity.

To examine a potential role for PIK3C3 in additional models of tumor dormancy, we used an inducible genetically engineered mouse model that exhibits key features of breast cancer progression as it occurs in patients with breast cancer, including the survival of residual/dormant cancer cells following therapy and their eventual spontaneous recurrence (49–51). In this model, MMTV-rtTA;TetO-Her2, doxycycline (dox) administration induces expression of the Her2/neu oncogene, leading to mammary tumor formation. Subsequent withdrawal of dox induces oncogene downregulation and complete tumor regression, mimicking targeted therapies (Fig. 8A). However, a population of residual cells survives oncogene downregulation and resides in a dormant, nonproliferative state in the mammary gland, and these cells eventually reinitiate proliferation to form a recurrent tumor. Thus, in this model, primary cells represent a proliferative state, and recurrent cells represent a postdormancy state (Fig. 8A; refs. 51, 52).

In line with the rationale behind 4T1-4T07 CRISPR screens, we reasoned that fitness genes exclusively required for growth or survival in recurrent, but not primary, cells might regulate survival during dormancy or outgrowth in permissive conditions. To identify fitness metabolic genes in primary and recurrent tumor cells, we performed a CRISPR-Cas9 knockout screen on two cell lines derived from primary MMTV-rtTA;TetO-Her2 tumors (54074 and 99142) and two cell lines (42929 and 48316) derived from recurrent tumors. Cells were transduced with Cas9 and an sgRNA library targeting 421 metabolic genes. Cells were harvested to extract gDNA at day 0 and after 14 population doublings, and the abundance of sgRNAs was determined using NGS to identify depleted sgRNAs that target genes required for survival or growth (Fig. 8B). Interestingly, the Pik3c3 gene was identified as an essential gene (i.e., its targeting gRNAs were significantly depleted) only in the two recurrent lines but not in the primary lines (Fig. 8C). Furthermore, after merging the screen results from the four cell lines, Pik3c3 emerged as one of the 14 recurrent cell-selective hits (Fig. 8D). We next sought to validate the selective dependence of recurrent tumor cells on Pik3c3 using pharmacologic approaches. We treated the four primary or recurrent cell lines with increasing doses of VPS34IN1 for 3 days and measured cell viability at the experimental endpoint. Consistent with the CRISPR screen results, recurrent cells were significantly more sensitive to VPS34IN1 than primary cells (Fig. 8E). Collectively, these results suggest that Pik3c3 is indeed a fitness gene in dormancy-prone HER2-dependent breast cancer cells.

We further tested whether Pik3c3 inhibition could directly target and eliminate dormant cells in this HER2-dependent model. To tackle this, we deprived the 54074 primary cell line of dox for a week to induce HER2 downregulation, hence forcing them into dormancy, as established before (53). Then, cells were treated with either VPS34IN1 or DMSO, and the burden of cell death was quantified after 24 or 48 hours (Fig. 8F). Expectedly, in line with prior results, VPS34IN1 did not induce any significant change in the cell death burden after 24 or 48 hours in cells maintained in dox (i.e., proliferative state). Strikingly, however, 24- and 48-hour treatments with VPS34IN1 induced about 3- and 10-fold increases in cell death burden in dox-deprived dormant cells, respectively (Fig. 8G). Taken together, these results strongly suggest that Pik3c3 is a therapeutic vulnerability for dormancy-prone and dormant HER2-driven breast cancer cells.

The landscape of the cancer cells’ intrinsic factors contributing to their metastatic fate and response to therapies remains poorly outlined. In this study, we compared two related basal B–like breast cancer models, 4T1 and 4T07, which differ in their metastatic capabilities and dormancy potential. We show that metastatic cells exhibit higher PI3K activity and depend on it more than dormancy-prone cells. These observations align with earlier reports demonstrating that bone marrow–derived DTCs from patients with breast cancer during the metastatic latency period show low PI3K activity (32, 54). Notably, despite the lower PI3K-AKT activity observed in the dormancy-prone 4T07 cells (this study) and in D-HEp3 cells (a model of dormant human head and neck carcinoma; ref. 32), both cell lines exhibited higher mTORC1 activity compared with their metastatic counterparts, 4T1 and T-HEp3 (32), respectively.

A surprising finding in our study is that peripheral lysosomal positioning contributes to elevated mTORC1 activity in the dormancy-prone cells despite their low PI3K activity. Previous studies have established that peripheral localization of lysosomes enhances mTORC1 activation, whereas perinuclear clustering reduces this activity and promotes macroautophagic dynamics (38, 39). Our results suggest that the 4T1-4T07 cell line pair may follow this binary model. Exploring whether there is a differential autophagic potential between these two lines would be valuable in future studies, given the conflicting reports on the role of autophagy in dormancy (22, 55).

PIK3C3 activates mTORC1 (36) through various mechanisms (28, 31, 38, 56), including promoting the anterograde transport of lysosomes to the cell periphery, in which mTORC1 becomes activated (38). We found that inhibiting Pik3c3 activity reversed the persistent peripheral lysosomal positioning and the associated high mTORC1 activity in dormancy-prone cells. This inhibition also reduced the ability of these cells to colonize lung tissues, suggesting that this molecular machinery plays a role in successful metastasis. Consistent with a conserved role in human tumors, PIK3C3 inhibition mediated similar effects in three human breast cancer lines with different genomic alterations and intrinsic subtypes. In one of the six murine and human cell lines we tested, CAMA-1, the reduction in mTORC1 activity was not accompanied by a shift in lysosomal positioning following PIK3C3 inhibition. This not only highlights the heterogeneity between different models but also indicates that PIK3C3 may regulate mTORC1 activity through mechanisms independent of lysosomal positioning (31). Future research in this area could enhance our understanding of the genetic and signaling dependence of dormancy-prone cells.

mTORC1 and PIK3C3 have an entangled bilateral relationship. Although we focused on PIK3C3’s role upstream of mTORC1, it has been reported that mTORC1 also acts upstream of PIK3C3 to mediate lysosomal regeneration following starvation-induced autophagy, supporting cell survival (45, 57). These studies have identified tubulated lysosomes as transient autophagy-induced structures that serve as proto-lysosomes during lysosomal regeneration (57). However, these structures are also observed under nutrient-rich conditions (45), with their formation being dependent on mTORC1 activity. In the PDX1915 cells, we found that the presence of these structures was augmented upon treatment with VPS34IN1 despite the reduction in mTORC1 activity. It is unclear whether this increased tubulation is a feedback mechanism induced by reduced mTORC1 activity to maintain a basal lysosomal pool for mTORC1 activation. The potential connection among lysosomal positioning, mTORC1 activity, autophagy, and the formation of lysosomal tubules warrants further investigation.

Our findings provide insights into research directions of potential clinical relevance (58). Gene signatures associated with either mTORC1 or PIK3C3 correlated with clinical outcomes in breast cancer. However, when stratifying patients by intrinsic breast cancer subtype, no statistically significant correlation was observed with either signature in the TCGA or METABRIC datasets. One exception is a significant correlation between the PIK3C3 signature and outcomes in patients with luminal A breast cancer (not shown). This might be due to the sample size, suggesting that analyzing larger cohorts with different subtypes could provide more insights. It is important to note that the PIK3C3 signature used was generated from datasets from pancreatic and lung cancers. However, notably, the identified 183 genes seem to be specific to PIK3C3 because only 4 of them were shared with the 200-gene mTORC1 Hallmark gene signature, arguing against a promiscuous set of genes for highly proliferative tumors. Given the relatively unexplored roles of PIK3C3 in breast cancer pathogenesis, future studies aiming to generate breast cancer–specific PIK3C3 signatures might be significant. From a therapeutic perspective, gedatolisib, a dual class I PI3K/mTOR inhibitor, failed to reduce the metastatic burden in two breast cancer models when used alone or in combination with standard-of-care chemotherapy (59). Another study demonstrated that long-term pharmacologic inhibition of PI3K or AKT induces a PIK3C3-SGK3 signaling axis to activate mTORC1, and targeting this axis can revert the induced resistance to PI3K-AKT inhibition (31). Our work demonstrating that dormancy-prone cells do not depend on class I PI3K and are resistant to its pharmacologic inhibition, while being dependent on class III PI3K, complements these studies. Future studies should explore whether the combinatory inhibition of class I and class III PI3K and mTORC1 could effectively reduce the metastatic burden in vivo. Importantly, the contribution of the microenvironment to the observed effect of inhibiting PIK3C3 on breast cancer metastasis should be considered.

Although our CRISPR screening approach revealed novel insights into the intrinsic differences between cells of differential metastatic capacity, it has limitations. The screens were performed for technical feasibility in an in vitro setting that does not necessarily recapitulate the impact of the cross-talk with the microenvironment on gene essentiality and signaling pathways’ activity. Nonetheless, we extensively validated the PI3K(s)-mTORC1 signaling circuits in vitro and in vivo. Furthermore, we extended and validated our main finding, the essentiality of PIK3C3 in breast cancer metastatic dormancy, in cells originating from the MMTV-rtTA;TetO-Her2 model. Importantly, the transcriptomic profile of dormant residual tumor cells from the MMTV-rtTA;TetO-Her2 model mimicked that of cells in a state of metastatic dormancy (51). This provides support for the notion of PIK3C3 being a potential therapeutic vulnerability during the clinical dormancy phase preceding clinically detectable metastasis.

In summary, the work described here provides novel insights into the intrinsic differences between cells of differential metastatic capacity that can guide future explorations of the biology and therapeutic vulnerabilities of breast cancer metastatic relapses.

C.L. Kleinman reports grants from the Canadian Institutes of Health Research and other support from the Fonds de Recherche du Québec - Santé during the conduct of the study. A.P. Gomes reports other support from MetroBiotech outside the submitted work. J.A. Aguirre-Ghiso reports personal fees from the Samuel Waxman Cancer Research Foundation and Astrin Biosciences and nonfinancial support from HiberCell LLC outside the submitted work and has a patent for WO2019191115A1/EP-3775171-B1 issued. No disclosures were reported by the other authors.

I.E. Elkholi: Conceptualization, formal analysis, supervision, investigation, writing–original draft, writing–review and editing. A. Robert: Formal analysis, investigation, writing–review and editing. C. Malouf: Formal analysis, investigation, writing–review and editing. J.L. Wu: Formal analysis, investigation. H. Kuasne: Formal analysis, investigation, writing–review and editing. S. Drapela: Formal analysis, investigation, writing–review and editing. G. Macleod: Formal analysis, investigation, writing–review and editing. S. Hébert: Formal analysis, investigation, writing–review and editing. A. Pacis: Formal analysis, investigation, writing–review and editing. V. Calderon: Formal analysis, writing–review and editing. C.L. Kleinman: Formal analysis, writing–review and editing. A.P. Gomes: Formal analysis, writing–review and editing. J.V. Alvarez: Formal analysis, investigation, writing–review and editing. J.A. Aguirre-Ghiso: Formal analysis, writing–review and editing. M. Park: Resources, formal analysis, writing–review and editing. S. Angers: Formal analysis, writing–review and editing. J.-F. Côté: Conceptualization, formal analysis, supervision, funding acquisition, project administration, writing–review and editing.

We would like to thank Dr. Marie-Anne Goyette for providing constructive feedback on the manuscript. We acknowledge Dr. Flippo Giancotti (4T07-TGL cell line) and Dr. Vladimir Ponomarev (TGL reporter) for providing reagents. We thank Drs. Sandra Turcotte, Philippe Roux, Matthew Smith, Anne-Marie Fortier, and Deepak Singh for fruitful discussions. We acknowledge the “Genome Engineering using CRISPR/Cas Systems” Google discussion group and Drs. Julia Joung and Jonathan Boulais for helpful discussions on technical steps and the bioinformatic analysis of the CRISPR screens. We express our gratitude to the Montreal Clinical Research Institute (IRCM) facilities staff (Dr. Dominic Filion, Manon Laprise, and Dr. Odile Neyret) for their expert technical help involved in this work. This work was supported by operating grants from the Cancer Research Society (operating grant #25244 to J.-F. Côté), Canadian Institutes of Health Research (foundation grant #FDN-143281 to M. Park and operating grant #PJT-156086 to C.L. Kleinman), Canadian Cancer Society and Oncopole grants (to M. Park), NIH/NCI R01CA292658 (to J.V. Alvarez), and an American Cancer Society Research Scholar Award (RSG-22-164-01-MM to A.P. Gomes). PDX cell lines were obtained from the breast tissue bank at McGill University, supported by grants to the Réseau Cancer banque de tumeurs from the Fonds de la Recherche du Québec - Santé (FRQS) and the Quebec Breast Cancer Foundation (to M. Park). Data computational analyses were enabled by compute and storage resources provided by Compute Canada and Calcul Québec. I.E. Elkholi was a recipient of an FRQS Doctoral Scholarship, an IRCM Foundation-TD scholarship, and a Peter Quinlan Postdoctoral Research Fellowship in Oncology (McGill University). S. Drapela is a recipient of a Miles for Moffitt postdoctoral fellowship. C.L. Kleinman is the recipient of an FRQS Salary Award. J.A. Aguirre-Ghiso is supported by the following awards: NIH/NCI (CA109182, CA253977, CA284085, P30CA013330), the Mark Foundation ASPIRE program, the Gurwin Foundation, DoD-MRP, DoD-BCRP, Melanoma Research Alliance, and the Rose C. Falkenstein Chair in Cancer Research. J.A. Aguirre-Ghiso is a Samuel Waxman Cancer Research Foundation investigator. M. Park is a distinguished James McGill professor and holds the Diane and Sal Guerrera Chair in Cancer Genetics. J.-F. Côté holds the Canada Research Chair Tier 1 in Signalling in Cancer and Metastasis and the Alain Fontaine Chair in Cancer Research from the IRCM Foundation.