Macropinocytosis and Vascularization Determine Response to mTOR Inhibitors in Lung Squamous Cell Carcinoma.

Introduction
Lung cancer is a highly heterogeneous disease that is histologically separated into non-small cell lung cancer (NSCLC, ~85% of lung cancer) and small cell lung cancer (SCLC, ~15% of lung cancer) (1). NSCLC is further divided into lung adenocarcinoma (LUAD), lung squamous cell carcinoma (LUSC), and large cell carcinoma. Lung squamous cell carcinoma is an aggressive subtype of NSCLC with limited precision medicine treatment options leading to poor clinical outcomes (2),(3). Specifically, LUSC tumors have fewer genetic mutations in classical oncogenic drivers that are commonly found in LUAD, such as KRAS, EGFR, ALK or ROS1 (4),(5), and instead display a high frequency of amplifications and activating mutations in the PI3K-AKT-mTOR signaling pathway, which supports tumor growth and metabolism (4). Data from the clinical trials with the current frontline therapy of chemo-immunotherapy regiments, show that the great majority of patients with LUSC eventually experience progression of disease with limited effective later line therapy options (6),(7).
The mechanistic target of rapamycin (mTOR) signaling pathway is a central regulator of cellular growth, metabolism, and survival, integrating diverse signals from nutrients, growth factors, energy status, and stress to control critical cellular processes such as protein synthesis, autophagy, and metabolism (7),(8). The mTOR pathway is frequently activated in LUSC tumors, with 50% of LUSC cases harboring amplifications and mutations in the PI3KCA and PTEN genes (4), leading to the activation of the mTOR pathway which drives tumor proliferation. Among many targets of the active mTOR signaling pathway are proteins involved in the control of protein translation, including p70 ribosomal protein kinase S6 (S6K), ribosomal protein S6 (S6), and eukaryotic initiation factor 4E-binding protein 1 (4E-BP1); their phosphorylation leads to increased protein translation and cell growth. This mTOR activation is often linked to the reprograming of cellular metabolism, promoting anabolic processes that support accelerated growth and cell survival. Given its integral role in tumorigenesis, the mTOR pathway has represented an attractive therapeutic target, with several mTOR inhibitors currently under investigation in clinical trials (9) (10).
One of the small molecule inhibitors of mTOR is TAK228, a highly selective and potent mTORC1/2 inhibitor (11) currently in phase I/II clinical testing (NCT02417701, NCT04250545, NCT05022394, NCT02503722). However, the latest data from the TAK228 clinical trials indicate that the efficacy of this regimen varies among patients (12),(13). As such, it is critical to gain a deeper pre-clinical understanding of the molecular mechanisms underlying resistance to mTOR pathway inhibition in lung cancer to inform the development of more effective strategies to block mTOR signaling in lung cancer and provide meaningful clinical benefit to patients whose tumors are reliant on this pathway for growth.
LUSC are frequently hypermetabolic due to high levels of glucose and glutamine uptake (14),(15), (16),(17),(18) which not only fuel tumor cell growth, but also create an unfavorable tumor microenvironment, contributing to the poor response to immunotherapy (19),(20) and poor overall survival (21). Glucose and glutamine function as critical carbon sources that drive tumor biosynthesis and energy production through the essential pathways of glycolysis and glutaminolysis. These metabolic processes prove fundamental to sustaining aggressive cancer cell growth and proliferation (22, 23). The recognition that these metabolic alterations play a central role in tumorigenesis has led to the emergence of metabolism-based therapies as a promising therapeutic strategy in oncology (24),(25),(26),(27),(28),(29),(30),(31),(32). Multiple studies have demonstrated that concurrent targeting of both glycolysis and glutaminolysis pathways substantially impairs cancer cell viability (26, 33–35). Using personalized precision medicine based combinations of inhibitors of the major metabolic pathways is an attractive alternative to chemotherapy, as these therapies seek to disrupt the metabolic networks that sustain tumor growth, thereby inhibiting tumor progression and improving clinical outcomes. Here we present data that identified two key mechanisms that LUSCs use to evade targeted metabolic treatment with mTOR and GLS inhibitors.

Materials and Methods

Cell lines and reagents
Following cell lines were used (source; DepMap ID; histology; mutations; RRID): H1703 (American Type Culture Collection (ATCC): CRL-5889; DepMap ID: ACH-000747; LUSC; CDKN2A D84V; RRID:CVCL_1490), H520 (ATCC: HTB-182, DepMap ID: ACH-000395; LUSC; ATM P383A, TP53 W146Stop, CDKN2A G45VfsStop; RRID:CVCL_1566), H460 (ATCC: HTB-177, DepMap ID: ACH-000463; Large Cell Lung Carcinoma (LCLC); KRAS Q61H, PI3KCA E454K, STK11 Q37Stop, KEAP1 D236H; RRID:CVCL_0459), H2170 (ATCC: CRL-5928, DepMap ID: ACH-000481; LUSC; RHOA G17V, TP53 R158G, KEAP1 R336Stop; RRID:CVCL_1535); HCC15 (DSMZ: ACC 496, DepMap ID: ACH-000878; LUSC; NRAS Q61K, TP53 259V, CTNNB1 S45F, KEAP1 G364C; RRID:CVCL_2057), SW900 (ATCC: HTB-59; DepMap ID: ACH-000669; LUSC; KRAS G12V, TP53 Q167Stop; RRID:CVCL_1731). All cell lines tested negative for mycoplasma contamination and authenticated by short tandem repeats (STR). Cells were maintained DMEM (Corning, Cat # 10–013-CV) with 5% FBS (Gemini) and penicillin/streptomycin (Gibco, Cat # 15140163) in a humidified 5% CO2 incubator. Plasmids encoding VEGFA and VEGFC were obtained from Addgene (pQCXIP-VEGFA121 was a gift from Michael Grusch (Addgene plasmid # 73017, RRID:Addgene_73017); pQCXIP-VEGFC was a gift from Michael Grusch (Addgene plasmid # 73012, RRID:Addgene_73012). Viral preps were made using standard protocols with 293FT cells (Life Technologies, (RRID:CVCL_6911) and pCL-Ampho (Novus Bio). Viral preps were used to create H2170, HCC15 and H1703 cells overexpressing either VEGFA or VEGFC.
TAK228 was purchased from Chemietek (Cat # CT-INK128); for in vitro studies it was dissolved in DMSO, for in vivo studies it was dissolved in 1-methyl-2-pyrrolidinone (NMP), then diluted in 15% PEG-400 dissolved in water (25, 26). EIPA was purchased from Cayman Chemicals (Cat # 14406); for in vitro studies it was dissolved in DMSO, for in vivo studies it was dissolved in DMSO, then diluted in 15% PEG-400 dissolved in water. Pazopanib was purchased from Chemietek (Cat # CT-PAZ). For in vivo studies it was dissolve in DMSO, then diluted in 0.5% methylcellulose. DQ-BSA Red (Cat # D12051), Lysotracker Red (Cat # L7528), Lysotracker Green (Cat # L7526) were purchased from Life Technologies. Prolong Gold Antifade with DAPI (Cat # P36935) was purchased from Thermo Scientific. PROTAC for SLC9A1 (d9A-2) was purchased from Sigma (Cat # SML2978).

Metabolomics
Tissues (15–20 mg) were placed in a pre-filled bead mill tube with metal beads and 1 mL of ice-cold 80% methanol solution with 10nM trufluoromethanesulfonate was added, followed by tissue homogenization using a Bead Mill Homogenizer. Samples were then centrifuged at 12,000 rpm (4 °C) for ten minutes to remove precipitated cell material. Supernatant equivalent to 5–10 ug of protein was transferred to a glass tube and samples were dried using EZ-2 Elite evaporator (Genevac). Dried samples were stored in −80 °C freezer until further processing. Dried metabolites were reconstituted in 100 μL of a 50% acetonitrile (ACN) 50% dH20 solution. Samples were vortexed and spun down for 10 min at 17,000g. 70 μL of the supernatant was then transferred to HPLC glass vials. 10 μL of these metabolite solutions were injected per analysis. Samples were run on a Vanquish (Thermo Scientific) UHPLC system with mobile phase A (20mM ammonium carbonate, pH 9.7) and mobile phase B (100% ACN) at a flow rate of 150 μL/min on a SeQuant ZIC-pHILIC Polymeric column (2.1 × 150 mm 5 μm, EMD Millipore) at 35°C. Separation was achieved with a linear gradient from 20% A to 80% A in 20 min followed by a linear gradient from 80% A to 20% A from 20 min to 20.5 min. 20% A was then held from 20.5 min to 28 min. The UHPLC was coupled to a Q-Exactive (Thermo Scientific) mass analyzer running in polarity switching mode with spray-voltage=3.2kV, sheath-gas=40, aux-gas=15, sweep-gas=1, aux-gas-temp=350°C, and capillary-temp=275°C. For both polarities mass scan settings were kept at full-scan-range = (70–1000), ms1-resolution=70,000, max-injection-time=250ms, and AGC-target=1E6. MS2 data was also collected from the top three most abundant singly-charged ions in each scan with normalized-collision-energy=35. Each of the resulting “.RAW” files was then centroided and converted into two “.mzXML” files (one for positive scans and one for negative scans) using msconvert from ProteoWizard (RRID:SCR_012056). These “.mzXML” files were imported into the MZmine 2 software package. Ion chromatograms were generated from MS1 spectra via the built-in Automated Data Analysis Pipeline (ADAP) chromatogram module and peaks were detected via the ADAP wavelets algorithm. Peaks were aligned across all samples via the Random sample consensus aligner module, gap-filled, and assigned identities using an exact mass MS1(+/−15ppm) and retention time RT (+/−0.5min) search of our in-house MS1-RT database. Peak boundaries and identifications were then further refined by manual curation. Peaks were quantified by area under the curve integration and exported as CSV files. If stable isotope tracing was used in the experiment, the peak areas were additionally processed via the R package AccuCor 2 to correct for natural isotope abundance. Peak areas for each sample were normalized by the measured area of the internal standard trifluoromethanesulfonate (present in the extraction buffer) and by the number of cells present in the extracted well.

RNASeq data analysis
RNA was extracted from snap frozen tissues and processed for NGS RNA Sequencing by UCLA The Technology Center for Genomics & Bioinformatics (TCGB) core. FASTQ files were aligned and quantified using the Galaxy platform (RRID:SCR_006281) (36–41). GSEA analysis was completed using the Galaxy platform (RRID:SCR_006281) (36)(bioRxiv 2021.060012).

Xenograft and PDX studies
Mice were housed in pathogen-free facilities at University of California Los Angeles (UCLA). All experimental procedures were approved by the UCLA Animal Research Committee (ARC). PDXs were established in our lab using 6–12 week old NOD-scid IL2Rgammanull (NSG) mice implanted with tumor tissue from patients undergoing surgeries. All patients provided written informed consent for use of samples in research under protocols approved by Institutional Review Boards at University of California, Los Angeles IRB # 10–001096 and at Long Beach Memorial Hospital IRB #208–13 (26). All studies were conducted in accordance with the Belmont Report and U.S. Common Rule. All three PDXs were established from patients with LUSC. PDX005: BRAF I300V, KEAP1 S275A R362Q N438Q, TP53 P72R, ATM M1040V. PDX007: HRAS Q61K, PIK3CA E545K, KEAP1 E306Q V316A M317V.
NOD.Cg-Prkdc scid IL2rg tm1Wjl / SzJ (NSG) mice were purchased from Jackson Laboratory (Cat # 005557) or from UCLA Radiation Oncology mouse core (ROC NSG). Female NSG mice between 6–12 weeks old were used for patient derived xenograft (PDX) studies and xenograft studies, except for experiment in Figure S5H, where male NSG mice were used.
NSG mice were injected subcutaneously into two lower flanks with 1×10^6 cells (H1703 and H2170) or 3.5×10^6 cells (HCC15 and H520) suspended in 75% PBS, 25% Matrigel (Corning) and allowed to grow to ~100–200 mm3 before treatment was started. Animals were randomly assigned to one of two groups: Vehicle or TAK228 (1 mg/kg, i.p., q.d.). For experiments in Figure 2K and S2F, mice were randomly assigned to one of four groups: Vehicle, TAK228 (1 mg/kg, i.p., q.d.), EIPA (7.5 mg/kg, s.c., q.d.), or TAK228+EIPA. For experiments in Figure 4A, mice were randomly assigned to one of four groups: Vehicle, TAK228 (1 mg/kg, i.p., q.d.), pazopanib (30 mg/kg p.o., q.d.), or TAK228+pazopanib. For experiments in Figure 4D and S5H, mice were randomly assigned to one of four groups, Vehicle, TAK228 (1 mg/kg, i.p., q.d.) + EIPA (7.5 mg/kg, s.c., q.d.), pazopanib (30 mg/kg p.o., q.d.), or TAK228+EIPA+pazopanib. For experiments in Figure 5A–C, mice were randomly assigned to one of two groups: Vehicle or TAK228 (1 mg/kg, i.p., q.d.) + 200 mg/kg CB-839. Body weights (g) and caliper measurements ((L×Ŵ2)/2 = mm3) were recorded three times a week. At the end of the study animal tissue was fixed in 10% buffered formalin or snap frozen in liquid nitrogen 2 hr after the final dose. For experiments with H2170, HCC15 and H1703 cell lines overexpressing VEGFA and VEGFC, equal numbers of VEGFA- and VEGFC-expressing cells were mixed prior to implantation. The total number of VEGFA- and VEGFC-expressing cells implanted was equal to the total number of cells expressing the control vector only. Investigators were not blinded to the treatments.

Immunohistochemistry and IHC analysis
Tissue isolation, fixation and staining procedures were performed as previously described (24–26, 33). Briefly, tumors were fixed for 24 hr in 10% buffered formalin and transferred to 70% ethanol. Tissues were processed and embedded by Translational Pathology Core Laboratory (TPCL) at UCLA. The following antibodies were used: phospho-4EBP1 (Thr37/46) (Cell Signaling Technology, #2855 1:800, RRID:AB_560835), anti-CD31 (Cell Signaling Technology, #77699, 1:400, RRID:AB_2722705). Slides were scanned onto a ScanScope AT (Aperio Technologies). Digital slides were analyzed with QuPath software (RRID:SCR_018257) (42) to determine percent positive area/cells for CD31 and p4EBP1.

Flow cytometry and microscopy assays
For flow cytometry with DQ-BSA Red and Lysotracker Green, cells were treated with Vehicle or 100 nM TAK228 for 72 hr. Next, cells were trypsinized, counted, and 250,000 cells were treated with 25ug/ml DQ-BSA Red or 1 uM Lystotracker Green for 30 min in the presence of either Vehicle, 100 nM TAK228, or 25 uM EIPA. Cells were washed twice with phenol-red free DMEM and fluorescence was acquired on BD LSRII flow cytometer in the UCLA Jonsson Comprehensive Cancer Center and Center for AIDS Research Flow Cytometry core facility. Data was analyzed using Flowing software. For microscopy with Lysotracker Red, cells were treated with Vehicle or 100 nM TAK228 for 72 hr, then stained with 250 nM Lysotracker Red for 30 min in presence of either Vehicle or 100 nM TAK228, washed in PBS, fixed in 4% paraformaldehyde (Electron Microscopy Sciences, Cat # 15713), mounted using Prolong Gold Antifade with DAPI (Thermo Scientific, Cat # P36931), and imaged on a Zeiss LSM 880 Confocal microscope at UCLA Broad Stem Cell Research Center Microscopy core facility.

Immunoblotting
After treatment, cells were quickly rinsed with PBS and then lysed in RIPA buffer (20 mM Tris pH 7.5, 150 mM NaCl, mM Na2EDTA, 1 mM EGTA, 1% NP-40, 1% sodium deoxycholate, 2.5 mM sodium pyrophosphate, 1 mM beta-glycerophosphate, 1 mM Na3VO4, 1 μg/ml leupeptin). Protein content was normalized using BCA protein kit (Pierce). Samples were resolved by SDS-PAGE, transferred to PVDF membrane (Thermo Scientific, Cat # 88518), and blocked in 5% non-fat dry milk. PVDF membranes were incubated with primary antibodies overnight: SLC9A1 (1:1000, Santa Cruz, sc-136239, RRID:AB_2191254), actin (1:5000, Cell Signaling Technologies, CST #3700, RRID:AB_2242334).

Cell viability
Cell viability was measured using trypan blue exclusion assay (ViCell XR, Beckman Coulter).

NanoString
NanoString analysis of tumors was carried out as previously described (43).

Quantification and statistical analysis
The statistical tests were calculated using GraphPad Prism 6 (RRID:SCR_002798). Details of the specific statistical analysis are indicated in the figure legends.

Data availability
RNASeq data generated by this study is available in the Gene Expression Omnibus (GEO) under GSE292426. All other materials are available upon reasonable request from the corresponding author. The TCGA LUSC data analyzed in this study were obtained from cBioPortal at www.cbioportal.org (44, 45).

Results

Global increase in amino acids identified in TAK228 resistant LUSC tumors
We previously demonstrated that LUSC tumors exhibit remarkable metabolic plasticity, enabling adaptation to TAK228 mTOR inhibition through enhanced glutamine uptake (26). Specifically, following TAK228 treatment, the contribution of glucose to tricarboxylic acid (TCA) cycle activity decreased, while the contribution of glutamine increased (26). Combined administration of TAK228 with CB-839, a small molecule inhibitor of glutaminase (GLS), which is a key enzyme in the glutaminolysis pathway, effectively halted tumor growth (26). Based on these findings, we hypothesized that TAK228-resistant tumors may upregulate the utilization of additional amino acids beyond glutamine. To test this hypothesis, we conducted amino acid profiling on tumor tissue extracted from xenografts following treatment with either vehicle or TAK228. Our analysis focused on comparative amino acid analysis between xenografts that were resistant to TAK228 vs xenografts in which TAK228 significantly slowed the rate of tumor growth. We examined the LUSC cell lines H1703, which did not show reduced tumor growth following treatment with TAK228 (26), and H2170, which did show reduced tumor growth following TAK228 treatment in xenograft models (26). As reported previously, H1703 xenografts did not exhibit a reduction in tumor volume after 3 weeks of TAK228 treatment (Figure 1A). Metabolomic analysis of tumor amino acid levels revealed a significant increase in glutamine, consistent with prior reports (26). Additionally, we observed a notable elevation in the relative levels of several other amino acids in TAK228-treated tumors compared to Vehicle-treated controls, including alanine, serine, proline, valine, threonine, leucine/isoleucine, asparagine, phenylalanine, arginine, tyrosine, and tryptophan (Figure 1B).
In contrast, H2170 xenografts, which showed reduced rate of tumor growth following 3 weeks of TAK228 treatment (Figure 1C), showed a significant decrease in glutamine levels, along with reductions in proline, valine, threonine, arginine, aspartate, glutamate, and tyrosine (Figure 1D). We next performed gene set enrichment analysis (GSEA) of RNA-Seq data, which revealed significant enrichment in the “Reactome Metabolism of Amino Acids and Derivatives” and “Reactome Cellular Response to Starvation” pathways in the TAK228 resistant H1703 xenografts (Figure 1E). In contrast, these pathways were not enriched in the H2170 xenografts (Figure 1F). These results demonstrate that tumors that failed to show reduced growth following TAK228 treatment upregulate multiple amino acids in addition to glutamine, while tumors that show reduced rate of tumor growth following TAK228 treatment fail to increase amino acid levels in response to TAK228 treatment.

Macropinocytosis as a mechanism of mTOR inhibitor evasion
To further investigate mTOR inhibitor resistance in LUSC, we considered possible supply routes by which the H1703 xenografts could increase amino acid levels following TAK228 treatment. First, we examined whether TAK228 treatment increased tumor vascularization, which could increase amino acid levels in these tumors and support growth. However, TAK228 treatment decreased the density of CD31 positive cells, marker of endothelial cells, in H1703 xenografts (Figure S1A, B), and TAK228 treatment did not significantly change density of CD31 positive cells in H2170 xenografts (Figure S1C, D). This suggested that alternative mechanisms are responsible for increased amino acid levels seen in H1703 xenografts.
We next hypothesized that the increased amino acid levels seen in mTOR resistant H1703 xenografts were driven by macropinocytosis, a form of endocytosis that has been reported in oncogene or tumor suppressor mutated cancers, including both pancreatic and lung cancer (46),(47),(48),(49),(50). Macropinocytosis is a multi-step process through which tumor cells internalize large molecules, such as albumin, which are engulfed into a macropinosome (51). This process requires activity of the sodium/hydrogen (Na+/H+) exchanger (SLC9A1) (52),(53). Upon fusion with the lysosome, albumin is degraded by lysosomal proteases, releasing amino acids (Figure 2A). Our hypothesis that macropinocytosis is contributing to the amino acid increase in mTOR resistant H1703 tumors (Figure 1A), was further supported by Gene Set Enrichment Analysis (GSEA) of RNA-Seq data obtained from H1703 xenografts, which revealed a lack of enrichment for pathways associated with amino acid transport across the cell membrane (Figure S2A).
To directly evaluate the extent of TAK228-induced macropinocytosis, we employed DQ-BSA, a fluorescently labeled version of bovine serum albumin (BSA). The fluorescence of DQ-BSA is quenched when the BSA is intact and becomes detectable only upon cleavage by lysosomal proteases (54) (Figure 2B). After treating cells with either vehicle or TAK228 for 72 hours, we added DQ-BSA with or without EIPA (5-[N-ethyl-N-isopropyl] amiloride), an inhibitor of Na+/H+ exchanger (SCL9A1) that is required for acidification near the plasma membrane and the ruffling of the cell membrane. EIPA can act as a efficient inhibitor of macropinocytosis (52),(53) (Figure 2C). Fluorescent signal intensity of DQ-BSA was measured by flow cytometry in six LUSC cell lines (55),(56). Increased DQ-BSA fluorescence was observed in H1703 and H520 cells treated with TAK228 (Figure 2D, E) compared to H2170 and HCC15 cells (Figure 2F, G). Importantly, EIPA treatment significantly decreased the fluorescent signal in all cell lines tested (Figure 2H). Along with the increase in DQ-BSA fluorescence, we observed a concurrent rise in the intensity of LysoTracker Green and LysoTracker Red signals, dyes that accumulate in lysosomes, with the most prominent increase seen in H1703 cells treated with TAK228 (Figure S2B–D). These results demonstrate that TAK228 treatment in H1703 and H520 cells upregulates macropinocytosis in response to TAK228, while TAK228 treatment in H2170 and HCC15 cells fail to substantially increase macropinocytosis.
Given that EIPA inhibits SLC9A1 function (57), we also utilized d9A-2, a recently described PROTAC (proteolysis-targeting chimera) that targets the SLC9A family of transporters (58), including SLC9A1. We first confirmed that d9A-2 treatment reduced SLC9A1 protein levels in H460, H1703, H2170, and H520 cells (Figure 2I). Next, we assessed the viability of eight human LUSC cell lines treated with either TAK228 or d9A-2 alone or in combination for 72 hours (Figure 2J). Although, both TAK228 and d9A-2 treatments as single agents significantly reduced cell viability, the combination of TAK228 and d9A-2 resulted in a further decrease in cell viability compared to TAK228 alone (Figure 2J).
Finally, we examined the effect of inhibiting both the mTOR pathway with TAK228 and macropinocytosis with EIPA on tumor growth in vivo. TAK228 resistant H1703 xenografts were treated with vehicle, TAK228, EIPA, or the combination of TAK228+EIPA for 18 days (Figure 2K). Consistent with our in vitro data, combination treatment significantly impaired tumor growth, resulting in a marked reduction in the increase of the tumor volume compared to either the vehicle or TAK228 groups, which coincided with changes in GSEA for the genes represented in the KEGG_Lysosome geneset (Figure 2L). The combination therapy was well tolerated and did not induce significant weight loss in tested mice (Figure S2E). Additionally, similar results were seen in H520 xenografts (Figures S2F, S2G). Collectively, our data suggest that macropinocytosis supports LUSC tumor proliferation in the presence of TAK228, while combinatorial inhibition of mTOR and macropinocytosis can significantly reduce growth in tumors that were resistant to TAK228. These results highlight a potential new therapeutic strategy for targeting both mTOR and macropinocytosis in LUSC.

Increased vascularization modulates response to TAK228 in vivo
While our data suggests that macropinocytosis contributes to LUSC mTOR inhibitor resistance, both in vitro and in vivo growth conditions fail to fully replicate the complexities of the tumor microenvironment observed in patients with lung tumors. Xenograft and patient-derived xenograft (PDX) models rely on subcutaneous tumor growth, which favors growth of tumors that are proliferative in hypoxic and poorly vascularized environments (59),(60),(61). Given the poorly vascularized nature of H1703 and H2170 xenografts (Figure S1A–D). we hypothesized that enhancing vascularization in these tumors could serve as an additional mechanism for mTOR resistance by increasing the availability of nutrients, such as amino acids.
To test this hypothesis, we overexpressed both VEGFA and VEGFC in two LUSC cell lines with low TAK228-induced macropinocytosis, H2170 and HCC15 (Figure 2F,G), and evaluated the in vivo response to TAK228. Since both VEGFA and VEGFC can bind to VEGFR2, a key receptor that promotes angiogenesis (62),(63), we overexpressed both VEGFA and VEGFC in the cells to maximize tumor vascularization. Remarkably, enhancing vascularization converted H2170 xenografts from tumors where TAK228 was able to slow the rate of tumor growth (Figure 3A) into tumors where TAK228 failed to reduce the rate of tumor growth (Figure 3B). Further, VEGFA/C expression led to almost four-fold increase in vascular density in these xenografts (Figures 3C–F). In both low- and highly vascularized tumors (Figure 3C–F, S3A–D), TAK228 treatment was able to efficiently suppress activity of mTOR pathway, as evidenced by decreased p4EBP1 staining (Figure 3C–F). Similar results were observed in HCC15 xenografts, which showed reduced rates of tumor growth after TAK228 treatment when the tumors were poorly vascularized but failed to show reduced rates of tumor growth after TAK228 treatment upon increased vascularization (Figure 3G–L, S3E–H). Overall, these findings support our hypothesis that the extent of tumor vascularization can influence the response to TAK228, even when mTOR pathway is effectively inhibited.

Inhibiting vascularization improves response to TAK228 in cells with both inactive and active macropinocytosis
Based on our findings that experimentally increased vascularization can attenuate the response to TAK228, we next investigated whether combining anti-angiogenic therapy with TAK228 could improve efficacy in TAK228 resistant LUSC tumors. We first assessed whether highly vascularized H2170 xenografts would respond to the combination of TAK228 and pazopanib, a FDA-approved pan-VEGFR1/2/3 inhibitor (64),(65). H2170 cells overexpressing both VEGFA and VEGFC were implanted into the flanks of NSG mice, and once tumors reached approximately 150 mm³, mice were treated with Vehicle, TAK228, pazopanib, or TAK228+pazopanib for 18 days. There was no significant response in the single-agent TAK228 or pazopanib groups (Figure 4A). However, the combination of TAK228 and pazopanib effectively slowed the rate of tumor growth (Figure 4A). We next evaluated the extent of vascularization following treatment in all four groups: TAK228 alone had a modest impact on the percentage of CD31-positive cells, while pazopanib alone significantly reduced vascularization. Notably, the combination of TAK228 plus pazopanib did not further decrease the percentage of CD31-positive cells compared to pazopanib alone (Figure 4B,C, S4A–D). Taken together, these results suggest that co-targeting of mTOR and VEGFR in heavily vascularized tumors can significantly slow the rate of tumor growth in response to TAK228 in LUSC tumors.
Next, we investigated whether co-targeting angiogenesis and macropinocytosis in highly vascularized, macropinocytosis-competent tumor xenografts with a combination of anti-angiogenic therapy, TAK228, and EIPA would inhibit tumor growth. We hypothesized that highly vascularized, macropinocytosis-competent H1703 xenografts would exhibit reduced responsiveness to the TAK228+EIPA combination compared to poorly vascularized H1703 xenografts (Figure 2K). To test this hypothesis, we overexpressed both VEGFA and VEGFC in H1703 cells and implanted them into flanks of NSG mice. As anticipated, overexpression of both VEGFA and VEGFC abrogated the effects of TAK228+EIPA in highly vascularized H1703 xenografts (Figure 4D), when compared to poorly vascularized H1703 xenografts (Figure 2K), suggesting that increased vascularization can overcome the nutrient limitations imposed by the combined inhibition of mTOR and macropinocytosis. Further combining pazopanib with TAK228 and EIPA led to a significantly slower rate of tumor growth (Figure 4D). Pazopanib was particularly effective in reducing the area of tumors that stained positive for the endothelial marker CD31 (Figure 4E,F, S4E–H), while TAK228 demonstrated on-target activity, as indicated by p4EBP1 staining (Figure 4G,H). Of note, in agreement with previous reports (66, 67) pazopanib alone was able to significantly reduce cell growth in a dose-dependent manner in LUSC cell lines in vitro (Figure S4I). This data suggests that LUSC tumors can overcome mTOR inhibition by two mechanisms, either by upregulating macropinocytosis or tumor vascularization or both to support nutrient requirements for growth and proliferation.
Together, these data suggest that poorly vascularized (CD31 low) LUSC tumors will be most responsive to mTOR inhibition when combined with a macropinocytosis inhibitor or glutaminase inhibitor (Figure 4I). For highly vascularized (CD31 high) tumors, however, the addition of an anti-angiogenic agents, such as the pan-VEGFR inhibitor pazopanib, to TAK228, with or without a macropinocytosis inhibitor, would result in more effective tumor management, as outlined in the Figure 4I schematic.

Increased vascularization overcomes glucose and glutamine inhibition in LUSC PDXs
Combining small molecule inhibitors targeting glycolysis and glutaminolysis has shown promise in limiting tumor growth in preclinical lung cancer models (26). Recent clinical trials examining the combination of TAK228 and CB-839 in NSCLC have demonstrated that this approach can be effective in certain clinical contexts (68). However, not all patients responded to the TAK228+CB-839 regimen regardless of NFE2L2/KEAP1 mutational status (13); mutations in the NFE2L2/KEAP1 have been reported to induce mTOR pathway dependency (69). This raises questions about the broader effectiveness of metabolism-based therapies in real-world clinical settings. As such, we aimed to investigate whether increased vascularization may reduce the efficacy of the TAK228+CB-839 combination in patient-derived xenografts (PDXs).
First, we compared the effectiveness of TAK228+CB-839 treatment in three LUSC PDXs with different vascularization levels, measured by CD31-positive staining, namely PDX005, PDX007 and PDX025. We found that the response to TAK228+CB-839 combination (Figure 5A–C) was significantly attenuated in PDX025 which had the highest CD31 density, when compared to PDX005 and PDX007 (Figure 5D–F). Specifically, PDX025 tumors, which exhibited the highest CD31 density, were resistant to TAK228+CB-839 combination therapy, while PDX005 and PDX007 tumors, with lower CD31 density (Figure S5A–C), showed significant reduction in the rate of tumor growth following TAK228+CB-839 treatment. All PDXs were positive for p63, PDX005 was positive for p40, and PDX025 had high levels of FGFR1, likely reflecting FGFR1 amplification (Figure S5D). NanoString data showed that mRNA levels of VEGFA were highest in PDX025 compared to PDX007 and the H2170 xenograft, respectively, (Figure S5E), suggesting that the increased vascular density observed in PDX025 is VEGFA-driven.
Next, we examined whether prolonged treatment with TAK228+CB-839 would induce increased vascularization in these LUSC PDXs. We treated CD31 low PDX005 with TAK228+CB-839 for 22 days and found that after an initial reduction in tumor volume, these tumors began to gradually regrow by the second week of treatment (Figure 5G). Immunostaining for CD31 revealed a significant increase in the percentage of CD31-positive cells in the TAK228+CB-839 treatment group compared to the vehicle group, suggesting increased vascularization (Figure 5H,I). We next performed RNA sequencing to assess the expression of VEGFA and VEGFC following TAK228+CB-839 treatment (Figure 5J,K). While VEGFA levels were not significantly different between vehicle and TAK228+CB-839 groups, VEGFC mRNA levels were significantly higher in the TAK228+CB-839 group, suggesting that VEGFC likely drives increased vascularization following treatment. Interestingly, VEGFA expression did not correlate with CD31 or CD34 levels (markers of endothelial cells) in the LUSC TCGA dataset (Figure 5L, S5F). However, there was a positive correlation between VEGFC expression and both CD31 and CD34 expression (Figure 5M, S5G), raising the possibility that VEGFC may be a more relevant driver of vascularization in LUSC patients. To directly evaluate if co-targeting a highly vascularized PDX with metabolic and anti-angiogenic therapy would slow tumor growth, we treated VEFGA/CD31 high PDX025 with Vehicle, TAK228+EIPA, pazopanib, or TAK228+EIPA+pazopanib for 2 weeks. Single treatment with pazopanib had a modest effect in reducing tumor growth, however the combination of TAK228+EIPA+pazopanib was significantly more effective at reducing tumor growth compared to either TAK228+EIPA or pazopanib treatments in these highly vascularized tumors (Figure S5H).
Finally, we examined the impact of high VEGFA and VEGFC expression on disease-free survival (DFS) and overall survival (OS) in the LUSC TCGA dataset. Patients with the highest expression of both VEGFA and VEGFC had significantly worse DFS compared to those with the lowest expression of these genes (Figure 5N,O), with a non-significant trend toward shorter OS (Figure 5P).
In summary, our data highlight vascularization as a key mechanism of resistance to metabolism-based therapies targeting glucose and/or amino acid utilization in LUSC. These findings suggest that nutrient availability, particularly that of amino acids, plays a critical role in modulating response to small molecule inhibitors targeting metabolic pathways in LUSC.

Discussion
Small molecule inhibitors which target key metabolic and bioenergetic pathways such as glycolysis (TAK228) and glutaminolysis (CB-839), represent a potent class of drugs with selective on-target activity that can be utilized for cancer treatment (11),(25),(26),(29), and as such represent an emerging therapeutic approach for the management of advanced solid tumors, given the importance of these critical pathways to cancer growth. To date, these inhibitors have demonstrated encouraging preclinical efficacy (26), but clinical results have been mixed with many failed advanced solid tumor trials in this space (70, 71) (72, 73) (NCT04265534, NCT04471415), limiting effective use of these drugs. As a result, there is a critical need to gain a better understanding for tumor resistance mechanisms, which will allow for improved clinical trial design moving forward.
A major contributing factor to the limited clinical efficacy of these therapies lies in our incomplete understanding of the adaptive mechanisms that tumor cells employ to circumvent therapeutic efficacy. Tumor cells exhibit high levels of metabolic flexibility and have the capacity to rewire their cellular metabolism in response to therapeutic pressure(s), thereby maintaining their proliferative and survival capabilities despite the inhibition of key metabolic enzymes. In this context, cancer cells often activate compensatory metabolic pathways or engage alternative signaling networks that bypass the primary targets of inhibition, reducing the therapeutic efficacy of these drugs. In this study, we show that LUSC models (human cell lines, xenografts, and PDXs) readily adapt to treatment with TAK228, a potent mTOR inhibitor, via two distinct mechanisms, despite robust on-target mTOR inhibition.
First, first observed an increase in the relative levels of multiple amino acids following treatment with TAK228 in H1703 xenografts that were resistant to TAK228 treatment. In contrast, H2170 xenografts, which showed a significantly reduced rate of tumor growth following TAK228 treatment, failed to increase amino acids pools in their tumors. These results were consistent with the expected outcome that inhibition of mTOR signaling, and the subsequent suppression of anabolic processes are reliant on amino acid metabolism (Figure 1). This finding highlights distinctly different metabolic responses between TAK228 resistant tumors vs tumors in which TAK228 significantly reduced the rate of tumor growth. Additionally, our results suggest that TAK228 resistant tumors continued to grow in the presence of TAK228 through activation of alternative mechanisms of nutrient uptake in order to sustain cellular homeostasis and growth.
Next, we examined the role of macropinocytosis, a form of bulk endocytosis enabling cells to uptake extracellular fluid and nutrients such as amino acids, as a source of adaptive nutrient uptake in TAK228 resistant LUSC tumor cells, hypothesizing that macropinocytosis may contribute to metabolic adaptation following TAK228 treatment. Macropinocytosis has been reported to be active in KRAS-mutant (33, 34) and NRF2/KEAP1-mutant cancers (54), both basally (33, 34) and in response to glutamine limitation (35). In the current study, we show that macropinocytosis is induced to a significantly greater extent in H1703 and H520 cell lines as compared to H2170 and HCC15 LUSC cell lines (Figure 2D–G). Importantly, LUSC lines with the largest increase in TAKK28-induced macropinocytosis demonstrated the greatest resistance to TAK228 when implanted into xenografts (Figures 2D–H and Figures 2K, S2F, 3A, 3G). These findings suggest that TAK228 resistant tumors rely on macropinocytosis to replenish amino acids and support their growth despite the inhibition of mTOR signaling by TAK228. This finding was confirmed by testing TAK228 in combination with EIPA, a small molecule that leads to macropinocytosis inhibition via downregulation of SLC9A1 activity (57). In these experiments we demonstrated that the combination of TAK228 + EIPA significantly inhibited tumor growth in TAK228-resistant H1703 and H520 xenografts (Figure 2K, S2F). These findings underscore the importance of metabolic reprogramming in mediating resistance to mTOR inhibition and highlight that targeting macropinocytosis in combination with TAK228 may provide a promising future therapeutic strategy to overcome resistance in solid tumors. By addressing the adaptive mechanisms that sustain tumor growth despite metabolic inhibition, this approach could enhance the efficacy of mTOR inhibitors and potentially improve outcomes for patients with resistant tumors.
We also explored additional mechanism(s) that tumor may utilize to facilitate the availability of proteins and amino acids to support tumor growth and proliferation in the context of mTOR inhibition. We hypothesized that in addition to macropinocytosis, tumors may increase angiogenesis as a mechanism to provide plentiful nutrients that support tumor growth and proliferation. This predicted that for tumors deficient in macropinocytosis, an increase in angiogenesis would provide an escape from the effects of TAK228 mediated mTOR inhibition. We showed that despite robust on-target inhibition of mTOR, as indicated by the reduced p4EBP1 levels, tumors with enhanced angiogenesis were able to bypass the metabolic stress induced by mTOR inhibition (Figure 3A–F, G–L). These results suggest that augmented blood vessel formation enables tumor cells to maintain access to essential nutrients, including amino acids, which are critical for sustaining cellular function and promoting tumor proliferation despite metabolic blockade. In macropinocytosis-competent H1703 xenografts, which possess the ability to scavenge nutrients via macropinocytosis, increased angiogenesis blunted the efficacy of the TAK228 and EIPA combination therapy (compare Figure 2K vs. Figure 4D). New blood vessel formation in these tumors ensured a sufficient nutrient supply, thereby supporting tumor growth and survival, which diminished the therapeutic impact of targeting both glycolysis (via TAK228) and macropinocytosis (via EIPA). These observations underscore the notion that while metabolic therapies can reduce the availability of critical metabolic intermediates, the presence of an efficient and robust vasculature can counteract these effects by ensuring the continuous delivery of nutrients to the tumor microenvironment.
To investigate the potential beneficial effects of combining anti-angiogenic therapy with metabolism-based treatments, we evaluated the therapeutic efficacy of pazopanib, a pan-VEGFR1/2/3 inhibitor (74), in conjunction with either TAK228 or TAK228+EIPA. Pazopanib was utilized to inhibit angiogenesis, and its effects were assessed in combination with metabolic inhibitors targeting glycolysis (via TAK228) and macropinocytosis (via EIPA). Our data demonstrate that the addition of pazopanib effectively suppressed angiogenesis and, when combined with TAK228 or TAK228+EIPA, resulted in a significantly slower rate of tumor growth compared to TAK228 or TAK228+EIPA groups. This combination proved remarkably effective in tumors that were previously resistant to metabolism-based treatments, including those with functional macropinocytosis. Collectively, these findings highlight the pivotal role of angiogenesis in enabling tumors to evade the therapeutic effects of metabolism-based therapies. As a powerful compensatory mechanism, angiogenesis maintains robust nutrient delivery to tumors, enabling sustained cellular proliferation under severe metabolic stress. Consequently, simultaneous targeting of angiogenesis and metabolic pathways may be a highly promising future therapeutic strategy.
Our study also highlights the importance of evaluating tumor vascularization when assessing the efficacy of novel metabolism-based therapies, particularly in resistant patient populations. Given the substantial metabolic heterogeneity observed across solid tumors, our findings suggest that an assessment of the extent of tumor vasculature could serve as a valuable prognostic tool in predicting therapeutic response. Tumor vasculature plays a crucial role in regulating the availability of essential nutrients, such as glucose and glutamine, which are critical for sustaining tumor growth. As such, the degree of vascularization could significantly influence the efficacy of therapies that limit these key nutrients. These insights point toward a potential clinical strategy involving the combination of metabolic inhibitors with angiogenic therapies, either as part of an upfront treatment regimen or as a follow-up option for patients whose tumors fail to respond to initial therapies targeting glucose and/or amino acid metabolism. The rate of nutrient delivery to the tumor microenvironment is largely determined by the extent and functionality of the tumor vasculature. Tumors with extensive vascular networks are better equipped to overcome metabolic stress, as these vessels ensure continuous nutrient flow, even under conditions where metabolic inhibitors are targeting key pathways such as glycolysis and glutaminolysis (55, 56). In such cases, enhanced angiogenesis can counteract the effects of metabolic therapies by providing a sustained supply of glucose, amino acids, and other essential nutrients required for tumor cell survival and proliferation. Our data suggest that increased vascularization may be a critical mechanism of resistance to metabolism-based therapies that limit the availability of glucose and/or amino acids.
Given the prominent role of many novel classes of therapies, such as antibody drug conjugates, bispecific T-cell engagers and other immunotherapy approaches, it will be important to evaluate the role of macropinocytosis across a variety of treatment approaches. For example, recent studies suggest that macropinocytosis can enhance uptake of peptide-drug conjugate by pancreatic cancer cells (75) and a monobody-drug conjugate by multiple myeloma cells (bioRxiv 2025.04.04.647098). However, macropinocytosis can also contribute to off-target effects, such as ADC-induced thrombocytopenia mediated by megakaryocyte internalization (76) and ocular toxicity (77). For bispecific T- cell engagers the role of macropinocytosis is less clear and future work will be required to understand its implications.
With respect to other immunotherapy approaches, dendritic cells and macrophages, utilize macropinocytosis to take up large amounts of extracellular fluid and antigens. These antigens are then processed and presented to T cells via major histocompatibility complex (MHC) class II molecules, initiating adaptive immune responses. This process is constitutive in immature dendritic cells, allowing continuous sampling of the environment for potential threats, including tumor-associated antigens. Treatments that enhance anti-tumor immunity, such as cancer vaccines, can leverage macropinocytosis to improve antigen delivery to dendritic cells. For example, nanoparticle-based vaccines designed to be taken up via macropinocytosis have shown promise in promoting dendritic cell maturation and antigen presentation, leading to stronger T cell activation against tumors. Combining these approaches with checkpoint inhibitors, like anti-PD-1 antibodies, may further enhance efficacy by preventing T cell exhaustion. However, the complexity of tumor microenvironments and varying macropinocytic activity across cancer types suggest that more research is needed to optimize these strategies.
While our study showed that inhibiting glucose and amino acid utilization along with vascularization can limit tumor growth in xenografts and PDXs of LUSC, there are several limitations. First, while our study focuses on LUSC cell lines, xenografts and PDXs, we cannot exclude the possibility that other NSCLC tumor types might exploit these mechanisms to circumvent tumor growth in the presence of mTOR inhibitor(s), necessitating future studies on other types of NSCLC, such as lung adenocarcinoma. Second, treatment duration in our pre-clinical models were short and therefore it is unclear if they can translate to the extended survival in patients. Third, although TAK228, EIPA, and pazopanib treatments limited tumor growth, these treatments did not result in tumor regression. Fourth, EIPA, a tool compound derivative of FDA approved diuretic amiloride, is an inhibitor of Na+/H+ exchanger SLC9A1 and potent inhibitor of macropinocytosis, however it also can cause cell cycle arrest (78) and enhance autophagy (79).

Supplementary Material
Figure S1Figure S2Figure S3Figure S4Figure S5