Multinuclear NMR and MRI Reveal an Early Metabolic Response to mTOR Inhibition in Sarcoma

Sarcomas are rare heterogeneous mesenchymal malignancies arising from different tissues (11,000 cases per year in United States; ref. 1), accounting for 21% of all pediatric solid malignant tumors and 38% of soft-tissue tumor at 65+ years old (2). More than 50 subtypes of sarcoma differing by origin, localization, and grade of invasiveness have been identified. Because of this heterogeneity, sarcomas are considered multiple malignancies rather than a single tumor type, though targeting such a vast array of malignancies poses a challenge both in approaching a treatment strategy and assessing response.

Deregulation of several elements in the phosphatidylinositol-3-kinase/Akt/mammalian target of rapamycin (PI3K/AKT/mTOR) pathway has been identified in many types of sarcoma. An analysis performed using the c-bioportal database (TCGA dataset) shows that the PI3K/AKT/mTOR pathway is altered in 74% of sarcoma patients (n = 265 patients; Supplementary Fig. S1; refs. 3, 4). Given the frequency of alterations in this pathway, multiple clinical trials have been initiated to target mTOR in the setting of sarcoma, though with limited success. The SUCCEED trial being one of the largest aimed at utilizing Ridaforolimus as a maintenance therapy, and ultimately was deemed unsuccessful. A major critique of this study and many others has been the heterogeneity of the treatment response, making it difficult to assess tumor dependence on the targeted pathway and predict possible therapeutic benefit. This coupled with the challenge of treating these tumors further exemplifies the need for approaches to stratify patients noninvasively as well as possibly predict treatment response outcomes (5).

The role of the PI3K/mTOR pathway is well established in cancer and directly controls protein and lipids synthesis, autophagy, and glucose metabolism (6). mTOR is a protein kinase composed of two distinct multiprotein complexes, mTORC1 and mTORC2, that act as master regulators of cellular metabolic homeostasis (7–9). Rapamycin, an FDA-approved drug, specifically targets mTORC1 and was used in this study as a tool compound to evaluate its effect on tumor metabolism in vivo. Inhibition of mTORC1 leads to the downregulation of the phosphorylation status of the ribosomal protein S6 (S6) with a subsequent reduction of energy (ATP) and cofactors (NADPH), both essential for glucose metabolism and for other biosynthetic processes. The metabolic reprograming of cancer cells is essential to sustain the energy needs for cell survival and proliferation. A key metabolic hallmark of carcinogenesis is the shift from oxidative phosphorylation (OXPHOS) to aerobic glycolysis (Warburg effect) with an increase in glucose consumption and an elevated rate of lactate production (10, 11). This has been probed in vitro most routinely by mass spectrometry methods and in vivo using ¹H and ¹³C magnetic resonance spectroscopy (MRS). Although ¹H and ¹³C MRS can assess the metabolic status of a specific tissue at a steady state, it is severely limited by both scan time and sensitivity (12).

Currently, visualization and quantification of glycolytic metabolic flux can be achieved using hyperpolarized MRI (HP MRI). This technique overcomes the sensitivity issue of conventional MRS techniques by dramatically increasing the spin population, beyond the Boltzmann distribution for a given magnetic field strength, of a target molecule prior to its introduction into the system of interest. In this work, we take advantage of dissolution dynamic nuclear polarization (dDNP) as a methodology to hyperpolarize our molecules of interest (13). Using dDNP, many substrates have been hyperpolarized and a strong enhancement of MR signal can be obtained (>10⁴-fold increase) at various magnetic field strengths, permitting the evaluation of metabolism noninvasively (14). Moreover, imaging of glycolytic metabolism with hyperpolarized [1-¹³C]pyruvate has shown considerable potential in preclinical oncology studies, particularly for the assessment of treatment response (15–17), though early response and the delineation of reduction in cell proliferation versus cell death remain a challenge. In this work, we utilize these approaches, along with molecular biology approaches, to characterize in vitro and in vivo early metabolic responses to mTOR inhibition in sarcomas, demonstrating the potential to utilize HP lactate as a marker to separate eventual cell death from reduced proliferation.

Unless otherwise indicated, all chemicals and solvents were purchased from Sigma Aldrich, including 99% enriched [1,6-¹³C]glucose.

Gastrointestinal Sarcoma Tumor (GIST-T1), DDLS, JJ012, and CS1 cell lines were cultured under standard conditions in DMEM supplemented with 10% FBS and penicillin streptomycin. GIST-T1 and DDLS were kindly provided by laboratories at Memorial Sloan Kettering Cancer Center (Chi Lab and Koff Lab at MSKCC in 2015). JJ012 and CS1 were kindly provided by the Thompson Lab at Memorial Sloan Kettering Cancer Center in 2014. All cell lines were authenticated prior to publication by short tandem repeat DNA fingerprinting.

The IncuCyte HD system (Essen BioScience) was used to evaluate the effect on proliferation of rapamycin on GIST-T1, DDLS, JJ012, and CS1 cells. Briefly, the day before the experiment, 5 × 10⁴ cells were plated in 24-well plates and incubated overnight to allow cell adherence. At the day of the experiment, the media were exchanged with complete media containing 50 nmol/L of rapamycin or DMSO alone (1 μL/well) as a treatment and vehicle condition, respectively. Frames were captured at 6-hour intervals from 4 separate 950 × 760 μm² regions per well using a 20× objective. Confluence was measured using the incuCyte software. Values from all four regions of each well were pooled and averaged across all four replicates. Results were expressed graphically as fold increase of cell growth normalized to day 1 versus time.

Cell lysate for vehicle-treated (6% DMSO) or rapamycin-treated (50 nmol/L) cell lines was prepared as previously published (18). For Western blot analysis, 20 μL of 2 mg/mL of cell lysate were mixed with 5 μL of 5× sample loading buffer. Separated proteins in the gels were electrophoretically transferred onto PVDF membrane. After washing and blocking, pS6K, PFK, pPKM2, pLDH-a, and β-actin antibody (1 μg/mL; Cell Signaling Technology), diluted in TBS-T containing 5% BSA, were added and incubated for overnight at 4°C. The bound antibodies were detected by horseradish peroxidase–conjugated anti-goat Ig secondary antibody (Santa Cruz Biotechnology) followed by ECL detection system (Thermo Scientific) according to the manufacturer's instruction.

Experiments were performed on a 14.1T NMR spectrometer equipped with an autosampler and ¹H cryoprobe (Bruker Biospin). ¹H NMR spectra for each cell extract and media samples were acquired with a water presaturation recycle delay of 4 seconds, acquisition time of 2.67 seconds, 90° pulse, and 512 averages (18). Resonances of each metabolites were identified and quantified using Chenomx NMR Suite 8.0 professional (Chenomx Inc.), with 0.5 mmol/L DSS as a known reference standard to determine the concentration of individual metabolites and 10 mmol/L of imidazole as a pH reference (19). In order to quantify and distinguish total pool of lactate from ¹³C lactate, ¹H NMR has been used. Proton resonance line shapes are split into well-established patterns when J coupled to a ¹³C nucleus. Comparing experiments performed with and without ¹³C decoupling during acquisition indirectly measures fractional ¹³C enrichment (18).

The animal portion of this study was performed under a protocol approved by the Institutional Animal Care and Use Committee of Memorial Sloan Kettering Cancer Institute. GIST-T1, JJ0012, and CS1 xenografted mice were used as a tumor model for in vivo HP pyruvate MRI (n = 3 mice/group). DDLS cells did not develop a xenograft tumor, and therefore the in vivo analysis in this tumor model was not assessed. A total of 5 × 10⁶ GIST-T1, 10 × 10⁶ JJ012, and 10 × 10⁶ CS1 cells were trypsinized and resuspended in a 1:1 solution of complete media:Matrigel before being injected subcutaneously on the flank of immune compromised NOD. CB17-Prkdcscid mice (Jackson Laboratory). Twenty-four hours before the HP experiment, mice were treated with 15 mg/kg of rapamycin for comparison with 6% DMSO (vehicle). The starting tumor volume for GIST-T1, JJ012, and CS1 was 0.23 ± 0.05 cm³ and 0.21 ± 0.04 cm³, 0.35 ± 0.03 cm³ and 0.34 ± 0.02 cm³, 0.17 ± 0.05 cm³ and 0.21 ± 0.06 cm³ for vehicle- and rapamycin-treated animals, respectively.

All of the animal experiments were carried out on a permanent 1T MRI system (nanoScan PET/MRI, Mediso). Tumor regions were identified in axial anatomic images acquired using T₂- weighted fast-spin-echo (FSE) acquisition (EchoTime/Repetition Time = 88.5/200 ms). Fifteen 2 mm-slices were acquired with 50-mm FOV and 256 × 252 matric to cover the whole tumor. The images were acquired at days 1, 3, 5, 7, 9, 11, 13, and 15. Osirix (Pixmeo) was used for data analysis. A region of interest was manually drawn within the tumor edge, and a composite image was created to determine the total tumor volume. Results were expressed graphically as fold increase of tumor growth normalized to day 1.

Note that 100 μL of 14.2 mol/L [1-¹³C]pyruvate (GE healthcare) was mixed with 15 mmol/L of trityl radical (GE healthcare, prepared as previously published) and polarized on Spin lab (GE healthcare; ref. 20). The frozen sample was dissolved in 10 mL of 40 mmol/L of TRIS buffer. The dissolution was neutralized in a receiving vial containing sodium hydroxide. Two hundred microliter of 100 mmol/L hyperpolarized [1-¹³C]pyruvate was injected intravenously in the tail vein of a catheterized animals in 10 seconds. After 15 seconds of pyruvate distribution, a 2D-CSI image was acquired. The radiofrequency coil used in this experiment was dual-tuned ¹H/¹³C coil. Before hyperpolarized study, T₂-weighted FSE images were acquired for anatomical localization (EchoTime/Repetition Time = 88.5/200 ms). Slices of 2 mm were acquired with 50-mm FOV and 256 × 252 matrix to cover the whole tumor. 2D-CSI sequence was used to acquire the HP [1-¹³C]pyruvate MRS. In vivo [1-¹³C]pyruvate spectroscopy data were processed using a custom software (Matlab R2015b, Mathworks). The peak area of HP [1-¹³C]pyruvate to [1-¹³C]lactate was used to calculate relevant ratios. The reduction of HP [1-¹³C]pyruvate was expressed using the following formula: HP [1-¹³C] lactate/(HP [1-¹³C]pyruvate+ HP [1-¹³C]lactate). Each tumor voxel was coregistered anatomically using a T₂-weigthed image, and a mean value was assigned to each tumor. Only voxels with 80% of tumor tissue were included to limit the influence of partial volume effects.

Tumor tissues were embedded in paraffin, and 4-μm-thick sections were prepared and stained with hematoxylin–eosin. After deparaffinization, tissue sections were stained with monoclonal rabbit anti-Ki67 antibody and cleaved caspase-3 to detect proliferation and cell death, respectively. Quantification of stained cells was achieved using FiJi, and the ratio between stained and unstained cells was established.

All other methods are described in Supplementary Information.

Four cell lines have been included in this study to survey whether tumor origin or aggressiveness has a differential effect after rapamycin treatment. In identical medium conditions, the four cell lines grew at significantly different rates with CS1 (chondrosarcoma) cells and demonstrated the fastest growth, whereas JJ012 (chondrosarcoma) grew moderately rapidly, and GIST T1 (gastrointestinal soft-tissue sarcoma) and DDLS (liposarcoma) showed the slowest growth rates (Fig. 1A). After suppression of mTORC1-dependent signaling by treatment with 50 nmol/L of rapamycin (Fig. 1B), a differential effect on cell growth was observed. The cell growth arrest becomes significant at 54 for GIST T1, 96 hours for DDLS, 36 hours for JJ012, and 36 hours for CS1. (Fig. 1A). mTOR plays a central role in cell growth and proliferation, and as expected, the fastest proliferating cells relied heavily on this pathway (CS1 and JJ012 cells) and were the most affected by PI3K/mTOR inhibition (Fig. 1B), though cell death was not observed in vitro. These results confirm interdependence of the PI3K/mTOR pathway and cellular proliferation.

To verify inhibition of the mTOR pathway, the phosphorylation status of S6 and 4EBP-1 was evaluated in rapamycin-treated GIST-T1, DDLS, JJ012, and CS1 as compared with control DMSO-treated cells. In all cell lines, we observed a significant decrease in phosphorylation of S6 meaning an effective target inhibition. No significant effect was observed in the phosphorylation status of 4-EBP1 as expected for each cell type. In order to then characterize the effect of rapamycin treatment on sarcoma cell glucose metabolism, we first interrogated multiple proteins responsible for glucose uptake and subsequent phosphorylation. The GLUT-1 transporter, typically the main uptake pathway for glucose, and hexokinase-2 (HK-2), routinely upregulated in cancer, were considered as markers for upstream glycolysis, whereas downstream lactate production was measured as an indicator of total glycolysis (21, 22). No changes in GLUT-1 and HK-2 expression were detected in all the four sarcoma cell lines after 24-hour rapamycin treatment (Fig. 1B). Moreover, no changes in total protein and phosphorylation level of lactate dehydrogenase (LDH) were detected (Supplementary Fig. S2).

In vitro metabolic analyses were performed in order to assess changes in metabolic pools sizes in all cell lines. In vitro, the total metabolites pool size concentrations are highest in JJ012 and CS1 as compared with GIST-T1 and DDLS. Changes in intracellular and extracellular metabolites are also the largest in those two cell lines with rapamycin treatment. Several metabolic pathways, such as glutaminolysis, and amino acid uptake are affected by rapamycin treatment (Supplementary Fig. S3A and S3B). However, in all cell lines analyzed, the largest difference was a decrease in glycolytic flux from glucose to lactate after mTOR inhibition.

Figure 1C shows the changes in glucose consumption after 24 hours of treatment with rapamycin. We observed a significant decrease in glucose consumption of 35.8%, 35.0%, 58.5%, and 54.5% for GIST-T1, DDLS, JJ012, and CS1, respectively (P < 0.05). Importantly, this change in glucose consumption was observed before any demonstration of significant growth arrest (Fig. 1A). With treatment, glucose consumption significantly decreased in all cell lines with a corresponding decrease in both total intracellular and extracellular pool size of lactate in all the 4 cell lines (intracellular: 49%, 48%, 56%, 49% and extracellular: 54%, 60%, 51%, 47% for GIST-T1, DDLS, JJ012, and CS1, respectively; P value < 0.05; Fig. 2A).

To determine whether the changes in lactate production were specifically derived from glucose, we traced [1,6-¹³C]glucose labeling to lactate in culture. The same degree of decrease was observed in ¹³C-labeled lactate derived from 3 hours of incubation with [1,6-¹³C]glucose (Fig. 2B). After 24 hours of rapamycin treatment, we did not observe a significant change in either the extracellular or intracellular fractional enrichment of lactate. These data demonstrate that the inhibition of the mTOR pathway modulates lactate production, which derives predominantly from glycolysis.

In order to assess tumor metabolism noninvasively in vivo, hyperpolarized [1-¹³C] pyruvate MRS was used to monitor early changes in metabolism. DDLS cells did not efficiently xenograft and were not used for in vivo analysis. Figure 3A demonstrates a representative T₂-weigthed ¹H image with corresponding HP [1-¹³C] pyruvate MRS in vehicle- and rapamycin-treated tumor-bearing mice. Upon entering the cell, HP [1-¹³C]pyruvate can be rapidly reduced to [1-¹³C] lactate, transaminated to [1-¹³C] alanine, and oxidized, resulting in the formation of CO₂ and later bicarbonate (HCO₃) via carbonic anhydrase (23). Although at baseline, GIST-T1, JJ012, and CS1 tumors exhibited similar HP [1-¹³C]lactate/total ¹³C ratios, a significant decrease in [1-¹³C] lactate production is observed at 24 hours of rapamycin treatment in all models with 36.8%, 30.9%, and 47.3% observed in GIST-T1, JJ012, and CS1, respectively (P < 0.001 for all rapamycin tumor treated vs. vehicle, Fig. 3B). When quantitatively comparing the HP lactate/total ¹³C ratio after treatment, a significant difference was observed between GIST-T1, JJ012, and CS1 sarcomas, providing a means of separating these groups (0.46 ± 0.01, 0.50 ± 0.04, and 0.35 ± 0.01 for rapamycin-treated GIST-T1 and JJ012 vs. CS1, P < 0.001, Fig. 3C). Moreover, in the setting of CS1 sarcomas, a decrease in HP [1-¹³C]lactate and a significant increase in both [1-¹³C]alanine and [1-¹³C]HCO₃ were observed in treated tumors (P < 0.05 and P < 0.01 for alanine and bicarbonate, respectively; Supplementary Fig. S4). These data show a clear metabolic shift from aerobic glycolysis of the DMSO-treated CS1 tumors to a partial re-establishment of oxidative phosphorylation after rapamycin treatment in vivo.

Both in vitro NMR experiments and in vivo HP pyruvate suggest an inhibition of glycolytic flux in short-term rapamycin treated cells, prior to a change in cell growth. Thus, to evaluate the long-term effect in vivo, xenograft mice were treated systemically for 15 days with 15 mg/kg of rapamycin without any off target side effect reported (24–26). In all sarcoma tumor models, rapamycin markedly reduced tumor volume compared with vehicle-treated animals. Remarkably, we observed differences in treatment efficacy depending on the model used. GIST-T1 showed a significant inhibition of tumor growth at 13 days after starting rapamycin treatment (P < 0.05). As expected, for the in vitro cell experiment, we observed a greater effect of rapamycin treatment in JJ012 and CS1 compared with GIST-T1. Tumor growth arrest was observed at 5 and 3 days of rapamycin treatment for JJ012 and CS1 tumors, respectively (P < 0.001, for both JJ012 and CS1; Fig. 4A). The sensitivity profile of these xenografts mirrors their glycolytic metabolic status. Moreover, it seems that the more dependent the tumor is on glycolysis, the more sensitive it is to rapamycin-induced growth inhibition. The present in vivo tumor growth data correlate with the histopathologic analysis performed at 15 days of rapamycin treatment. As expected, we observed in all rapamycin-treated tumors an antiproliferative effect, identified by a decrease of Ki-67 staining (Fig. 4B). However, only JJ012 and CS1 tumors had a significant decrease in cell proliferation (P < 0.05 for both; Fig. 4C). Moreover, a significant increase in cell death, detected by cleaved caspase-3, was observed only CS1 tumors (3.7-fold higher, P < 0.001, Fig. 4B and C).

Our findings indicate that rapamycin induces a rapid metabolic effect and later a biological response in short- and long-term treatment. In vitro, we observed an effective inhibition of the mTOR pathway in sarcomas by rapamycin, leading to a significant change in glucose metabolism as early as 24 hours after treatment. Interestingly, short-term treatment in these cells models did not affect the expression or phosphorylation of major glycolysis-involved proteins. However, a decrease in intracellular ATP and NADPH was observed in all cell lines analyzed in this study (Supplementary Fig. S5), and this most likely drives the metabolic effects observed. We were also able to detect early metabolic changes in glucose metabolism in vivo (observed as early as 24 hours) using HP pyruvate. A significant decrease in glycolytic flux with treatment using HP MRI correlates with a decrease of tumor growth in long-term treatment in all sarcoma models imaged. Interestingly, a decrease in approximately 50% of HP lactate production at 24-hour rapamycin treatment correlates with induction of apoptosis in long-term treatment.

To date, cumulative data demonstrate a strong interdependence between metabolic changes and cell death. Specifically, various pharmacologic treatments in cancer cells have shown altered glucose metabolism observed prior cell death (27–29). Zhao and colleagues have demonstrated that the upregulation of glycolysis in cancer cells attenuates cell death by stabilizing the antiapoptotic Bcl-2 family protein Mcl-1 (30). Therefore, a decrease of glycolytic flux induced by rapamycin treatment in vivo can activate the proapoptotic pathway as shown by the increase of caspase-3 staining in the tumor sections of treated animals (Fig. 4B). Interestingly, quantitative changes in the glycolytic state leading to a specific utilization of glucose and their direct relationship to induction of apoptosis are not well understood in vivo. This is mostly due to the lack of tools available to accurately study metabolism noninvasively. The most commonly utilized metabolic imaging approach, 2-deoxy-2-[¹⁸F]fluoro-D-glucose positron emission tomography (FDG PET), has shown use in the setting of AKT, but lacks predictive power in the setting of mTOR inhibitor therapy (31). Although FDG PET has provided a window into our understanding of glucose metabolism in vivo, it is limited due to its inability to directly resolve the substrate (FDG) from its compartmentalized downstream product (e.g., FDG-6-phosphate). This makes it difficult to discern uptake kinetics via glucose transporters (predominantly GLUT1) from metabolic conversion by hexokinase. Moreover, recent work has demonstrated that directly assessing the downstream metabolic products of glycolysis using HP MRI may provide a sensitive marker for signaling-dependent changes in glycolysis (32).

Further studies are needed in order to address the differential induction of cell death in rapamycin-treated tumors, but this work provides evidence that the magnitude of reliance on metabolic flux, assessed in vivo, may indicate degree of treatment response to targeted inhibitors. Ultimately, the successful development of targeted therapies is dependent on effective noninvasive approaches to predict and monitor response in patients. Playing a complimentary role to the use of genomic testing to place patients in to phase I clinical trials, HP MRI assessment of mTOR-dependent changes in metabolism could predict which patients would benefit from targeted treatment. In summary, this would allow for better stratification of patients into appropriate clinical trials, assessment of possible heterogeneous tumor response in current trials as well as determine potential treatment efficacy.

No potential conflicts of interest were disclosed.

Conception and design: V. Di Gialleonardo, H.N. Aldeborgh, W.D. Tap, W.A. Weber, K.R. Keshari

Development of methodology: V. Di Gialleonardo, H.N. Aldeborgh, W.D. Tap, K.R. Keshari

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): V. Di Gialleonardo, K.M. Folkers, J.S. Lewis

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): V. Di Gialleonardo, H.N. Aldeborgh, V. Miloushev, K. Granlund, J.S. Lewis, W.A. Weber, K.R. Keshari

Writing, review, and/or revision of the manuscript: V. Di Gialleonardo, W.D. Tap, J.S. Lewis, W.A. Weber, K.R. Keshari

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J.S. Lewis

Study supervision: V. Di Gialleonardo, K.R. Keshari

We gratefully thank Dr. Vanessa S. Rodrik-Outmezguine for her helpful scientific comments.

This study was supported by the NIH/NCI Cancer Center Support Grant P30 CA008748 (K.R. Keshari) and NIH/NIBIB R00 EB014328 (K.R. Keshari) as well as Memorial Sloan Kettering's Center for Molecular Imaging and Nanotechnology (CMINT; K.R. Keshari), Cycle for Survival (K.R. Keshari), and the American Italian Cancer Foundation (V. Di Gialleonardo).

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