Predictive Value of MRP-1 in Localized High-Risk Soft Tissue Sarcomas: A Translational Research Associated to ISG-STS 1001 Randomized Phase III Trial

High-risk localized soft tissue sarcomas (STS) of the limbs or trunk wall encompass patients with malignancy grade 3, tumor size larger than 5 cm, and deeply seated tumors. This subset of patients with STS, with histologies limited to the most frequent subtypes in these locations, such as leiomyosarcoma, undifferentiated pleomorphic sarcoma (UPS), synovial sarcoma, myxoid liposarcoma, and malignant peripheral nerve sheath tumors (MPNSTs), consistently showed a 5-year overall survival (OS) of around 70% when treated with adequate surgery, radiation therapy, and perioperative full-dose chemotherapy with anthracycline plus ifosfamide (1, 2). Recent attempts to improve this outcome failed to demonstrate a survival advantage for histotype-tailored regimens in a phase III randomized trial (ISG-1001) over the control arm with epirubicin plus ifosfamide (3, 4). Moreover, anthracycline plus ifosfamide showed a superimposable survival curve to that observed with the same scheme tested a decade ago in the same patient population. Despite a remarkable survival, this indicates that there is a drug ceiling effect for chemotherapy in these sarcoma subtypes. Multidrug resistance (MDR) mechanisms, which refers to the resistance exhibited by tumor cells to a broad kind of mechanistically and structurally different anticancer drugs, could explain, at least to some extent, the ceiling efficacy observed in these large randomized trials with perioperative chemotherapy in STS. Efflux pumps constitute a critical defense of cells against xenobiotics entailing a relevant MDR mechanism in tumor cells (5). A renewed interest has been recently observed with MDR and cancer, focusing on agents that can modulate efflux transporters for increasing the efficacy of anticancer drugs (6, 7) as well as the relevance of such transporters in the cancer stem cells (8).

MRP-1 overexpression, at protein and mRNA levels, led to a worse prognosis in a correlative study of a previous randomized trial comparing three versus five cycles of epirubicin and ifosfamide in high-risk STS. In contrast, P-glycoprotein (P-gp) and glutathione S-transferase pi (GST-π) expressions had no prognostic influence in the same series (9). The prognostic prominence of MRP-1 could be related to the fact that epirubicin and ifosfamide are both substrates of MRP-1 (in the case of ifosfamide through the GSH conjugates of metabolites), whereas epirubicin is also a substrate of P-gp, but not GST-π, and ifosfamide is a substrate of GST-π, but not P-gp (10, 11).

This research aimed to analyze the prognostic and predictive value of MRP-1 as a correlative prespecified translational study in the ISG-STS 1001 trial. As the patient population enrolled in this trial was the same in terms of risk (high-risk), subtypes, and treatment (in the control arm), we estimated the current series as the validation set for the prognostic/predictive role of MRP-1 in localized sarcomas treated with anthracyclines plus ifosfamide.

In addition, new candidates for MRP-1 modulators, such as avapritinib, ripretinib, and selumetinib, were preclinically tested in sarcoma cell lines, to identify the most potent inhibitors of MRP-1 activity. These latter drugs have been probed in cell lines treated with palifosfamide (an active metabolite of ifosfamide) and doxorubicin, with the aim of translating this potentially enhancing combination into clinical research.

Patients in this series were prospectively enrolled into the ISG-STS 1001 trial (Clinical Trial.gov no. NCT01710176, and the European Union Drug Regulating Authorities Clinical Trials, no. EUDRACT 2010–023484–17) and randomly assigned to receive three cycles of neoadjuvant standard chemotherapy (epirubicin 60 mg/m²/day, days 1 and 2 plus ifosfamide 3 g/m²/day, days 1, 2 and 3) scheduled every 3 weeks (standard arm) or three cycles of neoadjuvant histotype-tailored chemotherapy that consisted of: trabectedin 1.3 mg/m²/day every 3 weeks for high-grade myxoid liposarcoma; gemcitabine 1,800 mg/m²/day plus dacarbazine 500 mg/m²/day repeated every 14 days for leiomyosarcoma; high-dose ifosfamide 14 g/m² in continuous infusion every 28 days by portable pump for synovial sarcoma; etoposide 150 mg/m²/day plus ifosfamide 3 g/m²/day, both from day 1 to 3 repeated every 21 days for MPNST and gemcitabine 900 mg/m²/day on days 1 and 8 plus docetaxel 75 mg/m²/day on day 8 for UPS (experimental arm). Patients diagnosed with myxofibrosarcoma, pleomorphic liposarcoma, pleomorphic rhabdomyosarcoma, spindle cell nonspecified sarcoma were not randomized but were registered and treated with the standard chemotherapy (registry arm). Likewise, cases that could be allocated into gemcitabine or trabectedin arms, and for whom preoperative radiation therapy was deemed necessary, were enrolled for only registry with standard chemotherapy. Additional details of the study have been reported previously (3). Patients were followed up with at least chest imaging every 3 months during the first 3 years, every 6 months for the 4th and 5th years, and from then on yearly. MRI was recommended for local control assessment every 6 months for the first 5 years and from then on, every year. Central pathology confirmation and central radiology assessment were compulsory. Patients enrolled in the trial consented to participate in the translational study presented here and signed an informed consent form. Procedures were performed in accordance with guidelines established by the hospitals’ Ethics Committee, as well as in conformance with the Declaration of Helsinki.

MRP-1 assessment was performed in all cases who entered the study for which there was adequate tissue available. The tumor areas chosen for the tissue microarrays (TMAs) were selected by pathologists blinded to clinical study results. TMA building was performed using an automated tissue arrayer (Minicore Excilone). TMA cores of 1.0 mm in diameter were sampled from paraffin tumor blocks corresponding to each included patient. Afterward, 4-μmol/L sections were cut and stained with hematoxylin and eosin and IHC staining. IHC used the QCRL-1 (Santa Cruz Biotechnology) MRP-1 mAb. MRP-1 expression was grouped as low (≤25% positive cells) versus high (>25% positive cells) extension, whereas the strength of IHC was grouped as of negative/weak intensity (0, + or ++) or strong intensity (+++ or ++++). Kidney was used as positive control of IHC, according to the manufacturer's instructions. Two pathologists with great expertize in sarcomas (PC and RR) were responsible for reviewing MRP-1 protein expression, blinded to clinical data. Each pathologist independently reviewed all the TMA cores.

Clinical and demographic data were expressed in terms of frequency and percentage for categorical variables, median, and interquartile range (IQR) for quantitative ones. Follow-up completeness (median follow-up time) was evaluated using the “reverse Kaplan–Meier” method on OS (12). OS time was measured from enrollment until death. Survival time of patients who did not experience the event considered during follow-up observation was censored at the time of the last follow-up. Disease-free survival (DFS) was calculated from enrollment to progression or death, whichever came first. For DFS, patients who did not experience progression or death, during the follow-up period, were censored at the time of the last follow-up. Survival functions were estimated using the Kaplan–Meier method and compared using the log-rank test. Histotype-based subgroup explorative analysis was performed estimating Cox regression HR and accompanying confidence intervals, and by plotting them in a forest-plot style graph. A prognostic analysis on both OS and DFS was performed analyzing clinically suggested possible factors by Cox models analyses. Cohen kappa coefficient (κ) was performed to evaluate the concordance between the pathologists’ analyses. Unless otherwise specified, confidence intervals were two-tailed and calculated considering a 0.95 confidence level (CI). Performed tests were considered statistically significant where the P values were <0.05. Statistical analysis was performed using R 3.5.2 R Core Team (2020).

The synovial sarcoma SW982 (HTB-93), liposarcoma SW872 (HTB-92), and leiomyosarcoma SK-UT-1 (HTB-114) human cell lines were obtained from the ATCC. The CP0024 leiomyosarcoma and ICP060 MPNST human primary cell lines were established from fresh tumor samples in our laboratory. SK-UT-1 cell line was cultured in DMEM medium supplemented with 10% FBS, 1% penicillin/ampicillin (P/S), 1% sodium pyruvate, 0.1% MEM-nonessential amino acid solution, and 0.1% HEPES buffer. Both CP0024 and ICP060 were cultured in RPMI medium supplemented with 10% FBS, 1% P/S, and 1% Fungizone. All these cell lines were maintained at 37°C with 5% CO₂. SW982 and SW872 were cultured in L-15 medium supplemented with 10% FBS, 1% P/S, 0.1% HEPES buffer, and in 0% CO₂ air conditions. Tissue culture supplements were all purchased from Sigma-Aldrich. Cells were checked routinely and found to be free of contamination by mycoplasma or fungi and their authenticity checked prior to experiments. All the cells lines were discarded after seven to eight passes and new lines were obtained from frozen stocks.

IC₅₀ concentrations, which induced 50% of cell death, were calculated for doxorubicin (Sigma-Aldrich) and palifosfamide (AdoqBioscience) in SK-UT-1 and CP0024 cell lines. Probenecid at 200 μmol/L was used as a typical positive control of MRP-1 inhibition as well as MK-571 (1 μmol/L), another standard MRP-1 inhibitor. Nilotinib (NVP-AMN107-AA; Sigma-Aldrich) was tested at 10 and 5 μmol/L concentrations (13). Ripretinib (DCC-2618) and avapritinib (BLU-285) were selected as new tyrosine kinase inhibitor (TKI) compounds and purchased from MedChemExpress and used at the same concentration as nilotinib to check differences in activity between them. Selumetinib (AZD6244), a selective inhibitor of the enzyme mitogen-activated protein kinase kinase (MEK) subtypes 1 and 2 that has been described to reduce MPR-1 expression (14) was also used at 10 and 5 μmol/L concentrations. DMSO was used as a drug vehicle and negative control.

SK-UT-1 (1.5×10³) and CP0024 (2×10³) cells were seeded in 96-well plates and treated during 72 hours with drug concentrations ranging from 10^–5 to 10^–10M. Cell viability was measured using 3-(4,5-dimethylthiazol-2-yl)-5-(3-carboxymethoxyphenyl)-2-(4-sulfophenyl)-2H-tetrazolium, inner salt (MTS) method (Promega), recording absorbance at 490 nm in iMarkTM Microplate Absorbance Reader (Bio-Rad) spectrophotometer. Data were analyzed using Prism 6.0 (GraphPad).

RNA extraction from tissue culture was performed with RNA PureLink RNA Mini KIT (Invitrogen) following manufacturer's instructions. RNA quantification was carried out in NANOdrop spectrophotometer (NanoDrop One C; Thermo Fisher Scientific) and the ratios 230/280 nm and 260/230 nm were analyzed. RNA reverse transcription to cDNA was carried out with High Capacity Reverse cDNA Transcription Kit (Applied BioSystems; Thermo Fisher Scientific). Expression levels of the genes selected were measured by means of qRT-PCR using the following TaqMan RNA assays (Applied Biosystems): ABCB1 (Hs01067802_m1), ABCC1 (Hs00219905_m1), and ABCG2 (Hs01053790_m1) whereas B2M (Beta-2-microglobulin; Hs99999907_m1) was used as housekeeping gene for data normalization. An ABI Prism 7900HT (Applied BioSystems) real-time PCR system was used. After data normalization, the relative expression of the genes was evaluated, applying the 2^−ΔCt method.

Pools were used for transient gene silencing assays for ABCC1 [ON-TARGETplus SMARTpool ABCC1 small-interfering RNA (siRNA) (L-007308–00–0005)] and a siRNA negative control [ON-TARGETplus Non-Targeting Control Pool (D-001810–10–05); GE Healthcare; Dharmacon, Inc.]. Transfections were carried out using Lipofectamine RNAiMAX transfection reagent (Invitrogen; Thermo Fisher Scientific), according to the manufacturer's instructions. All siRNAs were used at 15 nmol/L concentration. The cells were subjected to the different treatments after 24 hours of silencing.

The fluorescent dyes 5(6)-carboxyfluorescein diacetate (CFDA) and rhodamine 123 (Rho-123) were used to examine the effect of inhibitors on MRP-1 and P-gp activity, respectively. SK-UT-1, CP0024, SW872, SW982, and ICP060 were incubated with 0.5 mmol/L CFDA dissolved in medium without serum or 0.5 mmol/L Rho-123 dissolved in complete medium for 30 minutes at 37°C. After removing the extracellular-free dye, cells were incubated in dye-free media with DMSO, probenecid (200 μmol/L), MK-571 (1 μmol/L), and 5 or 10 μmol/L of nilotinib, ripretinib, selumetinib, or avapritinib. In the case of ABCC1 silencing, CFDA was measured 24 hours after transfection. The efflux of CFDA and Rho-123 was analyzed by flow cytometry (Canto II flow cytometer; BD Biosciences) and data analyzed with BD FACS Diva (BD Biosciences) and FlowJo (FlowJo LLC) software. Mean CFDA or Rho-123 fluorescence of each compound was relativized to the DMSO sample. MRP-1 inhibition score was obtained after calculating the CFDA retention mean of all cell lines for each inhibitor, the highest value denoting the more active inhibiting MRP-1.

Intracellular doxorubicin fluorescence was quantified by flow cytometry in SK-UT-1 and CP0024 cell lines. Cells were pretreated with vehicle or with 10 or 5 μmol/L nilotinib, ripretinib, selumetinib, or avapritinib for 24 hours, and 500 nmol/L doxorubicin for SK-UT-1 and 1 μmol/L for CP0024 was then added for an additional 24 hours treatment. At this point, cells were harvested, washed, and resuspended in PBS at 37°C. Intracellular doxorubicin fluorescence was immediately measured by flow cytometry (Canto II flow cytometer), and the results were analyzed with both BD FACS Diva and FlowJo software.

The levels of apoptotic, early apoptotic, and necrotic cells were evaluated in SK-UT-1 and CP0024 cell lines. The doxorubicin and palifosfamide concentrations were adapted for each cell line. Thus, for the SK-UT-1 cell line, IC₅₀ concentrations were 100 nmol/L doxorubicin and 20 μmol/L palifosfamide and lower concentrations were 50 nmol/L doxorubicin and 10 μmol/L palifosfamide. For CP0024, IC₅₀ concentrations were 50 nmol/L doxorubicin and 100 μmol/L palifosfamide and lower concentrations were 25 nmol/L doxorubicin and 50 μmol/L palifosfamide. Avapritinib and nilotinib (5 or 10 μmol/L) or ABCC1 siRNA was administrated to the cells 24 hours before doxorubicin and palifosfamide. An FITC annexin V Apoptosis Detection Kit with PI was used to determine cell death (Immunostep) following the manufacturer's instructions. Apoptosis levels were determined by flow cytometry (Canto II flow cytometer) and data analyzed with both BD FACS Diva and FlowJo software.

SK-UT-1 (1.5×10³) and CP0024 (2×10³) cells were seeded in 96-well plates, 24 hours before being treated with nilotinib or avapritinib (5 or 10 μmol/L concentrations). Then, SK-UT-1 cell line was treated with DMSO, doxorubicin or/and palifosfamide at IC₅₀ concentrations (100 nmol/L doxorubicin–20 μmol/L palifosfamide) or lower concentrations (50 nmol/L doxorubicin–10 μmol/L palifosfamide) for 48 hours. For CP0024 cell line, IC₅₀ concentrations were 50 nmol/L doxorubicin–150 μmol/L palifosfamide and lower concentrations were 25 nmol/L doxorubicin–75 μmol/L palifosfamide. After 72 hours, cell viability was measured using MTS method (Promega), recording absorbance at 490 nm.

Cells (10³) were seeded in 10-cm dishes and treated with same scheme used in cell viability experiments. For this experiment, doxorubicin was used at 10 and 1 nmol/L, and palifosfamide used at 10 and 1 μmol/L. Each condition was tested in triplicate. After 10 days, colonies were fixed and stained with crystal violet assay. After extensive-washing, colonies were counted and the relative number of observed colonies were represented in a graph.

Data of biological replicates were grouped and presented as mean ± SE. Differences between two treatment conditions were statistically analyzed, when indicated, using the unpaired Student t test. Differences were considered significant when P < 0.05. Statistical analysis was performed using Prism 6.0 (GraphPad).

In silico binding affinity predictions for each drug were calculated by docking. Since there are no complete crystallized forms of human MRP-1, we used SWISS-MODEL (15) to predict the 3D structure of MRP-1 by searching the SWISS-MODEL template library (SMTL version 2020–08–17, PDB release 2020–08–17) with BLAST (16) and HHBlits (17) for evolutionary related structures. To reduce possible artifacts, we performed the analysis for 23 modeled MRP-1 structures based on the following templates: 6uy0, 6ujr, 6pzi, 6pzc, 6bhu, 5uja, 5kpi, 5kpd, 5koy, 5ko2, 4lsg, and 3g5u. Bovine MRP-1 structures (n = 8), which had been described to be highly similar (91% sequence identity) to human MRP-1 (18), were included in the analysis. In all cases, the co-crystallized hetero atoms were removed and the Autodock tools v1.5.7 (19) was used to prepare structures and ligands. The docking was performed with Autodock Vina v1.1.2 (20) with an exhaustiveness of 500. To reduce the search space, we manually inspected homologous MRP-1 proteins co-crystallized with other inhibitors to determine the binding sites and restricted the analysis to the transmembrane channel domain.

A total of 264 patients enrolled in the ISG-STS 1001 trial were included in this translational research. Of them, 175 had been randomly assigned to the standard arm (n = 88) or to the experimental arm (n = 87), whereas 89 were registered and treated with standard chemotherapy. The median age at diagnosis was 53, with a male predominance, 62% vs. 38%. Lower extremities accounted for 68% of cases, and the three most frequent subtypes were UPS (28%), high-grade myxoid liposarcoma (21%), and synovial sarcoma (16%; Table 1).

The MRP-1 expression was high in 112/264 (43%), low in 119/264 (44%), and nonevaluable in 33/264 (13%) patients. The intensity of MRP-1 expression was strong in 81/264 (31%), weak-negative in 150/264 (56%), and nonevaluable in 33/264 (13%) cases. More detailed information is presented in Table 1 and examples of MRP-1 immunostaining displayed in Supplementary Fig. S1. Higher MRP-1 expression was seen in synovial sarcoma (26/42, 62%), myxofibrosarcoma (17/33, 52%), and UPS (34/74, 46%), and similarly, the strong intensity was most frequently found in synovial sarcoma (19/42, 45%), myxofibrosarcoma (15/33, 46%), and UPS (24/74, 32%). Additional information of MRP-1 expression among different subtypes is shown in Supplementary Table S1. The level of concordance between both pathologists was excellent, for both MRP-1 expression (κ = 0.83; 95% CI, 0.71–0.95; P < 0.001) or intensity (κ = 0.90; 95% CI, 0.86–0.95; P < 0.001) evaluations.

In the univariate analysis, higher MRP-1 expression significantly correlated with worse DFS (HR, 2.71; 95% CI, 1.31–5.62; P = 0.007) in the patients randomly assigned to the standard chemotherapy arm. A trend toward a worse OS (HR, 2.75; 95% CI, 0.97–7.81; P = 0.058) was also observed in the same population.

Similarly, strong MRP-1 intensity was significantly related to both DFS (HR, 2.04; 95% CI, 1.03−4.06; P = 0.042) and OS (HR, 2.92; 95% CI, 1.11−7.67; P = 0.030) in the same subset of patients randomly assigned to the standard regimen of a full-dose anthracycline plus ifosfamide (Fig. 1).

In contrast, protein expression of MRP-1 (expression or intensity) did not show any prognostic relevance in those patients randomly assigned to the experimental arm (Supplementary Fig. S2). Likewise, no significant differences were seen in MRP-1 expression in the group of registered patients treated with standard chemotherapy (Supplementary Fig. S3). In this latter subset of patients, the most frequent subtype was myxofibrosarcoma (37%) for which MRP-1 expression did not show consistent reliability (Supplementary Fig. S4). Of note, no patients affected by myxofibrosarcoma were included in either randomized arm.

In the whole series, variables such as performance status (ECOG 0 vs. 1), gender, age, tumor size, and tumor site had no prognostic influence, whereas RECIST response had prognostic influence in both DFS and OS (Supplementary Table S2).

The strong intensity of MRP-1 expression was the only predictive factor independently associated with DFS (HR, 1.78; 95% CI, 1.11–2.83; P = 0.016) in the multivariate analysis, also showing a trend for OS (HR, 1.78; 95% CI, 0.97–3.25; P = 0.062; Supplementary Table S3).

ABCC1, the gene coding for MRP-1 protein, is expressed in all of the cell lines tested; SK-UT-1 and CP0024 expressed higher levels of the ABCC1 gene: 0.030 ± 0.009 and 0.023 ± 0.005, respectively, compared with the other cell lines used. On the other hand, they showed a lower expression of ABCB1 (2.137E^–05 ± 5.263E^–07 and 3.575E^–05 ± 2,582E^–05, respectively) or ABCG2 (3.725E^–04 ± 4.592E^–05 and 1.072E^–04 ± 2.553E^–06), compared with SW872, SW982, and ICP060 (Supplementary Fig. S5).

The effect of different compounds and TKIs on the activity of the MDR-relevant efflux pumps MRP-1 and P-gp was evaluated using the fluorescent dyes 5(6)-carboxyfluorescein diacetate (CFDA) and Rhodamine 123 (Rho-123) as MRP-1 or P-gp substrates, respectively. In the STS cell line panel used, we observed that nilotinib and avapritinib were able to reduce CFDA efflux, consequently increasing its intracellular concentration. In the SK-UT-1 cell line, the CFDA intracellular fluorescence was 10.08 ± 6.85-fold (P = 0.003) and 4.54 ± 3.15-fold (P = 0.007) superior, compared with the control cells treated with DMSO, when cells were treated with 10 or 5 μmol/L nilotinib, respectively. In the same cell line, CFDA dye retention was 9.69 ± 3.84-fold (P < 0.001) and 7.82 ± 4.12-fold (P < 0.001) higher with respect to DMSO treatment, when cells were treated with 10 and 5 μmol/L avapritinib, respectively. Similar results were also attained in the CP0024 leiomyosarcoma cell line: the increase in CFDA intracellular fluorescence was 6.29 ± 1.82-fold (P < 0.001) and 3.23 ± 1.13-fold (P = 0.001) higher compared with DMSO treatment, with 10 and 5 μmol/L nilotinib, respectively and 6.00 ± 0.47-fold (P < 0.001) and 5.32 ± 1.58-fold (P < 0.001) higher with 10 and 5 μmol/L avapritinib (Fig. 2A–D). The MRP-1 inhibitor probenecid, which was used as a positive control, induced an increase in the intracellular CFDA of 1.24 ± 0.32-fold (P = 0.04) in SK-UT-1 and 1.77 ± 0.42-fold (P = 0.002) in CP0024, relative to DMSO-treated cells. MK-571, ripretinib, and selumetinib effects in SK-UT-1 and CP0024 cells lines are also shown in Table 2 and Fig. 2. On the other hand, Rho-123 efflux was not significantly altered by any of the molecules tested in both SK-UT-1 and CP0024 cell lines (Table 2).

The effect of nilotinib and avapritinib on the intracellular accumulation of CFDA or Rho-123 dyes was also similar in ICP060, SW872, and SW982 (Supplementary Fig. S6; Table 2), except for SW872, in which an effect of the inhibitors on Rho-123 intracellular fluorescence was observed. In the SW982 cell line, the CFDA intracellular retention was 6.70 ± 3.21-fold (P = 0.004) and 3.69 ± 1.57-fold (P = 0.040) superior, compared with the control cells treated with DMSO, when cells were treated with 10 and 5 μmol/L nilotinib, respectively. In the same cell line, CFDA dye retention was 11.71 ± 4.58-fold (P = 0.020) and 7.07 ± 3.70-fold (P = 0.050) higher, compared with DMSO treatment, when cells were treated with 10 and 5 μmol/L avapritinib, respectively. Similar results were also attained in the ICP060 cell line: the increase in CFDA intracellular retention was 8.61 ± 2.73-fold (P = 0.008) and 5.59 ± 1.19-fold (P = 0.003), higher comparing to DMSO treatment, with 10 and 5 μmol/L nilotinib, respectively and 7.13 ± 4.17-fold and 7.53 ± 0.6-fold (P = 0.050) with 10 and 5 μmol/L avapritinib. The MRP-1 inhibitor probenecid, which was used as a positive control, induced an increase in the intracellular CFDA of 2.84 ± 0.83-fold (P = 0.01) in SW982 and 3.63 ± 1.11-fold (P = 0.01) relatively to DMSO-treated cells. On the contrary, the others inhibitors used did not show a clear inhibitory effect upon MRP-1. In the SW982 cell line, the CFDA intracellular fluorescence was 2.70 ± 0.5-fold, higher than in the control, with 1 μmol/L MK-571; 3.74 ± 0.96-fold and 2.57 ± 0.54-fold in respect to the control group, with 10 and 5 μmol/L ripretinib, respectively; and 1.28 ± 0.22-fold and 0.82 ± 0.14-fold with 10 and 5 μmol/L selumetinib, respectively. Similar results were also obtained for CDFA intracellular fluorescence in the ICP060 cell line: 3.51 ± 0.77-fold with 1 μmol/L MK571; 3.81 ± 0.69-fold and 2.39 ± 0.33-fold with 10 and 5 μmol/L ripretinib, respectively; and 1.3 ± 0.19-fold and 0.74 ± 0.39-fold with 10 and 5 μmol/L selumetinib, respectively (Table 2; Supplementary Fig. S6).

On the other hand, Rho-123 efflux was not significantly altered by any of the molecules tested in both SK-UT-1 and CP0024 cell lines. In the SK-UT-1 cell line, the fold of Rho-123 intracellular accumulation, compared with DMSO-treated cells was 0.96 ± 0.07-fold with 200 μmol/L probenecid; 0.92 ± 0.004-fold with 1 μmol/L MK-571; 1.18 ± 0.17-fold and 1.15 ± 0.11-fold with 10 and 5 μmol/L nilotinib, respectively; 0.92 ± 0.01-fold and 0.95 ± 0.07-fold with 10 and 5 μmol/L ripretinib, respectively; 0.99 ± 0.06-fold and 0.83 ± 0.08-fold with 10 and 5 μmol/L selumetinib, respectively; and 0.94 ± 0.09-fold and 1.07 ± 0.12-fold with 10 and 5 μmol/L avapritinib, respectively and in the CP0024 cell line was 0.88 ± 0.02-fold with 200 μmol/L probenecid; 0.77 ± 0.02-fold with 1 μmol/L MK-571; 1.09 ± 0.06-fold and 1.07 ± 0.09-fold with 10 and 5 μmol/L nilotinib, respectively; 1.05 ± 0.27-fold and 1.13 ± 0.15-fold with 10 and 5 μmol/L ripretinib, respectively; 0.93 ± 0.02-fold and 0.91 ± 0.1-fold with 10 and 5 μmol/L selumetinib, respectively; and 0.99 ± 0.05-fold and 1.03 ± 0.19-fold with 10 and 5 μmol/L avapritinib, respectively (Table 2; Fig. 2). On the other hand, Rho-123 efflux was not significantly altered by any of the molecules tested in both SW982 and ICP060 cell lines. In SW982 cell line, the fold of Rho-123 intracellular accumulation, comparing with DMSO-treated cells was 0.89 ± 0.01-fold with 200 μmol/L probenecid; 0.91 ± 0.03-fold with 1 μmol/L MK-571; 2.19 ± 0.37-fold and 2.26 ± 0.43-fold with 10 and 5 μmol/L nilotinib, respectively; 1.78 ± 0.35-fold and 1.56 ± 0.30-fold with 10 and 5 μmol/L ripretinib, respectively; 0.64 ± 0.10-fold and 0.67 ± 0.09-fold with 10 and 5 μmol/L selumetinib, respectively; and 1.74 ± 0.56-fold and 2.04 ± 0.37-fold with 10 and 5 μmol/L avapritinib, respectively and in ICP060 cell line was 1.01 ± 0.03-fold with 200 μmol/L probenecid; 0.84 ± 0.08-fold with 1 μmol/L MK-571; 3.12 ± 1.09-fold and 2.38 ± 0.52-fold with 10 and 5 μmol/L nilotinib, respectively; 1.98 ± 0.85-fold and 1.50 ± 0.26-fold with 10 and 5 μmol/L ripretinib, respectively; 0.96 ± 0.12-fold and 1.04 ± 0.18-fold with 10 and 5 μmol/L selumetinib, respectively; and 2.56 ± 0.67-fold and 3.14 ± 0.54-fold with 10 and 5 μmol/L avapritinib, respectively. However, in SW872, we weren't able to see differences between MRP-1 inhibitors and DMSO-treated cells in CFDA and Rho-123 intracellular fluorescence: 1.26 ± 0.34-fold CFDA and 1.01 ± 0.33-fold Rho-123 with 200 μmol/L probenecid; 1.71 ± 0.18-fold CFDA and 1.00 ± 0.32-fold Rho-123 with 1 μmol/L MK-571; 1.60 ± 0.49-fold CFDA - 2.16 ± 1.15-fold Rho-123 and 1.47 ±0.76-fold CFDA – 2.26 ± 0.88-fold Rho-123 with 10 and 5 μmol/L nilotinib, respectively; 1.16 ± 0.08-fold CFDA – 1.95 ± 0.53-fold Rho-123 and 1.07 ± 0.11-fold CFDA – 1.77 ± 0.47-fold Rho-123 with 10 and 5 μmol/L ripretinib, respectively; 0.99 ± 0.13-fold CFDA – 0.67 ± 0.08-fold Rho-123 and 0.93 ± 0.17-fold CFDA – 0.65 ± 0.09-fold Rho-123 with 10 and 5 μmol/L selumetinib, respectively; and 2.54 ± 0.82-fold CFDA – 1.93 ± 0.75-fold Rho-123 and 1.38 ± 0.32-fold CFDA – 1.96 ± 0.78-fold Rho-123 with 10 and 5 μmol/L avapritinib, respectively (Table 2; Supplementary Fig. S6).

When a score was applied to the MRP-1 inhibitory effect of these drugs, avapritinib and nilotinib were the most effective inhibitors of this efflux pump, with a mean of 6.62- and 5.18-fold of CFDA retention in the cell line panel used.

Also, ABCC1 transient silencing led to an accumulation of intracellular CFDA both in SK-UT-1 and CP0024 cell lines [1.51 ± 0.18-fold (P = 0.001) and 1.41 ± 0.05-fold (P < 0.001), respectively] compared with control cells treated with DMSO (Supplementary Fig. S7).

The effect of TKIs on intracellular doxorubicin fluorescence was determined, by flow cytometry, in SK-UT-1 and CP0024, which were the cell lines that expressed higher levels of ABCC1. In the SK-UT-1 cell line, avapritinib was able to augment intracellular doxorubicin accumulation after 48 hours of treatment: 1.49 ± 0.07-fold (P < 0.001) with 10 μmol/L avapritinib and 1.43 ± 0.06-fold (P < 0.001) with 5 μmol/L avapritinib (Fig. 2E). The effect of avapritinib on the intracellular accumulation of doxorubicin was not significant in CP0024 (Fig. 2F), neither with 5 μmol/L nor with 10 μmol/L, although a tendency was observed with 5 μmol/L avapritinib (1.13 ± 0.16-fold). On the other hand, nilotinib was capable of increasing doxorubicin accumulation in both SK-UT-1 and CP0024 cell lines: 1.25 ± 0.10-fold (P = 0.015) with 10 μmol/L nilotinib and 1.30 ± 0.1-fold (P = 0.006) with 5 μmol/L nilotinib in SK-UT-1 cells, and 1.19 ± 0.04-fold (P = 0.022) with 10 μmol/L nilotinib and 1.04 ± 0.01-fold (P = 0.036) with 5 μmol/L nilotinib in CP0024 (Fig. 2E and F).

The percentage of apoptotic/necrotic cells with each of the treatments was evaluated in SK-UT-1 and CP0024 cell lines, by flow cytometry, selecting 5 μmol/L avapritinib or 5 μmol/L nilotinib as working concentrations for this assay. In SK-UT-1, the percentage of apoptotic/necrotic cells increased from 24.9 ± 7.94% with doxorubicin and palifosfamide, in IC₅₀ concentrations, to 52.9 ± 18.08% when cells were treated with triple combination with 5 μmol/L (P = 0.037) of nilotinib, and from 21.2 ± 6.29% to 39.78 ± 13.8% (P = 0.049) when comparing doxorubicin plus palifosfamide in lower concentrations to the triple combination (Fig. 3A). A similar effect was also observed with 5 μmol/L avapritinib in combination with IC₅₀ concentrations of doxorubicin and palifosfamide (from 18.5 ± 5.35% to 40.3 ± 9.92%, P = 0.029) or with these two latter drugs in lower concentrations (from 16.03 ± 3.33% to 44.93 ± 5.37%, P = 0.001; Fig. 3C).

In the CP0024 cell line, there was also a statistically significant increase in the percentage of apoptotic/necrotic cells, when the activity of doxorubicin plus palifosfamide was compared with doxorubicin and palifosfamide with nilotinib. This difference was observed either when cells were treated with IC₅₀ concentrations of the drugs (65.85 ± 4.6% to 82.8 ± 3.34%, P = 0.006) or with lower concentrations of doxorubicin and palifosfamide (13.8 ± 5.09% to 33.13 ± 3.40%, P = 0.005; Fig. 3B). Likewise, similar differences were observed when double combination at IC₅₀ concentrations was compared with the triple combination with 5 μmol/L avapritinib: 30.35 ± 11.81% to 80.7 ± 18.1% (P = 0.042) and when comparing the effect of both drugs at lower concentrations with the triple combination (14.95 ± 3.94% to 31.55 ± 3.79%; P < 0.001; Fig. 3D). Combining ABCC1 siRNA with doxorubicin and palifosfamide resulted in increased apoptosis, either using both drugs at IC₅₀ (65.57 ± 4.94 vs. 43.40 ± 2.50; P = 0.045) or at lower concentrations (54.17 ± 1.94 vs. 37.75 ± 3.15; P = 0.001), compared with the double combination of doxorubicin and palifosfamide in the SK-UT-1 cell line (Supplementary Fig. S8). This also occurred in CP0024: 91.73 ±2.82 vs. 62.60 ± 9.30 (P = 0.034) at IC₅₀ concentrations of doxorubicin and palifosfamide, and 46.60 ± 10.15 vs. 18.83 ± 2.17 (P = 0.038) with lower concentrations of doxorubicin and palifosfamide (Supplementary Fig. S8).

The higher efficacy of the triple combinations of nilotinib or avapritinib with doxorubicin and palifosfamide compared with doxorubicin plus palifosfamide was also validated in cell viability and clonogenic assay studies, in the SK-UT-1 cell line (Supplementary Figs. S9 and S10). In SK-UT-1 cells, cell viability decreased significantly, when we compared the triple combination of 10 μmol/L avapritinib and doxorubicin and palifosfamide IC₅₀ concentrations (37 ± 2.72%) with the double combination of doxorubicin and palifosfamide IC₅₀ concentrations (60.18 ± 1.02%; P < 0.001). Cell viability was also lower when we decreased the concentration of avapritinib to 5 μmol/L for the triple combination (42.9 ± 3.26%; P < 0.001; Supplementary Fig. S9). Similar results were also attained when SK-UT-1 cells were treated, in combination with avapritinib, with doxorubicin and palifosfamide lower concentrations. Accordingly, cell viability was 36.99 ± 2.67% for the triple combination with 10 μmol/L avapritinib versus 68.09 ± 5.35% for double combination (P < 0.001), whereas cell viability was 46.05 ± 2.58% for doxorubicin plus palifosfamide lower concentrations and 5 μmol/L avapritinib versus 68.09 ± 5.35% for lower concentration double combination (P = 0.003; Supplementary Fig. S10).

Cell viability was also lower when SK-UT-1 cells were treated with 10 μmol/L nilotinib and lower concentrations of doxorubicin and palifosfamide (31.23 ± 6.00%), compared with double combination at lower concentrations: (56.19 ± 10.35%; P = 0.023; Supplementary Fig. S10). However, no statistically significant differences were observed in terms of cell viability, comparing the triple combinations versus the double combination, when SK-UT-1 cells were treated with the IC₅₀ concentrations of doxorubicin and palifosfamide (Supplementary Fig. S9).

Moreover, clonogenic assay studies carried out in SK-UT-1 cell line showed that 5 μmol/L nilotinib (358.25 ± 24.8; P < 0.001) or 5 μmol/L avapritinib (35.33 ± 33.49; P = 0.011; in combination with doxorubicin and palifosfamide, reduced significantly colony formation compared with doxorubicin and palifosfamide combination (Supplementary Fig. S9).

The in silico docking results showed that nilotinib (−10.54 kcal/mol) and avapritinib (−9.70 kcal/mol) had a lower mean of binding energy for MRP-1, compared with ripretinib (−9.41 kcal/mol), probenecid (−6.91 kcal/mol), MK-571 (−8.57 kcal/mol), and selumetinib (−7.50 kcal/mol). These values correspond to the mean affinity energies of the best dock pose among the MRP-1 modeled structures used. The results obtained for each structure analyzed are shown in Supplementary Table S4, whereas the docking of avapritinib and nilotinib to MRP-1 protein, as well as the amino acid residues participating in the TKI-protein binding are represented in Fig. 4.

The predictive role of the protein expression of MRP-1 has been validated, for DFS and OS in high-risk localized STS patients treated with epirubicin plus ifosfamide in the prospective randomized trial ISG-STS1001. These results confirm those previously reported within a retrospective exploratory analysis of a previous randomized trial, in the same patient population, treated with the same chemotherapy (9). Of note, the significant predictive correlation with the clinical outcome was found with data from two independent pathologists in a peer review blinded to clinical data within a multicentric trial. This analysis was a preplanned translational research project linked to the ISG-STS1001 trial.

Between the proteins associated to MDR, we selected only MRP-1 expression, based on our previous translational study (9). The prominent role of MRP-1 in extruding conjugates of oxazaphosphorines, including ifosfamide (11, 21), and anthracyclines (22) has resulted in a key predictive factor for resistance, in the context of patients affected by the most frequent limb sarcoma subtypes and treated with the combination of anthracycline plus ifosfamide. MRP-1 intensity expression was shown to be an independent factor in the multivariate analysis in our series. In contrast, in a small series of 29 localized STS treated with doxorubicin, dacarbazine, and ifosfamide in the neoadjuvant setting, pathologic response was correlated to P-gp expression, the only multidrug resistance (MDR) factor analyzed. Nine of 10 cases (90%) with positive P-gp expression had a poor response, whereas 7 over 19 (36%) P-gp negative cases had a poor response, P = 0.0078 (23). However, in another short series of 44 patients with STS, no significant prognostic correlation was found between P-gp expression and clinical outcome (24). Likewise, our previous analysis did not show any association of P-gp or GST-π, another protein related to MDR, expressions to outcome (9). These conflicting results about P-gp might be explained by the heterogeneity in the sarcoma subtypes and grades included in these analyses, as well as the experimental tools used for protein detection. Of note, MRP-1 overexpression has been directly correlated with a higher grade in STS (25).

Higher protein expression of MRP-1 has been significantly related to a worse clinical outcome in retrospective series of neuroblastoma (26, 27), whereas P-gp showed no prognostic correlation. Comparable with our study, in a series of 516 early breast cancer patients enrolled in the randomized ABCSG 5 trial comparing CMF versus endocrine adjuvant treatment, high expression (>30%) of MRP-1 by immunostaining was significantly related to worse relapse-free survival and OS. Interestingly, higher expression of MRP-1 showed an independent prognostic value only in patients treated with chemotherapy but not in those treated with just endocrine therapy (28).

An attractive field of investigation is exploring the combination of an MRP-1 inhibitor plus systemic agents that are substrates of certain efflux pumps. Nilotinib was shown to synergize with doxorubicin in sarcoma cell lines, observing a significant increase of apoptosis, a significantly higher decrease cell proliferation after withdrawal of compounds at day 12, probably related to an increase of doxorubicin accumulation in sarcoma cells and an increase of CDFA (specific MRP-1 substrate) by more than 11-fold with respect to the accumulation achieved with probenecid (13). In line with this, triterpenes maslinic acid and oleanolic acid were shown to enhance the antitumoral effect of doxorubicin through the inhibition of MRP-1 activity, but not of P-gp, in sarcoma cell lines (29). Among the plethora of compounds that have shown inhibitory effects on MDR efflux pumps, some of the most recent agents such as TKIs and other kinase inhibitors exhibited higher inhibition of activity than earlier used compounds (i.e., probenecid, verapamil, MK-571; ref. 30). In our study, avapritinib and nilotinib showed higher MRP-1 inhibition, compared with probenecid and MK-571, suggesting that the efficacy of these classic efflux pump inhibitors is in some way limited in the context of STS. In addition, nilotinib among other TKIs, has been shown to downregulate the expression of ABC transporters including MRP-1 (13, 31). In chronic myeloid leukemia cell lines, nilotinib was shown to be the most potent inhibitor of P-gp, breast cancer resistance protein (BCRP) and MRP-1, over imatinib and dasatinib. The authors suggested that therapeutic doses of nilotinib may diminish the resistance of MDR, especially P-gp and BCRP (31).

The first convincing evidence of clinical advantage through ABC transporter modulation came from three randomized trials, reporting significant longer OS in the arm treated with an ABC transport inhibitor. Patients with acute myeloid leukemia that were treated with daunorubicin plus cytarabine along with cyclosporine, a P-gp inhibitor, had significantly better relapse-free survival and OS rates compared with the arm without cyclosporine, 34% versus 9% and 22% versus 12%, respectively. The maximum benefit was detected in patients with the highest P-gp expression (32). Two additional randomized trials demonstrated significant better OS in the arms where verapamil was added to the chemotherapy regimens in patients with anthracycline-resistant breast cancer and non–small cell lung cancer (33, 34). In patients with sarcoma, the combination of nilotinib (400 mg/12 hours from day 1 to 6 of the cycle) plus doxorubicin (60, 65, and 75 mg/m² on day 5 of the cycle, as different dose-levels) was tested in a phase I trial. The recommended dose for the phase II part was 75 mg/m² for doxorubicin. Interestingly, patients had an average decrease in ABCC1 and ABCB1 RNA expression, measured in blood samples, between days 1 and 5 in all but one patient, who experienced disease progression (74% increase). A similar correlation between RNA levels and the radiologic response was seen in protein expression between basal biopsy and after fourth cycle. There were two partial responses (one of them in the dedifferentiated component of liposarcoma), eight stable diseases, and three progressions among seven advanced chondrosarcomas, four WD/DD liposarcomas, and three leiomyosarcomas. Of note, no cardiac toxicity was reported, neither in the left ventricular fraction nor clinically (35). The phase II trial with the same scheme was conducted in a neoadjuvant setting in retroperitoneal liposarcoma and leiomyosarcoma tumors and was already closed. In both parts of the trial, the scheme was feasible and safe, with the use of preventable GCSF after detecting some incomplete neutrophil recovery at day 21.

Noteworthy, the predictive of the resistance value of MRP-1 varied across sarcoma subtypes, suggesting that our hypothesis cannot be generalized to all STS histologies treated with anthracycline and ifosfamide. Thus, the overexpression of this efflux pump did not show any worse predictive role for myxofibrosarcoma in this series. Furthermore, it is doubtful that it would exert relevant predictive influence in high-grade myxoid liposarcoma in accordance with the experience of previous and current series. Liposarcoma showed the lowest MRP-1 protein expression in this series, which is in line with other observations measuring mRNA expression of MRP-1 (36). Future work should focus on the biological processes underlying drug resistance in these subtypes, which seems to be through different mechanisms, other than MRP-1 activity. In contrast, MRP-1 expression showed a significant predictive role in correlation some specific STS subtypes, such as synovial sarcoma and MPNST.

Limitations of this study include the exclusive focus on MRP-1, whereas no BCRP assessment was performed even if it targets both anthracyclines and ifosfamide. In addition, we did not assess MRP-1 mRNA expression. Constraints derived from a high number of competitive preplanned correlative translational studies explain these restrictions. In addition, the potential prognostic value of MRP-1 cannot be completely ruled out based on our results since an arm without perioperative chemotherapy was not included in the trial.

In the in vitro experiments performed in several sarcoma cell lines, nilotinib and avapritinib were shown to be the most potent MRP-1 inhibitors and demonstrated more efficacy for MRP-1 inhibition than for P-gp. Furthermore, using the same type of drugs tested in the randomized trials, where the predictive role of MRP-1 was validated, the co-administration of nilotinib or avapritinib with doxorubicin and palifosfamide resulted in a significant increase of apoptosis in sarcoma cell lines. The specific role of MRP-1 in the chemotherapy resistance of these compounds was validated using siRNA, obtaining similar results to those observed with the use of nilotinib or avapritinib. In the in silico analysis, the higher efficacy of MRP-1 inhibition by nilotinib and avapritinib over MK-571, ripretinib, selumetinib, or probenecid was supported. It is true that ripretinib showed an expected affinity for MRP-1 which was similar to avapritinib, but in the experiments, its inhibitory activity for MRP-1 was clearly inferior to nilotinib or avapritinib in sarcoma cell lines. This finding should be further studied. Moreover, in vivo experiments testing the combination of these MRP-1 inhibitors with anthracyclines and ifosfamide should also be performed in future preclinical studies.

These experiments provide us with the rationale for designing a phase I/II trial in advanced STS with a view to implementing it if signs of increased activity are detected, in a high-risk localized STS setting through a randomized trial. In this sense, the MRP-1 inhibitor would act as co-adjuvant therapy to standard chemotherapy and, hence, it would be administered during a limited number of days with the main purpose simply being to retain the active drugs in the tumor cells for a longer time.

J. Martin-Broto reports grants and personal fees from PharmaMar, Eisai, Eli Lilly, and Bayer and grants from Immix Biopharma, Novartis, BMS, Pfizer, AROG, Lixte, Karyopharm, Deciphera, GlaxoSmithKline, Blueprint, Nektar, Forma, Amgen, and Daichii-Sankyo outside the submitted work. M. Lopez-Alvarez reports nonfinancial support from Eisai and Eli Lilly and grants from Immix Biopharma, Pharmamar, and Novartis outside the submitted work. D.S. Moura reports grants and nonfinancial support from PharmaMar and Eisai, grants from Novartis and Immix Biopharma, nonfinancial support from Celgene, Bayer, and Pfizer outside the submitted work. C. Romagosa reports personal fees from Pharmamar and Roche outside the submitted work. M. Gambarotti reports personal fees from 6768 (ex6754) - Piero Picci Attivita Di Revisione I Tumori Maligni Occorsi Nello Studio Amgen 20062004DITT Amgen (Ricerca Commissionata) outside the submitted work. E. Palmerini reports personal fees from SynOx Therapeutic, Eusa Pharma, and Deciphera; nonfinancial support from Daiichi Sankyo, Pharmamar, Bristol Myers Squibb, Incyte, and Daiichi Sankyo outside the submitted work. S. Stacchiotti reports personal fees and other support from Novartis, Pharmamar, and Daiichi Sankyo, GlaxoSmithKline, and Bayer, Epizyme, Dechiphera, and Eli Lilly; personal fees from Maxivax, Ikena, Bavarian Nordic; other support from Springworks, Karyopharm, Advenchen, and Amgen outside the submitted work. G. Grignani reports grants from Bayer, Novartis, and PharmaMar; other support from Bayer, Eisai, Eli Lilly, Merck, and GlaxoSmithKline outside the submitted work. J. Blay reports grants and personal fees from Pharmamar, Bayer, Novartis, GlaxoSmithKline, and EISAI during the conduct of the study. A. Brunello reports other support from Glaxo-Smith Kline, Eli lilly, and Pharmamar outside the submitted work. C. Valverde reports personal fees and other support from PharmaMar, Pfizer, and Bayer, and personal fees from GlaxoSmithKline outside the submitted work. N. Hindi reports grants from European Union during the conduct of the study; grants, personal fees, and other support from PharmaMar; grants and personal fees from Eli Lilly; grants from Eisai, Novartis, AROG, Bayer, Lixte, Karyopharm, Deciphera, GlaxoSmithKline, Blueprint, Nektar, FORMA, Amgen, and Daichii-Sankyo outside the submitted work. A.P. Dei Tos reports personal fees from Bayer, Roche, and PharmaMar during the conduct of the study. A. Gronchi reports personal fees from Novartis, Pfizer, Bayer, Eli Lilly, SpringWorks, and Nanobiotix and grants and personal fees from PharmaMar outside the submitted work. No disclosures were reported by the other authors.

J. Martin-Broto: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. M. Lopez-Alvarez: Data curation, formal analysis, investigation, visualization, methodology, writing–original draft, writing–review and editing. D.S. Moura: Conceptualization, formal analysis, investigation, visualization, writing–original draft, writing–review and editing. R. Ramos: Formal analysis, investigation, methodology, writing–review and editing. P. Collini: Formal analysis, investigation, visualization, methodology, writing–review and editing. C. Romagosa: Resources, investigation, writing–review and editing. S. Bagué: Resources, investigation, writing–review and editing. S.L. Renne: Resources, investigation, writing–review and editing. M. Barisella: Resources, investigation, writing–review and editing. V. Valasco: Resources, investigation, writing–review and editing. J.-M. Coindre: Resources, investigation, writing–review and editing. D. Lopez-Lopez: Formal analysis, investigation, methodology, writing–review and editing. J. Dopazo: Formal analysis, supervision, investigation, writing–review and editing. M. Gambarotti: Resources, investigation, writing–review and editing. L. Braglia: Formal analysis, investigation, methodology, writing–review and editing. D.F. Merlo: Formal analysis, supervision, investigation, writing–review and editing. E. Palmerini: Resources, investigation, writing–review and editing. S. Stacchiotti: Resources, investigation, writing–review and editing. V.L. Quagliuolo: Resources, investigation, writing–review and editing. A. Lopez-Pousa: Resources, investigation, writing–review and editing. G. Grignani: Resources, investigation, writing–review and editing. J.-Y. Blay: Resources, funding acquisition, investigation, writing–review and editing. A. Brunello: Resources, investigation, writing–review and editing. A. Gutierrez: Formal analysis, investigation, methodology, writing–review and editing. C. Valverde: Resources, investigation, writing–review and editing. N. Hindi: Resources, investigation, writing–review and editing. A.P. Dei Tos: Resources, investigation, writing–review and editing. P. Picci: Resources, investigation, writing–review and editing. P.G. Casali: Resources, investigation, writing–review and editing. A. Gronchi: Conceptualization, resources, supervision, funding acquisition, investigation, visualization, project administration, writing–review and editing.

The study was partially funded through a European Union grant (EUROSARC FP7 278472). The authors would like to thank Patricio Ledesma Figueroa for Data Management. J. Martin-Broto, D.S. Moura and N. Hindi would like to thank the International Accelerator Award funded by Cancer Research UK [C56167/A29363], Associazione Italiana per la Ricerca sul Cancro [AIRC - 24297] and Fundacion Científica – Asociacion Espanola Contra el Cancer [Foundation AECC - GEACC19007MA]. D.S. Moura was supported by a Sara Borrell grant (CD20/00155) funded by the Instituto de Salud Carlos III. C. Romagosa was supported by a grant from the “Programa d'impuls del talent i de l'ocupabilitat del PERIS 2016–2020.” J-Y. Blay would like to express gratitude for the support from NetSARC (INCA and DGOS) EURACAN (EC 739521) and LYRICAN (INCA-DGOS-INSERM 12563). The ISH1001 study was supported by Eurosarc (FP7–278742).

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.