Autophagy Inhibition to Augment mTOR Inhibition: a Phase I/II Trial of Everolimus and Hydroxychloroquine in Patients with Previously Treated Renal Cell Carcinoma

Everolimus inhibits the mTOR, blocking a key downstream effector of growth factor signaling. Inhibition of mTOR causes a rapid decline in protein translation, including key glucose and amino acid transporters, resulting in abrogation of extracellular nutrient uptake. Thus, through inhibition of mTOR, everolimus is a potent initiator of metabolic stress for cancer cells (1).

A phase III randomized controlled trial (2) of everolimus 10 mg daily versus placebo in patients with advanced clear-cell renal cell carcinoma (ccRCC), who progressed on an approved tyrosine kinase inhibitor (TKI) demonstrated a median progression-free survival (PFS) of 4.01 months for everolimus-treated compared with 1.87 months for placebo-treated patients [HR, 0.30; 95% confidence interval (CI), 0.22–0.40; P < 0·0001; ref. 2]. Consequently, everolimus was commercially approved for the treatment of ccRCC after failure of treatment with sunitinib or sorafenib. Although safe and well tolerated, the 1% response rate and 4-month median PFS with everolimus reflects modest activity of this mTOR inhibitor on RCC as a single agent (2).

Autophagy is an intracellular process characterized by the formation of autophagic vesicles, which sequester cytoplasmic contents and target them for degradation in lysosomes. This process is activated by metabolic stress and also by most cancer therapies (3, 4). Autophagy is used by cancer cells to remove damaged organelles and recycle macromolecules that serve as an internal reservoir of fuel in the face of a crisis of nutrient unavailability. Inhibition of mTOR is one of the most potent inducers of autophagy. Therapeutic activation of autophagy may be a key resistance mechanism to mTOR inhibition with everolimus. Our initial in vivo studies demonstrate that inhibition of therapy-induced autophagy with chloroquine derivatives enhances cell death and tumor regression in a mouse lymphoma model (5). Furthermore, combining agents that target mTOR signaling with hydroxychloroquine results in enhanced cell death compared with each single agent alone in multiple cancer types (6, 7). Synergistic cell death was observed when the rapamycin analogue temsirolimus was combined with hydroxychloroquine in renal cell carcinoma (RCC) cell lines and in an orthotopic mouse model of RCC (8). In addition, we demonstrated the safety and preliminary activity of combining temsirolimus and hydroxychloroquine in patients with solid tumors (9). In this phase Ib study, the highest FDA-approved dose of hydroxychloroquine 600 mg orally twice daily was combined safely with temsirolimus. We hypothesized that combining mTOR and autophagy inhibition would be most effective in a disease such as ccRCC, where the mTOR inhibitor everolimus has single-agent activity. On the basis of this hypothesis, we conducted a multi-institution open-labeled phase I/II trial of everolimus in combination with hydroxychloroquine in patients with advanced clear-cell RCC (ccRCC).

Three centers (University of Pennsylvania Abramson Cancer Center, West Chester, PA; the University of Pittsburgh, Pittsburgh, PA; and Rutgers Cancer Institute of New Jersey, New Brunswick, NJ) enrolled patients with histologic evidence of metastatic RCC. For the phase I portion of the trial, patients could have any number of prior therapies and any RCC histology, but in the phase II portion, patients were required to have RCC with some clear-cell features and have previously received 1–3 prior regimens, including a VEGF receptor–TKI (VEGF-TKI).

This study was conducted in accordance with the U.S. Common Rule and received institutional review board approval at each respective center with informed written consent obtained from each subject. All patients had at least one measurable site of disease per RECIST 1.1 criteria (10), as well as normal organ function and an ECOG performance status of 0–2. Also patients had fasting serum cholesterol ≤ 300 mg/dL or ≤7.75 mmol/L and fasting triglycerides ≤ 2.5 × ULN, due to anticipated disruption of the glucose and lipid axis from everolimus.

For the phase I portion of the trial, patients received 10 mg everolimus for 1 week followed by the addition of 400 mg (dose level 1) or 600 mg (dose level 2) hydroxychloroquine twice daily beginning 1 week later. These doses continued for the duration of therapy. Cycle 1 included a 1-week run-in of everolimus alone to obtain pharmacodynamic markers. In subsequent cycles, both everolimus and hydroxychloroquine were given without interruption. Cycle length for the first cycle for all patients was 35 days. All subsequent cycles were 28 days of therapy. Imaging was obtained every two cycles for the assessment of response. Response was assessed by RECIST 1.1 criteria (10). Toxicity was assessed using the Common Terminology Criteria for Adverse Events version 4.0. Additional blood samples were obtained for pharmacokinetic and biomarker analyses described below. In addition, available archival tissue was obtained for predictive biomarker studies.

Dose-limiting toxicities (DLT) were defined by toxicity occurring during the first 5 weeks of this study in the phase I portion of the study. A DLT was any nonhematologic adverse event (AE) of grade 3 or higher that was at least possibly treatment-related with the exception of nausea and vomiting not treated with optimal antiemetic therapy. For hematologic toxicity, a DLT was defined as grade 4 neutropenia lasting more than 7 days or febrile neutropenia or platelet count less than 25,000/mm³.

Any DLT that caused a patient to miss >28 consecutive days of hydroxychloroquine resulted in the patient being taken off treatment. A DLT was defined as a grade 3 or 4 toxicity considered at least possibly related to hydroxychloroquine. Known toxicities specific to everolimus such as rash were not considered dose limiting unless the treating physician considered the toxicity to be exacerbated by hydroxychloroquine. Toxicities attributable to hydroxychloroquine included, but were not limited to, nausea, vomiting, diarrhea, and visual field deficit.

In the phase I portion of the study, the target DLT rate was ≤33%. The MTD was defined as (i) the dose producing DLT in 2 of 6 patients, or (ii) the dose level below the dose that produced DLT in ≥2 of 3 patients, or in ≥3 of 6 patients. Patients were evaluable for a DLT if they finished 5 weeks of combined therapies. Patients removed from the study due to clinical progression prior to the 5-week period in the phase I portion were replaced. Patients were evaluable for response if they completed 90% of their expected dose of hydroxychloroquine for the 8 weeks. Patients were evaluable for toxicity if they had received at least one dose of hydroxychloroquine.

If a DLT was observed in 1 patient per cohort, the cohort was expanded to 6. If a DLT occurred in 2 or more patients per cohort, then the cohort one dose below was the declared MTD provided that at least 6 patients had been treated at that level with no more than one-third having DLTs. No intrapatient dose escalation was allowed.

Hydroxychloroquine population pharmacokinetics was conducted as described previously (14). Briefly, whole blood was collected in tubes containing sodium heparin, and stored at −70°C until analysis. Whole-blood concentrations of hydroxychloroquine were measured using high-performance liquid chromatography with tandem mass spectrometry detection. Sample aliquots containing 500 ng of internal standard (IS; d4-HCQ) were vortexed with acetonitrile/methanol and then centrifuged. An aliquot of the supernatant was withdrawn, dried under nitrogen gas, reconstituted with mobile phase, and 10 μL injected onto a Kinetex 50 × 3 mm 2.6 μm 100A HPLC Column (Phenomenex). Samples were eluted with a gradient mobile phase of 0.1% formic acid in acetonitrile and water using a 1200 Series Agilent HPLC system with an API 4000 (AB SCIEX) mass spectrometer and electrospray interface operated in positive mode with multiple reaction monitoring detection. The capillary voltage was 4,000 V with a source temperature of 500°C. Mass spectrometer parameters were adjusted to maximize the intensity of the [M + H]⁺ ions in quadrupole 1 and the m/z transition ions of hydroxychloroquine (337.275 → 248.152) and IS (341.150 → 252.035) in quadrupole 3.

The mass spectrometers were controlled by AB SCIEX Analyst software (version 1.6.1) and data collection and analyses were conducted with the same software. Standard curves were constructed by plotting the analyte to IS ratio versus the known concentration of hydroxychloroquine (x) in each sample. Standard curves were fit by linear regression with weighting by 1/x. Samples were assayed in duplicate; samples for which the percent difference exceeded 15% were reanalyzed and the samples for which concentrations exceeded the range values for the calibration curve were diluted appropriately and reanalyzed. The calibration curve was linear from 1 to 7,500 ng/mL with correlation coefficients ranging from 0.9990 to 0.9999. The lower limit of quantitation was 1.0 ng/mL. The correlation coefficients for both inter- and intraday variability were <5.6% for each concentration (15 ng/mL, 150 ng/mL, 1,500 ng/mL, and 5,000 ng/mL) studied. The mean accuracy for inter- and intraday evaluations was between 97.2% and 102%.

Serial peripheral blood mononuclear cells (PBMC) were collected from patients before treatment, after one cycle of treatment, and after two cycles of treatment. Fixation and analysis by electron microscopy (EM) was conducted as described previously (11)

DNA was extracted from formalin-fixed, paraffin-embedded tissue and sequenced via the 147 gene UCM OncoPlus panel, as described previously (12). Briefly, DNA was fragmented (Covaris) and used for library preparation using the KAPA HTP Library Preparation Kit (Kapa Biosystems). Libraries were quantified via qPCR (Kapa Biosystems), pooled and captured using a custom-designed SeqCap EZ capture panel (Roche), and supplemented with select xGen Lockdown Probes (IDT). Amplified postcapture libraries were sequenced via HiSeq 2500 (Illumina) with rapid run v2 reagents to generate 2 × 101 bp sequencing reads. After demultiplexing, data were analyzed via custom bioinformatics pipelines, as described previously (12).

The primary objective for the phase I portion of the trial was to determine the MTD of hydroxychloroquine when administered with daily everolimus in patients with advanced RCC. The primary phase II objective was to estimate the rate of 6-month PFS in patients with RCC receiving everolimus and hydroxychloroquine who have had between one and three prior treatment regimens for advanced disease. Secondary objectives were (i) to estimate the response rate of this combination, (ii) to measure evidence of autophagy inhibition by EM to characterize significant associations between baseline genetic mutations and outcome.

On the basis of a Simon two-stage design for a single-arm phase II trial if greater than 5 patients were progression-free at 6 months, an additional 15 patients would be enrolled to complete this single-arm phase II trial. With α = 0.05 and 80% power with a total of 35 patients, if 14 or more patients were progression free (out of 35), the study would be considered a success and the regimen worthy of further investigation. This two-stage design was based on Kaplan–Meier survival curves indicating 6-month PFS of roughly 40% in RECORD-1, METEOR, and CHECKMATE 025 (2, 13, 14). Under this design, the probability of stopping the study in the first stage is 58% if the true 6-month PFS rate was 26% or less and the probability was 5% when the true 6-month PFS rate was 46% or more. For each gene, the binary variable was created; the presence or absence of the gene mutations was coded as 1 and 0, respectively. The heatmap and hierarchical clustering analysis was done by R package gplots heatmap.2. The order of the genes and the samples were reorganized by the hierarchical clustering. Kaplan–Meier curves were computed and an exact log-rank test was used to compare survival curves.

A total of 38 patients from the three participating centers were enrolled in this phase I/II trial from February 9, 2012 until January 16, 2017. Complete demographics for these patients are provided in Table 1. An additional 2 patients were screen failures and did not receive treatment with study agents. The majority of patients were Caucasian males with metastatic ccRCC who had received at least two prior therapies. Three patients were enrolled in dose level 1 (oral everolimus 10 mg daily plus oral hydroxychloroquine 400 mg twice daily) and 3 patients were enrolled at dose level 2 (oral everolimus 10 mg daily plus oral hydroxychloroquine 600 mg twice daily). Because no DLTs occurred, the phase II portion proceeded at dose level 2. Patients represented a refractory treatment population: 14 patients had received at least one prior regimen of therapy for their metastatic RCC disease, 14 patients had received two prior lines of therapy, and 9 patients had received at least three prior regimens.

No DLTs were observed in the phase I portion of the study. Hydroxychloroquine 600 mg orally twice daily in combination with standard dose everolimus 10 mg daily was identified as the recommended phase II dose and further evaluated for safety in the expansion cohort. Among 38 patients evaluable for AEs in both phase I and phase II, the most common grade 1–2 AEs attributable to treatment were nausea, fatigue, anemia, diarrhea, and rash (Table 2). All grade 3–4 AEs occurred at a rate of <10%. No grade 5 AEs were observed. Importantly the DLTs of everolimus were not significantly worsened by the addition of hydroxychloroquine.

The primary endpoint was to estimate the rate of 6-month PFS in patients with ccRCC receiving everolimus and hydroxychloroquine who have had between one and three prior treatment regimens for advanced disease. The median PFS and best overall response rate were the secondary clinical endpoints. When 15 patients had enrolled, a total of 5 patients achieved at least stable disease (SD) exceeding 6 months, and therefore met the desired endpoint, so the trial proceeded to the second stage. Of 38 patients treated on study, 33 were assessable for efficacy. Four patients withdrew from the trial for intolerable side effects that did not meet the definition of grade 3 or higher AEs (including bodyaches or nausea), and 1 patient deteriorated due to rapid disease progression within the first cycle of treatment. A total of 33 patients received a median of five treatment cycles. Among these patients, 22 of 33 (67%) experienced either partial response (2, PR) or SD (20) more than 3 months, as their best response per RECIST v1.1. Of the 22 patients who achieved SD or PR, 15 exceeded 6 months (Fig. 1A). The range was 7–21 months. Eleven patients developed disease progression at the time of first disease assessment (9 weeks). The median PFS was estimated to be 6.3 months (Fig. 1B).

We have previously performed population pharmacokinetic studies of hydroxychloroquine in combination with other cancer drugs in numerous phase I studies (9, 11, 13, 14). These studies have all used a limited blood sampling approach (one blood draw at the time of treatment visits) appropriate for a compound that has a long half-life. Pharmacokinetic studies of hydroxychloroquine in patients with rheumatoid arthritis have been performed with early intensive sampling following drug administration and showed a multiphasic exponential decline in hydroxychloroquine blood concentrations (15, 16). In this study, we therefore included early intensive sampling (predose, 15 minutes, 30 minutes, 1 hour, 2 hours, 3 hours, 4 hours, and 6 hours after dosing) in a small number of patients, providing concentration–time data during the absorption phase of hydroxychloroquine dosing. We performed population pharmacokinetic analysis using 163 nonbaseline blood samples from 27 patients collected over a period up to 616 days (average, 143; median, 89). The population model pharmacokinetic parameters do not specifically represent steady-state values, as they were determined from multiple repeated single doses taken by individual patients during their period of participation in the study. The model that best described the disposition of hydroxychloroquine blood concentrations was a 2-compartment model with first-order absorption with no lag time. No covariate interactions were identified that significantly improved the model. A nondiagonal fit was superior to a diagonal fit based on −2(LL). The final model was as follows: first-order absorption rate constant (Ka) = typical value (tv)Ka * exp(nKa); apparent volume of distribution in central compartment (V/F) = tvV * exp(nV)/F; apparent volume of distribution in peripheral compartment (V2/F) = tvV2 * exp(nV2)/F; apparent oral clearance (Cl/F) = tvCl * exp(nCl)/F; intercompartmental clearance (Q) = tvQ * exp(nQ); and lag time (tLag) = tvTlag = exp(nTlag). Visual inspection of conditional weighted residuals versus individual predicted values plots suggested that an additive error model was appropriate for intraindividual error (residual error). The first-order conditional maximum likelihood estimation, extended-least squares first-order conditional estimation (FOCE-ELS) method was used for modeling. A visual predictive check with 200 replicates was performed to assess the model performance. A total of 1,000 bootstrap runs were performed to provide estimates of the precision of parameter estimates and the 95% CIs for the pharmacokinetic parameters (Table 3). Figure 2 shows the individual predicted concentrations versus the individual observed concentrations from the population pharmacokinetic model.

Serial PBMCs were collected in a limited number of patients and fixed for quantitative EM as described previously (Supplementary Fig. S1A; ref. 13). Among 3 patients treated with hydroxychloroquine 400 mg and 8 patients treated with hydroxychloroquine 600 mg on the phase II portion, there were no significant drug-induced changes in autophagic vesicles observed in serially collected PBMCs, either after 1 week of everolimus alone or after 1 month combined everolimus and hydroxychloroquine. Although there was a trend of therapy-induced accumulation of vesicles in a minority of individual patients, there was no correlation between increased mean vesicle count/cell and response or PFS. (Supplementary Fig. S1B).

To determine genetic determinants of PFS, genomic DNA from archival tumor tissue was subjected to 147-gene panel massively parallel sequencing panel. Out of 33 evaluable patients, 29 patients had viable archival tumor for DNA and RNA sequencing analysis. DNA sequencing was successful in 22 of 29 samples, with the main reason for unsuccessful DNA sequencing being failed library preparation or inadequate DNA content. Hierarchical clustering was performed on known pathogenic mutations identified in the 147-gene panel (Supplementary Fig. S2A). Among the 22 patients, 6 patients had a potentially pathogenic-mutated gene in the PI3K/mTOR pathway. Two of the 6 mutations would likely not result in constitutively activated mTORC1 signaling (PIK3CA^F791I, and PTEN^L928M). The remaining four mutations were mTOR^G2223L (17), TSC1^L462K (18), FBXW7^G233A (19), and PIK3CB^D1067H (20), all of which produce truncating frameshifts. These four gene mutations are predicted to result in constitutively activated mTORC1 signaling and respond to mTOR inhibitors in vitro (17–21). There was a significant difference between the PFS survival curves of the patients that harbored these four pathogenic PI3K/mTOR mutations compared with the 18 patients that did not have these mutations (P = 0.033). The median PFS for the 4 patients that harbored these pathogenic PI3K/mTOR mutations and the 18 patients without a pathogenic PI3K/mTOR mutation was 55 days and 145 days, respectively (Supplementary Fig. S2B; Table 4).

The combination of everolimus and hydroxychloroquine was well tolerated with only everolimus-related toxicities observed. In this phase I/II study, the grade 3–4 AE rate was <10%. Although our observed 6-month PFS (15/33, 43%) met the primary endpoint, it was close to the observed results with everolimus alone in the three phase III studies (2, 13, 14). On the other hand, our median PFS was 6.3 months, which is longer than the median everolimus PFS for RECORD-1 of 4 months; METEOR, 3.8 months; and CHECKMATE, 025-4.6 months (2, 13, 14). During the course of this clinical trial, cabozantinib, nivolumab, and the combination of lenvatinib and everolimus have all become commercially available for treatment of metastatic RCC following treatment with a VEGF2-TKI (19–21). The combination of lenvatinib plus everolimus demonstrated a survival advantage in a randomized phase III study compared with either agent alone. The median PFS of this regimen was 14.6 months making it a very attractive combination treatment for previously treated advanced RCC (24). However, 71% of patients treated with lenvatinib and everolimus developed grade 3 or higher AEs on the lenvatinib/everolimus arm of this trial (24). Hydroxychloroquine did not significantly worsen the toxicity of everolimus in this study. Therefore, while further development of the everolimus and hydroxychloroquine combination studied here is not warranted in advanced RCC, the addition of hydroxychloroquine to the potent lenvatinib plus everolimus combination might further control disease progression without excess toxicity. Consideration of adding hydroxychloroquine to other RCC combinations is also provocative.

The hydroxychloroquine population pharmacokinetic parameters are comparable with previously reported analyses in patients with advanced cancer receiving oral hydroxychloroquine, where there is large interpatient variability in hydroxychloroquine pharmacokinetics (9, 11, 13, 14). Hydroxychloroquine is distributed widely in body tissues as reflected by its large apparent volume of distribution, particularly of the peripheral compartment. Hydroxychloroquine has a long half-life, which can be attributed to extensive distribution into tissues and partitioning into red blood cells (15). Because hydroxychloroquine is not extensively bound to plasma proteins, alterations in protein binding are unlikely to affect its disposition. Factors identified as having an effect on hydroxychloroquine disposition have not been extensively studied, but include body weight (14, 25). We did not find weight to have a significant effect on hydroxychloroquine parameter estimates in this analysis.

Pharmacokinetic studies with early sampling following drug administration show a multiphasic exponential decline in hydroxychloroquine blood concentrations (5, 7). We performed extensive sampling in a small number of patients after the first dose and at steady state, providing concentration–time data during the absorption phase of hydroxychloroquine dosing. Thus, we found that a 2-compartment model best fit the data. A lag time did not significantly improve the model fit, and thus it was not incorporated into the final population pharmacokinetic model.

Efficacy and toxicity are unlikely to be linked to differences in hydroxychloroquine pharmacokinetics. Although higher hydroxychloroquine doses and exposure have been associated with increases in markers of autophagy in PBMCs, relationships between blood–hydroxychloroquine concentrations with efficacy in limiting tumor progression or toxicities have not been consistently observed in combination anticancer studies and are generally absent with hydroxychloroquine monotherapy.

Our analysis of a limited set of serial PBMCs did not reveal a significant accumulation of autophagic vesicles in this study, despite both everolimus and hydroxychloroquine being well-accepted autophagy modulators, and the maximal dose of hydroxychloroquine allowed by the FDA being achieved. This is in contrast to what was observed with a phase I trial of temsirolimus and hydroxychloroquine in solid tumors (9), in which 600 mg twice daily hydroxychloroquine there was a significant accumulation of autophagic vesicles in the PBMCs of patients treated on the combination regimen. The sample size of this analysis was limited in this study, and a larger sample size may have found a significant therapy-associated change in vesicle accumulation. Another explanation for this difference is that pharmacodynamic studies of everolimus in mice and humans have demonstrated that measureable changes in phosphoS6 kinase, the downstream effector of mTOR signaling, following treatment with everolimus, are much more pronounced in tumor tissue than in PBMCs. In contrast, temsirolimus treatment produces sustained mTOR inhibition in PBMCs (26). Importantly, significant accumulation of autophagic vesicles has only been detected after one cycle of combined therapy across trials and never with single-agent therapy (11, 13, 14), suggesting that the vesicle accumulation observed in previous studies was reflecting combined autophagy induction and inhibition. If an agent is less effective at pathway blockade in PBMCs, this may explain why the combination of everolimus and hydroxychloroquine did not produce a significant accumulation of autophagic vesicles. This suggests that the EM assay used here may not be sensitive or specific enough to detect autophagy induction or inhibition well in PBMCs, or that the time-points chosen are too late to detect these changes. Alternatively, the use of PBMCs may be broadly problematic as a surrogate tissue to study autophagy dynamics, and as previous experience suggests, therapy-induced changes in autophagy may be occurring in a more striking fashion within the tumor tissue (9).

Who are the patients likely to respond to mTOR inhibitor combinations? In a previous study, analysis of a limited panel of cancer cell lines found that cell lines that harbor activating mutations in the mTOR kinase domain were more sensitive to rapamycin analogs than cell lines that harbored other mTOR pathway mutations (21). Clinical correlative studies in RCC have found that mutations in mTOR, TSC1, and TSC2 are found in higher frequency in patients who respond to rapamycin analogues (17, 18). Unexpectedly, in this trial, next-generation sequencing identified a number of mutations predicted to activate mTOR signaling in patients with the shortest PFS. This finding suggest that somatic mutations that have been suggested as predictive markers for rapamycin analogues fail to predict a combination involving a rapamycin analogue and a lysosomal inhibitor. It could be that hydroxychloroquine somehow limits the efficacy of everolimus in the presence of mTOR pathway mutations. Alternatively, this association between mTOR pathway–activating mutations and shortened PFS could be confounded by other genomic or epigenetic alterations that were not tested for in our study. Finally, only 22 patients were sequenced in this study, therefore our findings could be limited by small sample size.

In summary, the combination of hydroxychloroquine 600 mg and everolimus 10 mg daily was safe and prolonged SD was seen in a subset of patients. Activating mutations in the mTOR signaling pathway were associated with shorter PFS with this regimen.

N.B. Haas is a consultant/advisory board member for Merck and Exelixis. M. Stein reports receiving commercial research grants from Bristol-Myers Squibb, Merck Sharp & Dohme, Genocea Biosciences, Lilly, Nektar, Seattle Genetics, Harpoon, Advaxis, Janssen Oncology, Oncoceutics, and Medivation/Astellas and is a consultant/advisory board member for Merck Sharp & Dohme and Exelixis. J.P. Segal reports receiving commercial research grants from Abbvie. R.K. Amaravadi reports receiving other commercial research support from Novartis, holds ownership interest (including patents) in Immunaccel and Pinpoint Therapeutics, and is a consultant/advisory board member for Sprint Biosciences. No potential conflicts of interest were disclosed by the other authors.

Conception and design: N.B. Haas, M. Stein, L.E. Davis, R.K. Amaravadi

Development of methodology: N.B. Haas, X. Xu, L.E. Davis, R.K. Amaravadi

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): N.B. Haas, L.J. Appleman, M. Stein, M. Redlinger, M. Wilks, X. Xu, A. Onorati, J.P. Segal, L.E. Davis

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): N.B. Haas, L.J. Appleman, M. Stein, M. Wilks, A. Kalavacharla, T. Kim, S. Kadri, J.P. Segal, P.A. Gimotty, L.E. Davis, R.K. Amaravadi

Writing, review, and/or revision of the manuscript: N.B. Haas, L.J. Appleman, M. Stein, X. Xu, A. Kalavacharla, S. Kadri, P.A. Gimotty, L.E. Davis, R.K. Amaravadi

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): N.B. Haas, L.J. Appleman, M. Redlinger, M. Wilks, A. Onorati, C.J. Zhen, R.K. Amaravadi

Study supervision: N.B. Haas, L.J. Appleman, M. Redlinger, R.K. Amaravadi

This work was supported by investigator-initiated study funded by Novartis.

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.