Liquid Biopsies Using Plasma Exosomal Nucleic Acids and Plasma Cell-Free DNA Compared with Clinical Outcomes of Patients with Advanced Cancers

Mutations in the KRAS, BRAF, and EGFR genes are prevalent in many cancer types and can predict outcomes of targeted therapies (1–4). Therefore, accurate assessment of the mutation status is of utmost importance. Mutation testing is usually done using archival formalin-fixed, paraffin-embedded (FFPE) tumor tissue, which is not always available or is of inadequate quality (5). Furthermore, the molecular profiles of primary and metastatic sites differ (6, 7). In addition, the tumor molecular profile can evolve over time, which can be difficult to monitor in clinical practice, as serial tissue biopsies add to the cost of care and can lead to complications (8). Liquid biopsy–based approaches utilizing circulating sources of tumor DNA can provide alternative materials for molecular testing in cancer (9, 10). Plasma-derived cell-free DNA (cfDNA) is most commonly used for genomic testing; however, it originates from cells undergoing apoptosis or necrosis, which may not reflect the viable cell population of the tumor. In contrast, exosomal nucleic acids (exoNAs) such as DNA and RNA are actively secreted from living cells and might better correspond with tumor dynamics (11, 12). Molecular testing of plasma exosomal RNA (exoRNA) was described by Skog and colleagues, who demonstrated the presence of EGFRvIII mutations in serum exoRNA from patients with glioblastoma (13). In addition, Kahlert and colleagues reported detecting KRAS mutations in serum exosomal DNA (exoDNA) from patients with pancreatic cancer (11). San Lucas and colleagues demonstrated the next-generation sequencing (NGS)–based genomic and transcriptomic profiling of plasma exoDNA and exoRNA from patients with advanced pancreatic or biliary cancers (14). Finally, Allenson and colleagues detected KRAS-mutant plasma exoDNA in 80% to 85% of patients with locally advanced/metastatic pancreatic cancer and 44% of patients with early pancreatic cancer (15).

The purpose of the current study was to compare the assessment of common hotspot mutations in BRAF, EGFR, and KRAS using NGS of extracted exoNA and cfDNA with that using standard clinical testing of FFPE archival tumor samples in a Clinical Laboratory Improvement Amendments–certified laboratory and with that using droplet digital PCR (ddPCR) and BEAMing digital PCR testing of plasma cfDNA. In addition, we sought to determine whether the quantity of mutant exoNA and/or cfDNA is correlated with clinical outcomes and survival.

The study enrolled consecutive patients with progressing advanced cancers and BRAF^V600, KRAS^G12/G13, or EGFR^{exon19del/L858R} mutations (except for two patients without mutations, who were enrolled for specificity purposes) detected with clinical testing of their FFPE archival tumor specimens (Supplementary Methods) who were referred to MD Anderson's Department of Investigational Cancer Therapeutics (Houston, TX) for experimental therapies from February 2010 to April 2014 and consented to the protocol LAB10-0334. Whole blood was collected in EDTA-containing tubes and centrifuged twice within 2 hours to yield plasma. The study was conducted in accordance with MD Anderson's Institutional Review Board guidelines and the U.S. Common Rule.

Plasma exoNA (exosomal DNA and RNA, along with present cfDNA) were coisolated using the ExoLution Plus Isolation Kit (Exosome Diagnostics; Supplementary Methods, Supplementary Fig. S1). Subsequently, a quantitative NGS method EXO1000 was used to detect any hotspot mutations in BRAF exon 15, KRAS exon 2, and EGFR exons 19-21. The lower limit of detection is a mutation allele frequency (MAF) of approximately 0.05% in the wild-type allele background.

Plasma cfDNA was isolated using the QIAamp Circulating Nucleic Acid Kit (Qiagen) according to the manufacturer's instructions. Then, if available, 16 ng of unamplified cfDNA was assessed for BRAF^V600 or KRAS^G12/G13 mutations with a multiplex ddPCR Screening Kit or for EGFR^exon19del and EGFR^L858R mutations with mutation-specific assays (Bio-Rad) using the QX200 Droplet Digital PCR platform (Bio-Rad). The lower limit of detection is an MAF of approximately 0.2% for the multiplexed assay and an MAF of <0.1% for the single-well mutation-specific assay. Analysis of plasma cfDNA for common hotspot mutations in BRAF, EGFR, and KRAS using the BEAMing assays was conducted as described previously (16). Briefly, individual DNA molecules were attached to magnetic beads in water-in-oil emulsions and then subjected to compartmentalized PCR amplification. The mutational status of DNA bound to beads was determined by hybridization to fluorescent allele-specific probes for mutant or wild-type alleles of the gene of interest. Quantification of mutant DNA was performed using flow cytometry. The lower limit of detection is an MAF of approximately 0.02%.

Overall survival (OS) was defined as the time from the date of exoNA and cfDNA collection to the date of death or last follow-up. Time to treatment failure (TTF) was defined as the time from the date of systemic therapy initiation to the date the patient was taken off the treatment or last follow-up. The Kaplan–Meier method was used to estimate OS and TTF, and a log-rank test was used to compare OS and TTF among patient subgroups. Cox proportional hazards regression models were fit to assess the association between patient characteristics and OS or TTF. The Mann–Whitney U test was applied to assess the association among the response on imaging during therapy per RECIST 1.1 and plasma mutation status and quantity (17). All tests were two-sided, and P values <0.05 were considered statistically significant. All statistical analyses were performed with the SPSS 23 (SPSS) software program.

Of the 43 patients, 41 (95%) had a mutation of interest in their tumor tissues, including 20 (47%) with a BRAF^V600 mutation, 17 (40%) with a KRAS^G12/G13 mutation, and four (9%) with an EGFR^{exon19del/L858R} mutation. Most patients were Caucasian [29 (67%)] and male [24 (56%)]. The patients' median age was 57 years (range, 30–76 years). The most common tumor types were colorectal cancer in 20 patients (47%), melanoma in eight patients (19%), and non–small cell lung cancer (NSCLC) in six patients (14%). The median time from tissue sampling to blood sampling was 20 months (range, 0.1–140.4 months). Detailed patient characteristics are depicted in Table 1 and Supplementary File S1.

Mutation testing of plasma exoNA from 43 patients using NGS detected BRAF^V600 mutations in 19 of 20 patients with BRAF^V600 mutations in tumor tissue [sensitivity, 95%; 95% confidence interval (CI), 75–100], KRAS^G12/G13 mutations in all 17 patients with KRAS^G12/G13 mutations in tumor tissue (sensitivity, 100%; 95% CI, 80–100), and EGFR^{exon19del/L858R} mutations in three of four patients with EGFR^{exon19del/L858R} mutations in tumor tissue (sensitivity, 75%; 95% CI, 19–99; Supplementary Table S1). In total, NGS detected 39 of 41 mutations present in tumor tissue for an overall sensitivity of 95% (95% CI, 83–99; Table 2). NGS of plasma exoNA yielded two false-negative results: one for an appendiceal cancer patient with a BRAF^V600E mutation in tumor and one for NSCLC patients with an EGFR^L858R mutation in tumor (Table 3). None of the samples from patients with no mutation in tumor tissue detected a mutation in exoNA (specificity, 100%; 95% CIs, 85–100, 87–100, and 91–100 for BRAF^V600, KRAS^G12/G13, and EGFR^{exon19del/L858R} mutations, respectively; Supplementary Table S1).

Mutation testing of plasma cfDNA from 41 of 43 patients (no samples were available for two patients) using ddPCR detected mutations in 17 of 19 patients with BRAF^V600 mutations in tumor tissue (sensitivity, 89%; 95% CI, 67–99), KRAS^G12/G13 mutations in all 17 patients with KRAS^G12/G13 mutations in tumor tissue (sensitivity, 100%; 95% CI, 80–100), and EGFR^{exon19del/L858R} mutations in two of three patients with EGFR^{exon19del/L858R} mutations in tumor tissue (sensitivity, 67%; 95% CI, 9–99; Supplementary Table S1). In total, ddPCR detected 36 of 39 mutations present in tumor tissue for an overall sensitivity of 92% (95% CI, 79–98; Table 2). Testing of cfDNA using ddPCR yielded three false-negative results: One for a patient with Erdheim–Chester disease and a BRAF^V600E mutation in the tumor tissue, one for a patient with NSCLC and an EGFR^19del mutation in the tumor tissue, and one for a patient with rectal carcinoma and a BRAF^V600E mutation in the tumor tissue (Table 3). Testing with ddPCR detected a KRAS mutation in a prostate cancer patient that was not present in the patient's tumor tissue. The respective specificities were 100% (95% CI, 85–100) for BRAF^V600 mutations, 96% (95% CI, 79–100) for KRAS^G12/G13 mutations, and 100% (95% CI, 90–100) for EGFR^{exon19del/L858R} mutations (Supplementary Table S1).

Testing of plasma cfDNA from 37 of 43 patients (no samples were available for six patients) using BEAMing detected BRAF^V600 mutations in 13 of 14 patients with BRAF^V600 mutations in tumor tissue (sensitivity, 93%; 95% CI, 66–100), KRAS^G12/G13 mutations in all 17 patients with KRAS^G12/G13 mutations in tumor tissue (sensitivity, 100%; 95% CI, 80–100), and EGFR^{exon19del/L858R} mutations in all four patients with EGFR^{exon19del/L858R} mutations in tumor tissue (sensitivity, 100%; 95% CI, 40–100; Supplementary Table S1). In total, BEAMing detected 34 of 35 mutations present in tumor tissue for an overall sensitivity of 97% (95% CI, 85–100; Table 2). BEAMing of cfDNA yielded only one false-negative result for an appendiceal cancer patient with a BRAF^V600E mutation in tumor but yielded two false-positive KRAS mutation results: one for a patient with prostate cancer and one for a patient with ovarian cancer (Table 3). The respective specificities were 96% (95% CI, 78–100) for BRAF^V600 mutations, 85% (95% CI, 62–97) for KRAS^G12/G13 mutations, and 100% (95% CI, 88–100) for EGFR^{exon19del/L858R} mutations (Supplementary Table S1).

We analyzed whether the amount of mutant plasma exoNA or cfDNA was associated with OS. For plasma exoNA tested with NGS, the median OS duration of 21 patients with an MAF lower than or equal to the median MAF of 4.22% (11.8 months; 95% CI, 2.4–21.2 months) was significantly longer than that of 22 patients with an MAF higher than the median MAF (5.9 months; 95% CI, 4.7–7.1 months; P = 0.006; Fig. 1A). For plasma cfDNA tested with ddPCR, the median OS duration of 21 patients with an MAF lower than or equal to the median MAF of 6.1% (8.5 months; 95% CI, 2.1–14.9) was significantly longer than that of 20 patients with an MAF higher than the median MAF (5.9 months; 95% CI, 3.6–8.2 months; P = 0.023; Fig. 1B). For plasma cfDNA tested with BEAMing, there was a trend toward a longer median OS duration in 19 patients with an MAF lower than or equal to the median MAF of 7.22% (7.4 months; 95% CI, 4.3–10.5) compared to 18 patients with an MAF higher than the median MAF (6.5 months; 95% CI, 4.8–8.2 months; P = 0.066; Fig. 1C).

Next, to analyze the prognostic impact of exoNA and cfDNA MAFs on OS, we used a multicovariable analysis that included Royal Marsden Hospital (RMH) prognostic scores (18). The RMH score is a prospectively validated prognostic tool for patients with advanced cancers who are referred for early-phase clinical trials, which is calculated using lactate dehydrogenase levels (greater than the upper limit of normal vs. normal), albumin levels (<3.5 g/mL vs. ≥3.5 g/mL), and the number of metastatic sites (>2 sites vs. ≤2 sites). Scores of 0 or 1 are associated with longer OS than scores of 2 or 3. The median survival duration of 25 patients with RMH scores of 0 or 1 (8.4 months; 95% CI, 6.0–10.8) was significantly longer than that of 18 patients with RMH scores of 2 or 3 (6.0 months; 95% CI, 2.3–9.7 months; P = 0.017; Supplementary Fig. S2). A multicovariable Cox regression analysis that included RMH scores as well as exoNA, ddPCR cfDNA, and BEAMing cfDNA results demonstrated that an exoNA MAF of ≤4.22% was the only independent prognostic factor for OS (HR, 0.15; P = 0.026; Table 4).

We analyzed whether the amount of mutant plasma exoNA or cfDNA was associated with TTF in 32 patients who received systemic therapy. For exoNA tested with NGS, the median TTF of 15 patients with an MAF lower than or equal to the median MAF of 4.22% (7.4 months; 95% CI, 1.5–13.3) was significantly longer than that of 17 patients with an MAF higher than the median MAF (2.3 months; 95% CI, 1.2–3.4; P = 0.009; Fig. 2A). For plasma cfDNA tested with ddPCR, the median TTF of 14 patients with an MAF lower than or equal to the median MAF of 6.1% (8.6 months; 95% CI, 2.4–14.8) was significantly longer than that of 17 patients with an MAF higher than the median MAF (2.3 months; 95% CI, 1.2–3.4; P = 0.001; Fig. 2B). For plasma cfDNA tested with BEAMing, there was a trend toward longer median TTF in 12 patients with an MAF lower than or equal to the median MAF of 7.22% (3.7 months; 95% CI, 1.8–5.6) compared to 15 patients with an MAF higher than the median MAF (2.3 months; 95% CI, 1.3–3.3; P = 0.086; Fig. 2C).

We analyzed whether mutations in plasma exoNA or cfDNA were associated with response to therapy assessed by imaging using RECIST 1.1 in 32 patients who received systemic therapy (17). For exoNA tested with NGS, the median mutated exoNA MAF of 12 patients with partial response (PR) or stable disease (SD) ≥6 months (0.43%) was significantly lower than that of 20 patients with progressive disease (PD) or SD <6 months (14.74%; P = 0.006; Fig. 3A). For plasma cfDNA tested with ddPCR, the median mutated cfDNA MAFs of 12 patients with PR or SD ≥6 months (0.70%) and 19 patients with PD or SD <6 months (16.00%) did not differ significantly (P = 0.24; Fig. 3B). For plasma cfDNA tested with BEAMing, the median mutated cfDNA MAFs of 10 patients with PR or SD ≥6 months (5.44%) and 17 patients with PD or SD <6 months (12.88%) did not differ significantly (P = 0.24; Fig. 3C).

Our findings demonstrate that NGS testing for common hotspot mutations in plasma exoNA from patients with progressing advanced cancers has very good sensitivity overall (95%) compared with the standard testing of archival FFPE samples of discordantly obtained tumor tissue. This sensitivity of plasma exoNA testing was similar to those of simultaneous testing of plasma cfDNA with ddPCR and BEAMing (92% and 97%, respectively). Although BEAMing of plasma cfDNA had the highest sensitivity overall, this was offset by its lower overall specificity; of the three methods used, BEAMing detected the most mutations not detected in the archival tissue.

To our knowledge, ours is the first study to assess agreement between mutation testing of plasma exoNA and tumor tissue; however, our results with plasma exoNA appear to be similar to those of previously published data with plasma cfDNA (16, 19–24). In a previous study, we demonstrated in a similar patient population that the testing of plasma cfDNA with BEAMing has sensitivities of 76%, 100%, and 80% for BRAF, EGFR, and KRAS mutations, respectively, compared with the mutation analysis of FFPE primary or metastatic tumors (16). Similarly, other studies have reported that ddPCR testing of plasma cfDNA compared with tumor tissue has sensitivities of 84%, 69% to 86%, and 64% to 96% for BRAF, EGFR, and KRAS mutations, respectively (20, 21, 24). A certain level of disagreement between the results of mutation testing of plasma exoNA and testing of tumor tissue in our study could be explained by the discordant collection of both materials, as the median time from tumor tissue acquisition to plasma acquisition was 20 months. A small retrospective study of patients with metastatic breast cancer tested with BEAMing for PIK3CA mutations in plasma cfDNA and simultaneously collected tumor tissue demonstrated 100% agreement between the methods; however, the sensitivity decreased to 57% in a prospective cohort when plasma cfDNA and tissue were collected at different points (25). In another study of 100 patients with advanced colorectal cancer, the ddPCR detection of RAS mutations in plasma cfDNA was in agreement with that in archival tissue in 97% of cases (26). This rate was favorable compared with most other studies, including ours; however, the median time from tissue collection to plasma collection was only 43 days, which could explain the high agreement rate. In addition, tumor heterogeneity, clonal evolution, and preanalytic factors such as suboptimal specimen collection can also contribute to discrepancies (6, 27). Of interest, in our study, two of the methods used missed a BRAF^V600E mutation that was present in the tumor of a patient with appendiceal cancer, but this could have been due to the biology of that specific disease.

We found that patients with a low amount (i.e., that below or equal to the median amount) of mutated plasma exoNA had longer survival than patients with a high amount of mutated exoNA did (P = 0.006), which was confirmed in a multicovariable analysis. Patients with a low amount of mutated plasma cfDNA tested with ddPCR had significantly longer survival than patients with a high amount of mutated plasma cfDNA (P = 0.023), and there was a trend for patients with a low amount of mutated plasma cfDNA tested with BEAMing to have longer survival than patients with a high amount of mutated plasma cfDNA (P = 0.066). To our knowledge, this is the first report suggesting a relationship between the amount of mutated plasma exoNA and survival. In our study, exoNA fared better then cfDNA, possibly because, unlike cfDNA, exoNA originates from living cancer cells (11, 12). However, our findings have to be interpreted cautiously, as our sample size was small. An earlier study using BEAMing PCR to detect KRAS mutations in plasma cfDNA from patients with advanced cancers found that a low amount of KRAS-mutant cfDNA was associated with longer median survival (7.3 vs. 4.8 months; P = 0.008; ref. 16). Another study that used the Idylla system to detect BRAF^V600 mutations in plasma cfDNA from patients with advanced cancers showed that a low percentage of BRAF^V600-mutant cfDNA was associated with longer survival (10.7 vs. 4.4 months; P = 0.005; ref. 28). Similarly, baseline samples collected from patients with advanced colorectal cancer treated in a phase III randomized trial of regorafenib versus placebo showed that low baseline levels of KRAS-mutant cfDNA were associated with longer survival durations (29). In addition, low amounts of KRAS-mutant cfDNA were associated with longer survival durations in patients with advanced colorectal cancer treated with irinotecan and cetuximab and in patients with advanced NSCLC treated with carboplatin and vinorelbine (30, 31). Similarly, in a combined analysis of clinical trials of BRAF and MEK inhibitors in patients with advanced melanoma and BRAF mutations in their tumors, the absence of BRAF^V600E mutations in plasma cfDNA was associated with longer median survival (22).

We also demonstrated that patients with a low amount of mutated plasma exoNA identified with NGS or cfDNA identified with ddPCR had a significantly longer TTF than patients with a high amount of mutated exoNA or cfDNA did (P = 0.009 and 0.001, respectively). In addition, patients with a low amount of mutated plasma cfDNA identified with BEAMing had a trend to longer TTF than patients with a high amount of mutated cfDNA did (P = 0.086). Allenson and colleagues demonstrated that the amount of mutated exoDNA can be predictive of disease-free survival in patients with localized pancreatic cancer (15). In a previous study, using allele-specific quantitative PCR (Idylla) to test for BRAF mutations in plasma cfDNA, we demonstrated that patients with advanced cancers and BRAF mutations in their tumors but not plasma cfDNA had a longer median TTF on BRAF and/or MEK inhibitors than patients with detected BRAF mutations in cfDNA did (3.0 vs. 13.1 months; P = 0.001). In addition, a retrospective analysis of plasma samples collected from patients with malignant melanoma and BRAF mutations in their tumors who were treated with BRAF and MEK inhibitors in clinical trials also demonstrated that the absence of BRAF mutations in cfDNA predicted longer progression-free survival (22).

Moreover, we showed that patients who received systemic therapy who had a PR or SD ≥6 months had a significantly lower median amount of mutated plasma exoNA than patients without a PR or SD ≥6 months did (P = 0.006). This was not observed for cfDNA tested with ddPCR or BEAMing (P = 0.24 and 0.24, respectively). Weiss and colleagues demonstrated that quantitative chromosomal instability in plasma cfDNA can predict immunotherapy response in patients with advanced cancers (32). Although some data suggest that changes in mutated cfDNA during therapy can correspond with treatment response, our study is the first to our knowledge to show that pretreatment levels of mutated exoNA and cfDNA can be associated with response to systemic therapy (33).

Our study had several potential limitations. First, the study was retrospective and enrolled relatively few patients. Second, its patient population was heterogeneous, with diverse tumor types that were treated with diverse therapies. Third, we tested for only selected common hotspot mutations in the BRAF, KRAS, and EGFR genes. These mutations are relevant in only a limited number of patients with certain cancer types; thus, whether our findings can be extrapolated to other molecular abnormalities remains unclear. Fourth, nearly half of the patients (47%) had colorectal cancer, which could have influenced the results. Fifth, we used archival tumor tissue, which was not collected at the same time as plasma samples, and this may have impacted our sensitivity assessments. Finally, our observation that increased plasma MAF can be associated with worse outcomes irrespective of disease burden (using the RMH score) is intriguing; however, it needs to be interpreted with caution because our tools to estimate disease burden are not optimal.

In conclusion, our findings suggest that molecular testing of plasma exoNA has potential in cancer molecular diagnostics and can be predictive of clinical outcomes. Thus, future clinical studies to assess its utility in furthering personalized cancer therapy are warranted.

D.S. Hong has ownership interests (including patents) at OncoResponse, reports receiving speakers bureau honoraria from LOXO and miRNA, is a consultant/advisory board member for Baxter, Bayer, and Guidepoint Global, and reports receiving commercial research grants from Amgen, Bayer, Daichii-Sanko, Eisai, Genentech, Ignyta, Lilly, LOXO, Merck, Mirati, and Pfizer. V. Subbiah reports receiving commercial research support from Bayer, Blueprint, LOXO, Novartis, and Roche/Genentech. C. Eng reports receiving speakers bureau honoraria from Genentech, and is a consultant/advisory board member for Bayer and Sirtex. F. Meric-Bernstam is a consultant/advisory board member for Clearlight Diagnostics, Darwin Health, Dialecta, GRAIL, Inflection Biosciences, Pieris, and Sumitomo Dainippon and reports receiving commercial research grants from Aileron, AstraZeneca, Bayer, Calithera, Curis, Debiopharma, eFFECTOR, Genentech, Novartis, Pfizer, PUMA, Taiho, and Zymeworks. F. Janku has ownership interests (including patents) at Trovagene, is a consultant/advisory board member for Deciphera, Guardant Health, Illumina, Trovagene, and reports receiving commercial research grants from Agios, Astellas, BioMed Valley Discoveries, Deciphera, Genentech, Novartis, Piqur, Plexxikon, Symphogen, and Upsher-Smith Laboratories. No potential conflicts of interest were disclosed by the other authors.

Conception and design: J. Skog, F. Janku

Development of methodology: H.J. Huang, V. Subbiah, A. Krug, D. Enderle, M. Noerholm, C. Coticchia, J. Skog, F. Janku

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): L. Möhrmann, H.J. Huang, D.S. Hong, A.M. Tsimberidou, S. Fu, S.A. Piha-Paul, V. Subbiah, D.D. Karp, A. Naing, D. Enderle, M. Noerholm, E. Eitan, G. Stoll, L.-M. Jordan, E.S. Kopetz, J. Skog, F. Meric-Bernstam, F. Janku

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): L. Möhrmann, H.J. Huang, D.S. Hong, A.M. Tsimberidou, S. Fu, V. Subbiah, A. Naing, A. Krug, D. Enderle, T. Priewasser, M. Noerholm, E.S. Kopetz, J. Skog, F. Meric-Bernstam, F. Janku

Writing, review, and/or revision of the manuscript: L. Möhrmann, D.S. Hong, A.M. Tsimberidou, S. Fu, S.A. Piha-Paul, V. Subbiah, D.D. Karp, A. Naing, D. Enderle, T. Priewasser, M. Noerholm, L.-M. Jordan, C. Eng, E.S. Kopetz, J. Skog, F. Meric-Bernstam, F. Janku

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): L. Möhrmann, A.M. Tsimberidou, V. Subbiah, M. Noerholm, J. Skog, F. Janku

Study supervision: J. Skog, F. Janku

The authors would like to thank Prof. Dr. Gerhard Wanner and Alexandra Spiel for the scanning electron microscopy on the eluted vesicles and Prof. Dr. Jörg Kleiber for providing access to the second NanoSight instrument for cross-instrument comparisons. The authors would also like to acknowledge Joseph Munch from the Department of Scientific Publications, the University of Texas MD Anderson Cancer Center for his editorial and grammar assistance. L. Möhrmann was supported by scholarships of the German Academic Exchange Service (Deutscher Akademischer Austauschdienst, DAAD), the German Academic Scholarship Foundation (Studienstiftung des deutschen Volkes), and the Evangelisches Studienwerk e.V. Villigst.

This study was supported by the Sidney Kimmel Foundation for Cancer Research, the Sheikh Khalifa Al Nahyan Ben Zayed Institute for Personalized Cancer Therapy, and the National Center for Advancing Translational Sciences (grant No. UL1 TR000371). This study was also supported by the National Institutes of Health through MD Anderson's Cancer Center Support Grant (P30 CA016672).

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.