Clinical Utility of Circulating Cell-Free DNA Mutations in Anaplastic Thyroid Carcinoma

Abstract

Background: Anaplastic thyroid carcinoma (ATC) is an aggressive thyroid cancer that requires a rapid diagnosis and treatment to achieve disease control. Gene mutation profiling of circulating cell-free DNA (cfDNA) in peripheral blood may help to facilitate early diagnosis and treatment selection. The relatively rapid turnaround time compared with conventional tumor mutation testing is a major advantage. The objectives of this study were to examine the concordance of ATC-related mutations detected in cfDNA with those detected in the corresponding tumor tissue, and to determine the prognostic significance of cfDNA mutations in ATC patients.

Methods: The ATC patients who were diagnosed and treated at The University of Texas MD Anderson Cancer Center between January 2015 and February 2018 and who had cfDNA testing were included in this study. cfDNA was collected by blood draw and was analyzed by next-generation sequencing (NGS) using the Guardant360-73 gene platform.

Results: A total of 87 patients were included in the study. The most frequently mutated genes detected in cfDNA were TP53, BRAF, and PIK3CA. In 28 treatment naive ATC patients, the concordance rate of detected mutations in TP53, BRAF^V600E, and PIK3CA between cfDNA and matched tissue NGS was 82.1%, 92.9%, and 92.9%, respectively. Patients with a PIK3CA mutation detected on cfDNA had worse overall survival (OS) (p = 0.03). This association was observed across various treatment modalities, including surgery, cytotoxic chemotherapy, radiation, and BRAF inhibitor (BRAFi) therapy. With regard to treatment, BRAFi therapy significantly improved ATC OS (p = 0.003).

Conclusions: cfDNA is a valuable tool to evaluate a tumor's molecular profile in ATC patients. We identified high concordance rates between the gene mutations identified via cfDNA analysis and those identified from the NGS of the corresponding tumor tissue sequencing. Identified mutations in cfDNA can potentially provide timely information to guide treatment selection and evaluate the prognosis in patients with ATC.

Introduction

Anaplastic thyroid carcinoma (ATC) is rare but is one of the most aggressive solid tumors in humans. The ATC accounts for 1–2% of all thyroid cancers (1). The historical median overall survival after diagnosis is 3–6 months and only 10–15% ATC patients have a survival of 2 years after presentation. The disease-specific mortality rate approaches 100% (1).

The clinical course of ATC is characterized by rapid and invasive local tumor growth, frequent distant metastases, and fatal outcomes. Therefore, ATC requires rapid diagnosis and prompt treatment. Targeted therapy, such as BRAF inhibitor (BRAFi) treatment for patients with BRAF^V600E mutated ATC, shows promising results in ATC treatment (2) and the U.S. Food and Drug Administration (FDA) has approved the BRAF/MEK inhibitor combination dabrafenib/trametinib for ATC patients with BRAF^V600E mutated tumors. Understanding ATC genetic features and their clinical significance has become essential for ATC management.

Liquid biopsies to detect circulating cell-free DNA (cfDNA) for cancer genotyping have been increasingly used in clinical practice as reliable and minimal invasive methods for many solid tumors, including lung cancer and gastric cancer (3). Compared with the molecular testing performed on tissue biopsies, cfDNA-based analysis has several advantages, including the fact that the sampling procedure is minimally invasive and that serial samples are easily obtained. The method also allows a rapid turn-around time and therefore captures a real-time tumor molecular profiling (3).

We previously reported the utility of using cfDNA to obtain a rapid and reliable molecular profiling in 23 ATC patients, particularly in those who were treatment naive at the time of sampling (4). In the current study, we have expanded this cohort to 87 cases (including the previous 23 patients) to explore the clinical significance of cfDNA testing.

Methods

Patients

The ATC patients who were diagnosed and treated at The University of Texas MD Anderson Cancer Center (MDACC) between January 2015 and February 2018 and who had cfDNA testing and next-generation sequencing (NGS) on tumor tissue samples were included in this study. cfDNA was collected during the patient's first visit to MDACC. If the patient had more than one cfDNA test performed, only the first test was used for analysis.

Molecular testing

cfDNA was isolated from the patients' peripheral blood and analyzed by using NGS by Guardant Health Inc., as previously described (4). The Guardant360^®73 gene platform was utilized for the test and the cfDNA assay was reported to have a sensitivity and specificity of 85%+ and >99.9999%, respectively (5). Genomic alterations covered by Guardant360 included 73 somatic point mutations, amplifications in 18 genes, and fusions in 6 genes (Supplementary Table S1). Tumor tissue genotyping was performed by using an in-house NGS platform (CM50), which detects point mutations, amplifications, short insertions, and deletions in a total of 50 genes (6,7).

Summarized in Supplementary Tables S2 and S3 are all mutations identified in primary tumors and cfDNA samples, including gene and protein coordinates, cDNA and amino acid changes, as well as their oncogenic score using TransVar (8) and CScape (9). Supplementary Table S4 lists all primary tumor mutations for each patient with amino acid changes and variant allele frequency. All testing and data collection were performed as part of standard-of-care treatment and under an MDACC Institutional Review Board approved protocol.

Statistical analysis

Descriptive statistics were used for demographics and cfDNA mutations analysis. Survival analysis was performed by using the Kaplan–Meier method and the log rank test. All analyses were performed by using R software (Version 3.5.1). When appropriate, p-values were adjusted for multiple testing by using the Benjamini–Hochberg approach. A p < 0.05 was considered statistically significant.

Results

Patient characteristics

Eighty-seven ATC patients were included in the analysis. The median age was 65 years (range, 35–88 years). Of the 87 patients, 72 (83%) had cervical lymph node metastases and 56 (64%) had distant metastases at the time of diagnosis. Sixty-two patients (71%) underwent surgery with at least total thyroidectomy, 64 (74%) patients had cytotoxic chemotherapy, and 61 (70%) patients had external beam radiation to the neck. Median follow-up duration was 11.7 months (range, 1.2–45.4 months). Thirty of the 87 patients (35%) were alive at the time of analysis. Patient information is summarized in Table 1.

Table 1.

Patient Clinical Characteristics

Characteristics	No. of cases	%
Age (years)	Mean: 65 Range: 35–88
Sex
Male	59	67.8
Female	28	32.2
History of thyroid disease
PTC	27	31.0
PDTC	5	5.8
Goiter	4	4.6
Hürthle cell	3	3.4
FTC	2	2.3
None	46	52.9
T stage
T4a	11	12.6
T4b	72	82.8
NA	4	4.6
N stage
N0	8	9.2
N1	72	82.8
Nx	7	8.0
M stage
M0	29	33.3
M1	56	64.4
Mx	2	2.3
Surgery
No	25	28.7
Yes	62	71.3
Systemic treatment
Cytotoxic chemotherapy	64	73.6
Radiation therapy	61	70.1
BRAF-directed therapy	32	36.8
Immunotherapy	38	43.7
Kinase inhibitor therapy	28	32.2

Cytotoxic chemotherapy, paclitaxel, carboplatin, cisplatin, doxorubicin; FTC, follicular thyroid carcinoma; Immunotherapy, pembrolizumab, atezolizumab, ipilimumab, nivolumab; Kinase inhibitor therapy, lenvatinib, everolimus; NA, not available; PDTC, poorly differentiated thyroid carcinoma; PTC, papillary thyroid carcinoma.

Cell-free DNA results

Eighty of 87 (92%) patients had at least one mutation detected in their cfDNA, and 7 patients had no mutations detected. No fusions were detected in the cfDNA analyzed. The median number of mutated genes detected per patient was 3 (range, 0–16 mutated genes). As shown in Figure 1, the most frequently mutated genes detected in cfDNA were TP53 (60.9%), BRAF^V600E (34.4%), PIK3CA (16.1%), EGFR (14.9%), RAS (N-, H-K-; 13.8%), and NF1 (13.8%). RAS and BRAF^V600E mutations were mutually exclusive. The specific mutations detected for each gene are shown in Table 2.

FIG. 1.

Oncoprint of cfDNA mutations. Gene mutations found on cfDNA in patients with ATC are shown and listed by frequency. TP53 and BRAF^V600E were the most common mutations, followed by PIK3CA, EGFR, RAS, and NF1. RAS includes NRAS 9 cases (10%), KRAS 2 cases (2%), and HRAS 2 cases (2%). RAS and BRAF mutations were mutually exclusive. Red boxes with an X in the box indicate truncating mutations. ATC, anaplastic thyroid carcinoma; cfDNA, cell free DNA.

Table 2.

Cell-Free DNA Mutations Detected by the G360 Platform

Genes	Mutations (no. of cases)
TP53	R273H (6), R175H (4), Q331* (4), R248Q (2), R248G (2), R248W (2), D208V (2), R280T (2), Y220C (2), G266R (2), A189D (1), A83fs (1), D259V (1), E258fs (1), E271K (1), E56* (1), Exon 7 Insertion (1), F134L (1), G108S (1), G266E (1), R158H (1), H168R (1), H193R (1), K132M (1), Q192* (1), Y234S (1), Q136Q (1), K132R (1), L130F (1), P177R (1), R175_C176delinsG (1), R213fs (1), R267W (1), R196Q (1), R282W (1), S127F (1), T125T (1), T253P (1), V157F (1), V203L (1), W146* (1), W91* (1), Y163C (1), Y205* (1)
BRAF	V600E (30), L441V (1), G469V (1), D576E (1)
PIK3CA	E542K (5), H1047R (4), E525K (1), E707K (1), L455L (1), H1047L (1), H160R (1), T1025S (1)
EGFR	I1050V (1), R776H (1), P518P (1), G322S (1), Y727C (1), Exon 20 Insertion (1), A419A (1), Exon 19 Deletion (1), A653T (1), D587D (1), D321E (1), I1186T (1), H590Q (1), T678T (1)
RAS	NRAS: Q61R (4), Q61K (3), S106L (1), K5T (1), T158T (1)
	KRAS: K117R (1), G12S (1)
	HRAS: Q61K (2)
NF1	F2014F (1), R1276Q (1), I1983M (1), E725* (1), F1413fs (1), K2273M (1), N1885H (1), I679fs (1), R1970fs (1), R2637* (1), R1362* (1), K874fs (1), P1084R (1), Splice Site SNV (1)
APC	A2219V (1), P1853P (1), K2357fs (1), S32R (1), P2271del (1), Q789* (1)
GNAS	R201C (3), R201H (2), Q227E (1)
ATM	G2695V (1), R3008C (1), G2695S (1), R337C (1), 3057C (1)
BRCA1	K1489I (1), R1028H (1), V1842G (1), Q94*(1), S768S (1)
BRCA2	I2209T (1), N1201K (1), E2959fs (1), P2734S (1), F2011F (1), K21K (1)
CCND2	Amplification (1), P281R (1), T280N (1), L98L (1), T280A (1)
ERBB2	C584G (1), R456C (1), N124S (1), R499W (1), R143Q (1)
TERT	Promoter (5)
CDKN2A	D74N (1), P81fs (1), V51D (1), V28G (1)
FGFR1	R718C (1), Amplification (1), Y236Y (1), M535I (1)
FGFR2	D304N (2), L10L (1), P708S (1)
KIT	Amplification (3), N828S (1)
MET	T230M (1), R21K (1), Exon 14 Skipping (1), D1099V (1)
NOTCH1	F436F (1), I2109I (1), A2374T (1), A2257T (1), N454S (1), V2119M (1)
NTRK1	H571Y (1), S394T (1), T470T (1), S304F (1)
AR	A441A (1), A403V (1), T231T (1)
CCNE1	P106H (1), S308L (1), Amplification (1)
mTOR	V169I (1), P555R (1), D1005G (1)
PDGFRA	Amplification (2), F490F (1)
RB1	K122E (1), E34A (1), T682fs (1)
RET	R693H (1), H594P (1), K662E (1)

The number of patients harboring a specific mutation is in parentheses. Only genes that were reported in three or more cases are listed. A comprehensive list of mutations is presented in Supplementary Table S3.

Concordance between cfDNA and tumor tissue detected gene mutations

We examined the concordance between mutations detected in cfDNA and in matched tumor tissue samples. We analyzed 28 treatment-naive ATC patients separately to avoid confounding the results due to the possibility of treatment effect on the mutation profile. Our results showed that the concordance of cfDNA and tumor mutations of the most frequently mutated gene mutations (>3 cases of 28 patients), including TP53, BRAF^V600E, PIK3CA, and NRAS were 82.1% (23/28), 92.9% (26/28), 92.9% (26/28), and 92.9% (26/28), respectively (Table 3). In addition, both the positive predictive value (PPV) and specificity of cfDNA to detect BRAF^V600E and NRAS mutations are 100%. The sensitivity for BRAF^V600E mutation was 88.2%, and negative predictive value (NPV) was 84.6%. For PIK3CA mutation detection, both the specificity and the NPV was 95.7%, as shown in Supplementary Table S5. For previously treated patients the concordance rates for TP53, BRAF^V600E, and PIK3CA were 59.2%, 83.7%, and 87.8%, respectively. The median turn-around time of cfDNA test was 11 days (range, 5–16 days), compared with 21 days (range, 11–63 days) for tumor tissue NGS testing.

Table 3.

Concordance of Cell-Free DNA and Tissue Next-Generation Sequencing Results in Treatment-Naive Patients

Same mutations found in both cfDNA and tissue samples are highlighted in gray.

cfDNA, cell free DNA; Pl, plasma; Tu, tumor.

Prognostic significance of BRAFi treatment

Forty-one out of 87 (47.1%) patients had BRAF^V600E mutations, including 30 cases from cfDNA testing and 11 cases from tumor tissue sample analysis. Thirty-two of these 41 patients (78%) received a BRAFi treatment (full dose of dabrafenib 150 mg po twice daily or vemurafenib 960 mg po twice daily with or without a MEK inhibitor) after diagnosis. Patients with BRAF^V600E mutated ATC who received a BRAFi had a median OS of 15.5 months compared with 3.4 months in those who did not receive a BRAFi. The hazard ratio was 0.3 (95% confidence interval [CI] 0.111–0.679) (p = 0.003) (Fig. 2). There was no statistical difference in the number of mutations between the BRAFi-treated group (average of 3.0 mutations per case) and the no-BRAFi group (average of 3.0 mutations per case); p = 1.000. Out of 32 BRAFi treated patients, 20 patients (63%) also received immunotherapy. Although not statistically significant, the combination of BRAFi and immunotherapy trended toward better OS compared with BRAFi alone (p = 0.051) (Supplementary Fig. S1). There was no statistical difference in the number of mutations between the BRAFi group (average of 3.25 mutations per case) and the BRAFi + immunotherapy group (average of 2.85 mutations per case); p = 0.529.

FIG. 2.

BRAFi treatment efficacy in ATC patients with BRAF^V600E mutations. Out of 87 patients, 41 patients harbored the BRAF^V600E mutation. Survival of patients who were treated with BRAFi (n = 32) and patients who were not (n = 9) were compared. BRAFi, BRAF inhibitor.

Prognostic significance of cfDNA mutations

We evaluated the prognostic significance of the most common gene mutations detected from cfDNA testing, which included TP53, BRAF^V600E, PIK3CA, EGFR, NF1, and RAS. Only the PIK3CA mutation was associated with OS (p < 0.05). In the entire 87 ATC patient cohort, there were 13 patients with a PIK3CA mutation and an oncogenic score ≥0.80 detected with cfDNA testing (Supplementary Table S3). These patients with PIK3CA mutations had a worse OS compared with patients with PIK3CA wild-type (wt), despite most of them having received BRAFi therapy for BRAF^V600E mutated ATC. The median OS was 9.57 months (CI is [5.03, not available]) for the PIK3CA mutated group and 14.83 months (CI is [11.67, 24]) for the PIK3CA wt group. The hazard ratio was 2.077 for PIK3CA mutated versus PIK3CA wt, p = 0.03 (Fig. 3).

FIG. 3.

PIK3CA mutation is associated with worse overall survival in patients with ATC. Of 87 patients with ATC, 13 patients were positive for PIK3CA mutation. The survival of PIK3CA-mutated ATC patients (n = 13) and those with PIK3CA wild type (n = 74) were compared.

Next, we evaluated the PIK3CA mutation for its impact on treatment outcomes. Currently, the most commonly used treatment options for ATC patients are surgery, cytotoxic chemotherapy, radiation therapy (RT), and BRAFi for BRAF^V600E mutated disease. We evaluated the prognostic value of PIK3CA mutation within each treatment modality. For all these four ATC treatment modalities, PIK3CA mutation was associated with worse OS than PIK3CA wt (p < 0.05) (Fig. 4). In the group of surgical patients, the median OS was 9.2 versus 22.7 months (hazard ratio 3.4, CI 1.435–7.839, p = 0.003) for PIK3CA mutated versus PIK3CA wt. For the cytotoxic chemotherapy group, the median OS was 9.2 versus 18.9 months (hazard ratio 3.2, CI 1.479–7.118, p = 0.002) for the PIK3CA mutated versus PIK3CA wt patients. For RT, the median OS was 9.2 versus 18.9 months (hazard ratio 2.8, CI of ratio 1.192–6.810, p = 0.014) for the PIK3CA mutated versus PIK3CA wt patients. In the BRAFi group, the median OS was 9.2 versus 21.7 months (hazard ratio 5.7, CI of ratio 2.075–15.889, p = 0.0002) for the PIK3CA mutated versus the PIK3CA wt patients. The BRAFi group included only patients with BRAF^V600E mutated tumors. The corresponding Benjamini–Hochberg adjusted p-values for the earlier four groups are: 0.004 for the surgery group, 0.004 for the cytotoxic chemotherapy group, 0.014 for the RT group, and 0.0008 for the BRAFi therapy group.

FIG. 4.

Association of PIK3CA mutation and treatment outcomes. For each ATC therapeutic modalities including surgery (A), cytotoxic chemotherapy (B), radiation therapy (C), and BRAFi therapy (D), the survival between PIK3CA mutated ATCs and PIK3CA wild-type ATCs was compared.

Discussion

To our knowledge, this is the largest cfDNA study among ATC patients and the first to evaluate ATC prognosis by cfDNA mutation testing. cfDNA is made of double-stranded DNA fragments that are found in the noncellular component of the blood, and they are believed to be released into the bloodstream through cell apoptosis or necrosis. To analyze tumor-derived DNA from blood samples, instead of an invasive tumor biopsy, represents a critical advance in clinical practice (3). Molecular profiling using cfDNA enables early cancer detection and rapid treatment decisions (3). It has also been used to track therapeutic response and monitor the disease (3). Since ATC is a very aggressive cancer that requires rapid intervention, the value of using cfDNA in ATC patients becomes critical in clinical practice because of the rapid turnaround time.

The current study demonstrated that the most frequent mutations are TP53, BRAF, and PIK3CA, which is consistent with previous reports. Landa et al. investigated 33 patients with ATC and found that the most common mutations detected by NGS of tissue samples were in the TERT promoter (72.7%), as well as in the TP53 (69.7%), BRAF (45.5%), PIK3CA (18.2%), and NRAS (18.2%) coding sequences (10). Pozdeyev et al. reported tissue NGS results based on 196 ATC specimens showing a TP53 mutation frequency of 65%, a TERT promoter mutation frequency of 65%, a BRAF mutation frequency of 41%, and a PIK3CA mutation frequency of 14% (11). Duan et al. studied the mutational profile of 25 ATC patients by tissue NGS and found that the most common mutations were in the TP53 (60%), BRAF (56%), TERT promoter (56%), and PIK3CA (44%) genes (12). Rao et al. investigated tissue samples of 54 patients with ATC and showed that the most common mutations tissues were in the BRAF^V600E (48%), TP53 (34.7%), PIK3CA (27.7%), and RAS (21.3%) genes (13). In 2017, our group described for the first time the use of cfDNA testing for genotyping in patients with ATC, and they found that the most common mutated genes were TP53 (65%) and BRAF^V600E (48%) (4). We acknowledged the low occurrence of TERT promoter mutation detected in cfDNA (5.8%); however, the reason for this was unclear. One explanation could be that the TERT promoter mutations might not be captured well by cfDNA NGS.

Based on the current study as well as previous reports, cfDNA testing had a high concordance with tissue sample NGS of ATCs. This study showed a concordance rate of above 90% for BRAF, PIK3CA, and NRAS and a concordance of 82% for TP53. Sandulache et al. compared 23 ATC cfDNA results with tissue NGS results and found a high concordance rate for BRAF, PIK3CA, and NRAS, and a moderate concordance rate for TP53. The concordance was highest in patients who underwent sequencing before initiation of definitive treatment (4). In addition, using cfDNA for molecular testing showed very high sensitivity, specificity, PPV, and NPV for the BRAF^V600E mutation. The current study showed a sensitivity of 88% and specificity and PPV of 100%, respectively. Iyer et al. (14) reported a concordance of 93% for BRAF^V600E between droplet digital polymerase chain reaction and tissue testing with a sensitivity of 85% and specificity of 100%. The reasons for these discordances are unknown. Several studies investigated the correlation between levels of cfDNA and tumor burden in solid tumors. In pancreatic cancers, cfDNA levels generally correlate with tumor burden (15). In a breast cancer study, cfDNA levels correlated with tumor burden only post-treatment and at progression (16). In non-small cell lung cancer, results of these correlation studies disagree (17,18). One large series showed that about 15% of patients with metastatic cancer may not shed sufficient cfDNA for plasma mutational testing (3). Therefore, cell-free DNA levels might reflect mechanisms of tumor biology more complex than simple cell lysis. Also, if tumor and plasma samples were not collected at the same time, that is, before and after treatment, molecular evolution of the tumor could lead to discordance (3). We also acknowledge that the amount and quality of the cfDNA analyzed are critical parameters, which could not be evaluated in the current study as the studies were done by using a commercial test. Zviran et al. recently demonstrated that genome-wide mutational integration might be necessary to overcome the fundamental limitation of low input cfDNA (19).

ATC management requires a multidisciplinary approach. Besides conventional therapies such as surgery, cytotoxic chemotherapy, and RT, targeted therapy has become increasingly important in recent years. BRAFi therapy proved to be effective in BRAF^V600E mutated ATC patients based on previous clinical trials (2), and the FDA has approved the dabrafenib/trametinib (BRAF/MEK inhibitor) combination for BRAF^V600E mutant ATC (20). Although limited by its retrospective nature, the present report shows that BRAFi improves the OS of ATC patients whose tumors harbor a BRAF^V600E mutation when compared with those not treated with BRAFi. In addition, the combination of BRAFi and immunotherapy may provide additional survival benefit for BRAF^V600E ATC patients when compared with BRAFi alone, but larger patient series will need to be studied to determine the added benefit of immunotherapy.

We also investigated individual gene mutations and their association with ATC prognosis, and we found that the presence of a PIK3CA mutation inversely correlated with patients' OS, and this association remained consistent across the four different treatment types. Most driver (initiating event) mutations in thyroid cancer occur in genes involved in the MAPK pathway (21). The PIK3CA gene encodes a 110 KD catalytic subunit of phosphatidylinositol 3-kinase (PI3K). Mutations in this gene activate the PI3K pathway, resulting in cell growth and proliferation (22). As one of the most recurrently mutated genes in different human cancers, PIK3CA mutations have been previously reported in ATC (23).

To explore the optimal treatment strategy for PIK3CA mutated ATC, we also found that traditional ATC treatments such as cytotoxic chemotherapy, RT, and surgery are less effective in PIK3CA mutated ATC. BRAFi therapy appears to benefit all patients with BRAF^V600E mutation regardless of the PIK3CA mutation status. These results suggest that targeting both the MAPK and PI3K pathways, such as with an mTOR inhibitor, might be of interest in patients with both BRAF^V600E and PIK3CA mutated ATC, warranting investigation in future prospective trials (24–26). Lastly, the stratification of PIK3CA as a high-risk prognostic factor may provide insight into the development of new targeted therapies for patients with PIK3CA mutated ATC.

With a total of 87 ATC patients, this study is so far the largest prognostic analysis of ATC using cfDNA profiling. However, the number of patients is still too low to fully investigate the significance of low prevalence mutations when using cfDNA testing. In addition, this study is a single-center study that investigated only 73 genes from the Guardant360 platform. Therefore, in addition to the need for a larger study cohort, an expanded analysis of genes by cfDNA may provide a more precise molecular evaluation of ATCs.

In summary, cfDNA profiling showed high concordance with profiling of tissue biopsies. Further, the data obtained from cfDNA profiling may be used to guide treatment selection and evaluate ATC patients' prognosis.

Authors' Contributions

Y.Q.: designed and performed the study, acquired and analyzed patient and NGS data, drafted and revised the article, and approved the final version; J.R.W.: analyzed patient and NGS data, and revised the article; Y.W.: performed biostatistical analysis; P.I.: acquired and analyzed patient and NGS data, and revised the article; G.J.C.: organized and analyzed data; N.L.B.: revised the article; R.D.: revised the article; M.Z.: revised the article; M.D.W.: performed pathological sample examinations; R.F.: revised the article; G.B.G.: revised the article; P.W.: provided biostatistical analysis; K.P.: acquired NGS data; M.-C.H.: analyzed NGS data, organized data, revised the article, and approved final version; and M.E.C.: designed the study, analyzed patient and NGS data, revised the article, and approved final version.

Author Disclosure Statement

No competing financial interests exist.

Funding Information

This work was funded by an MD Anderson Cancer Center Core Facilities Support Grant NIH CA16672 (M.D.W., P.W., K.P.), an ATA fellowship to Y.Q., and a SPORE bridge funding from MD Anderson Cancer Center (M.C.H., G.C., and R.D.).

Supplementary Material

Supplementary Figure S1

Supplementary Table S1

Supplementary Table S2

Supplementary Table S3

Supplementary Table S4

Supplementary Table S5