Abstract
A single arm, phase II trial of carboplatin, nab-paclitaxel, and pembrolizumab (CNP) in metastatic triple-negative breast cancer (mTNBC) was designed to evaluate overall response rate (ORR), progression-free survival (PFS), duration of response (DOR), safety/tolerability, overall survival (OS), and identify pathologic and transcriptomic correlates of response to therapy.
Patients with ≤2 prior therapies for metastatic disease were treated with CNP regardless of tumor programmed cell death-ligand 1 status. Core tissue biopsies were obtained prior to treatment initiation. ORR was assessed using a binomial distribution. Survival was analyzed via the Kaplan–Meier method. Bulk RNA sequencing was employed for correlative studies.
Thirty patients were enrolled. The ORR was 48.0%: 2 (7%) complete responses (CR), 11 (41%) partial responses (PR), and 8 (30%) stable disease (SD). The median DOR for patients with CR or PR was 6.4 months [95% confidence interval (CI), 4–8.5 months]. For patients with CR, DOR was >24 months. Overall median PFS and OS were 5.8 (95% CI, 4.7–8.5 months) and 13.4 months (8.9–17.3 months), respectively. We identified unique transcriptomic landscapes associated with each RECIST category of radiographic treatment response. In CR and durable PR, IGHG1 expression was enriched. IGHG1high tumors were associated with improved OS (P = 0.045) and were concurrently enriched with B cells and follicular helper T cells, indicating IGHG1 as a promising marker for lymphocytic infiltration and robust response to chemo-immunotherapy.
Pretreatment tissue sampling in mTNBC treated with CNP reveals transcriptomic signatures that may predict radiographic responses to chemo-immunotherapy.
Improved understanding of mechanisms underlying robust and durable responses to anti-programmed cell death protein 1 chemo-immunotherapy in metastatic triple-negative breast cancer (mTNBC) is desperately needed to better inform clinical decision-making and strategies for overcoming therapeutic nonresponse. Here, we present a single arm, phase II trial involving treatment of patients with mTNBC with carboplatin, nab-paclitaxel, and pembrolizumab to assess the overall response rate (ORR), overall survival (OS), and progression-free survival (PFS), and toxicity profile associated with this particular combination of chemo-immunotherapeutic agents. Further, we expand on clinical findings by using pretreatment tumor biopsies to identify pathologic and genomic correlates to treatment response.
An ORR of 48.0%, median PFS and OS of 5.8 and 13.4 months align with outcomes previously described in mTNBC treated with pembrolizumab and chemotherapy. RNA sequencing of pretreatment tumor tissue revealed transcriptional landscapes that varied by RECIST category of radiographic response. Complete response was enriched in pathways involving adaptive immune system priming and activation. These pathways were less enriched in partial response (PR). Stable disease showed increased enrichment of adaptive immune activating pathways compared with PR but were, conversely, also enriched in pathways of cellular senescence, defective apoptosis, and defective programmed cell death. PR showed significant heterogeneity in PFS and transcriptional landscape. Tumors from complete responders and partial responders with durable PFS showed enrichment in IGHG1, which was associated with increased tumoral B cell and follicular helper T cell presence and OS. Collectively, our findings provide transcriptional correlates to disparities in chemo-immunotherapy responses as well as potential response markers for future investigation.
Introduction
Triple-negative breast cancer (TNBC) represents approximately 15% to 20% of all breast cancer diagnoses and remains a major therapeutic challenge (1). Despite some progress with targeted biologics such as PARP inhibitors (olaparib; ref. 2) and antibody drug conjugates (sacituzumab govitecan; 3), cytotoxic chemotherapy remains the predominant treatment paradigm for metastatic TNBC (mTNBC; 1). However, tumors become quickly resistant to chemotherapy. Consequently, median progression-free (PFS) and overall survival (OS) in patients with mTNBC who receive standard chemotherapy are only 5 and 15 months, respectively (4, 5).
The phase III KEYNOTE-355 trial evaluated the addition of pembrolizumab, an anti-programmed cell death protein-1 (PD-1) monoclonal antibody, to standard cytotoxic chemotherapy in the first-line setting for patients with mTNBC. Compared with chemotherapy alone, pembrolizumab plus chemotherapy resulted in significantly longer median PFS [9.7 vs. 5.6 months, HR 0.65, 95% confidence interval (CI), 0.49–0.86; P = 0.0012] and median OS (23 vs. 16 months, 0.73; 95% CI, 0.55–0.95; P = 0.0185) among patients with programmed cell death-ligand (PD-L1)-positive tumors with combined positive score (CPS) of ≥10 (6). Consequently, pembrolizumab received accelerated approval in 2020 by the FDA, followed by full approval in 2021 based on confirmatory results from KEYNOTE-522, for use in combination with first-line chemotherapy in patients with locally recurrent or mTNBC whose tumors possessed a positive PD-L1 status with a CPS ≥10 (7).
Despite promising results, it is sobering that only 39% of trial patients with PD-L1+ tumors (CPS ≥ 10) who received pembrolizumab plus chemotherapy were progression-free at 12 months (6). Therefore, more research is urgently needed to better understand mechanisms of response and resistance to immune checkpoint inhibition in mTNBC. Given that combination chemo-immunotherapy regimens in advanced TNBC are superior to single agent checkpoint inhibition alone (8), one strategy to build on the positive results from KEYNOTE-355 was to investigate particular chemotherapeutic regimens in combination with pembrolizumab that may be more synergistic and/or less immunosuppressive (8, 9).
We hypothesized that weekly nab-paclitaxel and carboplatin may be a superior chemotherapy regimen to combine with pembrolizumab compared with solvent based paclitaxel or gemcitabine/carboplatin, the two regimens used in KEYNOTE-355, given the enhanced antitumor activating properties of nab-paclitaxel and lack of requirement for premedication with corticosteroids (10). Therefore, we conducted a prospective, single-arm study in 30 patients with mTNBC who received carboplatin, nab-paclitaxel and pembrolizumab (CNP). Given that not all patients with mTNBC respond to immune checkpoint blockade and the wide variance in degree and duration of response (DOR) in those who do respond, we also performed pretreatment biopsies of metastatic sites in all participants. From these biopsies, correlative PD-L1 testing, bulk RNA sequencing (RNA-seq), and transcriptomic analysis were performed to investigate mechanisms of therapeutic response and resistance.
Patients and Methods
Patients and treatment
Patients with mTNBC who had received no more than two prior lines of therapy in the metastatic setting were eligible for this clinical trial (NCT03121352) in which J. Baar served as the overall principle investigator. Primary eligibility criteria included: age ≥18 years, histologically or cytologically confirmed mTNBC, Eastern Cooperative Oncology Group (ECOG) performance status of 0–1, and measurable disease as per RECIST v1.1 (11, 12). Patients who had previously received anti–PD-1/PD-L1 therapy were excluded. As this trial commenced prior to the initial FDA approval of pembrolizumab for mTNBC in 2020, tumor PD-L1 expression was not required for enrollment. A summary of representativeness of study participants is listed in Supplementary Table S1.
Patients meeting eligibility initially received three cycles of CNP. Each cycle initially comprised of C (AUC 6 on day 1 of a 21-day cycle), N (100 mg/m2 i.v. on days 1, 8 and 15 of a 21-day cycle), and P (200 mg i.v. on day 15 of each cycle). However, due to significant bone marrow toxicity with the first two enrolled patients, dosing of C was reduced to AUC 4.4 and N was reduced to 75 mg/m2. Chemo-immunotherapy was continued either until a diagnosis of progressive disease (PD) or intolerable adverse events (AE). Concomitant use of growth factor (usually G-CSF or GM-CSF) support was permitted on the basis of medical necessity as determined by the treating physician, which was independent of decisions regarding dose reduction. If it was considered medically necessary to discontinue one study drug, the others could be continued. Particularly, the protocol allowed for continuance of single agent pembrolizumab alone if significant toxicity that was attributable to carboplatin and/or nab-paclitaxel arose and resulted in treatment being held for 4 or more weeks.
Trial design and oversight
This study was designed as a prospective phase II clinical trial of CNP in women with mTNBC to evaluate the overall response rate (ORR), PFS, disease control rate (DCR), DOR, safety/tolerability, and to identify pathologic and transcriptomic correlates of response to CNP. With an expected ORR of 35% and precision of the two-sided 95% CI for the ORR set to 0.17 (the distance to the expected ORR of 35%), the sample size required for the study was determined to be 30 patients (nQuery Advisor).
As part of initial screening, patients underwent baseline bloodwork, biopsy, and radiological staging to determine eligibility for study enrollment. After completing three cycles of CNP, patients were radiologically restaged by CT chest, abdomen, pelvis, and bone scan before starting the next three cycles of treatment and clinical responses were scored according to standard RECIST v1.1 (12, 13). Patients with stable or regressing disease were eligible for three additional cycles of CNP, whereas patients with PD were taken off the study.
The clinical trial was sponsored by Merck Sharp & Dohme LLC and conducted in accordance with Good Clinical Practice and the Declaration of Helsinki. It was approved by the University Hospitals/Seidman Cancer Center and Cleveland Clinic/Taussig Cancer Institute—members of the NCI-designated Case Comprehensive Cancer Center—institutional review boards. All patients provided written informed consent in accordance with the Declaration of Helsinki and the International Conference on Harmonization Guidelines for Good Clinical Practice. AEs were recorded regularly according to protocol and were subject to mandatory reporting by the investigators. AEs were graded according to the NCI Common Terminology Criteria for AEs, version 4.0 (14). Patients were followed for survival every 3 months for at least 12 months after their post study 30-day follow-up.
Statistical analysis
The true response rate of therapy was estimated on the basis of the number of responses using a binomial distribution and its CIs were estimated using Wilson's method. OS was measured from the date of biopsy to date of death and was censored at the date of last follow-up for survivors. PFS was measured from the date of biopsy to disease progression or the date of death and was censored at the date of last follow-up for those alive without disease progression. The DOR for those with complete response (CR) and partial response (PR) was measured from the date of response to the date of progression or death, whichever occurred earlier, and was censored at the date of last follow-up for those alive and without disease progression. Survivor distribution was estimated via the Kaplan–Meier method.
PD-L1 testing
Baseline PD-L1 expression on formalin-fixed tumor samples was assessed retrospectively (QualTek Molecular Laboratories, Newtown, PA) by means of the PD-L1 IHC 22C3 pharmDx assay (Agilent Technologies, Carpinteria, CA), and characterized by the CPS score, defined as the number of PD-L1–positive cells (tumor cells, lymphocytes, and macrophages) divided by total number of tumor cells × 100.
Bulk RNA-seq of pretreatment tumor biopsy tissue
RNA was isolated from frozen core biopsies of primary or mTNBC tissue (Supplementary Table S2) using the Qiagen AllPrep DNA/RNA kit. Libraries were prepared using the Takara SMART-Seq v4 Ultra Low Input RNA kit and sequenced on an Illumina NovaSeq for a target of 40 million reads per sample.
Raw FASTQ files were aligned to the hg38 genome using STAR aligner v2.7.6a with default parameters and mapped reads were counted using featureCounts. Count files were merged and normalized using DESeq2. DESeq2 was used to perform differential gene expression analysis.
Gene set enrichment analysis
DESeq2 identifies differentially expressed genes (DEG) between two groups by using gene expression count data to create a generalized linear model where significant DEGs are identified using the Wald test. For all pretreatment RNA samples, we assessed comparisons between all RECIST categories of radiographic response [CR, PR, stable disease (SD), PD]. The 100 DEGs with the highest log fold change values for each comparison were used to perform gene set enrichment analysis (GSEA) using the R package clusterProfiler. GSEA analysis was implemented on Hallmark and REACTOME pathway genesets downloaded from GSEA-MSigDB and reactome.org respectively. These results were visualized using the enrichplot R package. A heat map was generated using R package ComplexHeatmap Version2.13.1. The compareCluster function from enrichplot R package was used to calculate enriched profiles of top 500 upregulated genes from each comparison and aggregate these results. The P-adjusted value was calculated during this step using bonferroni correction. GeneRatio was calculated using the dotplot function and it was by default (count of core enrichment genes)/(count of pathway genes). The enricher function with the Hallmark pathways from msigdbr R library was used to perform enrichment analysis for each gene list. The functions dotplot and cnetplot from the enrichplot R package were used to visualize the pathway profiles and enrichment results of the multiple gene clusters.
Gene expression intersection
For hierarchical clustering, a Likelihood Ratio Test was run in DESeq2, and clustering was performed using the pheatmap R package for the top 200 most significant DEGs. The UpSetR R package was implemented on upregulated and downregulated genes from every comparison to visualize intersections of genes that were simultaneously upregulated or downregulated between comparisons.
DEGs associated with survival
The glmnet R package was used to fit a regularized Cox proportional hazards model to study the relationship between predictor variables and duration of PFS. The interquartile range (IQR) of expression for each gene across samples was calculated with IQR function from stats R package. The top quartile of these calculated IQRs was chosen to reduce the number of predictor variables going into the cox model. The cv.glmnet function with 10 fold cross validation was used to implement the Cox model where the count matrix of 7402 genes for 15 baseline patients was the covariates matrix. The Surv function from the survival R package was used to create a response matrix with PFS data. Genes with active beta coefficients were chosen for further analysis. Positive Cox regression coefficients indicated a negative association with PFS and negative coefficient indicated a protective effect of the variable with which it was associated. Normalized expression of the genes chosen from the Cox model were visualized via heat map for all pretreatment patient samples using the pheatmap R package. Survival curves were created using the surv_fit function and plotted using the ggsurvplot function from the survminer R package.
Immune deconvolution
To deconvolute tumoral cell types represented in each biopsy sample, single-sample GSEA (ssGSEA) was used to calculate enrichment scores for cell type–specific marker genes within each sample. The scores represent the degree to which genes of a gene set are coordinately up- or downregulated. The R package GSVA was used to generate enrichment scores.
Data availability
The data generated in this study are publicly available in Gene Expression Omnibus (GEO) at accession number GSE241876. The single cell sequencing data employed to validate the findings of this study are available at GEO and BioTuring at accession number GSE169246 and BioTuring ID ayse_et_al_2021_cohort1_all. Both sources of validation data were derived from the following published resources: https://doi.org/10.1016/j.ccell.2021.09.010 and https://doi.org/10.1038/s41591-021-01323-8.
Results
Patients
Between February 2017 and April 2019, 35 patients with histologically confirmed mTNBC were screened. Of these, 30 patients with a median age of 55 (range: 36–80) were enrolled into the clinical trial (Table 1). Among them, 25 had recurrent disease (83.3%) and five had de novo mTNBC (16.6%). Among the patients with recurrent disease, 22 (73.3%) had previously received anthracycline and taxane-based chemotherapy either in the neoadjuvant or adjuvant setting. Most patients (76.6%) had not received any prior chemotherapy in the metastatic setting; only seven (23.3%) patients had received one or two prior lines of systemic chemotherapy. In 16 patients (53.5%), there was insufficient tissue for PD-L1 assessment. In the 14 patients (46.6%) with sufficient tissue, seven (50%) had CPS scores of 0, four had CPS scores of 1–9, and 3 patients (21.4%) had tumors with CPS scores of ≥ 10 (Table 1).
Patient characteristics.
Characteristics . | All patients (n = 30) . |
---|---|
Age | |
Median (range), years | 55 (36–80) |
Race, n (%) | |
Asian | 2 (6.7) |
Black-African American | 4 (13.3) |
White | 24 (80) |
ECOG performance status, n (%) | |
0 | 15 (50) |
1 | 15 (50) |
Stage, n (%) | |
De novo stage 4 | 5 (16.6) |
Recurrent breast cancer | 25 (83.3) |
Prior neo/adjuvant chemotherapy, n (%) | 25 (83.3) |
Prior neo- or adjuvant anthracycline & taxane | 22 (73.3) |
Prior line(s) of therapy for metastatic cancer, n (%) | |
0 | 23 (76.6) |
1–2 | 7 (23.3) |
PD-L1 status | |
Insufficient tissue/unevaluable | 16 (53.3) |
PD-L1 status known | 14 (46.6) |
CPS = 0 | 7 (50) |
CPS = 1–9 | 4 (29) |
CPS ≥ 10 | 3 (21.4) |
Characteristics . | All patients (n = 30) . |
---|---|
Age | |
Median (range), years | 55 (36–80) |
Race, n (%) | |
Asian | 2 (6.7) |
Black-African American | 4 (13.3) |
White | 24 (80) |
ECOG performance status, n (%) | |
0 | 15 (50) |
1 | 15 (50) |
Stage, n (%) | |
De novo stage 4 | 5 (16.6) |
Recurrent breast cancer | 25 (83.3) |
Prior neo/adjuvant chemotherapy, n (%) | 25 (83.3) |
Prior neo- or adjuvant anthracycline & taxane | 22 (73.3) |
Prior line(s) of therapy for metastatic cancer, n (%) | |
0 | 23 (76.6) |
1–2 | 7 (23.3) |
PD-L1 status | |
Insufficient tissue/unevaluable | 16 (53.3) |
PD-L1 status known | 14 (46.6) |
CPS = 0 | 7 (50) |
CPS = 1–9 | 4 (29) |
CPS ≥ 10 | 3 (21.4) |
Efficacy
Of 30 participants, 27 were evaluable for treatment response. Two patients (7%) had CR, 11 (41%) had PR, 8 (30%) had SD, and 6 (22%) had PD (Fig. 1A). Accordingly, the ORR was 48%. The median DOR among the 13 patients with a CR or PR was 6.4 months (95% CI, 4–8.5 months). The median PFS for the entire 30-patient cohort was 5.8 months (95% CI, 4.7–8.5 months; Fig. 1B); the median OS was 13.4 months (95% CI, 8.9–17.3 months; Fig. 1C).
Survival in patients with mTNBC treated with CNP. A, Swimmer plot of OS by treatment response. The x-axis represents survival in months and y-axis represents individual patients. Bars are colored by RECIST v1.1 response with red representing nonresponders and blue representing responders. Crosses represent patients with PD-L1+ tumors (CPS ≥ 10). For responders (blue bars): green circles represent CRs, purple squares represent PRs, and yellow triangles represent SD. For nonresponders, orange diamonds represent PD. A black chevron at the end of a bar indicates patient with “alive” vital status at the end of the study period in December 2022. Designation of NE (e.g., NE-08) before patient number represents a patient whose tumor was not evaluated in correlative studies. Tumors from 3 patients, labeled as “unevaluable,” were evaluated for correlative studies but not included due to insufficient tumor tissue. B, Kaplan–Meier plot of OS for the overall cohort of 30 patients. The median OS was 13.4 months with 95% CI of 8.9 to 17.3 months. C, Kaplan–Meier plot of PFS for the overall cohort of 30 patients. The median PFS was 5.8 months with 95% CI of 4.7 to 8.5 months.
Survival in patients with mTNBC treated with CNP. A, Swimmer plot of OS by treatment response. The x-axis represents survival in months and y-axis represents individual patients. Bars are colored by RECIST v1.1 response with red representing nonresponders and blue representing responders. Crosses represent patients with PD-L1+ tumors (CPS ≥ 10). For responders (blue bars): green circles represent CRs, purple squares represent PRs, and yellow triangles represent SD. For nonresponders, orange diamonds represent PD. A black chevron at the end of a bar indicates patient with “alive” vital status at the end of the study period in December 2022. Designation of NE (e.g., NE-08) before patient number represents a patient whose tumor was not evaluated in correlative studies. Tumors from 3 patients, labeled as “unevaluable,” were evaluated for correlative studies but not included due to insufficient tumor tissue. B, Kaplan–Meier plot of OS for the overall cohort of 30 patients. The median OS was 13.4 months with 95% CI of 8.9 to 17.3 months. C, Kaplan–Meier plot of PFS for the overall cohort of 30 patients. The median PFS was 5.8 months with 95% CI of 4.7 to 8.5 months.
As of December 2022 (the trial's cut-off date for survival analysis), four patients (two with CR and two with PR) were still alive. Patient R022, with a right chest wall mass measuring 5.47 cm as well as pericardial and mediastinal adenopathy, achieved CR. The right pericardial and mediastinal lymph nodes completely regressed after three cycles of therapy; the chest wall mass completely regressed after six cycles of treatment. This patient was treated for 12.5 months, with several interruptions and eventual discontinuation due to grade 2 pneumonitis. Patient R020, with a 4.7 cm retrosternal mass that completely regressed after three cycles, also achieved a CR. This patient's treatment was discontinued after 4.5 months due to grade 3 neutropenia and grade 2 thrombocytopenia.
Of 30 enrolled patients, data on subsequent therapy post-progression was available for 19 patients. Most received eribulin as next line therapy. Of the four long-term responders (CR or durable PR), subsequent treatment information was available for three. Only one of these patients continued to receive chemotherapy. This patient (R017) was continued on weekly nab-paclitaxel for approximately 12 more months and pembrolizumab was discontinued due to autoimmune hepatitis. Subsequently, nab-paclitaxel was discontinued due to neuropathy and the patient remains without evidence of progression. The other two patients have remained on observation without any evidence of progression and have not received any systemic therapy since trial discontinuation.
Safety
Treatment-related AEs were observed in 24 patients (80%); most were classified as grade 1–2 (Table 2). The most common grade 1 AEs were nausea (70%), diarrhea (57%), fatigue (57%), anemia (50%), and anorexia (50%). Grade 3–4 AEs were primarily hematologic and included neutropenia (66%), leukopenia (43%), anemia (33%), lymphopenia (20%), and thrombocytopenia (7%). Non-hematologic grade 3–4 AEs, occurring in more than 10% of patients, included pain (13%), dyspnea (10%), and muscle weakness (10%). Treatment was discontinued in two patients due to definite immune-related pneumonitis (one grade 2 and one grade 3) and in 1 patient due to prolonged grade 3 neutropenia. There were no treatment-related deaths.
Adverse drug reactions in the safety data set.
Characteristics n (%) of patients with event . | Grade 1–2 . | Grade 3 . | Grade 4 . |
---|---|---|---|
Gastrointestinal Disorders | |||
Constipation | 14 (47) | 0 | 0 |
Diarrhea | 17 (57) | 1 (3) | 0 |
Mucositis | 4 (13) | 0 | 0 |
Nausea | 21 (70) | 0 | 0 |
Vomiting | 9 (30) | 1(3) | 0 |
General Disorders | |||
Allergic rhinitis | 4(13) | 0 | 0 |
Dysgeusia | 10 (33) | 0 | 0 |
Edema | 7 (23) | 0 | 0 |
Fatigue | 17 (57) | 2(7) | 0 |
Fever | 10 (33) | 0 | 0 |
Hot flushes | 5 (17) | 0 | 0 |
Hypertension | 7 (23) | 0 | 0 |
Blood and Lymphatic Disorders | |||
Anemia | 15 (50) | 10 (33) | 0 |
Epistaxis | 3 (10) | 0 | 0 |
Febrile neutropenia | 0 | 1 (3) | 0 |
Leukopenia | 9 (30) | 12 (40) | 1 (3) |
Lymphopenia | 5 (17) | 5(17) | 1 (3) |
Neutropenia | 1 (3) | 16 (53) | 4 (13) |
Thrombocytopenia | 13 (43) | 2 (7) | 0 |
Metabolism and abnormal laboratory values | |||
Acute kidney injury | 5 (17) | 1 (3) | 0 |
Alanine aminotransferase increased | 6 (20) | 1 (3) | 0 |
Alkaline phosphatase increased | 3 (10) | 0 | 0 |
Aspartate aminotransferase increased | 6 (20) | 0 | 0 |
Dehydration | 2 (7) | 1 (3) | 0 |
Hyperthyroidism | 2 (7) | 0 | 0 |
Hypoalbuminemia | 7 (23) | 0 | 0 |
Hypokalemia | 10 (33) | 0 | 0 |
Hypomagnesemia | 14 (47) | 0 | 0 |
Hyponatremia | 7 (23) | 0 | 0 |
Hypothyroidism | 7 (23) | 0 | 0 |
Musculoskeletal and connective-tissue disorders | |||
Arthralgias | 2 (7) | 0 | 0 |
Back pain | 3 (10) | 0 | 0 |
Bone pain | 1 (3) | 1 (3) | 0 |
Breast pain | 5 (17) | 1 (3) | 0 |
Muscle weakness | 6 (20) | 3 (10) | 0 |
Myalgias | 3 (10) | 0 | 0 |
Nervous system disorders | |||
Anorexia | 15 (50) | 0 | 0 |
Anxiety | 6 (20) | 0 | 0 |
Headache | 6 (20) | 0 | 0 |
Insomnia | 6 (20) | 0 | 0 |
Peripheral neuropathy | 11 (37) | 1 (3) | 0 |
Respiratory | |||
Cough | 11 (37) | 0 | 0 |
Dyspnea | 7 (23) | 3 (10) | 0 |
Pneumonitis | 1 (3) | 1 (3) | 0 |
Sinus tachycardia | 6 (20) | 0 | 0 |
Others | |||
Dizziness | 7 (23) | 0 | 0 |
Infusion reaction | 3 (10) | 0 | 0 |
Pain | 11 (37) | 4 (13) | 0 |
Characteristics n (%) of patients with event . | Grade 1–2 . | Grade 3 . | Grade 4 . |
---|---|---|---|
Gastrointestinal Disorders | |||
Constipation | 14 (47) | 0 | 0 |
Diarrhea | 17 (57) | 1 (3) | 0 |
Mucositis | 4 (13) | 0 | 0 |
Nausea | 21 (70) | 0 | 0 |
Vomiting | 9 (30) | 1(3) | 0 |
General Disorders | |||
Allergic rhinitis | 4(13) | 0 | 0 |
Dysgeusia | 10 (33) | 0 | 0 |
Edema | 7 (23) | 0 | 0 |
Fatigue | 17 (57) | 2(7) | 0 |
Fever | 10 (33) | 0 | 0 |
Hot flushes | 5 (17) | 0 | 0 |
Hypertension | 7 (23) | 0 | 0 |
Blood and Lymphatic Disorders | |||
Anemia | 15 (50) | 10 (33) | 0 |
Epistaxis | 3 (10) | 0 | 0 |
Febrile neutropenia | 0 | 1 (3) | 0 |
Leukopenia | 9 (30) | 12 (40) | 1 (3) |
Lymphopenia | 5 (17) | 5(17) | 1 (3) |
Neutropenia | 1 (3) | 16 (53) | 4 (13) |
Thrombocytopenia | 13 (43) | 2 (7) | 0 |
Metabolism and abnormal laboratory values | |||
Acute kidney injury | 5 (17) | 1 (3) | 0 |
Alanine aminotransferase increased | 6 (20) | 1 (3) | 0 |
Alkaline phosphatase increased | 3 (10) | 0 | 0 |
Aspartate aminotransferase increased | 6 (20) | 0 | 0 |
Dehydration | 2 (7) | 1 (3) | 0 |
Hyperthyroidism | 2 (7) | 0 | 0 |
Hypoalbuminemia | 7 (23) | 0 | 0 |
Hypokalemia | 10 (33) | 0 | 0 |
Hypomagnesemia | 14 (47) | 0 | 0 |
Hyponatremia | 7 (23) | 0 | 0 |
Hypothyroidism | 7 (23) | 0 | 0 |
Musculoskeletal and connective-tissue disorders | |||
Arthralgias | 2 (7) | 0 | 0 |
Back pain | 3 (10) | 0 | 0 |
Bone pain | 1 (3) | 1 (3) | 0 |
Breast pain | 5 (17) | 1 (3) | 0 |
Muscle weakness | 6 (20) | 3 (10) | 0 |
Myalgias | 3 (10) | 0 | 0 |
Nervous system disorders | |||
Anorexia | 15 (50) | 0 | 0 |
Anxiety | 6 (20) | 0 | 0 |
Headache | 6 (20) | 0 | 0 |
Insomnia | 6 (20) | 0 | 0 |
Peripheral neuropathy | 11 (37) | 1 (3) | 0 |
Respiratory | |||
Cough | 11 (37) | 0 | 0 |
Dyspnea | 7 (23) | 3 (10) | 0 |
Pneumonitis | 1 (3) | 1 (3) | 0 |
Sinus tachycardia | 6 (20) | 0 | 0 |
Others | |||
Dizziness | 7 (23) | 0 | 0 |
Infusion reaction | 3 (10) | 0 | 0 |
Pain | 11 (37) | 4 (13) | 0 |
Correlative studies
PD-L1 assessment
Two of the 3 patients with confirmed PD-L1+ tumors (CPS ≥ 10) were evaluable for treatment response. Both of these patients achieved CR: R020 (CPS of 70) and R022 (CPS of 10). Four patients had PD-L1 low tumors (CPS ≥ 1 and ≤ 9): R009, R013, R017, and R021.
Distinct tumor landscapes by radiographic response revealed by bulk transcriptomics
Given the variability of chemo-immunotherapy responses observed in mTNBC, we speculated that bulk RNA-seq of pretreatment biopsies may elucidate differences in intra-tumoral transcriptional landscapes that underlie varied treatment responses as well as unique markers that may predict response or resistance. We first performed GSEA, which revealed unique gene enrichment signatures by radiographic response category (RECIST 1.1). Differential gene expression between all four RECIST categories of response (CR, PR, SD, or PD) were compared using the best response of each pairwise comparison as group 1. From these results, we performed functional GSEA using enlisted genes ranked by log2-fold change, first using the REACTOME database of gene-sets. The largest of normalized enrichment scores in selected (Supplementary Table S3) REACTOME pathways as well as the top and bottom 25 pathways with differential enrichment seen across the 2,513 gene-sets (Supplementary Fig. S1) are shown. Selected REACTOME gene sets of interest were chosen to highlight any differences in baseline intra-tumoral immune cell activation and cell turnover between response groups.
GSEA for this study compares degrees of enrichment of particular gene sets between all pairwise comparisons of radiographic responses. We noted differential patterns of enrichment which differed between response categories (CR, PR, SD, and PD). For example, all pairwise gene set enrichment comparisons involving CR as a comparison group showed increased enrichment for antigen activates B cell receptor leading to generation of second messengers, PD-1 signaling, downstream TCR signaling, TCR signaling, MHC Class II antigen presentation, and Antigen processing-cross presentation (Fig. 2A) among patients with CR. These enrichment patterns were, therefore, shown to be characteristic for patients who achieved a CR. These pathways were more enriched in CR even relative to PR. In fact, the REACTOME pathway enrichment signatures of tumors from patients with CR were more similar to SD than PR for enrichment in immune-stimulatory pathways. Compared with SD, PR was associated with decreased PD-1 signaling, downstream TCR signaling, TCR signaling, extension of telomeres, downstream signaling events of B cell receptor, and CD28 co-stimulation. Concurrently, tumors from patients with SD showed greater enrichment for pathways involving cellular senescence, defective intrinsic pathway for apoptosis, and diseases of programmed cell death compared with all other groups. To summarize, GSEA of REACTOME pathways revealed key differences in pretreatment tumor transcriptional landscapes between each category of radiographic response, including CR and PR, with decreased enrichment of adaptive immune system-stimulating pathways observed in PR.
GSEA reveals distinct pathway enrichment signatures unique to RECIST categories of radiographic response. A, Differential expression was performed between each RECIST category of response, and from these DEG lists log2FC was used to rank genes as the input to GSEA analysis using the REACTOME gene sets. Heat map values are shown as Normalized Enrichment Score (NES) values from the output of each pairwise comparison. B, Differential expression outputs were similarly used to assess enrichment of Hallmark pathways by the top 500 DEGs for each comparison, ordered by log2FC; the size of each dot represents the gene ration (ratio of core/enriched genes over count of pathway genes). C, CNET plot demonstrating linkages of hallmark pathway genes differentially enriched according comparisons to degree of response to therapy. The size of circles demarcating pathways represent the number of DEGs belonging to that pathway.
GSEA reveals distinct pathway enrichment signatures unique to RECIST categories of radiographic response. A, Differential expression was performed between each RECIST category of response, and from these DEG lists log2FC was used to rank genes as the input to GSEA analysis using the REACTOME gene sets. Heat map values are shown as Normalized Enrichment Score (NES) values from the output of each pairwise comparison. B, Differential expression outputs were similarly used to assess enrichment of Hallmark pathways by the top 500 DEGs for each comparison, ordered by log2FC; the size of each dot represents the gene ration (ratio of core/enriched genes over count of pathway genes). C, CNET plot demonstrating linkages of hallmark pathway genes differentially enriched according comparisons to degree of response to therapy. The size of circles demarcating pathways represent the number of DEGs belonging to that pathway.
We next performed GSEA of Hallmark pathways as a well characterized set of cancer related genes (Supplementary Fig. S2), generating a dot plot to visualize differentially enriched pathways with significant adjusted P values (P < 0.05) for each pairwise comparison (Fig. 2B). This reinforced our finding of unique signatures specific to RECIST categories of chemo-immunotherapy response. Compared with tumors from patients with PR and PD, pretreatment tumor samples from patients who attained CR showed significantly greater enrichment for allograft rejection. In CR, there was also increased enrichment in interferon gamma response compared with PR, reinforcing comparative reduction of baseline intra-tumoral immune cell activation in PR. Pretreatment biopsies from patients who developed PR, in contrast, were enriched in xenobiotic metabolism and late estrogen response compared with PD, and enrichment in bile acid and fatty acid metabolism compared with SD. A CNET plot, a visualization tool for functional enrichment, of Hallmark pathways (Fig. 2C) demonstrated clustering of comparisons involving CR as best response among genes associated with allograft rejection. The CNET plot highlighted that genes particularly enriched in CR compared with PD included IFNG, CD3D, CCR2, CCR5, and CXCL9, while those enriched in CR compared with PR included GZMA, IL2RB, PRF1, major histocompatibility genes HLA-DRB1 and HLA-B, and CD38. It similarly highlighted distinction and separate clustering of genes enriched in CR and PR, with comparisons involving PR as best response clustering among coagulation (e.g., TF, MMP8, ANG), xenobiotic (ESR1, ALDH2) and fatty acid (ALDH1A1, and ADH7) metabolism-associated genes. Meanwhile, overlap was observed between CR and SD among genes differentially enriched compared with PD and/or PR, particularly those associated with allograft rejection, such as IL2RB, CD8B, and CD79A. There was also overlap between genes enriched in CR versus SD and SD versus PD among the Hallmark pathway of genes downregulated by KRAS activation (KRAS Signaling DN). Finally, the CNET plot also revealed genes distinctly enriched in SD versus PD in inflammatory response (e.g., IL1B, IL18RAP, CSF3R, and CXCL18) and epithelial mesenchymal transition (e.g., WNT5A, PTHLH, CRLF1, and LOXL2).
We next used DEG analysis to investigate intersecting gene expression and genes associated with PFS using linear regression. Upset plots were used to show intersections of upregulated genes between all pairwise response comparisons (Supplementary Fig. S3). This analysis revealed that three genes were upregulated in four of the six response comparisons, namely those involving CR or PR as best response compared with SD and PD: cathelicidin antimicrobial peptide (CAMP), LINC020616, and MAGEA12 (Supplementary Fig. S3A). Conversely, five genes were downregulated in the same four pairwise comparisons: CRHR1, GABRA5, KRT5, OTOP1, and TFAP2D (Supplementary Fig. S3B). As such, we found multiple intersections of up- and downregulated genes in pretreatment tumor biopsies from patients who eventually experienced tumor regression, whether CR or PR.
In seeking to further elucidate genes driving therapeutic response, we employed a Cox hazard regression model to help identify genes associated with PFS (Fig. 3A). While these methods are difficult in low numbered studies, we identified a significant association in IGHG1, while other identified genes expectedly turned out to be sample specific. IGHG1, Immunoglobulin Heavy Constant Gamma 1, was differentially enrichened in five patients, with the greatest enrichment seen in both complete responders (R020 and R022), followed by 3 patients with PR, two of which remained progression-free by last data analysis (R017, R012). While the median normalized expression count for IGHG1 in the overall cohort was 46,885.84, the threshold expression count differentiating the top tercile (comprised of patients R020, R022, R017, R012, and R008) of expression was 364,131.52. Survival analysis revealed statistically significant differences in OS between patients with IGHG1 expression in the highest tercile (IGHG1_exp = high) compared with those in the lowest terciles (IGHG1_exp = low; P = 0.045; Fig. 3B). Supplementary Figure S4 confirms statistically significant differences in IGHG1 expression between these groups on Wilcoxon test, P = 0.00067. We also observed a strong trend toward increased PFS in the IGHG1_exp = high group, though statistical significance was not reached (Fig. 3C).
Genes associated with survival in patients administered CNP. A, Heat map of genes derived from a regularized cox model showing gene expression associated with PFS per patient. Univariate overall (B) and progression-free (C) survival analysis according to IGHG1 expression. IGHG1_exp = high reflects patients with IGHG1 expression highest tercile; IGHG1_exp = low reflects patients with expression in the lowest terciles.
Genes associated with survival in patients administered CNP. A, Heat map of genes derived from a regularized cox model showing gene expression associated with PFS per patient. Univariate overall (B) and progression-free (C) survival analysis according to IGHG1 expression. IGHG1_exp = high reflects patients with IGHG1 expression highest tercile; IGHG1_exp = low reflects patients with expression in the lowest terciles.
Given observed enrichment of immune-activating pathways and IGHG1 expression in patients with durable response, we hypothesized that pretreatment tumors from patients who developed long-term response may have contained greater abundances of effector lymphocytes, especially B cells. To genomically test this hypothesis, we performed immune cell deconvolution (Fig. 4A) to estimate immune cell abundance in each sample. Deconvolution of DEGs showed that the same five patients in the top/highest tertile for IGHG1 expression also had concurrent enrichment in tumor infiltration of B cells and follicular helper T (Tfh) cells, which was not observed in tumor biopsies from any other patients. Accordingly, IGHG1 expression showed concordance with coinciding intra-tumoral B and Tfh presence. A heat map of the top 50 DEGs between the IGHG1high and IGHG1low cohorts (Fig. 4B) showed that IGHG1high tumors also showed enrichment for many other immunoglobulin-related genes, such as immunoglobulin heavy variable genes and immunoglobulin lambda loci.
Differences in immune cell deconvolution, DEG expression, and chemo-immunotherapy response from validating studies by IGHG1 expression. A, Immune deconvolution analysis was performed using GSEA analysis estimating immune subtypes and immunotherapy related signatures. These estimates are scaled by cell type and plotted as a heat map to represent the distribution of scores across the cohort. Red boxes indicate patients with differential expression of IGHG1. B, Differential gene expression between patients with IGHG1 expression greater than the median (IGHG1high) versus patients with IGHG1 expression below median expression (IGHG1low) was performed and the top and bottom 50-fold change genes are plotted as a scaled heat map. Color legend at the top of the heat map indicates sample ID by response (multi-color) and category of IGHG1 expression (pink = IGHG1high and red = IGHG1low).
Differences in immune cell deconvolution, DEG expression, and chemo-immunotherapy response from validating studies by IGHG1 expression. A, Immune deconvolution analysis was performed using GSEA analysis estimating immune subtypes and immunotherapy related signatures. These estimates are scaled by cell type and plotted as a heat map to represent the distribution of scores across the cohort. Red boxes indicate patients with differential expression of IGHG1. B, Differential gene expression between patients with IGHG1 expression greater than the median (IGHG1high) versus patients with IGHG1 expression below median expression (IGHG1low) was performed and the top and bottom 50-fold change genes are plotted as a scaled heat map. Color legend at the top of the heat map indicates sample ID by response (multi-color) and category of IGHG1 expression (pink = IGHG1high and red = IGHG1low).
Discussion
mTNBC is the most aggressive breast cancer subtype, amenable to few targeted therapeutic options, and is accordingly associated with significantly shorter OS. Checkpoint inhibitors, specifically anti–PD-1 monoclonal antibodies, have shown promise in treating mTNBC. Pembrolizumab in combination with first-line chemotherapy is considered the standard of care for PD-L1+ mTNBC. However, most patients do not respond to immune checkpoint blockade, leaving an unresolved question of what underlying mechanisms and associated markers determine treatment response in this population.
Several studies have attempted to identify predictive biomarkers for chemo-immunotherapy response in mTNBC, but results have often been inconsistent. Tumor mutational burden (TMB) has shown promise as a predictive biomarker for benefit of checkpoint inhibition in mTNBC. However, high TMB is uncommon in breast cancer and primarily observed in lobular breast cancer (15, 16). Furthermore, variability in methodology (from whole-genome sequencing to targeted gene panels), lack of consensus on how many or which genes to include in assessment, as well as a clear definition of a “TMB–high” threshold which varies widely by tumor type, limits TMB as a predictive immune biomarker in clinical practice (17). Others have described markers of the immune cell microenvironment to be predictive of immunotherapy response in mTNBC, such as TILs, CD4 and CD8 expression, TCR diversity, and MHC Class II expression (18–24). However, these findings are still in the early stages of investigation. In this phase II clinical trial, the two patients who achieved CR had pretreatment tumor biopsies that were PD-L1 positive by IHC staining and also had high TIL scores. The positive relationship between increased presence of TILs and tumor PD-L1 expression is well-described (25–28). Still, our cohort demonstrates that not all patients with increased TILs have PD-L1+ tumors, and although PD-L1+ tumors clearly demonstrate enhanced response to checkpoint blockade, not all patients with PD-L1+ tumors have durable responses to therapy. An absence of robust biomarkers for response to anti–PD-1 immunotherapy in mTNBC remains a significant clinical challenge.
To address this challenge, we performed correlative bulk RNA-seq on tumor biopsy tissues obtained prior to CNP administration. We found distinct transcriptional landscapes between radiographic response categories. For example, GSEA showed distinct differences in enrichment between CR and PR among pathways involved in activation of the adaptive immune system. Tumors from patients who went on to achieve CR showed enrichment for REACTOME pathways involving antigen activation of B cell receptor, PD-1 signaling, and TCR signaling, MHC Class II antigen presentation, and antigen processing-cross presentation. Compared with CR, tumors from patients who achieved PR displayed decreased enrichment in genes of immune-activating pathways. Genes in these pathways were even downregulated in PR when compared with tumors from patients who maintained SD. These observed differences in the transcriptional immune landscapes between CR, PR, and SD likely underlie differences in degree and durability of response as well as differences in survival between groups. Our findings appear to indicate that distinct tumor landscapes established before initiation of chemo-immunotherapy likely drive variances in radiographic response and, outside the setting of a CR, the degree of radiographic response neither fully recapitulates the DOR nor OS.
Despite divergence in characteristic GSEA signatures between CR and PR, there were examples of transcriptional intersections between groups. Three genes were found to be upregulated in both CR and PR compared with SD and PD: CAMP, LINC020616, and MAGEA12. The CAMP exclusively seen in humans, human cationic antibacterial protein (also known as hCAP18, LL-37, and FALL39), is released from monocytes, macrophages, and some epithelial cells and has been described as a chemoattractant for neutrophils as well as T cells. The chemoattractive abilities of LL-37 may increase immune cell infiltration of tumors, which is known to be essential criteria for immunotherapy response (29–32). Though prior studies investigating long noncoding RNA (lncRNA) LINC020616 are currently lacking, there have been studies showing diagnostic, prognostic, and potential immunotherapeutic target value of other lncRNAs, such as LINC02126 in lung cancer, which was correlated with greater immune cell infiltration of tumor microenvironments (32). MAGEA12 was found to be particularly upregulated in aggressive breast cancer subtypes, including some TNBCs, and was proposed as a potential breast cancer subtype biomarker (33).
Conversely, we found the following five genes to be similarly downregulated in responders (CR/PR): CRHR1, GABRA5, TFAP2D, OTOP1, and KRT5. Others have reported that CRH binding to CRHR1 in ovarian cancer cells promotes tumor privilege and serves as a potential therapeutic target (34). CRHR1 immunoreactivity was also described as a poor prognostic factor in pancreatic cancer (35). Similarly, GABRA5 and TFAP2D have been described as contributing either to tumorigenesis, aggressive tumor growth, or metastasis through p53 degradation and transcription of oncogenic genes, respectively (36–38). Interestingly, octopetrin 1, OTOP1, modulates purigenic control of intracellular calcium accumulation (39), which many have described propagates tumorigenesis, angiogenesis, epithelial-to-mesenchymal transition, (40, 41) and cancer therapy resistance (42). In fact, others have offered calcium signaling as a potential therapeutic target in cancer therapy (43, 44). KRT5 is a marker for Basal-like breast cancer subtypes and is, therefore, upregulated in many TNBCs (45, 46); its downregulation in responders could indicate alternate TNBC subtypes in these patients, such as immunomodulatory or mesenchymal TNBC.
Our data highlights that though, clinically, complete and partial responders to chemo-immunotherapy are collectively considered responders by RECIST criteria, there is significant genomic tumor heterogeneity among partial responders. Some partial responders showed tumor transcriptional landscapes and durability of treatment response similar to patients who achieved a CR, while others more closely mirrored transcriptional landscapes characteristic of PD. Regularized Cox regression showed enriched expression of immunoglobulin heavy constant gamma 1, IGHG1, in complete responders as well as nonresponders who experienced durable PFS. Others found IGHG1 mRNA expression to correlate strongly with tumor plasma cell density and positive prognosis in TNBC (47). A recent study also corroborates an association of IGHG1 with superior OS in HER2low TNBC (48). All patients with highly enriched IGHG1 expression also showed concurrent enrichment of B cells and Tfh cells in their biopsy samples. Assessment of the top 50 DEGs between IGHG1 high versus low tumors revealed enriched expression of a tremendous number of other immunoglobulin-related genes in IGHG1high tumors, supporting presence of plasma cells in these tissue samples. Findings of improved response to immunotherapy among TNBC patients with high intra-tumoral infiltration of B and Tfh cells have been well-described (49–51). Collectively, this is, to our knowledge, the first description of IGHG1 as a potential marker for durable response and survival for patients with mTNBC treated with anti–PD-1 immunotherapy and chemotherapy. Using publically available single cell RNA-seq data from BioTuring, a single cell browser and database, we were able to validate our finding of IGHG1 as a potential marker for immunotherapy response using data from a previous study by Zhang and colleagues (2021), which explored immune cell subsets associated with response to PD-L1 blockade in advanced TNBC (Supplementary Fig. S5). Our analysis showed that, compared with other previously described intra-tumoral immune cell markers associated response such as CD8 and IFNG, differences in IGHG1 expression between treatment PR, SD, and PD were highly significant.
Limitations
Our study has several important limitations. First, the small sample size of this trial limits the overall generalizability of our results. Second, significant tumor heterogeneity between the site of biopsy and areas measured for RECIST may explain some of our results. As these were research biopsies, a third limitation is tissue availability. Precedence for tissue was given to standard histopathological analysis, followed by RNA correlative analysis; only remaining tissue was used for PD-L1 assessment. At the time of this clinical trial, pembrolizumab had not yet been FDA approved for mTNBC and PD-L1 testing was not standard of care and therefore was given lowest priority tissue allocation. For this reason, patients that had very limited tissue on the tumor biopsy were less likely to have sufficient remaining tissue available for PD-L1. This largely explains why PDL1 testing wasn't performed in approximately 50% of patients in the study. Furthermore, the location of metastatic sites may have limited the amount of core needle biopsy tissue obtained. As this was a single arm, non-randomized clinical trial, prognoses cannot be compared by treatment exposure (CNP) and no causation can be established between treatment and response. Here, we are highlighting significant associations between transcriptomic landscapes and clinical responses in patients treated with CNP. It is also essential to acknowledge that response in this phase II trial was determined using RECIST 1.1 rather than iRECIST criteria, and was investigator assessed rather than centrally reviewed. Finally, the significance of IGHG1 expression must be tempered given no internal validation, such as IGHG1 staining via IHC, was performed. Further studies are needed to corroborate our findings.
Authors' Disclosures
N. Stabellini reports other support from Sociedade Beneficente Israelita Albert Einstein outside the submitted work. P. Fu reports grants from NIH during the conduct of the study. H. Moore reports personal fees from Myovant and grants from Sermonix Pharmaceuticals, AstraZeneca, Daiichi-Sankyo, Seattle Genetics, and Roche/Genentech outside the submitted work. T.A. Chan reports other support from Grtistone Bio, grants from Pfizer and AstraZeneca, and personal fees from Illumina and LGChem outside the submitted work; in addition, T.A. Chan has a patent for use of TMB as biomarker with royalties paid. A.J. Montero reports grants from Merck during the conduct of the study and other support from AstraZeneca, New Century Health, and Paragon healthcare outside the submitted work; and Alberto Montero is the spouse of coauthor and collaborator Claudia Marcela Diaz-Montero. No disclosures were reported by the other authors.
Authors' Contributions
A.D. Wilkerson: Conceptualization, data curation, validation, methodology, writing–original draft, writing–review and editing. P.B. Parthasarathy: Data curation, formal analysis, validation, visualization. N. Stabellini: Writing–original draft, writing–review and editing. C. Mitchell: Writing–original draft, writing–review and editing. P.G. Pavicic: Writing–review and editing. P. Fu: Formal analysis, writing–review and editing. A. Rupani: Data curation, formal analysis. H. Husic: Writing–review and editing. P.A. Rayman: Project administration, writing–review and editing. S. Swaidani: Project administration. J. Abraham: Writing–review and editing. G.T. Budd: Writing–review and editing. H. Moore: Writing–review and editing. Z. Al-Hilli: Writing–review and editing. J.S. Ko: Writing–review and editing. J. Baar: Funding acquisition, writing–review and editing. T.A. Chan: Writing–review and editing. T. Alban: Methodology, project administration, writing–review and editing. C.M. Diaz-Montero: Supervision, project administration, writing–review and editing. A.J. Montero: Resources, supervision, project administration, writing–review and editing.
The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.
Note: Supplementary data for this article are available at Clinical Cancer Research Online (http://clincancerres.aacrjournals.org/).