Exploratory biomarker analysis from a phase II clinical trial of eribulin plus gemcitabine versus paclitaxel plus gemcitabine for HER2‑negative metastatic breast cancer patients (KCSG BR13‑11)

Abstract

Purpose: We conducted an exploratory biomarker study from a phase II clinical trial of eribulin plus gemcitabine (EG) versus paclitaxel plus gemcitabine (PG) in HER2-negative metastatic breast cancer (BC) patients.

Methods: We performed targeted deep sequencing with a customized cancer gene panel and RNA expression assay. Tumor mutation burden (TMB) and mutation signatures were determined based on genetic alteration in targeted regions.

Gene set variation analysis was performed with PanCancer Immune Profiling and PanCancer Pathway Panels. Statistical analyses were conducted to identify the associations between genetic alterations and clinical outcomes.

Results: Of 119 patients, 40 had available biomarker data. Among the 40 patients, 4 supported their post-treatment tissues.

In targeted deep sequencing, FAT3 (48%) was the most frequently mutated gene, followed by PKHD1, TP53, GATA3, PARP4, and PIK3CA. In terms of gene expression, low expression of epithelial-mesenchymal transition (EMT) pathway genes was associated with prolonged progression-free survival (PFS) in the EG group, while high expression of the EMT pathway was associated with good prognosis in the PG group.

Median TMB was 6.5 (range 2.44–46.34) and there was no relationship between TMB and patient prognosis. Analysis of mutation signatures showed that signatures 3, 20, and 26 were frequently observed in our cohort.

Further survival analysis according to mutation signature showed that mutation signature 3, as a homologous recombinant deficiency-related signature, was highly associated with disease progression (hazard ratio (log2 scale) 8.21, 95% confidence interval 2.93–13.48, p = 0.002). Kaplan–Meier plot also showed that BCs with signature 3 had short PFS compared to those without these signatures (median PFS (months) for signature 3 (low vs. high): 17.2 vs. 8.1, p = 0.0026).

Conclusions Mutation signature 3, found in about 30% of MBCs regardless of hormone receptor status, was associated with short PFS for patients with cytotoxic chemotherapy.

Introduction

Exploratory biomarker analysis has been performed in many clinical trials to predict response to drugs [1, 2]. Most recently, development of next-generation sequencing (NGS) techniques allows prompt multigene assays; there- fore, exploratory translational research has been actively performed.

In breast cancer (BC), parallel translational research has revealed that PIK3CA and ESR1 mutations are associated with drug resistance and patient prognosis [1, 2]. Patients with hormone receptor-positive and human epidermal growth factor receptor 2 (HER2)-positive BC harboring PIK3CA alteration had shorter progression-free survival compared to BC patients without PIK3CA mutation under everolimus and trastuzumab combination treatment [1].

ESR1 and PIK3CA mutations were frequently observed in patients with short progression-free survival (PFS) [2]. In addition, ERBB2 and ERBB3 mutations are potential biomarkers of neratinib, a pan-HER kinase inhibitor [3]. Germline BRCA1/2 mutations also had a predictive role as platinum chemotherapy and PARP1 inhibitors [4–6].

In general, triple-negative breast cancer (TNBC) and hormone receptor-positive BC that must be treated with cytotoxic chemotherapy has a relatively aggressive pheno- type and worse prognosis rather hormone receptor-positive BC treated with an anti-hormonal agent [7, 8].

However, regarding cytotoxic chemotherapy, parallel translational research has been rarely performed. Based on the BR9601 trial, HER family genes (Epidermal growth factor recep- tor; EGFR, HER2, and HER3), Ki67, and topoisomerase IIa were evaluated to identify associations with anthracycline treatment response [9].

In this study, BC without HER fam- ily gene expression had a higher response rate to epirubicin plus cyclophosphamide, methotrexate, and fluorouracil (CMF) chemotherapy compared with CMF. Besides, other microarray-based parallel translational research has been performed to identify chemotherapy biomarkers [10, 11].

In terms of genetic mutation, BC harboring germline BRCA mutation increased response rate to platinum-based chemotherapy compared with anthracycline-based chemo- therapy [12]. Recently, cancer with germline BRCA1/2 mutation had double response rate of carboplatin compared with docetaxel [6].

Therefore, biomarkers predicting treat- ment response can aid when choosing effective chemothera- peutic agents and prolong survival in BC patients treated with a cytotoxic agent.

We previously conducted a phase II clinical trial of eribu- lin plus gemcitabine (EG) versus paclitaxel plus gemcitabine (PG) as first-line chemotherapy in HER2-negative metastatic breast cancer (MBC) patients [13].

The result of this clini- cal trial showed that EG chemotherapy has similar efficacy and less neurotoxicity compared to PG chemotherapy. In this biomarker analysis, we aimed to identify biomarkers that predict drug response and prognosis using targeted deep sequencing and gene expression assays.

Patients and methods

Patients

Patients with histologically confirmed HER2-negative MBC and with no prior history of chemotherapy for metastatic disease were eligible for the study. Patients were eligible for the study if at least 12 months had passed since the com- pletion of prior chemotherapy, even if they had received an anthracycline- or taxane-containing regimen as neoadjuvant or adjuvant therapy.

This parallel biomarker study was con- ducted using tumor samples at diagnosis and after disease progression. All patients consented to provide tumor sam- ples.

Study design

This study was a multicenter, randomized phase II, open- label study of the Korean Cancer Study Group (KCSG) (KCSG-BR13-11). Patients were randomly assigned, in a 1:1 ratio, to either the EG or PG chemotherapy group.

EG chemotherapy comprised of intravenous (IV) admin- istration of 1.0 mg/m2 eribulin and administration of 1000 mg/m2 gemcitabine as a 30-min IV infusion on days 1 and 8 q 3 weeks.

PG chemotherapy comprised of IV admin- istration of 175 mg/m2 paclitaxel on day 1, with 1250 mg/m2 gemcitabine administered as a 30-min IV infusion on days 1 and 8 q 3 weeks.

The study was conducted in full accordance with the guidelines for Good Clinical Practice and the Declaration of Helsinki and was approved by the Institutional Ethics Committees of each hospital and the KCSG Institutional Review Board. The ClinicalTrials.gov identifier number was NCT02263495. Written informed consent was obtained from each participant.

Biomarker analysis

We evaluated genetic alterations and gene expression. Tumor DNA was extracted from formalin-fixed, paraffin-embedded (FFPE), or fresh frozen tissue.

To perform targeted deep sequencing, we used Cancer- Scan™, a 375-cancer gene panel (Table S1). After enriched exome libraries were multiplexed, they were sequenced on a HiSeq 2500 sequencing platform (Illumina). A paired-end DNA sequencing library was prepared through gDNA shear- ing, end-repair, A-tailing, paired-end adapter ligation, and amplification.

After hybridization of the library with bait sequences for 27 h, the captured library was purified and amplified with an index barcode tag, and the library qual- ity and quantity were assessed.

Sequencing of the exome library was performed using the 100-bp paired-end mode of the TruSeq Rapid PE ClusterKit and TruSeq Rapid SBS Kit (Illumina). This targeted deep sequencing panel included entire exons and some upstream areas.

Sequence reads were mapped to the human genome (hg19) using Burrows-Wheeler Aligner (BWA, v0.7.5a), and duplicates were removed by Picard (v1.93) [http:// broadinstitute.github.io/picard/] and SAMtools (v0.1.19).

The Genome Analysis Toolkit (GATK, v2.4-7) was used to recalibrate base quality and optimize local realignment. Variant calling was performed only in the targeted region. To detect single-nucleotide variants (SNVs), we integrated the results of the variant callers LoFreq (v0.6.1) and Mutect (v.1.1.4) to increase sensitivity.

Indels and copy number variations (CNVs) were called using Pindel (v0.2.4w) and GATK (v3.1.1), respectively. Variants were annotated using ANNOVAR [14], with gene and chromosome information, exonic function (synonymous, nonsynonymous, stop gain, nonframeshift, or frameshift indel), amino acid changes, and allele frequencies extracted from public databases such as the 1000 Genomes Project (2015 August version), dbSNP (version 137, 138), and ClinVar (2017 January version).

To remove common germline polymorphisms, variants were filtered using public and in-house databases. Variants were chosen if they were located in an exonic region, and syn- onymous variants were filtered out.

SNVs with a frequency of 1% with at least five supporting reads were chosen for further analyses. Indels having at least 8 supporting reads were reported.

For gene expression analysis, we performed an nCounter expression assay (NanoString Technologies, Seattle, WA) using a 730-gene PanCancer Immune Profiling Panel and a 730-gene PanCancer Pathway Panel. These two gene panels shared 130 genes (Table S2).

An nCounter CodeSet (NanoString Technologies) containing a biotinylated cap- ture probe for 170 genes and five housekeeping genes and reporter probes attached to color-barcode tags, according to the nCounter CodeSet Design, was hybridized in solu- tion to 200 ng of total RNA for 18 h at 65 °C according to the manufacturer’s instructions. Hybridized samples were loaded into the nCounter Prep Station for post-hybridization processing.

On the deck of the Prep Station, hybridized sam- ples were purified and immobilized in a sample cartridge for data collection, followed by quantification of target mRNA in each sample using the nCounter Digital Analyzer. Quan- tified expression data were analyzed using NanoString nSolver analysis software.

After performing image quality control using a predefined cutoff value, outlier samples were excluded using a normalization factor based on the sum of positive control counts greater than threefold. The counts of the probes were then normalized to the geometric mean of the five housekeeping genes.

Accordingly, expression level represents normalized mRNA transcript counts per 200 ng of total RNA extracted from tumor tissue. Gene set analyses were performed with PanCancer Immune Profiling Panel and PanCancer Pathway Panel using GSVA [15], which esti- mates gene set enrichment for each sample based on gene expression data.

Tumor mutation burden and mutation signature analyses

Tumor mutation burden (TMB) was defined as the number of somatic variants per megabase of genome. We assessed TMB based on variants detected by targeted deep sequenc- ing, CancerScan™. Among variants in coding regions, nonsynonymous SNVs and frameshift indels were counted

Germline variants with population frequency > 0.001 in the Exome Aggregation Consortium (ExAC) [16] database and Korean population were also excluded. Variants listed in the Catalogue of Somatic Mutations in Cancer (COSMIC) [17] were filtered. To calculate the TMB per megabase, the total number of mutations counted was divided by the size of the coding region of the targeted region.

We obtained a set of 30 mutation signatures that represent distinct characteristics of human cancer types based on base substitution at the mutation site [18]. Mutation signature was inferred from all base substitutions detected by targeted deep sequencing.

We calculated 30 mutation signatures for each sample with R package deconstructSigs [19], which identi- fies the weighted combination of predefined signatures to best explain the mutational profiles used.

Statistical analysis

We evaluated the impact of genetic alterations on treatment efficacy in each treatment group. Progression-free survival (PFS) and overall survival (OS) according to genetic infor- mation in each treatment group were estimated using the Kaplan–Meier method.

A Cox proportional hazards model was used to test the interactions between treatment and bio- markers and to compute hazard ratios (HRs) and the associ- ated 95% confidence interval (CI) for the biomarker-defined subpopulations. We also evaluated correlation between genetic alteration and overall response rate (ORR) using Fisher’s exact test.

Results

Population with biomarker analysis

From 52 patients, we obtained 57 tumor tissues. Tumor DNA was extracted from 53 FFPE tissues from 43 patients, and RNA was extracted from 50 FFPE tumor tissues from 46 patients.

Finally, 38 baseline and 6 progression tissues from 40 patients were analyzed using nCounter expression assays, and 26 baseline and 5 progression tissues from 28 patients were analyzed using targeted ultra-deep sequencing. Twenty-two patients were enrolled in the EG group, and 18 patients were in the PG group.

Genetic alterations and gene expression assay

The most commonly mutated gene was FAT3 (48%), followed by PKHD1 (42%), TP53 (39%), GATA3 (29%), PARP4 (29%), and PIK3CA (29%). In terms of type of alteration, single-nucleotide substitutions were frequent, while frame shift deletions were commonly detected in TP53, and only frame shift insertions were seen in GATA3.

Further survival analysis showed that genetic alterations were not associated with patient survival. FAT3 mutation seemed to be related to poor prognosis, but not statistically significant (p = 0.079).

In terms of gene expression, DEG analysis and GSVA were performed to determine a relationship between gene expression and patient prognosis. We chose a 24-months PFS (the median follow-up duration) as a measure of disease progression. According to this criterion, 8 MBCs had not recurred (3 in the EG group and 5 in the PG group).

In DEG analysis using cancer panels, high expression of SHC3, ITGB3, IL1B, COL6A6, and FGF10 was associated with long duration of treatment response, while expression of CCNE1, LEFTY2, BAMBI, and RNF43 were related to disease progression (p < 0.05, respectively).

In the immune panels, DEG analysis showed that BCL6, CSF1, and IL6ST expression predicted no disease progression in contrast to CD24 and C8G expression, which was associated with disease progression (p < 0.05, respectively).

Among all gene panels, genes related to angiogenesis were most intensively associated with disease prognosis. To identify the impact of expression of angiogenesis associ- ated genes on patient prognosis, further DEG analyses were performed according to treatment group.

In the EG group, tumors with disease progression expressed low levels of CD34, TNFSF12, VEGFA, FOXO4, and HDAC5, and high levels of NODAL, LAMA5, and S100A7 (p < 0.05, respec- tively). In the PG group, low expression levels of 13 genes including IL18 and IL1B were associated with disease progression.

In GSVA using cancer panels, the expression levels of gene sets associated with tumor invasion, PI3K, and ECM remodeling were associated with lack of disease progression. Among immune panels, hypoxia, angiogenesis, and leukocyte function were associated with tumors with better prognosis.

According to treatment group, low expression of epithelial-mesenchymal transition (EMT) pathway genes and high expression of Notch pathways were associated with prolonged PFS in the EG group In contrast, high expression of the EMT pathway was associ- ated with good prognosis in the PG group.

In conclusion, mutation signature 3, found in about 30% of MBCs regardless of hormone receptor status, was associ- ated with short progression-free survival for patients with cytotoxic chemotherapy.

Considering the rarity of genetic studies for cytotoxic chemotherapy, although most MBC patients are eventually treated with cytotoxic agents, com- prehensive genetic analysis would help many MBC patients to expect their response to cytotoxic chemotherapy. Eribulin