Validation of the 18-gene classifier as a prognostic biomarker of distant metastasis in breast cancer

We validated an 18-gene classifier (GC) initially developed to predict local/regional recurrence after mastectomy in estimating distant metastasis risk. The 18-gene scoring algorithm defines scores as: <21, low risk; ≥21, high risk. Six hundred eighty-three patients with primary operable breast cancer and fresh frozen tumor tissues available were included. The primary outcome was the 5-year probability of freedom from distant metastasis (DMFP). Two external datasets were used to test the predictive accuracy of 18-GC. The 5-year rates of DMFP for patients classified as low-risk (n = 146, 21.7%) and high-risk (n = 537, 78.6%) were 96.2% (95% CI, 91.1%–98.8%) and 80.9% (74.6%–81.9%), respectively (median follow-up interval, 71.8 months). The 5-year rates of DMFP of the low-risk group in stage I (n = 62, 35.6%), stage II (n = 66, 20.1%), and stage III (n = 18, 10.3%) were 100%, 94.2% (78.5%–98.5%), and 90.9% (50.8%–98.7%), respectively. Multivariate analysis revealed that 18-GC is an independent prognostic factor of distant metastasis (adjusted hazard ratio, 5.1; 95% CI, 1.8–14.1; p = 0.0017) for scores of ≥21. External validation showed that the 5-year rate of DMFP in the low- and high-risk patients was 94.1% (82.9%–100%) and 80.3% (70.7%–89.9%, p = 0.06) in a Singapore dataset, and 89.5% (81.9%–94.1%) and 73.6% (67.2%–79.0%, p = 0.0039) in the GEO-GSE20685 dataset, respectively. In conclusion, 18-GC is a viable prognostic biomarker for breast cancer to estimate distant metastasis risk.


Introduction
Breast cancer is the most common cancer in women, accounting for more than one-tenth of cancers worldwide [1]. The incidence of breast cancer is rapidly rising in developing countries, and will become a major health burden in both developed and developing countries.
Owing to the advancement of breast cancer screening in recent decades, double the number of early-stage breast cancer cases are being detected as compared to the rates in the 1980s [2]. This has led to debates on the overtreatment of early-stage breast cancer [3].
Overtreatment not only increases social and family burden, but also causes unnecessary harm to patients with a favorable prognosis [4]. Early-stage breast cancer patients often receive adjuvant chemotherapy, with only a small proportion (2-20%) deriving benefit, while the others are at risk of toxic side effects [5]. Chemotherapy can lead to acute side effects (nausea and vomiting, fatigue, leukopenia, thrombocytopenia, infection, etc.) and sustained late complications (cognitive dysfunction, heart failure, or treatment-related secondary cancer) [6]. Radiotherapy increases (up to six times) the risk of death from ischemic heart disease and secondary lung cancer after 10 years [7,8]. Therefore, cancer societies have focused on additional and new/novel adjuvant chemotherapy with fewer side effects in the past decades, and efforts to customize therapy and reduce chemotherapy for patients unlikely to benefit will probably be the main goal of future research [6].
The key to preventing early breast cancer overtreatment and providing a more individualized treatment is to utilize molecular diagnostics to determine whether the cancer is likely to be indolent or aggressive [9,10]. In western countries, several multigene panels (Oncotype Dx, MammaPrint, and EndoPredict) have been developed to predict the possibility of distant metastasis in hormonal receptor-positive and human epidermal growth factor receptor type 2 (HER2)-negative breast cancer patients [11,12]. These low-risk patients may omit adjuvant chemotherapy but use adjuvant hormone therapy alone [13]. Risk stratification of non-luminal breast cancer by the currently available multigene panels is limited [12]. Moreover, racial disparity of breast cancer subtypes exists [14]; whether the multigene panels developed from women of European descent are applicable to the Asian population warrants further analysis [15].
We previously reported an 18-gene classifier (18-GC) derived from a 34-gene panel initially applied to estimate the local/regional recurrence (LRR) risk in breast cancer patients after mastectomy [16]. This 18-GC includes BLM, TCF3, PIM1, RCHY1, PTI1, DDX39, BUB1B, STIL,  TPX2, CCNB1, MMP15, CCR1, NFATC2IP, TRPV6, OBSL1, C16ORF7, DTX2, and ENSA. The expression of RCHY1, PTI1, ENSA, and TRPV6 is associated with better tumor biology and disease control. The remaining 14 genes are associated with poor outcomes. The optimal cutoff to predict LRR after mastectomy was set at !44. Patients with scores <21 were generally free from recurrence (both locally and distantly) [17]. LRR has long been thought to be a sign of distant metastasis; patients with LRR would eventually develop systemic recurrence [18]. Thus, our gene set may not only be related to LRR, but also represent a prognostic factor for distant metastasis.
A substantial portion of breast cancer patients treated with surgery alone was deemed curable before the era of chemotherapy [19]. In this subset of patients, tumor biology is localized. We hypothesized that our LRR-gene panel can identify this disease entity, and aimed to validate the 18-GC as a prognostic factor of distant metastasis after primary breast surgery.

Methods and patients
Between 2005 and 2014, patients with clinical stage I-III breast cancer who underwent primary surgery in a freestanding cancer center in Taiwan, and had frozen tissue of their primary tumor stored during the operation, were considered potential candidates. The flow chart for patient selection is shown in Fig 1. Events of recurrences in primary operable breast cancer were relatively small. Patients with recurrent disease were preferentially enrolled. Others were randomly selected from our database. The last follow-up date was October 30, 2016. All patients received treatment and care in accordance with contemporary, evidence-based medicine hospital practice guidelines, which are largely similar to the National Comprehensive Cancer Network guidelines. The adjuvant chemotherapy drugs used in the distant recurrent patients (n = 101) were: dose-dense ATC in 32 patients, 3 cycles of doxorubicin plus 6 cycles of cyclophosphamide/methotrexate/5-fluorouracil in 13 patients, 4 cycles of doxorubicin/ cyclophosphamide and 4 cycles of taxol or taxotere in 19 patients, 6 cycles of cyclophosphamide/ doxorubicin/cyclophosphamide in 22 patients, others (mixed regimens due to intolerance of initial treatment) in 6 patients, and no adjuvant chemotherapy in 9 patients (Table 1). Eligible patients met the following inclusion criteria: (1) invasive carcinoma of the breast, (2) clinical stages I-III, (3) first treatment being surgery (modified radical mastectomy, total mastectomy and sentinel lymph node biopsy, and breast conserving surgery with sentinel node biopsy and/or axillary lymph node dissection), (4) frozen fresh tissues available, and (5) provided written informed consent. Patients who had pre-operative chemotherapy and/or radiotherapy were excluded. The study was approved by the Biobank Ethics Committee and Institutional Review Board of Koo Foundation Sun Yat-Sen Cancer Center (IRB 20131001A).
Procedures for mRNA extraction from frozen fresh tissues (tumor tissue, at least 50%) and microarray analysis have been reported previously [16,20]. Details on the Affymetrix protocol can be found on the following website: https://www.thermofisher.com/order/catalog/product/ 900470?SID=srch-srp-900470. Frozen tissue samples (n = 818) from surgical specimens of primary tumors taken from patients prior to any systemic treatment were used. In brief, total RNA was isolated using Trizol (Invitrogen, Carlsbad, CA) and purified with the RNeasy Mini Kit (Qiagen, Valencia, CA). RNA quality was assessed using an RNA 6000 Nano Kit and Agilent 2100 Bioanalyzer (Agilent Technologies, Waldbronn, Germany). The RNA samples had an average RNA integrity number of 7.85 ± 0.99 (mean ± SD). Biotinylated targets were prepared according to published methods (Affymetrix, Santa Clara, CA, USA) and hybridized to the Affymetrix microarray U133 2.0 GeneChip [21]. Arrays were scanned using standard Affymetrix protocols. Raw data from Affymetrix chips were then normalized with the robust multi-array average (RMA) for background adjustment.

Statistical methods
Cox proportional hazards regression models were used to assess the prognostic significance of age at diagnosis, primary tumor size, the number of axillary lymph nodes involved, tumor grade, lymphovascular invasion (LVI), ER/PR status, HER2 overexpression, breast cancer subtypes, and the 18-gene score (stratified by either a categorical factor or a continuous variable).  Breast cancer subtypes were classified via immunohistochemistry [22]: luminal A subtype, ER or PR (+), HER2 (-), and grade 1 to 2; luminal B subtype, ER or PR (+), HER2 (-), and grade 3. HER2 subtype means ER or PR (+/-) and HER2 (+). Triple-negative subtype represents ER (-), PR (-), and HER2 (-). Duration of freedom from distant metastasis was defined as that from the first day of treatment to the day of diagnosis for any distant metastasis, or to the last follow-up. Probability of freedom from distant metastasis (DMFP) was calculated according to the Kaplan-Meier method. Patients with LRR or those who died before distant recurrence were censored. A logrank test was performed to assess the statistical significance of the differences in DMFP between patient subsets. All statistical analyses (P < 0.05) were performed using SAS, version 9.4.

Internal and external validation
We excluded 135 mastectomy patients who were included for the development of the 18-gene scoring algorithm in predicting LRR. Quantile normalization using the 135 patients' gene expression levels was applied to the remaining 683 patients. These patients were then included in the internal validation for the estimation of distant metastasis risk (Fig 1).
We also used a dataset from GEO (GSE20685, ID: 200020685) from patients treated between 1991-2004, and a dataset from the National Cancer Center Singapore for external validation (Singapore CIRB 2016/2449), from which we obtained a microarray dataset without knowledge of patient outcomes. After mean shift of quantile normalization using our dataset [23], each patient was assigned a "score." Survival analysis was performed independently by the Singapore group.
The patient factors strongly related to distant metastasis included age <40 or >60, more advanced tumor stage and nodal status, hormonal receptor-negative, prominent LVI, higher tumor grading, and 18-gene scores (Table 1). Treatment factors associated with distant metastasis were primary surgical type, adjuvant hormonal therapy, and adjuvant radiotherapy. Time to distant recurrence in each subtype was different (S1 Fig). According to the 18-gene scoring algorithm, 21.4% (n = 146) of the patients were classified as low-risk (scores of <21) and 78.6% (n = 537) of the patients were classified as high-risk (scores of !21). The distant metastasis rates in the low-and high-risk groups were 2.7% and 18.1%, respectively (p < 0.0001, Table 1).

Univariate and multivariate analysis
Univariate analysis for distant metastasis by Cox proportional hazards regression revealed that age, subtype, tumor stage, nodal stage, tumor grade, LVI, and adjuvant treatments were prognostic factors. The hazard ratio (HR) was 7.1 (95% CI, 2.6-19.2) for patients with 18-gene scores of !21 compared to patients with scores of <21 (HR 1.0). In the multivariate analysis, the hazard ratio was 4.9 (95% CI, 1.8-13.5; Table 2). Other risk factors independently related to distant metastasis included T stage (T2, HR 1.  Table 2). Breast cancer subtypes by immunohistochemistry were prognosticative after univariate analysis, but not significant after multivariate analysis.

Variable Hazard ratio (95% CI) (Crude) P-value Hazard ratio (95% CI) (Adjusted)* P-value
The study group (135 training, 683 validation) was selected from an initial cohort of 8,155 patients, which represented a randomly selected breast cancer population in a free-standing cancer center where about one-tenth of breast cancer patients in Taiwan are treated [24,25]. Unlike other multigene panel trials that focused on a specific subtype or stage of breast cancer, our study shows that 18-GC is a sound classifier predicting favorable or unfavorable prognosis regarding distant metastasis in a general breast cancer patient population.
The distant metastatic rate was higher in our study group than in our initial group because patients with recurrences were preferentially enrolled (Fig 1). However, the result derived from this cohort is consistent with the common consensus that 1) higher T and N stages increase the likelihood of distant metastasis; 2) adjuvant treatments decrease the likelihood of distant metastasis; 3) triple-negative, HER2, and luminal-B subtypes are more aggressive than luminal-A breast cancer with higher 18-gene scores and recurrence rates (Table 2).
Overall, the 18-GC can identify 21% of low-risk breast cancer patients with a 5-year distant metastasis rate of <4%, regardless of stages and subtypes. There are three unique characteristics of the 18-GC: 1) it is the first Asian-based gene expression profiling validated in a large and diverse breast cancer population; 2) it could stratify the risk of relapse in both luminal and non-luminal breast cancer, although the proportion of low-risk patients with non-luminal cancer is small, and 3) it can simultaneously estimate the likelihood of LRR and distant metastasis.
During the past 15 years, there has been a consensus that early-stage breast cancer is a heterogeneous disease with different molecular subtypes and prognoses [9,26,27]. Several multigene panels for early-stage breast cancer have been developed [12]. The subsequent multigene panels (Prosigna, EndoPredict, Breast Cancer Index) were claimed to be better than the former ones (Oncotype Dx, MammaPrint, Genomic Grade Index) for the prediction of late distant metastasis [28,29]. These new panels focus mainly on luminal-like and node-negative breast cancer and could lead to the possibility of omitting chemotherapy in the low-risk group.
However, the possibility of stratifying the risk of relapse in non-luminal breast cancer patients remained an unsolved question. For example, the prognostic risk discrimination of the 70-gene and 76-gene panels is efficient among ER (+) patients, but <5% of ER (-) patients are classified as being in the low-risk group [9,10]. A relatively high false-negative rate (>50%) for the HER-2 subtype by the 21-gene panel was also noted [30]. The first-generation genomic panels rely largely on quantification of proliferation-related genes to determine the prognosis of ER (+) disease [31]. Consequently, these panels are limited by their clinical utility in other breast cancer subtypes. The gene expression profiles related to immune response and stromal invasion have prognostic value for ER (-) and high proliferative ER (+) breast cancer patients [32][33][34][35]. However, those classified by second-generation genomic panels as the lowrisk group had a 5-year relapse rate of up to 20%.
In this study, 18-GC identified 12% HER2 (+) and 10% triple-negative breast cancer patients as the low-risk group with a 5-year distant metastasis rate of 5% and 0%, respectively. This result opens the possibility of stratifying the relapse risk of non-luminal breast cancer patients. If this can be achieved, reduction of the chemotherapy dosage in this subset may be possible. Additionally, our study revealed that patients with high 18-gene scores would have a higher risk of distant metastasis than those with low scores; the 5-year rate of DMFP was 20%-30% regardless of treatments, and therefore, dose-dense chemotherapy or a novel clinical trial may be considered in these patients.
Although our research indicates great improvement in classifying breast cancer, there are limitations to our study. First, it is a single-centered, retrospective study. Up to 89% of the patients underwent adjuvant chemotherapy, and 81% underwent radiotherapy. Although our data showed that 18-GC is a prognostic tool in different treatment modalities, it is still hard to surmise that adjuvant therapy can be omitted in the low-risk group. However, our data would enable the identification of a group of high-risk patients who may benefit from dose-dense chemotherapy, higher dose of radiotherapy, or novel clinical trials. The 5-year distant metastasis rate in the low-risk group was only 2.7% (4/146). Meta-analyses of long-term outcomes have suggested that adjuvant chemotherapy may reduce the recurrence risk by 30%-50% [36]. Even if we assumed the greatest odds, say 50%, the initial distant metastasis rate without chemotherapy in the low-risk group would be around 5%. Thus, only 3% may benefit from the procedure, and it could expose the other 95% of patients to the toxic effect of chemotherapy in the low-risk group. A prospective, multi-centered study is warranted to validate whether the low-risk-group patients can forego chemotherapy for luminal breast cancer and whether the chemotherapy dosage may be decreased in non-luminal breast cancer. Second, some selection bias existed in our study. For the mRNA microarray, we preferentially enrolled those patients with recurrence or mortality. Therefore, our distant metastasis rate may overestimate the recurrence risk.
In conclusion, the risk stratification limitations of clinical and pathological parameters still leave a significant number of patients with breast cancer at risk of overtreatment or suboptimal treatment. However, the 18-GC can predict LRR and distant metastasis simultaneously, and will aid the advancement into a new era of precision medicine and breast cancer treatment.