Predicting early brain metastases based on clinicopathological factors and gene expression analysis in advanced HER2-positive breast cancer patients

The overexpression or amplification of the human epidermal growth factor receptor 2 gene (HER2/neu) is associated with high risk of brain metastasis (BM). The identification of patients at highest immediate risk of BM could optimize screening and facilitate interventional trials. We performed gene expression analysis using complementary deoxyribonucleic acid-mediated annealing, selection, extension and ligation and real-time quantitative reverse transcription PCR (qRT-PCR) in primary tumor samples from two independent cohorts of advanced HER2 positive breast cancer patients. Additionally, we analyzed predictive relevance of clinicopathological factors in this series. Study group included discovery Cohort A (84 patients) and validation Cohort B (75 patients). The only independent variables associated with the development of early BM in both cohorts were the visceral location of first distant relapse [Cohort A: hazard ratio (HR) 7.4, 95 % CI 2.4–22.3; p < 0.001; Cohort B: HR 6.1, 95 % CI 1.5–25.6; p = 0.01] and the lack of trastuzumab administration in the metastatic setting (Cohort A: HR 5.0, 95 % CI 1.4–10.0; p = 0.009; Cohort B: HR 10.0, 95 % CI 2.0–100.0; p = 0.008). A profile including 13 genes was associated with early (≤36 months) symptomatic BM in the discovery cohort. This was refined by qRT-PCR to a 3-gene classifier (RAD51, HDGF, TPR) highly predictive of early BM (HR 5.3, 95 % CI 1.6–16.7; p = 0.005; multivariate analysis). However, predictive value of the classifier was not confirmed in the independent validation Cohort B. The presence of visceral metastases and the lack of trastuzumab administration in the metastatic setting apparently increase the likelihood of early BM in advanced HER2-positive breast cancer.


Introduction
The overexpression or amplification of the human epidermal growth factor receptor 2 gene (HER2/neu) is associated with high risk of brain metastasis (BM). Approximately 30-50 % of advanced HER2-positive breast cancer patients will develop BM, with an annual risk of around 10 % [1][2][3][4][5]. It has been speculated that improvements in systemic therapy resulting in greater numbers and more durable systemic responses may permit more time for BM relapse. Trastuzumab, a monoclonal antibody that targets the extracellular domain of HER2, is used in combination with chemotherapy to improve the survival of patients with HER2-positive tumors [6][7][8][9][10]. However, owing to its high molecular weight, penetration of trastuzumab into the central nervous system is extremely low, 1/420th of serum levels [11], and this compound is ineffective in treating established BM.
The development of BM predictors in advanced breast cancer patients might have practical clinical implications. First, the use of imaging to detect occult BM in unselected patients is controversial, whereas this strategy may be reasonable in patients at highest immediate risk. Second, reliable predictive factors may improve selection of patients in clinical trials assessing the efficacy of putative BM prevention strategies, such as prophylactic cranial irradiation or the use of brain-permeable compounds. Finally, these studies may prompt new therapeutic strategies.
In the present study we analyzed the risk of early BM according to gene expression, and clinical and pathological variables in two well annotated cohorts of advanced HER2positive breast cancer patients.

Patients
This study was approved by the Institutional Review Board of the coordinating centers (Medical University of Gdańsk, Poland and Indiana University, USA). Two patient cohorts were derived from a consecutive series of 315 advanced HER2-positive breast cancer patients treated in nine oncology centers in Poland and Serbia between 1993 and 2010 (consort diagram; Fig. 1). Discovery Cohort A (n = 167) and an independent validation Cohort B (n = 148) were collected between 2006-2008 and 2008-2010, respectively. According to standard clinical practice, no screening for occult BMs was used, therefore almost all BM were symptomatic. BM were defined as metastatic lesions involving the brain parenchyma, with or without accompanying leptomeningeal disease. Demographic and clinicopathologic data, as well as treatments and clinical follow-up were extracted from institutional databases or original patient files. Treatments were rule based (Table 1). Dominant metastatic sites were assigned into three categories: soft tissue, bones and viscera. Dominant metastatic site was classified by the category associated with the worst prognosis in the following order of increasing gravidity: soft tissue, bones, viscera [12].

Pathology review
The starting material from each patient was a formalinfixed, paraffin embedded specimen of primary breast cancer. A pre-cut section of each tumor, stained with hematoxylin and eosin, was reviewed by two pathologists (SB and WB) to confirm the presence of sufficient invasive breast cancer component (1 cm 2 invasive tissue, C30 % tumor cells). In Cohorts A and B, 90/167 and 75/148 tumors, respectively, had sufficient material for molecular analysis. Expression of ER and PR was determined using immunohistochemistry (IHC), with 10 % of nuclear staining considered as a positive result. HER2 protein expression was determined using semiquantitative IHC (HercepTest, Dako A/S, Glostrup, Denmark) or HER-2/ neuTest 4B5 (Ventana Medical Systems, Inc.). Only samples showing strong expression (scored 3?), defined as uniform, and intense membrane staining of at least 10 % of invasive tumor cells, were considered positive. The samples showing intermediate expression (scored 2?) were subjected to additional analysis of HER2 gene copy number using fluorescence in situ hybridization (FISH). Gene amplification by FISH was defined as a FISH ratio (HER2/ centromeric probe for chromosome 17 ratio) of greater than 2.0. FISH-positive patients were considered HER2positive.

RNA extraction
Tumor cells were processed using macrodissection to enrich their population for analysis. Sections were deparaffinized with CitriSolv clearing agent (Fisher Scientific Company, Fair Lawn, NJ) and scraped off from the slide into a microcentrifuge tube. Total RNA was extracted from three 10 lm thick whole tissue sections from each sample using the Roche high pure RNA paraffin kit according to manufacturer's instructions (Roche Applied Science, Indianapolis, IN). Purified total RNA samples were stored frozen at -80°C until needed for quality control (QC) analysis and subsequent gene expression profiling and quantitative reverse transcription PCR (qRT-PCR). The concentration of RNA was measured using Nanodrop Ò ND-1000 spectrophotometer (ThermoScientific, Wilmington, DE). RNA (200 ng) was reverse-transcribed to complementary deoxyribonucleic acid (cDNA) using iScript cDNA synthesis kit (Bio-Rad Laboratories, Inc., Hercules, CA). To prequalify RNA samples, SYBR Green-based qRT-PCR (Applied Biosystems, Foster City, CA) was performed for RPL13A ribosomal protein gene according to Illumina's instructions (San Diego, CA).

DASL analysis
Cohort A samples were analyzed by annealing, selection, extension and ligation (DASL) assay using Cancer Panel v1 to provide expression data on 502 known cancer genes. DASL was performed with the Sentrix universal array (Illumina, San Diego, California) as per the manufacturer's instructions [13] and blinded to patient outcome. Shortly, a 20-ll RT reaction containing a reaction mix (MMC; Illumina, San Diego, CA), biotinylated random hexamers and oligo-d(T) 18 , and total RNA, was incubated at room temperature for 10 min and then at 42°C for 1 h. Pooled assay oligos were annealed to their sequence-specific targets on the cDNA under a controlled hybridization program. The cDNA was immobilized on paramagnetic beads and washed to remove any excess or mis-hybridized oligos. Hybridized oligos were then extended and ligated to generate amplifiable templates, using Illumina-supplied reagents and conditions (BeadStation User's Manual, Illumina). A PCR reaction was performed with Cy3 labeled universal PCR primers. Single-stranded PCR products were prepared by denaturation, and were then hybridized to Sentrix arrays under a temperature gradient program. The arrays were imaged using a BeadArray Reader scanner (Illumina). The DASL assay was performed three times independently, and samples were hybridized to three different array matrices. The 502-gene assay was available in   Generation of the 13-gene signature Cohort A samples were divided into an internal training set and an internal testing set. Predictive analysis of microarray analysis (http://www-stat.stanford.edu/*tibs/PAM/) was performed to identify multigene profiles predictive for BM. The best gene-expression signature was selected based on a built-in 10-fold cross-validation analysis in PAM. Then the gene-signature was output as a single variable from the PAM. Its association with the BM free survival (BMFS) was analyzed in the internal testing set with a Cox regression analysis, in which clinical and demographic variable effects were justified. This analysis was performed with the R function, coxph. The gene signature construction from the internal training set used the optimal variable selection strategy in PAM, and p value was not considered. Then, the correlation between the gene signature and BMFS was assessed by the Cox regression model, and the p value\0.05 was considered as statistically significant. Real-time qRT-PCR analysis Owing to the abandoning of the 502-gene DASL assay by the manufacturer, and to increase the potential utility of the profile, we switched to a qRT-PCR assay. Apart from its clinical applicability, this method allows precise quantification of transcriptional abundance of identified genes. TaqMan reactions were performed in triplicates using custom array microfluidic cards preloaded with TaqMan gene expression assays containing 16 genes (13 discriminant genes and 3 reference genes) on an ABI Prism 7900HT fast real-time platform according to the manufacturer's instructions. The primer sequences are listed in Table 2. Transferrin receptor (TFRC), beta cytoskeletal actin (ACTB) and glyceraldehyde-3-phosphate dehydrogenase (GAPDH) were used as endogenous reference controls for normalization. Delta threshold cycle (DC t ) values for each of the 13 genes of interest were normalized using the three endogenous reference controls according to the method of Applied Biosystem's DataAssist TM Software. All procedures were performed blinded to patient outcomes. After normalization, 2 ÀDC t values were subject to

Discovery Cohort A
Of the 84 primary tumors subjected to analysis in the Cohort A, 83 were analyzable (Fig. 1)

Determinants of BMFS and OS
Performed in Cohort A binary comparison for presence or absence of BM among 502 analyzed genes did not show any differential gene expression ( Table 3]. The microarray data have been deposited in NCBI's gene expression omnibus (http://www.ncbi.nlm.nih.gov/geo; GSE38057). In order to increase the potential clinical applicability of this signature, a qRT-PCR based analysis of the 13 genes (and 3 references) was performed and showed promising preliminary results [15,16]. The TaqMan gene expression assay IDs for each gene was chosen to meet FFPE sample requirements for custom TLDA based on Applied Biosystems guidelines. As expected, DASL and qRT-PCR had inherent differences related to the platform (Fig. 2). As the next step, a leave-one-out LDA was performed using an updated database that had a longer follow-up (5 years) data. A predictive model that included only 3 of the original 13 genes: HDGF, RAD51 and TPR, with corresponding LDA coefficients of 1.06, 0.35 and -1.08, respectively, was developed. The 3-gene classifier was highly predictive of early BM both in univariate (HR 3.7, 95 % CI 1.3-11.1; p = 0.01) and multivariate analysis (HR 5.3, 95 % CI 1.6-16.7; p = 0.005; Table 3). High 3-gene classifier was associated with tumor grade 3, ERnegativity and less frequent use of endocrine treatment and trastuzumab in the adjuvant and/or metastatic setting (Table 4). Additionally, patients with high 3-gene classifier were more likely to develop the first relapse in the visceral organs.
In an independent Cohort B the mean qRT-PCR expression of 13 genes was different compared to Cohort A, and only 16 % of patients (compared to 59 % in Cohort A) were assigned to the high-risk group (Table 4). Accordingly, the 3-gene classifier was not predictive of early BM (HR 1.2, 95 % CI 0.3-20.0; p = 0.8; Table 3). In this cohort the high 3-gene classifier was associated with less frequent use of induction chemotherapy and more lung and liver metastases (Table 4).
In both cohorts the independent variables associated with shorter OS included higher tumor grade (HR 1.

Discussion
The aim of this study was to identify molecular predictors of the BM development in advanced HER2-positive breast cancer patients. This subset of breast cancer patients carry particularly high risk of BM. Additionally, some studies suggested increased risk of BM associated with the use of trastuzumab [17].
The current study employed a high throughput DASL technology based on the expression of 502 cancer related genes in addition to analysis of the clinicopathologic variables. This targeted gene analysis did not demonstrate any differential gene expression in patients who did and did not develop BM. This may likely be due to the limited number of genes analyzed, but it is also possible that BM in advanced HER2-positive breast cancer patients is a biologically determined, stochastic and inevitable event. Further analysis of the DASL led to identification of a 13-gene profile that was apparently predictive for development of early BM [15]. For precise quantification of transcriptional abundance of identified genes, we employed qRT-PCR technology, which identified a 3-gene classifier (RAD51, HDGF, TPR), also seemingly predictive for early BM. However, the significance of this classifier was not confirmed in the independent cohort.
The retrospective design of this study made it difficult to control for major clinicopathologic differences between Cohorts A and B. In consequence, patients in Cohort B had fewer ductal carcinomas and, even more importantly, less frequently received neoadjuvant chemotherapy. Gene expression alterations of breast cancer were recently demonstrated to be drug-specific, and drug-induced tumor gene signatures may be more informative than unchallenged signatures in predicting treatment outcomes [18,19]. The study by Bos et al. [20] showed that BM gene set tested in various breast cancer cohorts was less BM predictive in patients whom received postoperative systemic therapy compared to those whom did not. This confirms the hypothesis that systemic therapies, apart from their preventive effect, may also alter the pattern of relapse in breast cancer. In this study, patients in Cohort B, compared to Cohort A, had also infrequent first relapse at distant sites and significantly fewer visceral metastases. Furthermore, much more patients in this cohort received lapatinib at trastuzumab relapse (32 %, compared to 14 % in Cohort A). The pivotal study by Geyer et al. [21] showed that the addition of lapatinib to capecitabine after progression on trastuzumab resulted in decreased BM occurrence, and preclinical studies show that lapatinib prevents BMs formation by 53 % in a HER2-transfected model system [22]. The abovementioned differences between both cohorts led to better general prognosis in Cohort B compared to Cohort A, expressed by longer OS and time to diagnosis of BM. Finally, the imbalanced proportion of patients with high gene classifier in both cohorts (59 % in Cohort A vs. 16 % in Cohort B) might have largely impacted study results.
Although the gene signature could not be validated, it identified a number of genes that could be important in the development of BM. The most important of which is RAD51, a gene involved in homologous recombination in DNA double strand breaks repair [20]. RAD51 expression has been linked to response to neoadjuvant therapy [23][24][25]. We have previously reported that high cytoplasmic expression of RAD51 in breast cancer is associated with significantly increased risk of BM, particularly in combination with high Ki-67 index and ER-negativity [26]. Further, in other study demonstrated that BARD1 and RAD51 are frequently overexpressed in BMs from breast cancer and may constitute a mechanism to overcome reactive oxygen species-mediated genotoxic stress in the metastatic brain [27]. Taken together, this data suggest that RAD51 targeting might be important in HER2-positive breast cancer. High nuclear expression of HDGF, another gene constituting our 3-gene signature, was earlier found to associate with high tumor grade, Ki-67[20 %, lymph node involvement and poor prognosis in breast cancer patients [28,29]. Chen et al. [29] demonstrated that nuclear HDGF over-expression stimulates epithelial-mesenchymal transition of breast cancer cells by down-regulation of E-cadherin and up-regulation of vimentin. The third gene of our signature-TPR, a translocated promoter region nuclear basket protein, is poorly characterized but has a normal function in nuclear pore function and is the target of oncogenic fusions [30].
In the current study, the clinical factors associated with early development of BM were visceral location of first relapse and, at a borderline level, ER-negativity, the two hallmarks of tumor aggressiveness. This is partly consistent with our earlier study in advanced HER2-positive breast cancer patients, showing the association between the risk of BM and shorter time to first extracranial progression [5]. The association between ER-negativity and the occurrence of BM in HER2-positive breast cancer patients was earlier reported by other authors [2,4,31,32]. Indeed, the clinical behavior including tumor kinetics and sites of recurrence in ER-positive/HER2 positive (HER2-positive luminal B) breast cancer is different compared to that in non-luminal HER2 enriched subtype [31][32][33][34]. We also showed that trastuzumab administration in the metastatic setting may reduce the risk of early BM. This is in line with two other studies, that noticed shorter time to development of BM in HER2-positive patients who never received trastuzumab [35,36].

Conclusions
We demonstrated that the presence of visceral metastases and the lack of trastuzumab administration in the metastatic setting apparently increase the likelihood of early BM in advanced HER2-positive breast cancer, and the 3-gene classifier does not improve their predictive value. Our study also illustrates the difficulties in developing clinically useful predictive markers in the retrospective setting [37]. In our case these included problems associated with archival tissue collection, heterogeneity of patient populations and inconsistent therapeutic approaches over the study period. Further studies, including larger and more homogeneous groups, are necessary to identify biomarkers, which may help in designing BM preventive trials and prompt new treatment strategies.