MicroRNA Signature for Early Prediction of Clear Cell Renal Cell Carcinoma Metastasis and Prognosis Using Formalin-Fixed Paraffin-Embedded Specimens

1. Introduction

Kidney cancer, mainly renal cell carcinoma (RCC), accounts for approximately 438,000 new cases and 155,953 deaths annually worldwide and represents the 6th and 9th leading cancers in American men and women, with an estimated incidence of 61,560 and expected mortality of 14,080 in 2026 [1,2]. Approximately 51% of RCC patients initially present with stage I disease, 11% with stage II, 16% with stage III, and 23% with stage IV disease [3]. More than 40% of patients suffer from metastatic RCC (mRCC), either presenting with metastases at diagnosis or even 30 years post nephrectomy [4]. Up to 80% of RCC metastasis occurs within 3 years and 93% within 5 years.

Currently, few patients with metastasis can be cured, but better understanding of signaling pathways of RCC progression resulted in significant development of targeted agents for metastatic disease. The targeted therapies improved clinical outcomes in randomized clinical trials by inhibiting the vascular endothelial growth factor (VEGF), mammalian target of rapamycin (m-TOR) and other pathways [5]. Several clinical prognostic factors stratify mRCC patients to low, intermediate, and high risk groups with median survival of 43.0, 22.5, and 7.8 months, respectively [6,7]. Clinical trials are assessing newly developed targeted therapies for mRCC as adjuvant treatment for high risk localized RCC to prevent recurrence or metastasis [8,9]. These clinical trials require reliable biomarkers to identify high-risk patients early, but currently, there are no reliable molecular biomarkers clinically available.

MicroRNAs (miRNAs) play important roles in cancer metastasis [10]. As functionally mainly bound to Ago2 protein, miRNAs are relatively stable even in archival formalin-fixed paraffin-embedded (FFPE) tissue blocks, making them robust bio-molecules for developing large cohort-based biomarkers [11,12]. Several studies have shown dysregulated miRNA expression patterns to be specifically correlated with RCC metastasis and prognosis [13,14,15,16,17]. However, these studies have not progressed to clinical applications due to lack of either well-established algorithms for reliable risk evaluation or sufficiently large cohort validation for the assays developed [18].

About 70-80% of RCCs are clear cell type, and we previously demonstrated the dysregulation of miR-10b-5p, miR-130b-3p, miR-139-5p and miR-199b-5p to be highly correlated with clear cell RCC (ccRCC) metastasis [12]. This original microarray-based study successfully led to the identification of a 4-miRNA signature for ccRCC metastasis and prognosis in frozen tumor specimens [12]. In the current study, we used the biostatistical approach and archival FFPE specimens from the same training cohort and built a signature algorithm including optimized risk score cutoffs specifically for FFPE specimens, with further validation in a 265-case FFPE tissue test cohort of primary ccRCC samples.

2. Materials and Methods

2.1. ccRCC Tissue Cohorts

In this study, we used the FFPE specimens from patients in the previous training cohort with localized (pT1/Stage I; n=13) and metastatic (M1; n=15) ccRCCs [12]. As described, we used pT1 and metastatic ccRCCs to represent low and high risk diseases, respectively. A miRNA signature algorithm was then rebuilt and specifically optimized for FFPE tissue. We established a separate FFPE tissue test cohort of primary ccRCC samples (pT1-4/stages I-IV; n=265) for validation. This test cohort included all available and qualified nephrectomy cases of ccRCC patients at City of Hope National Cancer Center (COH) from 1986 to 2012. The cases were collected adhering to the COH Cancer Protocol Review and Monitoring Committee and Institutional Review Board. All patients had been followed for at least 3 years if no metastasis reported. Each case was separately reviewed by two genitourinary pathologists with a consensus on the diagnosis. No cases of primary ccRCCs with sarcomatoid components were included. The clinical characteristics of patients in the training and test cohorts are summarized in Table 1. For each case, a tumor area with the highest nuclear grade on the H&E slide was marked by the same genitourinary pathologist.

2.2. Tissue and Total RNA Preparation, and miRNA Quantitative RT-PCR Testing

Tissue sample preparation, total RNA extraction, and miRNA RT-PCR testing were described previously [11,12,19]. Briefly, up to 3 cores (1 mm in diameter, 2-5 mm in thickness) of tumor samples from FFPE blocks were punched using biopsy needle (Miltex, York, PA, USA) and total RNA were then extracted using the RecoverAll^TM Total Nucleic Acid Isolation Kit (Life Technologies, Grand Island, NY, USA). The expression of miRNA was measured by quantitative RT-PCR (qRT-PCR) using the TaqMan MicroRNA Assay and 7900 HT Fast Real-time PCR System (Applied Biosystems, Carlsbad, CA, USA). Briefly, 100 ng of total RNA were converted to first strand cDNA with miRNA-specific primers, followed by real-time PCR with TaqMan probes. PCR reactions for each sample were carried out in triplicate. As we reported previously, miR-24-3p was one of the most stably expressed miRNAs in human benign renal epithelial and ccRCC tissue and was used as the internal reference [12]. In the current study, the miRNA expression, normalized with miR-24-3p, was quantified using the formula X=2-∆ CT, where ∆CT=CT(miR-X)–CT(miR-24-3p) [20]. Each miRNA qRT-PCR test was done in triplicate and an average of the ∆CT values obtained was used in the study.

2.3. Method for Building an FFPE Tissue-Specific 4-miRNA Signature

As previously described [12], we used the R statistical language for data analysis and risk score method to build a mathematical formula [21]. This formula was a linear combination of the expression levels of the 4 miRNAs, weighted by the regression coefficients derived from the univariate logistic regression analysis. We further validated the performance of the signature using separate test cohort cases. Each patient's risk status determined by the calculated risk score was compared to her/his outcome using clinical follow-up information.

2.4. Statistical Analysis

We used univariate Cox regression models to evaluate the association between the summary value and event-free survival in the validation group. A multivariable Cox regression model was developed by using the backward-selection method for event-free survival. The candidate variables that were initially considered to be included in the model were those that were significant at an alpha level of 0.20 in the univariate models and satisfied by proportional-hazard assumption. Variables remaining in the final model were significant at an alpha level of 0.05. Estimates for hazard ratios and corresponding 95% confidence intervals were obtained for each significant outcome factor. Kaplan-Meier curves were generated for event-free survival in the validation group, with data stratified according to the miRNA signature risk score. All statistical analyses were performed by using SAS 9.4.

3. Results

3.1. Rebuild a miRNA Signature for FFPE Specimens

We used the FFPE specimens from patients in the previous training cohort that had localized (pT1/Stage I; n=13) and metastatic (M1; n=15) ccRCCs12 (Table 1). We measured the expression of miR-10b-5p, miR-130b-3p, miR-139-5p and miR-199b-5p by qRT-PCR, normalized with miR-24-3p. We used the same method as described [12] to build a FFPE metastatic risk score formula (see Materials and Methods). This newly developed FFPE tissue-specific signature algorithm is described as following:

Risk Score = 15+1.4159×XmiR-10b–1.6789×XmiR-130b+1.1338×XmiR-139-5p–0.5994×XmiR-199b-5p. We added “15” in the formula as a correction factor to make the risk score a positive value.

Based on the risk score values, the training cohort localized and metastatic ccRCC cases formed two distinct, nonoverlapping groups with a gap in between (Figure 1). These two scores (rounded to the nearest hundredth place) were set as cutoff points: ≥7.74 for high risk, ≤6.08 for low risk. For a test cohort sample, if the risk score falls into this gap between the cutoff points (between 6.08-7.74), its risk status (biologic behavior) is difficult to predict, and should be interpreted as “equivocal”.

3.2. miRNA Signature Reliably Predicts ccRCC Metastasis

Among all 265 primary ccRCC specimens in the test cohort, 209 cases (79%) were predicted to have high or low risk whereas 56 (21%) were considered equivocal. For each of the 209 patients, the predicted risk status was compared with the patient’s clinical outcome. The overall sensitivity and specificity (n=209) (concurrent or subsequent metastasis) were 80% (53/66) and 76% (109/143). For patients presenting with localized disease (n=173), the sensitivity and specificity were 73% (22/30) and 76% (109/143). The receiver operating characteristic (ROC) curve is shown in Figure 2. The area under curve (AUC) was 0.82. Specifically, the sensitivities and specificities were 72% (8/11) and 78% (90/116) for stage I, 70% (7/10) and 67% (10/15) for stage II, and 78% (7/9) and 75% (9/12) for stage III patients. The sensitivity for association with stage IV (M1) disease was 86% (31/36).

Univariate logistic regression analysis showed the signature correlated well with the metastatic status. Multivariable analysis further confirmed that the risk score was the best predictor of ccRCC metastasis (OR=6.60, 95%Cl=2.54-17.13, p=0.0001), compared to tumor sizes or stage alone (Table 2).

In our test cohort, there were 109 patients with primary tumors 4 cm or smaller, comprising 41% (109/265) of this cohort population. Among these, 22 (22/109; 20%) samples were predicted as high risk, 64 (64/109, 59%) as low risk, and 23 (23/109, 21%) as equivocal. Given the clinical outcome, in this limited cohort, the sensitivity and specificity to stratify high and low risk patients with ccRCC 4 cm or smaller were 60% and 77%, respectively.

3.3. miRNA Signature Correlates with Cancer-Specific Survivals

We then examined the correlation between the signature and patient’s prognosis. For the 209 cases with high or low risk, the univariate Cox regression tests showed the signature to be highly associated with cancer-specific mortality in ccRCC patients (Table 3). Multivariable Cox regression analysis confirmed that patients predicted as high-risk had worse cancer-specific survival (HR=3.06, 95%Cl 1.49-6.93, p<0.005) than those with low risk (Table 3 and Figure 3). Stage IV disease and large tumor size (>10 cm) were the only additional significant prognostic factors (p<0.01).

In the test cohort, 78 of 265 (29%) cases presented with concurrent or subsequent metastasis. Within metastatic cases. 53 (68%) of those were classified as high risk, 13 (17%) as low risk, and 12 (15%) as equivocal. Interestingly, univariate Cox regression analysis showed that those cases predicted as high risk had worse cancer-specific survival than those predicted as low risk or equivocal (p<0.05) (Table 4 and Figure 4).

3.4. Biological Significance of the 4-miRNA Signature

To examine the functional significance of the four miRNAs involved in ccRCC metastasis, we applied a Context Likelihood of Relatedness (CLR) network modeling algorithm to generate pair-wise measures of associations between miRNAs and mRNAs based on Mutual Information (MI) through calculation of the entropy [22]. Using 540 cases with matching miRNA-seq and mRNA-seq data from the TCGA database (https://tcga-data.nci.nih.gov/tcga/), we identified all mRNAs significantly associated with the 4 miRNAs in the signature (Table A1), with positive correlation with miR-130b-3p and miR-199b-5p, and negative correlation with miR-10b-5p and miR-139-5p (p<0.05). Interestingly, Ingenuity Pathway Analysis showed that many of these genes are related to angiogenesis, cell migration, cell survival, and invasion, which are known to be related to metastatic process (Table A2).

4. Discussion

We rebuilt a ccRCC metastasis prognostic miRNA signature specifically for clinically available FFPE ccRCC specimens. The signature can stratify primary ccRCC patients into 3 groups: high and low risk for metastasis and equivocal. Using the risk score, clinicians can predict ccRCC metastasis and patient prognosis, but for those cases categorized as equivocal, clinicians need to correlate with other clinical manifestations to make treatment decisions.

Adjuvant therapies using agents for metastatic disease after resection of localized cancer are aimed to clear micrometastasis and benefit cancer patients [23]. An important focus of RCC clinical research is to develop effective adjuvant therapies to prevent or minimize ccRCC recurrence/metastasis and improve survival in high risk patients [9]. Currently more efforts have focused on transposing the new targeted agents into the adjuvant setting for localized RCC and developing several new clinical trials [23]. One of the major applications of this signature is to discover a biomarker to reliably identify high risk ccRCC patients for proper adjuvant therapies. Tumor size/stage correlates well with but cannot specifically predict metastasis in an individual patient. In addition to exposing patients to unnecessary and toxic treatment, overtreating low risk patients may also dramatically dilute the positive effect of the adjuvant agent(s) and falsely lead to failure of the clinical trials. Occasionally, clinical trials fail or have suboptimal results due to poor patient selection, so we look forward to using this signature to reclassify the patients’ risk status for reevaluation of the treatment effects. In a proper setting, this signature could play a role of “drug saver”.

Small renal masses (SRMs) are considered tumors 4 cm or smaller. With advanced imaging technology, 85% of asymptomatic RCCs were found to be SRMs [24]. Patients with SRMs diagnosed as ccRCC are currently managed by mainstay partial/radical nephrectomy, ablative therapy (AT) if the patients are elderly or not candidates for surgery, or active surveillance (AS) if the tumors are likely indolent [25,26]. However, only about 60% of SRMs are indolent and the other 20% are potentially aggressive including 5% with future metastasis [25]. Partly due to tumor biology, even less aggressive AT have a recurrence rate approximately of 10% and mortality rate of 5%, which were slightly higher than conventional nephrectomy [27,28,29,30]. Therefore, it will be very helpful if we could identify aggressive tumor using SRM biopsy specimens and exclude those high risk patients for AS or possibly AT. Although there were limited cases in the test cohort, in the group of patients with SRMs without information of tumor involvement in lymph nodes or distant organs, the signature predicted tumor metastasis with the sensitivity of 60% and specificity of 77%. With further validation studies, we anticipate this miRNA signature can identify high risk patients in biopsy SRM tissue to guide proper treatments.

In our study, among metastatic patients, those who were predicted to have high risk had significantly worse cancer-specific survival than those predicted not (low risk or equivocal) (Table 4 and Figure 4), indicating biologic subgroups of ccRCCs with metastasis. This feature makes it possible to differentiate mRCC patients with different prognosis for appropriate aggressive or palliative treatments.

Bioinformatics analysis with TCGA data has shown that the 4 miRNAs are significantly associated with genes involved in cell movement, development and differentiation. These specifically include cell migration and invasion and angiogenesis, which are characteristic of metastatic cancer cells. Further characterization of these genes/pathways may further broaden our understanding of the mechanisms of RCC metastasis. Meanwhile, any available therapeutic agents targeting these gene products for tumors other than ccRCC could be considered as potential targeted therapies for high risk patients.

5. Conclusions

We developed and validated an FFPE sample-specific, qRT-PCR-based molecular assay and risk score formula to predict ccRCC metastasis and prognosis. This test may help to identify high risk patients for clinical trials of adjuvant therapies after nephrectomy, examine SRM biopsy specimens to exclude high risk patients for AS, and differentiate mRCC patients with different prognosis for further aggressive or palliative treatment.

6. Patents

Part of this reported work has been patented by Beckman Research Institute of City of Hope (Patent No. US 9,771,619 B2; Date of Patent Sept. 26, 2017)