A Three-microRNA Signature Predicts Responses to Platinum-Based Doublet Chemotherapy in Patients with Lung Adenocarcinoma

Lung adenocarcinoma (LADC) is the most common type of non–small cell lung cancer (NSCLC) and is a leading cause of cancer mortality worldwide (1). Surgical resection is the best curative treatment for NSCLC; however, patients that experience recurrence after surgery and those with advanced disease receive chemotherapy to slow tumor growth and improve survival. LADCs harboring an EGFR mutation or an ALK fusion are primarily treated with specific tyrosine kinase inhibitors (TKI), with response rates of approximately 60% (2, 3). Other oncogene aberrations, such as BRAF, HER2, RET, and ROS1, are also targeted by specific TKIs (4–6), but a major barrier to curative treatment of LADC using TKIs is innate and acquired drug resistance (7, 8). Furthermore, more than 60% of USA/European and more than 30% of Japanese LADC cases do not harbor the oncogene aberrations listed above (9, 10). Such resistant and oncogene-negative LADC cases are treated with chemotherapy. The standard regimens comprise platinum-based doublets, that is, a combination of platinum and another agent; the drugs paired with platinum (cisplatin or carboplatin) include microtubule-targeted agents (paclitaxel, docetaxel, or vinorelbine) and DNA-damaging agents (gemcitabine or irinotecan). A series of trials in unselected patients revealed that the efficacy of each combination is similar, with response rates of 30% to 40% (11–13); thus, identifying biomarkers that can discriminate between patients that will respond to platinum-based doublet chemotherapy and those who may not before treatment will help improving the efficacy of chemotherapy for LADC.

microRNAs (miRNA) are small noncoding RNAs that posttranscriptionally regulate the translation of target genes; these miRNAs show altered expression in a variety of cancers and can modify the malignant properties of cancer cells, including the response to DNA damage (14–16). In fact, functional studies in LADC cell lines have identified a number of miRNAs that modulate sensitivity to platinum-based agents (17–23). Furthermore, miRNAs are stable in formalin-fixed paraffin-embedded (FFPE) tissues, that is, materials used for daily pathologic diagnosis (24, 25). Indeed, intratumor miRNA expression is a promising prognostic marker in patients that have undergone surgical resection (26–31), but few studies have examined the utility of miRNAs as a predictor of chemotherapeutic responses (32). To the best of our knowledge, only two such studies have been reported: one shows that a two-miRNA signature (miR149 and miR375) is associated with responses to platinum-based chemotherapy in NSCLC (n = 38), and the other shows that miR92a-2* expression is associated with chemoresistance in small-cell lung cancer (n = 34; refs. 33, 34).

Here, we investigated the utility of intratumor miRNAs as a biomarker for predicting responses to platinum-based doublet chemotherapy. We examined the expression of miRNAs in 103 surgically resected specimens obtained from patients who received platinum-based chemotherapy upon LADC recurrence to ascertain whether miRNA expression in primary tumors could predict responses to platinum-based doublet chemotherapy in patients who experienced LADC recurrence after surgery. First, a two-step screening process involving 904 miRNAs was performed to identify an miRNA signature with predictive value. The first step was performed using a test cohort comprising 40 frozen tumor samples from which RNAs were isolated, and the second step was performed using a validation cohort comprising 63 cases, for which RNAs from FFPE tissues were available. We identified a three-miRNA signature that predicted responses to platinum-based doublet chemotherapy with an accuracy of >75%.

Of note, 103 surgically resected LADC tissues were examined in the present study (Fig. 1A). Briefly, 643 Japanese patients with NSCLC received platinum-based doublet chemotherapy at the National Cancer Center Hospital (NCCH; Tokyo, Japan), between 2000 and 2008, and the therapeutic response was evaluated using the Response Evaluation Criteria In Solid Tumors (RECIST) guidelines (35). None of the patients had received prior treatment with platinum-based chemotherapy. Of the 643 cases, 118 were recurrent cases that had undergone surgical resection at NCCH, and all were pathologically diagnosed with adenocarcinoma. Tumor tissues for RNA extraction were available for 103 of 118 cases; these cases were examined in the present study. Information regarding age, gender, pathologic TNM stage (the 7th classification), smoking habits, postoperative chemotherapy regimens and responses to platinum doublet therapy, and performance status (PS) were retrospectively collected. RNAs isolated from fresh-frozen tissues were available for 40 of 103 cases; these were defined as the test cohort. The RNAs from the test cohort were subjected to miRNA microarray analysis followed by verification by quantitative reverse transcriptase-PCR (qRT-PCR) analysis. RNAs from FFPE tissues were available for the remaining 63 cases; these were defined as the validation cohort. In addition, patients were classified into two categories according to RECIST guidelines: those that responded to platinum-based doublet chemotherapy [complete response (CR) or partial response (PR)] and those that did not [stable disease (SD) or progressive disease (PD)].

RNA was extracted from snap-frozen tissues (test cohort) using TRizol reagent (Thermo Fisher Scientific). The quality and quantity of the RNAs were examined using a Bioanalyzer (Agilent). All RNA samples showed an RNA integrity number RIN >6.0; therefore, they were subjected to microarray analysis. For the validation cohort, RNA was isolated from three unstained FFPE sections (5-μm thick). The area of the carcinoma in the three unstained sections was outlined by referring to a sequential section that was stained with hematoxylin and eosin. Each marked area was macrodissected using a sterile disposable scalpel and RNA was isolated using the RecoverAll Total Nucleic Acid Isolation Kit (Ambion). Total RNA was quantified using a NanoDrop ND-1000 spectrometer (Thermo Fisher Scientific). The optical density (OD) 260/280 and OD 260/230 ratios were used for quality control.

The Human miRNA Microarray Kit release 14 (8 × 15 K; Agilent Technologies), covering 904 miRNAs, was used to screen for miRNAs in the test cohort samples (n = 40). Data were normalized and analyzed using GeneSpring GX software (version 12.5; TOMY Digital Biology). The fold change in expression was defined as the ratio of expression in responders to that in nonresponders. Normalized and raw expression data were deposited in the Gene Expression Omnibus at the National Center for Biotechnology Information (GSE56264).

All 40 test cohort samples were also screened for oncogene fusions (EML4– and KIF5B–ALK, KIF5B– and CCDC6–RET; and CD74–, EZR–, and SLC34A2–ROS1) by reverse transcription-PCR as previously described (4, 36). Genomic DNA was extracted from fresh or frozen samples from all 103 subjects using a QIAamp DNA Mini Kit (QIAGEN) and then analyzed for EGFR, KRAS, BRAF, and HER2 hot spot mutations using the high resolution melting method (37, 38).

qRT-PCR of mature miRNA was performed using TaqMan MicroRNA assays (Thermo Fisher Scientific) and the 7900 HT Fast Real-Time PCR System (Thermo Fisher Scientific). cDNA was synthesized using miRNA-specific primers and a TaqMan MicroRNA Reverse Transcription Kit (Thermo Fisher Scientific). RNA (40 ng) was reverse transcribed in a 20 μL reaction containing gene-specific RT probes. All assays were performed in triplicate and investigators were blinded to the clinical outcome. All TaqMan probes were purchased from Thermo Fisher Scientific: hsa-miR135a-3p (ID 002232), hsa-miR196b-5p (ID 002215), hsa-miR1181 (Assay ID 241045_mat), hsa-miR31-5p (ID 002279), hsa-miR31-3p (ID 002113), hsa-miR1290 (ID 002863), hsa-miR598 (ID 001988), hsa-miR1 (ID 002222), hsa-miR144-5p (ID 002148), hsa-miR628-5p (ID 002433), hsa-miR449a (ID 001030), and hsa-miR34b-3p (ID 002102). RNU66 (ID 001002) was used as a normalization control. Relative expression of miRNAs was calculated using RQ manager 1.2 (Thermo Fisher Scientific).

Differences in miRNA expression levels between responders and nonresponders were tested by the Mann–Whitney U test using Graphpad Prism v5.0 (Graphpad Software Inc). Spearman correlation analysis was used to examine the correlation between microarray and qRT-PCR data (Graphpad Prism v5.0). Linear discriminant analysis was performed for each cohort to distinguish responders from nonresponders (JMP 10 software; SAS Institute) based on miRNA expression (i.e., the miRNA signature). Continuous expression values for a single miRNA or for plural miRNAs, that is, ΔC_t values obtained by qRT-PCR against RNU66, were included as variables in the analysis. Receiver operating characteristic (ROC) curves were generated to evaluate response sensitivity and the area under the curve (AUC) was calculated (JMP 10). Principal component analysis (PCA) of the expression of three miRNAs (miR1290, miR196b, and miR135a*) was performed using JMP 10.

The aim of this study was to identify biomarkers in patients with metastatic LADC who relapsed following potential curative surgical resection. Therefore, surgically resected primary tumor tissues from 103 LADC patients who were treated with platinum-based doublet chemotherapy upon recurrence were selected for miRNA profiling (Fig. 1A). The cases were assigned to a test cohort (n = 40; RNAs from frozen tissue available) or a validation cohort (n = 63; RNAs from FFPE tissues available) according to the availability of tumor tissue samples. Patients in both cohorts were classified as responders (CR and PR) or nonresponders (SD and PD) to platinum-based doublet chemotherapy according to the RECIST criteria (Materials and Methods; Supplementary Table S1). In this study, platinum-based doublet chemotherapy includes several different regimens. The cohorts were similar in terms of clinicopathologic characteristics such as age, gender, smoking habits, pathologic stage, representative oncogene mutations, therapeutic regimen, and therapeutic response (Table 1). The samples were subjected to a two-step screening procedure to identify miRNAs whose expression is associated with responses to platinum-based doublet chemotherapy (Fig. 1B).

First, we used microarray analysis to identify miRNAs in test cohort samples that were differentially expressed in LADC tissues from responders and nonresponders to platinum-based doublet chemotherapy. Fifty-nine miRNAs were identified (P < 0.05; Welch t test; Supplementary Table S2). Of these, 28 were upregulated in responders and 31 were downregulated.

Next, to identify the limited number of miRNAs that can be used to deduce responsiveness in the clinic, we searched for miRNAs showing highly differential expression between responders and nonresponders. Twelve miRNAs (miR135a*, miR196b, miR1181, miR31, miR31*, miR1290, miR598, miR1, miR144*, miR628-5p, miR449a, and miR34b) showing a >5-fold change in expression were identified as potential candidates and investigated further (Fig. 2A; Supplementary Table S2). The expression of these 12 miRNAs was reanalyzed by qRT-PCR using the same RNA samples used in the microarray experiments. A good agreement between the qRT-PCR (ΔC_t value) and microarray (log₂ signal) data (as indicated by the Pearson correlation coefficient) was observed for 10 of the 12 miRNAs (excluding miR1181 and miR598; Supplementary Fig. S1). qRT-PCR identified three miRNAs (miR1290, miR196b, and miR135a*) as differentially expressed in responders and nonresponders (P < 0.001, P < 0.001, and P < 0.008, respectively; Fig. 2B; Supplementary Table S3). The fold changes observed in the qRT-PCR experiment were lower than those observed in the microarray experiment. This may be due to the higher sensitivity of the former. Several samples that gave no significant signal (calculated as zero) in the microarray experiment yielded ΔC_t values in the qRT-PCR experiment (Supplementary Fig. S1), leading to a reduction in the fold change values (Supplementary Table S3).

We next used linear discriminant analysis to examine these three miRNAs for their potential to discriminate responders from nonresponders. For this, continuous expression values for a single miRNA or a combination of two or three miRNAs, that is, ΔC_t values obtained by qRT-PCR, were included as variables in the analysis. ROC curves were plotted to examine both sensitivity and specificity. The results showed that a combination of all three miRNAs provided the best discrimination, with an AUC of 0.893 (Fig. 2C). This combination of three miRNAs is, henceforth, referred to as the “three-miRNA signature.”

To validate the findings in the test cohort, we next examined the expression of the three-miRNA signature in the validation cohort (Supplementary Fig. S2 and Supplementary Table S3). We then performed a linear discriminant analysis to evaluate the potential of the three-miRNA signature as a biomarker. The mean expression levels of the three miRNAs were higher in responders than in nonresponders in the validation cohort, although the difference in the expression of miR135a* did not reach a statistical significance. However, a combination of all three miRNAs again was better able to distinguish responders from nonresponders (AUC = 0.837) than a single miRNA or a combination of two miRNAs (Supplementary Fig. S3).

PCA and support vector machine (SVM) analysis showed a predictive response of 37.5% for the test cohort containing 40% responders and a predictive response of 31.7% for the validation cohort containing 28.6% responders. The signature predicts responders and nonresponders in the test and validation cohorts with an accuracy of 82.5% and 77.8%, respectively (Fig. 3). The sensitivity, specificity, and positive and negative predictive values were similar for both cohorts (Supplementary Table S4). There was no significant difference in the clinical characteristics between true responders/nonresponders and predicted responders/nonresponders (Supplementary Table S1). Taken together, these results show that the predictive ability of the three-miRNA signature was confirmed in the independent validation cohort, and that the signature is still predictive even if archive FFPE tissues are used for analysis. Specificity and negative predictive values greater than sensitivity and positive predictive values suggest that the three-miRNA signature predicts nonresponders better than responders (Supplementary Table S4).

We previously showed in the same study population (i.e., the 640 cases shown in Fig. 1A) that the Arg72Pro polymorphism in the TP53 gene in noncancerous (germline) DNA is associated with responses to platinum-based doublet chemotherapy: The response rate is higher in those harboring the TP53-72Pro polymorphism (35). Therefore, we combined the three-miRNA signature with the TP53-Arg72Pro genotype data to ascertain whether the predictive power of the miRNA signature was enhanced. We dichotomized the study cohorts into two subgroups (patients with the TP53-72Pro allele and those without) and examined the predictive accuracy of the three-miRNA signature. We found that the predictive accuracy marginally improved in both the test (85.0%) and validation (82.5%) cohorts (Fig. 1B).

Driver oncogene aberrations in LADC are a critical factor that determines the therapeutic strategy for each patient. In addition, such aberrations are associated with clinical characteristics, as represented by the predominance of EGFR mutation in females and never-smokers. Therefore, we next addressed whether the ability of the three-miRNA signature to predict responses to chemotherapy was affected by driver oncogene alterations or clinical characteristics. The three-miRNA signature failed to predict responses in 21 of 103 cases (21%). The test and validation cohorts contained seven (one with an EGFR mutation, one with a HER2 mutation, and five aberration-negative cases) and 14 (six with EGFR mutations, one with a KRAS mutation, and seven aberration-negative cases) nonpredicted cases, respectively (Fig. 4). The nonprediction rate was 15% (7/47 cases) for patients harboring the EGFR mutation, 10% (1/10) for those harboring the KRAS mutation, 33% (1/3) for those harboring the HER2 mutation, and 32% (12/38) for aberration-negative cases; therefore, there was no significant correlation between driver gene status and nonprediction. Similarly, there was no significant association between clinical factors and nonprediction; thus, the three-miRNA signature may be a useful biomarker for predicting the responses of patients with LADC to chemotherapy, irrespective of driver oncogene aberrations and clinical characteristics.

Here, we performed miRNA expression profiling of patients who initially underwent surgical resection for primary LADC and were then treated with platinum-based doublet chemotherapy upon recurrence. We identified a three-miRNA signature (miR1290, miR196b, and miR135a*) that predicts whether patients with recurring LADC respond to and, therefore, will benefit from platinum-based doublet chemotherapy. Even patients with LADC harboring druggable oncogene aberrations (that are resistant to treatment with TKIs) may be treated with platinum-based doublet chemotherapy; therefore, platinum-based doublet chemotherapy is a major therapeutic strategy for almost all patients with LADC (39–41). Personalized therapy, in which a drug with the greatest chance of eliciting a response (i.e., tumor shrinkage) is chosen specifically for each patient, is the first critical step toward improved prognosis for patients with LADC with advanced and recurrent disease; indeed, clinical trials examining the effect of new drugs on NSCLC have set improved response rates as their primary endpoint (10); thus, response to treatment according to the RECIST criteria rather than survival was the outcome measure selected for the present study. The three-miRNA signature will facilitate personalized therapy for LADC and will include platinum-based doublet therapy as an option.

Here, we examined the three-miRNA signature of primary tumors to predict the responsiveness of recurrent tumors. It is noteworthy that the biologic characteristics of recurrent tumors are not the same as those of primary tumors due to tumor cell heterogeneity in the primary lesions and the accumulation of additional genetic/epigenetic changes during progression. Thus, at present our findings are applicable to the treatment of recurrent tumors for which corresponding primary tumor tissue samples are available. The finding that the three-miRNA signature is predictive in archived primary tumor tissues is an advantage; patients are spared the additional burden of further tissue sampling for genetic analysis. However, it is also worth analyzing recurrent tumors and inoperable advanced tumors to find out whether the three-miRNA signature is applicable to patients for whom archived surgical tissues are not available. The finding that the signature can be identified in archived FFPE tissues is also an advantage and will facilitate translation to the clinic.

We also examined the combination of the three-miRNA signature with the TP53-Arg72Pro polymorphism genotype to see whether this provided greater predictive accuracy, as blood cells used for genotyping polymorphisms are easily obtained from patients. However, the improvement was only marginal. Therefore, more polymorphisms associated with responses to platinum-based doublet therapy must be identified if we are to achieve any marked improvement over the three-miRNA signature alone. In addition, we used a fold change >5 as a criterion for identifying candidate miRNAs that are differentially expressed between responders and nonresponders. Using less stringent or other statistical criteria may lead to the identification of more miRNAs that are useful for prediction.

The three-miRNA signature predicted responses irrespective of the presence of driver oncogene aberrations. This is important when we consider that LADCs that have acquired resistance to specific TKIs are treated with platinum-based doublet chemotherapy. However, unfortunately, the present study cohort did not include samples from patients with EGFR- or ALK-positive LADC that received TKIs before platinum-based doublet chemotherapy. Such cases should be examined to address this issue.

This study has several limitations. First, it was retrospective in nature, so the ability of the three-miRNA signature to predict responses needs to be further validated using more samples. Here, different types of tumor tissue for the test and validation sets (bulk frozen and microdissected FFPE tissues, respectively), which contained patients that had undergone several different chemotherapeutic regimens, were subjected to analysis; therefore, we may have over- or underestimated predictive value. Thus, further studies that use a larger number of samples obtained according to a defined experimental procedure and take factors such as previous treatment regimen, disease stage, and PS into account are required. In addition, prospective studies, for example, studies using samples from patients treated with a single therapeutic regimen, and the analysis of primary and recurrent tumors and inoperable advanced tumors, should be conducted to confirm the utility of the three-microRNA signature. Second, although the three-miRNA signature was significantly associated with response to chemotherapy, differences in progression-free survival were only suggestive (Supplementary Fig. S4). We chose the response as the primary endpoint of efficacy to identify subgroups for which chemotherapy does work (35). However, treatment is continued after the failure of platinum-based doublet chemotherapy; therefore, clinical response alone would not be enough to improve the outcome. Third, the functional relevance of miR1290, miR196b, and miR135a* to the chemosensitivity of LADC remains unclear. Interestingly, a recent study shows that the expression of miR196b is upregulated in patients with rectal adenocarcinoma that respond to neoadjuvant chemoradiotherapy (capecitabine or 5-fluorouracil), which supports the findings of the present study (42). However, preliminary experiments examining the exogenous expression of the three miRNAs in LADC cell lines did not show increased sensitivity to a platinum agent, cisplatin (CDDP). Therefore, the direct or indirect effects of miRNAs on chemosensitivity should be further investigated.

No potential conflicts of interest were disclosed.

Conception and design: M. Saito, K. Shiraishi, C.C. Harris

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): K. Shiraishi, K. Matsumoto, H. Kunitoh, H. Nokihara, S.-I. Watanabe, K. Tsuta, C.C. Harris

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M. Saito, A.J. Schetter, H. Ogata-Kawata, N. Tsuchiya, K. Kumamoto

Writing, review, and/or revision of the manuscript: M. Saito, A.J. Schetter, H. Ogata-Kawata, H. Kunitoh, H. Nokihara, C.C. Harris, T. Kohno

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): K. Shiraishi, S. Takenoshita

Study supervision: J. Yokota, T. Kohno

The authors thank Yoko Shimada for technical assistance.

This work was supported in part by the Advanced Research for Medical Products Mining Program of the National Institute of Biomedical Innovation (NIBIO), Grants-in-Aid from the Ministry of Health, Labor, and Welfare for the Third Term Comprehensive 10 year Strategy for Cancer Control, and Grant-in-Aid from Japan Society for the Promotion of Science (JSPS) for Scientific Research (B). Grant-in-Aid for Young Scientists (B; 24790340) and Third-Term Comprehensive Control Research for Cancer (H25-Young Scientists-009).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.