| Abstract |
Background: Lung cancer is the most lethal malignant neoplasm worldwide, with non-small cell lung cancer (NSCLC) accounting for approximately 85% of the cases. More than half of NSCLC patients are diagnosed at an incurable stage. In the era of targeted and contemporary immunomodulatory therapies, cytotoxic chemotherapy retains an integral role in the management of advanced-stage/metastatic NSCLC, especially for patients lacking access to innovative anticancer drugs due to ineligibility and/or unaffordability. Moreover, as the addition of anti-EGFR agents, such as necitumumab, or immune-checkpoint inhibitors, such as pembrolizumab, to classical cytotoxic therapy have become novel treatment options in the first-line setting, customisation of chemotherapy would further improve the clinical effect of such combination strategies. The antimetabolite gemcitabine holds an essential role in the first-line setting either as a component of platinum-based regimens in eligible patients or, alternatively, combined with non-platinum compounds, and as single-agent in unfit patients. Given that standard gemcitabine-based chemotherapy has reached an efficacy plateau, the identification and clinical implementation of appropriate predictive biomarkers would help optimizing upfront treatment. Based on limited, predominantly retrospective clinical data, tumoural transcriptional levels of RRM1, RRM2, CDA, dCK and hENT1 genes -all being involved in the intracellular activity and metabolism of gemcitabine- are potential biomarkers for the drug’s efficacy in advanced-stage NSCLC. At the same time, BRCA1 mRNA expression is a candidate molecular predictor of tumour susceptibility to both platinum compounds and anti-mitotic chemotherapeutic agents, such as taxanes and vinorelbine, that are commonly combined with gemcitabine in the first-line setting of NSCLC. Aim: We explored the potential predictive value of the combined transcriptional patterns of an expanded panel including all five aforementioned gemcitabine sensitivity-related genes for clinical outcome to standard gemcitabine-based chemotherapy. Individual mRNA expression of the five studied genes was dichotomised by the corresponding median value into “high” versus “low”. The categorical trasnscription patterns of the potentially unfavourable genes CDA, RRM1 and RRM2, and the potentially favourable genes dCK and hENT1 were combined within each of the two groups, thus classifying the respective gene co-expression into “concordant high”, “concordant low” and “discordant/mixed”. In turn, the co-expression patterns of the favourable and unfavourable genes were used to generate a five-level, ordinal classification of tumoural gemcitabine sensitivity (5L-GSC), ranging from “very low” to “very high”. At the same time, we assessed the potential predictive value of BRCA1 mRNA expression (dichotomized as “high” versus “low”) separately in the two subgroups of patients treated with gemcitabine in combination with either platinum compounds or antimitotic drugs. In addition to their individual effects, we investigated the potential interaction of the two studied biomarkers (5L-GSC and BRCA1 expression) on clinical outcome measures. Materials and Methods: We retrospectively analyzed each gene’s relative mRNA expression by quantitative, real-time polymerase chain reaction (RT-qPCR) in microdissected, formalin-fixed, paraffin-embedded (FFPE) primary-tumour specimens from 219 chemonaïve patients with histologically confirmed stage IV or recurrent NSCLC unamenable to local therapy, treated with gemcitabine -either as monotherapy or combined with platinum compounds, taxanes, vinorelbine or pemetrexed- in the context of 10 phase II, III and IV clinical trials conducted by the Hellenic Oncology Research Group and the Hellenic Cooperative Oncology Group, as previously described. Tumour specimens were collected for central laboratory testing, while clinicopathological data were retrieved from the corresponding trial databases. Both were obtained with the patients’ signed informed consent. The study was approved by the Ethics and Scientific Committee of the University General Hospital of Heraklion. (FFPE) specimens were pathologically reviewed for validity and adequate cellularity, while tumour cells were procured using a piezoelectric microdissector. Τoumoral RNA was extracted by the trizol LS method, treated with DNase and assessed for purity at 260/280 nm with NanoDrop-1000 Spectrophotometer. This was followed by complementary DNA (cDNA) synthesis using SuperScript III reverse transcriptase. TaqMan-based, relative RT-qPCR analysis was performed on ABI Prism 7900HT Sequence Detection System with appropriate non-template and non-reverse-transcribed RNA controls. Gene-specific, intron-spanning forward/reverse primer and probe sets were designed based on the NCBI Reference Sequence database and using Primer Express 2.0 Software. PCR experiments were run in triplicate for each target gene, while cDNA input corresponding to 15 ng RNA in a 12.5 μl final volume was used per reaction. The mRNA relative expression levels of each gene were calculated as a continuous variable by the 2-ΔΔCt method, using β-actin as a normalizing reference gene and commercial RNA as quantification calibrators. The target threshold cycle values were acceptable provided that the standard deviation between triplicates was below 0.3, according to the manufacturer’s suggestions. Molecular analysis was performed blinded to the clinical data. Statistics: The expectation-maximization approach was employed to impute missing transcript values of the dataset. Overall response rate (ORR), i.e.,, percentage of patients achieving complete and partial response (CR/PR), and disease control rate (DCR), i.e.,, percentage of patients achieving objective response and disease stabilization (SD), were defined by the RECIST v1.0 criteria, while time to progression (TTP) and overall survival (OS) were calculated from treatment initiation to the first documented disease progression (PD) or death from any cause, respectively. Multivariate analysis of variance (MANOVA) with type IV sum of squares and Pillai's trace as criterion was used to determine potential effects of patients’ and tumoural characteristics, as well as their interactions on the mRNA levels of all six studied genes, following rankit-based transformation of the corresponding values to achieve distribution normality. The Spearman’s rank coefficient test was used for pairwise correlation analysis of the gene expression profiles. The potential associations of BRCA1 expression, defined as a two level categorical variable, and 5L-GSC with ORR and DCR were tested univariably by the chi-square test with exact probabilities, as appropriate, in the corresponding groups of patients. The Kaplan–Meier method was used for plotting the TTP and OS curves, and estimating the corresponding median values, while the log-rank test evaluated their univariable association with the two biomarkers, separately, in the corresponding patient groups. In the total study population (219 patients), multivariable binary logistic regressions using a simultaneous (forced entry) model were carried out to calculate the odds ratios (OR) and 95% confidence intervals (CIs) for ORR defined as a dichotomous variable (i.e.,, objective response versus nonresponse) and DCR across the five levels of gemcitabine-sensitivity, with adjustment for several potential confounding factors, including performance status (PS), squamous (SCC) versus non-squamous (non-SCC) cell histology, addition of biological agents, and gemcitabine-based regimen grouped into monotherapy and combinations with either platinum compounds, antimitotic drugs (taxanes and vinorelbine) or antimetabolites (pemetrexed). Likewise, simultaneous, multivariable Cox proportional-hazards regression models assessed the potential independent predictive significance of 5L-GSC for PD and death, while estimating the corresponding adjusted hazard ratios (HR) and 95%CIs for each chemosensitivity level. Bias-corrected and accelerated (Bca) bootstrapping with 1000 replications was employed as an internal validation method for the multivariable regression-analysis results, while potential multicollinearity of the model predictors was assessed by the tolerance and variance inflation factor values. Furthermore, categorical regression with optimal scaling transformation (CATREG), combined with elastic-net regularization and the 0.632 bootstrap method with 200 iterations were applied to identify the most parsimonious model with the highest prediction accuracy for overall tumour response, the latter defined as a three-category variable (i.e.,, CR/PR versus SD versus PD). A similar multivariate statistical approach and validation methodology were used to explore the correlation of BRCA1 mRNA expression with clinical outcome in the two above-mentioned chemotherapy subgroups, i.e, in the 76 patients treated with gemcitabine-plus-platinum doublets, as well as in the 122 patients treated with gemcitabine-plus-antimitotic regimens. For all statistical analyses, the two-tailed level of significance was set at p-value ≤ 5%. Results: PCR amplification was successful for all six genes under study in approximately 95% of tumour specimens. Patients were predominantly male (88%), with a median age of 66 years and a good performance status (ECOG PS 0-1, 91%). Around 69% of participants had stage IV-M1b and 22% stage IV-M1a disease (7th UICC TNM edition). Non-SCCs comprised ~60% of the tumours. In more than half of the patients, gemcitabine was combined with anti-mitotics, whereas more than one-third received gemcitabine/platinum doublets. Biological agents, such as bevacizumab and bortezomib, were added to first-line regimens in 8% of cases. Data on subsequent therapy were available for 85% of patients, of whom 35% received no other treatment. Among the clinico-histopathological features studied, only PS emerged as an independent prognostic factor for clinical outcome, in particular TTP and OS. Although no statistically significant correlations were revealed between any of the clinic-histopathological features and the linear combination of the transcriptional levels of the six studied genes (i.e, as a whole), a significant but limited and disordinal (cross-over) interaction effect of gender and histological type was observed on CDA mRNA levels, with a differential expression between squamous and nonsquamous carcinomas in male patients. Significant and moderate (i.e, r-value ≥ 0.3), positive, pairwise expression correlations were seen between the unfavourable genes RRM1 and RRM2 and between the favourable genes dCK and hENT1. Furthermore, there were significant positive pairwise correlations between the mRNA levels of both former genes and the two latter, indicating a simultaneous activity of the two functionally competitive gene groups. More specifically, dCK mRNA expression correlated moderately with RRM1 and strongly with RRM2 expression, while hENT1 mRNA expression correlated weakly to moderately with the expression of RRM1 and RRM2, respectively. Besides, BRCA1 mRNA levels were positively and moderately correlated with those of RRM1, RRM2 and dCK. 5L-GSC was independently associated with all four clinical endpoints -ORR (p=0.03), DCR (p=0.004), TTP (p<0.001) and OS (p<0.001)- in multivariable regression models adjusting for other factors such as tumour histology and chemotherapy regimen, and after excluding collinearity among the predictor variables. Furthermore, the magnitude of chemotherapy efficacy increased progressively across the five classification groups and in accordance with the scale level of gemcitabine-susceptibility, with the values of outcome measures for the moderate-chemosensitivity group resembling those for the total study population. These, in turn, were as expected by current standards for the first-line chemotherapy setting. The multivariable regression analysis results regarding the effect of 5L-GSC on clinical outcome retained their statistical significance after Bca bootstrapping validation, while multicollinearity was excluded. Penalised, optimally-scaled, categorical-regression modelling of overall response identified 5L-GSC as the most stable predictor, and the corresponding expected prediction error was estimated at 0.615 (±0.033). BRCA1 mRNA expression defined as a two-level categorical variable was significantly correlated with all outcome measures (ORR, DCR, TTR and OS) in multivariable regression analyses and independently of 5L-GSC in both chemotherapy subgroups, but in a differential manner. In particular, whereas low BRCA1 expression had a favourable impact on clinical outcome in the subgroup of patients treated with gemcitabine-plus-platinum combinations, its corresponding effect among those receiving gemcitabine and antimitotic agents was detrimental. As in the case of 5L-GSC, the above-mentioned results remained statistically significant by internal validation (Bca bootstrap) and were unaffected by multicollinearity. No significant differences regarding the frequency or type of second-line therapy were seen among the levels/categories of either of the two biomarkers. A statistically significant and favourable, synergistic, interaction effect was observed between low BRCA1 expression and 5L-GSC status on OS, as reflected by a progressive increase in the latter’s median from 4.9 months among patients with tumours of “very low-gemcitabine sensitivity” and low BRCA1 expression to 41.6 months among those with tumours of “ very high-gemcitabine sensitivity” and low BRCA1 expression. Lastly, among the 117 patients for whom data on the type of disease response to 2nd-line therapy were available, no statistically significant correlation was observed between clinical outcome and either 5L-GSC or BRCA1 mRNA expression. Conclusions: Combined transcriptional-expression profiling of CDA, RRM1, RRM2, dCK and hENT1, along with BRCA1 mRNA expression in the primary tumour are promising biomarkers that could serve as predictive tools for guiding front-line gemcitabine-based chemotherapy in advanced-stage NSCLC. Their potential clinical utility -either individually and/or in combination- merits external validation and prospective evaluation.
|
Collections
