The first two authors contributed equally to this article.
Predictive and prognostic roles of ribonucleotide reductase M1 in resectable pancreatic adenocarcinoma
Article first published online: 26 JUN 2012
Copyright © 2012 American Cancer Society
Volume 119, Issue 1, pages 173–181, 1 January 2013
How to Cite
Xie, H., Jiang, W., Jiang, J., Wang, Y., Kim, R., Liu, X. and Liu, X. (2013), Predictive and prognostic roles of ribonucleotide reductase M1 in resectable pancreatic adenocarcinoma. Cancer, 119: 173–181. doi: 10.1002/cncr.27715
- Issue published online: 17 DEC 2012
- Article first published online: 26 JUN 2012
- Manuscript Accepted: 30 MAY 2012
- Manuscript Revised: 27 MAY 2012
- Manuscript Received: 16 APR 2012
- pancreatic cancer;
- ribonucleotide reductase M1;
- adjuvant therapy;
Ribonucleotide reductase M1 (RRM1) is an important molecule in different types of cancer. The objective of this study was to evaluate the predictive roles of RRM1 in the survival of patients with resectable pancreatic adenocarcinoma who received treatment with gemcitabine or nongemcitabine adjuvant therapy.
In total, 122 patients underwent tumor resection for pancreatic adenocarcinoma at the authors' institution from October 1999 to December 2007. Total RNA was isolated from microdissected, paraffin-embedded tumors. RRM1 expression levels were measured using quantitative reverse transcriptase-polymerase chain reaction (QRT-PCR) and were dichotomized using recursive partitioning analysis. The Kaplan-Meier method was used to estimate overall survival and progression-free survival, and the predictive value of RRM1 expression on survival was examined using Cox proportional hazards regression.
RRM1 expression did not have significant prognostic value in the entire cohort regarding overall survival (P = .2) or progression-free survival (P = .7). In the subgroup of 44 patients who received adjuvant gemcitabine, patients who had low RRM1 expression had longer overall survival (median, 47.8 months vs 14.1 months; P = .005) and a trend toward longer progression-free survival (median not reached vs 12.9 months; P = .06). In contrast, in the subgroup of 35 patients who received nongemcitabine adjuvant therapy, patients who had high RRM1 expression had significantly longer overall survival (median, 41.9 months vs 19.8 months; P = .01) and progression-free survival (median, 70.0 months vs 11.8 months; P = .04). These results were confirmed in Cox proportional hazards multivariable analysis.
In patients with resectable pancreatic adenocarcinoma, low RRM1 expression in the tumor predicted an overall survival benefit of adjuvant gemcitabine; and high RRM1 expression predicted the survival benefit of nongemcitabine adjuvant therapy. Cancer 2013. © 2012 American Cancer Society.
Pancreatic ductal adenocarcinoma is the fourth leading cause of cancer deaths in the United States. Approximately 40,000 new cases are diagnosed every year, and the 5-year overall survival (OS) is only 4%.1 Patients with pancreatic adenocarcinoma often do not have symptoms until their disease reaches an advanced stage. Even for the 15% to 20% patients who have resectable tumors at initial presentation, the 5-year survival rate is only 18% to 24%.2 Many patients die from local recurrence or metastasis. Although adjuvant chemoradiation or chemotherapy with gemcitabine has been established as the standard of care,3 in a significant portion of patients, the disease still recurs. Therefore, identifying predictive markers of response to adjuvant treatment is essential not only to individualized patient care but also to the stratification of patients in future clinical trials.
It has been demonstrated that the substrate and allosteric regulator binding subunit (M1) of ribonucleotide reductase, RRM1, plays 2 important roles in various types of cancer. First, RRM1 is a cellular target for gemcitabine, which is a deoxycytidine analog commonly used as a chemotherapeutic agent for solid tumors. The inhibition of RRM1 has been implicated in tumor sensitivity toward gemcitabine in both preclinical and clinical settings. In vitro, RRM1 expression has been related inversely to gemcitabine sensitivity in various tumor cell lines.4, 5 Transfection of gemcitabine-resistant tumor cells with small-interfering RNA (siRNA) into RRM1 increased their sensitization to gemcitabine.6 Clinically, randomized controlled trials have demonstrated that low RRM1 mRNA levels predicted longer OS in patients with advanced nonsmall cell lung cancer (NSCLC) who received gemcitabine and cisplatin.7 However, unlike NSCLC, both the prognostic role and the predictive role of RRM1 in the response to gemcitabine among patients with pancreatic cancer, especially those have resectable tumors, have been controversial.8-10
Second, RRM1 is a tumor suppressor. Preclinical studies have demonstrated its involvement in the suppression of cancer cell proliferation, migration, and metastasis.4 Overexpression of RRM1 in mouse inhibited lung tumor formation and metastasis, presumably through phosphatase and tensin homolog (PTEN)-regulated pathways.11 In line with these findings, high RRM1 expression levels determined by either real-time reverse transcriptase-polymerase chain reaction (RT-PCR) or immunohistochemistry were associated with longer OS and progression-free survival (PFS) in patients with early stage NSCLC who underwent tumor resection only.12 However, to our knowledge, the role of RRM1 has not been studied in the setting of nongemcitabine-containing adjuvant therapy for patients with resectable pancreatic adenocarcinoma.
On the basis of the results from our pilot study, we hypothesized that low expression of RRM1 is predictive of adjuvant gemcitabine treatment benefit in patients with resectable pancreatic adenocarcinoma, whereas this does not hold true for nongemcitabine adjuvant therapy. In the current study, our objective was to determine whether RRM1 expression correlates with survival in patients with resectable pancreatic adenocarcinoma and in those who received adjuvant gemcitabine or nongemcitabine-containing adjuvant therapy.
MATERIALS AND METHODS
This was a retrospective study of 122 patients who underwent surgical resection for resectable pancreatic adenocarcinoma at a single institution from October 1999 to December 2007. The inclusion criteria included a diagnosis of pancreatic ductal adenocarcinoma and pathologic stage of T1-T3N0-N1M0. Exclusion criteria included unresectable or metastatic disease, R2 resection, ampullary carcinoma, and indolent pancreatic tumors, such as mucinous cystadenoma, mucinous cystadenocarcinoma, and islet cell tumor. The institutional review board at the Cleveland Clinic approved and monitored this study.
All 122 pancreatic adenocarcinoma tissue samples were either formalin-fixed (104 patients; 85.2%) or Hollande fixative (18 patients, 14.8%), paraffin-embedded tissues. We reviewed the tumor-containing hematoxylin and eosin-stained slides and microdissected the area with the greatest tumor density (range, 30%-80%). Then, we isolated total RNA from microdissected tumor samples with 10-μm tissue sections according to a modified protocol using a high pure RNA paraffin kit (Roche Applied Sciences, Indianapolis, Ind). After deparaffinization of tissue sections, the pellet was dried at 55°C for 10 minutes and then suspended in 100 μL lysis buffer, 80 μL proteinase K, and 16 μL 10% sodium dodecyl sulfate. The mixture was swirled and incubated in a thermomixer at 55°C and 400 rpm for 3 hours. The RNA samples were quantified by optical density 260/280 readings using a spectrophotometer, diluted to a concentration of 50 ng/μL, and stored at −80°C in RNase-free water.
We used a 100-ng RNA sample along with 2 reference genes (hydroxymethylbilane synthase [HMBS] and ribosomal protein L13A [RPL13A]) in a 1-step, multiplex, quantitative RT-PCR (QRT-PCR) assay, which was performed at Veridex LLC (San Diego, Calif). PCR primers and probes were designed to span an intron to prevent amplification of residual genomic DNA. Their sequences for RRM1, HMBS, and RPL13A from 5′ to 3′ were as follows: RRM1 forward, CACATCAGAACACACATACGACTTTAA; RRM1 reverse, TTCAAGTTTCGGACAACGACTTT; RRM1 probe, FAM-AGTTGGCTGAAGTCAC-MGB; HMBS forward, CCTGCCCACTGTGCTTCCT; HMBS reverse, GGTTTTCCCGCTTGCAGAT; HMBS probe, FAM-CTGGCTTCACCATCG-BHQ; RPL13A forward, CGGAAGAAGAAACAGCTCATGA; RPL13A reverse, CCTCTGTGTATTTGTCAATTTTCTTCTC; RPL13A probe, FAM-CGGAAACAGGCCGAGAA-BHQ. We optimized all primers and probes to the same amplification efficiency. Then, we used the ABI 7900HT sequence detection system (Applied Biosystems, Frenso, Calif) with a 384-well block format and a 10 μL reaction volume to conduct PCR amplifications. The RNA samples, 4 μmol/L of each primer, and 2.5 μmol/L of each probe were incubated at 48°C for 30 minutes for reverse transcription. This was followed by Amplitaq activation at 95°C for 10 minutes and then 40 cycles at 95°C for 15 seconds for denaturing and at 60°C for 60 seconds for annealing and extension. We calculated ΔCt, defined as the difference between the crossover threshold (Ct) of the target gene and the Ct average of HMBS and RPL13A for each sample.13 In this study, a larger ΔCt value indicates lower expression of the targeted genes.
Statistical analysis for this retrospective study was primarily descriptive in nature. Thus, we did not perform any sample size calculation or power estimation. We summarized categorical data as frequency counts and percentages and continuous measures as means, standard deviations, standard errors, medians, and ranges. All statistical tests were 2-sided. OS was calculated from the date of tumor resection to the date of death. PFS was calculated from the date of tumor resection to the date of either disease progression or death. Time-to-event data were summarized using the Kaplan-Meier method with surviving patients censored at the date of last follow-up. Different groups of patients were evaluated using log-rank tests. We used Cox proportional hazards model for multivariate analyses of potential prognostic indicators in a stepwise fashion. The criterion for entry variables was P < .1, and the retention criterion was P < .05. All Cox proportional hazards regression results are presented as hazard ratios (HRs), 95% confidence intervals for the HR, and corresponding P values. The proportionality assumption was assessed graphically using log (−log) plots and quantitatively using the Z statistic. We used recursive partitioning analyses (RPA) in a univariate fashion to separate patients into the most homogeneous subgroups with respect to survival (OS or PFS). Patients were assigned into high and low RRM1 expression subgroups, which were carried forward into the univariate and multivariate analyses. Statistical analyses were performed using SAS statistical software (version 9.2; SAS Institute, Cary, NC).
Prognostic Value of RRM1 in the Entire Cohort of Patients With Resectable Pancreatic Adenocarcinoma
In total, 122 patients were diagnosed with pancreatic ductal adenocarcinoma and underwent tumor resection between October 1999 and December 2007. Overall, 68 patients (55.7%) were men, and the median age at the time of tumor resection was 65 years (range, 40-93 years), as indicated in Table 1. The surgical procedures included a Whipple procedure in 100 patients (82%), distal pancreatectomy in 20 patients (16.4%), subtotal pancreatectomy, and uncinate excision. Thirty-eight patients (31.1%) had a positive surgical margin. The mean tumor size (±standard deviation) was 3.4 ± 1.5 cm. Both TNM stage and overall stage were examined. No patient had metastatic disease. The tumor grade was determined as well differentiated, moderately differentiated, and poorly differentiated. The median number of positive lymph nodes was 1 (range, 0-21 lymph nodes). The tumors in 57 patients (46.7%) demonstrated lymphovascular invasion, and the tumors in 77 patients (63.1%) had perineural invasion. Forty-four patients (36.1%) received adjuvant gemcitabine treatment or gemcitabine-based regimens, and 35 patients (28.7%) received nongemcitabine-containing adjuvant therapy. The median follow-up was 14.5 months (range, 0.2-130 months). At the time of last follow-up, 106 patients (86.9%) had died.
|Variable||No. of Patients (%)|
|Median [range]||65 [40-93]|
|Distal pancreatectomy||20 (16.4)|
|Subtotal pancreatectomy||1 (0.8)|
|Uncinate excision||1 (0.8)|
|Tumor size, cm|
|Median [range]||3.5 [0.1-8]|
|No. of positive lymph nodes|
|Median [range]||1 [0-21]|
|Status at follow-up|
RPA identified a cutoff value of 1.13465 for the association of RRM1 expression with OS and a cutoff value of 1.952 for the association of RRM1 expression with PFS. For OS analysis, 96 patients were segregated into a low RRM1 expression group, and 26 patients were segregated into a high RRM1 expression group. For PFS analysis, 37 patients were segregated into a low RRM1 expression group, and 43 patients were segregated into a high RRM1 expression group. Among the variables that were included in univariate analysis, number of positive lymph nodes, overall stage IIA/I and IIB/I, and perineural invasion were associated with decreased OS, as indicated in Table 2. Similarly, the number of positive lymph nodes, overall stage IIB/I, lymphovascular invasion, perineural invasion, and lymph node status were associated with decreased PFS. Patients who received adjuvant therapy had longer OS (median, 19.0 months vs 7.0 months; P = .0008) and PFS (median, 15.4 months vs 3.6 months; P < .0001) than patients who underwent surgical resection only, as depicted in Figure 1. RRM1 expression was not a prognostic factor for either OS or PFS.
|OS, n = 122||PFS, n = 80|
|Variable||HR||95% CI||P||HR||95% CI||P|
|RRM1 expression (ΔCt): High vs lowa||1.3||0.8-2.1||.2||0.9||0.5-1.5||.7|
|Tumor size per 1-cm increase||1.1||1.0-1.3||.06||1.2||1.0-1.4||.07|
|No. of positive lymph nodes per 1-lymph-node increase||1.1||1.04-1.2||.001||1.1||1.01-1.2||.02|
|Lymphovascular invasion: Yes/no||1.4||1.0-2.1||.06||1.9||1.1-3.2||.02|
|Perineural invasion: Yes/no||1.7||1.1-2.6||.009||2.4||1.3-4.4||.004|
|Adjuvant therapy: Yes/no||0.5||0.3-0.7||.001||0.3||0.1-0.5||< .0001|
|Age per 10-y increase||1.0||0.8-1.2||.9||0.8||0.6-1.1||.1|
|Lymph node status: N1/N0||1.4||0.9-2.0||.1||2.1||1.2-3.7||.01|
|Surgical margins: Positive/negative||1.3||0.9-1.9||.2||0.9||0.5-1.7||.8|
In multivariable analyses (Table 3), adjuvant therapy was associated with both increased OS and increased PFS. The number of positive lymph nodes was associated with both decreased OS and decreased PFS.
|Variable||HR||95% CI||P||HR||95% CI||P|
|Adjuvant therapy: Yes/no||0.5||0.3-0.7||.001||0.3||0.1-0.5||< .0001|
|No. of positive lymph nodes per 1-lymph-node increase||1.1||1.07-1.2||< .0001||1.1||1.02-1.2||.01|
|Patients who received adjuvant gemcitabine|
|RRM1 expression: Low/higha||0.3||0.1-0.8||.01||NA||NA||NA|
|No. of positive lymph nodes per 1-lymph-node increase||1.2||1.04-1.4||.01||NA||NA||NA|
|Perineural invasion: Yes/no||NA||NA||NA||7.1||1.6-30.7||.009|
|Patients who received nongemcitabine adjuvant therapy|
|RRM1 expression: high/lowb||0.3||0.1-0.7||.005||0.3||0.1-0.8||.02|
|No. of positive lymph nodes per 1-lymph-node increase||1.2||1.04-1.3||.007||1.2||1.04-1.3||.01|
Predictive Value of RRM1 in the Subgroup of Patients With Resectable Pancreatic Adenocarcinoma Who Received Adjuvant Gemcitabine
Next, we performed similar analyses for a subpopulation of 44 patients who received adjuvant gemcitabine treatment. RPA identified a cutoff value of 2.4778 for the association of RRM1 expression with OS and a cutoff value of 2.2479 for the association of RRM1 expression with PFS. For OS analysis, there were 10 patients in the low RRM1 expression group and 34 patients in the high RRM1 expression group. For PFS analysis, there were 11 patients in the low RRM1 expression group and 23 patients in the high RRM1 expression group. In the univariate analysis depicted in Table 4, high RRM1 expression (HR, 3.3; P = .007), higher T-classification (T2/T1, T3/T1), and the number of positive lymph nodes were associated with decreased OS. Only perineural invasion was associated with decreased PFS. The high RRM1 expression group had a trend toward worse PFS (P = .07). Patients who had tumors with low RRM1 expression had significantly longer OS than patients who had tumors with high RRM1 expression, as depicted in Figure 1C (median OS, 47.8 months vs 14.1 months; P = .005). Patients who had tumors with low RRM1 expression had longer PFS than patients who had tumors with high RRM1 expression (median PFS, not reached vs 12.9 months; P = .06), although the result was not statistically significant (Fig. 1D).
|OS, n = 44||PFS, n = 33|
|Variable||HR||95% CI||P||HR||95% CI||P|
|RRM1 expression (ΔCt): High/lowa||3.3||1.4-7.8||.007||2.8||0.9-8.5||.07|
|Tumor size per 1-cm increase||1.3||1.0-1.8||.05||0.9||0.6-1.4||.8|
|No. of positive lymph nodes per 1-lymph-node increase||1.2||1.1-1.3||.003||1.1||0.9-1.4||.3|
|Lymphovascular invasion: yes/no||0.9||0.5-1.7||.7||1.5||0.6-3.4||.4|
|Perineural invasion: yes/no||1.5||0.7-3.1||.3||7.1||1.6-30.7||.009|
|Age per 10-y increase||1.0||0.7-1.5||.9||0.6||0.4-1.0||.07|
|Lymph node status: N1/N0||1.4||0.7-2.7||.3||2.2||0.9-5.3||.08|
|Surgical margins: Positive/negative||1.5||0.8-3.1||.2||1.2||0.5-3.2||.7|
The multivariable analyses depicted in Table 3 demonstrated that the number of positive lymph nodes and perineural invasion were predictors of shorter OS and shorter PFS, respectively. Low RRM1 expression predicted a benefit from adjuvant gemcitabine with longer OS (HR, 0.3; P = .01).
Predictive Value of RRM1 in the Subgroup of Patients With Resectable Pancreatic Adenocarcinoma Who Received Nongemcitabine-Containing Adjuvant Therapy
In a subpopulation of 35 patients who received nongemcitabine-containing adjuvant therapy, 32 patients (91.4%) received chemoradiation with 5-fluorouracil. RPA identified a cutoff value of 2.00195 for the association of RRM1 expression with OS and a cutoff value of 1.98315 for the association of RRM1 expression with PFS. For OS analysis, there were 15 patients in the low RRM1 expression group and 20 patients in the high RRM1 expression group. For PFS analysis, there were 11 patients in the low RRM1 expression group and 16 patients in the high RRM1 expression group. In univariate analysis (Table 5), the number of positive lymph nodes was associated with decreased OS; and the number of positive lymph nodes and lymphovascular invasion were associated with decreased PFS. The high RRM1 expression group had both a favorable OS (HR, 0.4; P = .01) and a favorable PFS (HR, 0.4; P = .047). Patients who had tumors with high RRM1 expression had significantly longer OS than patients who had tumors with low RRM1 expression (median OS, 41.9 months vs 19.8 months; P = .01), as depicted in Figure 1E. Patients who had tumors with high RRM1 expression had longer PFS than patients who had tumors with low RRM1 expression (median PFS, 70.0 months vs 11.8 months; P = .04) (Fig. 1F).
|OS, n = 35||PFS, n = 27|
|Variable||HR||95% CI||P||HR||95% CI||P|
|RRM1 expression (ΔCt): High/lowa||0.4||0.2-0.8||.01||0.4||0.1-0.99||.047|
|Tumor size per 1-cm increase||1.2||0.9-1.5||.2||1.3||0.9-1.7||.1|
|No. of positive lymph nodes per 1-lymph-node increase||1.1||1.01-1.3||.03||1.1||1.01-1.3||.03|
|Lymphovascular invasion: Yes/no||2.2||1.0-5.0||.06||2.9||1.08-7.7||.04|
|Perineural invasion: Yes/no||2.2||1.0-4.8||.06||1.7||0.7-4.5||.3|
|Age per 10-y increase||0.9||0.6-1.3||.5||0.8||0.5-1.4||.5|
|Lymph node status: N1/N0||1.8||0.8-4.3||0.2||2.6||0.9-7.6||.07|
|Surgical margins: Positive/negative||1.2||0.5-2.9||.6||0.8||0.2-3.0||.8|
The multivariable analyses illustrated in Table 3 confirmed that high RRM1 expression predicted a benefit from nongemcitabine-containing adjuvant treatment of both longer OS (HR, 0.3; P = .005) and longer PFS (HR, 0.3; P = .02). In addition, the number of positive lymph nodes was a predictor of worse OS and PFS.
Several potential prognostic and predictive factors have been studied extensively to help guide individualized therapy and to facilitate patient stratification for future clinical trials in pancreatic adenocarcinoma.14 The results from univariate analyses in the current study of patients with resectable pancreatic adenocarcinoma recapitulated common histopathologic prognostic factors. They were tumor size, disease stage, the number of positive lymph nodes, lymphovascular invasion, perineural invasion, and adjuvant therapy. These results indicate that the cohort in the current study is a representative group of patients with resectable pancreatic adenocarcinoma and is similar to the patients reported in previous studies.
Since gemcitabine became the mainstay of therapy for advanced pancreatic adenocarcinoma,1 several key enzymes involved in the transportation and metabolism of gemcitabine have been studied.9 Although it has been demonstrated that RRM1 is a prognostic and predictive marker in patients with NSCLC who received gemcitabine-based therapy,7 neither the prognostic ability nor the predictive value of RRM1 have been clearly demonstrated in patients with resected pancreatic cancer or in patients who received adjuvant gemcitabine. In the study by Giovannetti and coworkers,9 RRM1 expression was stratified into tertile and two groups by median as the cutoff in 81 patients treated with gemcitabine. However, there were no significant correlations between RRM1 expression and survival in pancreatic adenocarcinoma. A later study from Fujita and coworkers in Japan supported similar conclusion.10 Of 70 patients with resectable disease, including 40 patients who received adjuvant gemcitabine, RRM1 expression had neither prognostic significance nor predictive value. In contrast, Akita and coworkers8 evaluated 68 patients with resectable pancreatic adenocarcinoma and demonstrated that patients with high RRM1 protein expression had significantly longer OS. Among the 28 patients in their study who received adjuvant gemcitabine, low RRM1 protein expression was associated with better response and longer OS.
The findings in our study suggest that RRM1 expression determined by QRT-PCR did not appear to be prognostic for the entire cohort of 122 patients with resectable pancreatic adenocarcinoma. However, low RRM1 expression predicted an adjuvant gemcitabine benefit of prolonged OS for 44 patients who received this treatment in the current study. This finding is consistent with the results reported by Akita et al in 28 patients with resectable stage I and stage II pancreatic adenocarcinoma. The use of automated quantitative analysis based on immunohistochemistry by Akita et al8 also substantiates our conclusion regarding the value of RRM1 in predicting response to adjuvant gemcitabine in patients with resectable pancreatic adenocarcinoma.
Several possible rationales may explain these apparent discrepancies. First, patients with different stages of pancreatic adenocarcinoma were studied. The studies by Giovannetti et al9 and Fujita et al10 included patients with stage III and IV disease. Our study was similar to the study by Akita et al,8 which involved only patients with stage I or II disease. The prognosis and natural biology of resectable and unresectable pancreatic adenocarcinoma are different, as demonstrated in clinical practice.1, 2 Therefore, studies of these 2 different groups of patients were not expected to yield similar results. Second, genotypic variations between Japanese patients in previous studies8, 10 and Caucasian patients in the current study may explain in part the different results. Recent studies on RRM1 single nucleotide polymorphisms (SNPs) demonstrated an association between several SNPs collectively and gemcitabine toxicities as well as clinical outcomes.15 Third, different quantification methods were used to examine RRM1 expression. Fujita et al10 used QRT-PCR for RNA quantification in their study, although they used only 1 reference gene: β-actin. The study by Giovannetti et al9 included frozen biopsy specimens in addition to resection tissue samples with glyceraldehyde 3-phosphate dehydrogenase (GAPDH) as the QRT-PCR reference gene. In the current study, we used QRT-PCR on paraffin-embedded resection tissues with normalization to 2 reference genes (RPL13A and HMBS). We also evaluated β-actin as the reference genes. However, further analyses with the Norm-Finder (version 19; Molecular Diagnostic Laboratory, Aarhus, Denmark) and geNorm (version 3.5; Zoological Institute, Leuven, Belgium) software packages demonstrated that β-actin was not as stable as the 2 reference genes we used in pancreatic cancer (unpublished data). Finally, different methods were used to choose the cutoff value for RRM1 stratification. In the current study, we used RPA for dichotomization of the RRM1 level. The same method was used in a previous study.10 Compared with arbitrary cutoff level selections, such as medians and tertiles in some of the previous studies,8, 9 RPA provides insights into the data structure and data partitioning scheme that do not require user-performed calculations.16 With a prespecified plan for statistical analysis in this study, RPA in a conditional inference framework17 allowed less variable selection bias and reportedly has better sensitivity and specificity18 and is accurate when it works with time-to-event data.19 In this study, the significance of these dichotomies also was confirmed in the multivariable Cox proportional hazards regression.
Our results from patients in this study who received nongemcitabine-containing adjuvant therapy—mainly chemoradiation and 5-fluorouracil—demonstrated a different predictive role of RRM1 expression in patients with resectable pancreatic adenocarcinoma. These findings indicate that, after surgical resection, to achieve the best survival, the level of RRM1 expression may be used to stratify patients to receive either adjuvant gemcitabine for the low RRM1 expression group (ΔCt ≥2.4778) or nongemcitabine-containing adjuvant therapy for the high RRM1 expression group (ΔCt < 2.00195). Although this implication is supported by the role of RRM1 as a tumor suppressor4 and the previous findings from early stage NSCLC studies,12 the conclusions should be interpreted with caution, because this was a retrospective study in a single institution, and our findings are not immune to selection bias or information bias. The sample sizes of our subpopulations for different adjuvant therapies were relatively small. Also, our results indicate that nongemcitabine-containing adjuvant therapy, such as chemoradiation and 5-fluorouracil, leads to favorable clinical outcome compared with surgical resection alone. Although this result is consistent with the previous retrospective studies,20 it falls into the current debate on the role of radiation therapy in patients with resectable pancreatic cancer brought about by the findings from previous randomized clinical trials.21-23 It is likely that more evidence will need to be generated to clarify the current controversy if the patients enrolled in clinical trials are stratified and treated based on RRM1 expression status.
RPA identifies the best cutoff point at which each subgroup is more homogeneous with respect to survival. This type of cutoff point is not equivalent to a biologic cutoff point, which is clinically meaningful but not usually available. The cutoff points identified by RPA for OS in both the gemcitabine and nongemcitabine groups (closer to the median of RRM1 level of ΔCt = 1.906) were much lower than the cutoff point for the entire cohort (ΔCt = 1.13465) in this study. The lower cutoff points for both the gemcitabine and nongemcitabine groups with respect to survival simply means that small numbers of patients with relatively low RRM1 levels have a more similar survival within each group. A validation study should be conducted before the current findings, including the cutoff points, can be clinically applied. Alternatively, semiquantitative analysis, such as immunohistochemistry, may be used to stratify RRM1 protein expression. Indeed, a recent study demonstrated a correlation between RRM1 mRNA levels and protein levels determined using an automated, quantitative RRM1 protein assay,12 opening the door for the potential application of a semiquantitative immunohistochemical analysis of RRM1 as a simple and easily interpretable tool in clinical practice. However, this approach needs extensive investigation, optimization, and standardization.
In conclusion, in this study, we demonstrated that low RRM1 expression determined by QRT-PCR on paraffin-embedded pancreatic adenocarcinoma tissue predicts an OS benefit for patients who receive adjuvant gemcitabine, whereas high RRM1 expression predicts a survival benefit for patients who receive nongemcitabine-containing adjuvant therapy. Further large, prospective studies are warranted to confirm the current findings and to provide guidance for individualized therapy.
No specific funding was disclosed.
CONFLICT OF INTEREST DISCLOSURES
Dr. Wang and Dr. Jiang are employees of Veridex.