Pharmacogenetic Analysis of INT 0144 Trial: Association of Polymorphisms with Survival and Toxicity in Rectal Cancer Patients Treated with 5-FU and Radiation Free

The efficacy of 5-fluorouracil (5-FU)–based chemoradiation (CRT) in rectal cancer has been well demonstrated by several large prospective studies, improving both local control and survival when compared with surgery alone or postoperative radiotherapy (1, 2). Consequently, postoperative 5-FU–based CRT was widely adopted for stage II and III rectal cancers in North America. More recently, based on the publication of two large phase III trials, 5-FU–based CRT was shifted to the neoadjuvant setting (3, 4). However, while the CRT sequence with regard to surgery has been shown to be important for local recurrence, its impact on overall survival (OS) is currently unclear. With regard to the chemotherapeutic agents, 5-FU remains the cornerstone agent that is combined with radiotherapy.

The Southwest Oncology Group (SWOG) initiated a clinical trial (S9304/INT 0144) on behalf of the Gastrointestinal Intergroup, comparing adjuvant CRT therapy using different 5-FU–based regimens to clarify their impact on clinical outcome and toxicity in patients with stage II and III rectal cancer. The SWOG S9304 study is the largest study reported so far in patients with locally advanced rectal cancer, with 1,917 registered patients. Patients were randomized to three treatment arms: bolus 5-FU before and after radiotherapy with 5-FU protracted venous infusion (PVI) during radiotherapy, 5-FU PVI only and bolus 5-FU only with leucovorin and levamisole. All arms provided similar disease-free survival (DFS), 3-year DFS of 67% to 69% and 3-year OS of 81% to 83%. In contrast, toxicity profiles were different with more grade 3 to 4 hematologic toxicities in the two bolus 5-FU arms (arm 1, 49% and arm 3, 55% vs. 4% in arm 2; ref. 5).

In contrast to the growing number of predictive biomarkers for anticancer agents, there are no established biomarkers to select patients who will benefit most from CRT, ultimately leading to higher cure rates (6). Patients having a worse outcome despite 5-FU–based CRT may need treatment intensification and more efficient monitoring strategies that could detect early recurrence and allow curative resection. It is recognized that germline variations in genes involved in the host microenvironment, DNA repair and drug metabolisms, as opposed to somatic mutations in the tumor, have a great influence for treatment efficacy or toxicity (7). We hypothesized that genetic variation within critical genes involved in major biologic capabilities of the rectal tumors might have prognostic value in patients treated with 5-FU–based CRT. We further assessed if these genetic variations would be associated with treatment toxicity. We evaluated 18 polymorphisms in 16 genes involved in detoxification (GSTP1, rs1695), angiogenesis (VEGFR2, rs1870377; VEGF, rs2010963 and rs3025039; CXCR2, rs2230054; IL8, rs4073), DNA repair (ERCC1, rs11615; RAD51, rs1801320; XRCC3, rs861539), adhesion (ICAM1, rs5498; MMP3, promoter 5A/6A), inflammation (COX2, rs20417 and rs5275; IL10, rs1800896), and tumor growth (EGFR, rs2227983; EGF, rs4444903; FGFR4, rs351855; CCDN1, rs17852153). The polymorphisms were selected on the basis of functionality and/or based on documented association with radiation, 5-FU–based CRT, or colorectal cancer patients' risk or prognosis (Supplementary Table S1).

The SWOG S9304 has been previously reported (5). A total of 1,917 patients with nonmetastatic rectal adenocarcinoma (T3-4, N0 or T1-4, N1-3) were enrolled in the SWOG S9304 study from March 1994 to August 2000. Patients were randomly allocated to one of three chemotherapy regimens and stratified according to nodal involvement, type of surgery, primary tumor stage, and time from surgery to registration.

Arm 1 consisted of induction bolus 5-FU followed by combined 5-FU PVI and radiation. Bolus 5-FU was then administered for two cycles.

Arm 2 consisted of induction 5-FU PVI followed by combined 5-FU PVI and radiation. Additional 5-FU PVI was given after completion of radiation.

Arm 3 consisted of induction bolus 5-FU, leucovorin, and levamisole. Radiation was combined with bolus 5-FU and leucovorin. Patients then received additional bolus 5-FU, leucovorin, and levamisole.

Radiation consisted in 45 Gy in single daily 1.8-Gy fractions followed by a boost volume of 5.4 Gy in 1.8-Gy fractions. Patients' characteristics, treatment schedules, and outcomes have been previously described (5).

The SWOG S9304 study used the Common Toxicity Criteria for Adverse Events (CTCAE) version 2.0. For this correlative study, we defined the clinical outcome of toxicity as any CTCAE grade ≥3 toxicity reported over the entire trial duration. Because of the distinct reported toxicity profile of each treatment regimen, toxicity outcomes were analyzed separately within each treatment arm. In addition to the overall toxicity, we explored the association of the polymorphisms with specific toxicities defined as: any grade ≥3 hematologic toxicity; any grade ≥3 gastrointestinal toxicity; and any grade ≥3 mucositis of the proximal upper gastrointestinal tract (PUGIT) defined by occurrence of any grade ≥3 stomatitis, pharyngitis, or esophagitis.

All patients signed an informed consent before entering this study that included information regarding the use of their tumor tissue to explore relevant molecular parameters. Formalin-fixed, paraffin-embedded (FFPE) surgically resected tumor tissue from the primary tumor was available for 746 patients. Blocks or 10-μm thick unstained slides were anonymized, and then forwarded to the laboratory, which was blind to treatment and outcomes. Samples were sent to the laboratory through multiple shipments over an extended period of time (from 2007 to 2010). Therefore, a screen testing with 241 available patients was performed and reported in 2008 (8). Polymorphisms with at least a trend for an association with outcome were expanded for an additional 505 patients. On the basis of newly available data, promising polymorphisms were tested in the latest cohort and correlated with survival and toxicity. Additional testing in the screen cohort was not conducted because of the lack of sufficient tumor tissue. Toxicity analysis for the polymorphisms evaluated solely in the screen cohort was not performed because of the resulting insufficient number of patients within each arm. This study was conducted adhering to the REporting recommendations for tumor MARKer prognostic studies (REMARK). The reference single-nucleotide polymorphisms (SNP) identification numbers, forward and reverse primer, and assay type are summarized in Supplementary Table S1.

Two 10-μm thick unstained slides were macrodissected by sterilized surgical scalpel. FFPE tumor tissues were collected in a sterilized eppendorf tube. DNA isolation was done by conventional proteinase K-phenol/chloroform—ethanol method using the QIAamp Extraction Kit (Qiagen) according to the manufacturer's protocol. Extracted DNA was stored at −80°C. Polymorphisms were determined using the polymerase chain reaction-restriction fragment length polymorphism (PCR-RFLP) technique. Briefly, 100-ng genomic DNA was amplified by PCR, followed by restriction enzyme digestion. Resulting DNA fragments were visualized on 3% to 4% ethidium bromide–stained agarose gels. If no matching restriction enzyme could be found, samples were analyzed by direct sequencing. For quality control purposes, 5% of the samples were analyzed by direct DNA-sequencing for each polymorphism to double-check the quality of the PCR-RFLP essays. Genotype concordance was >98%.

GSTP1 mRNA expression levels were measured at Response Genetics, Inc., a Clinical Laboratory Improvement Amendments–approved laboratory. Manual microdissection using a light microscope was performed on all tumor samples to ensure >80% tumor cells were dissected. After RNA isolation, cDNA was prepared from each sample as described previously (9, 10). Quantitation of GSTP1 and an internal reference (β-actin) cDNA were performed using a fluorescence-based real-time detection method [ABI PRISM 7900HT Sequence detection System (TaqMan) Applied Biosystems] as previously described (11). Primers and probe for detection of GSTP1 mRNA gene expression levels were as following: probe; TCACCTGGGCCGCACCCTTG; forward primer, CCTGTACCAGTCCAATACCATCCT; reverse primer, TCCTGCTGGTCCTTCC-CATA. All samples were amplified in triplicate. mRNA quantification was done by using the average of the triplicates.

The primary endpoints of this study were OS, DFS, relapse patterns, and toxicity. OS was defined as the date from registration until date of death from any cause. Patients still known to be alive at the time of analysis were censored at the last known date of contact. DFS was defined as date from registration to the first development of recurrent disease, or of death from any cause, whichever came first. Patients still known to be alive without recurrence at the time of analysis were considered to be censored at their last follow-up time. OS and DFS estimates and curves were generated using the Kaplan–Meier method. Polymorphisms were modeled using indicator variables for the heterozygous and homozygous variant genotypes. The associations of polymorphisms with OS and DFS were analyzed univariately using the Cox proportional hazards model and adjusting for the variables used in stratification at time of randomization: the type of surgery (abdominoperineal resection vs. anterior resection), nodal involvement (N0 vs. N1 vs. N2-3), primary tumor stage (T1-2 vs. T3 vs. T4b), and time from surgery to registration (20–45 days vs. 46–70 days), both in comparing pairwise genotype levels as well as the overall effect of the polymorphism on OS and DFS.

Toxicities, using the National Cancer Institute Common Toxicity Criteria (version 2.0), were categorized by maximum grade, evaluating at grades 0–2 vs. grades 3–5. The maximum grade of all reported toxicities was identified, as well as the maximum grade of specific toxicities including hematologic, gastrointestinal, and stomatitis or esophagitis. The association between the polymorphisms and toxicity was tested using the χ² test.

Given the number of analyses and comparisons, all unadjusted P values at or below the 0.05 level were adjusted for SNP-wise error rates using the Bonferroni–Holm method (12).

Patients' characteristics and outcomes included in this correlative study were similar to the overall SWOG S9304 population, and are summarized in Table 1. The large majority of studied patients were Caucasian (92%). At the time of data analysis, 383 out of 746 patients were still alive with a median follow-up of 123 months.

In univariate analyses, GSTP1-Ile105Val (rs1695) was significantly associated with DFS and OS. The 3-year and 5-year DFS was 60% and 53% for the low-activity genotype (G/G), 72% and 66% for G/A, and 65% and 57% for A/A patients (P = 0.02; Fig. 1A). The 3-year and 5-year OS was 78% and 58% for the low-activity genotype (G/G), 83% and 72% for G/A, and 79% and 66% for A/A patients (P = 0.02; Fig. 1B). In the multivariate model, adjusting for type of surgery, nodal status, primary tumor stage, and time from surgery until registration, rs1695 remained associated with both DFS and OS (Table 2; P = 0.01 and P = 0.01). The effect of rs1695 did not vary by treatment arm (P_interaction = 0.13) nor by gender (P = 0.10). Furthermore, its effect did not vary by either age or stage (stage II vs. III). None of the other tested polymorphisms were significantly associated with DFS or OS (Table 2).

The site of relapse of the studied population was similar to the overall SWOG S9304 population (Supplementary Table S2). rs1695 was associated with a significantly different rate of distant relapse (P = 0.03), consistent with the overall effect on DFS and OS (Supplementary Table S3). rs1695 was not associated with pelvic failure risk. In addition, MMP3 polymorphism showed a similar association that should be interpreted with caution, given the small number of patients and the lack of effect on DFS and OS.

Among patients with both rs1695 genotypes and gene expression levels assessed, there were 376 eligible patients with both genotype and expression data available. The distribution of rs1695 genotypes in these 376 patients was 171 (45%), 163 (43%), and 42 (11%) for A/A, G/A, and G/G, respectively. There was no significant association between rs1695 genotypes and gene expression levels (Supplementary Table S4).

In arm 2, the homozygous IL8 −251A/A genotype (rs4073) was associated with a lower risk of any grade 3–5 toxicities compared with A/T or T/T genotypes (P = 0.04; Supplementary Table S5). None of the evaluated polymorphisms predicted a significant relation with the incidence of overall toxicity in the two arms with bolus 5-FU–based CRT. Despite not being significantly associated with overall toxicity, several polymorphisms were related to specific grade 3–5 toxicities (Supplementary Table S5). Given the relatively small number of patients and events, these findings should be interpreted with caution. Overall, stomatitis and pharyngitis represented most of the PUGIT mucositis (stomatitis and pharyngitis = 425; esophagitis = 17). The homozygous VEGFR2 H472Q A/A genotype (rs1870377) was associated with a higher risk of grade 3–5 PUGIT mucositis compared with A/T or T/T genotypes in arm 2 (P = 0.04). However, the rs1870377 A/A genotype was significantly associated with a lower risk of PUGIT mucositis in arm 1 (P = 0.004). The EGFR +497G>A A/A genotype (rs2227983) was associated with higher risk of PUGIT mucositis only in arm 1 (P = 0.05). Although patients heterozygous for the EGF +61A>G polymorphism (rs4444903) experienced more hematologic toxicities in the arm 1 (P = 0.04), they experienced less PUGIT mucositis in arm 3 (P = 0.05). And finally, patients with the COX2 +8743 C/C genotype (rs5275) experienced less gastrointestinal toxicities in arm 1 (P = 0.03).

After adjusting rs1695 for multiple testing, the association did not remain univariately significant for OS, DFS (P = 0.36 for both), and site of relapse (P = 0.54). In the multivariate model adjusted for type of surgery, lymph node involvement, primary tumor stage, or time from surgery to registration, the P value was 0.18 for both OS and DFS. MMP3 polymorphism did not remain significant for site of relapse (P = 0.68). Moreover, none of the polymorphisms remained significantly associated with toxicity.

This study identified rs1695 as a prognostic biomarker, significantly associated with both DFS and OS. At 5 years, depending on the patient's genotype, this SNP could determine three distinct subgroups with a 14% and 6% absolute difference for the 5-year OS between the high- (G/G) and low-risk (G/A) groups and the intermediate- (A/A) and low-risk (Val/Ile) group, respectively. This association was independent of stratification factors that included prognostic factors such as tumor and nodal stage. Its effect did not vary by treatment arm, gender, age, or stage. We further showed that rs1695 was not associated with GSTP1 gene expression in the tumor, consistent with the known SNP's function affecting enzymatic activity.

The glutathione S-transferase (GST) superfamily is involved in cellular detoxification processes by catalyzing the conjugation of glutathione to a wide variety of xenobiotics and to reactive oxygen species (enzymatic role). The GST detoxification ability plays an important role in cellular protection, and is being implicated in drug resistance and, to a lesser degree, radiation resistance (13). The GSTPi class, encoded by the single gene (GSTP1), is the most highly expressed GST in human tumors, including colon cancers (14). Importantly, GSTP1 acts also as a regulator of the mitogen-activator protein (MAP) pathway (nonenzymatic role) by protein interaction (e.g., JNK1; ref. 15). This mechanism has been postulated for GSTP1-overexpressed drug-resistant cells when the drug is not a GSTP1 substrate, such as antimetabolites (e.g., 5-FU). In colon cancer, both GSTP1 enzymatic and nonenzymatic roles are critical, promoting proliferation and survival (16).

rs1695 leads to an amino acid change within the active site of the GSTPi protein, changing the GSTP1 catalytic activity in a substrate-specific manner. The catalytic activity and detoxification capacity in individuals with the G allele as compared with individuals with A allele was reduced for 1-chloro-2,4-dinitrobenzene (17). In contrast, the G allele conferred a higher catalytic efficiency for polycyclic aromatic hydrocarbon diol epoxides (18). Moreover, A-containing and G-containing haplotypes were shown to have differential effects on proliferation and apoptosis (19). These preclinical observations provide an explanation as to why the heterozygote G/A genotype has a differential outcome when compared with the homozygous genotypes. Most previous data are difficult to compare with this study because they included patients given platinum-based treatments, and included colon cancer or advanced diseases (20–25). However, our group previously reported clinical data supporting rs1695 allele-specific activity in stage II and III rectal cancer (26).

Because the locoregional failure (LRF) occurred in only approximately 10% in each arm and convincing studies testing the role of GSTP1 and radiation resistance were lacking, we speculated that rs1695 effect was either linked to the systemic therapy (antimetabolites) or to the tumor-inherent biologic behavior (prognostic effect). Consistently, with our hypothesis, we were able to demonstrate that the frequency of failure by rs1695 genotype was different only for distant relapses.

The higher rs4073 expression genotype (A/A) predicted a lower risk of overall high-grade toxicity only in patients given PVI 5-FU. It is critical to point out that fluoropyrimidines have different antitumor effects according to the mode of administration (27), which translates to specific side effects such as mucositis, stomatitis, and hand–foot syndrome. IL8 is a proinflammatory chemokine that is involved is several signaling pathways promoting cell survival, angiogenesis, and proliferation (28). We have previously published that overexpression of IL8 leads to cell proliferation and survival through the activation of the transcription factor NF-κB (29) and its inhibition overcame resistance to prolonged 5-FU exposure (30). We therefore postulate that high germline IL8 expression protects against cell death from PVI 5-FU through the activation of NF-κB. Alternatively, it may be related to the IL8 proangiogenic effect, as IL8 overexpression leads to higher vascular density (29). Increased VEGF-independent angiogenesis may translate to better healing processes and would therefore be particularly relevant for continuous 5-FU, given that its toxicity profile is predominant in the mucosae.

As bolus 5-FU is being replaced by infusional 5-FU in the treatment of colorectal cancer, we focused on the toxicity profiles on arm 2. Only rs1870377 was associated with PUGIT mucositis in patients given infusion 5-FU with radiotherapy. The homozygous A genotype associated with a lower binding affinity for VEGF (31) predicted a higher risk of mucositis. VEGFR2 is mostly expressed on vascular endothelial cells and is a crucial regulator of angiogenesis (32) and in wound-healing processes (33). Lower VEGF pathway activation may therefore interfere with this process and be associated with more important toxicities. Interestingly, our data suggest that this association depends on the mode of 5-FU administration, and is relevant only for PVI 5-FU, which is known for its mucosal toxicity. In contrast, for bolus 5-FU, with low mucosal toxicity, lower VEGF pathway activation was associated with lower mucosal toxicity. We speculated that lower VEGF pathway activation, affecting vascularity density or vascular permeability (34), led to poor drug distribution, reducing the risk of upper GI toxicities. However, the small number of events and the fact that no significant association was found in arm 3 raise the possibility of false-positive results.

This study has several limitations. Patients included in the SWOG S9304 trial were treated with postoperative 5-FU–based CRT. Although current standard of care in locally advanced rectal adenocarcinoma is preoperative 5-FU–based or capecitabine-based CRT because of better local control, it has never been shown to be associated with better distant control and longer OS. As rs1695 was linked only with distant relapses, it is therefore likely that its prognostic role can be translated to current practice. However, clinicians should be cautious when interpreting the toxicity data, as they may differ according to the fluoropyrimidine (e.g., capecitabine). Despite the biomarker evaluation being performed in a large prospective and well-designed phase III study, we acknowledge that we could collect only a subset of the 1,917 patients included in the trial and that we cannot exclude a sampling bias. However, we showed that patients' characteristics and outcomes included in this correlative study were similar to the overall SWOG S9304 population. Because of the number of studied variables, we adjusted our results for multiple testing. Whenever adjustments are made for multiple comparisons, the P values increase due to spreading 5% chance of error over more than one (multiple) comparison. None of the evaluated biomarkers remained significant raising the possibility of false-positive results (finding by chance). Nevertheless, in the context of a hypothesis-driven (nonagnostic) approach and because multiple testing adjusting techniques also increase false-negative results, we believe that rs1695 should be further studied. Therefore, given this limitation and despite the need of biomarkers in rectal cancer, we propose further validation of our results in another large adequately powered clinical trial before going into prospective validation. Prognostic biomarkers such as rs1695 are important, as they may help clinicians to evaluate and discuss the need of postoperative oxaliplatin-based chemotherapy based on the risk of distant relapse. Although the relative gain does not differ, the absolute gain is much higher when the prognosis worsens. Furthermore, patients with the unfavorable rs1695 G/G genotype may be encouraged to take part in clinical trials. Finally, prognostic biomarkers are often the key to find druggable targets in the biomarker itself or its related pathway.

In summary, our data show that the rs1695 is significantly associated with both DFS and OS in a representative subgroup of the large prospective phase III S9304 study. Additional clinical studies and further preclinical research should be performed to validate our data and better characterize this biomarker. Better understanding of rs1695 variants may help to optimize anticancer treatments for rectal cancers. Both rs4073 and rs1870377 may be associated with toxicity in patients given infusional 5-FU with radiotherapy.

K. Danenberg has ownership interest (including patents) at Liquid Genomics, Inc. No potential conflicts of interest were disclosed by the other authors.

Conception and design: C.M. Ulrich, S.R. Smalley, J.A. Martensen, R.J. Mayer, K. Danenberg, H.-J. Lenz

Development of methodology: P. Bohanes, K. Danenberg, W. Zhang, H.-J. Lenz

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): P. Bohanes, T. Winder, S.R. Smalley, A.B. Benson III, R.J. Mayer, C.M. Cripps, K.W. Makar, H.-J. Lenz

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): P. Bohanes, C.J. Rankin, C.D. Blanke, T. Winder, C.M. Ulrich, W. Zhang, J.K. Benedetti, H.-J. Lenz

Writing, review, and/or revision of the manuscript: P. Bohanes, C.D. Blanke, T. Winder, C.M. Ulrich, S.R. Smalley, J.A. Martensen, A.B. Benson III, R.J. Mayer, C.M. Cripps, K.W. Makar, H.-J. Lenz

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): C.D. Blanke, S.R. Smalley, T.A. Rich, K.W. Makar, H.-J. Lenz

Study supervision: S.R. Smalley, T.A. Rich, H.-J. Lenz

Other (principal investigator on the study from which the study materials were collected): S.R. Smalley

This work was supported, in part, by the following PHS Cooperative Agreement grant numbers awarded by the National Cancer Institute, Department of Health and Human Services (DHHS): CA32102, CA38926, CA058882, CA142558, CA21115, CA17145 CA25224, CA077202, CA32291, CCSRI021039, and CCSRI015469; and by NIH grants 5R01CA105145 and NIH P30CA014089. P. Bohanes was supported by a research grant from “Ligue Suisse contre le Cancer” and by the Daniel Butler Memorial Research Fund.

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.