GST Pi is a member of the glutathione-S-transferase superfamily of phase II xenobiotic-metabolizing enzymes that catalyse the conjugation of endogenous and exogenous electrophiles, including reactive oxygen species, toxins, carcinogens and anti-cancer agents, to the nucleophilic thiol group of reduced glutathione (GSH) [12
]. A number of studies have concentrated on the connection between aberrant expression of GST isozymes, including GST Pi isozymes, with the development and expression of resistance to chemotherapy drugs as reviewed by McIlwain et al.
]. From this perspective an expected result should have been a poor response to chemotherapy in patients with high expression of GST Pi. However our study has shown that, for stage C colon cancer patients, overall survival was significantly and markedly poorer in patients with high GST Pi who did not receive chemotherapy than in those with high GST Pi who did. Furthermore, survival in the latter group was no different from that in patients with low GST Pi, whether or not they received chemotherapy. Expressed differently, chemotherapy had no apparent effect on survival in patients with low GST Pi whereas those with high GST Pi appeared to benefit considerably. This finding could be explained by the fact that a major function of GST Pi is not only the phase II metabolism of drugs but to also contribute to the detoxification of endogenously generated reactive oxygen species and the associated potentially toxic macromolecules and lipid peroxides produced in cells. An example of this detoxificant/antioxidant role is the induction of GST Pi as part of the coordinated defence mechanism to protect colonic cells against oxidative injury [14
]. The presence of high GST Pi in colon tumours may confer a selective advantage on malignant colon cells, as increased production of reactive oxygen species is a feature of cancer progression [15
]. Therefore the poorer overall survival in patients with high GST Pi who did not receive chemotherapy could be attributed to the ability of GST Pi to counter the elevated oxidative stress, enhancing persistence of malignant colon cells and thereby leading to poorer survival. This is also consistent with our previous finding of increased GST Pi expression being an independent prognostic factor for reduced overall survival [5
]. Furthermore, treatment with diverse chemotherapy drugs tips the redox balance of already stressed cancer cells and elicits selective additional toxicity that ultimately kills them. In this regard the recent reports of the ability of 5-FU to promote oxidative stress in colonic cancer cells [16
], in addition to its multiple actions as an anti-metabolite, including inhibition of thymidylate synthase, may provide an explanation for the clear benefit of 5-FU based treatment in those patients with high GST Pi (Figure
). The additional generation of reactive oxygen species by 5-FU may overwhelm the advantage of high GST Pi in colon cancer cells either by directly increasing the molecular species detoxified by GST Pi enzyme activity and/or by depletion of the available pool of GSH required for conjugation reactions performed by GST enzymes. GSH is also the major anti-oxidant used by cells in non-enzymatic reactions to protect against reactive oxygen species.
Accordingly the implication of our finding is that stage C colon cancer patients with low intracellular concentrations of GST Pi may not need to be treated with 5-FU-based adjuvant chemotherapy whereas those with high GST Pi definitely should be treated. As far as we are aware, this is the first study to demonstrate the predictive value of GST Pi expression in regard to chemotherapy for stage C colon cancer.
A secondary finding was that GST Pi genotype had no apparent influence on survival or on the association between survival and the joint effects of GST Pi expression and adjuvant chemotherapy. Interest in the effect of GST Pi polymorphisms arose from the concept that variant GST Pi proteins may differentially affect the actions of chemotherapeutic drugs. The single nucleotide polymorphism A313G in exon 5 results in a valine for isoleucine substitution producing a Val/Val variant with reduced GST Pi enzyme activity [18
]. In metastatic colorectal cancer Stoehlmacher et al
. found poorer overall survival in patients with the Ile/Ile genotype than in those with the Ile/Val or Val/Val genotypes after combined 5-FU/oxaliplatin chemotherapy [9
] although this was not replicated by other researchers using the same agents [21
]. Also in 5-FU/oxaliplatin-treated patients poorer survival was seen in those with the Ile/Ile and Ile/Val genotypes compared with the Val/Val genotype [22
]. Similarly, significantly poorer survival associated with the Ile/Ile genotype as compared with the Ile/Val or Val/Val genotypes was reported in patients treated with FOLFOX-4 [10
], although two other studies using FOLFOX showed no difference in survival between genotypes [23
]. Additionally, a study of oxaliplatin treatment [25
] and two others using a mixture of agents and regimens showed no differences between GST Pi genotypes [11
]. All of the above were prognostic studies as all patients had received chemotherapy and there was no random allocation to different treatment groups. A truly predictive study that randomly allocated patients to one of three 5-FU-based treatment strategies also showed no differences between GST Pi genotypes [27
] while another study with patients randomised to first-line capecitabine alone versus capecitabine plus irinotecan showed a short, but significant improvement in progression-free survival for the Val/Val genotype only [28
]. Two remaining studies included patients with colorectal tumours of all Dukes/TNM stages, not just those with metastatic disease, but produced contradictory results. Holley et al. found no association between GST Pi genotype and overall survival [29
] whereas Jones et al. found that the Ile/Ile genotype was associated with decreased survival as compared with the Ile/Val and Val/Val genotypes [30
]. Thus the evidence for an association between GST Pi genetic polymorphism and chemotherapeutic treatment and survival (or other patient outcomes) is inconclusive. The reasons may relate to differing study designs; for example retrospective versus prospective, randomized versus non-randomized allocation to treatment; varying stage mixes of patients and hence different mixes of palliative and adjuvant treatment; various chemotherapeutic agents and regimens; and statistical issues such as limited numbers of patients available for analysis. The latter is particularly a problem in relation to the Val/Val genotype which has a prevalence of only 8% to 10% which may lead to type II errors in small patient samples.
A limitation of the present study is that patients were not randomly allocated to the treatment and control groups. Instead the treatment group comprised all patients selected for chemotherapy since April 1992 when it became available in this hospital, whereas the control patients were drawn from the period from 1979 to April 1992 and individually matched with treated patients on age and sex. However there was no significant difference between the treated group and the control group on 10 other background and tumour pathology characteristics. Comorbidity may have influenced survival and may have differed between the two groups but we were unable to compare them because details of comorbidity were not recorded in the database until 1995. However, as the survival of patients with low GST Pi did not differ significantly between the control and treatment groups, it is most unlikely that differential comorbidity could have accounted for the marked treatment/control survival difference in patients with high GST Pi alone. A further limitation was the relatively small number of patients available for analysis; there were 182 in total and 169 for the genotype analyses, which may have limited our ability to find statistically significant differences between some groups. Nevertheless the markedly poorer survival of the high GST Pi/non-chemotherapy group (Figure
) suggests that this interaction would be likely to persist in a larger study.
Earlier studies of the association between GST Pi expression and patient outcomes have yielded varying results. A significant independent association between high GST Pi and diminished overall survival was found by Mulder et al. and a similar independent association was found by Sutoh et al. for disease-free survival [3
]; both studies included colonic and rectal tumours and all four Astler Coller or Dukes stages. Contrary to these results Kim et al. found no association between GST Pi expression and overall survival in patients with stage IV colorectal cancer, all of whom had been treated by 5-FU/oxaliplatin palliative chemotherapy [21
]. Given that tumour stage is the most clearly established prognostic variable in colorectal cancer it is reasonable to expect that other prognostic factors, including biomarkers, may behave differently within different stages and also that they may behave differently between the colon and rectum or even at different sub-sites. In our study we focussed on stage C colonic tumours because of the known beneficial effects of chemotherapy in this patient group and the desire to find biomarkers which would allow more refined targeting of patients most likely to benefit from chemotherapy and to avoid costly and possibly toxic treatment of those who are unlikely to benefit.