This study applying the novel Bayes decision rule showed the RFS significance of the combination of CD4+ and FOXP3+ T-cell densities identified by IHC in T3/T4a stage II CRC patients. Adjuvant chemotherapy may be considered for patients with a combination of low CD4+ TIL density and low FOXP3+ TIL density, and fluoropyrimidines in combination with oxaliplatin regimens are recommended for such candidates.
We distinguished recurrence in T3/T4a stage II CRC by applying a novel discrete Bayes classifier. This classifier is unique and can deal with numerical and non-numerical data based on the Bayes decision theory using the posterior probability [21]. To find the optimal combination of markers and evaluate distinguishability, we applied the LOO method. Using this estimation method, we eliminated the arbitrariness in recurrence classification and obtained an objective probability distribution. Our results showed that CD4 and FOXP3 for two combination markers and CD4, FOXP3, and histologic grade for the three combination markers, respectively, could predict recurrence most significantly. If a clinician determines a patient as being recurrent, the clinician can differentiate the recurrence with a sensitivity of 68 and 71%, a specificity of 69 and 67%, and a diagnostic accuracy rate of 69 and 68%, for two and three combination markers, respectively (Table 3).
As shown in Table S2, the classification performance for the absence group of adjuvant therapy using two and three optimal combinations of markers against test samples showed sensitivities of 80 and 90% and a specificity of 70 and 68%, respectively. The discrete Bayes classifier resulted in high sensitivity, an important indicator for predicting CRC recurrence. Since we attempted not to miss recurrence, we adopted an evaluation standard in which sensitivity is maximised by maintaining the specificity at a certain level.
Consistent with our previous report [20], the present study confirmed the usefulness of CD4+ T-cell density itself as a prognostic factor (Fig. 2A-<link rid="fig1"><link rid="fig1">1</link></link>, 2B-<link rid="fig1"><link rid="fig1">1</link></link>, and 2C-<link rid="fig1"><link rid="fig1">1</link></link>), and the combination of low CD4+ cell infiltration and low FOXP3+ cell infiltration was a prognostic factor for low RFS in T3/T4a stage II CRC patients (Fig. 2A-4, 2B-4 and 2C-4).
We first showed that a high CD4+ T cell density was associated with a longer RFS. To our knowledge, this is the first report to suggest the usefulness of intratumoural CD4+ T-cell infiltration as a positive RFS factor in T3/T4a stage II CRC. It has been reported that in lung, renal, prostate, and breast cancers, CD4+ T-cell density is a negative prognostic factor [25–28]. The reason for this discrepancy between CRC and other cancers remains ambiguous. However, the function of CD4+ T cells within the tumour microenvironment, such as in immune response activation or immunosuppression, may be different depending on the cancer types can be one of the reasons, as we reported previously [20].
Poor clinical outcomes in many cancers are associated with the invasion of abundant FOXP3+ cells into the tumour tissue [29]. However, contradictory results have been reported in CRC; cases with high FOXP3+ T-cell infiltration showed a better prognosis in some studies [30–32]. In our results, a high FOXP3+ cell density was significantly associated with improved prognosis (Fig. <link rid="fig2"><link rid="fig2">2</link></link>A-<link rid="fig2">2</link> and <link rid="fig2">2</link>B-<link rid="fig2">2</link>).
Finally, we found that the combination of CD4+ and FOXP3+ cell densities was a precise prognostic marker (Fig. 2A-4, 2B-4 and 2C-4). CD4+ T cells that express the FOXP3 transcription factor function as Treg cells suppress effective immune responses against cancer cells [33, 34]. FOXP3+/CD4+ T cells are both functionally and phenotypically heterogeneous; FOXP3+/CD4+ T cells can be fractionated based on their expression levels of FOXP3 and CD45RA into FOXP3low/CD45RA+ naive Treg cells, FOXP3high/CD45RA− effector Treg cells, and FOXP3low/CD45RA− non-suppressive T cells that can secrete proinflammatory cytokines [34–36]. Our results indicate that the infiltration of only one type of immune cell, such as CD4+ or FOXP3+ cells, might be sufficient for a suitable tumour microenvironment to prevent recurrence.
In the QUASAR trial, the chemotherapy group (5-FU + LV\(\pm\)levamisole) had better RFS and OS compared to the surgery-only group in stages I, II, and III CRC, but no significant advantage in stage II CRC itself [37]. Furthermore, IMPACT B2 trial [38] and its meta-analysis [39, 40] and SEER database review [41] reported no significant advantage in the chemotherapy group (5-FU+LV) in RFS and OS for patients with T3N0 CRC. In addition, the SACURA trial showed no significant advantage for applying adjuvant chemotherapy (UFT) to stage II colon cancer in terms of RFS [42]. Hence, it is difficult to apply adjuvant therapy to every stage II CRC patient without considering recurrence risks.
Based on the American Society of Clinical Oncology (ASCO) and the European Society for Medical Oncology (ESMO) guidelines, it is recommended to apply adjuvant chemotherapy in high-risk stage II CRC patients. Poor prognostic features such as fewer than 12 retrieved lymph nodes, T4 lesions (defined as adherence to or invasion of local organs), tumour presentation with perforation and poorly differentiated adenocarcinoma/signet ring cell carcinoma/mucinous carcinoma, and T4, poorly differentiated adenocarcinoma/undifferentiated cancer, lymphovascular invasion, paranerve infiltration, intestinal obstruction/perforation, and fewer than 12 retrieved lymph nodes from the ASCO 2004 [43] and ESMO guidelines [15], respectively, are known as high-risk factors for stage II CRC. Thus, it is essential to identify prognostic factors that can influence the decision to apply adjuvant therapy in T3 and T4a stage II CRC. Our CD4 and FOXP3 combination could be a novel prognostic factor.
One of the main limitations of this study was the use of a small amount of sample data from a single institution. The LOO method was used to obtain highly accurate results, even with few samples. Each sample was used as a training sample and as a test sample, but never as a training sample or a test sample at the same time, and the LOO method was excellent in terms of use efficiency and independence. Therefore, high estimated accuracy is expected. There is room for developing more accurate criteria by assessing more cases with multiple institutions to establish more precise criteria, and we aim to carry out a prospective study to verify the validity of this study's findings.