All participants provided written informed consent prior to enrollment, and the Institutional Review Boards at the Tamana Regional Health Medical Center (Tamana, Kumamoto, Japan) approved the study protocol. All methods and procedures were consistent with Good Clinical Practice, the Declaration of Helsinki and local laws. In total, 55 patients with histologically and clinically diagnosed stage IV colorectal carcinoma, based on the unified Tumor-Node-Metastasis criteria (30), were enrolled at Tamana Regional Health Medical Center between July 2014 and July 2017. The specific inclusion and exclusion criteria were a performance status of ≥2 and <2, respectively. Among the patients with colorectal carcinoma, there were 21 men and 34 women, who ranged in age from 28 to 96 years, with a mean age of 65.7±14.8 years (Table I). The patients were treated with chemotherapy using XELOX (CapeOX) (1,200-1,800 mg capecitabine for 14 days continuously, followed by a 7-day rest period, and 85 mg/m² oxaliplatin at a 3-week interval) + bevacizumab (7.5 mg/m² at a 3-week interval). A total of 3 weeks was regarded as one cycle. Treatment was repeated until the cancer progression was confirmed by imaging. The patients inhaled hydrogen gas for 3 h daily at their own homes through a cannula or mask, rented or purchased by themselves, connected to a Hycellvator ET 100 (Helix Japan, Co., Ltd., Tokyo, Japan) (Fig. 1A). None of the patients reported any complaints regarding the daily 3-h hydrogen gas inhalation. Peripheral blood (10 ml) was collected from the patients prior to and 3 months after treatment with hydrogen gas.

The Hycellvator ET 100 (Helix Japan, Co., Ltd.) generates 1.67 l/min hydrogen gas (hydrogen purity, 99.99%) by electrolysis. As measured by gas chromatography at Kureha Special Laboratory (Iwaki, Fukushima, Japan), the gas generated consisted of 680,000 ppm hydrogen gas and 320,000 ppm oxygen gas. Recently, hydrogen gas inhalation was used in patients with post-cardiac arrest syndrome, and adverse events were not observed (31). Furthermore, no adverse events were observed in the 55 patients who inhaled hydrogen gas for 3 months in the present study.

Briefly, Ficoll-Hypaque solution (20 ml) was placed into a 50-ml conical centrifuge tube using a sterile pipette. Anti-coagulated blood (10 ml) mixed with an equal volume of PBS was then slowly layered over the Ficoll-Hypaque solution by gently pipetting down the side of the tube. Subsequently, samples were centrifuged for 40 min at 400 × g and 22°C for 30–40 min with no braking. Mononuclear cells that accumulated at the interface between the plasma (upper) and Ficoll-Hypaque layers (bottom) were carefully recovered using a Pasteur pipette and transferred to a 15-ml conical tube. Cells were then analyzed on a BD FACSCalibur (Nippon Becton Dickinson, Tokyo, Japan) with BD CellQuest software (version 5.1), using anti-CD57 conjugated to fluorescein isothiocyanate (clone NK-1; cat. no. 347393; Nippon Becton Dickinson), mouse anti-human CD27 conjugated to APC (clone M-T271; cat. no. B09983; Beckman Coulter, Tokyo, Japan), mouse anti-human PD-1 conjugated to PE (clone EH12.1; cat. no. 557946; Nippon Beckton Dickinson) and mouse anti-human CD8 conjugated to PerCP (clone SK1; cat. no. 347314) (BD Pharmingen, San Jose, CA, USA), incubated at 4°C for 30 min, following blocking with 1% γ-globulin for 15 min at 4°C. To determine the independent contributions of each marker to PFS and OS, FACS data were used to stratify patients based on the proportion of early, intermediate, terminal, and end PD-1⁺ and PD-1⁻ CD8⁺ T cells. All blood samples obtained from the patients were transferred to SRL, Inc. (Tokyo Japan), where lymphocyte separation and flow cytometry were performed. Therefore, the status of the laboratory data, reliable protocols and flow cytometry assays were certified by SRL, Inc., one of the most reliable clinical laboratory centers in Japan. The flow cytometry data was analyzed using SPSS version 19.0 for Windows (IBM Corp., Armonk, NY, USA).

The primary endpoints were PFS and OS time, which were measured from the date of randomization to the first recurrence and mortality regardless of cause, respectively. Patients were monitored by dynamic computed tomography or magnetic resonance imaging every 3 months from the baseline up to 60 months and every 3–6 months thereafter. Two independent and blinded radiologists, each with >5 years of experience, reviewed all scans at each site. In cases of discord, the two radiologists reviewed the images to reach the same conclusion following discussion. Adverse events were classified and graded every 2 months according to the Common Terminology Criteria for Adverse Events version 3.0 (National Cancer Institute, Bethesda, MD, USA) (32) from the day of consent until the end of the study at least 30 days after the treatment. Multiple events were counted once for each patient, of which the most severe was noted.

Testing of the significance of the differences between groups was performed using the χ² test. In case of persistent abnormal distribution, the linear correlation between two continuous variables was tested with the Spearman correlation coefficient. Receiver operating characteristic (ROC) analysis was used to determine optimal cut-off values for continuous variables. The ROC curve shows 1-specificity on the x-axis and sensitivity on the y-axis. The optimal cut-off value is calculated by maximizing the sensitivity and specificity across various cut-off points on the ROC curve. The probability of survival was estimated by the Kaplan-Meier method, and differences in survival were evaluated by the log-rank test. Prognostic factors were tested by univariate and multivariate Cox regression. All statistical analyses were performed using SPSS version 19.0 for Windows (IBM Corp.). Differences were considered statistically significant at P<0.05.

In the normal state, early CD8⁺ T cells abundantly express PD-1, which gradually diminishes with differentiation into terminal CD8⁺ T cells. However, loss of PD-1 is delayed in patients with cancer and may result in a poor prognosis. Hence, Cox proportional-hazards regression analysis was used to identify PD-1^+/− CD8⁺ T cell subsets (Fig. 1B) and clinicopathological factors [age, sex, primary tumor (T), regional lymph nodes (N), distant metastasis (M) and histology] associated with PFS and OS in patients with stage IV colorectal cancer. By univariate analysis of 18 factors, including 6 clinicopathological factors, terminal PD-1⁺ CD8⁺ T cells were found to be significantly associated with poorer PFS [hazard ratio (HR), 1.239; 95% confidence interval (CI), 1.106–1.389; P<0.0001] and OS (HR, 1.183; 95% CI, 1.066–1.314; P=0.002), as were end PD-1⁺ CD8⁺ T cells (PFS: HR, 1.296; 95% CI, 1.053–1.595; P=0.015; OS: HR, 1.333; 95% CI, 1.103–1.610; P=0.003). By contrast, early PD-1⁻ and end PD-1⁻ CD8⁺ T cells were associated with better (HR, 0.961; 95% CI, 0.926–0.997; P=0.036) and worse (HR, 1.044; 95% CI, 1.005–1.084; P=0.025) OS, respectively. The univariate analysis data of the other 14 factors for PFS and OS, respectively, were as follows: Age, P=0.270 and P=0.886; sex, P=0.894 and P=0.398; T factor, P=0.332 and P=0.664; N factor, P=0.080 and P=0.150; M factor (the univariate analysis could not be performed as all patients had distant metastasis); histology, P=0.503 and P=0.184; early CD8⁺ T cells, P=0.953 and P=0.273; early PD-1⁺ CD8⁺ T cells, P=0.757 and P=0.560; intermediate CD8⁺ T cells, P=0.434 and P=0.560; intermediate PD-1⁺ CD8⁺ T cells P=0.799 and P=0.505; intermediate PD-1⁻ CD8⁺ T cells P=0.137 and P=0.099; terminal CD8⁺ T cells, P=0.453 and P=0.595; terminal PD-1⁻ CD8⁺ T cells, P=0.681 and P=0.886; and end CD8⁺ T cells, P=0.285 and P=0.566.

Based on multivariate Cox regression, terminal PD-1⁺ CD8⁺ T cells were more strongly associated with PFS (HR, 1.239; 95% CI, 1.106–1.389; P<0.0001) and OS (HR, 1.136; 95% CI, 1.019–1.266; P=0.022) than others. The multivariate data of the other 3 factors were as follows: Early PD-1⁻ CD8⁺ T cells, P=0.677 for PFS and P=0.352 for OS; end PD-1⁺ CD8⁺ T cells, P=0.274 for PFS; HR, 1.247; 95% CI, 1.007–1.543; P=0.043; and end PD-1⁻ CD8⁺ T cells, P=0.561 for PFS and P=0.206 for OS). Accordingly, patients were stratified based on the abundance of terminal PD-1⁺ CD8⁺ T cells, using cut-off values of 8.18 and 6.81% for PFS and OS, respectively, as determined from receiver operating characteristic curves (Fig. 2A and B). There were no significant differences in clinicopathological factors between the patients with high and low terminal PD-1⁺ CD8⁺ T cells (Table I). The resulting stratified Kaplan-Meier survival curves revealed that PFS (log-rank test, P=0.001) and OS (log-rank test, P=0.008) were significantly worse for patients with terminal PD-1⁺ CD8⁺ T cells higher than the cut-off (Fig. 2C and D). The median PFS time was 18 months for the patients with a high-terminal PD-1⁺ CD8⁺ T cell ratio, but was not reached (>50% of patients remained alive) for those with a low-terminal PD-1⁺ CD8⁺ T cell ratio, while OS was 18 months for the patients with a high-terminal PD-1⁺ CD8⁺ T cell ratio and 46 months for those with a low-terminal PD-1⁺ CD8⁺ T cell ratio.

Molecular hydrogen was reported to activate PGC-1α (28), which enhances mitochondrial activity (29), thereby rescuing exhausted CD8⁺ T cells with inactive mitochondria following progressive loss of PGC-1α (15). Therefore, the present study investigated whether hydrogen gas alters the proportion of PD-1^+/− CD8⁺ T cell subsets and whether alterations, if any, are associated with prognosis in patients with stage IV cancer. Notably, hydrogen gas reduced the proportion of early, intermediate, terminal and end PD-1⁺ CD8⁺ T cells in 27 (49.1%), 28 (50.9%), 35 (63.6%) and 32 (58.2%) out of 55 patients, respectively. Conversely, hydrogen gas enhanced the proportion of early, intermediate, terminal and end PD-1⁻ CD8⁺ T cells in 32 (58.2%), 27 (49.1%), 39 (70.9%) and 31 (56.4%) patients, respectively. Univariate analysis showed that the ratio of the proportion of intermediate PD-1⁺ CD8⁺ T cells following treatment with hydrogen gas to that prior to treatment (intermediate PD-1⁺ CD8⁺ T cell ratio) was significantly associated with shorter PFS (HR, 2.286; 95% CI, 1.284–4.070; P=0.005) and OS (HR, 2.398; 95% CI, 1.384–4.156; P=0.002) times. Similar associations were observed for the intermediate PD-1⁻ CD8⁺ T cell ratio (PFS: HR, 2.286; 95% CI, 1.284–4.070; P=0.005; OS: HR, 2.398; 95% confidence interval, 1.384–4.156; P=0.002) and terminal PD-1⁺ CD8⁺ T cell ratio (PFS: HR, 6.459; 95% CI, 2.384–17.50; P<0.0001; OS: HR, 4.158; 95% CI, 1.718–10.06; P=0.002). By contrast, the early PD-1⁻ CD8⁺ T cell ratio was predictive of worse OS time (HR, 2.398; 95% CI, 1.384–4.156; P=0.002), whereas the terminal PD-1⁻ CD8⁺ T cell ratio was significantly associated with increased overall survival time (HR, 0.002; 95% CI, 0.000–0.131; P=0.004). For the multivariate analysis, the terminal PD-1⁺ CD8⁺ T cell ratio was found to be an independent predictor of poor PFS (HR, 6.459; 95% CI, 2.384–17.50; P<0.0001) and OS (HR, 2.957; 95% CI, 1.185–7.378; P=0.020). Based on these results, patients were stratified as having a high and low terminal PD-1⁺ CD8⁺ T cell ratio using cut-off values of 0.87 and 0.77% for PFS and OS, respectively (Fig. 3A and B). There were no significant differences in clinicopathological factors between the patients with high and low terminal PD-1⁺ CD8⁺ T cell ratio (Table II). The resulting stratified PFS and OS curves are plotted in Fig. 3C and D, and show that patients with a low terminal PD-1⁺ CD8⁺ T cell ratio have significantly increased PFS (P<0.0001) and OS (P=0.004) times compared with those with high ratios. The median follow-up time was 13 months for the patients with a high-terminal PD-1⁺ CD8⁺ T cell ratio, but was not reached (>50% of patients remained alive) for those with a low-terminal PD-1⁺ CD8⁺ T cello ratio, while that of OS was 15 months for the patients with a high-terminal PD-1⁺ CD8⁺ T cell ratio and 46 months for those with a low-terminal PD-1⁺ CD8⁺ T cell ratio.

Based on the hypothesis that hydrogen gas may activate mitochondrial function and thereby convert exhausted terminal PD-1⁺ CD8⁺ T cells into active terminal PD-1⁻ CD8⁺ T cells, the present study investigated whether the change in the abundance of the latter impacts prognosis in patients treated with hydrogen gas. Thus, patients were stratified by the terminal PD-1⁻ CD8⁺ T cell ratio based on cut-off values of 1.01 and 1.02 for PFS (Fig. 4A) and OS (Fig. 4B), respectively. There were no significant differences in clinicopathological factors between the patients with high- and low-terminal PD-1⁻ CD8⁺ T cell ratios (Table III). The resulting stratified survival curves are plotted in Fig. 4C and D, respectively, and indicate that patients with high ratios have significantly increased PFS (P=0.004) and OS (P=0.024) times compared with those with low ratios. The median follow-up time was 15 months for the patients with a low-terminal PD-1⁻ CD8⁺ T cell ratio, but was not reached (>50% of patients remained alive) for those with a high-terminal PD-1⁻ CD8⁺ T cell ratio, while that of OS was 16 months for the patients with a low-terminal PD-1⁻ CD8⁺ T cell ratio and 52 months for those with a high-terminal PD-1⁻ CD8⁺ T cell ratio. Furthermore, hydrogen gas treatment resulted in a significantly longer PFS time (P=0.014) and a generally longer, but non-significant, OS time (P=0.165) in patients with a high level of terminal PD-1⁻ CD8⁺ T cells compared with that in patients with a low level, although there was no significant difference between the groups prior to treatment (Fig. 5A and C).

As aforementioned, hydrogen gas reduces the proportion of terminal PD1⁺ CD8⁺ T cells, but increases the abundance of terminal PD1⁻ CD8⁺ T cells, and the magnitude of these changes is strongly associated with the prognosis. Notably, the terminal PD1⁺ CD8⁺ T cells were significantly and inversely correlated with the terminal PD1⁻ CD8⁺ T cells (Fig. 6A and B), suggesting that the dynamic balance between these subsets contributes strongly to prognosis in patients with advanced colorectal carcinomas. Therefore, patients were stratified based on all possible high/low combinations of these subsets [category (Cat) 1–4; Cat 1: Patients with low PD1⁺ and high PD1⁻ terminal CD8⁺ T cells; Cat 2: Patients with high PD1⁺ and high PD1⁻ terminal CD8⁺ T cells; Cat 3: Patients with low PD1⁺ and low PD1⁻ terminal CD8⁺ T cells; Cat 4: Patients with high PD1⁺ and high PD1⁻ terminal CD8⁺ T cells; Fig. 6A and B]. Kaplan-Meier analysis revealed that Cat 1 patients experienced significantly longer PFS times than all other groups, whereas Cat 3 patients experienced significantly longer OS times than the others. By contrast, Cat 4 patients experienced significantly worse PFS (Fig. 6C) and OS (Fig. 6D) times than others. Hydrogen gas also increased the number of patients with low PD1⁺ terminal CD8⁺ T cells (Cat 1 and 3), but decreased the number of patients with high PD1⁺ terminal CD8⁺ T cells (Cat 2 and 4), leading to an improved prognosis in patients with stage IV colorectal carcinoma (Fig. 6A and B; Table IV).