The eligibility criteria for part 1 were as follows: (1) pathologically confirmed solid tumor refractory to the standard treatment, (2) Eastern Cooperative Oncology Group performance status (ECOG PS) of 0 or 1, (3) age ≥ 20 years, (4) adequate organ function, (5) written informed consent from each patient, and (6) any evaluable lesions. The exclusion criteria included (1) active viral infections; (2) concomitant active diseases including synchronous cancer and diabetes mellitus (DM), except for DM secondary to pancreatic cancer; and (3) any prior history of biguanide use in the year before enrollment, cancer immunotherapy, and transplantation therapy.

All the study procedures were performed after written informed consent was obtained from each patient. This study was approved by the institutional review board of each participating hospital in accordance with the Declaration of Helsinki. The protocol summary, including participant and intervention details, has been described previously [18].

Three cohorts were organized based on metformin dose level in the part 1. Metformin administration in cohort 1 was started at a dose of 750 mg/day, we had defined as a clinically reasonable dose, on days 1–15 in each cycle, increased to 1,500 (cohort 2) and 2,250 mg/day (cohort 3) with fixed dose of nivolumab (3 mg/kg) on day 1. Based on the result of DLT evaluation in the part 1, the optimal dose of metformin was selected and administered in part 2. The treatment cycle was repeated every 2 weeks, continued until disease progression or unacceptable adverse effects in both of parts.

DLT was evaluated for the first two cycles at each cohort using the Common Terminology Criteria for Adverse Events v. 4.0 (CTCAE v4.0). The MTD and the recommended doses were determined using the conventional 3 + 3 cohort method. The cohort was elevated if none of the three patients developed DLT (Table 1). However, if one of the three patients experienced a DLT, the other three patients were treated at the same dose level. The MTD was defined as a dose level that produced any DLT in ≥ two of six patients. Patients were excluded and replaced from the DLT evaluation population, in cases of without occurring DLT events (1) lower dose intensity, defined as a single dose multiplied by the number of doses of metformin actually administered during the period, compared to 70% of the planned dose intensity during the period; (2) nivolumab not administered twice in the period; and (3) insufficient patient follow-up period.

PD-L1 expression in tumor specimens, irrespective of the time when obtained, was analyzed by immunohistochemical (IHC) staining with anti-PD-L1 antibodies 28-8 and 22C3.

Blood samples were collected at three points: before, and 2 h, and 4 h after metformin administration, on any day from Cycle 1 Day 8 to the day before nivolumab administration in Cycle 2 Day 1, and after confirming the stable administration of the prescribed dose of metformin for at least three consecutive days for pharmacokinetic analyses of metformin. The plasma concentrations were measured using high-performance liquid chromatography (HPLC).

Efficacy was determined by evaluating the overall response rate (ORR) according to the Response Evaluation Criteria for Solid Tumors (ver. 1.1). Progression-free survival (PFS) and overall survival (OS) were defined as the time from Cycle 1 Day 1 to the date of progressive disease or death by any cause, or until death, respectively.

The primary endpoint of part 1 is safety, evaluated in terms of the MTD and DLT, whereas the pharmacokinetics of metformin and adverse events profiling are endpoints of parts 1 and 2. The secondary endpoint was efficacy in terms of tumor shrinkage effects, PFS, and effective blood concentration of metformin. After identifying a suitable dose, an expanded assessment of safety and preliminary efficacy at the optimal dose was performed in part 2. OS was also added as a secondary endpoint after protocol amendment.

The data from all patients who received at least one dose of study drug were included in the analysis. Statistical analysis was performed as follows according to the characteristics of the data scale. For categorical data, frequencies and rates were calculated. The Clopper–Pearson method was used to estimate confidence interval for rate. For continuous data, summary statistics were estimated. For survival time data, the Kaplan–Meier method was used to estimate cumulative survival rate, and the Greenwood method was used to estimate confidence interval. Median survival time and confidence interval were also estimated. Scatter plot, box-and-whisker plot, waterfall plot, and Kaplan–Meier curve etc. were used for graphical presentation as appropriate. Statistical hypothesis testing was not applied in this study. The confidence coefficient for estimating two-tailed confidence interval was set at 95%. Statistical data analysis was performed using SAS software (version 9.04).