Programmed death-1 (PD-1) is one of several proteins that serve as checkpoints for immune cell activation. Primed T cells expressing PD-1 are inactivated by engagement of its ligand programmed death ligand 1 (PD-L1) expressed on tumor cells, thereby promoting tumor growth and metastasis.1 The anti-PD-1 antibody nivolumab was recently approved for previously treated non-small cell lung cancer (NSCLC). Several phase III studies have shown that nivolumab confers a superior survival benefit compared with docetaxel, a standard second-line cytotoxic agent, in both patients with squamous and non-squamous NSCLC.2 3 Based on these results, nivolumab is recommended as a standard second-line regimen in the recent guidelines for NSCLC.4

The proportion of tumor cells expressing PD-L1, as assessed by immunohistochemistry, is considered a predictive biomarker for the efficacy of anti-PD-1 therapy. However, objective response rates of only 19% and 37% have been reported for patients with high PD-L1-expressing (≥10% cells) squamous and non-squamous NSCLC, respectively, following nivolumab treatment.2 3 Several case reports have described a phenomenon known as pseudoprogression, in which initial tumor progression is followed by shrinkage during continuous treatment with immune-checkpoint inhibitors5 6; notably, this is not seen with cytotoxic drugs or other targeted agents. For these reasons, it is currently difficult to accurately predict or assess the early response to immune-checkpoint inhibitors. Additionally, these inhibitors are associated with debilitating side effects and are very expensive (eg, nivolumab is currently US$15,329 per month). Therefore, there is an urgent need to identify new predictive biomarkers of the early response to immune-checkpoint inhibitors, which will help to curtail ineffective and potentially toxic therapy and reduce unnecessary costs.

Currently, [¹⁸F]-fluoro-2-deoxy-d-glucose (¹⁸F-FDG) positron emission tomography (PET) is widely used for the diagnosis of malignancy, and it has recently been used to assess the early response of patients with NSCLC to chemotherapy,7 8 targeted agents9 and chemoradiotherapy.10 11 Diffusion-weighted MRI has also been used to assess the early response to chemotherapy in patients with lung cancer.7 The apparent diffusion coefficient (ADC) is a commonly used functional parameter and surrogate marker of tumor cellularity and proliferation in MRI,12 13 and a change in ADC has been reported to precede a reduction in tumor size.14 Furthermore, integrated PET/MRI technology has recently become available and can provide useful information about metabolic activity and tumor cellularity simultaneously. In the present study, we sought to determine whether changes in integrated ¹⁸F-FDG PET/MRI parameters after the first 2 weeks of nivolumab therapy could accurately discriminate between responders and non-responders, as assessed by Response Evaluation Criteria in Solid Tumor (RECIST) criteria, and/or predict patient survival.

The median follow-up period for all patients was 12.6 months (range 1.3–30.9 months). Twenty patients showed disease progression during the study period, and the median PFS was 7.2 months. Kaplan-Meier curves for PFS stratified by the imaging cut-off values are shown in figure 2. Significant differences in PFS were observed between the groups stratified by Δdiameter, ΔTLG and ΔADC_mean (figure 2A–C). The median PFS was also significantly longer for patients in the ΔTLG+ΔADC_mean<16.5 group than those in the ≥16.5 group (9.0 vs 1.8 months, p<0.00001, figure 2D). Conversely, no significant difference in PFS was observed between patients with PD-L1 TPS of <1% or ≥1% (p=0.24, online supplemnetary figure S2a).

A Cox proportional hazards model was used to evaluate predictors of PFS and OS (table 2). We found that older age (≥70, HR 2.83, 95% CI 1.00 to 7.98, p=0.049) and ΔTLG+ΔADC_mean≥16.5 (HR 37.7, 95% CI 4.50 to 316.2, p=0.00082) were indicators of shorter PFS by univariate analysis, whereas ECOG-PS, number of previous regimens, clinical stage, histology and TPS were not significant. In multivariate analysis, only higher ΔTLG+ΔADC_mean remained an independent predictor of PFS.

Among the 25 patients, 4 did not undergo ¹⁸F-FDG PET/MRI at 8 weeks after treatment initiation because of early disease progression. Of the remaining 21 patients, 4 showed PD in the CT images according to RECIST V.1.1 within 2 weeks after the 8-week ¹⁸F-FDG PET scan, which left 17 patients with no progression at 10 weeks for further analysis. Seven patients showed either increased SUV_max (≥25%) within the tumor region defined in the baseline scan or a new lesion on the 8-week ¹⁸F-FDG PET according to EORTC 1999 criteria.17 Kaplan-Meier PFS curves for ¹⁸F-FDG PET findings at 8 weeks are shown in figure 3. The median PFS for patients without relapse-associated findings was significantly longer than that of patients with PMD on ¹⁸F-FDG PET at 8 weeks (17.3 vs 4.2 months, p=0.000051).

In this prospective study of 25 patients with previously treated NSCLC, we found that changes in TLG, ADC_mean and TLG+ADC_mean between baseline and 2 weeks of nivolumab treatment were early predictive biomarkers for response, with ΔTLG+ΔADC_mean having the best discriminative power. To the best of our knowledge, this is the first report of the predictive utility of integrated ¹⁸F-FDG PET/MRI findings in this patient population, and additionally, the first report of changes in PET or MRI findings after a single dose of any immune-checkpoint inhibitor in patients with malignant tumors.

In recent years, ¹⁸F-FDG PET imaging has also been used as a predictive tool for the response to immune-checkpoint inhibitors in patients with melanoma, NSCLC and Hodgkin’s lymphoma.18–21 In those studies, the intervals between the baseline (before treatment) and first PET evaluation varied between 3–4 weeks and 1 year. The European Association of Nuclear Medicine symposium recommended that the first response evaluation PET scan should be performed after 8–9 weeks of immunotherapy, which is generally after three or four treatment cycles.22 However, two phase III studies of nivolumab for patients with NSCLC showed that the median time to response was only 2.1–2.2 months,2 3 which suggests that the recommended time of first response evaluation may be suboptimal for this patient population. Because our findings indicate that changes in integrated ¹⁸F-FDG PET/MRI parameters can successfully capture the response to nivolumab after a single dose, this method could be used to assist clinicians in designing the optimal therapeutic strategy for an individual patient, thus reducing ineffective treatment and associated drug costs.

Immune-checkpoint inhibitors have emerged as standard therapies for various types of malignant tumors. As noted earlier, a well-recognized atypical manifestation of immune-checkpoint therapy is the so-called tumor pseudoprogression. In patients with NSCLC, the reported rate of pseudoprogression is small (0.6%–4.7%)23–25; however, an increase in tumor size that does not meet the RECIST criteria of PD may be seen more frequently. Moreover, a recent study described the morphological and pathological changes after neoadjuvant nivolumab in patients with resectable NSCLC.26 In this report, despite apparent tumor enlargement on CT after preoperative administration of nivolumab, some patients exhibited major pathological responses in the resected specimens. Indeed, in our study, 4 of 16 (25%) patients with disease control (CR, PR or SD by RECIST) showed a Δdiameter of ≥10%. This phenomenon can potentially confound CT-based assessment of the early response to immune-checkpoint inhibitors.

One important problem in the use of ¹⁸F-FDG PET for the diagnosis of malignancy is false positive findings due to inflammatory disease and granulomas.27–29 Additionally, in the case of the early response to immune-checkpoint inhibitors, ¹⁸F-FDG PET can be complicated by the antitumor immune response and enhanced CD8⁺ T-cell infiltration in and around the lesions.5 26 30 In the present study, for example, we found that three of the nine responders (PR or CR) exhibited ΔTLG of >20% between baseline and 2 weeks, and three of the nine patients with PD showed a >50% increase in TLG. Therefore, it is important to discriminate between ‘metabolic’ pseudoprogression and authentic disease progression when increased ¹⁸F-FDG uptake in the lesion is observed after nivolumab treatment.

Diffusion-weighted-MRI is another imaging method used to diagnose and assess various malignant tumors. Water diffusion in tumors can be quantified by ADC, and this has recently been reported to correlate negatively with cell density12 31 32 and cell proliferation33 34 in several tumor types. Serial changes in ADC have previously been used to analyze the early response of patients with lung cancer to chemotherapy and chemoradiotherapy.7 35 However, little is known about how MRI parameters change after treatment with immune-checkpoint inhibitors. In the present study, we found that ADC_mean was significantly increased in patients with non-PD compared with PD. The early antitumor response to immune-checkpoint inhibitors is thought to reflect inflammation and edema induced by transient immune cell infiltration,36 and non-responders are therefore not expected to show changes in cell density or proliferation in the lesions. Based on these considerations, increased ADC_mean (ie, negative ΔADC_mean) could represent a change in intratumoral cell density and proliferation and may thus be a useful parameter for the assessment of early response to immune-checkpoint inhibitors.

Our study found that ΔTLG+ΔADC_mean was the most discriminative ¹⁸F-FDG PET/MRI parameter for distinguishing between patients with non-PD and PD. Two patients exhibited an increase in ΔTLG between scans of >20% and later achieved a PR (figure 1C), which represented metabolic pseudoprogression. Conversely, these patients showed a decrease in ΔADC_mean, which resulted in ΔTLG+ΔADC_mean below the cut-off value (<16.5, figure 1E). The amelioration of predictive value of ΔTLG+ΔADC_mean may result from the decreased ΔADC_mean, which counteracts the increased ΔTLG in the lesion with metabolic pseudoprogression.

Although PET/CT is recommended as a pretreatment evaluation for staging of patients with NSCLC,4 PET/MRI may have some advantages in detecting bone or brain metastases that are frequently seen in patients with NSCLC, as well as accurately combining the metabolic data obtained with the PET image with information about tumor cellularity from MRI. Furthermore, PET/MRI involves reduced exposure to radiation compared with PET/CT. Thus, serial PET/MRI could have widespread utility for evaluating the early response of various malignant tumors to immune-checkpoint inhibitors. However, PET/MRI has some limitations. First, the acquisition time of PET/MRI is longer than PET/CT (30 vs 15 min, in our institution). Second, because the cost of PET/MRI system is higher compared with PET/CT, PET/MRI systems are currently not widespread. Therefore, the predictive utility for the response to immune-checkpoint inhibitors using serial PET/CT and MRI should be verified.

Tumor PD-L1 expression level (measured as TPS) is a recognized predictive biomarker for the efficacy of anti-PD-1 therapy. However, in our study, TPS (<1% vs ≥1%) was not a significant predictor of PFS in nivolumab-treated patients. Some clinical trials of PD-L1 antibodies have reported response rates of only 30%–45%, even in patients with high tumor PD-L1 expression (≥50% TPS).37 38 Conversely, approximately 10% of patients with tumors lacking PD-L1 expression respond favorably to nivolumab.38 39 These unexpected responses suggest that tumor PD-L1 expression is highly heterogeneous, which renders PD-L1 IHC of biopsy samples an unreliable predictor of response. Indeed, PD-L1 expression has been shown to be heterogeneous within and among lesions40 41 and to change after chemotherapy.42

In the present study, we found that ΔTLG was a better predictor than ΔSUV_max of nivolumab efficacy, which is consistent with some previous studies on the ability of early phase PET to predict the response to immune-checkpoint inhibitor treatment.20 43 This finding may reflect heterogeneity in tumor infiltration of inflammatory cells in response to nivolumab treatment, which, in turn, may depend on the heterogeneity of tumor PD-L1 expression. Therefore, SUV_max after nivolumab treatment may reflect focal inflammation rather than tumor metabolic activity. Thus, despite being a simple and widespread method, measurement of SUV_max or SUL peak might be inappropriate for the early phase assessment of the tumor response to immune-checkpoint inhibitors.8 44

Our study has several limitations. The reliability of the cut-off value of this study is not certain because of its preliminary nature and the small sample size; thus, a validation cohort in a larger population is needed to confirm our findings. Furthermore, we could not accurately conclude the correlation between patients’ characteristics, such as age, ECOG-PS and tumor PD-L1 expression, and outcomes under nivolumab from this study. Additionally, we evaluated the response of patients with NSCLC to nivolumab, and the data may not be generalizable to other cancer types or immune-checkpoint inhibitors, such as cytotoxic T-lymphocyte antigen-4 targeted therapies. Further research is thus necessary to confirm our findings and to establish the utility of ¹⁸F-FDG PET/MRI for early responses to various types of immune-checkpoint inhibitors.

This prospective study showed that changes in TLG and ADC_mean measured by integrated ¹⁸F-FDG PET/MRI as early as 2 weeks after treatment initiation could have utility as predictors of the response and survival of patients with NSCLC treated with nivolumab. Although further clinical research is needed to confirm our results, our findings suggest that multidimensional functional imaging using integrated PET/MRI holds promise as a method to evaluate the activity of the intratumoral environment and predict the response to treatment with immune-checkpoint inhibitors.