Metabolic characteristics of programmed cell death‐ligand 1‐expressing lung cancer on 18F‐fluorodeoxyglucose positron emission tomography/computed tomography

Abstract Programmed cell death‐1 (PD‐1) and programmed cell death‐ligand 1 (PD‐L1) have been identified as novel targets of immunotherapy of lung cancer. In present study, we evaluated the metabolic characteristics of lung cancer by using 18F‐fluorodeoxyglucose positron emission tomography/computed tomography (18F‐FDG PET/CT) with regard to PD‐L1 protein expression. PD‐L1 protein expression was evaluated by immunohistochemistry with the antibody clone SP142 in 579 surgically resected primary lung cancer patients. Cases with less than 5% tumor membrane staining were considered negative. We examined the association between the frequency of PD‐L1 protein expression and the maximum standardized uptake value (SUVmax) in preoperative 18F‐FDG PET/CT. The cut‐off values for SUVmax were determined by receiver operating characteristic curve analyses. The SUVmax was significantly higher in nonsmall cell lung cancer (NSCLC) patients with PD‐L1 protein expression compared with those without PD‐L1 protein expression (P < 0.0001). However, there was no correlation between SUVmax and PD‐L1 protein expression in patients with neuroendocrine tumors (P = 0.6545). Multivariate analysis revealed that smoking, the presence of pleural invasion, and high SUVmax were independent predictors of PD‐L1 positivity. PD‐L1‐expressing NSCLC had a high glucose metabolism. The SUVmax in preoperative 18F‐FDG PET/CT was a predictor of PD‐L1 protein expression in patients with NSCLC.

Introduction lung cancer [10,11]. Subset analysis of CheckMate-057 showed a close correlation between PD-L1 protein expression and the efficacy of PD-1 antibodies [9]. Therefore, it is important to examine PD-L1 protein expression as well as the specific genotypes of EGFR mutations and ALK rearrangements in patients. 18 F-fluorodeoxyglucose positron emission tomography/ computed tomography ( 18 F-FDG PET/CT) scan is an essential imaging modality in diagnosis and staging of lung cancer [12,13]. Glucose metabolism in cancer tissues measured on 18 F-FDG PET/CT, is a significant biomarker to characterize cancers. Correlations between FDG uptake and biologic features of cancers such as proliferation, histologic type, tumor differentiation, and hypoxia were reported [14][15][16][17][18][19]. Moreover, associations between glucose metabolism and EGFR mutations/ALK rearrangements were also reported [20,21]. However, the relationship between glucose metabolism and PD-L1 protein expression is unclear. We recently revealed the Ki-67 index in PD-L1-positive patients was higher than in PD-L1negative patients in the analysis of 205 patients with squamous cell carcinoma (SCC) [22]. Furthermore, PD-L1 expression was significantly correlated with increased Ki-67 labeling index by multivariate analysis in patients with adenocarcinoma (ADC) [23] and significant differences in FDG uptake across histological subtypes and differentiation groups in NSCLC paralleled similar differences in the Ki-67 index [24]. Therefore, we speculated that FDG uptake might be associated with PD-L1 protein expression in patients with lung cancer.
This translational study examined PD-L1 protein expression in primary lung cancer patients who had undergone surgical resection and investigated the association between PD-L1 protein expression and the maximum standardized uptake value (SUVmax) in preoperative 18 F-FDG PET/ CT.

Patients and samples
We retrospectively examined patients with primary lung cancer who underwent complete surgical resection at the Department of Surgery and Science, Graduate School of Medical Sciences, Kyushu University. Among them, we selected patients who were investigated with chest CT and 18 F-FDG PET/CT before surgery. Four hundred and forty-one patients with ADC and 103 patients with SCC until December 2015, and four patients with large cell carcinoma (LCC), 16 patients with small cell lung carcinoma (SCLC) and 15 patients with large cell neuroendocrine carcinoma (LCNEC) until June 2016 were included in this study. A total of 579 paraffin-embedded specimens were retrieved from the registry of the Department of Anatomic Pathology, Graduate School of Medical Sciences, Kyushu University. Patients with a history of SCC of the head and neck or esophagus were excluded from this study because of the possibility of metastatic SCC from these cancers. Patients who received neoadjuvant therapy were also excluded because of inconsistency in the expression of PD-L1 on tumor cells before and after neoadjuvant chemotherapy [25]. Clinicopathological features, including age at surgery, gender, smoking status, pathologic tumornode-metastasis stage (seventh edition of the Lung Cancer Staging System), pleural or lymphovascular invasion and SUVmax were examined. Tumor differentiation, histological subtype of ADC (World Health Organization Classification 2015), and EGFR mutation status were also examined by subset analysis. EGFR status was determined in tumor tissues using the peptide nucleic acid-locked nucleic acid polymerase chain reaction clamp method (Mitsubishi Chemical Medience, Tokyo, Japan) in 377 specimens [26]. Clinical information and follow-up data were obtained from the patients' medical records. We obtained informed consent from each patient, and this study was approved by our institutional review board (Kyushu University, IRB No. 28-100).

Chest CT
Chest CT scanning was performed in a supine position during inspiratory breath-hold using various multi-detector row scanners: Aquilion 4 (Toshiba), Aquilion 64 (Toshiba), Aquilion ONE (Toshiba), Aquilion ONE Vision (Toshiba), SOMATOM Plus4 Volume Zoom (Siemens), Briliance CT (Phillips), and Briliance iCT (Phillips). Imaging parameters for thin-section CT were as follows: tube voltage 120 kVp; tube current 100-500 mA; scan field of view 320-360 mm; and slice thickness 2 mm. A real exposure control (Toshiba) of automatic exposure control (Siemens and Phillips) was added in each study. All CT data sets were transferred to a Picture Archiving and Communication System, which was accessible for workstations (Volume Analyzer, Synapse-Vincent, Fujifilm, Tokyo) with a specialized application for the lungs. and acquisition time per bed position was 3 min for Discovery STE and 2 min for Biograph mCT. We reconstructed PET images using the ordered-subset expectation-maximization method (VUE Point Plus) with two full iterations of 28 subsets for the Discovery STE and iterative True-X algorithm and TOF (Ultra HD-PET) with two full iterations of 21 subsets. The True-X algorithm incorporates an additional specific correction for the pointspread function. The full-width at half-maximum values of the Discovery STE and Biograph mCT were 5.2 and 4.4 mm, respectively. A low-dose 16-slice CT (tube voltage 120 kV; effective tube current 30-250 mA, Discovery STE) and a low-dose 32-slice CT (tube voltage 120 kV; use of angular and longitudinal dose modulation, CAREDose4D ® , Biograph mCT) from the vertex to the proximal thigh were performed for attenuation correction, and for determining the precise anatomic location of lesions before acquisition of PET images. CT scans were reconstructed by filtered back projection into 512 × 512 pixel images with slice thickness of 5 mm to match the PET scan. FDG uptake in lesions was evaluated using SUVmax, calculated by dedicated workstation for each scanner.

Immunohistochemical analysis
Immunohistochemistry was performed in 579 cases of surgically resected primary lung cancer using formalinfixed tissue sections according to our PD-L1 immunohistochemistry protocol as described previously [27,28].
The primary antibody was an antihuman PD-L1 rabbit monoclonal antibody (clone SP142, dilution 1:100; Spring Bioscience, Ventana, Tucson, AZ). Carcinoma cells showing membranous staining for PD-L1 were evaluated as positive cells. The proportion of PD-L1-positive cells was independently estimated as the percentage of total carcinoma cells in whole sections by three investigators (K.T., M.K., and G.T.). If the independent assessments did not agree, the slides were reviewed by all three investigators together to achieve consensus. The consensus judgments were adopted as the final results. Cases with <5% tumor membrane staining were considered negative in this study. Sections from human placentas were used as positive controls.

Statistical analysis
Associations between PD-L1 protein expression and patient characteristics were analyzed using Fisher's exact test, and univariate and multivariate analyses of the relationship between PD-L1 protein expression and other patient characteristics were performed by logistic regression analysis with backward elimination method. Cut-off values for SUVmax were determined by receiver operating characteristic curve analyses. We examined the association between the frequency of PD-L1 protein expression and SUVmax in preoperative 18 F-FDG PET/CT using Student's t-test. Correlations between the proportion of PD-L1positive carcinoma cells and SUVmax in preoperative 18 F-FDG PET/CT were assessed using Spearman's correlation coefficient test. All statistical analyses were performed by JMP Statistical Discovery Software (v11.0; SAS Institute, Cary, NC, USA). P values < 0.05 were statistically significant.

Results
Association between PD-L1 protein expression and SUVmax in patients with lung cancer A total of 579 patients with primary lung cancer (441 with ADC, 103 with SCC, four with LCC, 16 with SCLC and 15 with LCNEC) who underwent complete surgical resection were included in the present study (Table 1). Three hundred and thirty-eight (58.4%) patients were male, and 354 (61.1%) were smokers; the median age of the whole study population was 69 years (range, 36-89 years). Figure 1 and Figure S1 show representative thin-section CT, PET/CT, and immunohistochemistry staining for PD-L1 in patients without or with PD-L1 protein expression. Table 2

Univariate and multivariate analyses of the relationship between PD-L1 protein expression and other patient characteristics
We examined the association between PD-L1 protein expression and other patient characteristics, especially SUVmax. We determined the preferable SUVmax cut-off level using receiver operating characteristic curve analyses (Fig. 3, Figure S4). Multivariate analysis revealed that SUVmax in PET/CT was a predictor of PD-L1 protein expression in patients with lung cancer, especially NSCLC (Table 3). Additionally, smoking and the presence of pleural invasion were also predictors of PD-L1 protein expression. In the subset analysis of ADC and SCC, PD-L1 protein expression was also higher in patients with high SUVmax (Tables S1, S2). Table S3 shows the frequency of PD-L1 protein expression according to smoking history and SUVmax in PET/CT. The frequency of PD-L1 protein expression was very high in patients with smoking history (or ≥30 pack years in cases of SCC because almost all patients with SCC were smokers) and high SUVmax.

Discussion
We examined the relationship between PD-L1 protein expression and other patient characteristics. Glucose metabolism was generally higher in patients with PD-L1 protein expression than those without PD-L1. Additionally, multivariate analysis revealed that smoking, the presence of pleural invasion, and high SUVmax in PET/CT were predictors of PD-L1 protein expression in patients with lung cancer, especially NSCLC. Glucose metabolism of cancer cells measured on 18 F-FDG PET/CT is a significant biomarker for metabolic characteristics of cancer cells, and correlates with important features such as proliferation, histologic type, tumor differentiation, and hypoxia [14][15][16][17][18][19]. Moreover, the correlation between glucose metabolism and EGFR mutations/ALK rearrangements was also reported [20,21]. Recently, a correlation between metabolic information of FDG-PET and tissue expression of immune markers in patients with NSCLC who are candidates for upfront surgery was reported [29]. In that paper, they found a statistically significant correlation between SUVmax and SUVmean with the expression of CD8-tumor infiltrating lymphocytes (TILs) and PD-1-TILs, but no correlation between SUVmax and SUVmean with PD-L1 tumor expression. We hypothesize they might find a positive correlation between SUVmax and SUVmean with PD-L1 tumor expression if they performed a larger analysis. To the best of our knowledge, the present report is the first to show a statistically significant association between metabolic imaging parameter SUVmax and PD-L1 tumor protein expression in surgically resected lung cancer. PD-L1-expressing NSCLC demonstrated high SUVmax, suggesting PD-L1 protein expression is related to malignant features with high glucose metabolism.
Although 18 F-FDG PET/CT is valuable for evaluating PD-L1 protein expression in NSCLC, a mechanistic link between FDG uptake and PD-L1 protein expression remains unclear. The mechanism of FDG uptake within lung cancer cells involves glucose metabolism, hypoxia, angiogenesis, and the mammalian target of rapamycin (mTOR) signaling pathway; mTORC1 activity affected the amount of FDG uptake within lung cancer cells [14]. Moreover, AKT-mTOR pathway activation increased PD-L1 protein expression in NSCLC [30]. Thus, a correlation between high FDG uptake and PD-L1 protein expression might reflect AKT-mTOR pathway activation. Additionally, FDG is actively entrapped in tumor-related activated immune We recently examined PD-L1 protein expression in 417 surgically resected ADC specimens [28] and found it was related to smoking in ADC patients. Moreover, PD-L1 protein expression in patients with SCC, which has a close etiological relationship with tobacco smoking, is higher than for ADC patients [33,34]. In the present study, smoking was a predictor of PD-L1 protein expression in patients with lung cancer, especially NSCLC, suggesting PD-L1 protein expression might be associated with smoking and oxidative stress. Furthermore, intracellular oxidative stress was upregulated in malignant carcinoma cells with high proliferative capacity [35]. This report affirmed PD-L1 protein expression might be associated with oxidative stress, and the frequency of PD-L1 protein expression was higher in patients with both smoking history and high SUVmax than in other patients. Further in vitro or in vivo experiments are necessary to confirm these findings.  The present study had several limitations. First, this was a single institutional retrospective study and not a trial-based correlative study; however, 579 patients were examined for associations between PD-L1 protein expression and SUVmax in preoperative 18 F-FDG PET/CT. The data obtained might help identify patients with PD-L1 expression who would benefit from antiPD-1/PD-L1 antibodies. The second limitation was that we investigated the association between PD-L1 expression on tumor cells only and SUVmax. PD-L1 protein is expressed by tumor cells and immune cells such as macrophages [36]. Further study of the association between PD-L1 protein expression on immune cells or other immune markers, such as TILs and TAMs, and SUVmax are required to assess the whole tumor microenvironment. However, the present study is the first to show a statistically significant association  between SUVmax and PD-L1 tumor protein expression in surgically resected lung cancer. The third limitation is that PD-L1 immunohistochemistry was conducted using only one antibody and one cut-off value. Recently, McLaughllin et al. reported that discordance was observed for PD-L1 protein expression within a sample when using different antibodies [37]. Furthermore, a positive rate using SP142 was lower than that of other antibodies including 28-8, 22C3, and SP263. However, our previous study showed that SP142 had a higher positive rate in 40 patients with SCLC compared with other antibodies including 28-8 and E1L3N [27]. Thus, definitive antibody and cut-off values are not established. Several antibodies and cut-off values should be evaluated in future studies. The fourth limitation is the lack of analysis of advanced cases such as stage IV because we examined the association between PD-L1 protein expression and SUVmax in [18] F-FDG PET/CT using surgical specimens. The fifth limitation is that we did not exclude the cases with small tumor size to avoid the bias of partial volume effect [38]. In the analysis only of the patients whose CT findings revealed a primary tumor size smaller than 20 mm, however, the results were much the same (data not shown). Further analysis only of the cases with tumor size larger than 20 mm is therefore required. Immunohistochemistry is the main method for examining PD-L1 protein expression, which requires biopsied or surgical specimens from patients, and invasive procedures such as transbronchial lung biopsy, CT-guided biopsy, or surgery. However, this study suggests PD-L1 protein expression can be predicted by medical examination and radioisotope examination, which are noninvasive or minimally invasive and can be used even if specimens from patients are unavailable. Thus, SUVmax could be used as a noninvasive tool to assess the tumor microenvironment including the frequency of PD-L1 protein expression to predict benefit from antiPD-1/PD-L1 inhibitors. Standardized methods to define the cut-off value for SUVmax should be established using studies with larger sample sizes.
In conclusion, PD-L1-expressing NSCLC was related to high glucose metabolism. The SUVmax in preoperative 18 F-FDG PET/CT was a predictor of PD-L1 protein expression in patients with NSCLC.