Between February 2009 and May 2014, 1414 patients with primary lung cancers, including 1062 with adenocarcinomas, underwent pulmonary resection at our institution. Among these, we retrospectively reviewed 734 adenocarcinoma patients who underwent ¹⁸F-FDG PET-CT scanning within 2 months before surgery and whose surgically resected specimens were examined for EGFR and KRAS mutations. Patients who underwent induction chemotherapy and/or radiotherapy were excluded from this study. Patients were classified into three groups according to the mutation status of the tumors, namely EGFR mutation-positive (EGFR m⁺), KRAS mutation-positive (KRAS m⁺), and wild-type (WT) for both genes. Clinical characteristics such as age, gender, smoking status, preoperative serum carcinoembryonic antigen (CEA) level and SUV_max and pathological findings such as tumor size, nodal status, lymphatic permeation and vascular invasion of EGFR m⁺ and KRAS m⁺ tumors were compared to those of WT tumors.

This study was performed using surgical specimens in the tissue bank at our department, which was established with the approval of the institutional review board (IRB) of Juntendo University School of Medicine. Written consent was obtained from all patients prior to surgery for the procurement of tissue for the research purposes. The IRB approved the use of specimens stored in the tissue bank without obtaining new informed consent and deemed that the contents of this study were ethically acceptable.

As detailed previously [14], PET-CT scan was carried out with a Discovery ST PET/CT scanner (GE Medical Systems; Waukesha, WI, USA) at the Yotsuya Medical Cube (Tokyo Japan). Two experienced nuclear medicine radiologists (W. K. and M. A.) evaluated the PET-CT images, side by side, and reached a consensus on the findings.

Genomic DNA was extracted from frozen lung cancer tissues sampled from surgically resected specimens. EGFR mutations were analyzed using the peptide nucleic acid-locked nucleic acid polymerase chain reaction (PCR) clamp method [15], and KRAS mutations using the peptide nucleic acid-mediated PCR clamping method [16].

The Steel-Dwass test was used to compare SUV_max among multiple groups based on EGFR and KRAS mutation patterns. Receiver operating characteristic (ROC) curves were generated to obtain a cut-off for SUV_max of the primary tumor which maximizes the sum of sensitivity and specificity for predicting EGFR or KRAS mutation status. Correlations between EGFR or KRAS mutation status and clinicopathological factors were evaluated. Univariate analyses between SUV_max and each clinicopathological factor were performed by a logistic regression model. All of the variables identified to be significant in the univariate analyses were subsequently entered into the multivariate analyses using a bidirectional (i.e., forward and backward) step-wise logistic regression model. A P-value of < 0.05 was considered statistically significant. All statistical analyses were performed using the R statistical software package (version 3.0.2, http://www.r-project.org/).

CAGE data generated using the previously described protocol [17] were obtained from a previous study [13]. In brief, double-stranded RNA/cDNA produced by reverse transcription from total RNA extracts was purified, oxidized with sodium periodate, and biotinylated with biotin hydrazide. The single-stranded cDNA was recovered after digestion of the single-stranded RNA with RNase I, and ligated with 3’-end and 5’-end adaptors specific to the samples. Double-stranded cDNAs were synthesized and mixed for sequencing in one lane of an Illumina HiSeq2500 sequencer (Illumina; San Diego, CA, USA). The CAGE reads were aligned to the reference genome (hg19) with high mapping quality of ≥ 20.

The aligned CAGE reads were counted in each region of the FANTOM5 robust peaks [11], a reference set of TSS regions, as raw signals for the promoter activities. Expression (activity) levels of individual promoters were quantified as counts per million (CPM) after normalization by the relative log expression method [18], and subjected to differential analysis using edgeR (version 3.2.4) [19] in R/Bioconductor [20]. Associations between expression levels and SUV_max and their statistical significance were assessed by Spearman’s rank correlation. Only results with a false discovery rate (FDR) less than 1% were considered statistically significant, in both the differential and correlation analyses.

Patient characteristics are summarized in Table 1. Of 734 patients, 367 (50%) were male and 367 (50%) were female. Median age at the time of the operation was 68 years (range, 27–89 years). A total of 363 of 734 (49%) patients were smokers (pack-years > 5) and 371 (51%) were non-smokers (pack-years ≤ 5).

Of the 734 lung adenocarcinomas, EGFR and KRAS mutations were detected in 334 (46%) and 83 (11%), respectively. The EGFR mutation spectra were distributed as follows. The point mutation L858R in exon 21 and deletions in exon 19 were detected in 194 and 120 tumors, respectively, which together accounted for 94% of all EGFR alterations. The remaining 6% of the minor EGFR mutations were exon 18 G719A in 8 tumors, exon 18 G719S in 5, exon 18 G719C in 2 and exon 21 L861Q in 3. Double mutations were found in 2 tumors; 1 harbored exon 21 L861Q and exon 20 T790M and the other had exon 18 G719A and exon 20 T790M, simultaneously. With regard to KRAS, a point mutation in codon 12 was found in 81 (98%) tumors, and a point mutation in codon 13 in 2 (2%). G to T, or G to C transversions were found in 60 (72%) tumors, and G to A transition in 23 (28%). EGFR and KRAS mutations were mutually exclusive.

The median SUV_max of all primary tumors was 2.7 (range, 0–33.2). Median SUV_max in the EGFR m⁺ group, KRAS m⁺ group, and WT group were 2.1 (range, 0–23), 3.0 (range, 0–23.5), and 3.9 (range, 0–33.2), respectively. SUV_max of EGFR m⁺ tumors was significantly lower than that of WT and KRAS m⁺ tumors (Fig 1A). SUV_max of tumors with exon 21 L858R or exon 19 deletions was significantly lower than that of WT tumors. However, no significant differences were noted in SUV_max between tumors with minor mutations and WT tumors (Fig 1B). The SUV_max of KRAS m⁺ tumors did not significantly differ from that of WT tumors (Fig 1A). No significant differences were found in SUV_max between tumors with any KRAS mutation spectrum (G to T/G to C transversions or G to A transition) and WT tumors (Fig 1C).

Next, we evaluated the prediction of EGFR or KRAS mutation using SUV_max. A cut-off value of SUV_max ≤ 2.69 provided the highest area under the curve (AUC; 0.610) for predicting EGFR mutation, while SUV_max ≤ 3.40 provided the highest AUC (0.536) for KRAS mutation (Fig 2). Using these cut-off values, parameters for the prediction of EGFR mutations were sensitivity, 60%; specificity, 61%; accuracy, 60%; positive predictive value (PPV), 62%; and negative predictive value (NPV), 59%; and parameters for the prediction of KRAS mutations were sensitivity, 54%; specificity, 54%; accuracy, 54%; PPV, 23%; and NPV, 82%.

On univariate analysis, EGFR mutations were more frequent in females, non-smokers, patients with normal CEA levels, tumors without lymph node involvement or blood vessel invasion, and tumors with lower SUV_max. On multivariate analysis, significant predictors of EGFR mutation were smoking status and SUV_max (Table 2). The probability of EGFR mutation was inversely correlated with SUV_max. Univariate analyses showed that KRAS mutations were more frequent in males and smokers. On multivariate analysis, the only significant predictor of KRAS mutation was smoking history (Table 3). No relationship was found between the KRAS mutation status and SUV_max. The predictability of EGFR mutation status was compared between combinations of well-established clinical factors with or without SUV_max (Table 4). PPV of EGFR mutation status was increased by adding SUV_max to gender and smoking status.

Further, we examined expression levels of genes based on the CAGE results (Takamochi et al., submitted), in particular those related to glucose metabolism and the cell cycle, in association with SUV_max. We manually selected 7 genes associated with glucose metabolism: class I glucose transporters (GLUT1, GLUT2, GLUT3, GLUT4), hexokinase-II (HK-II), hypoxia-inducible factor-1 alpha (HIF-1α), and carbonic anhydrase IX (CAIX). Of these, 4 genes (GLUT1, HK-II, HIF-1α, and CAIX) showed positive correlations between their expression levels monitored by CAGE with SUV_max across 62 lung adenocarcinomas (Fig 3). Next, we selected 5 genes associated with cell growth: TP53, CCND1, BCL2, vascular endothelial growth factor (VEGF), and MKI67. Of these, expression of VEGF showed a positive correlation with SUV_max, while BCL2 showed an inverse correlation with SUV_max (Fig 3).