Different mutational characteristics of the subsets of EGFR-tyrosine kinase inhibitor sensitizing mutation-positive lung adenocarcinoma

A total of 74 lung adenocarcinoma tissues which met the following criteria were randomly selected from the tissue archives of two affiliated hospitals, Severance Hospital and Gangnam Severance Hospital, of Yonsei Medical Center (study cohort). Each specimen (1) is a pathologically confirmed lung adenocarcinoma, (2) had history of curative surgical resection, (3) has a submission of informed consent for sequencing of major cancer related genes, (4) was confirmed to have mEGFR either by Sanger sequencing or PNA clamping methods, and (5) has submission of informed consent for tissue collection. This study was approved by the IRB of Gangnam Severance Hospital (IRB #3–2017-0059) and was carried out in compliance with the Declaration of Helsinki (https://​www.​wma.​net/​policies-post/​wma-declaration-of-helsinki-ethical-principles-for-medical-research-involving-human-subjects/​) and Korean GCP guidelines. In order to further evaluate the mutational characteristics of lung adenocarcinoma with mEGFR, publicly available data were extracted from the GDC Data Portal of The Cancer Genome Atlas (TCGA https://​portal.​gdc.​cancer.​gov/​). Among the 585 TCGA-LUAD cases, 81 cases harbor mutations in the EGFR gene (ENSG00000146648) with 47 types listed in TCGA-LUAD (Additional file 1: Table S1). A total of 7577 mutations were found in 48 TCGA-LUAD cases; TCGA-55-8506 has 2305 mutations alone, accounting for 30.4% of all mutations, and was excluded from further analysis. For this analysis, we selected 47 cases with 7 EGFR subtypes known as common mEGFR (LUAD cohort) (Additional file 2: Table S2 and Additional file 3: Table S3) (8).

Next generation sequencing: A 0.62 Mb customized NGS panel containing 70 major cancer genes was constructed and sequenced using the Ion S5 NGS system (Thermo Fisher Scientific) (Additional file 4: Table S4). For the extraction of cancer-enriched gDNA, paraffin-embedded tissue samples were loaded onto silanated slides in 4-μm-thickness sections. Each slide was lightly stained with H&E and examined for the presence of cancer cells. The cancer-enriched area as determined by an independent lung pathologist was marked and scraped with clean blades. gDNA was extracted using a QIAamp DNA FFPE Tissue Kit (Qiagen, Valencia, CA, USA). Fifty ng of extracted DNA was reacted with fragmentizer for 12 to 50 min (Archer) and the section of cut DNA was blunted and 5′ phosphorylated with an end-repair enzyme (Archer). The end of the DNA was barcode-ligated through a reaction with an MBC adapter (Archer) for 15 min, and 1st and 2nd PCR were performed using a primer set for the selected target genes. After measuring the prepared library with Qubit^Ⓡ, 50 pmol of sample was obtained and mixed with mineral oil to normalize it with beads. The sample was applied to a 540 Chip (Life Technology) and sequenced with the S5 sequencer (Life technology). The obtained results were analyzed by Archer Analysis 5.1 (Archer) and the median depth of the sequencing was 308X [164.5 ~ 738.5].

Categorical and continuous variables were compared using χ²-tests and t-test, respectively. Differences in the distribution of continuous variables between the two independent samples and dependent samples were assessed by Mann–Whitney U test and Wilcoxon signed-rank test, respectively. The Kruskal-Wallis test was used to compare the medians of three or more groups, and the Bonferroni method was used for the post-hoc test. The mutational signature was analyzed using R software and the Maftools package [18]. Briefly, the pattern of the mutation is analyzed under the 3-nucleotide context, which includes one nucleotide immediately 5′ and 3′ of the mutated base [10, 17] to characterize single base substitution using a 96-mutation classification by combining 6 substitution types and the 5′ and 3′ adjacent bases to the mutated base. The mutation patterns found in cancers are analyzed by non-negative matrix factorization (NMF), and compared to 30 types of mutational signatures [8]. All analyses were performed using SPSS version 23 (SPSS Inc., Chicago, IL, USA). All statistical tests were two-sided, and a P-value < 0.05 was considered to be statistically significant.

The incidence of lung adenocarcinoma harboring EGFR-TKI-sensitizing mutations is higher in East Asia than in Western countries, and the proportion of non-smokers and women is high, suggesting the presence of mutagenic stressors other than cigarette smoking. We conducted targeted sequencing of 70 major cancer-related genes on a total 73 specimens recruited from 71 patients, including two cases with double primary cancer (Table 1 and Additional file 5: Table S5). These included 37 cases with EGFR exon19 deletion mutation, 35 cases with L858R mutation and one case with the L861Q mutation. The mean age of the patients was 60.34 ± 10.01 years, and they were comprised of 25 males and 46 females. Of these, 20 cases had a history of smoking, with a mean of 22.75 ± 13.33 pack-years.

The sequencing of 73 specimens revealed 530 mutations consisting of 490 single and double nucleotide substitutions, 1 insertion, and 39 deletions. The median of total mutations obtained from each sample was 6 [5 ~ 8] (Fig. 1A and B). These mutations were further classified according to the variant effect predictor (VEP) and the results are shown in Fig. 1C and Table 2. C > A transversion, which is attributed to cigarette smoking, is the most common type of SNV in lung cancer [5]; therefore, we questioned whether this finding also applies to Korean lung adenocarcinomas with mEGFR. Excluding the mutations which were assumed to be benign because they did not change protein behaviors by the VEP, and indels and doublets, a total of 363 SNVs were classified and analyzed (Fig. 1D and E). In this subset, C > T transition accounted for the largest portion (27.3%), followed by C > G transition (20.4%), whereas C > A transversion occupied the second smallest fraction (12.9%). The frequency of C > A transversion was statistically significantly lower than those of other mutations, except for T > A and T > G transversions, which differs from the general notion that NSCLC harbors higher C > A transversions due to cigarette smoking. The differences in the frequencies of Ti and Tv did not reach statistically significant level in each individual case (Fig. 1F).

To identify the presence of smoking effects, we examined whether there was a difference in the mutational spectrum between cases with vs. without a smoking history. There was no statistically significant difference in the total number of mutations between smoker and non-smoker groups and between those with more than a 20-pack-year history and the other cases. There was no significant difference in the frequencies of Ti, Tv, and Ti/Tv ratio according to the smoking history. These findings were similar to the comparison of mutation patterns according to gender, which showed no differences in the total mutational burden, frequencies of Ti and Tv, and Ti/Tv ratios between genders. Then, we questioned whether the subtypes of mEGFR are related to the difference in mutation burden, Ti, Tv, and Ti/Tv ratio. In order to accomplish this, the cases were classified into either the mEGFR E19del-subgroup or the mEGFR SNV-subgroup, which includes cases with L858R, L861Q, and other EGFR-TKI sensitizing SNVs. In the E19del-subgroup, there was no statistically significant difference in the frequency of Ti and Tv in each individual, whereas the frequency of Tv was significantly higher than Ti in the mEGFR SNV-subgroup (P = 0.0020, Wilcoxon signed rank test). There was a significant difference in the total number of mutations, as the E19del-subgroup had a lower mutational burden than the SNV-subgroup (P = 0.0326, Mann-Whitney U test). The E19del-subgroup had lower Tv and a higher Ti and Ti/Tv ratio than the SNV-subgroup (P = 0.0441, =0.0020, and = 0.0002, respectively, Mann-Whitney U test, Fig. 2A~D). As C > A transversion was less frequent than in general lung cancer, this suggests that some mutagenic stress other than cigarette smoking likely exists in this subset of lung adenocarcinomas. To identify the mutagenic stressors that influence mEGFR-positive lung adenocarcinomas, we analyzed the mutational signature proposed by Alexandrov et al. (4). When the three best-matched mutational signatures were estimated using non-negative matrix factoralization (NMF), signature 5 and signature 3 was derived (Fig. 2E and F). Among smoking history, gender, and EGFR-mutation subtype, only EGFR subtype showed a difference in the distributions of Ti and Tv. We further analyzed the mutational signatures between the E19del-subgroup and the SNV-subgroup using NMF. In the subgroup analysis on the mutational, signature 5 is the predominant mutational signature in this study cohort (Data not shown).

The mean age of the 47 selected cases was 66.20 ± 9.44 years, and the cases were comprised of 11 males, 34 females, and two cases in which the gender was not reported (Table 1). Of these, 19 had a smoking history. A total of 5272 mutations were recovered from 47 EGFR-TKI-sensitive mutation-positive TCGA-LUAD cases. The median number of total mutations obtained from each sample was 78 [56 ~ 112.5], which indicates that the mutational burden was significantly lower than that of other cancer types listed in the TCGA dataset and the TCGA-LUAD parent group (Fig. 3A and B). These 5272 mutations are classified into 4954 single nucleotide substitutions, 146 insertions, and 172 deletions (Fig. 3C). These mutations were further classified according to the VEP and the results are shown in the Table 3 and Fig. 3D.

Because C > A transversion, which is attributed to cigarette smoking, is the most common variation in NSCLC [5], we questioned whether this finding also applies to this subset of NSCLC. 1179 mutations, including 1071 synonymous mutations, 47 mutations in the splice region, and 1 stop retaining mutation, were assumed to be harmless or unlikely to change protein behaviors by the VEP; these SNVs along with insertions and deletions were excluded, and 3812 SNVs were classified and analyzed (Fig. 3E and F). In this subset, C > T transition accounted for the largest portion (32.7%), followed by C > A transversion (30.3%). There was no significant difference in the frequency of C > T vs. C > A mutations in each individual, which contradicts the general notion that NSCLCs harbor higher C > A transversions than any other SNV. The difference in the frequencies of Ti and Tv in each individual case was not statistically significant (Fig. 3G).

The differences in the mutational pattern observed in the study cohort were further verified in the LUAD cohort. The mutational burden did not differ between the groups with and without a smoking history. In addition, there were no differences in the Ti, Tv, and Ti/Tv ratio according to smoking history or between smokers and non-smokers. There was also no significant difference in the total mutational burden or the frequency of Ti, Tv, and Ti/Tv ratio between genders. Finally, we questioned whether the subtypes of EGFR-TKI-sensitizing mutations were related to the difference in mutational burden, Ti, Tv, and Ti/Tv ratio. As seen in the study cohort, there was no statistically significant difference in the frequency of Ti and Tv in each individual in the E19del-subgroup, whereas the frequency of Tv was significantly higher than Ti in the mEGFR SNV-subgroup (P = 0.0220, Wilcoxon signed rank test). In the E19del-subgroup, the total mutation burden and the number of Tv mutations were significantly lower than in the SNV-subgroup (p = 0.0497 and = 0.0220, respectively, Mann-Whitney U test) and the Ti/Tv ratio was significantly higher as well (P = 0.0055, Mann-Whitney U test) (Fig. 4A and B).

Next, in order to estimate the mutagenic stress that influenced EGFR mutations, we analyzed the LUAD cohort using the mutational signature proposed by Alexandrov et al. (4). When inferring the mutagenic stress using NMF, signature 4 (cosine similarity = 0.851), signature 30 (cosine similarity = 0.775), and signature 5 (cosine similarity = 0.713) were derived (Fig. 4C and D). Because the subset analysis of LUAD cohort according to smoking history, gender, and EGFR mutation subtype showed that only the subtype influenced the distribution of mutational burden, Ti, and Tv, we compared the mutational signatures between the E19del-subgroup and the SNV-subgroup. In the E19del-subgroup, signature 30 (cosine similarity = 0.741, unknown etiology) and signature 5 (cosine similarity = 0.725 and = 0.660, unknown etiology) were derived, whereas signature 4 (cosine similarity = 0.842, smoking effect), signature 30 (cosine similarity = 0.789, unknown etiology) and signature 5 (cosine similarity = 0.738 unknown etiology) were obtained from the SNV-subgroup (Fig. 4E and F). Taken together, the mutational signatures differ between the E19del-subgroup and the SNV-subgroup, as smoking related signature 4 was observed only in SNV-subgroup, and signature 5 and 30 are the prevalent underlying mutational signature in lung adenocarcinoma with mEGFR-TKI.