Heterogeneity of resistance mutations detectable by next-generation sequencing in TKI-treated lung adenocarcinoma

EGFR-mutated lung adenocarcinomas routinely develop resistance to tyrosine kinase inhibitors (TKI). To better characterize the relative frequencies of the resistance mechanisms, we analyzed 48 EGFR-mutated TKI-resistant specimens from 41 patients. Next-generation sequencing of post-treatment specimens detected EGFR p.T790M in 31 (79%) of 39 patients, PIK3CA mutations in 10 (26%), EGFR p.S768_V769delinsIL in one, and KRAS p.G12C in one. Five PIK3CA mutations were outside of codons 542, 545, and 1047. Three of four pre-treatment specimens did not carry the PIK3CA mutation found in the post-treatment sample. Small cell carcinoma transformation was identified in four patients; none had p.T790M, including two where p.T790M was identified in the co-existing adenocarcinoma. In p.T790M-mutated specimens, the allele frequency was less than 5% in 24% of cases. p.T790M allele frequency was usually lower than that of the sensitizing mutation indicating that the resistance mutation was present either in a subset of cells or, if the sensitizing mutation was amplified, in a subset of the sensitizing alleles of a dominant clone. Eight patients had multiple resistance mutations, suggesting either multiple separate resistant clones or a single clone harboring multiple resistance mechanisms. PIK3CA mutations appear to be a more significant resistance mechanism than previously recognized.


INTRODUCTION
The identification of targetable activating mutations in a subset of patients with adenocarcinoma of the lung has transformed therapy for patients with advanced disease. Approximately 10-15% of Caucasian patients and 30-40% of Asian patients with non-small cell lung cancers (NSCLCs) harbor activating mutations in the epidermal growth factor receptor (EGFR) gene [1][2][3]. Analysis of tumors for these mutations has become standard of care [1,2,4]. Treatment of patients with EGFR-mutated metastatic NSCLCs with gefitinib, erlotinib and afatinib, first-generation and second-generation EGFR tyrosine kinase inhibitors (TKIs), has improved response rates, time to progression, and overall survival [5]. Unfortunately, despite initial response to TKI therapy, acquired resistance develops after a median of approximately 10-13 months in almost all patients [1,2,4].
Next-generation sequencing (NGS) is a powerful tool both to identify low-level mutations in cancers and to increase the accurate assessment of small biopsy specimens, as is common after relapse. Because of its high sensitivity, NGS may detect the emergence of a resistant subclone within the tumor, even when it comprises a few percent of the tumor cells analyzed. The identification of these mutants will determine therapeutic options. In this retrospective cohort analysis using a validated clinical NGS assay, we survey our experience with detection of acquired resistance mutations to TKI therapy in a panel of 7 genes [20,21].

Positive control and negative control specimens
The peripheral blood negative control specimens showed no mutations in 115 runs; all mutations in the positive control specimens were detected over those runs. The observed mutant allele frequencies (MAFs) were highly consistent, demonstrating that NGS is quantitative and precise (Supplementary Table S1).

Level of background noise of EGFR p.T790M (c.2369C > T) mutation in FFPE specimens
In our previous clinical validation of this assay, the background noise for the EGFR c.2369C > T which results in p.T790M was calculated at 1.3% (mean plus 3 standard deviations (SD)), analyzing 16 FFPE nonneoplastic tissues [20]. For this study, a similar calculation of background noise for the EGFR c.2369C > T change was performed in 179 FFPE lung cancer specimens with an activating KRAS mutation. The C > T artifact (a deamination change) at position c.2369 was significantly higher than the C > A (P < 0.001) or C > G signal (P < 0.001) (Figure 1). The calculated background noise for c.2369C > T (mean plus 3 SD) decreased as read depth increased (0.77% for samples with 150-500 EGFR c.2369 reads, 0.42% for samples with 501-1,000 reads, and 0.37% for samples with more than 1,000 reads) ( Figure 1).

EGFR p.T790M mutation in pre-TKI specimens
Forty-one NSCLC patients who progressed after TKI treatment were included in this study. (Supplementary   Table S2). EGFR mutations before treatment were examined at the Johns Hopkins hospital in 21 patients, 8 by Sanger sequencing and 13 by NGS. EGFR mutations were retrospectively analyzed in patients 3 and 4 whose EGFR mutations were initially tested by Sanger sequencing. Co-existing p.E746_A750del (68%) and p.T790M (7.6%) mutations were detected in patient 3. Other specimens with a MAF in c.2369C > T of 0.25% or less were interpreted as negative for p.T790M mutation.

EGFR mutations in post-TKI specimens
Forty-eight post-TKI specimens were submitted from 41 patients whose NSCLCs progressed after TKI therapy. NGS failed in 5 of 48 specimens, including 2 from patients 16 and 20 who had only one specimen submitted for examination (Supplementary Table S2). TKI-sensitizing mutations were detected in all 39 patients, including 2 with an exon 18 mutation (p.G719C and p.E709_T710delinsD), 26 with an exon 19 deletion mutation, one with exon 20 p.A763_Y764insFQEA, and 10 with exon 21 p.L858R (Table 1). p.T790M was detected in 34 of 43 specimens or 31 (79%) of 39 patients (Tables 1 and 2), including specimen 2 with p.T790M at 1.1% MAF in 1059 reads. Retrospective analysis of 3 separate fragments from this pleural biopsy specimen showed relatively constant but low levels of mutation of p.T790M at 0.7% (1399 reads), 2.6% (1185 reads) and 2.6% (992 reads), respectively. Among the 6 additional EGFR mutations, 4 were present in both pre-and post-TKI specimens (patients 15, 17, 24 and 36), and one (p.S768_ V769delinsIL or p.S768I plus p.V769L) was present in only the post-TKI specimen ( Table 1). The type of EGFR mutation in the pre-TKI specimen was not known for patient 9, who had a p.K806I in the post-TKI specimen.
p.T790M and TKI-sensitizing mutations are in a cis arrangement By using amplicon-based NGS assays, the cis or trans relationship can be interpreted if co-existing variants are located within the same amplicon. p.T790M was located within the same allele with A763_Y764insFQEA in patient 4 and p.S768I in patient 17. The genotype of SNP rs1050171 was used to analyze 6 specimens in whom the variants were not on the same amplicon. By this method, a patient heterozygous for the inherited SNP will carry the initial, sensitizing mutation in linkage with only one of the SNP alleles (which are not on the same amplicon). If that chromosome is amplified, the SNP and sensitizing mutant alleles will be amplified together (> 50% allele frequency, similar in both SNP and mutant). A subsequent resistance mutation on the same amplicon with the SNP can then be imputed to be in either cis or trans with the sensitizing mutation. The MAF of the TKIsensitizing mutation was always consistent with that of the www.impactjournals.com/oncotarget SNP allele linked with p.T790M, indicating that p.T790M was in a cis configuration with the TKI-sensitizing mutant allele ( Figure 2 and Supplementary Table S3). In specimen 10 with two acquired EGFR resistance mutations, p.S768_ V769delinsIL was linked with the rs1050171 G allele, while p.T790M was linked with the A allele.

PIK3CA and KRAS mutations in post-TKI specimens
PIK3CA mutations were detected in 10 (26%) of 39 patients (Tables 1 and 2). The 3 most commonly reported codons (p.E542, p.E545 and p.H1047) for PIK3CA mutations accounted for 5 mutations. All except two with a PIK3CA p.E545K mutation (specimens 29 and 35) also carried an EGFR p.T790M. Four pre-TKI specimens were examined for the presence of the PIK3CA mutations known to be in their post-TKI specimens: p.G1049R was detected at both time points in patient 4 ( Figure 3), but pre-TKI specimens were negative in patients 3 (Supplementary Figure S1), 10 and 35. In specimen 28, 4 mutations were detected (Table 1), including KRAS p.G12C (Supplementary Figure S2). The pre-treatment specimen of this patient was reported to be positive for EGFR p.L858R, but negative for KRAS mutation using the therascreen KRAS test at another CLIA-certified laboratory. No mutations were detected in the AKT, BRAF, ERBB2 and NRAS genes in all post-TKI specimens examined.

Small cell carcinoma transformation
Transformation from adenocarcinoma to small cell carcinoma was observed in 4 (9.8%) of 41 patients (Supplementary Table S2 and Table 2). The TKIsensitizing mutation identified in the original pretherapy adenocarcinoma was detected in the small cell carcinomas of all 4 patients; p.T790M was not identified in any. However, two (patients 4 and 9) of these patients had a component of adenocarcinoma remaining after TKI therapy, and each of these harbored p.T790M in the adenocarcinoma. PIK3CA p.G1049R, seen in the pre-TKI specimen, was also detected in the small cell carcinoma component of patient 4. PIK3CA p.E545K was observed in specimen 29, but the PIK3CA mutation status in the pre-TKI specimen was not examined.

Mutant allele frequency of resistance mutations
p.T790M MAFs were 1-5% in 8 (24%), 1-10% in 16 (47%), and 1-20% in 27 (79%) of 34 specimens. In most post-TKI specimens, p.T790M MAFs were lower than the corresponding MAFs for sensitizing mutations ( Figure 4A), indicating p.T790M is most likely present in a resistant subclone. However, since EGFR mutations in lung cancers are commonly associated with gene amplification [22], p.T790M could be present in all tumor cells but only on one or a subset of the amplified EGFR TKI-sensitive mutant alleles. In specimen 7, for    Figure S3A). The detection of 31% p.T790M in the context of 51-70% estimated tumor cellularity indicates that p.T790M was present in a subset of the amplified p.L858R mutant alleles within a dominant resistant clone ( Figure 2). In support of this, the tumor was divided into five areas which were sequenced separately; the correlation of MAFs was analyzed by Spearman's rank correlation coefficient using GraphPad Prism software (GraphPad Software, ver5, La Jolla, CA) as described [23].  Figure S3B and S3C) PIK3CA MAFs were 1-10% in 3 (23%), and 1-20% in 8 (62%) of 13 PIK3CA-mutated specimens. As with p.T790M resistance mutations, PIK3CA MAFs in the post-TKI specimens were also equal to or lower than the TKIsensitizing EGFR MAFs ( Figure 4B), with the exception of two specimens (4A/4C). In this patient, a SNP array assay showed gain of the chromosomal region containing the PIK3CA gene in the pre-TKI specimen (Supplementary Figure S4). The PIK3CA p.A763_Y764insFQEA mutant allele ratio increased from 1.7 in the pre-TKI specimen to 2.1-3.7 in the post-TKI specimens ( Figure 3).

Co-existence of multiple of TKI resistance mutations
Two resistance mutations (EGFR p.T790M and PIK3CA mutations) were observed in 6 patients and 3 resistance mutations were observed in 2 patients (Tables 1  and 2). All 5 effusion specimens (3A, 3C, 4C, 9 and 10) from 4 patients showed 2 or 3 resistance mutations. Five specimens (13, 14, 15, 26 and 38) did not show mutations in EGFR p.790M, PIK3CA and KRAS genes nor transformation to small cell carcinoma; other proposed mechanisms of resistance (e.g., MET amplification) were not assessed.

DISCUSSION
In contrast to prior studies, we identified a high frequency of PIK3CA mutations in post-therapy lung adenocarcinoma patients. At 26% (10 out of 39 patients), this is much higher than the frequency of 0-5% reported The presence or absence of mutation in pre-treatment specimens was not known. Mutations were present in the pre-treatment specimens. f mut no.: The total number of resistance mutations in each patient's samples, including the EGFR 790M mutation, the EGFR p.S768_V769delinsIL mutation of specimen 10 (not present in the pre-treatment specimens), PIK3CA mutations, and the KRAS p.G12C mutation of specimen 28. The original TKI-sensitizing mutations and co-existing EGFR mutations in the pretreatment specimens of patients 15, 17, 24 and the EGFR p.K860I mutation of specimen 9 were not included. using a variety of non-NGS assay methods ( In this study, 9 of the 10 post-therapy cases with PIK3CA mutations also possessed another resistance mechanism. Eight harbored a concomitant EGFR p.T790M mutation and the other case showed small cell carcinoma transformation. The finding of coexisting mechanisms of resistance has been described in studies examining post-TKI specimens [6,26]. The significance of the PIK3CA mutation in these cases is uncertain. As EGFR acts through the PIK3CA/AKT pathway, mutations in that pathway might be anticipated to provide a resistance mechanism for the tumor cells. Interestingly, in the consortium study 2.7% of therapy-naïve specimens harbored multiple mutations [25]. Of these, PIK3CA mutation was a common target, found in association with another mutation in 48% of multiply mutated cases. A similar analysis of a broad variety of therapy-naïve cancers using a PCR-based mass spectrometry assay found that PIK3CA mutations are often associated with other driver mutations [27]. The significance of these mutations is uncertain. Several studies have suggested that its presence may affect outcome [28][29][30]. A group showed that EGFRmutated NSCLC cell lines with acquired resistance due to either EGFR p.T790M mutation or MET amplification had increased apoptosis when a PIK3CA inhibitor plus a MEK pathway inhibitor were used simultaneously. Single agent PIK3CA inhibitor did not show increased apoptosis [31]. These pre-clinical findings may provide a rationale for co-targeting mutant EGFR and PIK3CA pathways to overcome resistance.
As with other studies, our analysis found the most common post-TKI mutation to be p.T790M, occurring in 79% of patients compared to a reported range of approximately 40% to 60% [6,26]. Our higher percentage is most likely due to a higher analytic sensitivity of the NGS assay. Forty-seven percent and 79% of p.T790Mmutated specimens had MAFs less than 10% and 20%, respectively, which is below the limit of detection for Sanger sequencing. Twenty-four percent of p.T790Mmutated specimens had MAFs of 1-5% which is below the limit of detection for pyrosequencing, high resolution melting analysis or real-time PCR assays such as the therascreen EGFR RGQ PCR Kit and cobas EGFR Mutation Test [32][33][34]. In addition, these assays require separate runs for the detection of p.T790M and other EGFR mutations and therefore may not be suitable for comprehensive mutational profiling in core biopsy or fine needle aspiration specimens where tumor tissue is often limited. As shown in this study, many specimens were taken by core biopsy or fine needle aspiration.
The observation that p.T790M was present in only a small fraction of tumor cells of some patients implies the presence of additional resistant mechanisms in other subclones [35,36]. In this study, we showed that p.T790M MAFs lower than TKI-sensitive MAFs did not always indicate that p.T790M was present in a small subpopulation of resistant tumor cells. Specimen 7 from our study shows that an acquired p.T790M was present in a fraction of amplified TKI-sensitive alleles  of a dominant resistant clone and likely conferred TKIresistance. Our observation is consistent with an in vitro study demonstrating the dominant effect of a low fraction of p.T790M alleles acquired in TKI-resistant cell lines with a high level of amplified TKI-sensitive alleles [37]. Such "allelic dilution" may obscure detection of the biologically significant EGFR resistance mutation [37] if assays with a lower analytic sensitivity are used. Correctly identifying the proportion of a resistant tumor containing p.T790M may be clinically important as it may correlate with the response to third-generation TKIs. Factors that could affect this analysis include sampling bias in choosing which portion of the tumor to test and allele amplification which might lead to inaccurate and/ or imprecise estimation of tumor cellularity. Cell-free circulating tumor DNA has become an alternative source for non-invasive examination of p.T790M mutations [38,39], with potentially less sampling bias as compared with tissue biopsy. Serial blood draws and ultra-sensitive quantitative assays, such as digital droplet PCR, may provide a precise measurement of p.T790M MAF in the cell-free circulating tumor DNA to correlate with treatment outcomes.
p.S768I has been categorized as a primary resistant EGFR mutation [40], although there are conflicting opinions from other studies with small case numbers [41,42]. The largest cohort study showed TKIs were less effective in 7 patients with p.S768I mutation [43]. EGFR p.S768_ V769delinsIL (or compound p.S768I and p.V769L) has also been detected in a pre-treatment specimen of a patient who did not respond to gefitinib [44], but has not been reported in TKI-resistant specimens. In our specimen 10, linkage of the two acquired mutations p.S768_ V769delinsIL and p.T790M to the sensitizing exon 19 mutation could not be determined directly, but linkage of each with a different allele of SNP rs1050171 indicated that only one of these two mutants could be in cis with the TKI-sensitive mutant allele. A previous study showed a cis configuration between p.T790M and p.L8585R in two TKI-resistant cell-lines [37]. In vitro transfection also demonstrated greater TKI-resistance when the p.T790M allele was present in cis to the TKI-sensitizing mutation [37]. Further studies are warranted to elucidate if the p.S768_V769delinsIL confers TKI resistance and if an acquired resistance mutation is effective in the trans configuration with the TKI-sensitive mutant allele.
Mutations in KRAS and EGFR genes are mutually exclusive in therapy-naive NSCLCs, and KRAS mutations also appear to represent a negative predictor for TKI therapy in NSCLCs [45]. Surprisingly, acquired KRAS mutations were not detected in 4 previous larger cohort studies of re-biopsy tissues (Table 3) [6][7][8]24], although co-existing EGFR p.L858R and p.T790M and KRAS p.G12V mutations were reported in a Chinese patient who progressed after 3 months of TKI therapy [46]. In our patient 28, we identified a post-therapy KRAS mutation that was not detected by a CLIA-certified laboratory using a relatively sensitive therascreen KRAS test; both preand post-therapy specimens showed an EGFR p.L858R mutation. To our knowledge, this is the first convincing example of a KRAS mutation acquired following TKI therapy, presumably as a resistance mechanism. The role of KRAS mutations in acquired TKI resistance was also supported by a recent study demonstrating a high incidence of KRAS mutations in the cell-free circulating tumor DNA from TKI-resistant NSCLC patients [47]. Further studies are needed to evaluate the analytic specificity (or background noise) of an assay with relatively high analytic sensitivity and to confirm the clinical specificity of an assay in circulating tumor DNA detecting a mutation commonly seen in other neoplasms.
In summary, in the setting of post-therapy NSCLCs NGS demonstrates an excellent limit of detection, a broad reportable range of mutations, the capacity for

Materials
Between April 2013 and June 2015, 830 formalin-fixed paraffin-embedded specimens with lung adenocarcinoma were submitted to the Molecular Diagnostics Laboratory at the Johns Hopkins Hospital. Forty-eight specimens were submitted from 41 patients whose NSCLCs progressed after gefitinib (one patient) or erlotinib (40 patients) therapy (Supplementary Table S2). Three specimens were submitted from patients 3, 4 and 19 and two specimens were submitted from patient 24. Duration of TKI therapy ranged from 3 months to 10 years with a median of 12 months. EGFR mutations before treatment were examined at the John Hopkins Hospital in 21 patients, 8 by Sanger sequencing of exons 18-21 and 13 by NGS [20]. In the remaining 20 patients the EGFR mutations before therapy were documented at other institutions. In 3 of these 20 (patients 9, 16, and 18) the details of EGFR mutations were not known. Thirty two were biopsy specimens, 7 fine needle aspiration specimens, 6 pleural effusion or ascites specimens, 2 resection specimens and one was a bronchioloalveolar lavage specimen (Supplementary  Table S2). The Johns Hopkins Medicine institutional review board granted approval to this study One hematoxylin & eosin (H&E) slide followed by 5-10 unstained slides, each with 5 or 10-micron thick sections, and one additional H&E slide were prepared with PCR precautions. The H&E slide was marked for tumor enrichment by a pathologist. Macro-dissection of neoplastic tissue from 3-10 unstained slides was performed. DNA was isolated using the Pinpoint DNA Isolation System (Zymo Research, Irvine, CA), followed by further purification via the QIAamp Mini Kit (Qiagen, Valencia, CA) [48].

Next-generation sequencing (NGS)
NGS was conducted using the AmpliSeq Cancer Hotspot Panel (v2) for targeted multi-gene amplification, as described previously [20,49] Chip and Personal Genome Machine (PGM) as the sequencing platform (Life Technologies, Carlsbad, California). The DNA input was up to 30 ng, as measured by Qubit 20 Fluorometer (Life Technologies). Up to 8 specimens were barcoded using Ion Xpress Barcode Adapters (Life Technologies) for each Ion 318 chip. One to three controls (a non-template control, a normal peripheral blood control from a male, and/or an artificial positive control specimen) were included on each chip. Positive controls were mixed DNA specimens from several cell lines with known mutations at low mutant allele frequency (lung cancer panel control): AKT p.E17K and p.E49K, BRAF p.V600E, EGFR p.T790M and p.L858R, ERBB2 p.G776V or p.G776_V777insC, KRAS p.G12C and p.P121H, NRAS p.Q61R, and PIK3CA p.K111E and p.H1047R mutations (Supplementary Table S1).
Mutations were identified and annotated through both Torrent Variant Caller (Life Technologies) and direct visual inspection of the binary sequence alignment/ map (BAM) file using the Broad Institute's Integrative Genomics Viewer (IGV). All specimens were analyzed for AKT, BRAF, EGFR, ERBB2, KRAS, NRAS and PIK3CA genes. During our validation of this NGS assay, a cutoff of background noise at 2% was chosen for single nucleotide variations because of our study of 16 non-neoplastic formalin-fixed paraffin-embedded tissues [20]. With sufficient DNA input, the limit of detection is dictated by the depth of coverage (or number of sequencing reads). Approximately 150 and 500 reads are needed to detect a heterozygous mutation with 99% confidence in a specimen with 20% and 10% tumor cellularity, respectively. The reportable ranges for the AKT, ERBB2 and EGFR genes are summarized in Supplementary Table S4. Within these reportable ranges, rs1050171 (c.2361G > A, p.Q787 = of the EGFR gene) was the only single nucleotide polymorphism (SNP) with a minor allele frequency of more than 1% in the general population according to the dbSNP database of the National Center for Biotechnology Information. The reportable ranges and reference ranges for the BRAF, KRAS, NRAS and PIK3CA genes have been reported previously [21]. The nucleotide changes of the mutations detected in this study were summarized in Supplementary Table S5.

Single nucleotide polymorphism (SNP) array
SNP array analysis was performed as previously described [50]. Briefly, DNA samples extracted from FFPE tissues (optimally 200 ng) were treated with the Infinium HD FFPE NDA restore kit before running on the Illumina Infinium II SNP array (HumanCytoSNP-12 v2.1 DNA Analysis BeadChip, Illumina Inc., San Diego, CA) according to the manufacturer's standard protocol. The B allele frequency and Log R ratio data were analyzed using Illumina KaryoStudio software version 2.0 and CNV (copy number variation) partition V2.4.4.0.