KRAS mutation is predictive of outcome in patients with pulmonary sarcomatoid carcinoma

Abstract

Aims:

Pulmonary sarcomatoid carcinoma (PSC) is a poorly differentiated non-small-cell lung carcinoma (NSCLC) with aggressive behaviour. This study aimed to evaluate the prognostic clinicopathological and genetic characteristics of PSCs.

Methods and results:

Fifty-three cases of surgically treated PSCs were selected, 23 of which were subjected to mutation and copy number variation analysis using the 50-gene Ion AmpliSeq Cancer Panel. The majority of the patients were male (32 of 53, 60.3%) and smokers (51 of 53, 96.2%). Overall, 25 (47.1%) patients died within 2–105 months (mean = 22.7 months, median = 15 months) after diagnosis, and 28 were alive 3–141 months (mean = 38.7 months, median = 21.5 months) after diagnosis. Five-year overall survival was 12.5%. KRAS codon 12/13 mutation in adenocarcinomas (P = 0.01), age more than 70 years (P = 0.008) and tumour size ≥4.0 cm (P = 0.02) were associated strongly with worse outcome. TP53 (17 of 23, 74.0%) and KRAS codon 12 of 13 mutations (10 of 23, 43.4%) were the most common genetic alterations. Potentially actionable variants were identified including ATM (four of 23, 17.3%), MET, FBXW7 and EGFR (two of 23, 8.7%), AKT1, KIT, PDGFRA, HRAS, JAK3 and SMAD4 (one of 23, 4.3%). MET exon 14 skipping and missense mutations were identified in two (11.1%) cases with adenocarcinoma histology. Copy number analysis showed loss of RB1 (three of 23, 13%) and ATM (two of 23, 8.7%). Copy number gains were seen in EGFR (two of 23, 13.0%) and in one (4.3%) of each PIK3CA, KRAS, MET and STK11.

Conclusions:

Potentially targetable mutations can be identified in a subset of PSC, although most tumours harbour currently untargetable prognostically adverse TP53 and KRAS mutations.

Introduction

Pulmonary sarcomatoid carcinoma (PSC) is a highly aggressive type of non-small-cell lung carcinoma (NSCLC), composed of both epithelial and sarcoma-like components. There are five main histological subtypes in this category: pleomorphic carcinoma, spindle cell carcinoma, giant cell carcinoma, carcinosarcoma and pulmonary blastoma.¹ PSCs are rare and account for fewer than 1% of all pulmonary malignancies; however, compared to other stage-matched NSCLC, they are more resistant to conventional therapies and have poorer prognosis.^2,3

Although the molecular characteristics of the more common subtypes of NSCLCs, mainly adenocarcinomas, have been studied extensively, the genetic alterations in PSC have only recently become the target of studies.^3–7 This is due perhaps to the rarity of the disease and difficulty in diagnosing PSCs, particularly in small biopsies. The 2015 WHO¹ recommends molecular testing in PSCs according to known genetic abnormalities associated with the histological components in the tumour. KRAS mutation has been reported in up to 38% of PSCs and EGFR mutations in up to 25%.^6,8–10 Furthermore, few recent studies have identified targetable MET exon 14 skipping in a significant fraction of cases,^3,7,9 with a few case reports demonstrating a great response to targeted therapy with MET inhibitors.^11,12 Despite these advancements, there are still limited options available for treatment of these tumours. In addition, there are no clinicopathological or molecular features that could predict outcome reliably in PSC patients. In this work, using a targeted next-generation sequencing approach, we explored the genetic profile and clinicopathological characteristics of a cohort of surgically treated PSC.

Materials and methods

PATIENTS AND SPECIMENS

Of 53 consecutive, surgically treated PSCs during a 10-year period (from 2004 to 2014), 23 were selected based on tissue availability for additional studies. All cases were reviewed to confirm the diagnosis applying the 2015 WHO criteria¹ and were staged according to the American Joint Committee staging manual (8th edn).¹³ The study was conducted under an exemption approved by the University of Pittsburgh Institutional Review Board (PRO 12070229).

Clinical information including gender, age, tumour stage and smoking status was obtained from patients’ electronic medical records. Follow-up data regarding survival were collected through the institutional Network Cancer Registry.

IMMUNOHISTOCHEMISTRY

Immunohistochemistry (IHC) for Rb1 (Leica, Allendale, NJ, USA; clone 13A10, monoclonal mouse, 1:50) was performed.

FLUORESCENCE IN - SITU HYBRIDISATION ASSAYS

Fluorescence in-situ hybridisation (FISH) assays for amplification of KRAS, EGFR, PIK3CA and MET were performed as described previously.^14,15

NEXT – GENERATION SEQUENCING

DNA sequencing was performed using the Ion AmpliSeq Cancer Panel (Ion Torrent; Life Technologies, Thermo Fisher Scientific, Waltham, MA, USA), as reported previously.¹⁶ Briefly, 10 ng of DNA was amplified by polymerase chain reaction (PCR) using the AmpliSeq Cancer Panel Primers pool and Ion AmpliSeq Master Mix version 2.0. Multiplexed barcoded libraries were enriched by clonal amplification using emulsion PCR on ion sphere particles (ISPs) (Ion PGM template OT2 200 kit or Ion PI OT2 200 kit version 3) and loaded onto an Ion 318 chip or P1 chip (Thermo Fisher Scientific). Massively parallel sequencing was carried out on a personal genome machine sequencer or ion proton (Thermo Fisher Scientific).

The raw signal data were analysed using Torrent Suite (version 4.0.1; Life Technologies, Carlsbad, CA, USA). The short sequence reads were aligned to the human genome reference sequence (GRCh37 patch 13, GCF_000001405.25). Variant calling was performed using Variant Caller version 4.4.3.3 plugin (integrated with Torrent Suite) that generated a list of identified sequence variations [single nucleotide variants (SNV) and insertions or deletions (indels)] in a variant calling file (VCF version 4.2; https://samtools.github.io/hts-specs/VCFv4.2.pdf). After removing reference calls from the VCF files, variant calls in each VCF files were normalised¹⁷ and sorted based on the chromosome and genomic position. Variant calls were annotated using _ANNOVAR¹⁸ and the _HGVS python module.¹⁹ Several publically available databases were used for variant annotation: _COSMIC version 81 (http://cancer.sanger.ac.uk/cancergenome/projects/cosmic/; last accessed 8/30/2017), dbSNP build 137 (http://www.ncbi.nlm.nih.gov/SNP/; last accessed 8/30/2017), 1000 genomes (http://www.1000genomes.org/; last accessed 8/30/2017), Exome Variant Server (http://evs.gs.washington.edu/EVS/; last accessed 8/30/2017), Exome Aggregation Consortium (ExAC) (http://exac.broadinstitute.org/; last accessed 8/30/2017) and in-silico prediction scores (PolyPhen-2 and SIFT).^20,21 Sequence variants with at least 5% allelic fraction and at least ×200 depth of coverage were included for analysis. Integrated Genomics Viewer²² (IGV; Broad Institute, Cambridge, MA, USA) was used for manual review of the sequence read pile-ups to assess variant call quality. A joint cohort analysis of all variants across all samples were performed to identify recurrent low-frequency false positive variants. Variants were prioritised using the Association for Molecular Pathology, American Society of Clinical Oncology and College of American Pathologists joint consensus guidelines on variant interpretation in cancer.²³ Copy number analysis from next-generation sequencing data was performed using the copy number variation (CNV) kit.²⁴ A pooled normal reference was generated from targeted sequence analysis of 10 normal peripheral blood samples. Copy number variation (gains or losses) that was supported by deviation of all gene-specific amplicons from the baseline was prioritised and evaluated further. Sequence variants and CNVs were confirmed using DNA Sanger sequencing, FISH and IHC. Visualisation plots were created using JavaScript library jsComut (https://github.com/pearcetm/jscomut; last accessed 8/30/2017).

STATISTICAL ANALYSIS

Categorical data were presented as frequency and percentage, whereas continuous variables were described with mean. Overall survival (OS) was defined as the time from date of commencement of treatment (either surgical resection or beginning of radiation or chemotherapy) to the date of the last follow-up or death. Survival differences between groups for an individual risk factor were examined by the log-rank test. Statistical tests were performed using GraphPad Prism software version 7.03 (GraphPad Software, Inc., La Jolla, CA, USA). All tests were two-sided, and differences were considered significant at P-values ≤0.05.

Results

PATIENT CHARACTERISTICS

Clinicopathological characteristics of the 53 PSC cases are summarised in Table 1. Cases include surgically resected 52 pleomorphic carcinomas (98.2%) and one carcinosarcoma (1.8%), the latter composed of squamous cell carcinoma (SCC) and chondrosarcoma. Overall, adenocarcinoma was found in 35 (66.1%), SCC in 11 (20.7%), adenosquamous carcinoma (AdSC) in five (9.4%) and large cell neuroendocrine carcinoma (LCNEC) in two (3.7%) cases. All 52 cases of PSC had >10% of spindle cell carcinoma and giant cell carcinoma components. Tumours ranged in size from 1.0 to 10.0 cm in diameter with a median of 4.1 cm in greatest dimension.

Table 1.

Clinicopathological characteristics of the study cohort (n = 53)

Characteristics	Number (%)
Gender
Male	32 (60.3)
Female	21 (39.7)
Age range, median (years)	41‒84, 67
Smoking history
Current or former	51 (96.2)
Never smoker	2 (3.8)
Angiolymphatic invasion
Present	43 (81.1)
Absent	10 (18.9)
Visceral pleural invasion
Present	24 (45.2)
Absent	29 (54.8)
Histology
Adenocarcinoma	35 (66.1)
Squamous cell carcinoma	11 (20.7%)
Adenosquamous carcinoma	5 (9.4%)
Large cell neuroendocrine carcinoma	2 (3.7%)
Stage
I	29 (54.7%)
II	9 (17.0%)
III	7 (13.2%)
IV	8(15.1%)

MUTATIONS AND CNV

A total of 48 mutations (mean = 2.0; range = 0–6) were identified. The most commonly mutated gene was TP53 (17 of 23, 74.0%) followed by KRAS codon 12 of 13 (10 of 23, 43.4%). KRAS mutations were all found in smokers, distributed among eight (80%) PSCs with adenocarcinoma morphology, one (10%) AdSC and one (10%) SCC (Table 2). Figure 1 and Table 2 summarise the detected actionable and investigational variants by Ion AmpliSeq Cancer Panel. Only one PSC with adenocarcinoma histology had no identifiable mutation. Cases with frequent mutations (≥5) were all adenocarcinomas with a component of giant cell carcinoma.

Table 2.

Actionable and investigational genomic alterations detected by Ion AmpliSeq Cancer Panel^* among 23 cases

Tumour type	Ion AmpliSeq Cancer Panel actionable and investigational genomic alterations
	Gene	n (%)	Exon	Protein	cDNA	Mutation type
Adenocarcinoma (n = 18)	KRAS	8 (44.4%)	2	p.G12C (4) p.G12V (3) p.G13D (1)	c.G34T (4) c.G35T (3) c.G38A (1)	SNV missense
	EGFR	2 (11.1%)	21 20	p.L858R p.G779C	c.T2573G c.G2335T	SNV missense
	FBXW7	2 (11.1%)	10 7	p.L459V p.T267K	c.T1375G c.C800A	SNV missense
	CDKN2A	1 (5.5%)	2	p.P81S	c.C241T	SNV missense
	APC	3 (16.6%)	14	p.E1299Q p.T1538fs	c.G3895C (2) c.4613_4614insA (1)	SNV missense (2) Insertion (1)
	ATM	3 (16.6%)	17 9 50	p.F858L p.V410A p.E2446*	c.T2572C c.T1229C c.7335_7336TT	SNV missense (2) Substitution (1)
	MET	2 (11.1%)	14	p.T992I	c.C2975T	Splice (1) Missense (1)
	SMAD4	1 (5.5%)	12	p.I525V	c.A1573G	SNV missense
	JAK3	1 (5.5%)	16	p.V722I	c.G2164A	SNV missense
	PDGFRA	1 (5.5%)	15	p.H687R	c.A2060G	SNV missense
	PTEN	1 (5.5%)	6	p.R173L	c.G518T	SNV missense
Squamous cell carcinoma (n = 3)	HRAS	1 (33.3%)	3	p.Q61K	c.181A	SNV missense
	AKT1	1 (33.3%)	3	p.E17K	c. G49A	SNV missense
	KRAS	1 (33.3%)	2	p.G12D	c.G35A	SNV missense
Adenosquamous carcinoma (n = 2)	KRAS	1 (50%)	2	p.G12D	c.G35A	SNV missense
	ATM	1 (50%)	9	p.V410A	c.T1229C	SNV missense
	KIT	1 (50%)	18	p.S850I	c.G2549T	SNV missense

^*Ion Torrent, Life Technologies, Thermo Fisher Scientific, Waltham, Massachusetts.

Figure 1.

coMut plot representation of individual mutations and copy number variants (−c) present in 23 cases of pulmonary sarcomatoid carcinoma. Top: cases 1–23; left: percentages of alterations in each gene.

Among the 23 cases, Ion AmpliSeq Cancer Panel detected a total of 11 CNV (Figure 1 and Table 3). Gains in PIK3CA, EGFR, KRAS and MET were also confirmed by FISH. Additionally, there were copy number losses in RB1, confirmed by immunohistochemistry (Figure 1 and Table 3). There was no co-occurrence of MET amplification and MET exon 14 skipping mutation.

Table 3.

Copy number variants detected by Ion AmpliSeq Cancer Panel^* among 23 cases

Tumour type	Ion AmpliSeq Cancer Panel copy number alterations
	Gene	Gain (%)	Loss (%)
Adenocarcinoma (n = 18)	R81		2 (11.1)
	EGFR	2 (11.1)	-
	KRAS	1 (5.5)	-
	ATM		1 (5.5)
	MET	1 (5.5)	-
	STK11	1 (5.5)	-
Squamous cell carcinoma (n = 3)	Rs1	-	1 (33.3)
	P/K3CA	1 (33.3%)	-
Adenosquamous carcinoma (n = 2)	ATM	-	1 (50.0%)

^*ion Torrent, Life Technologies, Thermo Fisher Scientific, Waltham, Massachusetts.

SURVIVAL ANALYSIS

Mean follow-up was 28.8 months (range = 2–141, median = 16 months). Overall, 25 (47.1%) patients died within 2–105 months (mean = 22.7 months, median = 15 months) after diagnosis, and 28 were alive at 3–141 months (mean = 38.7 months, median = 21.5 months) after diagnosis. Five-year overall survival was 12.5% for the whole population. Kaplan–Meier survival analyses showed age greater than 70 years (P = 0.008), tumour size ≥4 cm (P = 0.02) and KRAS mutation (P = 0.01) among adenocarcinomas were associated strongly with worse overall survival (Figure 2A–C). There was no significant association between angiolymphatic invasion, visceral pleural invasion, tumour histology and stage with the outcome.

Figure 2.

Kaplan-Meier survival curves. A, Patient age >70 years; B, tumour ≥4 cm are associated significantly with worse overall survival. C, KRAS mutation in adenocarcinomas was associated significantly with poor survival.

Discussion

Pulmonary sarcomatoid carcinoma is a rare form of NSCLC characterised by high aggressiveness and mortality. The rare occurrence of PSC has restricted the characterisation of its genetic and molecular basis, thus impeding the development of targeted treatment protocols.

In this study, similar to previous reports, most patients were male, in the seventh decade of life and had a history of heavy smoking.^2,25 Tumours were found commonly as large masses, with a median diameter of 4.1 cm. We demonstrated that both older age (greater than 70 years) and large tumour size (greater than 4 cm) were associated with significantly worse survival (P = 0.008 and 0.02, respectively). With a mean follow-up period of 28.8 months the overall survival was poor, and only 12.5% of patients were alive at 5 years. Unlike previous studies,^2,25 we did not find a significant association between clinical stage and prognosis and this is perhaps because, for diagnostic purposes, we sought to include only surgically treated patients.

In our series, we demonstrated that PSCs harbour a broad spectrum of mutations, the most common being TP53 found in 74.0% of patients. These results are in accordance with those reported by Schrock et al.,³ who also identified TP53 mutations in 74% of their cases.^3,26 TP53 mutation often co-occurred with other mutations, with the most common being KRAS. We are uncertain about the significance of co-existing alterations in our study, but they were not of prognostic significance.

KRAS codon 12/13 mutations were the second most common mutation in our series, found in 43.4% of the overall cohort and 46.6% of PSC with adenocarcinoma component. This is slightly higher than the overall frequency of 33% in lung adenocarcinoma according to The Cancer Genome Atlas data;²⁶ however, it is in keeping with previous reports of KRAS mutations in PSC.^3,5,6,27 Prognostic significance of KRAS mutations in ‘pure’ lung adenocarcinomas is controversial,^28–32 with larger study cohorts indicating no apparent difference in outcome based on KRAS mutation status and subtype. In contrast, PSCs with adenocarcinoma morphology and KRAS codon 12/13 mutations in our study had a significantly worse outcome (P = 0.01) compared to KRAS wild-type. The number of cases is relatively small to make a reliable comparison based on KRAS mutation subtype. Interestingly, KRAS mutation was also identified in a single case of morphologically and immunohistochemically proven squamous cell carcinoma.

Recent studies indicate that inhibition of MET-driven oncogenic pathways has potential as a biomarker-driven targeted approach for PSC therapy.^{3,7,26,33–35} MET exon 14 mutations have been identified previously in up to 22% of PSC cases,^3,7,36 whereas others^3,4,9,27 have reported infrequent MET mutations, which may be due to differences in methodologies. In our series, MET amplification was seen in one case (5.5%) with an adenocarcinoma component and MET exon 14 skipping and missense mutations were identified in two (11.1%) cases with adenocarcinoma histology. In contrast to other studies, we did not find co-occurrence of MET amplification and mutation.^3,37 However, our findings further argue for the testing for MET mutations in PSC, as they may provide therapeutic options with MET inhibitors such as crizotinib in this setting.

Similar to other studies in the western population, the EGFR-sensitising mutation p.L858R was found in only one PSC (5.5%) with an adenocarcinoma component. Our data confirm previous observations that EGFR mutations are infrequent in PSCs,^3,6,27,37 limiting the clinical benefits from EGFR tyrosine kinase inhibitors in patients with PSC. Other targetable alterations, such as mutations in BRAF and HER2 or ALK and ROS1 gene rearrangements, were not identified in our cohort. Although these findings may be explained by a small number of cases, the rarity of these alterations and their associations with lack of smoking history and patient’s relatively younger age may be an alternative explanation. However, our study demonstrates that testing for genes outside the National Comprehensive Cancer Network and the College of American Pathologists/International Association for the Study of Lung Cancer/Association for Molecular Pathology guidelines may be potentially beneficial in this aggressive subtype of lung carcinoma.³⁸

Additional actionable and investigational variants were detected in ATM (17.3%), FBXW7 (8.7%), AKT1 (4.3%), PDGFRA (4.3%) and HRAS (4.3%), providing oncologists with options for potential therapeutic targets. In addition to mutations found in known cancer-associated genes, we detected and validated frequent copy number losses in RB1 (three of 23, 13.0%). RB1 deletion in one case was identified as an isolated event, but in the other two cases co-occurred with mutations, particularly p53 and KRAS. While the loss of RB1 has been reported recently in PSC,^3,7 its significance is uncertain.

Our study has some limitations. To increase the diagnostic accuracy we restricted our cases to only surgically treated patients, therefore decreasing the total number of cases for the study. Also, tissue blocks were not available for a large subset of cases, limiting molecular and statistical analyses. Further-more, the low number of MET exon 14 alterations in our study may be due to the limitation in coverage provided by the Ion AmpliSeq Cancer Panel. The majority of MET splicing mutations occur at the 3’ end of exon 14 in contrast to the 5’ end.³⁵ In our study and the one by Terra et al.,⁵ the amplicon in the Ion AmpliSeq Cancer Panel for MET exon 14 covers only the 5’ splice site and some intronic sequence but not the 3’ splice site (Supporting information, Figure S1). Therefore, an alternative sequencing approach may be considered if the initial results are negative for MET exon 14 alterations.

In summary, this study confirms that PSCs frequently harbour mutations in TP53 and KRAS genes among many others, probably contributing to patients’ decreased survival. Furthermore, we identified several actionable and investigational genomic alterations that could potentially increase targeted therapeutic options for these patients.

Supplementary Material

Supplemental Information

Figure S1. Coverage map for the Ion AmpliSeq Cancer Panel for MET exon 14.