Molecular profiling of lung adenosquamous carcinoma: hybrid or genuine type?

Lung adenosquamous carcinoma is a particular subtype of non-small cell lung carcinoma that is defined by the coexistence of adenocarcinoma and squamous cell carcinoma components. The aim of this study was to assess the mutational profile in each component of 16 adenosquamous carcinoma samples from a Caucasian population by a combination of next generation sequencing using the cancer hotspot panel as well as the colon and lung cancer panel and FISH. Identified mutations were confirmed by Sanger sequencing of DNA from cancer cells of each component collected by Laser Capture microdissection. Mutations typical for adenocarcinoma as well as squamous cell carcinoma were identified. Driver mutations were predominantly in the trunk suggesting a monoclonal origin of adenosquamous carcinoma. Most remarkably, EGFR mutations and mutations in the PI3K signaling pathway, which accounted for 30% and 25% of tumors respectively, were more prevalent while KRAS mutations were less prevalent than expected for a Caucasian population. Surprisingly, expression of classifier miR-205 was intermediate between that of classical adenocarcinoma and squamous cell carcinoma suggesting that adenosquamous carcinoma is a transitional stage between these tumor types. The high prevalence of therapy-relevant targets opens new options of therapeutic intervention for adenosquamous carcinoma patients.


INTRODUCTION
Lung cancer is the leading cause of cancer related death worldwide with a mean 5 year survival rate of less than 15% [1]. 85% of lung cancer is classified as nonsmall cell lung carcinoma (NSCLC). The two predominant histological subtypes are adenocarcinoma (AD), which account for 50% of NSCLC, and squamous cell carcinoma (SQ), which account for 40% of NSCLC [2]. AD have glandular histology and intracellular mucous production and express thyroid transcription factor 1 (TTF1) consistent with an origin in the distal lung. By contrast, SQ, which arise in more proximal airways, are more reminiscent of pseudostratified columnar epithelium of the upper airways and are characterized by keratinisation, intercellular desmosome formation and expression of p63 [3] and p40 [4]. In addition, both entities are characterized by a different set of driver mutations. Mutations of the oncogenes KRAS (25-40%), EGFR (10-15%), BRAF (2-4%), and HER2 (2%), and translocations of ALK (5-7%) and ROS1 (2%) are found predominantly in AD at frequencies indicated in brackets, most of which are targets for currently available or potential targeted therapies [5]. Potentially targetable driver mutations are less frequent in SQ than in AD. These include mutations of the discoidcontaining receptor 2 (DDR2) (3.8%), PTEN (10%) and PIK3CA (16%), or amplification of FGFR1 (20%) [6,7].
Adenosquamous carcinoma (ADSQ) of the lung is a rare subtype of NSCLC that accounts for 2-4% of lung cancer [8]. This type is characterized by the morphological presence of both SQ and AD patterns, each comprising at least 10% of the total tumor volume. The two components represent the typical immunoprofile of AD and SQ differentiation in the lung, i.e. TTF1 and napsin positivity (80% and 58% respectively) for the AD component and p63 and CK5/6 positivity (100% and 73% respectively) for the SQ component [9]. p40 is considered superior to p63 in terms of specificity of classical SQ [4], but whether this marker allows for the discrimination of AD and SQ components of ADSQ is not well studied. The prognosis of ADSQ is generally worse compared to classical AD or SQ, independent of the ethnic background [10][11][12].
ADSQ is a striking example of a morphologically dichotomous tumor whose genomic landscape has yet to be systematically probed for its contributory role to this dichotomy. In classical lung AD, multiregional sequencing revealed that the majority of mutations of cancer genes were trunk mutations, which represent ubiquitous mutations present in all regions of the tumor, and one out of 21 mutations within cancer genes was a branch mutation [13]. The latter class of mutations represents heterogeneous mutations present only in one region of the tumor. Intratumoral heterogeneity may have an impact on biopsy strategy, as single biopsies may be inadequate for identifying all cancer gene mutations. These findings may have consequences on treatment planning since important drug targets may be missed. The degree of tumor heterogeneity correlates with the likelihood of postsurgical relapse in patients with localized lung cancer [13].
The aim of this study was to assess tumor heterogeneity of ADSQ by a combination of next generation sequencing (NGS) using the Cancer hotspot panel as well as the Colon and Lung Cancer panel and fluorescence in situ (FISH) analysis. To verify mutations identified by NGS, tumor cells were collected from AD and SQ components by laser capture microdissection and analyzed by Sanger sequencing. This allowed us to assess if each component harbors a different set of mutations. To our knowledge, this is the first comprehensive study assessing the mutational status of cancer-related genes in ADSQ.

Patient's collective
A total of 16 patients diagnosed with adenosquamous carcinoma of the lung at the Institutes of Pathology in Bern and Locarno were enrolled in this analysis. The characteristics of the patients are shown in Supplementary Table 1. The mean age was 76 years and most patients were male. Thirty percent of tumors showed evidence of lymph node metastasis and the majority of tumors showed moderately differentiated histology.

Analysis of AD and SQ components of ADSQ
Serial tissue sections from paraffin blocks were performed as outlined in Supplementary Figure 1. The tumor region encompassing the AD and SQ components were defined by histological and immunohistochemical criteria. Morphological criteria for the AD component were glandular histology and intracellular mucous production. Morphological criteria for the SQ component were pseudostratified columnar histology, keratinisation and formation of intercellular desmosomes ( Figure 1A and 1B) (see WHO guide lines [1][2][3]). Immunohistochemical staining for Thyroid transcription factor 1 (TTF1), a marker specific for AD of the lung [9], gave rise to a significantly stronger signal in the AD component compared to the SQ component in 13 out of 16 ADSQ (Table 1 and Figure 1C and 1D). Tumors #3, 9 and 10 revealed either no staining or equal staining for TTF1 in both components. These tumors were analyzed for the expression of napsin, an alternative marker for classical AD. All tumors showed high level expression of napsin in the AD component, but no expression in the SQ component (Table 1). By contrast, all ADSQ samples revealed much stronger staining for p63 in the SQ component than in the AD component (Table 1 and Figure  1E and 1F). Consistent with this result, the p40 marker, which is considered a more specific marker than p63, gave rise to intensive staining of the SQ component but no staining of the AD component in 15/16 ADSQ (Table 1).
In conclusion, all tumors could be clearly classified as ADSQ based on immunohistochemical stainings.

Mutational analysis of AD and SQ components of ADSQ
In a next step, the mutational profile of each component of ADSQ was assessed. To this end, regions encompassing AD and SQ components, as indicated by immunohistochemical stainings, were macroscopically dissected from unstained tissue slides. Regions containing > 50% normal lung tissue or inflammatory infiltrates were excluded from the analysis. DNA was extracted from corresponding regions and subjected to NGS using the Cancer Hotspot panel, which covers most of the important cancer genes. DNA was also subjected to NGS using the Colon and Lung Cancer Research panel, which allows the assessment of the mutational status of additional genes including DDR2, MAP2K1 and FBX7, which are not covered by the Cancer Hotspot panel. Sequence analysis revealed an average number of 480, 000 mapped reads and an average mean depth of 2, 000 reads. Variant calling was filtered for nonsynonymous SNPs, indels and spliced variants. FFPE material may result in false positive low frequency (< 5%) calls owing to fixation artifacts. To remove false positive calls, variant allele frequency threshold was set to 5%. Common germline mutations that are predicted to have no effect on the protein function, as indicated by SIFT/polyphen analysis, were also discarded.
Supplementary Table 2 shows all predicted nonsynonymous somatic changes that were identified by NGS using the Cancer Hotspot and Colon and Lung Cancer panels. A total of 26 somatic mutations were identified in 15 out of 16 tumors while one tumor revealed no mutation. Among the identified mutations, 3 indel mutations, 2 splice-site mutations and 21 nonsynonymous SNPs were obtained. Nine out of 26 mutations revealed variant read frequencies that were > 3 times higher in one component compared to the other component of the same tumor sample, suggesting that these mutations are branch mutations (Supplementary Table 2). These include TP53 p.T155I, p.Y126C and p.P190L mutations in samples #4 and #13, CDKN2A p.W110* mutation in #5, PIK3CA p.E545K mutation in #4 and 10, and RB1 p.G203fs*8 and PTEN p.Q171* mutations in #16. All other mutations seem to be trunk mutations as they gave rise to variant read frequencies that were similar in both components.
Although corresponding regions were carefully macrodissected, it cannot be excluded that some DNA samples may be cross-contaminated by the DNA from the other component. To assess this possibility, cells were collected by laser capture microdissection from each component and analyzed by Sanger sequencing for mutations identified by NGS. To assess if identified mutations were indeed of somatic origin, normal tissue collected from the same patient was also included. There was a high agreement between the results obtained by NGS and Sanger sequencing: not only were the same mutations identified but also the assignment of these mutations to the trunk or branches, respectively, was consistent between both sequencing methods (Supplementary Table 2).
Somatic mutations identified in this study are presented in Figure 2. Our results indicate that mutations in the EGFR signaling pathway are among the most prominent mutations in ADSQ. Five out of 16 tumors harbored bona fide EGFR mutations including a p.E746_ E750 deletion in sample #1 and a p.L858R missense mutation in samples # 2, 6, 12 and 16 (Supplementary Table 2, Figure 2 and Figure 3A, 3B). These mutations confer increased sensitivity to tyrosine kinase inhibitors [14,15]. In each case, the same EGFR mutation was detected in both components of the tumor ( Figure 2 and Figure 3A, 3B) indicating a monoclonal origin of ADSQ. Interestingly, EGFR mutations were more prevalent than expected for a Caucasian population. By contrast, KRAS mutations, which are present in 25-40% of AD [5], were not detected in our collective of ADSQ samples.
Mutations in the PI3K signaling pathway are typical for SQ [7]. Four ADSQ tumors harbored mutations in the PIK3 pathway; one tumor harbored a p.C420R mutation in the C2 region and two tumors harbored a p.E545K mutation in the helical region of the PIK3CA gene, and one tumor harbored a p.Q171* mutation in the PTEN gene. All three mutations are predicted to result in constitutive activity of the PI3K signaling pathway [16]. In tumor sample #12, the same mutation was detected in both components ( Figure  2 and Figure 3C) while tumor samples #4, #10 and #16 harbored mutations that were only present in one, but not the other component ( Figure 2 and Figure 3D). Thus ADSQ show mutations typical for both classical AD and SQ.
TP53 mutations were also common in our collective of ADSQ: 6 out of 16 tumors harbored TP53 mutations, 4 of which were trunk mutations while 2 were branch mutations. In particular, both components of tumor #4 contained different www.impactjournals.com/oncotarget TP53 mutations. This suggests that TP53 mutations can either occur before or after separation of the AD and SQ components from a common precursor. Most hotspot mutations of the TP53 gene including some of the mutations identified in this study are gain of function mutations, which lead to increased expression of p53 protein in the tumor tissue [17]. To assess if the expression pattern for p53 reflects the mutational profile, immunohistochemistry was performed using a monoclonal -, negative; +, weak positive; ++, strong positive or amplified; na, not analyzed antibody directed against p53. As shown in Table 1 and Figure  4A (Table 1). In addition, mutations in cell cycle genes including those in Rb and CDKN2A were identified in 3 tumor samples (#5, 8 and 16). These mutations are predicted to lead to enhanced cell cycle progression of tumor cells. It remains to be shown if the p.P1369V missense mutation in the APC gene in tumor #3 results in enhanced activity of the β-catenin pathway. A G > C transversion at the 5′ splice site of exon 4 of STK11 was also identified in both components of tumor #5. Finally, a p.M541L mutation in the transmembrane domain of the KIT gene and a p.V722I mutation in the JAK3 gene were identified (samples #7, 14 and 15, Figure 2). These sequence alterations have been identified as somatic mutations as well as sequence variants occurring in the normal population, and it is not clear whether or not these sequence alterations may affect protein functions. The mutations identified in our collective were clearly somatic mutations as they were not present in normal tissue collected from the same patient.
ALK and ROS1 translocations are also important oncogenic drivers of NSCLC [5]. Since these genomic alterations cannot be detected by NGS using the Cancer hotspot or Colon and Lung Cancer panels, FISH analysis was performed. However, no ALK or ROS1 translocation was detected in our collective of ADSQ (Table 1). Tumor #2 gave rise to EGFR amplification in the AD component, but not the SQ component.

Analysis of classifier miR-205 in ADSQ tumor samples
MicroRNAs are important regulators of diff erentiation and their expression profile correlates significantly with the state of differentiation of various tissues [18]. Hence miRNAs proved to be very useful for tumor classification [19]. To assess if ADSQ components reflect the expression pattern of classical AD and SQ, classifier miR-205, which is upregulated in SQ [20], and miR-21, which is upregulated in both tumor subtypes, were analyzed. As shown in Figure 5, miR-21 was expressed at similar levels in both components of ADSQ as well as in classical AD and SQ. In agreement with published results, the expression level of miR-205 was 200 times higher in classical SQ than in classical AD (p = 0.05). In contrast, no significant difference in the expression level was observed between AD and SQ components of ADSQ.  In conclusion, although AD and SQ compo nents clearly differ by morphological as well as immunohistochemical criteria, they cannot be discriminated by means of mutational profiling of cancer genes or by expression analysis using classifier miRNAs.

DISCUSSION
Only a few studies exist which assessed the mutational profile of ADSQ [21][22][23][24][25]. In almost all studies Asian populations were analyzed, but it is not known if this ethnic group may exhibit a different mutational profile compared to Caucasian populations. In most cases mutational analysis was restricted to EGFR or KRAS. We are the first to provide a mutational analysis of ADSQ from Caucasian patients using an extended set of cancerrelated genes including those which are frequently mutated in classical AD or SQ. All somatic mutations which were identified by NGS were confirmed by Sanger sequencing in cancer cells collected from each component by laser capture microdissection. This approach ensures that there is little cross-contamination between ADSQ components and therefore allows an unequivocal assignment of mutations to either component. In contrast, most other studies used macroscopically dissected material, which may be potentially cross-contaminated.
We show for the first time that ADSQ harbor driver mutations, which are either specific for classical AD or

Figure 5: Expression analysis of miR-21 (upper panel) and miR-205 (lower panel) in ADSQ components. Expression of
miR-21 or miR-205 relative to the expression of RNU-48 is shown for classical AD and SQ and for AD and SQ components of ADSQ. The difference in the expression of miR-205 between classical AD and SQ was statistically significant (p = 0.05). www.impactjournals.com/oncotarget SQ including mutations in EGFR or genes in the PI3K pathway. The surprisingly high prevalence of EGFR mutations in ADSQ is a key finding of this study. This has significant therapeutic implications: small tyrosine kinase inhibitors gefitinib and erlotinib are clinically effective for the treatment of NSCLC harboring activating EGFR mutations [14,15]. Thus, although ADSQ is more aggressive compared to classical AD or SQ [12], more patients diagnosed with ADSQ may benefit from targeted therapy. In contrast, KRAS mutations were less prevalent than expected for a Caucasian population. The finding that EGFR was more prevalent and KRAS was less prevalent in ADSQ than in classical AD of patients of the Caucasian ethnic group suggests that carcinogenesis of ADSQ may be more similar between Western and Asian populations compared to classical AD. Tochigi et al. [23] reported similar EGFR mutation frequencies in classical AD and ADSQ within a Caucasian ethnic group which is in contrast to our findings. The reason for the discrepancy of these results is not clear, but it may either be explained by the relatively small cohort of patients used in these studies or, alternatively, by the analysis of different cohorts with specific risk factors. We show that EGFR mutations are present in both components suggesting that patients with ADSQ may be accessible to therapy using EGFR tyrosine kinase inhibitors. Indeed, several case studies reported good therapy responses of patients with ADSQ to EGFR inhibitors [24].
In addition to EGFR mutations, we observed a high incidence of mutations in genes of the PI3K signaling pathway, some of which were trunk mutations and some were branch mutations occurring in the AD or SQ component. Preclinical studies suggested that patients with PIK3CA or PTEN mutations may benefit from panclass I PI3K inhibitors [26,27]. In addition, lung cancer cell lines harboring PIK3CA or PTEN mutations were sensitive to dual PIK3CA/mTOR inhibitors [28,29], which raises the hope that patients with mutations in the PI3K pathway may benefit from this type of targeted therapy in the future. Introduction of activating PIK3CA mutants into EGFR mutant cell lines conferred resistance to EGFR inhibitors [29]. Based on these results we may conclude that patient #12, who contains an EGFR L858R mutation as well as a PIK3CA C420R mutation in the tumor DNA, may be less responsive to EGFR or PIK3CA inhibitors, but this has not yet been confirmed by clinical studies. It is possible, however, that this patient may benefit from new combinations of targeted drugs.
Our results indicate that driver mutations were normally in the trunk. This suggests that ADSQ are of monoclonal origin, which is in agreement with findings obtained by others [22,23,25,[30][31][32] and which may also be true for other heterogeneous lung tumors [31]. The molecular results support the histopathologist historical "morphology only" view of ADSQ being a real entity fitting in the "lung heterogeneity concept" [33,34]. Mather et al. [35] reported the isolation of cancer stem like cells from human ADSQ that recapitulated all features of ADSQ in mouse xenograft experiments, consistent with a monoclonal origin of ADSQ. However, the molecular mechanism of (trans)-differentiation into AD or SQ components is not clear. Loss of STK11 or targeted deletion of PTEN or TGFβR1 in the oncogenic KRAS G12D mouse model gave rise to ADSQ-like tumors [36]. Emerging tumors initially showed AD histology and expressed AD-specific markers, but STK11-deficient lung AD progressively transdifferentiated to SQ [37], suggesting that ADSQ are a transitional state between AD to SQ differentiation. Conversely, expression of SQ-related antigen in the AD component of ADSQ was higher while Mucin1 was lower than in classical AD suggesting a transition from SQ to AD [38]. Our findings that ADSQ can harbor mutations typically found in AD or SQ, respectively, may suggest that ADSQ have developed either from an AD-like or SQ-like precursor cell. Five ADSQ tumors (#1, 2, 6, 12 and 16) harbored EGFR mutations suggesting a transition from AD to SQ. In contrast, tumors #4, 10 and 16 harbored a PIK3CA/ PTEN mutation in the SQ branch suggesting that acquisition of this mutation induced the transition into the SQ component. Expression of classifier miR-205 in the AD and SQ components of ADSQ was intermediate between that of classical AD and SQ consistent with the finding that ADSQ is a transitional state between classical AD and SQ, but this possibility has to be confirmed by extended miRNA expression profiling or methylation analysis. By contrast, both components of ADSQ also clearly show hallmarks of classical AD or SQ as indicated by histomorphological criteria as well as expression of TTF1, napsin, p63 or p40.
Multiregional sequencing of classical AD revealed that only 1 out of 21 cancer genes were in the branch, while all other mutations were trunk mutations [13]. In contrast, our study revealed that one out of four mutations was a branch mutation. In particular, TP53 mutations were often identified as branch mutations. These results indicate that separation of AD and SQ components may have occurred very early during the development of ADSQ. The fact that ADSQ reveals a high incidence of branch mutations, makes it an attractive model for studying tumor heterogeneity.
In conclusion, we show that ADSQ harbors driver mutations specific for both classical AD and SQ. Owing to the high prevalence of these mutations, a higher proportion of patients may benefit form targeted therapy than expected for a Caucasian population. In addition, our results have diagnostic implications, especially in advanced tumors which are not resected and where only small biopsies are available for molecular testing: since the most relevant targetable driver mutations are mainly found in the trunk, they may not be missed by mutational analysis in such specimens. However, a problem that arises using small biopsies is that ADSQ can present as SQ if the AD component is missed. Consequently molecular testing for the most important drugable targets in AD like EGFR, ALK or ROS1 will not be performed, leaving the patient without this therapeutic option. Nevertheless, this issue may become less relevant in the future, when all NSCLC patients are analyzed by NGS.

Patients and samples
The ethical boards of the Inselspital Bern and the institute of Pathology of Locarno have approved the study as part of the general approval for research on formalin fixed paraffin embedded tissue (KEK Nr. 200/2014). Using the full text research tool of our database software (Pathowin +), we selected 35 cases of adenosquamous carcinoma patients diagnosed in 1993-2007. All cases were re-reviewed by two experienced pathologists for confirmation of tumor histology following the guidelines of the WHO 2004 classification [2]. After review of the histological slides 16 cases of operation specimens were selected for the study. Nineteen cases were excluded due to pretreatment, extensive necrosis or ambiguous immunohistochemical staining.

Serial tissue sectioning, laser capture microdissection and nucleic acid extraction
Serial 3-6 μm tissue sections were performed from paraffin blocks as outlined in Supplementary Figure 1. AD and SQ components of each tumor were macroscopically dissected for DNA and RNA extraction using the QIAamp DNA Micro Kit (QIAGEN) and RecoverALL Total Nucleic Acid Isolation kit (Ambion), respectively, following the manufacturer's instructions. Methylol groups that are covalently bound to nucleic acids during formalin treatment were cleaved from nucleic acids as described [39].
Laser capture microdissection was performed on 6 μm sections on Polyethylene Naphthalate (PEN)membrane slides [39]. Sections were deparaffinized and stained with cresyl violet following the manufacturer's instruction (Zeiss). Approximately 1000 tumor cells for each component were captured onto an adhesive cap using a PALM Microbeam (PALM, Zeiss). DNA was extracted from the dissected material as described above.

Fluorescence in-situ hybridization
The gene status of ALK (Vysis LSI ALK Dual Color, break apart rearrangement probe, Abbott Molecular, Des Plaines, IL, USA), ROS1 (ZytoLight SPEC ROS1 Dual Color Break Apart Probe, Zytovision, Bremerhaven, Germany) and EGFR (Vysis LSI EGFR spectrumOrange/ CEP7 spectrumGreen Probe, Abbott Molecular) were evaluated using commercially available FISH probes. 2-3 μm tissue sections were deparaffinized and pretreated using the commercial pretreatment kit Vysis (Abbott Molecular) and hybridized with the probe overnight in a humidified chamber at 37°C. Post hybridisation washes were performed using the Pathvysion kit (Abbott Molecular) and counterstained with DAPI. Thresholds for a positive signal constellation for the ALK or ROS1 probes were set to 15%. Threshold for a positive EGFR signal, as indicated by an EGFR to CEP7 ratio > 2.0, was set to 10%. At least 60 cells were counted.
Raw data analysis was performed using the Ion Torrent Suite software 4.2 relative to the human reference sequence (hg19). For variant calling (v4.2.1.0) somatic low stringency configuration was used. Detection of single nucleotide polymorphisms or indel polymorphisms was performed using the Torrent Reporter 4.2 and the AmpliSeq CHPv2 single sample workflow. Alternatively, sequence alignment and variant detection was performed using the CLC Cancer Research Workbench version 1.5.2. Variant allele frequency threshold was set to 5%. Identified mutations were visually inspected using the IGV software.

Sanger sequencing
Mutations identified by semiconductor sequencing were confirmed by Sanger Sequencing. To this end, DNA was extracted from laser capture microdissected material and amplified by PCR in a total volume of 30 μl containing 0.5 μM of the same primer pair as was used for library preparation for NGS and 1 Unit Hot Start Taq Polymerase (Qiagen) with an initial hold stage at 95°C for 10 minutes followed by 40 cycles of 94°C, 60°C and 72°C for 1 minute each. The sequence of the primers is available at the homepage from Life Technologies. PCR products were purified using the ExoSap-IT cleanup kit (Affymetrix) and subjected to sequencing using the Big Dye Terminator v1. 1 cycle sequencing kit (Applied Biosystems). Sequencing products were purified using the Big Dye x-Terminator purification kit (Applied Biosystems) and analysed using the Genetic Analyzer 3500 (Applied Biosystems). Sequence analysis was performed using the Sequencing Analysis Software 5.4 (Applied Biosystems).

Real-time PCR
Real-time PCR was performed as described [39]. Quantitative PCR was carried out using TaqMan assay (Applied Biosystems) and the One-step PCR system (Applied Biosystems). The mean C T was determined from triplicate experiments. miRNA levels were normalized to the level obtained for RNU48. Changes in expression were calculated using the ΔΔC T method.

Statistics
Statistical analyzes were performed using the GraphPAD prism software. Statistical differences were calculated using unpaired two-tailed Student's t test. A probability of p ≤ 0.05 was considered statistically significant.