Deciphering the clonal relationship between glandular and squamous components in adenosquamous carcinoma of the lung using whole exome sequencing

Adenosquamous carcinoma of the lung (ASC) is a rare subtype of non-small cell lung cancer, consisting of lung adenocarcinoma (LUAD) and lung squamous cell carcinoma (LUSC) components. ASC shows morphological characteristics of classic LUAD and LUSC but behaves more aggressively. Although ASC can serve as a model of lung cancer heterogeneity and transdifferentiation, its genomic background remains poorly understood. In this study, we sought to explore the genomic landscape of macrodissected LUAD and LUSC components of three ASC using whole exome sequencing (WES). Identified truncal mutations included the pan-cancer tumor-suppressor gene TP53 but also EGFR , BRAF, and MET , which are characteristic for LUAD but uncommon in LUSC. No truncal mutation of classical LUSC driver mutations were found. Both components showed unique driver mutations that did not overlap between the three ASC. Mutational signatures of truncal mutations differed from those of the branch mutations in their descendants LUAD and LUSC. Most common signatures were related to aging (1, 5) and smoking (4). Truncal chromosomal copy number aberrations shared by all three ASC included losses of 3p, 15q and 19p, and an amplified region in 5p. Furthermore, we detected loss of STK11 and SOX2 amplification in ASC, which has previously been shown to drive transdifferentiation from LUAD to LUSC in preclinical mouse models. Conclusively, this is the first study using WES to elucidate the clonal evolution of ASC. It provides strong evidence that the LUAD and LUSC components of ASC share a common origin and that the LUAD component ap- pears to transform to LUSC.


Introduction
Adenosquamous carcinoma (ASC) is a rare histological subtype of lung cancer accounting for 0.4 %-4 % of all lung cancers [1]. It consists of two morphologically distinct components including LUAD and LUSC [1]. From a clinical point of view, the biphasic ASC represents a challenge. It is more aggressive than its classical single components and has been shown to be associated with worse outcome [2]. There is no specific standard treatment for ASC, and current standard of care options rely on general non-small cell lung cancer (NSCLC) guidelines. Surgical resection is the only effective mean to treat patients with ASC, mostly in conjunction with platinum-based adjuvant chemotherapy [3]. Targeted treatments can be used as first-line therapy for advanced EGFR-mutant or ALK-rearranged ASC, but there are limited data on the efficacy of EGFR-or ALK-tyrosine kinase inhibitor (TKI) in ASC due to its rarity (reviewed in [4]). From a diagnostic perspective, it is challenging to grasp the biphasic nature of the tumor. Due to limited sampling by small biopsies, there is a high chance to miss one component so that the subsequent treatment will be based on only one of the components.
Previous studies focusing on oncogenic driver mutations revealed that ASC have similar mutation profiles and therapeutic targets as LUAD including EGFR mutations [5,6]. Yet, to the best of our knowledge, the genomic profiles of the LUAD and LUSC components within ASC and the inferred evolutionary trajectories have not been explored previously, beyond the use of single gene and gene panel testing [5,6]. Being defined by its distinct morphological heterogeneity, ASC represents an ideal model to study morphological transdifferentiation and clonal evolution in NSCLC. Common LUAD is considered to originate from stem cells at the bronchoalveolar junction, whereas LUSC derive from the more proximally located basal cell compartment of the bronchial epithelium [7]. Based on targeted sequencing, several mutations were detected in both components LUAD and LUSC of ASC suggesting the two entities share the same ancestor cell [5,6]. It has been hypothesized that they are likely to arise at the bronchoalveolar junction as LUAD and that the LUSC phenotype develops subsequently [5,6]. The molecular mechanisms of this adenosquamous transdifferentiation (AST) remain unknown.
This study aims at better understanding the genomic landscape and the evolutionary trajectories of ASC. To this end, we dissected the LUAD and LUSC components separately from three ASC patients for comprehensive genomic profiling of both components using whole exome sequencing (WES).

Patient cohort & immunohistochemistry
Three cases of ASC were retrieved from the pathology archive of the Institute of Pathology and Medical Genetics, University Hospital Basel, Switzerland. Representative formalin-fixed paraffin-embedded sections were cut (4 μm thick) and stained with hematoxylin and eosin. TTF-1, Napsin A, CK7 and p40 were used to distinguish LUAD and LUSC components. TTF-1 (Ventana catalog number 790-7456), Napsin A (Ventana catalog number 760-4867), CK7 (Ventana catalog number 790-4462) and p40 (Ventana catalog number 790-4950) staining was performed on the Benchmark Ultra (Ventana, Roche) as described previously [8]. The sections were reviewed by a pulmonary pathologist (LB), who marked distinct morphological regions for macrodissection. FISH analysis was performed according to the manufactures protocol (ZytoLight® SPEC SOX2/CEN3 dual color probe, catalog number Z-2127− 200, Zytovision, Bremerhaven, Germany). This study has been carried out under the ethical approval EKNZ 31/12.

Macrodissection and DNA extraction
Up to ten consecutive 10 μm thick unstained tissue sections were cut on glass slides. The distinct morphological components and matched normal lung tissue were identified and marked by a pathologist (LB), scratched from the glass slide using a 25 g disposable syringe under a stereoscope as previously described [9]. Scratched cells were subjected to 200 μl of ATL buffer and 40 μl Proteinase K and incubated overnight at 56 • C with 500 rpm using the reagents from DNeasy Blood and Tissue Kit (Qiagen, Germantown, MD, USA) according to the manufacturer's instructions. DNA was quantified using the Qubit Fluorometer assay (Life Technologies, Carlsbad, CA, USA) as previously described [9].

Whole exome sequencing and variant annotation
Extracted DNA from normal and macrodissected ASC components were subjected to WES. SureSelect Human All Exon V6 Kit (Agilent) was used for the whole exome capturing according to manufacturer's guidelines. Sequencing was performed on Illumina NovaSeq 6000 using paired-end 100-bp reads and yielded a median depth of coverage of 48x to 120x in tumor components and 49x to 122x in the corresponding germlines (Supplementary Table 1). Sequencing was performed by CeGaT (Tübingen, Germany).
Reads were aligned to the reference human genome GRCh37. Somatic single nucleotide variants (SNVs) and small insertions and deletions (indels) were detected using MuTect (v.1.1.7) and Strelka (v.2.9.10), respectively [10,11]. To reduce false positive results from artifacts caused by formalin fixation, specific SNVs C:G > T:A with variant allelic fraction (VAF) less than 10 % were discarded. Otherwise, SNVs or indels with a VAF < 1% or that were covered by fewer than 3 reads were discarded, if the SNVs were found in both components of one sample, a cut-off of two read was applied. We further excluded variants identified in at least two of a panel of 123 non-tumor samples, including the non-tumor samples included in the current study. Variant annotation was performed by SnpEff software v.4.1 [12].
The heatmap of non-synonymous mutations was generated using the R package maftools v.2.0.16 by selecting the genes in the Bailey et al. dataset that represents the significantly mutated genes in classic LUAD and LUSC obtained from TCGA [13].

Copy number aberration analysis and clonality analysis
Allele-specific CNAs were identified using FACETS v.0.5.14 [14]. Deletions/losses were defined with the log-ratio < -0.3 and amplifications/gains > 0.3. Cancer cell fraction (CCF) of each mutation was identified using ABSOLUTE v.1.0.6 [15]. A mutation was classified as clonal if its probability of being clonal was > 50 % or if the lower bound of the 95 % confidence interval of its CCF was > 90 %. Mutations were considered as subclonal, if they did not meet the mentioned criteria [16,17]. A mutation that was found in both components of one biopsy was considered as 'trunk'. Mutations that were detected in only one component of the tumor were granted as 'branch' or 'private'.

Phylogenetic analysis and mutational signatures
A maximum parsimony tree was built for each case using binary presence/absence matrices based on the repertoire of non-synonymous and synonymous somatic mutations, gene amplifications and homozygous deletions in the biopsies of the tumors, as described by Murugaesu et al. [18,19].
Decomposition of mutational signatures was conducted using the R package deconstructSigs by selecting mutational signatures based on the set of 30 mutational signatures ('signature.cosmic') that were observed in LUAD and LUSC [20,21].

Adenosquamous carcinoma have different growth patterns
The main clinico-pathologic characteristics of the three patients with ASC are summarized in Table 1. The three tumors were at stages pT2-pT4. In all three patients, the LUAD area was larger than the LUSC area. All LUSC areas were homogeneously p40-positive, while all LUAD areas were p40-negative by immunohistochemistry (Fig. 1). Two LUAD were TTF-1 positive (P118 & P119), while the other one was negative for both TTF1 and Napsin A. There was a predominant acinar growth pattern of the LUAD component in all three ASCs ( Fig. 1 and Supplementary Fig. 2). Additional LUAD patterns were lepidic (P118 & P119) and solid (P119 & P120). Keratinization in the LUSC component was observed in P118 and P119.
Across the samples, most of the mutations in driver genes were clonal (Fig. 2b). The analysis of P118 displayed four shared cancer gene mutations, of which only EGFR exon 19 deletion was clonal in both. TP53 p. Gln192* was mutated in both components and a further TP53 p. Gly108Asp mutation was detected only in the LUAD component. P119 shared three clonal mutations MET, BRAF and P119 ACVR1B mutations. We detected shared cancer gene mutations between the components of P120. Moreover, one clonal STRN mutation was detected in P120 LUAD. Further, four clonal and one subclonal private mutations were found in the LUSC component.
Gene copy number data from all patients and tumor components revealed high concordance between the two components (Fig. 2c).

Evolutionary history of genetic alterations
Most mutations in known cancer driver genes were truncal (Fig. 3). Most mutated cancer genes were known as typical for LUAD such as EGFR, MET and BRAF. P118 harbored a deletion in EGFR exon 19, a well-known driver gene in LUAD [6,22]. We found only KMT2D in P120, but not any other mutations in LUSC typical driver genes (Fig. 4). Interestingly, there were fewer mutations in the branches than truncal   mutations. Truncal copy number variations (CNVs) might also play a role as drivers of malignancy. Loss of regions at chr 3p, 15q, 19p and an amplified locus at chr 5p were identified across all samples. Losses in regions of chr 6, 17 and 20 in LUAD of P118 were normal in LUSC. LUAD P119 had private losses in regions of chr 19, whereas the LUSC component harbored private losses in chr 12. Further private losses in chr 12 were identified in P120 LUSC. Regions of chr 9 were gained in LUSC P120 but lost in the LUAD component. Notably, there was truncal loss of STK11 at 19p13 in P118 and P120. In P120, we detected a truncal amplified region on 3q including TP63 and SOX2 (Fig. 3). SOX2 amplification was confirmed using FISH ( Supplementary Fig. 1).
Given that truncal genomic alterations are likely acquired prior to the branching of the two components, we sought to investigate whether mutational signatures are shared or change over time (Fig. 3). For all our three ASC, we found different patterns of signatures as well as commonalities.
Trunks of all three tumors displayed signatures 1 and 4. The truncal mutations of P118 and P119 were also characterized by the APOBEC signature 13. In addition, trunks of P118 and P120 carried signature 5. The signatures changed drastically in the branches. Whereas, P119 LUAD was found to exhibit signatures 1, 2, 4 and 13 the LUSC component only exhibited signature 1 and 6. Signature 1 contributed to more than 50 % of the mutations of all LUAD/LUSC branches, except of P118 LUAD where it was not present at all. The smoking signature was also seen in both components of P120 and in P119 LUAD, but not P119 LUSC. Several signatures were observed exclusively in a single branch, including signature 13 in P119 LUAD, signature 6 in P119 LUSC and  signature 17 in P120 LUAD. Notably, signature 6 was observed in P119 LUSC, which is not common in classic LUSC, but prominent in LUAD.

Discussion
ASC had originally been assumed to be a mixed tumor with a polyclonal origin consisting of two separate tumors until recent gene panel sequencing demonstrated shared mutations in LUAD and LUSC [5,6]. Here we provided unequivocal evidence for a monoclonal origin of the LUAD and LUSC components in ASC based on comprehensive mutational patterns and CNVs obtained by WES.
In our study, truncal genetic alterations included not only mutations in common cancer genes such as TP53 but also the typical driver gene mutations of LUAD in EGFR, BRAF and deletions of STK11 (Fig. 3). This is in line with previous studies using NGS panels or selected single gene testing of the two ASC components [5,6]. In particular, activating EGFR mutations have been shown to be at least as common in ASC as in classical LUAD ranging from approximately 15 % in Western patients to 50 % in Asian patients, while they have been reported in only ~2% of LUSC [22][23][24]. This has also been emphasized by a recent study of Lin et al., who showed that the genomic ASC profiles by targeted sequencing looked more similar to classic LUAD than LUSC [6]. In accordance, typical LUAD mutations were predominant in our study. We identified only one KMT2D mutation as a putative LUSC driver [25,26]. KMT2D mutations are not exclusive for LUSC as they occur, albeit less frequently, also in LUAD according to TCGA (21 % vs. 8%) (http://cbiop ortal.org) [27]. In summary, the clonal relationship between the two components in ASC together with the presence of more LUAD than LUSC mutations suggest that ASC originates from an early LUAD ancestor. Interestingly, 23 lymph node metastases from 12 patients in the recent study by Lin et al. contained either LUAD or ASC (11, each), while only one consisted of pure LUSC [6]. This further supports that an early LUAD-like ancestor might not only be the driving force during tumor development but also during progression.
Tumor evolution is influenced by exogenous and endogenous mutational processes that shape the spectra of different mutational signatures [28]. It was previously demonstrated that LUAD and LUSC share common signatures of type 1 (CpG deamination and aging), 2 (APOBEC), 4 (smoking), 5 (transcriptional strand bias for T > C) 13 (APOBEC). In addition, LUAD was found to display signatures 6 (defective DNA-repair) and 17 (unknown origin) [21]. In our study, the trunks of all three ASC were dominated by aging and smoking-related signatures 1 and 4, respectively, which is partly explained by a smoking history of the patients. Interestingly, the distribution of signatures changed markedly in the LUAD and LUSC branches. This is in line with previous studies on LUAD, LUSC and breast cancer showing that mutational signatures may change from the trunk to branches [19,29,30]. Change in mutation signatures can provide additional insights into the development and evolutionary traits of malignant tumors [31,32]. Recently, Richard and colleagues found an association between mutational signatures and clinical efficacy of nivolumab, an immune checkpoint inhibitor, in lung cancer patients [33]. In our study, the trunks of all three ASC were dominated by aging and smoking-related signatures 1 and 4, respectively, which is partly explained by a smoking history of the patients. However, the distribution of signatures changed markedly in the LUAD and LUSC branches. Notably, we did not observe a systematic accumulation of the APOBEC signature, which was previously shown to drive mutagenesis and subclonal expansion in LUAD and LUSC [31]. Taken together, we found that most of the genomic alterations accumulated in the trunk before branching and changed over time. Aging and smoking-induced signatures 1 and 4 were the most prominent and present in almost all lineages suggesting a promoting role in tumor development and progression. Further studies are needed to better understand the biological significance and potential clinical impact of mutational signatures in lung cancer, and their association with histological transdifferentiation.
Our gene copy number data suggest persistent genomic instability as chromosomal aberrations are detected even after branching of the two components. The high overall concordance between the chromosomal copy number aberrations profiles of the components support our previous data from metastatic LUAD showing that most aberrant genomic events are truncal with only limited heterogeneity between primary tumors and matched metastases [34]. Nevertheless, the two ASC components had private CNVs, which were similar to the CNVs known to be typical for their pure morphological counterparts, i.e. classical LUAD and LUSC, respectively [35]. For example, the trunk of P120 and P119 LUSC had an amplified region on 3q including SOX2. SOX2 expression is known as a common feature of squamous differentiation, and SOX2 amplification occurs in 40 % of LUSC but is almost never seen in LUAD [36]. Thus, it is possible that SOX2 amplification served as a genomic precondition to facilitate the AST of truncal LUAD to LUSC in P119 and P120. This hypothesis would be in line with previous evidence from a mouse model, in which overexpression of SOX2 induced squamous differentiation [37,38]. In fact, AST was linked to silenced TTF-1 expression in the pre-clinical mouse model, suggesting that the differential regulation of TTF-1 and SOX2 expression might be involved in the transformation process. In P120, AST is difficult to prove. The truncal nature of SOX2 amplification and KMT2D mutation could point towards a reverse sequence with a LUSC-type ancestor clone despite the truncal STK11 loss. Deletion of STK11, which we found in P118 and P120, might pinpoint another mechanism of plasticity and AST. STK11 alterations in general, occur in up to 14 % in classic LUAD but in only 2% of classical LUSC (http://cbioportal.org) [39]. There is pre-clinical evidence from mouse models suggesting that different mechanisms in STK11 deficiency can lead to a transition from LUAD to LUSC [40,41]. Concluding, AST is a complex process that might be caused due to several different mechanisms that require further studies.
The main limitation of this study is the low number of patients reflecting the rarity of the tumor. In fact, this is the first study to perform comprehensive WES on the two components of ASC separately. Moreover, the usage of archived FFPE blocks and contamination with nontumor tissue negatively affect the resolution of our analysis. Nevertheless, our CNV data of SOX2 and STK11 support pre-clinical evidence of genomic driven mechanisms that could facilitate or mediate AST to LUSC. Although we discovered only KMT2D as a candidate mutation known to be typical but not exclusive for squamous differentiation, we cannot rule out a role of other mutations or mutational patterns. It is likely that AST in ASC is driven by both genetic and non-genetic mechanisms including transcriptional regulation and metabolic factors as discussed above. The role of epigenetic reprogramming has not yet been studied in the context of AST in ASC despite the paramount importance of epigenetics in shaping morphology [42].

Conclusion
Our whole-exome analysis showed compelling evidence for a clonal relationship between the LUAD and LUSC components of ASCs, and the emergence from a common LUAD-like ancestor clone. Even if the results are ambiguous, identification of SOX2 amplification and STK11 loss in two of the three LUSC components, each, supports previous pre-clinical evidence for a role of these genes in AST. Further and more comprehensive studies with a larger sample size are needed to gain a better understanding of the mechanisms that drive transformation from LUAD to LUSC or LUSC to LUAD, including transcriptomics and epigenomics.

Declaration of Competing Interest
The authors report no declarations of interest.