Molecular characterisation of TP53 mutated squamous cell carcinomas of the lung to identify putative targets for therapy.

Personalised cancer treatment depends on identification of therapeutically relevant biological subgroups of patients for assessing effect of treatment and to discover new therapeutic options. By analyses in heterogeneous patient populations, the effects may be lost in noise. Squamous cell carcinoma of the lung is a major killer worldwide. Despite recent advances, mortality is high and response to therapies varies greatly from patient to patient. Target search in biologically relevant subgroups may identify treatment options not so far discovered. A total of 198 patients undergoing surgery for squamous cell carcinomas of the lung were included in the study. The tumours were analysed for copy number alterations (n = 152) and gene expression from tumour (n = 188) and normal lung (n = 21), with both data levels present in 140 patients. We studied alterations in tumours harbouring mutations in TP53 and in previously published gene expression subtypes. Genes with consistent alterations in both genomic levels were identified as putative biomarkers. Results were validated in TCGA. The most convincing biomarker in TP53 mutated squamous cell carcinomas of the lung was BIRC5 with amplification in 36% of mutated samples, 5% in wild-type samples and a 17%-fold change of expression between TP53 mutated tumours and normal lung tissue. BIRC5 was significantly altered in the classical and primitive subtypes. We suggest BIRC5 as a putative predictive biomarker and putative druggable target in squamous cell lung carcinomas harbouring TP53 mutation or classified as classical and primitive subtypes.

the Database for Annotation, Visualisation and Integrated Discovery

false discovery rate

Oslo University Hospital

significance analysis of microarrays

The Cancer Genome Atlas

INTRODUCTION

Personalised cancer treatment means giving a treatment that is likely to be effective in a given patient based on individual characteristics of the tumour/patient. Studies testing inhibitors in unselected patient populations are likely to fail due to a proportion of the patients lacking a target for the treatment and hence an effect. This may hinder the identification of treatments that may be effective for smaller subgroups of patient. There is hence a need to enrich the study population for patients likely to benefit from the treatment by identifying useful predictive markers.

Squamous cell carcinoma (SCC) of the lung is a major killer worldwide. Despite recent advances, mortality is high and response to therapies varies greatly from patient to patient. For the recently studied combination of chemotherapy (carboplatin/paclitaxel) and immunotherapy (pembrolizumab) the response rate was found to be 58%. For more traditional chemotherapy‐regimens, the response rates are 20%‐40%. , Although immunotherapy has revolutionised lung cancer treatment, most patients do not benefit from the treatment with response rates of 46% for the group with PD‐L1 > 50% for monotherapy. The greatly varying response rates reflect different underlying biologies. Tumour characteristics are known to be associated with treatment response in many cancers. For lung SCC, tumour expression of PD‐L1 and tumour mutational burden have shown association with response to immunotherapy. ,

TP53 is a well‐known tumour suppressor gene and inactivation, primarily through genetic mutations or loss of heterozygosity, is frequent as an early step in carcinogenesis. The gene product is essential in cellular response to stress by cell cycle regulation and induction of DNA repair and apoptosis. As mutations in TP53 are common in many cancer types, treatments linked to this genotype could potentially be effective in similar subgroups of various cancer diseases. Mutations in TP53 have been linked to exposure of various carcinogens, including tobacco and the incidence of TP53 mutations is particularly high in smoking‐associated cancers, such as SCC of the lung. Although TP53 mutations have been widely studied in cancer, therapies targeting the gene alterations have not been introduced to the clinic and alternative approaches are needed to improve the predictive power and response to therapy in patients with TP53 mutated lung cancers.

Subgroups of lung cancer with biological differences relevant to cancer treatment can be identified in various ways. For targeted therapy, the presence of a targetable mutation is essential. Biological differences with therapeutic consequences other than targetable mutations require a different approach. For breast cancer, unsupervised clustering of the 535 most variable genes from a 44K gene expression microarray chip identified biological subgroups that have proven to be robust and clinically relevant. , Some of these subgroups have later been found to overlap with subgroups determined by genetic alterations suitable for targeted therapy. Similar subtypes with varying prognosis have been published for lung SCC. Wilkerson and colleagues also used expression of genes with high reliability and variability to identify subtypes, but used a total of 2307 genes for unsupervised clustering. The four subtypes identified had specific molecular and clinical characteristics. Patients with tumours classified as the primitive subtype have a poor prognosis and the tumours display features associated with high proliferation. The classical subtype is more common and tumours of this subtype overexpress genes associated with xenobiotic metabolism. The secretory subtype is characterised by an immune profile and the basal subtype is dominated by cell adhesion. These subtypes have, however, not yet reached the clinic. In current clinical practice, PD‐L1 and sometimes tumour mutational burden are the biomarkers used for SCC, both with limitations in predicting therapy response.

In our study, we explore the biology of TP53 mutated lung SCC to propose putative biomarkers or targets of therapy. We have studied early‐stage lung carcinomas integrating information at the genetic and transcriptomic levels to identify alterations that could be predictive biomarkers in the presence of TP53 mutations and in the published subtypes of SCC. Based on these subgroups, we have identified putative biomarkers and plausible targets of therapy in these predefined biological subgroups.

MATERIALS AND METHODS

Patients undergoing surgery for SCC from 2006 to 2015 were included in the study (n = 198). All patients signed informed consent. The study was approved by the Regional Ethical Committee (ref: S‐06402b). Clinical data from medical journals including follow‐up was available for all patients. Tumour tissue from the tumour centre was snap‐frozen in liquid nitrogen and stored at −80°C until DNA extraction. A pathologist reviewed all samples to ensure a tumour percentage above 70%. Matched normal lung tissue was collected from 21 patients. Clinical data is shown in Table 1.

TABLE 1

Patient characteristics

	Total	TP53 mutated	TP53 wild‐type
Patients included (total)	198	76	35
Gender
Male	133 (67%)	52 (68%)	23 (66%)
Female	65 (33%)	24 (32%)	12 (34%)
Age
Mean (min‐max)	67 (43‐82)	67 (43‐82)	68 (58‐80)
Pack‐years
Mean (min‐max)	40 (3‐145)	38 (9‐145)	37 (3‐79)
pStage
1	113 (57%)	39 (52%)	24 (69%)
2	59 (30%)	26 (34%)	7 (20%)
3	25 (13%)	10 (13%)	4 (11%)
4	1 (0%)	1 (1%)	0 (0%)
TP53
wt	35 (18%)
mut	76 (38%)
NA	87 (44%)
Survival status
Lung cancer	57 (29%)	28 (37%)	8 (23%)
Cardiovascular	9 (5%)	4 (5%)	4 (11%)
Other	39 (20%)	16 (21%)	8 (23%)
Alive	92 (46%)	28 (37%)	15 (43%)
Follow‐up time, months
Mean (min‐max)	56 (2‐143)	62 (2‐130)	68 (6‐143)

Copy number alterations

DNA was extracted from the lung tumours using Maxwell® 16 DNA Purification Kits according to the standard Technical Manual, Literature # TM284 (http://www.promega.com) and using a Maxwell®16 instrument. Affymetrix Genome‐Wide SNP Nsp/Sty 6.0 Arrays were used for copy number alterations performed by AROS Applied Biotechnology A/S (www.arosab.com; project no. A2714, samples no. 095‐188). Raw signal intensities were extracted and quantile normalised using Affymetrix Power Tools (APT) and were converted to Log R ratio (LRR) and B Allele Frequency (BAF) using PennCNV‐Affy package. Log R values then corrected for GC bias. A locus is denoted as amplified if the copy number is 3 or higher.

Gene expression

Gene expression analysis was performed as previously described. For RNA extraction from 188 tumours and 21 matched normal samples, standard TRIZOL methods (Invitrogen, Carlsbad, California) were used as specified by the manufacturer's instructions. NanoDrop ND‐1000 spectrometer (NanoDrop Technologies, Wilmington, Delaware) was used to determine RNA quantity and quality (yield, 260/280 ratio and 260/230 ratio). RNA integrity numbers (RIN) were measured by the 2100 Bioanalyzer (Agilent Technologies, Santa Clara, California) according to the manufacturer's protocol. Gene expression microarrays from Agilent technologies (SurePrint G3 human GE v3, 8 × 60 K) were used for gene expression. The raw data were processed with Agilent's Feature Extraction Software with default parameters (Agilent Feature Extraction version 10.7.3.1). Probes were collapsed by median, samples were quantile normalised, and the data were log2 transformed.

sequencing

Consecutive samples (n = 111) were tested for TP53 mutations by the Sanger Sequencing method as previously described. In brief, an Applied Biosystems 3730 DNA analyser was used according to the supplier's handbook, Applied Biosystems 3730/3730X/DNA Analysers Part 4331467 Rev.B (http://www.appliedbiosystems.com). Exons 2 to 11 were investigated. SeqScape v 2.5 was used to align and analyse the sequences (TP53 accession no: NM_000546 (http://www.ncbi.nlm.nih.gov/nuccore/NM_000546; TP53 ref_NC000017.9_NT010718.15). Two independent readers analysed each sequence.

Bioinformatic analyses

Allele‐specific copy number analysis of tumours (ASCAT) was applied to estimate copy number gains (amplifications) and losses (deletions). Segmentation of the genome was obtained using the Allele‐Specific Piecewise Constant Fitting (ASPCF) algorithm with penalty parameter γ = 25 and the data were used in ASCAT to estimate tumour ploidy, aberrant cell fraction and the allele‐specific copy number profile of the tumour samples. In the OUH cohort, flat profiles of 46 samples indicated insufficient percentage of tumour cells and these samples were excluded from further analysis, leaving available information about copy number alterations for 152 patients. Similarly, copy number aberration profiles were obtained for the TCGA cohort (n = 443).

Identification of copy number gains and losses

For each sample, we obtained segmentation of the genome where each segment corresponds to a genomic region with a specific copy number. The copy number of the segment is the sum of the copy number of the major and minor allele in that region. Tumour ploidy was calculated as the median total copy number across the complete genome and was rounded to the nearest whole number. Segments with copy number greater than the ploidy were called as gains (amplifications) whereas the segments with copy number less than the ploidy were called as losses (deletions). The coordinates given in the Affymetrix array annotation file were used to map aberrant genomic regions to the genes.

Frequency plots

Samples were assigned to two groups based on the presence of TP53 mutations, as TP53 mutant or wild‐type. Frequency of gains (or losses) at every genomic position in a group was calculated as fraction of number of samples with gain (or loss) at a position to total number of samples in the group. Frequencies of gains are plotted on the y‐axis in a positive scale while the frequencies of losses are plotted in a negative scale. Chromosome‐wise genomic positions are represented on the x‐axis. Regions with significant differences in gain or loss at given position in two groups were determined using two‐proportions z‐test implemented in prop.test function in R.

Correlation analysis

The Spearman correlation coefficient was calculated using the R (v3.5.0) to estimate the correlation between copy number alteration and mRNA expression in 140 OUH samples and 443 samples in TCGA cohort. Expression data for 12 samples in OUH cohort were unavailable. Copy number value for the gene was calculated by subtracting the absolute ploidy from total copy number of the sample at the gene position. The quantile‐normalised gene expression data in OUH cohort and the RSEM normalised gene values in TCGA cohort were used in correlation analysis. The P values and adjusted P values (FDR) were reported for the significant association at P < .05.

Other analyses

Subgroups of lung SCC were identified by the presence of TP53 mutations and by the previously published lung SCC subtypes.

The candidate genes were tested for survival in patients with TP53 mutated tumours receiving adjuvant platinum‐based chemotherapy. Gene expression of the transcripts was divided in high and low by the median value.

To identify genes associated with subtype, significance analysis of microarrays (SAM) was performed with the four subtypes as stratifiers and 100 permutations using the R‐package samr v3.0. Two‐sided SAM was used to identify genes differentially expressed between two groups of patients. A significance level of 0.05 was used and false‐discovery rate (FDR) was used to correct for multiple testing. Only genes with both gene expression and copy number alteration data available were included.

Lists of putative genes were analysed using Reactome Analysis Tool , to identify pathways overrepresented in candidate genes.

RESULTS

Gene expression data were available for 188 of 198 tumours and 21 matched normal samples. Information about copy number alterations was available for 152 patients, with both data levels available for 140 patients.

mutated tumours

Patient characteristics are listed in Table 1. TP53 mutation status was available for 111 of 198 patients. Of these 111, 76 (68%) were mutated. Of the 198 samples, information about copy number alterations was available for 152 samples. TP53 mutation status was available for 90 samples of which 72 (80%) were mutated. In addition, one silent mutation was counted as wild‐type. Of the 18 wild‐type samples, 10 (56%) had deletions in the region harbouring the TP53 gene, 8 (44%) had no genomic alteration in the gene and there were no amplifications in this specific region. Of the samples with a mutation in TP53, four samples (6%) also had a genomic amplification in the same region. The TP53‐mutations detected are listed in File S1.

Frequency plots of TP53 wild‐type (n = 18 OUH, n = TCGA) and TP53 mutant (n = 72 OUH, n = TCGA) tumours were generated separately (File S2, Figure S1). Significant differences in amplifications and deletions were observed in 9668 genes of 15 995 tested (60%) with a significance level of 0.05 (prop.test). The highest percentage of significant amplifications in TP53 mutated vs wild‐type samples was found on chromosomes 2 (10.7%), 12 (19.9%) and 17 (13.3%), respectively. TP53 mutant tumours demonstrate a higher level of aberrations than wild‐type tumours.

Putative target genes

To identify putative target genes in TP53 mutated tumours, we followed an approach specified in Figure 1. For TP53 mutated samples, putative genes were those with amplification in more than 25%, correlation between copy number alterations and gene expression and, compared to wild‐type tumours both overexpression of the transcript and more than the double frequency of amplifications. In the OUH cohort, 860 genes met the four criteria and 148 of these were validated in the TCGA cohort and are putative biomarkers. Of these 148 genes, 70 have differential expression between TP53 mutated tumours and normal lung tissue in the OUH cohort, of which 67 are upregulated in tumours and constitute our candidate biomarkers (File S3 and Figure 1).

FIGURE 1

Target gene search for TP53 mutated squamous cell carcinomas. Significant genes are validated in TCGA

The 10 genes with highest fold change in gene expression between normal and tumour samples are listed in Table 2. The single gene with the highest fold change gene expression between tumour and normal samples is BIRC5. TKI1 has the highest gene expression in TP53 mutated samples compared to all genes, but due to the relatively high expression in normal samples, the fold change is higher for BIRC5. All 67 genes are listed in Table 3 according to cytoband. In addition, three genes had significantly lower gene expression in normal tissue compared to tumour samples, notably CNP, FBXL20 and GABARAPL1.

TABLE 2

Top 10 putative targets genes amplified and overexpressed in TP53 mutant, ranked according to the highest fold change in mutant compared to wild‐type tumours by Significance Analysis of Microarrays (SAM)

	SAM fold change	Average absolute expression values for each gene				Average expression values for each gene divided by average expression value of all genes in normal samples
		Normal	TP53 wild‐type	TP53 mut	TP53 NA	Normal	TP53 wild‐type	TP53 mut	TP53 NA
BIRC5	16.8	145	1857	3122	2637	0.17	2.17	3.65	3.09
TK1	11.4	768	6750	11 125	10 025	0.90	7.90	13.02	11.73
EPN3	5.8	55	233	401	386	0.06	0.27	0.47	0.45
TOP2A	5.2	44	176	281	267	0.05	0.21	0.33	0.31
CBX2	5.2	121	538	900	879	0.14	0.63	1.05	1.03
JUP	5.2	1010	3825	6833	6290	1.18	4.48	7.99	7.36
BUB1	4.8	55	212	295	275	0.06	0.25	0.35	0.32
PDK1	4.7	173	669	837	803	0.20	0.78	0.98	0.94
SPAG5	4.2	258	766	1141	1061	0.30	0.90	1.33	1.24
CENPA	4.1	40	104	186	169	0.05	0.12	0.22	0.20

Note: All have FDR < 0.01. Mean absolute gene expression is listed for normal lung tissue samples, tumours harbouring TP53 mutations and TP53 wild‐type tumours and for tumours not analysed with respect to TP53. Average expression values for each gene are divided by the average expression value for all genes in all normal samples (854.7).

TABLE 3

Putative target genes amplified and overexpressed in TP53 mutated lung squamous cell carcinomas, listed according to cytoband

2p11‐13	2p14‐16	2p21‐25	2q11‐14	2q31	12p12‐13	17q11‐12	17q21	17q25	18q11
BOLA3	MSH2	AGBL5	ANKRD23	ATP5G3	KIAA1467	ALDOC	BRCA1	ACTG1	SNRPD1
CCT7	MSH6	CENPA	ANKRD39	OLA1	LTBR	ERAL1	EPN3	BIRC5
DGUOK	MTIF2	CGREF1	BUB1	PDK1	PLEKHA5	MRM1	ETV4	CBX2
FBXO41	PSME4	EPCAM	CKAP2L		STRAP	POLDIP2	JUP	EIF4A3
GMCL1		EPT1	CLASP1		STYK1	PSMD11	KLHL11	ICT1
KDM3A		GEMIN6				SPAG5	NME1	MRPS7
MTHFD2		KIF3C				TLCD1	NME2	TK1
POLR1A		NOL10				TMEM97	PSMD3
SMYD5		PREB					PSME3
SNRNP27		SLC5A6					TOP2A
SNRPG							UTP18
SPR

Reactome analysis of the 67 overexpressed genes identified 168 pathways overrepresented with an FDR < 0.05 (File S4). Among these, there was an overrepresentation of pathways involved in cell cycle, cell‐cell communication, DNA replication, immune system and metabolism of RNA.

Looking for association to relapse (local or systemic) in the OUH cohort, log‐rank tests were performed using gene expression from samples with TP53 mutations (n = 105) and the list of 70 validated genes with differential expression in normal and tumour tissue. Genes with P values <.05 are listed in Table 4. None of the genes was significant after correction for multiple testing or in the TCGA cohort. The top gene, CNP, displayed a trend toward higher risk of relapse with lower expression of the gene; see Kaplan‐Meier plots of the four top genes in File S2, Figure S2. The other top genes, SNRPD1, EPCAM and CGREF1, all showed a trend toward a higher risk of relapse with higher expression of the gene. Gene expression of BIRC5 showed no association with relapse‐free survival by log‐rank test.

TABLE 4

Log‐rank test results for progression‐free survival of the 70 putative target genes divided in high or low by the median gene expression

Gene	Median	P value	FDR
CNP	7.1	.01	0.25
SNRPD1	12.0	.01	0.25
EPCAM	11.7	.02	0.25
CGREF1	6.4	.02	0.25
SLC5A6	8.1	.03	0.25
MTIF2	9.1	.03	0.25
FBXO41	9.2	.03	0.25
PLEKHA5	6.7	.03	0.25
ATP5G3	13.2	.04	0.25
EPT1	7.4	.04	0.25
KLHL11	7.5	.04	0.25
BOLA3	11.6	.04	0.25
ANKRD23	7.3	.05	0.27

Analysing the expression of BIRC5 with respect to relapse‐free survival in all OUH samples with available gene expression and available survival data (n = 184), there is no significant difference in survival between patients with high and low expression of the gene (File S2, Figure S3). The same is seen using only patients with TP53 mutated tumours (n = 74). In the TCGA cohort, we lacked information about relapse‐free survival, but could confirm the lack of prognostic value of BIRC5 expression with respect to overall survival.

Target gene relation to survival in patients receiving adjuvant chemotherapy

To look for primary chemoresistance, the 70 putative genes were tested for relation to relapse‐free survival in patients receiving adjuvant platinum‐based chemotherapy (n = 38 including nine relapses). One gene, PDK1, had a P value < .05, but none reached statistical significance after correction for multiple testing (FDR; File S2, Figure S4). There is a trend toward worse prognosis in platinum‐treated patients if the tumour expresses high BIRC5 with low expression of PDK1. Analysing the expression of BIRC5 with respect to relapse‐free survival in patients receiving adjuvant or neoadjuvant platinum therapy (n = 48), there is no significant difference in survival between patients with high and low expression of the gene (File S2, Figure S3). Removing the three neoadjuvant samples does not alter the result. Performing the same analysis in patients with TP53 mutated tumours all receiving adjuvant chemotherapy (n = 19), there is a borderline significance toward better prognosis for patients with higher expression of the gene.

Target gene search within specific gene expression subtypes

The presence of TP53 mutations is one possible way of identifying subgroups for target gene search. Another plausible subgrouping of SCC lung carcinoma is based on gene expression and is published as intrinsic subtypes. Using these subtypes, target gene search was performed according to the following criteria: (a) Genes amplified in more than 25% of the samples assigned to the subtype. (b) Correlation between copy number alteration and gene expression with (FDR < 0.05). (c) Significant differential expression of genes between subtypes (SAM q < 5%) and contrast in the specific subtype > 0 (signifying a higher gene expression in this subtype). For validation in TCGA, the following criteria were applied: (a) Genes amplified in more than 25% of the samples assigned to the subtype. (b) Correlation between copy number alteration and gene expression with (FDR < 0.05). Genes for which gene expression was missing in either dataset were removed. The candidate genes for each subtype validated in the TCGA are listed in File S5. For the classical and primitive subtypes, BIRC5 is listed as a top candidate gene. Testing the occurrence of TP53 mutations in the gene expression subtypes, we found an uneven distribution of mutations with the highest frequencies in the classical and primitive subtypes (Fisher exact P value = .04; Table 5).

TABLE 5

The fraction of TP53 mutations vary across gene expression subtypes, with the highest fractions in the classical and primitive subtypes, Fisher exact P value = .04

	TP53
	Wild‐type, n (%)	Mutated, n (%)
Subtype
Basal	10 (38)	16 (62)
Classical	11 (23)	37 (77)
Primitive	0 (0)	5 (100)
Secretory	13 (50)	13 (50)
	34 (32)	71 (68)

The candidate genes for each subtype were entered into Reactome for pathway analyses. Pathways with FDR < 0.05 are listed in File S4. While most subtypes have a large representation of pathways involved in the cell cycle, for the Secretory subtype, all significant pathways are involved in the immune system. In addition to cell cycle‐related pathways, the Basal, Classical and Primitive subtypes have some pathways involved in mitochondrial elongation and metabolism of proteins.

We wanted to explore if the expression of BIRC5 would vary between previously published subtypes. The gene expression of the transcript was not significantly different between subtypes by Student's t test (File S2, Figure S5).

DISCUSSION

As an effort to meet the challenges of heterogeneity of squamous cell lung cancer and the lack of useful biomarkers, we have performed a focused target gene search in surgically removed samples harbouring TP53 mutations, using a TCGA cohort for validation. We have identified several putative target genes, focusing particularly on BIRC5 which is amplified and overexpressed in TP53 mutated squamous cell lung carcinomas as well as in the two gene expression subtypes with highest percentage of TP53 mutations; Classical and Primitive.

The gene BIRC5 (Baculoviral IAP Repeat Containing 5) is a member of the inhibitor of apoptosis (IAP) gene family and encodes a protein called survivin. The transcript is highly expressed during foetal development and in malignant tumours, but has a low expression in adult normal tissues. The gene is included in numerous prognostic and predictive cancer signatures such as the Oncotype DX used for node‐negative breast cancer, the proliferation score for breast cancer and found prognostic in stage III NSCLC. In NSCLC cells, BIRC5 has been identified as a target of miR‐195 and as a regulator of apoptosis, senescence, migration and invasion. , Analyses of gene expression pathways has identified BIRC5 as a hub gene and a promising therapeutic target in lung adenocarcinomas.

We do not find any indication that BIRC5 expression is prognostic in itself, but it remains worth exploring as a predictive marker and as a target of therapy.

Downregulation of BIRC5 by various micro‐RNAs have increased chemo‐sensitivity in cancer cell lines. , Similarly, one study found that inhibition of NF‐κB reduced expression of BIRC5 in lung cancer stem cells and hence induced apoptosis.

BIRC5‐targeting therapy includes various inhibitors and vaccines. Despite promising preclinical studies, none of the drugs have found their way into the clinic so far. Phase I‐II studies have been performed, but have lacked sufficient evidence of effect, and some have even proven to be too toxic. , Vaccines targeting survivin (BIRC5 protein) have been tested in patients with glioma (peptide based vaccine) and in patients with prostate cancer (dendritic cell based vaccine), both with good tolerance and effect.

One possible approach in the future studies of BIRC5‐targeting therapies could be to improve patient selection to include patients with tumours harbouring TP53 mutations or displaying the classical or primitive gene expression subtypes only, or to stratify the survival analyses on these features.

Targeting BIRC5 to increase chemosensitivity has been tried with success on cell lines from various cancer types chemotherapy. , We found a trend toward a prognostic value in the subpopulations of patients who had received adjuvant or neoadjuvant platinum‐based therapy and had a TP53 mutated tumour (P = .08) that could indicate a primary resistance to platinum‐based chemotherapy. To our surprise, higher expression of the gene conferred better prognosis in this population. The sample size was small (n = 19) and the results needs to be confirmed. As we lack information about adjuvant chemotherapy in the TCGA cohort, we could not validate this result. If high gene expression is actually associated with favourable prognosis after adjuvant chemotherapy, BIRC5‐inhibition to increase chemo‐sensitivity does not seem like a good idea. One study of acute myeloid leukaemia cells found that the combination of an inhibitor of BIRC5 (YM155) and chemotherapeutics were either synergistic or antagonistic depending upon the type of chemotherapy used and the type of cancer cell studied.

Other genes present themselves as candidate genes combining overexpression and amplification in TP53 mutated lung squamous cell carcinomas (Table 2). After BIRC5, Thymidine Kinase 1 (TK1) has the highest fold change between wild‐type and mutated samples. The gene is also a candidate for the primitive subtype (File S5). This gene resides in the same cytoband as BIRC (17q25) and is involved in DNA repair. It is a known biomarker of cancer which can be measured in the serum in early stages of a malignant disease and it has a prognostic value. There is, however, no treatment strategy available targeting this molecule. Other genes among those with highest overexpression also reside on 17q (TOP2A and JUP). These genomic regions are on the same chromosome as TP53 (17p13.1), but on a different chromosome arm. Since there is not only amplification, but also overexpression, we expect the genes to have an individual role, not only as passengers due to chromosomal localisation close to a driver.

Expression of BIRC5 was not associated with survival in our cohort. After correction for multiple testing, none of our putative target genes were significant (Table 4), and their prognostic role is therefore questionable. The therapeutic value is, however, independent of the prognostic value.

Several of the putative genes identified in our study (Table 2) have previously been tested as targets of therapy. Anticancer therapies targeting TOP2A have been developed (Delgado), but have not been widely used due to increased risk of developing leukaemia (Chen). Although there are no drugs targeting CBX2 directly, there are inhibitors of histone deacetylase that indirectly modulate the targets of CBX2. Increased effect of several anticancer drugs is seen after inhibition of BUB1. For further exploration of these drugs, identification of a relevant subtype for testing is crucial and our study may help in that regard.

Driver genes have previously been identified by using Pearson correlation between copy number alteration and mRNA expression value across different TCGA datasets. They used a study population of different cancer types, identifying genes that were amplified in two or more datasets, narrowing the list down to cancer‐associated genes with druggable properties. Searching for targets across datasets with several cancer types is reasonable, since similar genetic alterations may occur in different cancer diseases leading to a target‐driven treatment rather than diagnosis‐driven treatment. We have chosen a different approach, subgrouping the patients in order to identify less heterogeneous groups that may allow for identification of targets that may be missed due to noise in bigger populations.

Pathway analyses of candidate genes for each gene expression subtype identified an overrepresentation of pathways involved with the immune system for the samples assigned to the Secretory pathway. This could be linked to response of immunotherapy. Unfortunately, there is not enough information to analyse clinical effect of check‐point inhibition in our data, but this should be elucidated in future studies.

CONCLUSION

Personalised therapy requires identification of biologically relevant subgroups for discovery of new therapies. The gene TP53 is frequently mutated in various cancer diseases, including squamous cell lung carcinomas (68% in our cohort). As an important player in DNA repair and DNA maintenance, deficiency in TP53 affects the biology of the cancer cells profoundly. Indeed, we do observe a high level of genomic aberrations in the TP53 mutated cancers. Since therapies targeting TP53 have not found their way into the clinic, alternative strategies for developing new therapies in TP53 mutated tumours are needed. Our study presents a target gene search combining copy number alteration and gene expression to identify putative genes for therapeutic and predictive approaches. BIRC5 is confirmed as an appealing target of therapy in TP53 mutated squamous cell carcinomas and in the gene expression subtypes classical and primitive. Future clinical studies should consider stratifying on these parameters or preselecting patients to increase the possibility of identifying patients that will benefit from BIRC5‐directed treatment strategies. Other putative target genes are identified and our study may help decide which groups of patients should be selected for further testing of treatments targeting these genes.

CONFLICT OF INTEREST

The authors declare no potential conflicts of interest.

ETHICS STATEMENT

All patients signed informed consent. The study was approved by the Regional Ethical Committee (ref: S‐06402b).

File S1

Click here for additional data file.

File S2

Click here for additional data file.

File S3

Click here for additional data file.

File S4

Click here for additional data file.

File S5

Click here for additional data file.