Multilayered Omics-Based Analysis of a Head and Neck Cancer Model of Cisplatin Resistance Reveals Intratumoral Heterogeneity and Treatment-Induced Clonal Selection

Treatment of locally advanced squamous cell carcinomas of the head and neck (SCCHN) remains a challenge. Despite multimodal treatment of the primary disease by surgery combined with adjuvant (chemo) radiation or definitive platinum-based concurrent chemoradiation (CRTX), less than 50% of patients with SCCHN can be cured. Frequently, locoregional recurrence occurs within the high-dose irradiation field. These recurrences are most likely driven by tumor cell subpopulations within the bulk tumor tissue with primary resistance to CRTX. Recent next-generation sequencing (NGS) studies provided evidence of a considerable interpatient heterogeneity in SCCHN and identified novel actionable cancer drivers and predictive biomarkers for targeted therapies (1–4). However, most of these studies used whole-exome sequencing with a mean sequencing coverage of ∼80-fold, with limited sensitivity in the detection of minor tumor cell subclones. Moreover, comparative NGS analyses of primary versus recurrent tumor tissues are currently lacking. Thus, the extent of intratumoral heterogeneity, the subclonal structures and the mechanisms of treatment-induced clonal selection by platinum-based regimens remain largely unexplored.

The remarkable breakthroughs in omics technologies within the last 10 years have opened new opportunities to identify genes and pathways deregulated in cancer. In the current study, we used such a multifaceted omics approach for dissecting the molecular mechanisms involved in primary cis-diamminedichloridoplatinum(II) (CDDP) resistance and treatment-induced clonal evolution in the SCCHN cell line FaDu. Comprehensive molecular characterization of CDDP-sensitive FaDu parental cells, a resistant subline and single-cell derived clones with sensitive/resistant phenotypes included targeted NGS (tNGS), fluorescence in situ hybridization (FISH), microarray-based transcriptome, and mass spectrometry-based phosphoproteome analyses. Candidate therapeutic targets were validated for their clinical relevance using the publicly available data of the TCGA SCCHN cohort (3).

The human papilloma virus negative (HPV⁻) SCCHN cell line FaDu was purchased from ATCC. For establishment of single-cell derived subclones, FaDu cells were seeded at a density of 250 cells per 10-cm petri dish. After the formation of small colonies, 96 individual clones were picked under microscopic control, transferred to a 96-well plate, and separately expanded. Drug sensitivity testing and detailed molecular characterization was started at passage 3. The identity of FaDu and derived sublines was regularly checked by high-throughput SNP-based authentication (Multiplexion). Cells were cultured in minimal essential medium (MEM) supplemented with 10% heat-inactivated fetal bovine serum and 1× nonessential amino acids. All cell culture reagents were purchased from GIBCO (Life Technologies). Cell cultures were incubated at 37°C and 5% CO₂ in a humidified atmosphere. Testing for mycoplasma contamination was performed at monthly intervals by reverse-transcriptase polymerase chain reaction (RT-PCR; ref. 5). CDDP (NeoCorp 1 mg/mL, Hexal AG) was provided by the hospital dispensary. Working solutions were freshly prepared from the stock solution by dilution in cell culture medium on the day of the experiment.

Cells were seeded into 96-well plates at a density of 300 cells/well. Twenty-four hours after seeding cells were treated with different doses of CDDP. For knockdown experiments cells were treated for 24 hours with a nontargeting control or p53 siRNA before addition of CDDP. At the end of the experiment, 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide reagent (MTT) was added to the cells. After 1 hour of incubation, formazan complexes were dissolved in DMSO and absorbance was measured with the AR2001 microplate reader (Anthos Mikrosysteme GmbH). Survival fractions after CDDP treatment were calculated on the basis of the survival of untreated cells. Survival fractions at each time point and dose level were determined in sextuplets. At least three independent experiments were carried out. From the dose–effect curves, the IC₅₀ values for CDDP were calculated using the CalcuSyn software (version 2, Biosoft).

Cells were seeded into 12-well plates at a density of 300 cells/well. Twenty-four hours after seeding, cells were treated with CDDP at the indicated dose. Alternatively, irradiation was applied using a 250 kV deep X-ray unit (Philips RT250, Philips Medical Systems GmbH). Untreated cultures were processed along with treated cultures. Cells were then incubated for up to 14 days. At the end of the experiments, colonies were fixed, stained using 10% Giemsa stain solution, and colonies containing >50 cells were counted. Survival fractions were calculated on the basis of survival of untreated cells. Each dose level was analyzed in triplicates in at least three independent experiments.

Expression levels of total proteins in cell lysates were assessed by standard Western blot analysis. Briefly, cells were harvested by scraping in RIPA buffer. Standard SDS–polyacrylamide gel electrophoresis was performed using 40 μg of total protein per sample, followed by transfer to PVDF membranes (Millipore). For detection, the following antibodies were used: anti-GAPDH (clone 14C10, Cell Signaling Technology), anti-phospho-p70S6K (clone 108D2, Cell Signaling Technology), and anti-p53 (clone DO-1). As secondary antibodies peroxidase-conjugated goat anti-mouse and goat anti-rabbit (both from Jackson ImmunoResearch Laboratories) were used. The immunoreactivity was detected using the Pierce ECL Plus Western Blotting Substrate (Thermo Scientific).

FISH analysis was performed on cytospin samples using routine methods. The SpectrumGreen-labelled Vysis LSI CEP 17 (D17Z1) probe targeting the cytogenic localization 17p11.1-q11.1 and the SpectrumOrange-labelled Vysis LSI TP53 probe directed against 17p13.1 were used for detection of chromosome 17 and the TP53 gene, respectively. Both probes were from Abbott Molecular.

Mutational profiling was performed using an in-house gene panel targeting 327 genes (supplementary file 1) with the Haloplex-targeting enrichment system (Agilen). Genes had been selected based on the results from whole-exome sequencing of three independent SCCHN patient cohorts (1–3) and the COSMIC database (6). The complete coding sequence of all exons of the 327 genes was covered, resulting in a target region of 1.45 megabase pairs in total. Details on DNA isolation, library preparation as well as further details for targeted NGS of formalin-fixed paraffin-embedded tumor samples from patients with SCCHN are provided in the Supplementary Information. Pooled libraries from cell lines were used for paired end (PE) sequencing on the Illumina MiSeq platform with MiSeq Reagent Kits V2 or V3 (Illumina), both producing 125 bp PE reads. A mean sequencing depth of 320-fold (range, 42–6620) was achieved in the target region. Sequencing data have been deposited in the ArrayExpress database at EMBL-EBI (www.ebi.ac.uk/arrayexpress) under accession number E-MTAB-6021.

Processing of the raw fastq files, sequencing adaptor trimming, sequence alignment and variant calling was performed with Agilent SureCall Software (version 3.0.1.4). For alignment to the human reference sequence build 38 (hg19), the BWA-MEM algorithm was applied. Variant calling was performed with a cutoff of 0.05 for allele frequencies. For comparative analysis between sensitive and resistant cells, only nonsynonymous variants detected at an allelic frequency of >0.2 in one of the cell lines or with a difference in allelic frequency of >0.2 between two different cell lines were included. Unsupervised cluster analysis based on the allelic fractions of nonsynonymous single nucleotide variations (SNV) was performed using the Broad Institute Morpheus software (https://software.broadinstitute.org/morpheus/).

Basal mRNA expression levels of the four FaDu clones were determined in two independent analyses using the HumanHT-12 v4 Expression BeadChip array (Illumina). All experimental details are given in the Supplementary Information. The mRNA expression data were deposited in ArrayExpress under accession number E-MTAB-6022.

Unsupervised cluster analysis of mRNA expression levels was performed using the Broad Institute Morpheus software (https://software.broadinstitute.org/morpheus/). Genes with normalized expression values above the background value of 150 in at least 50% of samples were included in the analysis. In the average-linkage cluster algorithm, Pearson correlation was used as dissimilarity measure. For gene-set enrichment analysis, we used the Broad Institute GSEA software (https://software.broadinstitute.org/gsea/) with the MSigDB hallmark (h.all.v6.0.symbols.gmt), oncogenic (c6.all.v6.0.symbols.gmt), gene ontology (c5.all.v6.0.symbols.gmt), and immunological (c7.all.v6.0.symbols.gmt) gene sets. In this analysis, we included all genes for which a normalized expression value >150 was detected in at least half of the analyzed samples, comprising expression values of 12126 probes capturing 7590 genes.

Phosphorylated peptides from cell lysates were enriched based on the method of Thingholm and colleagues (7) using titanium dioxide (TiO₂) chromatography. Subsequently, phosphopeptides were characterized by mass spectrometry. All further experimental details are provided in the Supplementary Information.

The interference of molecular markers with overall and disease-free survival in the TCGA HPV⁻ SCCHN dataset was determined by Kaplan–Meier analysis and the log-rank test. Cox proportional hazard regression analysis was used for calculation of survival probabilities and hazard ratios. All statistical analyses were performed using the SPSS software (v.22, IBM Corp.). P values of <0.05 were considered significant.

We previously established a CDDP-resistance model by long-term treatment of the SCCHN cell line FaDu with increasing doses of CDDP (8). The median-effect doses of CDDP in the parental cell line (IC₅₀ FaDu_CDDP-S: 24.8 ng/mL) and its CDDP-resistant derivative (IC₅₀ FaDu_CDDP-R: 288 ng/mL) were significantly different (Two-way ANOVA P < 0.001; Supplementary Fig. S1). Comparative mutational analysis of CDDP-resistant versus sensitive FaDu cells by tNGS revealed 294 nonsynonymous single-nucleotide variants (SNV), which were present in both cell lines. For only two of these variants (0.7%), a difference of >0.2 in the allelic fraction was detected between FaDu_CDDP-S and FaDu_CDDP-R (Table 1). Further, nine SNVs were exclusively found with an allele frequency of >0.2 in either of the two cell lines (Table 1). Within this group of genetic variations potentially linked to CDDP resistance, we identified two SNVs in the TP53 gene: the intronic mutation c.673 G>A leading to an early stop codon was detected at decreased allele frequency in FaDu_CDDP-R compared with FaDu_CDDP-s (Table 1). The TP53 p.R248L missense mutation previously described in the FaDu cell line (9) was found in FaDu_CDDP-R but not in FaDu_CDDP-S cells (Table 1), even when mutant variants at allele frequencies down to 0.01 were considered in variant calling.

Based on the tNGS results, we hypothesized that the parental FaDu cell line consisted of various subclones with distinct genotypes interfering with sensitivity to CDDP and that the relative proportion of these subclones had changed due to the selection pressure by long-term CDDP treatment. To verify this hypothesis, we used the treatment-naïve parental FaDu cell line (FaDu_CDDP-s) for picking of 96 single cell-derived clones. After expansion as separate cultures, these clones were tested for their sensitivity to CDDP. By this approach, we isolated individual cell clones that were primarily resistant and others that showed high sensitivity to CDDP (Fig. 1A). Among the most resistant clones were the clones 5 and 78 with survival fractions (SF) of 0.68 and 0.89, respectively. Among the most sensitive ones were clone 46 (SF 0.05) and clone 54 (SF 0.04; Fig. 1A). The observed SF of the sensitive and resistant clones was similar to those observed after 24-hour treatment of FaDu_CDDP-S and FaDu_CDDP-R using the same concentration of CDDP (Fig. 1B). Of note, the individual clones also showed significant differences in radiosensitivity (Fig. 1C). To make sure that the observed differences in SF after CDDP treatment were not the result of differences in basal proliferation rates of the clones, cells were cultured over a time period of 96 hours, and cell numbers were determined every 24 hours. No correlation between CDDP sensitivity and proliferation rate was observed (Supplementary Fig. S2).

Given the higher likelihood for identifying molecular determinants of sensitivity/resistance by comparing cells displaying large differences in drug sensitivity, we selected these four clones for further comprehensive molecular characterization. Importantly, sensitivity/resistance characteristics of clones were repeatedly checked and remained stable throughout the study.

Comparative genetic analysis by tNGS confirmed that the majority of the detected nonsynonymous SNVs were present at similar allelic frequencies not only in FaDu_CDDP-S and FaDu_CDDP-R but also in the four single-cell derived clones (Fig. 2A). Despite of this homogeneity at the level of point mutations, unsupervised cluster analysis based on allelic frequencies of all SNVs clearly separated two clusters of CDDP-sensitive and -resistant cells (Fig. 2A). The largest differences in allelic frequency were observed for six mutant variants affecting five genes (CDH11, CPXCR1, FN1, TPO, TP53; Fig. 2B and Table 1). Consistently, the TP53 p.R248L mutation was detected in CDDP resistant but was absent in the CDDP-sensitive clones. These data support our hypothesis of intratumoral genetic heterogeneity and treatment-induced selection of subclones with genetic features endowing them with primary resistance to CDDP.

The allelic frequencies found in the resistant clones were suggestive of the existence of more than two copies of chromosome 17 at which TP53 is located. FISH analysis of 100 interphase and 10 metaphase nuclei confirmed the aneuploid status of chromosome 17 in the majority of cells (Supplementary Fig. S3). In the resistant clone 5, 88 of 100 interphase and 9 of 10 metaphase nuclei contained three copies of TP53, and in clone 78 all cells carried from three up to twelve copies of the gene. In contrast, more than two copies of the TP53 gene were detected in only four out of 100 interphase and none of the metaphase nuclei in the sensitive clones (Supplementary Fig. S3).

In order to determine which genes were differentially expressed between resistant and sensitive clones, microarray transcriptome analysis was performed. Normalized datasets of basal mRNA expression levels established in two independent microarray analyses for the four FaDu cell clones were subjected to unsupervised hierarchical clustering using the Morpheus online tool. Two clusters of CDDP-resistant and -sensitive clones were identified (Fig. 2C). In contrast to the homogeneity at the level of point mutations, the mRNA expression patterns were significantly more heterogeneous not only between but also within the sensitive and resistant clones (Fig. 2C).

In order to evaluate which signaling pathways were differentially activated with respect to the phenotype, we performed a gene set enrichment analysis. The heatmap of the top-50 features associated with the CDDP-sensitive and -resistant phenotypes is presented in Supplementary Fig. S4A and the list of genes in Supplementary File 2. Of note, ten of the top 50 (20%) upregulated genes in the CDDP-resistant clones have previously been reported to be transcriptionally regulated by p53 (10) and the majority (64%) have been shown to interact with CDDP (Supplementary File 2). The signature of the sensitive phenotype was enriched for genes downregulated in cell lines with mutant TP53 status (enrichment score 0.55; P = 0.06, FDR q = 0.079, Supplementary Fig. S4B). Unexpectedly, we also observed a correlation between the sensitive phenotype and an interferon gamma response signature (enrichment score 0.72; P = 0.03, FDR q = 0.088). In contrast, no gene sets were significantly enriched in cells with a resistant phenotype.

To further clarify whether clonal differences in CDDP sensitivity might be associated with variable loss in transcriptional activity of mutant p53, we analyzed expression levels of three known p53 target genes. We observed reduced basal expression levels of GADD45A and CDKN1A, two genes from the p53 core pathway involved in stress-induced cell cycle arrest and DNA damage repair in resistant compared with sensitive cells. In contrast, basal expression levels of SULF2, a p53 transcriptional target involved in the senescence response of cells to genotoxic stress (11) were higher in resistant compared with sensitive cells (Supplementary Fig. S5A). CDDP-induced upregulation of GADD45A and CDKN1A was also less pronounced in resistant versus sensitive cells whereas only minor differences were observed between all four clones for CDDP-induced changes in SULF2 expression (Supplementary Fig. S5B).

For further identifying any potential molecular differences of sensitive and resistant clones, we profiled their phosphoprotein expression by shotgun mass spectrometry. Overall, >1440 phosphopeptides were detected in a paired analysis of whole-cell lysates (clone 5 versus 54, clone 46 versus 78). Of these, 972 were consistently detected in the two independent experiments. When applying a twofold cutoff on fold changes, 24 phosphopeptides were found to be upregulated and 39 downregulated in the resistant compared with the sensitive clones in at least three of four possible paired analyses (Supplementary File 3). Kinase enrichment analysis based on the phosphopeptides upregulated in resistant clones revealed significant enrichment of nodes involving serine/threonine protein kinase D1 (PRKD1), AKT1, and ribosomal protein S6 kinase alpha-3 (RPS6KA3; Supplementary Fig. S6), linking the resistance phenotype of TP53^mut clones to the activation of the PI3K–AKT–mTOR pathway.

In line with the results from the phosphoproteome analysis, Western Blot analysis showed increased phosphorylation of p70S6K, a hallmark of mTOR activation in FaDu_CDDP-R compared with FaDu_CDDP-s cells. This significant difference in p-p70S6K expression persisted for >50 passages without CDDP treatment (Supplementary Fig. S7A) and was also observed in the single-cell derived clones (Supplementary Fig. S7B). In addition, the TP53 p.R248L genotype of FaDu_CDDP-R and the CDDP-resistant clones C5 and C78 was associated with strong p53 protein expression (Supplementary Fig. S7 B). In order to confirm a causal relationship between mutant p53 overexpression and CDDP resistance, we next tested whether siRNA-based knockdown of mutant p53 would sensitize resistant cells to CDDP treatment. Successful downregulation of p53 protein expression lasting for at least 72 hours after transfection was achieved by treatment with p53-targeting siRNA (Fig. 3A). In support of our hypothesis, resistant cells were significantly resensitized to CDDP by knockdown of mutant p53 (Fig. 3B–D). In contrast, transfection with NT-siRNA did not interfere with CDDP sensitivity of the resistant clones 5 and 78 (Fig. 3C and D) and had only a minor effect in FaDu_CDDP-R cells (Fig. 3B).

Because we had established the molecular basis of CDDP resistance in one single SCCHN cell line model, we aimed at validating the clinical relevance of our molecular findings. Recently, we performed a pilot-targeted NGS study in which the mutational profiles of primary, and recurrent tumors were compared in a small cohort of patients with SCCHN with early relapse after CDDP-based CRTX (N = 8; T. Eder and colleagues, 2017, manuscript in preparation). This dataset was used to address the question whether intratumoral heterogeneity and clonal selection also occur in patients. In four of the eight patients (50%), a TP53 mutant variant was detected in the tumor specimen collected at first diagnosis. Three of the four TP53 mutant cases (75%) showed the mutation at subclonal level (Table 2). In support of a role of GOF TP53 mutations in resistance to CDDP and radiation, we observed an increase in allelic frequency of GOF TP53 variants in the recurrent compared with primary tumor tissue. In contrast, no change or even a decrease in allelic frequency was observed for TP53 variants not previously associated with a gain of function (Table 2).

For further corroboration of a causal relationship between GOF TP53 variants and CDDP resistance, we then evaluated their impact on outcome in the TCGA SCCHN provisional dataset. HPV⁻ SCCHN cases with known TP53 status (N = 408, data download in March 2017) were stratified into two groups according to the TP53 mutation status. Group 1 harbored at least one of the following TP53 missense mutations (R175H, H193(L,P,R,Y), R248(Q,W), R249S, R273C, N = 48), which had been classified as high-risk mutation according to the Evolutionary Action Score of TP53 (12) and for which previous experimental proof for GOF has been provided (13, 14). Group 2 showed either any other type of TP53 alteration (N = 284) or wild-type (wt) TP53 (N = 76). Kaplan–Meier analysis revealed significantly reduced disease-free (log-rank P value: 0.009; HR for recurrence: 1.9, 95% confidence interval [CI], 1.2–3.0) and overall survival (log-rank P value: 0.024; HR for death: 1.6, 95% CI, 1.1–2.4) associated with TP53 GOF mutations (Fig. 4A and B).

For validation of the poor prognostic role of mTOR pathway activation, HPV⁻ SCCHN cases with known TP53 status and available protein expression data (N = 293, data download in March 2017) were included in Cox regression analysis. Normalized expression values of protein members of the mTOR pathway (Table 3) were used as continuous covariates for calculation of survival probabilities and hazard ratios. By this analysis, the regulatory-associated protein of mTOR (Raptor), which plays a pivotal role as scaffold protein in the mTOR signaling pathway, was identified as independent predictor of DFS and OS (Table 3, Fig. 4C and D). In contrast, no association with outcome was found for mTOR or the p70S6 and S6 kinases.

We here provide first evidence of intratumoral heterogeneity and CDDP-induced clonal selection as molecular mechanisms of drug resistance in SCCHN. Previous studies using either comparative genomic hybridization or massively parallel sequencing techniques have shown that multiple, genetically distinct clones can be detected within single tumor biopsies from patients with solid tumors such as breast (15) and renal cancer (16). Similar results have been observed in other solid tumors and leukemia, suggesting that heterogeneity is a common trait among cancers (17). In SCCHN, evidence for intratumor heterogeneity in a substantial proportion of tumors was provided by early karyotyping studies (18, 19). Comparable to the FaDu model, intrasample heterogeneity at the level of numerical chromosomal alterations was also observed in the majority of SCCHN cases in a study using specific FISH DNA probes binding to centromeric sites of chromosomes (20). However, the results from previous low-resolution genetic profiling studies remained often inconclusive and did not allow discriminating between multifocal tumorigenesis due to field cancerization in SCCHN and tumor evolution (21). Such discrimination became only possible by the introduction of NGS technologies in the assessment of the heterogeneity in genomic profiles of multiple spatially distinct tumor samples. The few available studies in SCCHN using such high-resolution approaches (22–24) confirmed the results on intratumoral heterogeneity of previous cytogenetic studies. Additionally, they showed that the extent of intratumoral heterogeneity differed considerably between individual patients, ranging from 100% down to only 30% of SNVs being shared among all tumor specimen from the same patient (22–24). Further studies including—similar to our approach—not only single-cell–based molecular profiling but also functional characterization of the genetically distinct subclones will certainly be required to understand the processes involved in tumor evolution and to clarify the clinical implications associated with clonal diversity within SCCHN tumors.

From previous sequential analyses of clonal evolution from primary to relapsed/metastatic disease, it became also evident that intratumoral heterogeneity constitutes a major factor of treatment resistance (25). In chronic lymphocytic leukemia, clones with TP53 mutations found at diagnosis exhibited faster time to relapse irrespective of clone frequency (26). Similar results have been reported in non–small cell lung cancer where clones harboring the EGFR T790M mutation—if detected at low frequency before start of tyrosine kinase inhibitor treatment—were associated with reduced progression-free survival (27). These observations are in line with a model in which inherently resistant clones expand after treatment to drive relapse growth.

We established the FaDu cell line as suitable model for studying CDDP-induced clonal evolution in SCCHN, by showing that this cell line is composed of multiple genetically different clones with varying CDDP sensitivity. Our preliminary results from pairwise comparison of primary/recurrent tumors indicate that intratumoral heterogeneity and clonal selection of therapy-resistant clones harboring GOF TP53 mutation can also occur in patients with SCCHN.

By comprehensive molecular and functional characterization of the parental cell line, a polyclonal subline and four single-cell derived clones, primary resistance to CDDP and radiation could be causally linked to a GOF TP53 mutation and increased activity of the PI3K–AKT–mTOR pathway. Accordingly, CDDP-based CRTX should represent an inefficient treatment for SCCHN cases with similar molecular traits. This hypothesis developed in the FaDu model could be confirmed in the TCGA SCCHN cohort of which a relevant portion of patients had received CDDP-based CRTX as adjuvant or definitive treatment.

Further support for the clinical relevance of our findings comes from the clinical observation that the risk of treatment failure after CRTX is especially high in the subgroup of patients with a previous history of excessive tobacco and alcohol consumption. Mutations in TP53 associated with the DNA-damaging effects of carcinogens from tobacco smoke (9) frequently occur in this patient population (3). TP53 mutations have not only been linked to the pathogenesis of SCCHN (28) but were also related to a treatment-resistant phenotype in this high-risk patient group. Indeed, the TP53 mutational status established from diagnostic tumor biopsies was identified as negative predictive biomarker for the efficacy of CDDP-based neoadjuvant chemotherapy (12, 29), concurrent (30) and adjuvant CRTX (31, 32). These studies also showed that the association between TP53 mutations and outcome depended on the mutation type and that patient stratification according to any type of TP53 mutations had only poor discriminative power (31, 32). Accordingly, the use of in silico algorithms such as the Evolutionary Action Score of TP53 that consider evolutionary variations (33) for the prediction of the functional impact of TP53 mutations have significantly improved risk classification according to the TP53 status.

Because no alternative treatment strategies are currently available for TP53^mut SCCHN patients, the clarification of the molecular basis of primary resistance and the identification of novel potential therapeutic targets for this patient population is of paramount importance. Reduced uptake, increased efflux, enhanced inactivation of CDDP by intracellular scavengers like glutathione, elevated DNA repair or tolerance to residual DNA lesions, overexpression of antiapoptotic proteins as well as further off-target effects which render cells globally less sensitive to cell death signals can contribute to a CDDP resistant phenotype (for a recent review see Galluzzi and colleagues (34)). Despite numerous studies on the role of TP53 mutations in chemoresistance, the exact molecular mechanisms by which mutant p53 interferes with CDDP sensitivity remains controversial. The loss of transcriptional activity of wt p53, a dominant-negative function mode of mutant p53 as well as independent oncogenic functions of distinct mutant variants have all been causally linked to CDDP resistance of tumor cells. Various GOF properties have been described, including the activation of genes normally unaffected or repressed by wt p53 through altered protein-protein interactomes and/or altered regulation of gene expression (14, 35). The results from our analysis of p53 transcriptional targets showing distinct expression patterns in p53^R248L-expressing resistant clones compared with p53^null sensitive clones both in normal and stress conditions are in line with such altered regulation of gene transcription by GOF TP53 variants. Their previous functional characterization also revealed that GOF phenotypes not only differ between individual TP53 variants (12, 14) but also different tissues (36), potentially reflecting differences in the expression of GOF TP53 target genes. This complexity poses relevant challenges in attempts to understand the role of TP53 mutations for treatment response, highlighting the importance of our current multilayered omics-based study. Future studies are warranted to further explore the potential of the FaDu and other SCCHN models for dissecting the drugs most indicated in the setting of primary resistance linked to GOF TP53 mutations.

In the FaDu model, we detected deregulated activity of the PI3K–AKT–mTOR pathway in CDDP-resistant cells overexpressing mutant GOF p53. Consistent with its possible role in aberrant mTOR activation, GOF mutant but not wt p53 was shown to activate mTOR by direct binding and inhibition of adenosine monophosphate (AMP)-activated protein kinase (AMPK; ref. 37), an energy sensor protein kinase that plays a key role in cellular energy metabolism by modulating mTOR pathway activity. In the latter study, the authors observed an invasive phenotype of SCCHN cells as result of mutant GOF p53-driven mTOR activation but did not address drug sensitivity (37). Indirect support for a role of mTOR in the efficacy of current standard treatment regimens including CRTX was provided by our analysis of the TCGA cohort of HPV⁻ SCCHN cases, by which we could establish the prognostic value of Raptor, a member of the mTOR signaling complex. Intriguingly, previous results from our group and others suggested that cells harboring mutant GOF p53-driven mTOR pathway activation are highly sensitive to the treatment with mTOR inhibitors like everolimus (38) or temsirolimus (8). Recently, higher response rates and improved overall survival were reported for paclitaxel combined with the pan-PI3K inhibitor buparlisib compared with paclitaxel plus placebo in patients with recurrent/metastatic SCCHN who had progressed on or after one previous platinum-based chemotherapy regimen (39). In line with the results from our current study, the benefit of buparlisib was significantly more pronounced in the HPV⁻ patient group displaying TP53 alterations compared with the TP53 wt group (39). Thus, the use of PI3K/mTOR inhibitors in primary treatment of HPV⁻TP53^mut SCCHN might represent an interesting option in this poor-outcome patient subgroup. Indeed, high sensitivity of SCCHN to mTOR inhibition has been demonstrated in experimental models (40, 41) and recent encouraging clinical studies (42, 43).

Our observation of a correlation between the CDDP-sensitive phenotype and an interferon gamma response signature is specifically intriguing in the light of recent results from biomarker development in clinical trials of immune checkpoint inhibitors (ICI). An interferon gamma signature previously established to be predictive for the efficacy of PD-1 blockade in melanoma (44) showed also significant associations with best overall response and progression-free survival after anti-PD-1 treatment of recurrent/metastatic SCCHN (44). The potential overlap in the predictive value of an interferon gamma signature for CDDP-CRTX and ICI provides a strong rationale to further develop such composite immune-related gene expression biomarkers in future trials of CRTX-ICI combinations.

In summary, we report a study of comprehensive molecular omics-based characterization of a SCCHN cell line model displaying intratumoral heterogeneity and varying sensitivity to CDDP and irradiation at clonal level. We show that tumors may contain minor subclones with primary drug resistance at levels only detected by deep sequencing which could drive tumor recurrence. We corroborate previous findings that GOF mutant TP53 variants are causally linked to drug resistance. Furthermore, we have identified the PI3K/mTOR pathway differentially upregulated between resistant and sensitive cells providing candidates for treatment optimization in SCCHN.

No potential conflicts of interest were disclosed.

Conception and design: F. Niehr, F. Klauschen, I. Tinhofer

Development of methodology: F. Niehr, T. Pilz, R. Konschak, D. Treue

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): F. Niehr, T. Eder, T. Pilz, R. Konschak, D. Treue, F. Klauschen, I. Tinhofer

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): F. Niehr, T. Eder, D. Treue, F. Klauschen, M. Bockmayr, S. Türkmen, K. Jöhrens, I. Tinhofer

Writing, review, and/or revision of the manuscript: F. Niehr, T. Eder, D. Treue, F. Klauschen, M. Bockmayr, V. Budach, I. Tinhofer

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): F. Niehr, T. Pilz

Study supervision: I. Tinhofer

We thank the microarray unit of the DKFZ Genomics and Proteomics Core Facility for providing the Illumina Whole-Genome Expression Bead chips and related services.

This work has partially been supported by a grant of the German Cancer Aid (DKH108791 to I. Tinhofer) and intramural funding of the German Cancer Consortium (DKTK) partner site Berlin.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.