Chemosensitivity Profiles Identify Polymorphisms in the p53 Network Genes 14-3-3τ and CD44 That Affect Sarcoma Incidence and Survival

The p53 regulatory network responds to cellular stresses by initiating processes such as cell cycle arrest and apoptosis. These responses inhibit cellular transformation and mediate the response to many forms of cancer therapies. Functional variants in the genes comprising this network could help identify individuals at greater risk for cancer and patients with poorer responses to therapies, but few such variants have been identified as yet. We use the NCI60 human tumor cell line anticancer drug screen in a scan of single nucleotide polymorphisms (SNP) in 142 p53 stress response genes and identify 7 SNPs that exhibit allelic differences in cellular responses to a large panel of cytotoxic chemotherapeutic agents. The greatest differences are observed for SNPs in 14-3-3τ (YWHAQ; rs6734469, P = 5.6 × 10⁻⁴⁷) and CD44 (rs187115, P = 8.1 × 10⁻²⁴). In soft-tissue sarcoma patients, we find that the alleles of these SNPs that associate with weaker growth responses to chemotherapeutics associate with poorer overall survival (up to 2.89 relative risk, P = 0.011) and an earlier age of diagnosis (up to 10.7 years earlier, P = 0.002). Our findings define genetic markers in 14-3-3τ and CD44 that might improve the treatment and prognosis of soft-tissue sarcomas. Cancer Res; 70(1); 172–80

The p53 protein plays a central role in eliciting cellular responses to a variety of stress signals and is crucial to an individual's ability to ward off cellular transformation and to respond to many forms of DNA damage–inducing cancer therapies (1, 2). A body of evidence is emerging in the literature that suggests that the inherited genetics of the p53 pathway could be used to further define patient populations in their abilities to respond to stress, suppress tumor formation, and induce p53 activity in response to DNA-damaging therapies (3, 4). The common inherited genetic variations in the p53 pathway most frequently studied are single nucleotide polymorphisms (SNP) in the p53 (p53 codon72; refs. 5, 6) and MDM2 (MDM2 SNP309; ref. 7) genes. The different alleles of p53 codon72 encode either a proline or arginine residue residing in a proline-rich region of p53 that has been shown to be important in mediating the apoptotic response (8). The MDM2 SNP309 locus results in either a thymine (T) or a guanine (G) in the intronic promoter/enhancer region of the MDM2 oncogene, which encodes a key negative regulator of p53 (7). In tumor-derived cell lines, the elevated levels of MDM2 associated with the G allele have been shown in multiple studies to result in the attenuation of the p53 apoptotic response after exposure to multiple types of chemotherapeutics compared with cells containing the T allele (7, 9, 10). These observations for both p53 codon72 and MDM2 SNP309 correlate well with studies of many types of cancers for which allelic differences have been noted for these loci in the onset of and risk for cancer, as well as in the response to therapy and survival (3, 4, 11).

The above-mentioned studies suggest that this important stress response pathway could harbor more functional inherited genetic variants, the study of which could help further define patient populations in their abilities to respond to stress, suppress tumor formation, and respond to DNA-damaging therapies. Recently, we proposed a methodology to uncover candidate functional SNPs using data derived from the NCI60 human tumor cell line anticancer drug screen (4). NCI60 is a panel of 60 cancer cell lines from different tissues of origin (12). Much is known of their genetics: the mutational status of more than 20 important cancer-related genes and the genotypes of 100,000 SNPs (13, 14). A great deal is also known of their response to therapies: both the growth response of these cell lines to more than 100 standard chemotherapeutic agents as well as to more than 40,000 publicly available potential anticancer agents (12). The growth response was measured 48 hours after treatment at a five-log concentration range for each compound. The negative log₁₀ (GI₅₀) of the concentration required to inhibit the growth of the given cell line by 50% relative to the vehicle-treated cells is calculated and reported (12). A previous study showed that the cells with mutations in the p53 gene are less sensitive to many standard chemotherapeutic agents that induce DNA damage, suggesting that the cellular response to these agents is p53 dependent (15). We developed an analytic framework that we implemented to mine the publicly available genotype, mutational, and cellular drug response data to identify SNPs in the p53 stress response that merited future study (4). In this report, we build upon, expand, and use this methodology and identify genetic variants in the CD44 and YWHAQ (14-3-3τ) genes that significantly affect both the cellular responses to many chemotherapeutics, as well as human cancer incidence and survival.

The mutational status of p53 and 20 important cancer-related genes, the genotypes of more than 100,000 SNPs (Affymetrix 125K chip), and the GI₅₀ data for the NCI60 cell panel were obtained from the National Cancer Institute (NCI)/NIH Developmental Therapeutics Program.⁸

The genomic DNA from the NCI60 panel of cell lines was a generous contribution from the NCI-Division of Cancer Treatment and Diagnosis Repository Molecular Characterization Program.

One hundred twenty-nine patients (73 females and 56 males; ages 14–87 y; mean 55.9 y) diagnosed with soft-tissue sarcomas (STS) in the years 1991 to 2001 at the Surgical Clinic 1, University of Leipzig, Germany, and at the Institute of Pathology of the Martin-Luther-University Halle, Germany, were included in the study (Supplementary Table S4). The mean observation time was 40.2 mo (range 2–198 mo). Sixty-three patients died from tumor-related causes within the observation time, whereas 66 patients were still alive at the time of follow-up. All patients underwent surgical treatment, and the subsequent mean survival time was 25.1 mo (range 2–119 mo). Eighty-three patients received postoperative radiotherapy and/or chemotherapy; 76 were treated with fractionated radiation (cumulative dose 60.4 Gy) and 26 received a combination treatment of doxorubicin and ifosfamide. The DNA of 66 patients was extracted from whole blood samples (n = 22) or from normal, pathologically confirmed tumor-free tissue, adjacent to the resection specimen (n = 44). The DNA of the remaining 63 patients was obtained from tissue within the confines of the tumor. A control cohort consisting of 498 blood donors (Germans of central European origin; 194 females and 304 males, ages 19–68 y; mean 44.0 y) from whom samples were obtained at the German Red Cross Blood Transfusion Service NSTOB (Springe, Germany) was included in the study. Each person gave written and informed consent. Approval from the local ethics committee was obtained.

Genomic DNA was extracted using the Innuprep Blood DNA mini-kit and the Innuprep DNA mini-kit (AJ Innuscreen GmbH). The CD44 SNP (rs187115) and the YWHAQ SNP (rs6734469) genotypes were determined by PCR amplification and subsequent allelic discrimination using the C_779820_10 and the C_29724290_10 genotyping assays. In brief, the allelic discrimination was performed as a multiplexed reaction in a 96-well format, using two primer/probe pairs in each reaction with a unique pair of fluorescent dye detectors to allow genotyping of the two possible variants at the SNP site in a target template sequence. Each well contained 20 ng DNA, 12.5 μL of TaqMan Universal PCR Master Mix (Applied Biosystems), 0.625 μL of the genotyping assay, and 9.875 μL distilled water. Following PCR, fluorescence was measured with the AB 7500 Real-Time PCR System (Applied Biosystems) and the genotype clusters were scored using Sequence Detection Software (version 1.3; Applied Biosystems). Both SNPs were genotyped successfully in all controls, but CD44 genotypes for one STS patient could not be determined due to irrecoverable DNA loss. Tumor tissue was available for 92 patients included in this study and the somatic p53 mutational status of these tumor tissues was determined by direct sequencing of exons 4 to 10 of the p53 gene as previously described (16).

The survival analysis was performed using the Cox multivariate proportional hazards regression model with the SPSS 16.0 software (SPSS, Inc.). A permutation test was performed to determine the statistical significance of the noted increase of the mean age of tumor diagnosis, whereby A/A < A/B < B/B. P values <0.05 were considered significant.

The genotypes for more than 100,000 SNPs are publicly available for the NCI60 cell line panel and were determined using the Affymetrix 125K genotyping platform.⁹

Two hundred sixty-four genotyped SNPs were found to reside in the 142 genes known to be important in the p53 stress response (Supplementary Table S1). We set out to explore possible allelic differences of SNPs residing in these genes in their growth responses to 132 standard chemotherapeutic agents. To do this, three statistical tests were performed (Supplementary Fig. S1A). First, a univariate test was undertaken for each drug and SNP pair and allelic differences were sought. Specifically, the average log GI₅₀ [X = −log₁₀(GI₅₀)] for cells for each of the three genotypes of a given locus (AA, Aa, and aa) were calculated for cells either wild-type or mutant for p53. Subsequently, the probability (P value) was computed that just by chance the difference for the following groupings either was equal to or larger than the actual measurement: (a) X_a-X_AA or (b) X_aa-X_A, or (c) X_AA-X_a, or (d) X_A-X_aa, or (e) [X_aa-X_aA and X_aA-X_AA], and (f) [X_AA-X_aA and X_aA-X_aa]. These probabilities were estimated using a permutation test (10⁶ permutations) that preserved the allele or genotype group sizes but permuted the samples among the groups. Results P < 0.05 were considered significant. Four different combinatorial hits were defined for each SNP-drug pair: 1, significant in all cells (SA); 2, significant in cells with a wild-type p53 gene (SWT); 3, significant in cells with mutant p53 (SMT); and 4, significant in cells with wild-type p53 but not significant or significant in the opposite allele orientation in cells with a mutant p53 (SWT-NRMT; Supplementary Fig. S1B).

Second, a multiple hypothesis test was performed for the noted observations. This test took advantage of the fact that 132 well-characterized compounds were tested against the NCI60 cell panel, which provided a set of independent measurements. The test incorporated all 264 p53 stress response SNPs and 132 agents and used a Fisher's exact test to compute the statistical significance of observing h univariate hits for a SNP (assuming one of the possibilities 1–4 listed above) on a total of D drugs, given that overall H significant hits are observed after testing S reference SNPs on the D drugs. We chose all 109,687 genotyped SNPs as a reference set.

The third statistical test served to help eliminate the possibility that a given allelic difference in the growth response to an agent could be due to the tumor type of the cells, other somatic mutations or other SNPs, or a combination of these variables. To do this, a two-step multivariate test was performed. Using a log-linear regression model, the contribution of each attribute [tissue of origin, somatic mutational status of 19 cancer-related genes (APC, BRAF, BRCA1, BRCA2, CDH1, CDKN2A, CTNNB1, EGFR, ERBB2, FLT3, HRAS, KRAS, NRAS, MADH4, PDGFRA, PIK3CA, PTEN, RB1, SKT11, VHL), and the SNP genotypes for the 264 p53 stress response SNPs] to the log GI₅₀ over the set of 132 drugs was computed. In the second step, identical computations were performed on a second set of 264 SNPs that were chosen randomly from the 109,687 genotyped SNPs. These calculations served as a reference set that allowed for the estimation of the statistical significance of the effects observed for a given SNP within the p53 stress response, specifically by defining the probability (P value) that a SNP in the reference set has an equal or larger contribution to the observed effect. The SNPs within p53 stress response with at least two genotypes demonstrating effects with P values lower than 5% in cells with wild-type p53 were deemed significant.

Only 8 of the 264 SNPs showed significant allelic differences in cellular growth responses to the standard chemotherapeutic agents according to the SWT-NRMT univariate analysis, the associated multiple hypothesis test, and the multivariate test in the cells wild-type for p53 (Supplementary Table S2). To complete and validate the genotyping of the eight candidate SNPs, we obtained the genomic DNAs from all 59 cell lines and regenotyped each SNP using accurate allelic discrimination assays (Applied Biosystems; Materials and Methods). The publicly available genotypes derived from the Affymetrix 125K genotyping platform had, on average, nine missing genotypes and, compared with the regenotyping, six missed calls (Supplementary Table S3). Therefore, we repeated the above-described genotype-response analysis using the new genotypes. All but one (rs2426127 in CSEIL) of the eight SNPs showed significant allelic differences in cellular growth responses to the standard chemotherapeutic agents in the cells that were wild-type for p53 (Table 1). Specifically, on average, the different genotypes of the seven significant SNPs associate with a different growth response for 79 (range 57–108; Table 1) of the 132 agents tested, whereby cells with the homozygote genotypes differed on average 4.9-fold (range 3.4- to 5.7-fold; Table 2) in their respective drug sensitivities in cells wild-type for p53.

The strongest effects in p53 wild-type cells were seen for the SNP in the YWHAQ (14-3-3τ) gene (rs6734469; Table 1). Specifically, the results of the univariate analysis suggest that cells with the different genotypes of the YWHAQ SNP significantly differ in their growth response to 108 of the 132 agents tested in wild-type p53 cells (P = 5.6 × 10⁻⁴⁷; Table 1), but only 3 of the 132 agents in mutant p53 cells. Significant allelic differences were observed with 35 alkylating agents, 20 and 15 topoisomerase I and II inhibitors, 14 and 15 RNA/DNA and DNA antimetabolites, and, to a lesser extent, with 10 antimitotic agents (Fig. 1A and B). Wild-type p53 cells with the homozygote genotypes for YWHAQ SNP differed on average 5.4-fold in their respective drug sensitivities (A/A-GI₅₀ > G/G-GI₅₀). For example, one of the topoisomerase II inhibitors for which the different genotypes of the YWHAQ SNP significantly differed was doxorubicin. Cells wild-type for p53 and A/A in genotype required 7-fold more doxorubicin to arrest growth compared with p53 wild-type cells G/G in genotype. Specifically, as seen in Fig. 1C, the 17 cell lines wild-type for p53 required, on average, 0.13 μmol/L of doxorubicin to arrest by 50%, and cells ranged from 0.01 to 0.64 μmol/L in their respective GI₅₀ values. Those five cell lines that required the highest concentration (average 0.33 μmol/L) were significantly enriched in cells containing the A allele compared with those five cell lines needing the least amount of agent (average 0.03 μmol/L, P = 0.0002). By contrast, no significant differences in the A-allele frequencies were seen in the extreme groups of the p53 mutant cell lines (Fig. 1C). Importantly, the allelic differences remain significant after performing a multivariate analysis, including the tissue of origin and the mutational status of the other cancer-related genes for those cells with a wild-type p53 gene (P_GG = 0.00004, P_AA = 0.0013; Table 1).

Interestingly, not all of the regenotyped SNPs retained the strong dependence of the wild-type p53 gene for the significant allelic differences in cellular growth responses to the standard chemotherapeutic agents (Table 1). For example, the allelic differences for the SNP in CD44, and to a lesser extent for the SNP in KDR, were seen with a similar number of agents and with similar strengths in both p53 wild-type and p53 mutant cells (Table 1). These data suggest that these allelic differences are not dependent on the mutational status of the p53 gene and that these two populations of cells can be grouped together in further analyses of these SNPs.

The allelic differences in cellular growth responses to the standard chemotherapeutic agents according to the SA univariate analysis, the associated multiple hypothesis test, and the multivariate test in all 59 cells regardless of the mutational status of the p53 gene are depicted in Table 2. Indeed, the strongest effects were seen for the SNP in the CD44 gene (rs187115). Specifically, the results of the univariate analysis suggest that cells with the different genotypes of the CD44 SNP significantly differ in their growth response to 96 of the 132 agents tested (P = 8.1 × 10⁻²⁴; Table 2 and Fig. 2A). Significant allelic differences were observed with 21 alkylating agents, 24 and 14 topoisomerase I and II inhibitors, 17 and 8 RNA/DNA and DNA antimetabolites, and with 20 antimitotic agents (Fig. 2A and B). Cells with the homozygote genotypes for CD44 SNP differed on average 2.8-fold in their respective drug sensitivities (C/C-GI₅₀ > T/T-GI₅₀). For example, one of the topoisomerase II inhibitors for which the different genotypes of the CD44 SNP significantly differed in their growth response was also doxorubicin. Cells C/C in genotype required 10-fold more of the doxorubicin to arrest growth compared with cells T/T in genotype. Specifically, as seen in Fig. 2C, those 10 cell lines that required the highest concentration (average 1.9 μmol/L) were significantly enriched in cells containing the C allele compared with those 10 cell lines needing the least amount of agent (average 0.3 μmol/L, P = 0.0134). Importantly, for all 59 cell lines, the allelic differences remain significant after performing a multivariate analysis, including the tissue of origin and the mutational status of the other cancer-related genes, and the 264 p53 stress response SNP genotypes (P_CC = 0.0000 and P_TT = 0.0003; Table 2).

The above-described allelic differences in cellular growth responses to chemotherapeutic agents for these SNPs suggest that individuals carrying the alleles that were associated with weaker growth responses to chemotherapeutics in cell culture could associate with poorer outcomes upon cancer onset due to a poorer response to therapies. To begin to test this, we analyzed these loci in 129 patients who were diagnosed with and treated for STS (Supplementary Table S4), a well-described, p53-surveilled tumor (17, 18).

Tumor tissue was available for 92 patients included in this study. The somatic p53 mutational status of these tumor tissues was determined by direct sequencing of exons 4 to 10 of the p53 gene, as previously described (16). Eighteen p53 mutations were identified from the 92 tumor DNAs, resulting in a p53 mutation frequency of 19.6% in this cohort, which is extremely similar to previously reported frequencies in STS (19.7%; International Agency for Research on Cancer database;¹⁰

ref. 19). The genotype frequencies in the 129 STS patients of the CD44 SNP were 11.7% for the C/C genotype, 45.3% for the T/C genotype, and 43.0% for patients T/T in genotype (Supplementary Table S5). The genotype distributions of the YWHAQ SNP were 27.1%, 47.3%, and 25.6% for the A/A, A/G, and G/G genotypes, respectively (Supplementary Table S5). The allelic frequencies of both SNPs did not differ significantly from the frequencies of the controls of the same ethnic origin nor did the frequencies differ between the patient's DNAs derived from nontumor or tumor tissues (Supplementary Table S5).

To assess the impact of both SNPs on patient outcomes, Cox multivariate regression analysis was performed, adjusting for known prognostic factors of STS: tumor stage, resection type (R-status), and gender (20, 21). Interestingly, the sarcoma patients harboring the alleles of the YWHAQ and CD44 SNPs, which in the NCI60 analysis were associated with weaker growth responses to chemotherapeutics, associated with poorer overall survival. Those patients homozygous for the C allele of the CD44 SNP were associated with a 2.16-fold increased risk for tumor-related death compared with individuals C/T and T/T in genotype (P = 0.041; Supplementary Table S7). Furthermore, patients carrying either one or two copies of the A allele of the YWHAQ SNP associated with a significantly worse prognosis compared with patients homozygous for the G allele (P = 0.043; relative risk, 1.92; Supplementary Table S7). As predicted by the results of the NCI60 analysis, the allelic differences in overall survival became notably stronger when the survival analysis was restricted to those 83 patients who received radiotherapy/chemotherapy. Patients C/C for the CD44 SNP associated with a 2.89-fold increased risk for tumor-related death compared with patients C/T and T/T for CD44 who also received DNA-damaging therapies (P = 0.011; Supplementary Table S7 and Fig. 3A). Patients harboring either one or two copies of the A allele of the YWHAQ SNP treated with DNA-damaging therapies associated with a 2.77-fold increased risk of tumor-related death compared with patients who are similarly treated but with a G/G genotype (P = 0.011; Supplementary Table S7 and Fig. 3B).

The above-described results support a model that these SNPs affect cellular stress responses, thereby resulting in altered survival rates. As cellular stress responses like those mediated by the p53 pathway are also important in tumor suppression, we explored the effects of these loci on the age-dependent incidence of these tumors. Interestingly, the alleles of both SNPs, which in the NCI60 analysis associated with weaker growth responses to chemotherapeutics and, in the survival analysis, associated with decreased survival rates, also associated with a significantly earlier age of tumor onset. Patients with a C/C genotype for the CD44 SNP were diagnosed on average at 49.1 years of age (range 16–83 years), patients C/T in genotype at 54.5 years (range 14–84 years), and patients T/T in genotype at 59.8 years of age (range 24–87 years; Table 3; P = 0.002). Thus, the C/C homozygotes showed on average a 10.7-year earlier tumor onset than individuals homozygous for the T allele (Table 3). Similar associations were observed for the YWHAQ SNP, whereby patients homozygous for the A allele were diagnosed with a mean age of 52.8 years (range 17–85 years), patients A/G in genotype at 55.5 years (range 14–85 years), and patients G/G in genotype at 59.9 years (range 22–87 years; Table 3; P = 0.006). Thus, patients with the A/A genotype showed on average a 7.1-year earlier age of onset than patients G/G in genotype (Table 3).

In this report, we build upon, expand, and use a methodology to uncover functional variants in p53 stress response genes with data from the NCI60 human tumor cell line anticancer drug screen. We scanned 142 genes known to affect the p53 stress response and identified seven SNPs in six genes that possess allelic differences in cellular growth responses to standard chemotherapeutic agents. The greatest differences in the p53 wild-type cells were observed for the YWHAQ SNP (rs6734469, P = 5.6 × 10⁻⁴⁷) and the strongest effects in all 59 cell lines, regardless of p53 mutational status, were seen for the CD44 SNP (rs187115, P = 8.1 × 10⁻²⁴). The CD44 gene encodes a transmembrane glycoprotein involved in a vast range of cellular processes, such as regulation of growth and survival, differentiation, and motility (22, 23). CD44 harbors tumor-promoting activities that include stimulating anchorage-independent cell growth and promote metastasis (24–26). Furthermore, its expression is characteristic for breast and prostate cancer stem cells (23, 27–29). Aberrated CD44 expression has also been reported to associate with tumor initiation and progression in various malignancies, such as colorectal, mammary, and prostate carcinomas, as well as neuroblastomas and sarcomas (25, 26, 30–32). Certainly, the precise effects of altered CD44 expression on the initiation of human cancer and patients' prognosis need to be further elucidated, but the reported studies and our data suggest a functionally and clinically important role of CD44 and its polymorphic variants in human malignancies.

The YWHAQ gene encodes 14-3-3τ, a member of the 14-3-3-protein family. These proteins interact with many signaling pathways that are critical for apoptosis and cell proliferation, mainly through binding phosphoserine/threonine motifs (33). 14-3-3 binding induces conformational changes in the target protein that can alter its stability, catalytic activity, cellular localization, or susceptibility to intracellular proteases, kinases, and phosphatases (33, 34). The 14-3-3 proteins are highly conserved and consist of seven family members in mammals: the β, γ, ϵ, σ, ζ, τ, and η isoforms (34). Interestingly, only YWHAQ (14-3-3τ) has been shown to promote apoptosis directly upon genotoxic stress (35–37). Furthermore, 14-3-3τ has also been shown to bind to wild-type p53 after ionizing radiation of cells and to stimulate transcriptional activity of p53 (35, 37). This observation is consistent with the observed dependence of wild-type p53 for the allelic differences in the cellular growth responses to chemotherapeutics for the YWHAQ SNP in our study. Importantly, the A allele, which associates with a significantly weaker drug response in the NCI60 panel, also associates with an early onset of disease and an increased risk of tumor-related death in STS patients, specifically in patients who received radiochemotherapeutic treatment (Fig. 3).

The observations reported here remain to be validated in other patient cohorts and, importantly, the regulatory changes associated with YWHAQ and CD44 SNPs remain to be determined. The identified YWHAQ SNP (rs6734469) is located in the second intron of the YWHAQ gene and the CD44 SNP (rs187115) in the first intron of CD44. Other SNPs in phase II of the HapMap project are closely linked (r² > 0.8)¹¹

to these two SNPs in the Caucasian population of northwestern European ancestry (CEU; Supplementary Table S6). Specifically, for the YWHAQ SNP, 10 SNPs in introns 2 and 3 are closely linked and, for the CD44 SNP, 3 SNPs in intron 1 are linked (Supplementary Table S6). To our knowledge, there have been no published attempts to characterize the regulatory regions of YWHAQ and no regulatory role of intron 1 of CD44 has been proposed. Hence, a discussion of whether one of these SNPs resides in a potential regulatory region would be premature. In addition, more exhaustive searches for other genetic variants closely linked to these SNPs, but not included in the HapMap project, will be necessary to develop a comprehensive list of candidate functional SNPs that warrant further experimental investigation into the molecular and cellular mechanisms underlying the significant allelic differences in cellular drug responses, sarcoma incidence, and survival reported in this work. However, these data strongly support the model that both CD44 and 14-3-3τ play a significant regulatory role in cellular stress responses, thereby affecting sarcoma incidence and survival.

No potential conflicts of interest were disclosed.

We thank Suzanne Christen, Claire Beveridge, and Mark Shipman for their help in preparation of the manuscript.

Grant Support: Ludwig Institute for Cancer Research, Simons Foundation, Helen and Martin Choolijian, and Leon Levy Foundation.

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.