Osteosarcoma is the most common malignant bone tumor in children and young people. Efficacy of multiagent MAP (methotrexate, doxorubicin [Adriamycin], cisplatin) chemotherapy may be influenced by multiple cellular pathways. This pilot study aimed to investigate the association of 36 candidate genetic polymorphisms in MAP pathway genes with histological response, survival, and grade 3-4 chemotherapy toxicity in osteosarcoma.
Blood samples were obtained from 60 patients who had completed MAP chemotherapy. All patients were manually genotyped for 5 polymorphisms. The remaining 31 polymorphisms were genotyped in 50 patients using the Illumina 610-Quad microarray. Associations between candidate polymorphisms and histological response, progression-free survival, and toxicity were estimated using Pearson chi-square and Fisher exact tests, the Kaplan-Meier method, the log-rank test, and the Cox proportional hazards model.
Poor histological response was increased in variants of ABCC2 c.24C>T (P = .011) and GSTP1 c.313A>G p.Ile105Val (P = .009), whereas MTHFD1 c.1958G>A p.Arg653Gln was protective (P = .03). Methotrexate toxicity was increased in variants of MTHFR c.1298A>C p.Glu429Ala (P = .038), ABCB1 c.3435T>C Ile145Ile (P = .027), and ABCC2 c.3563T>A p.Val1188Glu (P = .028). Variants of GSTP1 c.313A>G p.Ile105Val were at increased risk of myelosuppression (P = .024) and cardiac damage (P = .008).
Osteosarcoma is the most common malignant bone tumor in children and young people. Multiagent chemotherapy has dramatically improved survival from <20% after surgery alone to 55% to 70% at 5 years.1 Methotrexate, doxorubicin (Adriamycin), and cisplatin form the backbone of most standard treatment protocols as a triple-drug regimen (MAP). However, approximately 40% patients show a poor response to chemotherapy, with inferior 5-year survival of 45% to 55%.2 Furthermore, 60% of bone tumor survivors experience severe or disabling chronic health conditions 30 years from diagnosis.3
Little is known about MAP chemotherapy pharmacogenomics, but information gained from the use of these agents in other malignancies provides insight into potentially critical pathway components. All 3 drugs are substrates of adenosine triphosphate-binding cassette (ABC) membrane efflux transporters. Functional polymorphisms in several of these proteins influence both chemotherapy toxicity and outcome in several malignancies but have not been investigated in osteosarcoma.4-12
Methotrexate, transported by the reduced folate carrier (RFC) protein, disrupts the folate cycle by direct inhibition of dihydrofolate reductase (DHFR), consequently depleting methylene tetrahydrofolate reductase (MTHFR) and thymidylate synthetase (TYMS). Adverse survival in childhood acute lymphoblastic leukemia (ALL) is associated with several single nucleotide polymorphisms (SNPs), including MTHFR c.677C>T p.Ala222Val, also known to enhance methotrexate toxicity in both ALL and solid tumors.2, 3, 6, 13-16 In osteosarcoma, MTHFR c.677C>T p.Ala222Val increased hematological toxicity but did not influence survival.17
Cisplatin and doxorubicin form helix-distorting DNA adducts resulting in strand breakage and inhibition of replication. Restoration of damaged DNA is by nucleotide excision repair (NER), whereas glutathione-S-transferase (GST) enzymes play a detoxification role. The influence of NER and GST polymorphisms are well described in solid and hematological malignancies.5, 9, 18-21 Increased lung relapse and improved overall survival in GSTM1 null and GSTT1 null patients, respectively, was recently reported in 80 osteosarcoma patients.22 A further small study reported an association between excision repair cross-complementation group 2 (ERCC2) c.2251A>C p.Lys751Gln, poor histological response, and inferior survival.23
We hypothesized that pharmacogenomic profiling of osteosarcoma patients may facilitate optimization of current regimens with the aim of improved outcome and a decreased burden of late effects. For this pilot study, 21 genes were selected from MAP pharmacological pathways (www.pharmgkb.org). From these genes, 35 candidate polymorphisms were identified on the basis of previously described associations or putative functional effects (Table 1). One cell-cycle protein polymorphism was also included, based on previously described associations. The primary aim was investigation of possible associations between genetic polymorphisms in drug target and metabolizing pathways of MAP chemotherapy and response to treatment in osteosarcoma; the secondary aim was investigation of polymorphic associations with treatment-related renal, cardiac, hematological, and ototoxicity.
Table 1. Clinical Effects of Candidate Polymorphisms
Abbreviations: A, doxorubicin; ABC, adenosine triphosphate-binding cassette; ACT, anthracycline cardiotoxicity; ALL, acute lymphoblastic leukemia; AML, acute myeloid leukemia; EFS, event-free survival; GST, glutathione-S-transferase; LD, linkage disequilibrium; M, methotrexate; MAP, methotrexate, doxorubicin, cisplatin; NHL, non-Hodgkin lymphoma; NSCLC, nonsmall cell lung cancer; OS, overall survival; P, cisplatin; SNP, single nucleotide polymorphism. *Now known as CD3EAP 1510>A Gln504Lys.
T ↑ relapse and death in ALL,15 M toxicity in ovarian cancer,16 hematological toxicity in osteosarcoma.17
T allele carriers associated with acute and chronic ACT.10
MATERIALS AND METHODS
Patients aged >16 years who had completed MAP chemotherapy for histologically proven osteosarcoma were eligible for enrollment. The study protocol was approved by the local research ethics committee, and patients provided written informed consent. Clinical data recorded at study entry included age at diagnosis, sex, ethnic group, tumor site, histology, presence of metastasis, surgical intervention (reconstruction or amputation), histological response (degree of tumor necrosis found histologically in the surgical resection specimen defined as good response [>90% necrosis] or poor response [≤90%]2), time to and site of relapse, and length of follow-up. Treatment-related data included number of administered chemotherapy cycles, delays and dose modifications, total cumulative doxorubicin and cisplatin exposure (mg/m2), adjuvant radiotherapy, and therapy at relapse.
The MAP regimen comprises 6 chemotherapy cycles, 2 preoperative and 4 postoperative. In week 1 of cycles 1 to 4, patients receive doxorubicin 75 mg/m2 and cisplatin 120 mg/m2, followed by methotrexate 12 g/m2 at weeks 4 and 5. In cycles 5 and 6, patients receive doxorubicin 75 mg/m2 alone at week 1 followed by methotrexate 12 g/m2 at weeks 3 and 4. Toxicity assessment was performed before each cycle, graded according to Common Terminology Criteria for Adverse Events version 3.0 (CTCAE).
CTCAE grades for infection, mucositis, and myelosuppression were recorded for each drug and cycle. Renal, cardiac, and auditory function was recorded at diagnosis, preoperatively, and at end of treatment measured by isotopic glomerular filtration rate (GFR; mL/min/1.73m2), cardiac ejection fraction (EF), and audiogram, respectively. Treatment-related cardiac toxicity, nephrotoxicity, and ototoxicity were defined as: early = decrease in EF, GFR, and audiogram by ≥1 CTCAE grade from diagnosis to cycle 2; and end of treatment = decrease in EF, GFR, and audiogram by ≥1 CTCAE grade from diagnosis to end of treatment.
DNA Extraction and Amplification
Two 5-mL ethylenediaminetetraacetic acid venous blood samples were obtained from all subjects at recruitment. Lymphocytic genomic DNA was extracted from 5 mL whole blood using the Qiagen FlexiGene kit (Qiagen, Crawley, UK). DNA yield, integrity, and protein contamination were checked by spectrophotometry. Manual genotyping was performed for thymidylate synthase variable numbers of tandem repeats, G/C SNPs, and DHFR insertions/deletions using standard polymerase chain reaction (PCR) and PCR-restriction fragment length polymorphism techniques as previously described.14, 24 Multiplex PCR was used to detect homozygous deletions of GSTT1 and GSTM1 with primers for the housekeeping gene BCL2 as an internal control.25 Fragments were analyzed by electrophoresis on 4% agarose gel stained with ethidium bromide. Final assignment of genotypes was confirmed by a clinician scientist blinded to patient outcomes.
SNP Microarray Analysis
The Illumina 610-Quad SNP array was used for microarray analysis of DNA in accordance with Illumina guidelines (Illumina, San Diego, Calif). After an initial quality check on 1% agarose gel, 300 ng of high-quality DNA was whole genome amplified, fragmented, precipitated, and resuspended. Samples were denatured and arrays loaded by a Tecan liquid handling robot (Tecan Group, Männedorf, Switzerland). After hybridization for 16 to 24 hours at 48°C, single base extension and labeling with Cy3 and Cy5 monoreactive dyes was performed. All reagents were supplied by Illumina. Genotypes were assigned using BeadStudio version 3 software (Illumina).
Candidate SNPs with minor allele frequency <0.05 or deviating from Hardy-Weinberg equilibrium P < .001 were excluded. Pearson chi-square or Fisher exact tests were used to analyze demographic, clinical, and genotypic data. In addition to wild-type, heterozygous, and variant homozygous forms, genotypes were grouped as dichotomous variables on the basis of their phenotypic consequences.
Highest recorded CTCAE grades were grouped as binary variables (grade 0-2; grade 3-4) for each drug cycle. Associations between genotypes and toxicity were calculated by logistic regression analysis and presented as odds ratios (ORs) and their 95% confidence intervals (CIs). Where small sample size resulted in the statistical phenomenon of separation, ORs and CIs were calculated by penalized log-likelihood regression.26 Multivariate logistic regression analysis of cardiotoxicity and nephrotoxicity included doxorubicin and cisplatin exposure (mg/m2) as covariates, respectively. Student t test was used to investigate differences between mean diagnostic and end of treatment EF and GFR with respect to genotype.
Progression-free survival (PFS) was calculated from the start of chemotherapy to first disease recurrence. Patients without disease recurrence at final analysis were censored at last follow-up. The Kaplan-Meier method was used to calculate survival probabilities, and log-rank test was used to compare differences in PFS. Overall survival was not analyzed, as because of small patient numbers it was not informative. Cox proportional hazards model was used for calculation of hazard ratios (HRs) in multivariate analysis, with predictors of survival identified in univariate analysis as covariates (tumor primary site, metastasis at diagnosis). All statistical analyses were performed using SPSS version 14.0 (SPSS Inc., Chicago, Ill) and Stata 10 (StataCorp, College Station, Tex). Reported P values were 2-sided, and a value of .05 was considered statistically significant. Adjustment for multiple testing was not performed, as P values and CIs were felt to be more informative in the setting of a small pilot study.
Sixty patients were enrolled; 2 patients not treated with MAP were excluded. Clinical and pathological characteristics of study patients are shown in Table 2. Microarray availability limited analysis to 50 patients, but demographic data for the microarray samples did not differ significantly from the entire cohort.
Table 2. Clinical and Pathological Characteristics of Study Patients
Total No. patients
Age at diagnosis, y
Metastasis at diagnosis
No significant associations were observed between histological response and sex, age group, ethnic group, primary site, histological subtype, metastasis at diagnosis, and relapse. Decreased PFS was significantly associated with axial tumors (HR, 10.0; 95% CI, 2.4-42.6; P = .002), metastasis at diagnosis (HR, 6.3; 95% CI, 6.3-21.7; P = .004), and amputation (HR, 5.5; 95% CI, 1.2-25.3; P = .028). Age, sex, histological subtype, and histological response did not influence survival.
Chemotherapy toxicity was recorded for 211 doxorubicin/cisplatin, 575 methotrexate, and 94 doxorubicin cycles. Three patients were excluded because of incomplete data; 2 patients treated with additional ifosfamide/etoposide were excluded from renal toxicity analysis. Ototoxicity analysis was not performed because of incomplete data. At least 1 episode of grade 3-4 toxicity was experienced by 46 (79%) patients after methotrexate, 54 (93%) after doxorubicin/cisplatin, and 40 (69%) after doxorubicin cycles.
Median diagnostic EF was 63% (interquartile range [IQR], 57-66), and doxorubicin exposure was 444 mg/m2 (IQR, 386-453). Early and end of treatment cardiotoxicity was recorded in 16 (29%) and 25 (45%) patients, respectively. A deterioration in cardiac function by ≥2 CTCAE grades from diagnosis to end of treatment was seen in 10 (18%) patients.
Median diagnostic GFR was 110 mL/min/1.73 m2 (IQR, 97.5-127), and cisplatin exposure was 474 mg/m2 (IQR, 443-480 mg/m2). Early and end of treatment nephrotoxicity was recorded in 18 (33%) and 33 (60%) patients, respectively. A deterioration in renal function by ≥2 CTCAE grades from diagnosis to end of treatment was seen in 5 (9%) patients.
Genotype frequencies were consistent with previously reported studies, with no deviation from Hardy-Weinberg equilibrium. Two SNPs with minor allele frequency <0.05 (NQO1 c.415C>T p.Arg139Trp and GSTP1 c.341C>T p.Ala114Val) were excluded from further analysis. Genotype showed no association with age, primary site, or histology. TYMS group 2 (2R/3G, 3G/3C, 3G/3G) was over-represented in males (P = .016) and wild-type MTHFR c.677C>T p.Ala222Val over-represented in non-Caucasians (P = .015). Metastasis at diagnosis was associated with CCND1 c.870A>G p.Pro241Pro wild-type homozygotes (OR, 9; 95% CI, 1.4-57.9; P = .021) and ERCC1 c.354T>C p.Asn118Asn variant homozygotes (TT; OR, 12; 95% CI, 1.3-112; P = .03).
Poor histological response was significantly increased in variants of ABCC2 c.24C>T (OR, 6.3; 95% CI, 1.4-28.5; P = .017) and GSTP1 c.313A>G p.Ile105Val heterozygotes (OR, 7.8; 95% CI, 1.6-37.5; P = .01), whereas MTHFD1 c.1958G>A p.Arg653Gln was protective (OR, 0.2; 95% CI, 0.05-0.8; P = .03). Data for significantly associated polymorphisms are presented in Table 3.
Table 3. Polymorphisms Significantly Associated With Histological Response
In univariate analysis, decreased PFS was seen in wild-type homozygotes of CCND1 c.870A>G p.Pro241Pro (P = .018), in variant carriers of GSTP1 c.313A>G p.Ile105Val (P = .025) and RFC c.80G>A p.Arg27His (log rank P = .02), and in the GSTT1 null allele (P = .0006) (Fig. 1). GSTT1 null allele suggested a trend in multivariate analysis (HR, 3.2; 95% CI, 0.9-11.6; P = .075). Multivariate logistic regression suggested a trend between relapse and RFC c.80G>A p.Arg27His (OR, 17.7; 95% CI, 0.8-417; P = .075). Full data are presented in Table 4.
Table 4. Genotypic Associations With Progression-Free Survival
Multivariate analysis of genotypic associations with cardiotoxicity and nephrotoxicity are shown in Table 6.
Table 6. Multivariate Analysis of Genotypic Associations With Cardiotoxicity and Nephrotoxicity
OR (95% CI)
EoT Functional Decrease, EF/GFR
P, t Test
Abbreviations: CI, confidence interval; EoT, end of treatment; GFR, glomerular filtration rate; OR, odds ratio.
GSTP1 c.313A>G p.Ile105Val
GSTP1 c.313A>G p.Ile105Val
ERCC2 c.2251A>C p. Lys751Gln
4 mL/min/1.73 m2
23 mL/min/1.73 m2
MTHFR c.677C>T p. Ala222 Val (sex additional covariate)
20.5 mL/min/1.73 m2
33 mL/min/1.73 m2
Understanding and overcoming the interindividual variation in drug response and toxicity remains among the major challenges in cancer chemotherapy. This pilot study explored the pharmacogenomics of osteosarcoma chemotherapy and although limited by small sample size secondary to retrospective recruitment and the rarity of this tumor, it remains the most comprehensive study to date. Herein, we observed several novel polymorphic associations as well as confirming several previously reported findings. Cautious interpretation is required, but we believe further investigation is warranted.
Several strengths have optimized the value of the dataset obtained. First, the selection of candidate polymorphisms was based on the established pharmacological pathways of each MAP drug, functional data, and previous associations reported in the literature. Second, detailed recording and analysis of chemotherapy toxicity according to drug adds rigor, as many studies combine toxicities of multidrug regimens despite actions on several different pharmacological pathways. Third, MAP is a prolonged and intense chemotherapy regimen with extensive patient monitoring, providing a comprehensive overview of individual toxicity. Finally, predictors of survival identified by univariate analysis were used as covariates in multivariate analysis of survival.
Multidrug chemotherapy has significantly improved survival in nonmetastatic osteosarcoma, but the proportion of patients cured has remained static for 2 or more decades. The identification of markers of subclinical metastatic or resistant disease may allow early treatment intensification, potentially salvaging some patients. Wild-type CCND1 c.870A>G p.Pro241Pro was over-represented in patients with metastatic disease at diagnosis, which is intriguing, as this genotype is also associated with poor survival in ALL.27 Cyclin D1 assumes a key role in cell cycle progression; the wild-type allele of c.870A>G p.Pro241Pro encodes a truncated CCND1 protein resistant to nuclear export, enhancing its oncogenicity. The importance of CCND1 in osteosarcoma has previously been reported, both in Notch-mediated tumor growth and by RNA interference inhibiting proliferation of tumor cell lines.28, 29 Thus c.870A>G p.Pro241Pro may impose far-reaching cellular consequences, suggesting that the association with metastatic disease deserves more extensive investigation.
The relationship between poor histological response and inferior survival is well established in osteosarcoma, with the concept of later treatment intensification to rescue these patients currently being explored in an international clinical trial.2, 30 Lacking at present is a means of identifying patients at risk of poor response at diagnosis. This study observed 2 novel polymorphic associations with poor response but did not replicate a reported association with ERCC2 c.2251A>C p.Lys751Gln.23
Poor response was associated with ABCC2 c.24C>T in a dose-dependent manner. Previously in 44 pediatric ALL patients treated with high-dose methotrexate, female carriers of the c.24C>T variant allele showed a 2-fold increase in plasma levels, suggesting reduced efflux.11 High plasma methotrexate levels confer a greater risk of toxic side effects, possibly decreasing subsequent chemotherapy dose intensity, an important factor in osteosarcoma response.31 There are further treatment implications, as cisplatin and doxorubicin are also effluxed by ABCC2.9, 12 However, in contrast to our findings, enhanced response to platinum-based chemotherapy was reported in Chinese patients with nonsmall cell lung cancer.9
Poor response was also associated with GSTP1 c.313A>G p.Ile105Val. GSTP1 is widely expressed in human epithelium, with overexpression contributing to both cisplatin and doxorubicin resistance in osteosarcoma.32GSTP1 has 2 common polymorphisms, c.313A>G p.Ile105Val and c. 341C>T p.Ala114Val, the 4 alleles showing significant differences in their abilities to protect against anticancer agents.33 There are no published data on the influence of GSTP1 c.313A>G p.Ile105Val in osteosarcoma disease response, although extensive research exists in other malignancies, much of it conflicting.5, 9, 18-20, 34-36
A trend toward osteosarcoma relapse in variant homozygotes of RFC c.80G>A p.Arg27His replicated a previous report in ALL, where higher plasma methotrexate levels suggested reduced RFC function.13 In osteosarcoma, compromised intracellular methotrexate transport, enzyme inhibition and drug polyglutamylation have potentially devastating consequences for disease control as the efficacy of high-dose methotrexate relies heavily on polyglutamylation. Furthermore, sensitivity to cisplatin, a pivotal drug in MAP chemotherapy, correlates with RFC expression in human tumor cell lines.37
MAP chemotherapy imposes significant early and long-term side effects. The early identification of patients at risk of severe toxicity may facilitate improved monitoring and early intervention to minimize morbidity. In addition, as the absence of chemotherapy toxicity may predict poor outcome in osteosarcoma, the identification of those at risk of chemoresistance is of equal importance.38 Several ABC polymorphisms influenced methotrexate toxicity, an observation not previously reported. In ALL, high methotrexate levels were associated with ABCC2 polymorphisms, although not c.3563T>A p.Val1188Glu as was found in our study. The gene however exhibits a high degree of linkage disequilibrium, perhaps explaining this association.11 The possible association of ABCG2 with toxicity may be explained by its function as the only folate transporter capable of effluxing polyglutamates, the latter critical for prolongation of methotrexate activity.
ERCC1 c.354T>C p.Asn118Asn and ERCC1 c.1510C>A p.Gly504Lys were associated with gastrointestinal and hematological toxicity, as might have been predicted from the cytotoxic effects of doxorubicin/cisplatin. The former showed no association in >1000 ovarian and colorectal cancer patients, but the latter was in concordance with previous reports of increased gastrointestinal toxicity in cisplatin-treated nonsmall cell lung cancer.5, 19, 39 One study in osteosarcoma reported no association of ERCC1 c.354T>C p.Asn118Asn with cisplatin-induced hearing loss but did not investigate other chemotherapy toxicities.23
Myelosuppression was increased in variants of GSTP1 c.313A>G p.Ile105Val, the same genotype implicated in poor histological response and PFS. Previously reported associations are inconsistent, with enhanced neurotoxicity observed in platinum-treated colorectal cancer but no association with toxicity in a large ovarian cancer patient cohort.5, 19 We speculate that increased myelosuppression with c.313A>G p.Ile105Val reduces dose intensity of MAP, increasing the likelihood of poor histological response. Intriguingly, variants of GSTP1 c.313A>G p.Ile105Val were also at increased risk of cardiotoxicity, a novel association with significant implications for patient management if validated. Highly oxidative metabolism and poor antioxidant defenses render cardiac myocytes extremely susceptible to anthracycline-induced oxidative stress; GSTP1 is pivotal in myocyte protection through its role in glutathione conjugation and peroxidase activity. Reduced function by GSTP1 c.313A>G p.Ile105Val may compromise the already reduced oxidative stress capacity of doxorubicin-exposed cardiac myocytes, facilitating cardiac damage. As a proven cardioprotectant is available, pretreatment identification of individual cardiotoxicity risk offers the potential to reduce the burden of late cardiac damage, particularly as high-dose doxorubicin is likely to remain a fundamental component of osteosarcoma therapy.
High-dose methotrexate and cisplatin may both cause acute renal damage and long-term insufficiency. Early nephrotoxicity was increased in variants of ERCC2 c.2251A>C p.Lys751Gln, and although a small decrease in GFR is of arguable clinical significance, the impact on early chemotherapy dose intensity is potentially important. Cisplatin nephrotoxicity is mediated through renal ischemia, a tumor necrosis factor-α inflammatory response, and generation of reactive oxygen species by apoptotic cellular signaling pathways.40ERCC2 removes oxidative DNA damage; thus, altered function by c.2251A>C p.Lys751Gln may expose renal tubules to greater oxidative stress, enhancing renal toxicity.
A trend toward early nephrotoxicity was seen with MTHFR c.677C>T p.Ala222Val, previously associated with acute renal failure after high-dose methotrexate.41 The variant allele has been associated with diabetic nephropathy but also correlates with hyperhomocysteinemia that is directly toxic to glomerular cells.42 It is possible that variant carriers of MTHFR c.677C>T p.Ala222Val are more susceptible to both methotrexate toxicity and direct renal damage.
This exploratory study has demonstrated several polymorphic associations with disease outcome and chemotherapy toxicity in osteosarcoma, but it must be re-emphasized that the study is retrospective and limited by small sample size. Nonetheless, some may have important implications for patient management and deserve validation in a larger, prospective cohort. It is hoped that prospective pharmacogenomic profiling will be undertaken in patients entered into future international clinical trials, the ultimate goal being the prediction of aggressive disease, poor clinical outcome, or severe toxic effects.
Supported by the Bone Cancer Research Trust (R.E.W.) and University College London Hospitals/University College London Comprehensive Biomedical Research Centre (S.J.S., J.S.W.).