Pharmacogenomic associations of cyclophosphamide pharmacokinetic candidate genes with event-free survival in intermediate-risk rhabdomyosarcoma: A report from the Children’s Oncology Group

Abstract

Background:

In vitro data suggest that the growth of rhabdomyosarcoma (RMS) cells is suppressed in a concentration-dependent manner by 4-hydroxycyclophosphamide (4HCY), the principal precursor to the cytotoxic metabolite of cyclophosphamide (CY). Various retrospective studies on the relationship between genes encoding proteins involved in the formation and elimination of 4HCY (i.e., 4HCY pharmacokinetics) and cyclophosphamide (CY) efficacy and toxicity have been conflicting.

Procedures:

We evaluated germline pharmacogenetics in 262 patients with newly diagnosed intermediate-risk RMS who participated in one prospective Children’s Oncology Group clinical trial, ARST0531. Patients were treated with either vincristine/actinomycin/cyclophosphamide (VAC) or VAC alternating with vincristine/irinotecan (VAC/VI). We analyzed the associations between event-free survival and 394 single-nucleotide polymorphisms (SNP) in 14 drug metabolizing enzymes or transporters involved in 4HCY pharmacokinetics.

Results:

Eight SNPs were associated (p-value < .05 by univariate analysis) with 3-year event-free survival; no SNPs survived a false discovery rate < 0.05.

Conclusions:

Our data suggest that a pharmacogenomic approach to therapy personalization of cyclophosphamide in intermediate-risk rhabdomyosarcoma is not viable. Other methods to personalize therapy should be explored.

1 |. INTRODUCTION

Rhabdomyosarcoma (RMS) is the most common soft tissue sarcoma in children and adolescents.¹ The intensity of RMS therapy is tailored to the likelihood of disease recurrence. The Children’s Oncology Group (COG) defines intermediate-risk RMS to include both embryonal RMS arising at an unfavorable primary site, unresected prior to systemic therapy, and without distant metastases and alveolar RMS without distant metastases regardless of the primary site or resection status.² For over 40 years, the North American standard therapy for intermediate-risk RMS has been the combination chemotherapy regimen of vincristine, dactinomycin, and cyclophosphamide (VAC), with radiation and/or surgery for local control.²

In vitro, the growth of RMS cells is suppressed in a concentration-dependent manner by 4-hydroxycyclophosphamide (4HCY), the principal precursor to the cytotoxic metabolite of cyclophosphamide (CY).³ Pharmacogenetic associations reported between clinical outcomes and genes expected to affect 4HCY pharmacokinetics have predominantly been conducted in adults, who often receive different CY-based combination chemotherapy regimens than children.⁴ The genotype–phenotype relationships may differ with CY dose and age, as we have previously reviewed.⁵ The contribution of the cytochrome P450 enzymes involved in 4HCY formation varies with CY concentration. The timing of when children attain the enzyme activity of adults is enzyme-specific, with hepatic CYP2B6, CYP2C9, CYP2C19, CYP3A4/5, and GST activity (normalized for bodyweight) being greater in children between the ages of 1 and 14 years.^6–12 We hypothesized that genetic polymorphisms in 4HCY-pharmacokinetic enzymes are associated with event-free survival (EFS) in children with newly diagnosed intermediate-risk RMS. As a correlative study to the most recent COG phase III trial for intermediate-risk RMS (ARST0531), we evaluated germline genetic factors associated with the efficacy of VAC. The polymorphisms of interest were those associated with 4HCY formation^13–21 or 4HCY elimination.^22–31 In choosing the candidate genes, there are no pharmacodynamic-based gene candidates specific to RMS sensitivity to chemotherapy.

2 |. METHODS

2.1 |. Patient population selection

A full description of the selection of the patient population and treatment arms for ARST0531 (ClinicalTrials.gov identifier: NCT01222715) has been previously published.³² Briefly, participants were diagnosed with intermediate-risk RMS and subsequently randomized to receive intravenous (IV) CY (1.2 g/m² each course) as part of the VAC regimen (cumulative CY dose, 16.8 g/m²) or the VAC regimen alternating with vincristine/irinotecan (VI) (cumulative CY dose, 8.4 g/m²) for 14 total courses of therapy over 42 weeks (Table S1).³² All participants and/or parents in this pharmacogenetic analysis consented as an optional component of the trial. Data were collected at the COG participating institutions and entered electronically within the COG electronic remote data entry system (eRDES).

2.2 |. Candidate gene selection

We analyzed the associations between EFS and the single-nucleotide polymorphisms (SNPs) in 14 drug metabolizing enzymes or transporters involved in CY metabolism (Figure S1). These genes of interest were cytochrome P450 (CYP) 2B6, CYP2C9, CYP2C19, CYP3A4, CYP3A5, myeloperoxidase (MPO), aldehyde dehydrogenase (ALDH) 1A1, ALDH3A1, glutathione S-transferase (GST) A1, GSTM1, GSTP1, GSTT1, ATP-binding cassette (ABC) ABCC2, and ABCC4 (Table S2).

2.3 |. DNA genotyping

Total genomic DNA was isolated from a 10-ml whole blood sample using the Gentra Purgene DNA purification kit according to the manufacturer’s instructions (Gentra Systems, Minneapolis, MN). DNA was genotyped using Illumina’s GoldenGate genotyping assay on the VeraCode Platform (Illumina, San Diego, CA) following manufacturer’s recommendation (VeraCode Assay Guide 11312819 rev A1).³³ 250 nanograms of genomic DNA was aliquoted into 96-well plates, processed accordingly, and scanned on the BeadXpress reader using GenomeStudio software (v2011.1). To ensure quality control (QC) of the data, 90 samples representing 30 parent–parent–child Centre d’Étude du Polymorphisme Humain (CEPH) trios (Utah residents with ancestry from northern and western Europe)³⁴ were genotyped to assess the performance of the Illumina oligonucleotide pool assay (OPA) before analysis of study samples. Assay accuracy was verified by comparing genotypes to publicly available genotype data for these samples from HapMap (http://www.hapmap.org/), 1000Genomes (http://www.1000genomes.org), dbSNP (http://www.ncbi.nlm.nih.gov/projects/SNP/), and by assessing inheritance errors. Two external control samples from the HapMap project were included on each plate to confirm the reliability and reproducibility of the genotyping across the study plates. Intra- and inter-plate duplicates comprised >10% of all samples. Laboratory personnel were blinded to all research information about the samples. Other QC procedures included the use of barcodes on samples and plates, dedicated materials and working space, and a visual review of SNP cluster plots by two laboratory staff members. Samples with weak signals, discordant duplicates, and outliers were repeated at least once.

Any SNPs with less than 90% concordance with public data for CEPH controls were considered failed and excluded from the analysis. Among successfully genotyped SNPs, sample and SNP call rates were >85%. All nonancestry SNPs were in Hardy–Weinberg equilibrium (p> .001). All SNPs with allele frequencies included in the analysis are given in Table S3.

2.4 |. Statistical analysis

EFS was defined as the time from study enrollment to disease progression, disease recurrence, second malignant neoplasm, or death as a result of any cause.³² There were 105 events. Cox proportional hazards models were used to estimate the hazard ratios and p-values for the associations of each SNP and EFS. All analyses were adjusted for prognostic variables for intermediate-risk RMS, including age, histology, clinical group, regional lymph node involvement, primary tumor size, and primary site).³² No adjustment was made for the treatment arm because the EFS was similar for both the VAC and VAC/VI arms on the study.³² Our analysis did not assume that particular genotypes would affect outcomes in any particular direction or by any joint effects; rather, we took an unbiased approach of testing for difference in EFS across the three genotypes (or two, in some cases) for each SNP. Variants with less than five observations were combined with the other less common variant for SNPs with three genotypes. SNPs with only two genotypes were dropped if the variant had less than five observations. For those SNPs nominally significant at p< .05, contrast was carried out to evaluate which variant had the greatest risk or protection. The Benjamini–Hochberg algorithm was used to control the false discovery rate (FDR < 0.05).³⁵ All analyses were conducted in Stata (v16.1, StataCorp, College Station, TX).

3 |. RESULTS

3.1 |. Patient characteristics

The demographic characteristics of the entire ARST0531 study cohort have been described previously.³² The demographics of the 262 participants for which DNA was available are shown in Table 1.

TABLE 1.

Clinical characteristics of participants (n= 262)

Clinical characteristics	VAC (n= 128)		VAC/VI (n= 134)
	Count	%	Count	%
Gender
Male	69	54	72	53
Female	59	46	62	46
Age, years
0–0.99	6	5	9	7
1–9.99	86	67	85	63
10+	36	28	40	30
Self-reported race
Caucasian	91	71	103	77
African American	16	13	10	47
Asian	2	2	6	11
Other/unknown	19	15	15
Self-reported ethnicity
Hispanic or Latino	22	17	13	10
Non-Hispanic/Latino	102	80	116	87
Unknown	4	3	5	4
Histology
Embryonal/botryoid/spindle cell	67	52	80	60
Alveolar	565	444	50	37
Other			4	3
Clinical group
I	4	3	4	3
II	17	13	14	10
III	107	84	116	87
Stage
1	15	12	10	7
2	43	34	40	30
3	70	55	84	63
Maximum tumor size, cm
<5	61	48	60	45
5–9.99	50	39	59	44
10+	17	13	15	11
Regional lymph node status
N0	103	80	103	77
N1	23	18	30	22
Nx	2	2	1	1
Primary site
Orbit	3	2	2	1
Head or neck	9	7	8	6
Parameningeal	59	46	62	46
GU, bladder/prostate	15	12	22	16
GU, non-bladder/prostate	1	1	1	1
Extremity	16	13	16	12
Retroperitoneal/perineal	15	12	17	13
Trunk	9	7	4	3

Note: Percentages may not total to 100 due to rounding off.

Abbreviations: GU, genitourinary; N0, no clinical or radiographic evidence of regional lymph node involvement; N1, clinical and/or radiographic evidence of regional lymph node involvement; Nx, regional lymph nodes not evaluated; RMS, NOS, rhabdomyosarcoma not otherwise specified; VAC, vincristine, dactinomycin, and cyclophosphamide; VI, vincristine and irinotecan.

3.2 |. SNP associations with event-free survival

We analyzed the associations between the EFS and the SNPs in genetic polymorphisms encoding for one of the following 14 drug metabolizing enzymes or transporters. These genes of interest were CYP2B6, CYP2C9, CYP2C19, CYP3A4, CYP3A5, MPO, ALDH1A1, ALDH3A1, GSTA1, GSTM1, GSTP1, GSTT1, ABCC2, and ABCC4. Of the 394 SNPs assayed, 363 had at least two variants for evaluation. Table S4 shows the univariate hazard ratios for the risk of EFS for all SNPs tested. Eight SNPs were nominally significant at p< .05 (Table 2). However, none survived the FDR < 0.05 threshold. Five of these were variants in the ABCC4 gene. Based on power analyses of the differences in risk between genotype groups, we were underpowered to detect a significant association in this prospective trial.

TABLE 2.

SNPs associated with EFS overall at a nominal p< .05 and within genotype

rs#	Gene	Genotypea	N b	HR	SE	p-Valuec	Linear gene dose from HR?
rs1729747	ABCC4	Overall	262	1.56	0.25	.007	Yes
		CC	183	Ref
		CG	70	1.76	0.37	.007
		GG	9	1.85	0.87	.19
rs931111	ABCC4	Overall	260	1.55	0.25	.008	Yes
		TT	182	Ref
		TC	69	1.75	0.37	.008
		CC	9	1.84	0.87	.008
rs1751057	ABCC4	Overall	243	2.15	0.63	.009	No
		AA	222	Ref
		AG/GG	19/2	2.15	0.63	.009
rs1629441	ABCC4	Overall	262	1.83	0.53	.02	Yes
		CC	237	Ref
		TC/TT	23/2	1.93	0.52	.02
rs2304925	Intron variant	Overall	23	11.94	12.31	.02	N/A
		TT	16	Ref
		TG/GG	5/2	11.94	12.31	.02
rs2065982	Ancestry information marker	Overall	262	0.61	0.14	.03	Yes
		TT	212	Ref
		TC	41	0.79	0.22	.4
		CC	9	0.15	0.15	.06
rs4148418	ANKRD40	Overall	259	0.75	0.1	.04	N/A
		AG	111	Ref
		AA	96	0.88	0.19	.55
		GG	52	0.52	0.16	.03
rs9590183	ABCC4	Overall	262	0.49	0.17	.04	N/A
		TT	223	Ref
		AT/AA	38/1	0.49	0.17	.04

Abbreviations: EFS, event-free survival; HR, hazard ratio; N/A, not applicable; Ref, reference group; SE, standard error.

^aSNP variants with less than five patients were combined with the other variants for the genotype.

^bSeveral assays with quality control samples were used, however, there were a small number of failures.

^cThere were no SNPs significant at FDR <0.05 (363 SNPs evaluated). A linear gene dose–response relationship, in which the heterozygotes and homozygotes for the variant allele’s change in the EFS are in a similar direction.

4 |. DISCUSSION

Among 262 patients with intermediate-risk RMS treated with VAC or VAC/VI, we did not find an association between EFS and germline SNP that encodes an enzyme involved in 4HCY formation^13–21 or 4HCY elimination.^22–31 Previous data from smaller studies have been conflicting. From a clinical intervention trial testing almost 400 gene variants, our analysis provides evidence that the germline pharmacogenomics of 4HCY pharmacokinetics are not informative enough to personalize treatment for patients with intermediate-risk RMS treated with CY-containing chemotherapy.

Since the 1970s, newly diagnosed RMS patients have received VAC (with or without additional agents) in conjunction with radiation therapy.^36–40 Dose escalation of vincristine or dactinomycin has not been investigated due to dose-limiting neurotoxicity and hepatotoxicity, respectively.⁴¹ CY dose intensification was investigated in a pilot study but abandoned due to excessive toxicity.⁴² VAC is an excellent treatment model for evaluating the pharmacogenomics of CY in children because it has clearly shown age-dependent toxicity with empiric CY dosing.⁴³ In an attempt to reduce CY toxicity, ARST0531 used a lower CY dose per course (1.2 g/m²) and cumulative CY dose (8.4 and 16.8 g/m²) than the two prior North American clinical trials (2.2 g/m² per course and 25.1–30.8 g/m² cumulatively).^40,44 We sought to identify germline polymorphisms associated with EFS.

CY is a pro-drug with a complex metabolic schema (Figure S1), with a wide range of parent drug area under the plasma concentration–time curve (AUC) and even greater variability in the AUCs of its metabolites, such as acrolein and phosphoramide mustard.⁴⁵ Of the pharmacogenetic studies in children receiving CY (references highlighted in blue in Table S2), the majority have focused upon GST polymorphisms.^46–50 Pharmacogenetic studies in adults receiving conventional-dose CY also included genetic polymorphisms of the CYP enzymes relevant to 4HCY formation (Table S2).^13–20 Many of these studies were limited by their small sample size and inconsistency in CY dosing, which could affect the predominant CYP enzyme involved in 4HCY formation^14,51–53 and thus, genotype-clinical outcome phenotype associations. The adult data cannot, however, be directly translated to children as children have faster CY clearance than adults.^45,54 Overall, children have impaired hepatic metabolism over the first 6–12 months of life, followed by a period up to adolescence when drug metabolism is more rapid relative to adults. The timing of when children attain the enzyme activity of adults is enzyme-specific, with hepatic CYP2B6, CYP2C9, CYP2C19, CYP3A4/5, and GST activity (normalized for bodyweight) being greater in children between the ages of 1 and 14 years.^6–12 Data regarding the ontogeny of the other metabolizing enzymes and transporters relevant to 4HCY elimination in vivo are not available. Thus, the predominant cytochrome P450 catalyzing 4HCY formation may differ between children and adults, potentially influencing the genotype–phenotype relationship.

Furthermore, the administered CY dose also influences the extent of 4HCY formation in children (Fig. 2b of Raccor et al.),⁵ which qualitatively agrees with adult data that the extent of bioactivation is greater with conventional-dose CY (0.5 g/m²) compared to myeloablative-dose CY (100 mg/kg).^55,56 Thus, there are potential differences in the quantitative contribution of CY-metabolizing enzymes toward 4HCY formation based on the CY dose. There is a shift in CYP contribution based on CY dose, with CYP2C9 and CYP2C19 being the major hepatic cytochrome P450 responsible for 4HCY formation at low concentrations and CYP3A4/5 and CYP2B6 assuming greater roles at high CY concentrations.^14,51–53 We did consider using archived tumor samples from the prior COG study (D9803) for this pharmacogenetic study,^57,58 however, this approach is suboptimal because this prior study (D9803) used a higher CY dose of 2.2 g/m², which would have a lower ratio of 4HCY/CY AUC.

There are limitations to these data. Given the relative rarity of RMS, patient numbers for this trial were limited and likely hampered our ability to detect meaningful differences in outcome between genotypic groups. We chose pharmacokinetic-based candidate genes, which have been associated with efficacy, toxicity, or pharmacokinetics (Table S2). Many of the studies in Table S2 are from small populations of heterogeneous CY-based combination chemotherapy regimens with incomplete candidate gene lists. We sought to overcome these limitations by systematically evaluating all of these candidate genes in a homogenous population of children. We evaluated many of the candidate genes in children receiving conventional-dose CY, as only the CYP3A4, CYP3A5, and GST polymorphisms have been evaluated in children with either immunologic conditions or acute lymphoblastic leukemia.^46–50 Another potential confounder is drug–drug interactions between CY and other chemotherapy or supportive care agents.^59–61 Our in vitro data suggest that 4HCY formation is not influenced by concomitant chemotherapeutics, including those concomitant with CY in the VAC regimen - vincristine and dactinomycin.⁵ Furthermore, the inclusion/exclusion criteria for ARTS0531 recommend avoiding strong inhibitors of CYP3A4, which is the anticipated mechanism of most drug–drug interactions.

Over the past few decades, there has been substantial enthusiasm for germline pharmacogenomics improving the treatment of patients with pediatric cancers.⁶² The benefits of mercaptopurine dosing based on thiopurine methyltransferase genotype are clear for children with acute lymphoblastic leukemia (ALL).⁶³ The treatment of ALL is similar to other pediatric cancers: multiagent chemotherapy given over an extended duration. Patients diagnosed with intermediate-risk RMS receive either three (VAC) or four (VAC/VI) anticancer agents. With this relatively homogenous chemotherapy regimen and the in vitro data showing that growth of RMS cells is suppressed in a concentration-dependent manner by 4HCY,³ we took a candidate gene approach based on genes regulating proteins involved in 4HCY formation and 4HCY elimination to evaluate their association with EFS. Our analysis suggests that there are no promising germline SNPs to improve the effectiveness of the VAC or VAC/VI regimen in children with intermediate-risk RMS. These negative genotype–phenotype associations indicate that germline pharmacogenetic testing is likely not a viable approach to the personalization of chemotherapy in patients with intermediate-risk RMS treated with VAC or VAC/VI. In contrast, the recent identification of somatic MYOD1 and TP53 mutations as biomarkers for inferior outcome in RMS⁶⁴ provides an opportunity to incorporate a molecular feature in risk stratification of RMS, with the long-range goal of improving EFS.