IMPACT OF POLYMORPHIC VARIATIONS OF GEMCITABINE METABOLISM, DNA DAMAGE REPAIR, AND DRUG RESISTANCE GENES ON THE EFFECT OF HIGH-DOSE CHEMOTHERAPY FOR RELAPSED OR REFRACTORY LYMPHOID MALIGNANCIES

Abstract

Purpose

The goal of this study was to determine whether single-nucleotide polymorphisms (SNPs) in genes involved in gemcitabine metabolism, DNA damage repair, multidrug resistance (MDR) and alkylator detoxification influence the clinical outcome of patients with refractory/relapsed lymphoid malignancies receiving high-dose gemcitabine/busulfan/melphalan (Gem/Bu/Mel) with autologous stem-cell support.

Experimental Design

We evaluated 21 germline SNPs of the gemcitabine metabolism genes CDA, dCK, and hCNT3, DNA damage repair genes RECQL, XRCC1, RAD54L, ATM, ATR, MLH1, MSH2, MSH3, TREX1, EXO1, and TP73, and multidrug resistance genes MRP2 and MRP5, as well as glutathione-S-transferase GSTP1 in 153 patients with relapsed or refractory lymphoma or myeloma receiving Gem/Bu/Mel. We studied the association of genotypes with overall survival (OS), progression-free survival (PFS) and nonhematological grade 3–4 toxicity.

Results

CDA C111T and TREX1 Ex14-460C>T genotypes had a significant effect on OS (P=0.007 and P=0.005, respectively), and CDA C111T, ATR C340T and EXO1 P757L genotypes were significant predictors for severe toxicity (P=0.037, P=0.024 and P=0.025, respectively) in multivariable models that adjusted for clinical variables. The multi-SNP risk score analysis identified the combined genotypes of TREX1 Ex14-460 TT and hCNT3 Ex5 +25A>G AA as a significant predictor for OS and the combination of MRP2 Ex10 +40 GG/GA and MLH1 IVS12-169 TT as significant predictor for PFS.

Conclusions

Polymorphic variants of certain genes involved in gemcitabine metabolism and DNA damage repair pathways may be potential biomarkers for clinical outcome in patients with refractory/relapsed lymphoid tumors receiving Gem/Bu/Mel.

INTRODUCTION

Gemcitabine is a pyrimidine nucleoside analogue with broad antitumor activity and wide clinical use. This prodrug requires cellular uptake and intracellular phosphorylation before incorporation into DNA, which is believed to be its mechanism of cytotoxicity.^{1, 2} The profile of gemcitabine, with dose-dependent cytotoxicity and few nonhematological side effects, has prompted its study at high doses with autologous stem-cell support.³ Since gemcitabine inhibits DNA damage repair,⁴ combinations of this agent with alkylators should have synergistic or additive antitumor activity and a favorable therapeutic index.

We have developed a new high-dose combination of infusional gemcitabine with busulfan and melphalan (Gem/Bu/Mel) for lymphoid tumors, with promising results in relapsed Hodgkin’s and diffuse large B-cell lymphoma (DLBCL).^5,6 The extramedullary toxicity profile of Gem/Bu/Mel includes mucositis, skin rash and transaminase elevation. Since these nonhematological side effects can be severe, it would be helpful to predict their occurrence for a given patient. Unfortunately, no patient or clinical features have been associated with toxicity.⁵

The cellular pharmacodynamic effect of gemcitabine depends on multiple enzymes, such as those involved in its intracellular metabolism, DNA damage repair and multidrug resistance mechanisms, whose activity may depend on their genetic polymorphic variants. We have previously identified single nucleotide polymorphisms (SNPs) of key enzymes in these pathways with a major impact on clinical outcome or toxicity on patients with pancreatic cancer undergoing gemcitabine-based chemoradiation^{7,8,9,10,11,12,13,14} In contrast, the pharmacogenomics of high-dose gemcitabine have not been adequately studied. Since the effect of gemcitabine on normal and tumor cells is greater at higher doses, it is conceivable that the impact of polymorphic genetic variation of relevant enzymes may be greater in the transplant setting.

The electrophilic alkylators busulfan and melphalan are detoxified inside the cell by reduced glutathione (GSH). GSH conjugation of alkylating agents is mediated by glutathione S-transferase (GST), whose activity also depends on polymorphic variations.^15,16,17 GST pi 1 (GSTP1) is the most abundant GST class found in many normal cell and malignant tissues.^18,19 The GSTP1 Ile105Val polymorphism has been associated with improved outcomes in patients with myeloma receiving high-dose melphalan.²⁰

We hypothesized that polymorphic variations of genes involved in gemcitabine metabolism, DNA damage repair, multidrug resistance and glutathione detoxification correlate with the toxicity and outcome of patients with relapsed/refractory lymphoid tumors receiving Gem/Bu/Mel.

Materials and Methods

Patient recruitment and data collection

This prospective study involved patients with relapsed/refractory lymphoid malignancies, including Hodgkin’s lymphoma or DLBCL and myeloma, with refractory or poor-risk features that made them eligible for clinical trials of Gem/Bu/Mel with autologous stem-cell transplantation at our institution.^5,6 This laboratory study was approved by the Institutional Review Board and all patients provided informed consent prior to enrollment. All patients received the same treatment doses and schema of Gem/Bu/Mel, as previously described.⁵ Overall survival (OS) and progression-free survival (PFS) were calculated from the date of diagnosis to date of death and progression/death, respectively. Living patients and patients without progression at the last follow up time were censored. Nonhematological toxicities, including mucositis, skin rash and transaminase elevation, were graded according to the Common Terminology Criteria for Adverse Events (CTCAE 3.0).²¹

DNA extraction and genotyping

We selected 21 SNPs of the deoxycytidine deaminase (CDA), deoxycytidine kinase (dCK), human concentrative nucleotide transporter (hCNT3), RECQL, X-ray repair complementing (XRCC)1, RAD54L, ATM, ATM and Rad3-related (ATR), mutL homolog (MLH)1, mutS homolog (MSH)2, MSH3, three prime repair exonuclease (TREX)1, exonuclease I (EXO1), tumor protein (TP)73, multidrug resistance-associated protein (MRP)2, MRP5, and GSTP1 genes according to the following criteria: 1) Minor allele frequency of the SNP >15% among Caucasians, 2) coding SNPs including nonsynonymous or synonymous SNPs, and 3) association with cancer risk or clinical outcome in previous studies. The genes, chromosome locations, nucleotide substitutions, function (such as encoding amino acid changes), reference SNP identification numbers, and minor allele frequencies of the 21 SNPs evaluated in this study are summarized in Table 1.

Table 1.

Single nucleotide polymorphisms (SNPs) evaluated

Gene	Chromosome	SNP	RS No.	Minor Allele Frequency Observeda	Minor Allele Frequency Reportedb
CDA	1p36.12b	Ex4 +111C>T, T145T	1048977	0.31	0.28
		Ex2 − 76A>C, K27Q	2072671	0.30	0.44
dCK	4q13.3b	IVS6 − 1205C>T	4694362	0.41	0.45
		IVS2 +9846A>G	12648166	0.43	0.43
hCNT3	9q21.32c	Ex14 − 69C>T, L461L	7853758	0.18	0.15
		Ex5 +25A>G, T89T	7867504	0.44	0.39
RECQL	12p12	Ex15 +159A>C	13035	0.40	0. 28
XRCC1	19q13.2	Ex6 − 22C>T, R194W	1799782	0.09	0.15
RAD54L	1p32	Ex18 +157C>T, A730A	1048771	0.10	0.15
ATM	11q22 - q23	IVS22 − 77 C>T	664677	0.43	0.29
		Ex38+61A>G, D1853N	1801516	0.07	0.16
ATR	3q22 - q24	Ex4 +340C>T, T211M	2227928	0.47	0.38
MLH1	3q21.3	IVS12 − 169C>T	2286940	0.37	0.41
MSH2	2p22 - p21	IVS12 − 6T>C	2303428	0.10	0.15
MSH3	5q11 - q12	Ex4 − 100G>A, P231P	1805355	0.14	0.15
TREX1	3p21	Ex14 − 460C>T	11797	0.38	0.43
EXO1	1q42 - q43	Ex15 +59C>T, P757L	9350	0.20	0.18
TP73	1p36.3	Ex2 +4G>A	2273953	0.23	0. 26
MRP2	10q24.2c	Ex10 +40G>A, V417I	2273697	0.19	0.25
MRP5	3q27.1b	Ex10 − 2A>G, Q382Q	7636910	0.36	0.35
GSTP1	11q13	Ex5 − 24A>G, I105V	1695	0.35	0.46

Abbreviation: SNP, single-nucleotide polymorphism; RS No., reference SNP identification number.

^aThe data observed in current study.

^bThe reported minor allele frequency was from SNP500 cancer database.

DNA was extracted from peripheral-blood lymphocytes in a single 10-cc blood sample of patients using Qiagen DNA isolation kits (Valencia, CA). Genotyping was performed using the Taqman 5′ nuclease assay. Primers and TaqMan MGB probes were provided by TaqMan SNP Genotyping Assay Services (Applied Biosystems, Foster City, Calif). The probes were labeled with the fluorescent dye VIC or FAM for each allele at the 5′ end. Polymerase chain reaction (PCR) was performed in a 5-μL total volume consisting of TaqMan Universal PCR Master Mix, 20 ng of genomic DNA (diluted with dH₂O), and TaqMan SNP Genotyping Assay Mix. Allele discrimination was accomplished by running endpoint detection using the ABI Prism 7900HT Sequence Detection System and SDS 2.3 software (Applied Biosystems). Twenty percent of the samples were analyzed in duplicate, with 100% concordance in genotype calling.

Statistical analysis

The distribution of genotypes was tested for Hardy-Weinberg equilibrium with the goodness-of-fit χ2 test. The association of clinical factors and genotypes with OS and PFS was evaluated using log-rank test and Kaplan-Meier methods. Hazard rations (HRs) and 95% confidence intervals (CIs) were estimated using univariable or multivariate Cox proportional hazard models. The association of genotypes and severe toxicity was estimated with odds ratios (OR) using univariable or multivariate logistic regression. Multivariate analyses of OS adjusted for age, number of prior chemotherapy lines, progression, and severe toxicity in this study. The effect of genotype on severe toxicity was adjusted for age, number of prior chemotherapy lines, and progression.

We estimated the false-positive report probability (FPRP) for the observed statistically significant associations using the Wacholder method.²² FPRP is the probability of no true association between a genetic variant and a phenotype given a statistically significant finding. It depends on the observed P value, on the prior probability that the association between the genetic variant and the phenotype is real, as well as on the statistical power of the test. In the current study, we set the HR and OR values of 2.0–4.0 as a likely threshold value. The prior probability used was 0.25 for all SNPs. The FPRP value for noteworthiness was set at 0.2.

To better assess combined genotype effects, we also conducted the risk score analysis as previously described.^23,24 Briefly, we selected SNPs with P ≤0.15 in likelihood ratio test derived from Cox or logistic regression models. We generated a risk score for these SNPs with deleterious genotype(s) as “1” and the reference genotype(s) as “0”. We explored all possible combinations of multiple SNPs to find the best-fitting models with consideration of the model likelihood and C statistic (Cox regression) or Area Under the Curve (AUC) (logistic regression).²⁵ In parallel, each selected SNP went through stepwise selection (α = 0.05) in 1,000 bootstraps. The SNPs surviving >50% bootstraps were most likely to be selected to the best-fitting model. A final multi-SNP risk score was developed based on the best-fitting SNPs through summation of risk scores. We compared the performance of the clinical factor based model, the SNP only based model and the clinical factor and SNP combined model using C-statistics or AUC. All statistical testing used SPSS Statistics, v17.0 (SPSS Inc., Chicago, IL) and R v.3.1.0 software packages. Statistical significance was defined as P<0.05.

RESULTS

Patient Characteristics and genotype frequencies

This prospective study included 153 patients with relapsed/refractory Hodgkin’s (N=54), DLBCL (N=64) and myeloma (N=35), enrolled in clinical trials of Gem/Bu/Mel from 08/2006 to 04/2011 and followed through 04/2014. Their median age was 47 years (range, 18–66 years). Their median number of pretransplant lines of treatment was 2 (range, 1–10). Their racial distribution was white (N=137), black (N=11), and Asian (N=5). Median follow-up time was 22 months (range, 1–42). The number of prior chemotherapy lines was significantly associated with PFS (P=0.006) and OS (P=0.002).

The 21 genotypes of interest were successfully amplified in 100% of the samples. No homozygous TT genotype was detected for RAD54L C157T SNP. The observed allele frequencies in this study population were comparable to previously reported allele frequencies in the general population (Table 1). Genotype frequencies of the 21 SNPs were found to be in Hardy-Weinberg equilibrium (χ²=0.006–3.304; P=0.069–0.936) except ATM A61G (χ²=15.118; P=0.0001) and MSH3 P231P (χ²=7.903; P=0.005).

Associations of genotypes with overall and progression-free survival

In the univariate analyses, 3 of the 21 SNPs evaluated (hCNT3 A25G, TREX1 Ex14-460C>T and MRP2 G40A) showed significant associations (P<0.05) with OS, and 7 SNPs (CDA C111T, dCK A9846G, hCNT3 C-69T, ATM C77T, ATM A61G, ATR C340T and GSTP1 Ex5-24A>G) showed nonsignificant associations with OS (Table 2). The association of CDA C111T and TREX1 Ex14-460C>T genotypes with OS remained statistically significant after adjusting for age, number of prior chemo lines, progression and severe toxicity (P=0.007 and P=0.005, respectively). The FPRP was 0.086 for CDA and 0.026 for TREX1, respectively, given a prior probability of 25%. Both are below the threshold of 0.20 indicating noteworthiness.

Table 2.

Characteristics of the study

Variable	No. of Patients	No. of Deaths (%)	P (log-rank)	No. of Progression (%)	P (log-rank)
Age, years			0.013		0.846
≦50	87	19 (21.8)		44 (50.6)
51–60	37	5 (13.5)		22 (59.5)
61–70	29	12 (41.4)		15 (51.7)
Diagnosis			0.102		0.744
Hodgkin’s lymphoma	54	7 (13.0)		27 (50.0)
Diffuse large B-cell lymphoma	64	18 (28.1)		32 (50.0)
Myeloma	35	11 (31.4)		22 (62.9)
Progression			<0.001		–
Yes (Progression)	81	36 (44.4)		81 (100)
No (No Progression)	72	0 (0)		72 (100)
Number of prior chemo lines			0.002		0.006
1	5	0 (0)		2 (40.0)
2	73	12 (16.4)		31 (42.5)
3	41	10 (24.4)		24 (58.5)
4	17	5 (29.4)		10 (58.8)
5	8	4 (50.0)		7 (87.5)
6	4	2 (50.0)		3 (75.0)
7	4	2 (50.0)		3 (75.0)
10	1	1 (100)		1 (100)
Grade 3–4 toxicity			0.591		0.633
Yes	47	12 (25.5)		58 (54.7)
No	106	24 (22.6)		23 (48.9)

To find the best multi-SNP predictor for OS and PFS, we conducted a risk score analysis. We first compared clinical factor-based, SNP-based and clinical factor and SNP combined model to find the best-fitting model (Table S1). SNP alone model had the highest predicting power compared to the other models. (Table S2). We found that the best multi-SNP risk score for OS is the TREX1 Ex14-460 TT genotype which had C-statistics as high as 0.91 in SNP alone model and 0.84 in combined model (Table S2). Patients carrying at-risk genotypes had HR=2.77 (95% CI, 1.26–6.11) (P=0.011) (Figure 1A). Furthermore, the best SNP combination of MRP2 Ex10 +40 GG/GA and MLH1 IVS12-169 TT remained a significant predictor of PFS after adjusting age, number of chemotherapy lines, and severe toxicity, with HR=2.06 (95% CI, 1.35–3.16) for patients with one or more variant alleles compared with those with no variant allele (P<0.001) (Figure 1B).

Figure 1.

Kaplan Meier plot of overall survival (panel A) and progression free survival (panel B) by risk score in the entire study population. The number 0 or 1 in panel A indicates the absence or presence of the at-risk TT genotype of TREX1 Ex14-460 (rs11797), respectively. The number 0–2 in panel B indicates the number of at-risk genotypes of MRP2 Ex10 +40 GG/GA and MLH1 IVS12-169 TT genotypes. The three lines for each genotype represent the survival curve and their 95% confidence intervals.

Subgroup analysis by disease

Risk score analyses in patients with HL and DLBCL showed a significant association of those 2-SNP genotypes with PFS within those patient subgroups [HR=2.08 (95% CI, 1.26–3.46), P=0.005], as well as a nonsignificant association with OS [HR=1.63 (95% CI, 0.88–3.03), P=0.1]. A borderline significant association of the risk score with PFS was also observed in patients with myeloma [HR=4.4 (95% CI, 0.92–21.15), P =0.064].

Association of genotypes with toxicity

The ATR C340T and EXO1 P757L genotypes showed significant associations with grade 3–4 nonhematologic toxicity, whereas CDA C111T, MRP2 G40A, and GSTP1 Ex5-24A>G had non-significant associations with severe toxicity (Table 4). The CDA C111T, ATR C340T and EXO1 P757L genotypes remained as significant predictors after adjusting for age, number of prior chemo lines, and progression (P=0.037, P=0.024 and P=0.025, respectively). The FPRP was 0.207 for CDA, 0.133 for ATR, and 0.137 for EXO1.

Table 4.

Extramedullary toxicity and genotype

Genotype		Grade 3–4 Toxicity, n (%)		P (χ2)	Univariate		Multivariate
		No	Yes		OR (95% CI)	P	OR (95% CI)*	P
Gemcitabine metabolic gene
CDA C111T (T145T)				0.139
CC		45 (61.6)	28 (38.4)		1.0		1.0
CT		49 (75.4)	16 (24.5)		0.53 (0.25–1.10)	0.086	0.49 (0.23–1.06)	0.070
TT		12 (80.0)	3 (20.0)		0.40 (0.10–1.55)	0.186	0.34 (0.80–1.40)	0.134
CC vs CT/TT					0.50 (0.25–1.01)	0.052	0.46 (0.22–0.95)	0.037 (0.043)
CDA A-76C (K27Q)				0.559
AA		51 (65.4)	27 (34.6)		1.0		1.0
AC		43 (72.9)	16 (27.1)		0.70 (0.34–1.47)	0.350	0.60 (0.28–1.31)	0.207
CC		12 (75.0)	4 (25.0)		0.63 (0.19–2.14)	0.459	0.56 (0.16–2.01)	0.380
AA vs AC/CC					0.69 (0.34–1.37)	0.288	0.60 (0.29–1.24)	0.169
hCWT3 A25G (T89T)				0.528
TT		38 (73.1)	14 (26.9)		1.0		1.0
CT		47 (70.1)	20 (29.9)		1.16 (0.52–2.59)	0.726	1.03 (0.45–2.40)	0.897
CC		21 (61.8)	13 (38.2)		1.68 (0.67–4.23)	0.271	2.04 (0.68–5.47)	0.155
TT/CT vs CC					1.55 (0.70–2.44)	0.283	2.01 (0.84–4.82)	0.119
DNA repair gene
RECQL A159C				0.398
AA		35 (66.0)	18 (34.0)		1.0		1.0
AC		57 (74.0)	20 (26.0)		0.33 (0.32–1.46)	0.326	0.66 (0.30–1.45)	0.296
CC		14 (60.9)	9 (39.1)		1.25 (0.45–3.44)	0.666	1.24 (0.41–3.69)	0.706
AA/CC vs AC					0.64 (0.32–1.27)	0.202	0.61 (0.30–1.26)	0.186
XRCC1194				0.306
CC		90 (69.8)	39 (30.2)		1.0		1.0
CT		15 (71.4)	6 (28.6)		0.92 (0.33–2.56)	0.878	0.97 (0.34–2.83)	0.960
TT		1 (33.3)	2 (66.7)		4.62 (0.41–52.4)	0.217	6.24 (0.42–92.6)	0.183
CC/CT vs TT					4.67 (0.41–52.8)	0.213	6.27 (0.42–92.5)	0.182
ATR C340T				0.087
CC		27 (60.0)	18 (40.0)		1.0
CT		50 (68.5)	23 (31.5)		0.69 (0.32–1.50)	0.348	0.60 (0.27–1.35)	0.219
TT		29 (82.9)	6(17.1)		0.31 (0.11–0.90)	0.031	0.28 (0.10–0.84)	0.024
CC/CT vs TT					0.39 (0.15–1.01)	0.053	0.39 (0.15–1.03)	0.057
EXO1 P757L				0.032
CC		63 (62.4)	38 (37.6)		1.0		1.0
CT		35 (81.4)	8(18.6)		0.38 (0.16–0.90)	0.028	0.38 (0.16–0.92)	0.032
TT	8 (88.9)		1 (11.1)		0.21 (0.03–1.72)	0.145	0.24 (0.03–2.10)	0.195
CC vs CT/TT					0.35 (0.15–0.79)	0.012	0.38 (0.16–0.88)	0.0258)
TP73 Ex2+4G>A				0.342
GG	65 (69.9)		28 (30.1)		1.0		1.0
GA	33 (64.7)		18 (35.3)		1.27 (0.61–2.62)	0.524	1.32 (0.63–2.77)	0.467
AA	8 (88.9)		1 (21.1)		0.29 (0.04–2.43)	0.254	0.27 (0.03–2.16)	0.206
GG/GA vs AA					0.27 (0.03–2.19)	0.219	0.22 (0.03–1.94)	0.174
Drug resistance gene
MRP2 G40A				0.083
GG	73 (73.7)		26 (26.3)		1.0		1.0
AG	30 (61.2)		19 (38.8)		1.78 (0.86–3.68)	0.121	1.72 (0.81–3.63)	0.158
AA	3 (60.0)		2 (40.0)		1.87 (0.30–11.8)	0.505	2.06 (0.28–15.3)	0.479
GG vs AG/AA					1.79 (0.88–3.62)	0.108	1.74 (0.84–3.63)	0.139
GSTP1 Ex5-24A>G				0.321
AA	46 (74.2)		16 (25.8)		1.0		1.0
AG	47 (63.5)		27 (36.5)		1.65 (0.79–3.46)	0.184	1.73 (0.80–3.73)	0.165
GG	13 (76.5)		4 (23.5)		0.89 (0.25–3.11)	0.848	0.67 (0.17–2.56)	0.554
AA/GG vs AG					1.70 (0.85–3.39)	0.136	1.87 (0.90–3.88)	0.092

Multi-SNP risk score analysis showed the clinical factor and SNP combined model moderately increased the power compared to clinical factor or SNP alone model (Table S2). The combined genotypes of EXO1 rs9350 (Ex15 +59C>T, P757L) and CDA rs1048977 (Ex4 +111C>T, T145T) had a HR of 2.47 (95% CI: 1.40–4.35, P = 0.0018) (Table 5).

Table 5.

Risk estimates and 95% confidence intervals for the best multi-SNP score models

Outcome	Regression	Model type	SNPs for Score	HR/OR (95% CI)	P
OS	Cox	1-SNP	TREX1 rs11797	2.77 (1.26–6.11)a	0.011
PFS	Cox	2-SNP	MRP2 rs2273697, MLH1 rs2286940	2.06 (1.35–3.16)b	0.0009
Toxicity	Logistic	2-SNP	EXO1 rs9350, CDA rs1048977	2.47 (1.4–4.35)c	0.0018

^aOS (overall survival) adjusted for progression free survival.

^bPFS (progression free survival) adjusted for age and number of chemotherapy lines.

^cToxicity adjusted for PFS and number of chemotherapy lines.

DISCUSSION

In this study, we evaluated the effect of polymorphic variants of genes involved in gemcitabine metabolism, DNA damage repair, drug resistance and glutathione detoxification on patient outcomes. The study was conducted in patients enrolled in clinical trials of our new HDC regimen Gem/Bu/Mel. Out of the 21 SNPs evaluated, CDA C111T and TREX1 Ex14-460C>T genotypes were independently associated with OS. The CDA C111T, ATR C340T and EXO1 P757L genotypes were independent predictors for severe toxicity. Furthermore, risk score analysis showed that the TREX1 Ex14-460 TT genotype and the combined genotypes of MRP2 Ex10 +40 GG/GA and MLH1 IVS12-169 TT were significant predictors for OS and PFS, respectively. These findings suggest that genetic variations in drug metabolism and DNA damage repair have value as prognostic biomarkers.

The selection of SNPs in this study was based on our previous observations in patients with pancreatic cancer receiving gemcitabine-based chemoradiation.^1–8 CDA, an enzyme involved in the salvage pathway of pyrimidine, is the major gemcitabine inactivation enzyme. Three potentially functional SNPs of the CDA gene, i.e. C111T (T145T), A-76C (K27Q), and G208A (A70T) have been clinically investigated in patients receiving gemcitabine.^{11,13, 26,27} The CDA C111T was significantly associated with severe toxicity in pancreatic cancer patients,⁵ as well as fewer tumor responses and worse outcomes in advanced non-small cell lung cancer.^{28, 29} Although CDA exon 4 C111T is a synonymous SNP that does not result in amino acid change, bioinformatics analysis predicted possible changes in splicing regulation and transcriptional regulation.³⁰ It is possible that the variant allele confers a reduced enzyme activity, making the variant allele carriers more susceptible to drug toxicity. Further functional studies are required to elucidate the mechanisms underlying the observed associations.

ATR, TREX1 and EXO1 are all DNA damage response genes that were significantly associated with either outcome or toxicity in the current study. Gemcitabine incorporation causes DNA replication arrest, and ATR/Chk1 signaling pathway plays a crucial role in the cellular response to the stalled DNA replication fork.³¹ Cells lacking ATR or Chek1 genes have been shown to be more sensitive to gemcitabine. The ATR gene encodes a protein kinase that is critically important in maintaining the integrity of the replication apparatus following damage that arrests the progression of the complex.³² ATR C340T (rs2227928) is a nonsynonymous SNP, and the replacement of threonine to methionine could have an impact on transcriptional regulation and post-translation consequence as predicted by bioinformatic models.³³ A lower level of expression or activity of ATR could explain the increased toxicity in patients with the variant allele observed in the current study. TREX1 is a major 3 prime exonuclease in mammalian cells. Loss of TREX1 leads to reduce the phosphorylation of the Chk1 gene in cells exposed to hydroxyurea,³⁴ which suggests a compromised ATR signaling pathway function. The TREX1 SNP (rs17971) investigated in the current study is an expression quantitative trait locus (eQTL).³⁴ As in or previous study, we saw a significant association of TREX1 Ex14-460C>T genotype with outcome. Thus, TREX1 is a critical determinant of efficacy of gemcitabine-induced DNA damage. EXO1 is a 5′′3′ exonuclease involved in the DNA mismatch repair and other DNA metabolic pathways affecting genomic stability, including homologous recombination and DNA damage repair.^35,36,37 EXO1 stability is dependent on ATR signaling.³⁸ The current study found a significant association of EXO1 P757L genotype with drug toxicity. The EXO1 P757L is a nonsynonymous SNP that result in replacement of amino acids, possibly affecting the protein functions.

In addition to the individual SNP effects, we have observed significant associations of the combined at-risk alleles of the TREX1, hCNT3 (involved in gemcitabine intracellular uptake), MRP2 (involved in exporting bilirubin and glucuronides of certain anticancer drugs) and MLH1 (DNA mismatch repair enzyme) genes with outcomes and toxicity. Although many of the at-risk alleles showed non-significant mild effect individually, the combined genotype had a strong effect on the clinical outcome, even within the disease subgroups. These observations support the concept that genes act in concert, and that the combined action of many genes exerts a greater influence on phenotype than individual SNPs. For future clinical applications, a battery of several genes/SNPs involved in the same pathway may have a better predicting power than relying on single gene/SNP.

Limitations to the present study include its moderate sample size and the heterogeneity of diagnoses. While the effect of the relevant SNPs was similar across patient diagnoses, our findings should be confirmed in disease-specific studies. Although our sample size is moderate and some observations might have occurred by chance, the consistency with previously reported associations, the functional basis of the observed associations, and the good performance of the risk scores argue for their potential importance.

In conclusion, we observed an important effect of polymorphic variants of genes involved in gemcitabine metabolism, DNA repair and multidrug resistance in a population of patients with lymphoid tumors receiving homogeneous HDC with Gem/Bu/Mel. The ultimate goal of this research is to identify genetic profiles that can be used in the clinic as predictors for therapy response or prognosis. If these findings are replicated in additional patient populations, such information may be helpful in stratifying patients for a more individualized high-dose cancer therapy.

Table 3.

Association of genotypes with overall survival

Genotype	No. of Cases	No. of Deaths	P (LR)	Univariate		Multivariate
				HR (95% CI)	P	HR (95% CI)*	P
Gemcitabine metabolic gene
CDA C111T (T145T)			0.210
CC	73	15		1.0		1.0
CT	65	19		1.61 (0.82–3.17)	0.168	2.35 (1.11–4.97)	0.025
TT	15	2		0.61 (0.14–2.69)	0.517	0.51 (0.11–2.31)	0.381
CC/TT vs CT				1.73 (0.90–3.33)	0.101	2.67 (1.31–5.43)	0.007
dCK A9846G			0.313
GG	51	13		1.0		1.0
AG	72	13		0.72 (0.33–1.55)	0.396	1.21 (0.53–2.77)	0.653
AA	30	10		1.34 (0.59–3.06)	0.487	1.38 (0.58–3.25)	0.464
GG/AA vs AG				0.64 (0.32–1.26)	0.195	1.05 (0.51–2.18)	0.889
hCNT3 C-69T (L461L)			0.332
CC	105	28		1.0		1.0
CT	42	7		0.58 (0.25–1.32)	0.190	0.63 (0.27–1.48)	0.288
TT	6	1		0.47 (0.06–3.48)	0.461	0.27 (0.04–2.03)	0.202
CC vs CT/TT				0.56 (0.23–1.23)	0.147	0.54 (0.24–1.19)	0.124
hCNT3 A25G (T89T)			0.081
TT	52	16		1.0		1.0
CT	67	10		0.42 (0.19–0.93)	0.033	0.53 (0.22–1.27)	0.153
CC	34	10		0.84 (0.38–1.85)	0.663	0.77 (0.34–1.78)	0.546
TT/CC vs CT				0.46 (0.22–0.94)	0.034	0.60 (0.28–1.30)	0.196
DNA repair gene
ATM C77T			0.224
TT	47	10		1.0		1.0
TC	81	17		1.00 (0.46–2.18)	0.996	1.10 (0.49–2.44)	0.819
CC	25	9		1.91 (0.78–4.71)	0.158	1.85 (0.71–4.81)	0.210
TT/TC vs CC				1.92 (0.90–4.08)	0.091	1.74 (0.78–3.88)	0.178
ATMA61G			0.376
GG	135	30		1.0		1.0
GA	14	5		1.93 (0.74–4.98)	0.177	1.72 (0.62–4.73)	0.297
AA	4	1		1.35 (0.18–9.89)	0.771	0.97 (0.12–7.67)	0.976
GG/AA vs GA				1.91 (0.74–4.93)	0.181	1.72 (0.62–4.73)	0.297
ATR C340T			0.326
CC	45	14		1.0		1.0
CT	73	16		0.67 (0.33–1.38)	0.281	0.96 (0.46–2.03)	0.922
TT	35	6		0.52 (0.20–1.35)	0.177	0.80 (0.30–2.14)	0.661
CC vs CT/TT				0.62 (0.32–1.22)	0.166	0.91 (0.46–1.81)	0.790
TREX1 Ex14-460C>T			0.078
CC	59	11		1.0		1.0
CT	73	17		1.29 (0.61–2.76)	0.506	0.87 (0.38–1.96)	0.732
TT	21	8		2.69 (1.08–6.69)	0.033	3.02 (1.17–7.81)	0.023
CC/CT vs TT				2.32 (1.06–5.09)	0.036	3.28 (1.44–7.51)	0.005
Drug resistance gene
MRP2 G40A			0.057
GG	99	23		1.0		1.0
AG	49	10		0.93 (0.44–1.95)	0.839	0.92 (0.42–2.03)	0.834
AA	5	3		3.71 (1.11–12.4)	0.034	1.80 (0.52–6.22)	0.353
GG/AG vs AA				3.80 (1.16–12.5)	0.028	1.85 (0.55–6.22)	0.320
GSTP1 Ex5–24A>G			0.216
AA	62	18		1.0		1.0
AG	74	13		0.54 (0.26–1.10)	0.090	0.65 (0.31–1.36)	0.253
GG	17	5		0.88 (0.33–2.37)	0.799	0.73 (0.25–2.14)	0.571
AA/GG vs AG				0.56 (0.28–1.10)	0.091	0.67 (0.34–1.31)	0.243

Abbreviations: P(LR), log-rank P; HR, hazard ratio; CI, confidence interval.

^*HR was from multivariate Cox regression model with adjustment for age, number of prior chemo lines, progression and severe toxicity.

HIGHLIGHTS.

• This prospective study shows that polymorphic variants of enzymes involved in gemcitabine metabolism, DNA damage repair and multidrug resistance pathways are associated with outcome and severe toxicity in patients with refractory/relapsed lymphoma or myeloma receiving high-dose chemotherapy with gemcitabine/busulfan/melphalan.

• These findings suggest the value of pharmacogenomics in stratifying patients for individualized high-dose cancer therapy.