Identification and Utilization of Donor and Recipient Genetic Variants to Predict Survival After HCT: Are We Ready for Primetime?

Overall survival following hematopoietic cell transplantation (HCT) has improved over the past two decades through better patient selection and advances in HLA typing, supportive care, and infection prophylaxis. Nonetheless, mortality rates are still unsatisfactory and transplant-related mortality remains a major cause of death after unrelated allogeneic HCT. Since there are no known pre-HCT, non-HLA biologic predictors of survival following transplant, for over a decade, scientists have been investigating the role of non-HLA germline genetic variation in survival and treatment-related mortality after HCT. Variation in single nucleotide polymorphisms (SNPs) has the potential to impact chemotherapy, radiation, and immune responses, leading to different post-HCT survival outcomes. In this paper, we address the current knowledge of the contribution of genetic variation to survival following HCT and discuss study design and methodology for investigating HCT survival on a genomic scale.


Introduction
The most successful curative therapy for many malignant hematologic diseases is hematopoietic cell transplantation (HCT). However, the success of this treatment is limited by transplant-related mortality (TRM). Better patient selection and advances in HLA typing combined with supportive care and infection prophylaxis have improved survival over the past two decades [1, 2•, 3]. Nonetheless, TRM remains a major cause of death, with disease-related mortality (DRM) the other largest contributor [4]. Although several clinical variables, including disease status at transplant, stem cell source, graft source, and CMV status are associated with survival outcomes [3], there are currently no established genetic predictors of survival after transplantation outside of HLA matching. Variation in single nucleotide polymorphisms (SNPs) may lead to differential gene transcription, translation, and protein structure. These changes have the potential to modify immune responses or side effects of chemotherapy and/or radiation, and thus, survival outcomes in HCT patients [5,6]. Various candidate genes have been tested for association with survival outcomes. Variants have been selected in genes relating to immune response to infection and inflammatory reactions, with the goal of understanding the biological basis of TRM and DRM and ultimately developing a better understanding of individual risk conveyed by genomic loci outside HLA. In studies of allogeneic HCT outcomes, the individual and joint effects of recipient and donor genotypes have been tested with the intent of developing a clinically applicable donor selection strategy to improve transplant success [7][8][9]. Another approach has focused on selecting and evaluating SNPs in genes within drug metabolism/detoxification pathways. By demonstrating genetic associations with transplant outcomes after exposure to various chemotherapeutic agents or combinations, the potential to assign a conditioning regimen based on genotype becomes a possibility [10,11]. To date, these candidate gene approaches have not been conclusive, in part due to sample size limitations and marked heterogeneity in population, disease, HLA matching, donor, and graft source. The objective of this article is to understand where studies of germline genetic variation have taken the transplant field in the hunt for clinically valid and actionable genetic variation associated with survival after HCT and how using genomics, versus genetics, may lead to better outcomes for patients [12••, 13]. We focus only on the role of germline genetic variation on survival after transplant since other studies have eloquently reviewed SNP associations with outcomes such as susceptibility to leukemia and chemotherapy toxicity [14] and the incidence and severity of graft-versus-host disease (GvHD) [15••]. We address current knowledge about the contribution of germline SNPs to survival following allogeneic HCT, first reviewing the published evidence for a role of non-HLA genetics in survival following transplant irrespective of, and considering, chemotherapeutic exposures. We then consider where these studies are in the translational research continuum and discuss genomic study design and methodological considerations when measuring competing risk outcomes.
Associations of NOD2/CARD15 with Survival After HCT An excellent example of this is NOD2/CARD15, selected for study in transplant outcomes as SNPs in NOD2/CARD15 were found to be associated with Crohn's disease, and there are similarities in this chronic inflammatory disorder of the gastrointestinal tract and GvHD symptoms. Few genes have been studied so extensively in relation to HCT with so many conflicting results, reviewed in [58]. While the initial association with an increased risk of TRM [42] was confirmed by a few studies [40,44,48], multiple other studies showed no association with survival following HCT [38,45,46,59], with additional work hinting that perhaps results varied by use of T cell-depleted grafts [43]. The largest NOD2/CARD15 study to date, 567 donor-recipient pairs both HLA matched and mismatched with primary diagnoses including hematologic malignancies, non-hematologic malignancies, and nonmalignant diseases, found only a borderline association (p=.049) of a recipient SNP with increased TRM and conflicting results in the non-malignant patient groups [60].
Collectively, these SNP-survival association studies appear to be dependent on the combination of transplant regimen, donor cell source, disease, and HLA matching and even further compounded by the fact that SNP frequencies vary significantly by race and ethnicity [61,62]. The intrinsic complexity of these results is further obfuscated as estimates of effect size and p values were obtained from prohibitively small studies, in most cases with 100-200 patients. In addition, only limited variants in this region were studied for associations with TRM, DRM, or OS. Given the polymorphic nature of NOD2 and the varying linkage disequilibrium structure by race and ethnicity [58], a more thorough investigation in homogenous appropriately sized populations is needed.
This extensive body of work is an excellent illustration of the consequences of heterogeneity and small sample size on studies of the relationship of germline genetics with survival following HCT. Definitive results from large-scale and replicated genetic studies are an imperative first step to finding clinically valid, and ultimately clinically actionable, variants for incorporation in treatment planning [12••, 63].

Pharmacogenetic Associations with Survival after HCT
Busulfan (Bu) and cyclophosphamide (Cy), the most commonly used alkylators in high-dose conditioning regimens prior to HCT, are associated with inter-individual variation in both relapse and toxicity [64]. While the unpredictable metabolism of these agents can lead to unintentional overdosing, an alternative explanation for excessive toxicity and differential adverse effects among patients given a Bu/Cy conditioning regimen is genetics [65]. SNPs in the CYP450 and glutathione enzyme families have been shown to affect clinical outcomes, relapse, and drug-related toxicities after HCT. Specifically response to Cy/Bu has been hypothesized to be due to variation in Cy metabolism by CYP2B6 and CYP2C19 which activates Cy to 4hydroxy Cy [66,67] and metabolizing enzymes involved in Bu conjugation, namely glutathione S-transferase (GST) isoenzymes A1 (GSTA1) and M1 (GSTM1). However, as with the studies of survival-SNP associations, pharmacogenetic studies of survival after HCT are plagued with inconsistencies due to small sample sizes compounded by the heterogeneity in exposure (drug) and dose.
Association of the CYP Gene Family with Survival CYP2C19 encodes a well-characterized and highly polymorphic enzyme that metabolizes Cy (among other drugs). Individuals can be grouped into poor (PM), intermediate (IM), and extensive (EM) metabolizers by p r e s e n c e o r a b s e n c e o f a c t i v e e n z y m e s . P M (CYP2C19*2/*2, *2/*3, and *3/*3) metabolize drugs more slowly, show prolonged side effects [68,69], and have a higher rate of TRM, but no association was seen with OS [59]. Although this was one of the larger studies of TRM, the low frequency of PM (approximately 3 % of the study population) necessitated a design with considerably larger sample sizes for the appropriate statistical power to detect the true effect size. Subsequent examinations of CYP2C19 have again suffered from the tyranny of small numbers. Melanson et al. [70] redefined the CYP2C19, PM and IM classification previously used [59] and showed associations of CYP2C19*2/*2 with worse progression-free survival (PFS) and OS but not with non-relapse mortality. This could be due to myeloablative doses of conditioning, the sequence of conditioning regimen drugs, or interactions with other drugs which may diminish or alter the detectable genetic effects. Consistent genetic and phenotype definitions are imperative for replications.
CYP2B6 is also an important enzyme that helps determine the rate of Cy clearance; however, only suggestive associations with survival following HCT have been shown among individuals who are considered ultra-rapid metabolizers, but no impact was seen on overall survival [70]. As with the CYP2C19 extensive metabolizer group, the CYP2B6 variants defining ultra-rapid metabolizers were not common, yielding a small comparison group. These CYP studies, like the NOD2/CARD15 research, highlight the importance, and challenge, in designing appropriately sized discovery and replication studies.
Association of Glutathione S-transferase (GST) Superfamily with Disease-Free Survival Bu-containing conditioning regimens show even greater interpatient variability in efficacy and toxicity than Cy. Intravenous vs oral administration of Bu, as well as differences in GI absorption, can generate variability in pharmacokinetics, drug clearance and drug activity. Therefore, patients receiving myeloablative Bu prior to HCT have therapeutic drug monitoring as standard clinical practice [71]. A high Bu steady-state plasma concentration can be toxic, whereas low concentrations are associated with poor engraftment and higher relapse risk [72]. Consequently, clinical outcomes are improved by targeting plasma concentrations and thus variants in the predominant metabolizing enzymes involved in Bu conjugation, glutathione S-transferase (GST) isoenzymes A1 (GSTA1) and M1 (GSTM1), have been tested for association with survival after transplant [10, 17,54,[66][67][68][72][73][74][75]. Yee et al. found a SNP in the GSTM1-GSTM5 locus, rs3754446, associated with an almost twofold shorter diseasefree survival in two cohorts of acute myeloid leukemia (AML) patients treated with chemotherapy-based autologous HCT. Despite replicating in both cohorts (p=.001 and p=.028), the finding was not significant after correction for multiple testing [10]. The authors found similar relationships with rs4148405 in ABCC3, although while passing multiple testing correction in cohort 1 (p<10e−06), it did not replicate in cohort 2. Additional genetic association studies have provided some evidence that variants in the genes in the ATP-binding cassette (ABC) family are associated with outcome [75,76].
While other adverse associations with GSTM polymorphisms have been shown with transplant-related toxicities [74] and in other cancers, reviewed in [77], the GSTM1 associations seen by Yee et al. have not replicated [54,74]. Again this could be due to the population studied (Rocha et al. analyzed associations in HLA-identical sibling donor-recipient pairs, while Hahn et al. analyzed overall survival in groups of autologous and related and unrelated donor allogeneic patients), the small heterogeneous sample sizes, the differences in the exact GSTM variation studied, or even subtle population substructure.
It is important to note that while this pharmacogenetic research is inconclusive, these drug metabolism pathways are well established and thus merit further investigation in larger cohorts. This approach has been done previously with success in relation to genetic associations with GvHD following HCT [78].
In 2007, an article in Blood was published, highlighting the rationale for a genome-wide approach in studying hematologic etiology and disease outcome [79•], and scientists have eagerly moved in this direction [15••]; however, since this time, only candidate gene studies have been published (Table 1). Of these, only two have replicated significant findings in a second independent population and one result has demonstrated enough validity to pursue clinical application in a multicenter trial currently underway.  [90][91][92]: the A haplotype, with at most a single activating gene, KIR2DS4, and the B haplotype, with KIR2DS4 plus at least one of the following activating genes: KIR3DS1, KIR2DS2, KIRDS3, or KIRDS5. Individuals who are B/x (either B/B or heterozygous B/A) are described as having an activating KIR genotype. The first study to demonstrate an association of recipient outcome with KIR activating genes included 209 HLAmatched and 239 HLA-mismatched T-replete URD-HCT for AML [88]. Three-year overall survival was significantly higher after transplantation from a KIR B/x donor afforded the recipients better overall and relapse-free survival when compared with A/A donors. The protective effect of donor B/x genotype was replicated in a larger cohort of early, intermediate, and advanced AML undergoing T-replete URD-HCT (n=1086). However this advantage was not seen in acute lymphoblastic leukemia (ALL) patients (n=323).

HLA-KIR Interactions Associated with Survival After HCT
The KIR locus contains genes that are centromeric (Cen) and telomeric (Tel). Dividing the B/x haplotype into Cen and Tel segments showed that the B haplotype genes in the Cen region had a stronger effect in improving the overall survival after transplantation than those in the Tel region. Specifically, individuals who were Cen-B/B/Tel-X/X (KIR2DS2 and/or KIR2DL2, no KIR2DL3/X) versus Cen-A/A/Tel-X/X (KIR2DL3 only/X) showed an increase in overall and disease-free survival. Again, no KIR effect was seen for ALL patients [87]. While promising, the proportion of Cen-B/B patients in the study was 11 %. When the cohort was expanded to include almost 500 additional AML patients (N=1532), the association of leukemia-free survival (LFS) with donors having two or more B motifs was maintained. In addition to testing the B motif association, the authors sought to assess donor KIR-HLA recipient combinations with outcome [85]. Individuals who are Cen-B/B have activating KIR genes that encode inhibitory KIRs specific for the C1 and C2 epitopes of HLA-C. The authors examined the interaction between donor KIR B genes and recipient class I HLA KIR ligands and found transplants mismatched at HLA-C1 experienced an almost twofold reduction in LFS by the KIR B/x donor. Interestingly this survival advantage in C1/x recipients compared with C2/C2 recipients was similar irrespective of the donor KIR B status. Unlike the frequency of Cen-B/B in the population, approximately 85 % of the US population is HLA C1/x, thus making these findings very generalizable to the transplant population.
While these studies assessed KIR haplotypes, it has been shown that both OS and TRM are affected by individual activating KIRs in the Tel region, KIR3DS1 and KIR2DS1 [89••, 93], and that like the motifs, these individual KIR may interact with recipient HLA. Venstrom et al. [89••] found that donor KIR2DS1 positive with HLA-C1/x conferred increased survival benefits (similar in effect size seen in Cen-B/B with HLA C/x) while KIR3DS1 impacted survival but donor HLA-C match or mismatch had no enhanced effect. Thus, it appears that both independent and joint donor and recipient genetics impact survival, making it particularly important to better understand this complex interplay in large comprehensive studies.

Selecting a Favorable KIR Donor in Unrelated HCT for AML
A prospective trial is currently accruing to test and validate the efficacy of choosing an URD for HCT based on KIR genotyping. Up to 600 AML patients will be enrolled to determine whether prospective selection of URDs based on a favorable KIR donor will reduce the cumulative incidence of relapse and improve LFS and overall survival. Preferred KIR donors are selected on the basis of a B content score. The KIR B-content score for each donor's KIR genotype is defined as the number of centromeric and telomeric genecontent motifs containing B haplotype-defining genes and is classed as BBest^, BBetter^or BNeutral^. This trial is not considering HLA-C/x donor variation. Once completed, this trial will provide the first prospective evidence on the utility of non-HLA genotypes to improve survival after URD-HCT for AML.

Are We Ready for Primetime?
To understand where these genetic studies have led the transplant field, we consider this research in a useful framework for characterizing the spectrum of translational research [12••]. It is important to not think of these phases as linear but rather as on a continuum, accessible at any point on the path to identifying, quantifying, and characterizing the relationship of genomics with survival after HCT. Four phases (T1-T4) comprise the framework, with T0 representing the discovery of new variant associations and other biomarkers of outcome following HCT. In the T1 phase, results from T0 are moved towards the implementation of interventions or diagnostic tests, e.g., evaluation of the function of genomic variants and analyses of gene-exposure interactions. T2 research is the clinical intervention used to determine if the proposed application of T1 findings is better than the standard of care. The KIR genomics studies began as T0 and moved into T1, and the trial is at present in T2. T3 assesses how to implement and integrate T2, e.g., should KIR-HLA matching improve overall survival, we must determine how this can be implemented into clinical practice at transplant centers nationwide. Lastly, the population health impact of implementation is evaluated in T4, e.g., what the overall reduction is in death attributable to the introduction of KIR matching in HCT.
Most of genomic variation research related to HCT survival is at the T0 and maybe T1 level; however, it is important to realize that while these are the first steps to translating results to routine clinical practice, it is also imperative researchers are simultaneously working with the fields of health economics, comparative effectiveness, disparities, and bioethics in an effort to understand and reduce barriers that will prevent the successful application of non-HLA genomics to transplant [63]. One example of opportunities in this arena rests with the scientists working in the field of statistical genetics. High-resolution typing is expensive and time consuming. To combat this time and cost issue, for the last decade, researchers have focused on the development of statistical methods to impute HLA regions with a high degree of accuracy [94]. In the European-American population for classical HLA class I and class II genes, there is approximately 97 % accuracy [95], and these methods are rapidly expanding to include other races and ethnic groups [96,97]. There are a number of translational research opportunities deriving from this T1 phase work, and with the accessibility of sequencing increasing, this work may move quickly through T2, as precision can be shown with studies in large sample sizes, and on to T3 and T4, implementation, e.g., where we must consider the how, where, and how much. We should not wait to think about these problems of implementation and assessment but rather understand the contribution and value of this particular research as it is moving towards clinical validation.

Genome-Wide Association Study of Survival After HCT
While heterogeneity and small sample sizes have plagued many of these studies, there is clear translational potential and further exploration of the contribution of non-HLA genetic variation in the recipient and/or donor to the risk of mortality after HCT is warranted. To examine this hypothesis in a well-sized and welldefined study population, we have begun a genome-wide association study (GWAS) to investigate the joint and independent genetic factors in recipients and donors contributing to death after URD-HCT, called Determining the Influence of Susceptibility-COnveying Variants Related to 1-Year mortality after unrelated donor Blood and Marrow Transplant (DISCOVeRY-BMT). The two independent cohorts used for analyses include patients diagnosed with AML, ALL, or myelodysplastic syndrome (MDS), reported to the Center for International Blood and Marrow Transplant Research (CIBMTR) with a banked National Marrow Donor Program (NMDP) biorepository sample available for both The DISCOVeRY BMT study design includes analyses of common and rare genetic variation in two cohorts. Specifically, we are undertaking a GWAS to map the independent and joint effects of recipient and donor genetic variation associated with survival outcomes after HLA-matched unrelated donor BMT in thousands of donor-recipient pairs. The purpose of specific aims 2 and 3 are to determine if conditioning regimens modify associations between recipient and/or donor genetic variants and TRM in the same population as aim 1 the recipient and donor. Cohorts 1 and 2 include 2609 10/10 HLA-matched, T cell-replete, URD-HCT recipients treated from 2000 to 2008 and 923 8/8 HLA-matched, T cell-replete, URD-HCT recipients treated from 2000 to 2011, respectively. Of the 2609 patients, 1116 (43 %) died before 1-year post-transplant, and of the 923 patients, 368 (40 %) died before 1-year post-HCT.

Competing Risk Models in GWAS
To analyze TRM, we must consider that TRM and DRM are competing risks for the outcome of death [98•]. Broadly, two approaches can be used to model competing risks: causespecific hazard functions, constructed using a Cox proportional hazard model, defining causes other than the one we are interested in as censored failures, or we can model the cumulative incidence functions of the different event types using a subdistribution hazard model which takes into account competing risks rather than merely censoring them [99]. The modeling of the cause-specific hazards is appropriate when the goal is to assess if a factor is associated with the risk of a specific cause of failure. However, when the goal is to compare the observed incidence of events from a given cause between groups, the subdistribution hazard should be used. Because the effect of a covariate on cause-specific hazard function for a particular cause can be different from its effect on the subdistribution function of the corresponding cause, the cause-specific hazard function and subdistribution can give different results [100,101]. These two analyses both provide important information, and it is recommended both methods are used when measuring associations with competing risk [102]. However, of the studies reviewed (Table 1), more often than not only cause-specific hazard was done and not both. This inconsistency in approach can be remedied going forward by leveraging both methods to better understand the contribution of genetic variation to survival after HCT.
To date, competing risk events have not been analyzed on a genome-wide scale and are often thought of as unique to the HCT field. However, as cancer treatments improve and more follow-up data are available, being able to analyze rare and common variation using competing risk analyses will be necessary in order to assess the role of genetic variation in competing clinical outcomes. At present, existing software scaled for millions of analyses is not available to test competing risk events; therefore, we are in the process of building custom R statistical packages to analyze both common and rare genetic variations in this setting.

DISCOVeRY-BMT Study Design and Power
The study design is illustrated in Table 2. We present some example power calculations for assessing the hazard of TRM overall as well as gene-drug interactions under a competing risk hazard model. Given that the minimum number of TRM subtype deaths (death due to GvHD, infection, and organ failure) and overall TRM is between 10 % and 40 % and assuming a minor allele frequency (MAF) of 0.40, we have excellent power to detect hazard ratios between 1.98 and 1.38, respectively (Fig. 1). Considering a reduced population comprising only patients given myeloablative cyclophosphamide (either TBI or Bu), approximately 70 % of all patients in the first cohort, and assuming 40 % of URD-HCT recipients experience TRM before 1-year post-transplant, we will have power to detect hazard ratios from 1.75 to 2.25 for MAF varying between 40 % to 10 %, respectively. Unfortunately, in contrast to the relative homogeneity of myeloablative conditioning regimens used in URD-HCT (BuCY or CyTBI), reduced intensity/nonmyeloablative conditioning regimens are heterogeneous in both drugs and doses therefore precluding a well-powered GWAS. DISCOVeRY-BMT is a large homogeneous population   Fig. 1. The x-axis, showing the proportion of events, can be used to determine power for a range of survival outcomes following HCT from 10 to 50 % in frequency. The dashed and solid lines reflect minor allele frequencies of 0.10 and 0.40, respectively. Thus, for example, given a survival outcome occurring in 25 % of DISCOVeRY-BMT cohort 1 and a minor allele frequency of 0.40, we have power to detect hazard ratios of approximately 1.5. A lower MAF of 0.10 yields power to detect hazard ratios of 2.0 that will be able to unambiguously identify clinically relevant hazards attributable to the joint and/or independent effects of genetic variation. Thus, it is the first step towards translating these data clinically and defining the biology of these devastating and poorly understood endpoints [12••].

Conclusion
The rapidly growing number of URD HCTs coupled with a high TRM yields an increasingly significant public health problem. Given TRM is the limiting factor to referring more potential patients and extending survival in existing patients, it is a clear target for translational research to increase the success and utilization of HCT as a curative therapy. So far, candidate gene studies have pursued discovery studies of genetic variation with the intent of making them clinically actionable but have had limited success in replication and thus are stuck in the T0 phase. When considering the KIR-HLA findings and the subsequent trial, we describe this as T2 research, with the understanding that the implementation (T3) and assessment of the impact (T4) will follow. Should this be successful, it will be the first demonstration that better matching using non-HLA alleles is not only possible but can also actively reduce death rates. DISCOVeRY-BMT is also designed to identify polymorphisms in recipients and/or unrelated donors with the potential for better recipient-donor pairing, the identification of patients at risk for TRM, or a high probability of DRM (T1 phase) with the intent of moving through the continuum of translational research. Importantly, the data generated from this project will be made publicly available to provide a very unique resource of a highly specialized, curative therapy where many additional hypotheses can be tested or confirmed.
Acknowledgments This study was supported by grants from the National Heart, Lung, and Blood Institute (HL102278 to LSC and TH) and the National Cancer Institute (CA139426 supporting AC). Polymorphisms in genes that regulate cyclosporine metabolism affect cyclosporine blood levels and clinical outcomes in patients