Hematopoietic Stem Cell Transplantation (HSCT) was an important breakthrough in the therapy of hematologic malignancies with up to 9 HSCT per 100.000 inhabitants currently being performed in Europe each year (1, 2). More than a million HSCTs have been carried out since the first successful transplants in 1968. Transplant-related mortality (TRM) rates are decreasing from about 50% in the 1970s to as low as 10% today (2). Human leukocyte antigen (HLA) minor and major incompatibilities between HSC donors and recipients can trigger allorecognition by T-cells. Relapse and graft-versus-host disease (GvHD) are the two major reasons for transplant (Tx) failure (3). A five-year rate of relapse after HSCT can be as high as 45%-50% (2). Relapse can be treated e.g. by donor lymphocyte infusion (DLI) or novel effective pharmacological compounds, such as tyrosine kinase inhibitors (TKI), hypomethylating agents or monoclonal antibodies (4, 5).

Acute GvHD (aGvHD) results from transplanted cells attacking alloantigens on the recipient’s tissues. aGvHD affects up to 35%–45% of HLA-matched siblings and 60%–80% of partly mismatched unrelated donor transplant recipients (6). Host antigen presenting cells are activated in response to damage-associated and pathogen-associated molecular patterns (DAMPs and PAMPs) released in the course of pretransplant conditioning and induce proliferation and activation of alloreactive donor T-cells. Subsequently, these T-cells mediate local tissue injury together with soluble inflammatory agents (e.g. IFN-γ, TNF-α) (3, 6). In contrast, chronic GvHD (cGvHD) is characterized by the presence of alloreactive, dysregulatory T- and B-cells. Almost every second HSCT recipient develops cGvHD, either de novo, during progression or after aGvHD. Furthermore, cGvHD is a major cause of late non-relapse associated mortality following HSCT (7). Prophylaxis for GvHD was introduced as an effective preventive action. Unfortunately, the lower incidence of severe GvHD was offset by high rates of relapse in malignancy.

During the past decade, a series of genomic variants have been discussed to be associated with GvHD or other transplantation (Tx)-related outcomes (8, 9). A first genome-wide association study (GWAS) associated genotype mismatch at HLA-DPB1 (10) and rs17473423 (12p12.1, now listed as rs1137282, located in the Kirsten rat sarcoma virus (KRAS) gene encoding K-Ras, a part of the RAS/MAPK pathway, were associated with GvHD in a Japanese population (11). Combining GWAS on a Finnish and a Spanish cohort with gene expression studies revealed further 51 genes potentially associated with GvHD (12). However, despite plausible molecular biological explanations, replication often fails, often due to incompatibilities between patient and donor cohorts with regards to HSCT treatment regimens and patient and donor characteristics (9, 13–15).

One major challenge in searching for genetic associations with post-Tx events are the different possible clinically relevant endpoints, the competing behaviour of target events in addition to the inherent differences between transplant cohorts (16, 17). Competing post-Tx events are events which prevent the occurrence or modify the risk of each other. These include the manifestation of relapse, or GvHD and death, and need to be taken into account when assessing cause-specific hazards, cumulative incidence, event-free or overall survival (18, 19). Both previously reported GWAS used a case-control design, meaning they compare, i.e., in the simplest case, the minor allele frequency (MAF) between GvHD cases and GvHD-free controls. Such comparisons may be biased if, for example, the time since Tx of cases and controls differ, or patients are prone to both GvHD and relapse, but the considered SNP is only associated with relapse (hence in the presence of competing risks) (20).

Here, we aim to identify genomic markers, associated with any of the relevant, post-Tx key events, death, relapse and GvHD, and establish a joint GWAS using recipient DNA samples from three transplantation centres. Complex survival analysis was carried out taking competing risks into account, as previously recommended (16, 17, 19).

We performed GWAS on post-Tx course with sGvHD, relapse, or death as the key events. Because these events compete, it is not advisable to focus on any one of these events alone, or examine composite endpoints (like EFS) or strongly simplified endpoints (like OS). In order to understand or biologically explain an observed association of a genomic marker, one needs to know which endpoint it is associated with and which one it is not. This is not possible, if only EFS, OS or one simple event is considered. Instead, we fitted models for aetiological and prognostic research, and for all competing key events to obtain an overall view of each SNP. We did not prioritize individual SNPs, nor did we include previous findings or speculation about a SNP in our analysis.

In general, identifying genes associated with the post-Tx events remains challenging due to disease- and treatment-related confounding, competing key events, multiple tests, among others. Examination of rare markers (MAF <5%) in the post-Tx course of HSCT is difficult and prone to inexplicably low p-values.

However, we found interesting genome-wide and suggestive significant signals for several genes as described below:

The gene SEMA3C was found to be positively associated with EFS or sGvHD in the final analysis, with genome-wide significance. Many types of cancer cells express class 3 semaphorins, to inhibit tumour angiogenesis and growth. In contrast, expression of SEMA3C is potentiating tumour progression in several cancer types. Subsequently, SEMA3C displays pro- and anti-tumorgenetic characteristics (38–40). Due to its role in neurodevelopmental processes which are frequently dysregulated in cancer, inhibiting SEMA3C signalling is proposed as therapeutic strategy to improve cancer control (39). SEMA3C may contribute to drug-induced resistance of cancer cells (27, 41). Interestingly, semaphoring C has also been shown to be involved in angiogenesis, and angiogenesis has been reported as a pathogenic factor in early GvHD, and it will be of interest to see whether there is a direct interaction of semaphoring C and endothelial repair in the context of allogeneic SCT (42–46). Another semaphoring Sema4D has been linked with T cell function in the setting of GvHD, but the mechanisms may be different form our observation (47).

The gene PXN was the only one that appeared at least suggestively significant regarding NRGM at the screening step and in the refined analysis. Paxillin functions as a docking protein for binding signalling molecules to a specific cellular compartment. As a central protein of focal adhesion, it is a common target of many different oncoproteins and involved in numerous processes. PXN is considered a susceptibility gene for many types of tumours, including large B-cell lymphoma (48, 49). Upregulated PXN expression was linked to poor OS in acute myeloid leukaemia (50) and has been associated with prediction of relapse in chronic myeloid leukaemia (51). Therefore, PXN seems to play a role in the post-Tx course, but for which key event remains unclear.

We were able to detect three further genes that are associated with the occurrence of severe GvHD: Plasmacytoma Variant Translocation 1 (PVT1), WW Domain Containing Oxidoreductase (WWOX) and Melanocortin 5 receptor (MC5R).

PVT1 is considered as a candidate oncogene, and was associated e.g. with Hodgkin lymphoma (HL) (27). PVT1 is closely located in the genome to the MYC gene, an universal amplifier of transcription associated with e.g. leukaemia or Burkitt lymphoma (52). They are considered “tween players among [some] haematological malignancies.” (53) Up-regulation of PVT1, mediated by MYC, is thought to confer a proliferation advantage to malignant cells, e.g. in acute myeloid leukaemia (AML), acute promyeloid leukaemia (APL), or multiple myeloma (MM) (53). PVT1 and the circularization of the exon 2 of the PVT1 gene (circPVT1), an alternative transcript, have been reported to enhance malignant cells and hinder the immune response to the tumour during cancer progression (53). Further, two independent variants of PVT1 (rs13255292 and rs4733601) were associated with diffuse large B-cell lymphoma (DLBCL), with genome-wide significance (54). These two markers, which were significant in our cohort, are in low LD to each other (D’ ≤ 0.357).

WWOX is a gene with many faces. Being located at a common fragile site (16q23.3) at the genome, it is considered a tumor suppressor gene. An increased incidence of lymphomas was observed in mouse experiments (55). However, WWOX is involved in several biologic functions interacting with many proteins (56). WWOX is involved in, inter alia, TGFβ1- and TNF-mediated cell death. WWOX inhibits Wnt-signaling and may control genotoxic stress-induced cell death synergistically with p53/TP53 (27, 57). WWOX has already been associated with severe aGvHD in a previous GWAS (p=1.75x10⁻⁷) (58). Defects in the gene WWOX are associated further with multiple types of cancer, i.a. hematopoietic malignancies (37, 59).

The protein encoded by MC5R is a receptor for melanocyte-stimulating and adrenocorticotropic hormones. It seems to play a role in sebum generation, but little is known about its involvement of tumorigenesis (27). In a small mouse experiment, MC5R was found as the only melanocortin receptor mediating the α-melanocyte-stimulating hormone (α-MSH) that promotes myelopoiesis, which in turn has been associated with GvHD in other animal models (60, 61).

Two other markers intronic of KCNS3 and LINC02570 were of genome-wide significance with respect to OS, but only at the screening step. The lack of a reasonable functional explanation and the lack of significance in the refined analysis limit the evidence for this finding.

Finally, the gene GABRG3 appeared suggestively significant related to relapse. The protein (a gamma-aminobutyric acid receptor) encoded by GABRG3 is the major inhibitory neurotransmitter in the mammalian brain (27). It contains the benzodiazepine binding site. Benzodiazepine use during HSCT was associated with adverse effects, as is the case in non-transplant patients (62, 63).. The GABAergic system is involved in immune cell functions, inflammatory conditions and diseases in peripheral tissues (64, 65). It was found that GABA protects human islet cells against the deleterious effects of immunosuppressive drugs and exerts immunoinhibitory effects (66).

Although we used all available data collected over a long period within three transplantation centres, the sample size is rather small for a GWAS considering three competing key events, due to the problem of multiple testing. We found it necessary to fit many and rather complex models in order to adequately apply Cox’s proportional hazard models to investigate the impact of individual SNPs on the course on the post-transplant outcome. While this is time-consuming and computationally expensive, it is the only way to separately assess their impact on key events while examine them in their entirety. Appropriate adjustment for clinical covariates appears to be another key to reliable estimation results. Thus, we were unable to run a single genome-wide analysis in suitable time with feasible use of computer resources. It is likely, that potentially associated markers were not selected during the screening step. We also restricted our investigation to recipient genotypes. A further differentiation between GvHD of different stages would be desirable, but only advisable in larger samples. Although we conducted some subgroup analyses, we did not fully investigate the interaction with clinical or environmental factors.

The observed suggestive association of a GABRG3 marker and relapse could be due to the administration of benzodiazepine during HSCT. Since the use of narcotic analgesics and psychotropic drugs was not considered as clinical covariate (data not available), the association, although unlikely, may be caused by residual confounding. If this is the case, it justifies the need for an appropriate adjustment when studying associations between gene and HST outcomes.

We found the recipient SNP rs17154454 of SEMA3C to be associated with EFS and severe GvHD. In addition recipient SNPs in the genes PXN, PVT1, MC5R and WWOX potentially influence the outcome of HSCT. All discovered genes are directly or indirectly involved in molecular pathways that could affect GvHD, relapse or other key events after HSCT. However, identifying genes associated with the post-Tx events remains challenging due to disease- and treatment-related confounders, competing key events, and multiple tests.

The datasets presented in this article are not readily available because of legal and privacy restrictions. Requests to access the datasets should be directed to arosenb@gwdg.de.

The studies involving humans were approved by the Ethical Review Committee of the University of Regensburg (approval no 02-220), the Ethical Review Committee of the University Medical Center Göttingen (approval no 6/7/10) and the Newcastle and North Tyneside Research Ethics Committee (REC Red: 14/NE/1136 and 07/H0906/131). The studies were conducted in accordance with the local legislation and institutional requirements. Written informed consent for participation in this study was provided by the participants’ legal guardians/next of kin.

AR: Formal analysis, Methodology, Software, Writing – original draft. RC: Writing – review & editing, Resources. RD: Writing – review & editing. DK: Resources, Writing – review & editing. DW: Resources, Writing – review & editing. GW: Resources, Writing – review & editing. HB: Funding acquisition, Writing – review & editing, Supervision. AD: Resources, Writing – review & editing, Conceptualization, Funding acquisition. EH: Funding acquisition, Writing – review & editing, Conceptualization, Resources.