Validation of single nucleotide polymorphisms in invasive aspergillosis following hematopoietic cell transplantation

Invasive aspergillosis (IA) remains a significant cause of infectious mortality in hematopoietic cell transplantation (HCT) recipients, despite improvements in transplantation practices and the introduction of mold-active azoles in the last decade. In contemporary patient cohorts, IA is associated with mortality rates that range from 19% to 60%.^1-5 

HCT recipients have increased susceptibility to IA as a result of profound immune defects from their underlying disease or the HCT process.^1-3,6  Despite similar risk factors, only approximately 3% to 15% of patients who undergo allogeneic HCT develop IA, indicating that other factors are important beyond universal Aspergillus exposure. The most robust identified risk factors relate to immune function in HCT recipients, and genetic variation within genes that regulate Aspergillus-specific responses or general immune reactivity are likely to contribute to IA susceptibility. In the last decade, more than 40 single nucleotide polymorphisms (SNPs) have been reported to influence the risk of Aspergillus-related diseases, and more than 20 have been associated with IAin patients with hematologic malignancies or after HCT (Table 1).^7-18  However, the majority of these studies were not validated independently, and sometimes the results of a study conflict with those of other studies. For example, although associations with Dectin-1/CLEC7A and IA have been reported in some studies (Table 1),^8,9  Chai et al¹⁹  found no association in their study. Additionally, the studies examined small patient cohorts and, at most, several SNPs simultaneously. Thus, published data are insufficient to estimate the pretransplantation risk for individual patients and to guide the clinical management of those at highest risk. Identifying patients at risk for IA through pretransplantation genetic analysis could improve patient care through targeted prophylaxis, surveillance, and treatment strategies.

The dominant role of the fungal biomarker galactomannan (GM) in the contemporary diagnosis of IA is another possible important factor in genetic analyses. The current IA classification system (proven, probable, possible, and no IA) is based on the 2008 revised European Organisation for Research and Treatment of Cancer/Invasive Fungal Infections Cooperative Group and the National Institute of Allergy and Infectious Diseases Mycoses Study Group (EORTC/MSG) criteria.²⁰  With the introduction of the GM assay, tissue-based and therefore proven diagnoses have declined sharply. Additionally, these classifications only inconsistently correlate with outcomes; conversely, several studies have shown that dissemination or a positive serum GM (SGM+) portends worse outcomes.^1,21  Furthermore, genetic studies that included patients prior to the widespread use of GM may have missed IA cases, given that the case definition would have relied heavily on less sensitive, culture-based methods, primarily from biopsy specimens.

In this study, we address several key limitations of prior studies by independently validating SNPs that were previously reported to be associated with IA using a large (>2500 patients) cohort of HCT donors and recipients. To minimize diagnostic bias, we retrospectively performed both bronchoalveolar lavage (BAL) fluid and SGM testing on banked patient samples when available. Retrospective testing was performed on samples from patients who met possible IA criteria but were diagnosed prior to the advent of routine GM assay testing. We also analyzed not only the proven and probable IA categories, but further refined “probable IA” to examine if there are different or stronger associations in those with a positive serum GM assay versus all patients with probable and proven IA.

We performed a comprehensive PubMed search using the keywords “aspergillosis” and “polymorphism” to identify studies published through November 2015 that reported an association between a genetic polymorphism and IA at a significance of P ≤ .05 in HCT recipients. Studies on SNPs in non-HCT populations and those that reported associations with other Aspergillus-related conditions (eg, allergic bronchopulmonary aspergillosis, colonization, etc) were excluded. Reported associations with nonsingle nucleotide polymorphism genetic variants, such as variable number tandem repeats, genetic deletions, insertions, or microsatellites were excluded because of limitations of our SNP genotyping array.

Patients who received an allogeneic HCT at the Fred Hutchinson Cancer Research Center (FHCRC) and Seattle Cancer Care Alliance between 1990 and 2011 were eligible. All included patients received T-cell–replete grafts. All recipient and donor samples were collected prior to HCT per approved research protocols. The FHCRC maintains an extensive electronic patient record database, and baseline pretransplantation demographic information, including age, sex, XY karyotype, ABO blood group, and race were prospectively collected through this clinical research database. Given that IA designations are not captured in this database and represent a combination of clinical, radiographic, and laboratory findings, IA phenotypes were based on retrospective chart review of all patients. Because only a small percentage of patients were nonwhite (9%), only white patients were included in the analysis. Genotyping for HLA-A, B, C, DRB1, and DQB1 was performed to determine patient-donor HLA matching. The study and use of these samples was approved by the FHCRC Institutional Review Board.

IA classifications were per the 2008 revised EORTC/MSG criteria.²⁰  The primary adjudication of IA phenotypes was performed by author C.E.F. on the basis of chart review, including review of images, reading of images by a radiologist, and review of laboratory/microbiologic information. In cases that were not clear, authors T.M.H. and M.B. also evaluated the data, and a consensus was reached through discussion. Patients with possible IA (typically because of the presence of clinical and radiographic criteria but lacking a microbiologic criterion) were not included. Other reasons for exclusion were: if patients had possible, probable, or proven IA pretransplantation, or another invasive mold infection either pre- or posttransplantation (n = 235); if patients received mold-active therapy or prophylaxis for their transplantation (voriconazole, amphotericin B deoxycholate or its derivatives, or posaconazole; n = 308), or if patients had insufficient follow-up information (n = 150).

All recipient and donor samples were collected before HCT according to research protocols approved by the FHCRC Institutional Review Board. Samples were prepared and genotyped using 3 different genotyping platforms, as described previously.²²  The genotypes of the candidate SNPs were determined separately for each platform. Candidate SNPs not genotyped on the array were imputed using the 1000 Genomes Project phase 1 SNPs as a reference panel and the software IMPUTE version 2 (http://mathgen.stats.ox.ac.uk/impute/impute_v2.html). For imputed SNPs, the posterior probability of the most probable genotype is calculated as the probability of observing an unobserved genotype at the imputed locus, given all of the observed genotypes in the flanking region. The imputed SNP genotype in a sample was retained only if the posterior probability of the most probable genotype exceeded 0.8; otherwise, it was treated as missing. On each platform, an imputed SNP was considered passing if the percent of missing genotypes was <10%, the average posterior probability of the most probable genotype for all samples on that platform was >0.9, and the Hardy-Weinberg equilibrium P value was <10⁻⁸. Five SNPs did not pass these quality control measures on the Affymetrix 5.0 platform (Thermo Fisher Scientific), and the samples genotyped on that platform were excluded from the analysis (supplemental Table 1, available on the Blood Web site). Supplemental Table 1 details whether each SNP was genotyped or imputed for each cohort.

FHCRC maintains an extensive specimen biorepository dating back to the early 1990s, with banking of excess serum and plasma samples, and in some cases, BAL fluid. For patients who were initially considered as having “possible IA” and who did not have a BAL and/or SGM assay performed clinically, we attempted to locate samples to test retrospectively. Serum and BAL fluid that was collected within 7 days of the possible IA diagnosis were eligible for testing. The GM testing was performed on serum and BAL fluid by a certified technician using the Bio-Rad Platelia Aspergillus EIA (Bio-Rad, Redmond, WA), per package insert. An index of ≥0.5 was considered positive for both serum and BAL.²³ 

We analyzed the association with 2 major phenotypes: proven/probable IA, according to EORTC/MSG criteria, and a novel phenotype based on SGM. Genetic associations with IA were evaluated using Cox regression, based on time from HCT to infection. Death without infection was a competing risk for all analyses. All analyses were adjusted for donor relationship and for the first 4 principal components of genome-wide association study SNPs to account for possible population stratification. We did not adjust for any baseline risk factors for IA, because these factors are not plausibly related to genotype and would not be true confounders. Despite this, given the large time period of this study, we did perform an analysis controlling for year of transplantation, and this did not change our results (data not shown). We also did not adjust for post-HCT factors that may influence the risk of infection, such as neutropenia or steroid use. First, because adjusting for factors potentially in the causal pathway for infection represents a study of mechanism, this was beyond the scope of this investigation. Second, when we analyzed the relationship between neutropenia and steroid use (using graft-versus-host disease as a proxy) and each SNP, no association was found with any of the SNPs. For each candidate SNP locus, we assessed allelic and genotypic (recessive and dominant) association, as previously described.^22,24  We analyzed the relationship of the different SNPs and development of IA during 2 time periods following HCT: within 3 months and 2 years. These end points were chosen on the basis of prior studies that looked at similar time frames and also to investigate if there is a different relationship for those who develop early IA versus those who develop IA later.

A 2-sided P ≤ .05 was used as the threshold of significance when the SNP was strictly replicated (eg, same genotype [patient or donor] and genetic model) using the standard phenotype of proven/probable IA at 2 years. When any of these factors differed from the original article (eg, SGM+ IA phenotype, IA at 3 months, different genetic model), we considered it an exploratory analysis. Although the SNPs analyzed were still previously reported in some manner and have biologic plausibility, if the analysis differed from that originally reported, we reported the results and displayed the P values separately if the P value was ≤.05.

Previously published studies of 25 SNPs in 14 genes met study inclusion criteria, as described in the Methods section. Table 1 outlines these SNPs and the reported degree of association with IA susceptibility. Of the 25 SNPs, 20 passed quality control by genotyping or imputation on all 3 platforms used in this study. Of the 5 that did not pass quality control, all problems occurred with the first platform (Affymetrix), and included 1 SNP with low posterior probability (TNFR1, rs767455), 1 SNP with low Hardy-Weinberg equilibrium (TLR1, rs4833095), and 3 SNPs with low call rates (CLEC7A, rs3901533; CD209, rs4804800 and rs7252229). Three SNPs had a minor allele frequency (MAF) of <0.05 in the Affymetrix cohort (TLR5, rs5744168, MAF, 0.03; AGER, rs2070600, MAF, 0.04; and CD209, rs11465384, MAF, 0.04), although the other 2 cohorts had an MAF of ≥0.05 for these SNPs, and these were included in the final analysis (supplemental Table 1).

Overall, 483 patients met criteria for either proven (n = 194) or probable (n = 289) IA (“proven/probable”), and 2126 HCT recipients had no evidence of IA during follow-up (“no IA”). Additionally, we separately analyzed the subset of patients who had positive SGM (“SGM+”), either performed prospectively as part of their diagnostic workup or retrospectively in patients for whom we had stored serum. Eighty additional probable IA cases (17%) were identified through retrospective serum and BAL GM testing, and in 58 patients who were proven/probable by a different method and did not have an SGM performed at the time, we confirmed a positive SGM. Table 2 summarizes clinical characteristics of these groups.

Using strict replication criteria (same genotype, genetic model, proven/probable IA end point at 2 years only) we validated 2 SNPs in 2 genes: rs1840680 in PTX3 and rs7309123 in CLECL7a (Table 3). Figure 1 illustrates the cumulative incidence of IA development following HCT in the validated SNPs.

In addition to using the strict replication criteria above, we also explored associations using the same SNPs but with deviations in the initial model, including different genotypes (donor vs recipient), genetic models (allelic, recessive, dominant), time to IA development (3 months vs 2 years), and end point (SGM+). Overall, 13 different SNPs in 9 genes had an association at P ≤ .05 with at least 1 of the models (Table 4): 5 SNPs in 4 genes in the proven/probable group, and 10 SNPs in 7 genes in the SGM+ group. Associations in 2 SNPs (rs2305619 in PTX3 and rs5743801 in TLR6) were found with both end points. Cumulative incidence curves for these SNPs are shown in Figure 2.

Several SNPs had a low MAF in our population. In order to investigate how this might affect our ability to successfully validate these SNPs, we performed an analysis of the minimal detectable hazard ratio for a range of MAF with a power of 80% and a type 1 error rate of 0.05 (supplemental Figure 1). For the available sample size and number of cases in our study, a low MAF, such as 0.05 for rs4986790 in TLR4, requires a very large association (hazard ratio of >12) to reliably identify the relationship.

Using a large cohort of HCT recipients with donor-recipient–paired DNA samples, we examined 20 candidate SNPs in 14 genes that had well-documented associations with IA susceptibility following HCT in previous studies. All of the SNPs examined are on genes that are important in immune function and therefore have biologic plausibility in IA susceptibility. Using the EORTC/MSG diagnostic categories of proven and probable to define our IA phenotype, we strictly replicated 2 SNPs from 2 genes. We also examined the association using not only the traditional EORTC/MSG case definitions of proven or probable, but also a novel phenotype based on a positive SGM assay. In modern cohorts, the diagnosis of IA increasingly relies on GM, and a positive SGM has been associated with worse prognosis, raising the possibility that patients with fungal angioinvasion, as judged by a positive SGM assay, may have more profound immune defects.²¹  However, because this is a novel phenotype in genetic analysis, we did not consider analysis of this end point as a strict replication. Using the SGM+ end point, we did find several associations at P ≤ .05, including SNPs in PTX3, CD209, CXCL10, TLR6, TNFR1, IFNG, and S100B. Additionally, using the proven/probable end point, but with genetic models that differed from the original studies, we also found associations in PTX3, CLEC7a, TLR6, and PLG. It is important to note that our statistical model purposely did not include adjustments for any baseline risk factors for IA, given that these factors are not plausibly related to genotype and would not be true confounders. We also did not adjust for post-HCT factors that may influence the risk of infection, such as neutropenia or steroid use, because adjusting for factors potentially in the causal pathway for infection represents a study of mechanism and was beyond the scope of this investigation.

Among the SNPs that were either strictly replicated or had an association in this study, a number occur in genes that have well-defined roles in host defense against aspergillosis, either in human cells challenged with A. fumigatus or in murine pulmonary challenge models. Pentraxin-3 is soluble collectin that binds to conidia and facilitates their uptake by myeloid cells via complement receptor 3 and Fc γ receptor IIA–dependent mechanisms.²⁵  Pentraxin-3–deficient mice are susceptible to pulmonary challenge with high doses of A. fumigatus conidia in the absence of exogenous immune suppression.^26,27  CD209/DC-SIGN is a type II C–type lectin receptor that mediates the in vitro binding and internalization of A. fumigatus conidia by human alveolar macrophages. This process can be inhibited by inclusion of A. fumigatus–derived GMs in the assay.²⁸  Dectin-1/CLEC7A, a well-characterized type II C–type lectin receptor with an intracellular immunoreceptor tyrosine–based activation motif, binds A. fumigatus β-glucan on germinating conidia and initiates a signal cascade that culminates in alveolar macrophage nuclear factor κB–dependent cytokine responses.^29-31  Dectin-1–deficient mice have variable susceptibility to A. fumigatus challenge in the absence of exogenous immune suppression,^32,33  in part due to molecular redundancy in A. fumigatus recognition by host cells. Humans with a Mendelian defect in Dectin-1/CLEC7A signaling (a stop codon polymorphism: Y238X) do not spontaneously develop IA,³⁴  although this study and others have defined a clear IA risk in the context of immune damage associated with HCT.⁸  CARD9 is a signal transducer that integrates signals from Dectin-1 and other CLRs (eg, Dectin-2/CLEC6A) that are implicated in Aspergillus recognition.^35,36  Mendelian defects in CARD9 can lead to the spontaneous development of IA in humans,³⁷  consistent with the model that multiple recognition events of A. fumigatus polysaccharide ligands by host receptors converge on CARD9 and individually contribute to innate defense against aspergillosis. The associations of other SNPs in this study (eg, in the CXCL10 promoter) provide a rationale for further investigations into the biologic role of associated genes in cultured human cells and animal models of disease.

A key strength of this study was the number of cases and controls and the presence of both donor and recipient samples for genetic testing. The aggressive standard diagnostic workup throughout the study period at our institution, with routine radiographic studies, lung fluid sampling, and serum and BAL GM testing in the era of GM assay availability, coupled with our ability to test banked serum and BAL samples retrospectively for the presence of GM, reduced the diagnostic bias over time and allowed for inclusion of more cases. We also manually reviewed all patient charts, including imaging, clinical history, and microbiologic results, in order to minimize phenotype misclassification bias.

This study also had several limitations and/or unanticipated results. Although the main goal of this analysis was to replicate previously reported associations, our analysis broadened the search for associations by considering recipient or donor genomes, 2 IA phenotypes, 2 time points, and 3 genetic models. The 2-sided significance level of .05 or less that we used should be considered the minimal threshold of significance. This threshold does not account for the multiplicity of SNPs examined nor the multiple association analyses performed for each SNP when the genetic model or end point deviated in any way from that in the primary article. The appropriate significance threshold to use in these cases is debatable, given that there is a higher pretest probability of these SNPs than in a true discovery study. The significance threshold of .0001, representing a Bonferroni correction for the 20 approximately independent SNPs × 24 genetic/end point combinations, would constitute the most stringent value. However, because this will likely be the largest IA genome-wide association study in this population, we presented all SNPs that had a P ≤ .05 in Table 4, under the premise that the results will provide some guidance for ranking candidate SNPs in future association studies. Although we were able to replicate 2 SNPs found in genes with strong biologic plausibility, we were unable to validate all of the SNPs tested. Some SNPs could not be validated because of quality control (eg, call rate, posterior probability, etc). However, an unanticipated result was our inability to validate the TLR4 SNP rs4986790. This SNP was reported as part of a haplotype (S4) with TLR4 SNP rs4986791 (which is in strong linkage disequilibrium with rs4986790) by our group as significantly associated with IA (adjusted hazard ratio, 2.49; 95% confidence interval, 1.15-5.41; P = .02 in the validation study for all donors),¹³  although the SNP rs4986791 was also found to be significant by itself (reported in their supplemental Table S3¹³  and in our Table 1). The low MAF of this SNP, 0.05 in our study, could explain our inability to replicate results of the previous study. As supplemental Figure 1 illustrates, even with our large cohort, with 80% power, our minimum detectable hazard ratio approached 13, which is much larger than the hazard ratio of 2.49 reported in the previous study. Additionally, we removed all of the patients that were included in both our cohort and the former cohort, which decreased our statistical power further. Therefore, this lack of validation may be a statistical issue and not a biologic one. Of note, de Boer et al¹²  similarly did not validate this SNP alone (only in combination with interferon γ). Last, the majority (>80%) of recipients in our study received myeloablative conditioning, and so the generalizability to other HCT settings is unclear.

This study is the largest to date, to our knowledge, to validate SNPs previously shown to confer an increased risk of IA in HCT recipients. Our findings suggest that the development of IA may be significantly influenced by genetic variation in genes important to immune function. Identifying patients who are at risk for IA pretransplantation through genetic analysis, or, in future studies, identifying those with IA who are at risk for poor outcomes, could improve patient care through targeted prophylaxis, surveillance, and treatment (ie, combination vs monotherapy) strategies, or even influence donor selection. This study constitutes a critical milestone toward designing such strategies. Future studies in large cohorts using a discovery approach may provide further insight into the genetic influences that underlie IA development.

The authors thank Zachary Stednick for database service; Terry Stevens-Ayers, Jessica N. Yi, Elsa Garnace, and Bradley Edmison for laboratory and repository assistance; Jenna Gravley for technical assistance; Anajane Smith for HLA typing; Sonia Goyal and Jessica N. Yi for medical chart assistance; Hu Xie for the creation of supplemental Figure 1; and Gary Schoch for data management.

This work was supported by grants from the National Institutes of Health (NIH) National Heart, Lung, and Blood Institute (HL087690 and HL105914 [J.A.H., P.J.M., E.H.W., M.B., B.E.S., D.M.L., and W.F.], K24HL093294 [M.B.]), National Cancer Institute (CA18029 and CA15704 [Fred Hutchinson Cancer Research Center]), NIH core grant P30 CA008748 [Memorial Sloan Kettering Cancer Center]), National Institute of Allergy and Infectious Diseases (T32 AI007044-37 [C.E.F.]), and by Joel Meyer’s Endowment Scholarship (C.E.F.), the Robert Sinskey Foundation (C.E.F., T.M.H.), and a Burroughs Wellcome Fund Investigator in the Pathogenesis of Infectious Diseases Award (T.M.H.).

Contribution: C.E.F. was responsible for literature search, study design, data collection, data analysis and interpretation, and writing; M.B., J.A.H., and T.M.H. were responsible for literature search, study design, data interpretation, and writing; P.J.M. and E.H.W. were responsible for study design, data interpretation, and writing; and D.M.L., W.F., L.P.Z., and B.E.S. were responsible for primary statistical analysis and interpretation, study design, and writing.

Conflict-of-interest disclosure: The authors declare no competing financial interests.

Correspondence: John A. Hansen, Fred Hutchinson Cancer Research Center, 1100 Fairview Ave N, D2-100, PO Box 19024, Seattle, WA 98109; e-mail: jhansen@fredhutch.org.