This study was reviewed and approved by the Institutional Review Board of Chang Gung Memorial Hospital, and its approval IDs were 201304949B0, 201700769B0, 201701849B0, 201801985B0, and 201901246B0. All donors and recipients except the donors of unrelated HSCT signed informed consents. All methods of the study were performed according to the ethical requirements and regulations. For unrelated donors, informed consents were exempted because the HLA-matched donors were selected by the Stem Cells Center in Taiwan and the identity of the donors were made anonymous and disconnected to the physicians and research team. A total of 163 patients receiving HSCT was enrolled in this study, in which 99 patients were diagnosed as acute myeloid leukemia (AML), and 64 patients were acute lymphoblastic leukemia (ALL). The clinical characteristics of the 163 patients are shown in Table 1 . All donor–recipient pairs had fully matched HLA as revealed by high-resolution HLA typing using the SeCore kit (Thermo Fisher, Waltham, MA). The MicroSSP Allele Specific Typing Tray (Thermo Fisher, Waltham, MA) was used to resolve ambiguous alleles of the SeCore typing with sequence-specific primers.

Mortality was referred to the state of patients who died in the duration of study. The presence of CMV antigen or DNA in the peripheral blood of recipients after transplantation was defined as a CMV-infected case. CMV antigen in the leukocytes was determined by CMV Antigenemia Assay (MONOFLUO™, Bio-Rad). The test was considered positive when more than two polymorphonuclear leukocytes (PMN) were positive for CMV antigen in a total of 50,000 PMN. CMV DNA Quantitative Amplification test is a real-time quantitative PCR assay (COBAS^® AmpliPrep/COBAS^® TaqMan^® CMV Test, Roche). The nucleic acid test was considered positive when the Ct < 37. These two assays can assist clinicians in monitoring the status of CMV infection.

According to the International Bone Marrow Transplant Registry, GVHD was considered as acute GVHD (aGVHD) when it occurred within 100 days after transplantation. It can be divided into four grades according to the clinical characteristics of organs as defined below. Grade I: maculopapular rash over <25% of body area with no liver or gastrointestinal involvement; Grade II: maculopapular rash over 25% to 50% of body area, diarrhea 500 to 1500 ml/day, and bilirubin 2 to 6 mg/dl; Grade III: maculopapular rash over >50% of body area, and severe diarrhea; Grade IV: skin blisters, bilirubin >15 mg/dl, severe diarrhea with pain, and life-threatening. Grades I–II were defined as mild GVHD, and Grades III–IV were defined as severe GVHD. Chronic GVHD (cGVHD) usually occurs more than 100 days after transplantation or occurs continually for more than 100 days without remission (26). Patients without any symptoms of aGVHD or cGVHD during the study period were defined as no GVHD.

Relapse was defined as recurrence of malignancy based on one or more of the following: bone marrow morphology, minimal residual disease by either flow cytometry, cytogenetics, imaging results, or short tandem repeat (STR) analysis. High-throughput amplicon sequencing (AmpFISTR Identifiler Amplification Kit, Thermo Fisher, Waltham, MA) was performed to analyze STR and to evaluate HSCT engraftments for identification of mixed chimerism (27–29) according to the manufacturer’s instruction. The presence of >5% recipient STR alleles in the chimeric test was considered as a surrogate marker of disease relapse.

Based on our initial screening of SNPs that were present in the CTLA4 gene of the Taiwanese population, and the studies demonstrating the importance of promoter and exon 1 in gene expression and the SNPs with clinical association (30), a total of eight DNA fragments of the four co-stimulatory genes (CTLA4, TNFSF4, CD28, and PDCD1) were selected for analyses of donor SNPs ( Table 2 ). A total of 17 SNPs in CTLA4, 3 SNPs in TNFSF4, 9 SNPs in CD28, and 5 SNPs in PDCD1 were subject to association study with the risk for relapse, mortality, GVHD, and CMV infection. All SNP variants were deposited to the NCBI database dbSNP and the accession numbers are provided in the Supplementary Table S1 . Data can be accessed with the following link: https://www.ncbi.nlm.nih.gov/SNP/snp_viewTable.cgi?handle=WANGWT.

Peripheral blood (3 ml) was collected from the corresponding donors of the recipients, and the genomic DNA was extracted by using QIAamp DNA Blood Mini Kit (Qiagen, Valencia, California, USA). A total of eight primer pairs ( Table 3 ) were used to amplify genomic DNA fragments covering the 34 SNPs as abovementioned ( Table 2 ). The PCR reaction volume was 25 μl, including 1 μl of forward and reverse primer (10 μM), 12.5 μl of 2X HotStart PCR Mix (BIOMAN, Taipei, Taiwan) containing HotStart Taq DNA polymerase, reaction buffer, and dNTP, 1 μl of sample DNA and 9.5 μl of ddH₂O. The PCR program was 1 cycle at 94°C for 4 min, 30 cycles of 94°C for 30 s, 58°C for 30 s, 72°C for 45 s, and 1 cycle of 72°C for 10 min. Subsequently, 5 μl of PCR products was fractionated on a 1.5% or 2% agarose gel to visualize the PCR products. The size of PCR products ranged from 1,039 bp to 2,234 bp ( Table 3 ). The Big Dye Terminator Cycle Sequencing kit (Thermo Fisher, Waltham, Massachusetts, USA) and the ABI PRISM genetic analyzer (Thermo Fisher, Waltham, Massachusetts, USA) were used for direct sequencing according to the manufacturer’s instructions. Because of the insufficient genomic DNA and failure of PCR reaction, not every donor had complete SNP data available.

Single-locus association tests were performed to identify the donor SNPs that were associated with the defined outcomes in the recipients. The allele or genotype frequencies between cases (patients with the indicated outcomes) and controls (patients without the indicated outcomes) were compared with Cochran-Armitage Trend test (or Trend test) and the allelic test using the PLINK software v1.07 (31). The allele effects of the SNPs on each outcome were further examined using the logistic regression analysis assuming three modes of inheritance: additive model, recessive model, or dominant models. p < 0.05 was considered statistically significant. The Haploview 4.2 (32) software was used to determine the linkage disequilibrium (LD). The pairwise linkage disequilibrium value D’ and the haplotype blocks of SNPs were determined. The haplotype blocks were defined as the SNPs in this region had no evidence for historical recombination.

A total of 163 patients receiving HSCT including 99 patients with AML and 64 patients with ALL were enrolled in this study. Clinical characteristics and the tracking data of mortality, CMV infection, relapse, and GVHD for these patients are listed in Table 1 .

With the importance of co-stimulatory signals in the immune system and transplantation tolerance, the associations of the four co-stimulatory genes including CTLA4, TNFSF4, CD28, and PDCD1 with the mortality, relapse, CMV infection, and GVHD after HSCT were analyzed in this study. Based on our initial screening of SNPs in the CTLA4 gene of the Taiwanese population, the importance of promoter and exon 1 in gene expression, and the SNPs with clinical association, the genomic regions covering the promoter, exon 1, exon 4, and 3’-UTR of CTLA-4 (17 SNPs), the promoter and exon 1 of TNFSF4 (3 SNPs), the promoter and exon 1 of CD28 (9 SNPs), and the promoter, exon 1, and exons 4–5 including intron 4 of PDCD1 (5 SNPs) for all donors were amplified by PCR using the forward and reverse primers ( Tables 2 , 3 ). The PCR amplicons were sequenced and the association of candidate SNPs with the adverse outcomes of patients with AML and ALL were analyzed by Trend test and allelic test, and by logistic regression analysis with additive, dominant, or recessive mode, respectively. The genotype and allele frequencies for all donors are summarized in the Supplementary Tables S2 – S5 .

By analyzing a total of 34 SNPs, 4 SNPs (rs733618, rs11571316, and rs3087243 in CTLA-4, and rs41386349 in PDCD1) and 1 SNP (rs11571315) elicited significant recessive allelic effects and contributed to the post-HSCT mortality for patients with AML and ALL, respectively ( Table 4 ). For patients with AML, the C allele of rs733618 (CC vs. CT+TT, p = 0.0376, OR = 2.77, and 95% CI = 1.07–7.22) and the G allele of rs11571316 (GG vs. AA+AG, p = 0.0441, OR = 2.32, and 95% CI = 1.03–5.24) located in the promoter region of CTLA4 were associated with higher risk for mortality. The SNP rs3087243 in the 3’-UTR of CTLA4 (GG vs. AA+AG, p = 0.0441, OR = 2.32, and 95% CI = 1.03–5.24) and rs41386349 in the intron 4 of PDCD1 (GG vs. AA+AG, p = 0.0362, OR = 2.62, and 95% CI = 1.07–6.42) also conferred recessive effects to the risk for mortality of patients with AML. In addition, the SNP of rs41386349 also elicited additive effects on post-HSCT mortality for patients with AML. For patients with ALL, the C allele of rs11571315 (CC vs. CT+TT, p = 0.0289, OR = 6.14, 95% CI = 1.21–30.99) in the promoter region of CTLA4 was found to associate with higher risk for mortality.

One SNP (rs6705653 in PDCD1) and four SNPs (rs36084323, rs41386349, rs6705653, and rs2227982 in PDCD1) were found to associate with CMV infection in patients with AML and ALL, respectively ( Table 5 ). For patients with AML, the C allele of rs6705653 (CC+CT vs. TT, p = 0.0138, OR = 7.91, and 95% CI = 1.54–40.71) elicited a dominant effect and contributed to the risk for CMV infection. For patients with ALL, the alternative T allele of the same SNP rs6705653 elicited an additive (Trend test: p = 0.0198, and additive effect: p = 0.0186) or a dominant effect to the risk for CMV infection (TT+CT vs. CC, p = 0.0201, OR = 3.95, and 95% CI = 1.25–12.49). The C allele of rs36084323, A allele of rs41386349, and G allele of rs2227982 in PDCD1 gene also associated with a higher risk for CMV infection (allele model: p = 0.0265, 0.0356, and 0.0252, respectively).

One SNP (rs200353921 in CD28) and three SNPs (rs5839828, rs36084323, and rs2227982 in PDCD1) were associated with the risk of disease relapse in patients with AML and ALL, respectively ( Table 6 ). For patients with AML, the T allele of rs200353921 located on the promoter region of CD28 gene was associated with a higher risk of relapse (allele model p = 0.0343 for T vs. A, OR = 2.1, and 95% CI = 1.06–4.18). For patients with ALL, the G7 allele of rs5839828, the C allele of rs36084323, and the G allele of rs2227982 in the PDCD1 gene were also associated with a higher risk for disease relapse (allele model: p = 0.0008, 0.0095, and 0.0018, respectively).

The status of GVHD was classified into four categories including GVHD III–IV (severe GVHD), GVHD I–II (mild GVHD), chronic GVHD (cGVHD), and no GVHD. Two SNPs (rs1234314 and rs45454293) in the promoter region of TNFSF4 were associated with the risk for severe GVHD ( Table 7 ) in patients with AML (rs1234314: Trend test, p = 0.006; allele model, p = 0.0114 for C vs. G, OR = 7.39, and 95% CI = 1.58–34.52; and rs45454293: Trend test, p = 0.0145, allele model p = 0.0100 for T vs. C, OR = 4.86, and 95% CI = 1.47–16.07). Both SNPs elicited additive and recessive effects on the risk for severe GVHD. No SNP was found to associate with the risk for severe GVHD in patients with ALL.

No SNP was associated with the risk for mild GVHD in patients with AML. Four SNPs (rs231775 in CTLA4, and rs41386349, rs6705653, and rs2227982 in PDCD1) were associated with the risk for mild GVHD (GVHD I–II) in patients with ALL. The A allele of rs231775 on exon 1 of CTLA4 (allele model: p = 0.0343 for A vs. G, OR = 2.28, and 95% CI = 1.07–4.89), the A allele of rs41386349 (allele model: p = 0.0436 for A vs. G, OR = 2.71, and 95% CI = 1.03–7.1), the T allele of rs6705653 (Trend test: p = 0.0086; allele model: p = 0.0039 for T vs. C, OR = 3.53, and 95% CI = 1.51–8.28) in the intron 4 of PDCD1, and the G allele of rs2227982 (Trend test: p = 0.0194; allele model: p = 0.0055 for G vs. A, OR = 3.4, and 95% CI = 1.44–8.03) in the exon 5 of PDCD1 gene contributed to a higher risk for mild GVHD. The three SNPs in the PDCD1 gene were also associated with CMV infection as above mentioned.

No SNP was found to associate with the risk for cGVHD in patients with AML. Five SNPs (rs5742909 and rs231775 in CTLA4, rs28541784 in CD28, and rs6705653 and rs2227982 in PDCD1) were associated with the risk for cGVHD in patients with ALL. The C allele of rs5742909 (allele model: p = 0.0465 for C vs. T) and the G allele of rs231775 (recessive model: p = 0.0279 for GG vs. AA+AG) on the CTLA4 gene contributed to the higher risk for cGVHD, yet in different modes. In addition, the T allele of rs28541784 on CD28 gene was associated with a higher risk for cGVHD (Trend test: p = 0.0473; allele model: p = 0.0303 for T vs. C, OR = 2.78, and 95% CI = 1.11–6.98). Of the SNPs located in the PDCD1 gene, the C allele of rs6705653 (allele model: p = 0.0066 for C vs. T) and the A allele of rs2227982 (allele model: p = 0.0305 for A vs. G) also contributed to the higher risk for cGVHD.

Two SNPs (rs3181096 and rs3181098 in CD28) and four SNPs (rs4553808, rs62182595, rs16840252, and rs5742909 in CTLA4) were associated with the protective effects on the development of GVHD in patients with AML and ALL, respectively. The dominant effects of the T allele of rs3181096 and the A allele of rs3181098 in the promoter region of CD28 conferred protective effects on the development of GVHD in patients with AML (dominant model: p = 0.0231 and 0.0235, respectively). The G allele of rs4553808 (allele model: p = 0.0452 for G vs. A, OR = 4.63, and 95% CI = 1.04–20.58), the A allele of rs62182595 (allele model: p = 0.0330 for A vs. G, OR = 5.19, and 95% CI = 1.15–23.4), the T allele of rs16840252 (allele model: p = 0.0350 for T vs. C, OR = 5.1, and 95% CI = 1.13–22.98), and the T allele of rs5742909 (allele model: p = 0.0479 for T vs. C, OR = 4.54, and 95% CI = 1.02–20.2) of the CTLA4 gene conferred better protective effects on the development of GVHD in patients with ALL.

The SNPs (n = 20) that were associated with the risk for adverse outcomes in patients with either AML or ALL were subject to LD analysis ( Figure 1 ). Several pairs of SNPs had high or complete LD including the rs3087243 with rs231775 (D’ = 0.97), with rs62182595 (D’ = 0.98), and with rs11571316 (D’ = 0.96) in the CTLA4 gene; the rs6705653 with rs41386349 (D’ = 1) in the PDCD1 gene; and the rs3181096 with rs3181098 (D’ = 0.96) in the CD28 gene. In addition, there were three haplotype blocks including the SNPs in the CD28, CTLA4 and PDCD1, respectively. These data imply a potential genetic linkage of these SNPs in the human genome.