For this study, we separately harvested approximately 8 ml of blood samples from 15 patients who developed GVHD and 13 patients who did not develop GVHD (no GVHD) at 21 to 100 days following allo-HSCT. In addition, we also collected samples from 15 HIs. The study design and patient characteristics are described in more detail in Table 1. The immune repertoire information of these patients is illustrated in Supplementary Table 1.

The analysis of CDR3 amino acid lengths across all productive reads in the samples obtained from patients with or without GVHD and HIs revealed no notable differences (Figure 1a, Supplementary Figure 1a). However, the count of clonotypes with low frequencies (less than 100) was significantly higher in patients with GVHD compared to those without GVHD (Figures 1b, c, Supplementary Figure 1b), indicating a relative enrichment of low-frequency clonotypes in GVHD patients during the early post- HSCT period. Additionally, no statistically significant differences were found in the enrichment of low-frequency clonotypes between patients with aGVHD grades ≤II and those with grades ≥III (Supplementary Figure 2).

To address the variations in the number of productive counts, a common limitation in TCR repertoire analyses, we performed a down-sampling procedure on the original dataset. This involved resampling the TCR data from both patients and HIs to match the lowest acceptable number of productive counts. After the normalization process, we characterized the general landscape of a “normal” immune repertoire in both patients with and without GVHD and HIs. Our results indicate that the clonotype distribution abundances exhibited no correlation with the donor’s age in patients with GVHD (Figure 1d, Supplementary Figure 1c). Distinct V and J gene usage was observed among groups, particularly for TRBV15 and TRAV2 between the GVHD and no GVHD groups. Additionally, differences in the TRAV1-1, TRAV1-2, TRAV5, TRAJ8, TRBV27, TRBV14-1, and TRBV5–6 gene segment levels were noted between the GVHD and HI groups. In addition, differences in the TRBV15 levels were also noted between the no GVHD and HI groups (Figures 1e, f; Supplementary Figures 1d, e).

The results indicated that the numbers of TRB and TRA clonotypes in the GVHD group were significantly higher than those in the no-GVHD group and the HI group (Figures 2a, b, e). The clonotype frequency distribution from all samples was transformed to z-scores. Clonotypes with a z-score greater than 2 were selected. The results of the clonotype analysis indicated that the GVHD group had fewer expanded clonotypes compared to the other two groups (Figure 2d). A comparative analysis using the Inverse Simpson Index (ISI) showed that the TRB and TRA diversity in the GVHD group was higher than that of the no-GVHD group (Figures 2c, f). The GVHD and HI groups showed similar richness and diversity in the TCR repertoire, while the TCR diversity of the no GVHD group showed much lower richness and diversity. Additionally, the numbers of TRA clonotypes in the GVHD group were higher than those in the no GVHD and HI groups, while the TRA diversity exhibited a similar trend (Figures 2e, f). We also analyzed the distribution of top TCR clones. We found that the no GVHD group exhibited an uneven and screwed TCR clonotype distribution, while the GVHD group showed similar trends for all the properties. The top 10 intervals of the TRB counts relative to the total counts of the no-GVHD group were significantly higher than those observed in the GVHD group (Supplementary Figure 3a). Conversely, for the top 11 to 100, and for the 101 to 1000 TRB counts relative to total counts, the GVHD group showed an increasing trend, although not significant (Supplementary Figure 3b, c). Compared with the other two groups, the GVHD group had slightly higher TCR diversity in the low-frequency range (0–0.1%) but slightly lower TCR abundance at high frequencies above 0.1% (Supplementary Figure 3d). However, the differences were not statistically significant. Importantly, these observations were made during early immune reconstitution (day 21–100 post-HSCT) and in bulk peripheral blood T cells, and thus may reflect composite effects of immune recovery, inflammation, and immunosuppression rather than GVHD-specific clonotypes alone.

Given the TCR clonal diversity among individuals, we explored whether there are common characteristics in the patients’ TCR in the same groups and across different groups, to identify any similar reactivity patterns. To evaluate this, we made use of the published Grouping of Lymphocyte Interactions by Paratope Hotspots algorithm version 2 (GLIPH2) to identify amino acid patterns in different TCRs that are likely to share antigen specificity to group the lymphocyte interactions according to the paratope hotspots (13–15). This approach revealed motif groups shared across patients. To assess the extent of sharing of the GLIPH groups across patients, we quantified overlap across all pairs of different groups via the Jaccard index (Figure 3a), a normalized measure of overlap that considers the presence versus absence of features. As shown in the results, the highest indices were observed in different samples within the GVHD group (Figure 3b).

To test whether the TCR repertoires have a predictive role in the occurrence of GVHD, 3 representative patients (Patient 31, Patient 32, Patient 33) who showed no clinical manifestations of GVHD (stage 0) after HSCT at the time of sampling but subsequently developed GVHD (stage 1) were selected. By comparing their TCR profiles before GVHD onset with those of the GVHD group and the non-GVHD group, we found that their TCRs were more similar to the GVHD group (Figures 3c–h). These findings suggest that TCR sequence profiles may have a predictive role in the occurrence of GVHD.

To further investigate the relationship between the TCR repertoire and the progression of GVHD, we compared samples from the same patient in the GVHD group at different stages: during GVHD (stage 1), and after GVHD resolution (stage 2). The results indicated that in the majority of patients (except for one patient) the number of TRB and TRA clonotypes declined and the ISI declined as the GVHD resolved, suggesting that the TCR clonotypes become less homogeneous (Figures 4a, b, Supplementary Figure 4). We further sorted the top 100 clonotypes by their abundance at different stages of GVHD occurrence and compared their presence and percentage through clonotype tracking. It was observed that more expanded clonotypes emerged after GVHD resolution, which is consistent with the previously mentioned decrease in TCR repertoire diversity (Figures 4c, d).

Interestingly, in two patients initially without GVHD who subsequently developed GVHD (as mentioned before, Patient 31, Patient 32), we observed that the dominant clones at the onset of GVHD diminished or disappeared as the GVHD resolved. After comparing the TCR clonotype differences across different GVHD stages we observed that the TCR clonotypes during the pre-GVHD (stage 0) and acute GVHD (stage 1) stages exhibited high similarity, whereas the TCR clonotypes during the resolution phase of GVHD (stage 2) show lower similarity to those of the previous stages (Figures 4e, f). This suggests that these clones are associated with GVHD and may aid in predicting its occurrence, allowing for the earlier identification of subclinical GVHD patients before symptoms manifest.

DeWolf et al. analyzed tissue samples from 7 GVHD patients, including blood, bone marrow, and intestines, to reveal specific expression patterns of TCRs in GVHD patients. Their study found specific TCR repertoire overlap across tissues and among different GVHD patients, with the same TCR sequences being expressed at higher frequencies across different patients (16). We first identified high-frequency TCR clones (CDR3s with frequency >0.01%) from the peripheral blood of our patient cohort (both GVHD and non-GVHD), then compared these clones with publicly available TCR repertoires derived from intestine, blood, and bone marrow samples of GVHD patients obtained via rapid autopsy from DeWolf’s study. For each tissue type, we computed the intersection of CDR3s with our high-frequency clones and summed their counts to evaluate the degree of clonal sharing. The validation studies in bone marrow samples showed that these same TCRs were expressed at higher proportions in GVHD patients compared to non-GVHD patients (Figure 5b). A similar trend was also observed in the blood (Figure 5a) and intestine (Figure 5c) samples. This result suggests that specific TCR clones may play a crucial role in the pathogenesis of GVHD. By comparing the TCR expression profiles of GVHD patients with those of non-GVHD patients, we were able to identify TCR sequences that are highly enriched in GVHD, which may be closely related to the pathological process of GVHD.

Compared to the no GVHD group, the patients in the GVHD group had higher levels of cytomegalovirus (CMV)-specific TCR, Epstein-Barr Virus (EBV)-specific TCR, and Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2)-specific TCR, while the trend for influenza-specific TCR was reversed. However, the differences were not statistically significant between the two groups (Figures 5d–g). This phenomenon may be associated with the use of immunosuppressants in GVHD patients, which makes these patients more prone to developing infections.

Human specimens were studied as part of a prospective observational study whose objective was to characterize the T cell repertoire in blood samples from allo-HSCT patients with or without a history of GVHD, and from 15 healthy individuals. The study design did not include blinding or randomization. The clinical features of 15 patients with GVHD and 13 patients without GVHD were compared as shown in Table 1. None of the patients exhibited active CMV or EBV infections at the time of blood sample collection. Pre-transplant conditioning varied according to patient age, diagnosis, remission status, extent of prior therapies, and co-morbidities; and the dosage and intensities of the myeloablative regimens (24). GVHD prophylaxis was administered using a calcineurin inhibitor combined with short-term methotrexate, with anti-thymocyte globulin added based on protocol or physician discretion. Posttransplant granulocyte colony-stimulating factor was used in all patients. Acute and late acute GVHD were diagnosed clinically with histological confirmation when possible. Staging of GVHD was graded according to standard criteria (25, 26). All patients underwent allo-HSCT at the Department of Hematology of The First Affiliated Hospital of Soochow University between 2022 and 2023. All normal samples were obtained from healthy volunteer donors. All subjects provided Institutional Review Board-approved informed consent for collection of blood samples.

Blood samples were collected at the time of aGVHD onset within 21 to 100 days following transplantation. In addition, blood samples were also collected after GVHD remission. The mononuclear cells were isolated by density centrifugation from the blood samples using a conventional protocol (8). The recovered cells were lysed in RLT buffer (QIAGEN), homogenized using QIA shredder columns (QIAGEN), and stored at −80 °C. The immune repertoire amplification and sequencing procedure described by van et al. (8) was modified as follows. Briefly, the mRNA of the TCR and BCR were purified and reverse transcribed using 5’RACE, and the V region of the TCR and BCR were amplified using two rounds of polymerase chain reaction (PCR). The specific reverse primers complementary to the C-region of the TCRs combined with an adapter were included in the reaction. The entire volume of the cDNA (10 µl) was used as a template for the 50 µl RACE-PCR. The generated amplicons were then added to the sequencing adaptors using paired-end Illumina sequencing. The raw data extraction was performed with PyIR—a minimally-dependent high-speed wrapper for the immunoglobulin Basic Local Alignment Search Tool (IgBLAST) (v1.20.0). The default options were used for the TCR sequence alignment and assembly. Before the downstream analysis, all samples were normalized to the same number (1 million) of productive reads to avoid bias as indicated in the study by Odak et al. (20). All bioinformatics and statistical analysis were performed on a Linux workstation equipped with the CentOS 7 operating system. The complete mapped sequencing data were transferred into immunarch for advanced analyses (https://immunarch.com/). The TCR clonality and diversity indices commonly used include clonal numbers/types, TCR repertoire richness/evenness (Chao1/Inverse Simpson Index), unique CDR3 species, and CDR3 sequence frequency were extracted and calculated. All TCR indices used in this study were calculated according to previous publications (8, 18, 27). Z-score normalization refers to the process of normalizing every value in a dataset such that the mean of all of the values is 0 and the standard deviation is 1.

All statistical analyses were conducted using the R software version 4.2.2. The Fisher’s exact test, two-sample Kolmogorov–Smirnov test, and Wilcoxon signed rank test were used to compare the variables of TCR repertoire between different groups. The results were illustrated using boxplots. The boxplots display the median, first quartile, and third quartile, with whiskers extending to the most extreme values within 1.5 times the interquartile range (IQR). For all statistical tests, a p-value below 0.05 was deemed statistically significant.

The GLIPH2 algorithm was applied to a dataset composed of all patient samples simultaneously (TCRs defined by the Vβ CDR3 amino acid sequences and whose counts constitute more than 0.01% of the total counts), with the following GLIPH2 parameters: simulation_depth = 1000, kmer_min_depth = 3, cdr3_length_cutoff = 8, refer_file = refererence_files/ref_CD48.txt. HLA data were not included. This approach identified 5,609 GLIPH groups, including groups defined by either local (“motif”) similarities or “global” similarities. These groups were included in subsequent analyses.

The CDR3 regions of our samples were compared to an existing database of TCR sequences with known antigen specificity. The database called VDJdb can be accessed at https://vdjdb.cdr3.net and https://github.com/antigenomics/vdjdb-db and was downloaded for the comparison mentioned here.