Single-cell analysis of adaptive NK and CMV-specific T-cell dynamics after allogeneic HSCT∗ Authors: Aino-Elina Häkkinen, Jani Huuhtanen, Sofie Lundgren, Riitta Niittyvuopio, Urpu Salmenniemi, Harri Lähdesmäki, Mikko A. I. Keränen, Mikko Myllymäki, Satu Mustjoki Published: 2026 ABSTRACT Key Points•Asymptomatic CMV viremia shapes post-HSCT immune reconstitution, driving oligoclonal CMV-specific CD8+ T-cell and adaptive NK cell responses.•Donor CMV serostatus strongly influences the timing, maturation, and function of CMV-specific T and adaptive NK cells after HSCT. Introduction Allogeneic hematopoietic stem cell transplantation (HSCT) is the only curative option for many hematological malignancies but is complicated by impaired early immune reconstitution, leading to treatment-related complications such as opportunistic cytomegalovirus (CMV) reactivation.1 CMV establishes lifelong latency,2 and its reactivation after HSCT is associated with significant morbidity, increased risk for acute graft-versus-host disease (GVHD), and reduced survival.3,4 CMV viremia occurs in 30% to 80% of seropositive recipients without prophylaxis and is routinely monitored to guide preemptive therapy.5,6 Recent letermovir prophylaxis has reduced clinically significant CMV reactivation and has improved survival.7, 8, 9 Protection against CMV reactivation depends on CMV-specific T-cell reconstitution,10, 11, 12 and the adoptive transfer of CMV-specific T cells is increasingly used after HSCT.13,14 CMV-specific CD8+ T cells can comprise 10% of the memory T-cell pool15 and display an effector memory phenotype.16 CMV also drives clonal expansion and adaptive differentiation of natural killer (NK) cells,17, 18, 19, 20, 21 which are associated with reduced relapse risk and improved immunotherapy responses.22, 23, 24, 25, 26 However, single-cell reconstitution of CMV-specific T cells and adaptive NK cells during early, subclinical CMV reactivation after HSCT remains poorly defined. Here, we dissect the cellular and molecular mechanisms underlying immune responses with single-cell RNA and T-cell receptor αβ sequencing (scRNA+TCRαβ-seq) of longitudinal blood samples (n = 32) systematically collected pre- and post-CMV viremia from HSCT donor-recipient pairs. We used CMV-epitope tetramers to sort and sequence CMV-specific T cells, enabling us to profile antigen-specific T cells more precisely than with in silico matching to databases of known TCR-antigen pairs. Importantly, comparing CMV-induced T-cell and adaptive NK cell development in patients receiving grafts from CMV-seropositive and CMV-seronegative donors allows us to specifically explore memory responses in vivo. We observed extensive anti-CMV immune responses during CMV viremia, highlighting the potency of CMV in modulating immune system recovery after HSCT. Methods More detailed descriptions of the experimental methods and analyses are available in the supplemental Methods. Patients and samples Peripheral blood samples from HSCT recipients and donors were collected at the Helsinki University Hospital, Finland, in the years 2018 to 2020 (supplemental Table 1). Peripheral blood samples from healthy blood donors were provided by the Finnish Red Cross Blood Service (supplemental Table 2). Single-cell RNA and TCRαβ sequencing and data analysis CD45+-sorted peripheral blood mononuclear cells (PBMCs) were subjected to scRNA+TCRαβ-seq using the Chromium Single Cell 5′ Library & Gel Bead Kit (10x Genomics). Raw sequencing data was processed using CellRanger version 3.1.0 pipelines, with GRCh38 as the reference genome. Downstream data analysis was performed in R version 4.2.1 with Seurat version 4.3.0.27 To overcome batch effects between samples, we used the probabilistic approach implemented in scVI version 0.13.0.28 Differentially expressed genes were identified using the Model-based Analysis of Single-cell Transcriptomics test29 with Bonferroni correction. Antigen-specific T-cell profiling with pMHC tetramer panel scRNA+TCRαβ plus peptide-major histocompatibility complex (pMHC) sequencing was performed using Immudex dCODE Dextramers (Immudex) from a total of 5 samples from 3 different patients. The panel consisted of 32 epitopes from endogenous viruses, common tumor-associated antigens, and 7 different negative control dextramers (supplemental Table 3). The dextramer-positive populations were determined from cells that passed transcriptomic quality control. For dextramer normalization, we used ICON version 0.0.0.9000.30 pp65NLV-specific TCRβ sequencing and data analysis We stained CD8+ T cells from 19 samples with the HLA-A02∗01-CMV pp65NLV-tetramer (Proimmune) before proceeding to flow sorting. TCRβ-seq from genomic DNA was conducted using the ImmunoSEQ assay at Adaptive Biotechnologies, and downstream analysis was performed with VDJtools version 1.2.1.31 pp65NLV-specific clones were filtered based on enrichment in the tetramer-sorted samples compared with single-cell TCRαβ data from the same time point. Statistics Nonparametric tests were used: the Mann-Whitney U test for pairwise comparisons and the Kruskal-Wallis test for multiple comparisons. Local ethical committee approved the study, and study participants provided written informed consent in accordance with the Declaration of Helsinki. Results Immune reconstitution after HSCT begins with monocyte expansion To characterize immune dynamics after HSCT, we prospectively collected blood from donor-recipient pairs, including pretransplant samples and frequent recipient follow-up samples up to 3 months after HSCT. Routine CMV monitoring enabled the selection of samples obtained before, during, and after CMV reactivation, alongside pretransplant donor and recipient samples. To enable tetramer sorting of CMV-epitope-specific T cells, we selected only patients with the HLA-A∗02 allele. In total, we performed single-cell RNA sequencing (scRNA-seq) and TCRαβ-seq on 32 longitudinal CD45+-sorted blood samples from 5 HSCT donor-recipient pairs and 6 healthy controls (Figure 1A-B). All recipients were diagnosed with a hematological cancer. Three patients received HSCT from a matched related donor, 1 from a matched unrelated donor, and 1 from a haploidentical (haplo) donor. Recipients with HLA-matched donors underwent myeloablative conditioning, whereas the haplo-HSCT recipient underwent reduced-intensity conditioning. No patients showed evidence of relapse at the time of their last sample collection for the study. Recipients of HLA-matched transplants were treated with corticosteroid and calcineurin inhibitor–based immunosuppressive regimens, and T-cell–depleting therapy with antithymocyte globulin (ATG) was administered in the matched unrelated donor, and posttransplant cyclophosphamide was administered in haplo-HSCT. All recipients were CMV seropositive, whereas donors were seropositive (CMVpos; n = 2) and seronegative (CMVneg; n = 3). Patients did not receive prophylactic CMV treatment. Laboratory-confirmed, asymptomatic CMV reactivation occurred in 4 patients at an average of 32 days after transplant. Ganciclovir was administered as the first-line antiviral therapy upon CMV reactivation. Grade I-II acute GVHD of the skin was noted in patients 2 and 3. Detailed clinical characteristics are presented in supplemental Table 1 and supplemental Figure 1.Figure 1.Immune landscape after allo-HSCT. (A) Schematic of study design. (B) Timeline of sample collection and key clinical features of each recipient-donor pair. The purple and green colors denote CMV-seropositive and CMV-seronegative individuals, respectively. A pre-HSCT sample was obtained from all D/R pairs except for pair 2, from which no donor sample was available. (C) Uniform manifold approximation and projection (UMAP) representation of CD45+-sorted PBMCs from all D/R pairs at all time points, with manual cell type annotations. (D) Proportion of selected cell types out of PBMCs over time. The solid lines represent averages across all patients, and the dashed lines denote individual patients. (E) Proportion of Souporcell-deconvoluted recipient genotype cells out of all PBMCs over time. The circles indicate results from routine clinical PCR chimerism testing of polymorphic gene markers from each patient. (F) Souporcell-predicted patient chimerism of individual cell types in each sample. The darker color indicates a higher share of cells within that cell type classified as patient genotype. (G) UMAP representation of PBMCs with detectable CMV reads called with Vulture. D/R, donor-recipient; PCR, polymerase chain reaction. Immune landscape after allo-HSCT. (A) Schematic of study design. (B) Timeline of sample collection and key clinical features of each recipient-donor pair. The purple and green colors denote CMV-seropositive and CMV-seronegative individuals, respectively. A pre-HSCT sample was obtained from all D/R pairs except for pair 2, from which no donor sample was available. (C) Uniform manifold approximation and projection (UMAP) representation of CD45+-sorted PBMCs from all D/R pairs at all time points, with manual cell type annotations. (D) Proportion of selected cell types out of PBMCs over time. The solid lines represent averages across all patients, and the dashed lines denote individual patients. (E) Proportion of Souporcell-deconvoluted recipient genotype cells out of all PBMCs over time. The circles indicate results from routine clinical PCR chimerism testing of polymorphic gene markers from each patient. (F) Souporcell-predicted patient chimerism of individual cell types in each sample. The darker color indicates a higher share of cells within that cell type classified as patient genotype. (G) UMAP representation of PBMCs with detectable CMV reads called with Vulture. D/R, donor-recipient; PCR, polymerase chain reaction. We pooled all longitudinal scRNA-seq samples, applied probabilistic modeling to mitigate batch effects, and identified 35 cell clusters, none of which were completely patient-specific. We annotated 197 887 immune cells, including 37 123 NK cells, 48 056 CD8+ T cells, 34 724 CD4+ T cells, and 66 195 myeloid cells (Figure 1C; supplemental Figure 2A-C). Consistent with prior studies,32,33 we observed an early increase in CD14+ monocytes after HSCT, followed by the expansion of T-cell and NK cell populations. B-cell expansion was marginal over the 3-month follow-up (Figure 1D). To compare immune cell states between healthy individuals and HSCT recipients, we mapped 43 148 single cells from healthy controls onto the integrated uniform manifold approximation and projection (UMAP) shown in Figure 1C using label transfer (supplemental Figure 3, supplemental Methods). Most immune cell states were shared between healthy controls and HSCT recipients, although several states appeared enriched or unique to the posttransplant setting. Consistent with ongoing immune reconstitution during the engraftment period, B cells constituted a proportionally larger subset in healthy controls than in HSCT recipients. To deconvolute patient- and donor-derived cells in posttransplant samples, we leveraged the genotype-free scRNA-seq demultiplexing tool Souporcell.34 We detected an average of 3.6% patient-derived cells 2 weeks after transplant, rapidly declining by D+30 (30 days after transplant; Figure 1E). At 2 weeks, donor chimerism was lowest in mature lymphocyte populations, including CD4+ and CD8+ memory T cells and adaptive NK cells (Figure 1F). Interestingly, we noted that the early mixed chimerism of CMV-seropositive patient 4 with a CMVneg donor consisted mostly of persisting CD4+ T cells (Figure 1E-F). CMV reads are expressed in up to 0.08% of PBMCs during reactivation After primary infection, CMV is thought to take up latency in monocytes, lymphocytes, dendritic cells, and CD34+ hematopoietic stem cells.2,35,36 Conjecturing whether CMV-infected circulating blood cells could be inferred in the scRNA-seq data, we called microbial reads from scRNA-seq reads in all samples with Vulture.37 CMV reads were detected in 5 cells in patient 1 at the onset of CMV viremia (D+31), representing 0.08% of CD45+ PBMCs at that time point. The CMV-infected cells were classified as CD14+ classical monocytes (n = 3), CD56dim NK cells (n = 1), and cytotoxic CD4+ T cells (n = 1) (Figure 1G). Of the 5 patients profiled, this patient withstood the highest CMV viral load (14 600 IU/mL; supplemental Figure 1A). No other herpesvirus reads were observed in the data set. Adaptive NK cells expand and evolve after CMV reactivation To understand the detailed transcriptional and phenotypic changes in immune cells responsible for CMV-mediated responses, we analyzed T-cell and NK cell phenotypes pre- and post-CMV viremia by separating and reclustering different immune cell lineages. NK cells comprise functionally distinct subsets along a maturation trajectory, from cytokine-producing CD56bright to cytotoxic CD56dim to adaptive NK cells; we defined these NK cell phenotypes based on previously described NK cell markers.38, 39, 40 CD56bright, CD56dim, and adaptive NK cells were present in all patients after HSCT, with patient-specific heterogeneity in phenotype and temporal dynamics (Figure 2A-B; supplemental Figure 4A-B). Based on Souporcell genotyping, most reconstituting NK cells were of donor origin (Figures 1F and 2D). A maximum of 4.3% of all NK cells, comprising mostly mature adaptive NK cells, were of patient origin 2 weeks after HSCT (patient 1), but patient-derived cells virtually disappeared thereafter (Figure 2D). Most NK cells developing earliest were immature CD56bright NK cells (Figure 2C-D), as expected.41 NK cells continued maturing to CD56dim and adaptive NK cells, resembling the NK cell landscape of healthy controls after 1 month (Figure 2D).Figure 2.Adaptive NK cells expand and evolve after CMV reactivation. (A) UMAP representation of NK cells from all donor, patient follow-up, and healthy control scRNA-seq samples. Clusters were annotated manually. (B) Heat map of key NK cell phenotype marker expression for each cluster identified in panel A. (C) Proportion of all NK cells and adaptive NK cells out of CD45+ PBMCs over time in each patient. The dashed lines denote timing of CMV reactivation, and the figure background color denotes donor CMV serostatus. (D) Proportion of NK cells belonging to different NK cell clusters in each sample (bottom). Bars in the top panel indicate total number of NK cells in each sample, with the colors representing Souporcell-deconvoluted patient- and donor-derived NK cells. (E) Volcano plot of differentially expressed genes between adaptive NK cells from CMVpos and CMVneg donors after onset of CMV viremia. (F) Adaptive NK cell signature, memory, cytotoxicity, and exhaustion scores for individual adaptive NK cells after viremia in patients with CMVpos and CMVneg donors and in healthy controls. P values were calculated using the Mann-Whitney U test. (G) UMAP representation of NK cells for patient 1 and patient 4 at each time point. The red stars denote timing of CMV reactivation. (H) Changes in log-normalized gene expression levels over time for selected genes in adaptive NK cells derived from patient 1. The dashed lines denote timing of CMV reactivation, and the color indicates whether the gene was found to be differentially expressed between any time points. (I) Changes in adaptive NK cell signature, memory, and cytotoxicity scores over time in adaptive NK cells from patient 1. The dashed lines denote timing of CMV reactivation. Adaptive NK cells expand and evolve after CMV reactivation. (A) UMAP representation of NK cells from all donor, patient follow-up, and healthy control scRNA-seq samples. Clusters were annotated manually. (B) Heat map of key NK cell phenotype marker expression for each cluster identified in panel A. (C) Proportion of all NK cells and adaptive NK cells out of CD45+ PBMCs over time in each patient. The dashed lines denote timing of CMV reactivation, and the figure background color denotes donor CMV serostatus. (D) Proportion of NK cells belonging to different NK cell clusters in each sample (bottom). Bars in the top panel indicate total number of NK cells in each sample, with the colors representing Souporcell-deconvoluted patient- and donor-derived NK cells. (E) Volcano plot of differentially expressed genes between adaptive NK cells from CMVpos and CMVneg donors after onset of CMV viremia. (F) Adaptive NK cell signature, memory, cytotoxicity, and exhaustion scores for individual adaptive NK cells after viremia in patients with CMVpos and CMVneg donors and in healthy controls. P values were calculated using the Mann-Whitney U test. (G) UMAP representation of NK cells for patient 1 and patient 4 at each time point. The red stars denote timing of CMV reactivation. (H) Changes in log-normalized gene expression levels over time for selected genes in adaptive NK cells derived from patient 1. The dashed lines denote timing of CMV reactivation, and the color indicates whether the gene was found to be differentially expressed between any time points. (I) Changes in adaptive NK cell signature, memory, and cytotoxicity scores over time in adaptive NK cells from patient 1. The dashed lines denote timing of CMV reactivation. CMV viremia considerably altered NK cell maturation, especially in patients with CMVpos donors (Figure 2C). The proportion of putative CMV-recognizing adaptive NK cells out of all PBMCs was higher after viremia in the 2 patients with CMVpos donors (18.9% and 7.4%) compared to the 2 patients with CMVneg donors (3.1% and 1.7%; Figure 2C). Adaptive NK cells derived from CMVpos donors underwent more transcriptomic changes over time compared to those from CMVneg donors, with an overall trend toward downregulation of gene expression (supplemental Figure 5A-C), consistent with previous reports of global downregulation of gene expression in adaptive NK cells compared to other CD56dim NK cells.21,42 After the onset of CMV viremia, adaptive NK cells from CMVpos donors expressed higher levels of KLRC2/NKG2C, CD3E, KLRG1, and LAG3 compared to those from CMVneg donors (Figure 2E). Moreover, adaptive NK cells derived from CMVneg donors expressed higher levels of granzymes and the transcription factors FCER1G/FcεRIɣ and SYK (Figure 2E), commonly downregulated in adaptive NK cells compared to conventional CD56dim NK cells.21,38 Postviremia adaptive NK cells derived from CMVpos donors demonstrated higher adaptive NK cell signature and memory scores (supplemental Methods), both in comparison with adaptive NK cells from CMVneg donors (P < .0001 and P < .0001; Mann-Whitney U test) and with healthy individuals’ adaptive NK cells (P < .0001 and P = .0001) (Figure 2F). Postviremia adaptive NK cells from CMVneg donors had higher cytotoxicity compared to those from CMVpos donors (P < .0001) and healthy individuals (P < .0001) (Figure 2F). In line with phenotype maturation, postviremia adaptive NK cells from CMVpos donors exhibited the highest overall expression of genes associated with T-cell exhaustion compared to CMVneg (P < .0001) and healthy donors (P < .0001) (Figure 2F). To understand the impact of donor CMV serology on adaptive NK cell development, we characterized in detail patient 1 and patient 4 with CMVpos and CMVneg donors, respectively (Figure 2G; supplemental Figure 6A). Patient 1, with a CMVpos donor and CMV reactivation 1 month after HSCT, had a large adaptive NK cell expansion, already seen a week after but most prominent 1 month after the onset of CMV viremia (Figure 2G). Compared to the previremia period, adaptive NK cells matured further, upregulating IFNG and the activating receptors KLRC2/NKG2C and KLRC3/NKG2E after viremia (Figure 2H). This translated into increases in core adaptive NK cell, memory, and cytotoxicity signature scores over time (Figure 2I). Adaptive NK cells began contracting after viremia but remained elevated 2 months after viremia onset (D+80; Figure 2C). Gene set enrichment analysis indicated that, compared with adaptive NK cells in the donor, reconstituting adaptive NK cells at D+80 downregulated pathways related to biosynthetic processes and protein metabolism (supplemental Figure 7A). Modeling of intercellular communication with NicheNet43 suggested that gene expression programs that were upregulated in adaptive NK cells during viremia were potentially induced by HLA-G, IFNG, and IFNL1 ligands through their respective NK cell receptors KIR2DL4, IFNGR1/2, and IL10RB (supplemental Figure 7B-C). Accordingly, a significant increase in KIR2DL4 expression was noted at the beginning of viremia (Figure 2H). In contrast, patient 4, with a CMVneg donor and 2 confirmed CMV reactivations after transplant, had a small expansion of adaptive NK cells and a notable increase in CD56bright NK cells (Figure 2G; supplemental Figure 6B). Developing adaptive NK cells in this patient showed modest transcriptomic changes over time and no differential enrichment in gene pathways, remaining relatively stable in phenotype after initial expansion and despite a second CMV viremia 1.5 months later (supplemental Figure 5D-E). CD4+ T cells exhibit the highest early mixed chimerism CD4+ T-cell help is crucial in controlling CMV reactivation through both cytokine production and direct cytotoxicity.44 We classified 6 distinct CD4+ T-cell clusters across patients and time points (Figure 3A-D; supplemental Figure 8A-B). We noted the lowest shares of CD4+ T cells in patient 2, the only patient who was administered ATG as lymphodepletion. ATG has previously been associated with delayed CD4+ T-cell reconstitution.45,46 Across all patients, 1.7% to 17.4% of CD4+ T cells present 2 weeks after transplant were patient-derived based on Souporcell genotyping, making CD4+ T cells the cell subset with the highest early mixed chimerism. Phenotypically, persisting patient cells belonged to CD4 naive/central memory T, T effector memory, and cytotoxic CD4+ clusters (Figure 3E). After 2 weeks, persisting CD4+ T cells declined in proportion in all except patient 4, in whom a large proportion of patient-derived cells belonged to a lower-quality effector memory T cell (TEM) cluster with upregulation of TNFAIP3 (Figure 3C-E; supplemental Table 4), a protein associated with negative regulation of inflammation.47Figure 3.CD4+ T cells exhibit the highest mixed chimerism. (A) UMAP representation of CD4+ T cells from all donor, patient follow-up, and healthy control scRNA-seq samples. Clusters were annotated manually. (B) UMAP of CD4+ T cell clone sizes. (C) Heat map of key CD4+ T cell phenotype marker expression for each cluster identified in panel A. (D) Proportion of CD4+ T cells belonging to different CD4+ T cell clusters in each sample (bottom). The bars in the top panel indicate the total number of CD4+ T cells in each sample, with the colors representing Souporcell-deconvoluted patient- and donor-derived CD4+ T cells. (E) UMAP representation of Souporcell-genotyped patient- and donor-derived CD4+ T cells. (F) Volcano plot of differentially expressed genes between Souporcell-genotyped patient- and donor-derived CD4+ TEM cells. (G) Changes in log-normalized gene expression levels over time for selected genes in cytotoxic CD4+ T cells in patient 1. CD4+ T cells exhibit the highest mixed chimerism. (A) UMAP representation of CD4+ T cells from all donor, patient follow-up, and healthy control scRNA-seq samples. Clusters were annotated manually. (B) UMAP of CD4+ T cell clone sizes. (C) Heat map of key CD4+ T cell phenotype marker expression for each cluster identified in panel A. (D) Proportion of CD4+ T cells belonging to different CD4+ T cell clusters in each sample (bottom). The bars in the top panel indicate the total number of CD4+ T cells in each sample, with the colors representing Souporcell-deconvoluted patient- and donor-derived CD4+ T cells. (E) UMAP representation of Souporcell-genotyped patient- and donor-derived CD4+ T cells. (F) Volcano plot of differentially expressed genes between Souporcell-genotyped patient- and donor-derived CD4+ TEM cells. (G) Changes in log-normalized gene expression levels over time for selected genes in cytotoxic CD4+ T cells in patient 1. Comparing gene expression between patient- and donor-genotype CD4+ TEM cells in post-HSCT time points, we found the proapoptotic protein BAX most significantly upregulated in persisting patient-derived cells (Figure 3F). Notably, we observed post-HSCT expansion of a cytotoxic CD4+ T-cell population in patient 1 with a CMVpos donor (Figure 3D). Cytotoxic CD4+ T cells upregulated STAT1, PRF1, LAG3, and PDCD1/PD-1 during CMV viremia, followed by downregulation of all T-cell exhaustion markers after viremia (Figure 3G; supplemental Figure 9A). This population comprised mostly 1 large donor-derived clone. CD8+ T-cell reconstitution is marked by clonal expansion and concomitant phenotype maturation Next, we characterized CD8+ T-cell reconstitution, identifying 7 distinct CD8+ T-cell clusters in the scRNA-seq data, ranging from naive to effector and memory T cells, along with mucosal-associated invariant T cell and ɣδ T cells (Figure 4A-C; supplemental Figure 10A-B). Compared to a healthy T-cell repertoire, early T-cell reconstitution favored CD8+ T cells. This resulted in an inverted CD4/CD8 ratio during the 3-month follow-up (Figure 4D), a characteristic posttransplant phenomenon reflecting rapid CD8+ T cell expansion, especially in CMVpos patients.48,49 Mucosal-associated invariant T cells and ɣδ T cells recovered faster than conventional CD8+ T cells, peaking ∼2 weeks with proportions relative to healthy controls (Figure 4E).Figure 4.CD8+ T-cell reconstitution is marked by clonal expansion and concomitant phenotype maturation. (A) UMAP representation of CD8+ T cells from all donor, patient follow-up, and healthy control scRNA-seq samples. Clusters were annotated manually. (B) UMAP of CD8+ T-cell clone sizes. (C) Heat map of key CD8+ T-cell phenotype marker expression for each cluster identified in panel A. (D) Proportion of CD4+ and CD8+ T cells and out of CD45+ PBMCs over time in each patient. The dashed lines denote timing of CMV reactivation, and the figure background color denotes donor CMV serostatus. (E) Proportion of CD8+ T cells belonging to different CD8+ T-cell clusters in each sample (bottom). The bars in the top panel indicate the total number of CD8+ T cells in each sample, with the colors representing Souporcell-deconvoluted patient- and donor-derived CD8+ T cells. (F) CD8+ TCR clonality measured using the Gini index, where higher values represent a more skewed TCR repertoire. (G) UMAP representation of CD8+ T cells from all donor samples and all patient post-HSCT samples, colored by donor TCR clone persistence. (H) UMAP representation of CD8+ T cells from all donor and recipient samples colored based on Souporcell genotyping. (I) Bar plot of CD8+ T-cell phenotypes identified in persisting and nonpersisting donor T-cell clones in donor samples. (J) Proportion of the 20 largest and all tetramer-sorted pp65NLV-specific CD8+ T-cell clones out of all CD8+ T cells in patient 1 scTCRαβ-seq data. scTCRαβ-seq, single-cell TCRαβ sequencing. CD8+ T-cell reconstitution is marked by clonal expansion and concomitant phenotype maturation. (A) UMAP representation of CD8+ T cells from all donor, patient follow-up, and healthy control scRNA-seq samples. Clusters were annotated manually. (B) UMAP of CD8+ T-cell clone sizes. (C) Heat map of key CD8+ T-cell phenotype marker expression for each cluster identified in panel A. (D) Proportion of CD4+ and CD8+ T cells and out of CD45+ PBMCs over time in each patient. The dashed lines denote timing of CMV reactivation, and the figure background color denotes donor CMV serostatus. (E) Proportion of CD8+ T cells belonging to different CD8+ T-cell clusters in each sample (bottom). The bars in the top panel indicate the total number of CD8+ T cells in each sample, with the colors representing Souporcell-deconvoluted patient- and donor-derived CD8+ T cells. (F) CD8+ TCR clonality measured using the Gini index, where higher values represent a more skewed TCR repertoire. (G) UMAP representation of CD8+ T cells from all donor samples and all patient post-HSCT samples, colored by donor TCR clone persistence. (H) UMAP representation of CD8+ T cells from all donor and recipient samples colored based on Souporcell genotyping. (I) Bar plot of CD8+ T-cell phenotypes identified in persisting and nonpersisting donor T-cell clones in donor samples. (J) Proportion of the 20 largest and all tetramer-sorted pp65NLV-specific CD8+ T-cell clones out of all CD8+ T cells in patient 1 scTCRαβ-seq data. scTCRαβ-seq, single-cell TCRαβ sequencing. CD8+ T-cell phenotypes were present in varying proportions in different patients and at different time points, but no clusters were completely patient-specific (Figure 4E; supplemental Figure 10A). Proportions of naive CD8+ T cells remained elevated during early immune reconstitution compared to healthy individuals. In line with maturation from naive to effector and memory phenotypes, the clonalities of TCR repertoires progressively increased. Using Gini index as a measure of immune repertoire evenness, the CD8+ TCR repertoire became more clonal in all patients after HSCT (Figure 4F), although this did not reach statistical significance (P = .38; Kruskal-Wallis test). Similarly, the proportion of large (>1%) and hyperexpanded (>10%) clones increased with time (Figure 4E). Hyperexpanded clones consisted mostly of CD8+ TEM/EMRA phenotypes characterized by the expression of effector molecules and downregulation of lymph node homing receptors (Figure 4B-C). The early-reconstituting CD8+ TCR repertoire consists mostly of donor-origin clones In the posttransplant setting, T-cell clones can originate either from the patient (patient-persisting; ie, noted in pre-HSCT samples), from the graft (donor-persisting; ie, noted in donor sample), or from T-cell maturation (donor de novo; ie, not noted in patient or donor samples). Based on a comparison of pre-HSCT patient and donor samples with post-HSCT patient samples, we found that donor-persisting CD8+ T cells were more prevalent (22-100 clones per patient) than patient-persisting CD8+ T cells (1-42 clones; Figure 4G,J; supplemental Figure 11A). Most reconstituting TCR repertoire was made up of clones not identified in pre-HSCT patient or donor samples; 98.4% of these clones were donor-derived based on Souporcell genotyping (Figure 4H). In general, the clones that appeared at the 2-week time point did not persist during follow-up. In particular, in patient 5, who received haplo-HSCT with posttransplant cyclophosphamide, only clonotypes that appeared after 2 months persisted, consistent with cyclophosphamide-mediated ablation of activated T cells when administered in the immediate days after HSCT50 (supplemental Figure 11A). When persisting donor-derived TCR clones were examined in the scRNA-seq data in donor pre-HSCT samples, persisting clones were transcriptionally CD8+ TEM and CD8+ TEM/EMRA phenotypes (Figure 4I). Comparing persisting and disappearing CD8+ T-cell clones in donor samples, persisting clones were more cytotoxic, with higher expression levels of granzymes, NKG7, FGFBP2, and CST7 (supplemental Figure 11B-C). pp65NLV-specific CD8+ T cells drive T-cell immunity toward CMV To characterize the antigen-specific CD8+ T-cell response to CMV, we screened 32 common viral epitopes presented on HLA-A∗02:01 from 5 post-HSCT samples (3 patients) with scRNA+TCRαβ plus pMHC-seq (Immudex dCODE Dextramers). Importantly, all patients and healthy controls in our study cohort shared the HLA-A∗02 allele. The NLVPMVATV epitope within pp65 (pp65NLV) was found to be the immunodominant CMV epitope, consistent with previous studies51,52 (supplemental Figure 12). To characterize dextramer-binding CD8+ T cells, we integrated the scRNA plus pMHC-seq data set with all CD8+ T cells from the scRNA-seq samples using scVI (supplemental Figure 13). Most dextramer-positive cells clustered within the CD8 TEM and terminally differentiated effector memory CD8 T-cells reexpressing CD45RA (TEMRA) populations identified in the scRNA-seq analysis regardless of antigen specificity, suggesting shared effector-memory transcriptional states across viral antigen specificities despite the limited number of non–NLV-specific cells. We proceeded to sort T cells specific to the immunodominant HLA-A∗02:01-CMV pp65NLV-tetramer from all donor-recipient pairs and seropositive healthy individuals (19 samples total; supplemental Figure 14). We then profiled the sorted T cells with TCRβ-seq and matched CMV-epitope–recognizing TCRs in the scRNA+TCRαβ-seq data (Figure 5A).Figure 5.pp65NLV-specific CD8+ T cells drive T-cell immunity against CMV. (A) Schematic of CD8+ T-cell HLA-A∗02:01-CMV pp65NLV-multimer sorting. (B) UMAP of multimer-sorted pp65NLV-specific CD8+ T cells in each subject. Multimer sorting was not performed on CMVneg healthy controls. (C) Bar plot of multimer-sorted pp65NLV-specific and VDJdb-queried CMV epitope-specific CD8+ T cells in each sample. (D) UMAP representation of CD8+ T cells belonging to the largest pp65NLV-specific clone in patient 1 at each time point. (E) Proportion of the largest pp65NLV-specific clone belonging to different CD8+ T-cell clusters in patient 1. (F) Changes in log-normalized gene expression levels over time for selected genes in the largest pp65NLV-specific clone in patient 1. The dashed lines denote timing of CMV reactivation, and the color indicates whether the gene was found to be differentially expressed between any time points. (G) Volcano plot of differentially expressed genes between NK-like CD8+ clusters 7 and 9. pp65NLV-specific CD8+ T cells drive T-cell immunity against CMV. (A) Schematic of CD8+ T-cell HLA-A∗02:01-CMV pp65NLV-multimer sorting. (B) UMAP of multimer-sorted pp65NLV-specific CD8+ T cells in each subject. Multimer sorting was not performed on CMVneg healthy controls. (C) Bar plot of multimer-sorted pp65NLV-specific and VDJdb-queried CMV epitope-specific CD8+ T cells in each sample. (D) UMAP representation of CD8+ T cells belonging to the largest pp65NLV-specific clone in patient 1 at each time point. (E) Proportion of the largest pp65NLV-specific clone belonging to different CD8+ T-cell clusters in patient 1. (F) Changes in log-normalized gene expression levels over time for selected genes in the largest pp65NLV-specific clone in patient 1. The dashed lines denote timing of CMV reactivation, and the color indicates whether the gene was found to be differentially expressed between any time points. (G) Volcano plot of differentially expressed genes between NK-like CD8+ clusters 7 and 9. We identified pp65NLV-specific TCRs in all patients and healthy controls profiled, with marked heterogeneity in clonotype abundance between subjects (Figure 5B-C). A total of 61 pp65NLV-specific TCR clones were identified in the pp66NLV-tetramer–sorted bulk TCRβ-seq data, only 8 of which were previously reported in the public database VDJdb.53 The number of pp65NLV-specific clones per patient or healthy control ranged from 1 to 12, representing 0.09% to 63.6% of the CD8+ T-cell repertoire (Figure 5C; supplemental Figure 11A). The largest clonal expansion was observed in patient 1, with a donor-derived, pp65NLV-specific clone encompassing 63.6% of the patient’s CD8+ TCR repertoire at viremia onset (Figure 4J). Furthermore, we used the single-cell TCRαβ data to predict TCR recognition of 6 different CMV epitopes with the Gaussian process method TCRGP54 (supplemental Methods). An additional 395 clones comprising at least 2 TCRs were predicted to be specific for 1 of the 6 CMV epitopes modeled by TCRGP (supplemental Figure 15A-B). pp65NLV-specific CD8+ T cells represent mature CD8+ T-cell phenotypes To understand the CD8+ T-cell phenotypes responding to CMV viremia after HSCT, we integrated scRNA-seq and TCRαβ data of pp65NLV-specific T cells. pp65NLV-specific CD8+ T cells belonged to TEM and TEM/EMRA clusters (Figure 5D-E; supplemental Figure 16A). Investigating the phenotype of the largest expanding pp65NLV-specific clone in patient 1, it was observed that the manifold clonal expansion was accompanied by upregulation of IFNG, STAT1, and the cytotoxicity markers PRF1 and GZMB, compared to the expression profile of the same clone in the donor sample (Figure 5F; supplemental Figure 9B). Expression of the coinhibitory receptors PDCD1/PD-1, LAG3, and TIGIT peaked during viremia and subsequently decreased (Figure 5F). Based on interaction analysis with NicheNet, gene expression changes in this clone before and after viremia were predicted to be modulated by ligands including IFNɣ and TNFα, which were most upregulated in naive/central memory CD4 T cells after viremia (Figure 6A), consistent with the role of CD4+ T cells in cytokine production.44Figure 6.Dynamic immune cell networks coordinate immune recovery in HSCT recipients. (A) Heat map of top-ranked ligands associated with gene expression changes in pp65NLV-specific T cells and corresponding ligand expression in sender-cell types before vs after viremia based on NicheNet analysis. Ligands are ranked based on the AUPR between a ligand’s target predictions and the observed transcriptional response. (B) Dynamic intercellular interactions in single cell transcriptomics (DIISCO)-predicted interaction networks in HSCT recipients with CMVpos donors across the 3-month follow-up period. (C) DIISCO-predicted interaction networks in HSCT recipients with CMVneg donors across the 3-month follow-up period. AUPR, area under precision-recall curve. Dynamic immune cell networks coordinate immune recovery in HSCT recipients. (A) Heat map of top-ranked ligands associated with gene expression changes in pp65NLV-specific T cells and corresponding ligand expression in sender-cell types before vs after viremia based on NicheNet analysis. Ligands are ranked based on the AUPR between a ligand’s target predictions and the observed transcriptional response. (B) Dynamic intercellular interactions in single cell transcriptomics (DIISCO)-predicted interaction networks in HSCT recipients with CMVpos donors across the 3-month follow-up period. (C) DIISCO-predicted interaction networks in HSCT recipients with CMVneg donors across the 3-month follow-up period. AUPR, area under precision-recall curve. In contrast, patient 4, with a CMVneg donor undergoing 2 subclinical CMV viremias, displayed deficient pp65NLV-specific CD8+ T-cell development, generating only 2 small anti-CMV clones after the second viremia (Figure 5C; supplemental Figure 6). A notable proportion of this patient’s cytotoxic T lymphocytes (CTLs), comprised of multiple small clones, belonged to a TEMRA cluster (cluster 3) characterized by upregulated IFNG expression compared to other CD8+ T cells (Figure 4E; supplemental Table 4). Terminally matured T cells express NK cell receptors Anti-CMV T-cell persistence has been associated with the acquired expression of activating and inhibitory NK cell receptors on CD8+ memory T cells.55,56 Importantly, we observed 2 distinct CD8 TEMRA clusters expressing NK cell receptors in our scRNA-seq data (Figure 4A,E; supplemental Table 4). We compared transcriptomic profiles between these NK-like CD8+ T cell clusters 7 and 9 (Figure 5G; supplemental Table 5). CTLs in cluster 7 upregulated KIR2DL1, KIR2DL4, KIR3DL1, NCAM1/CD56, SELL/CD62L, and KLRC1/NKG2A, whereas NK-like CTLs in cluster 9 had higher expression of KIR2DL3, KIR3DL2, KLRC2/NKG2C, and the exhaustion-associated TIGIT, potentially representing a more mature phenotype. A subset of the largest pp65NLV-specific clone in patient 1 was enriched in the cluster 7 phenotype after viremia, potentially representing terminal differentiation. Dynamic immune cell networks coordinate immune recovery in HSCT recipients The diverse immunological responses to post-HSCT CMV reactivation may arise from intricate cross talk between cells of the immune system over time. We applied dynamic intercellular interactions in single cell transcriptomics (DIISCO), a recently published Bayesian model based on Gaussian process regression,57 to model these temporally dynamic cell-to-cell interaction networks across our cohort. We modeled cell-to-cell interactions in grafts derived from CMVpos and CMVneg donors separately (supplemental Methods), identifying a more complex interactome in recipients with CMVpos donors (Figure 6B-C). Recipient-donor pairs with a CMVpos donor were predicted to have coordinated activation between adaptive NK cells, NK-like CD8 TEMRA cells, and cytotoxic CD4+ T cells, along with an inhibitory role for classical CD14+ monocytes. In recipient-donor pairs with CMVneg donors, CD4+ TEM cells were predicted to be more central, and a key regulatory network was identified between CD8+ TEM, cytotoxic CD4+, and CD8+ TEM/EMRA cells. Discussion We used single-cell transcriptomics and TCR sequencing to define post-HSCT immune recovery during CMV reactivation at unprecedented resolution. Our results detail adaptive NK cell and antigen-specific T-cell development in CMV-exposed and CMV-naive grafts, revealing patient-specific differences in the magnitude and transcriptional programs of antiviral responses. Overall, we observed early monocyte reconstitution, followed by T-cell and NK cell reconstitution, and only marginal B-cell reconstitution during the 3-month follow-up, in line with previous reports of post-HSCT immune reconstitution.10,32,33 As expected with myeloablative conditioning regimens,58 we observed only a few persisting patient-derived cells in the first weeks after transplant based on single-cell genotyping and T-cell clonotype matching. The highest early patient chimerism was observed in CD4+ T cells, a phenomenon previously reported after reduced-intensity conditioning.59,60 Phenotypically, patient-derived cells were naive/central memory T, TEM, and cytotoxic helper T cells and declined in numbers in all patients 2 weeks after transplant. In response to CMV reactivation, we observed adaptive NK cell expansion and phenotypic evolution to the greatest extent in patients with CMVpos donors. Adaptive NK cells from CMVpos donors displayed transcriptomic memory responses, whereas adaptive NK cells differentiating de novo from CMVneg donors were more cytotoxic. After HSCT, transferred T cells undergo homeostatic expansion, with large proportions of the T-cell repertoire directed against alloantigens and Herpesviridae.32 Correspondingly, we observed the largest clonal CD8+ T-cell expansion to be directed against the CMV epitope pp65. This clone upregulated cytotoxicity and exhaustion markers during CMV viremia but did not have an exhausted phenotype after viremia, concordant with earlier observations of the unique CTL response generated by CMV.16 After viremia, a subset of the largest pp65NLV-specific clone in patient 1 was enriched in an NK-like CD8+ T-cell phenotype, a mature phenotype associated with chronic inflammatory conditions, advancing age, and CMV seropositivity.56,61,62 Mechanisms by which CMV tries to escape T-cell recognition, including downregulation of major histocompatibility complex class I molecules in host cells, promote NK cell expansion and maturation.11,36 Overall, in our patient cohort, we did not observe an inverse relationship between the magnitude of the NK cell and T-cell response to CMV reactivation. Rather, patients displaying the largest expansions of adaptive NK cells had the most robust development of pp65NLV-specific CTLs, coinciding with donor CMV seropositivity. The major limitation of this study is the small and heterogeneous patient cohort. The donor-recipient pairs profiled exhibited considerable heterogeneity in hematological malignancy (acute myeloid leukemia, B-cell acute lymphoblastic leukemia, and multiple myeloma), donor subtype (matched related, matched unrelated, and haplo), and CMV donor serostatus. In addition, HSCT recipients in our cohort were treated with varied conditioning regimens and GVHD prophylaxis (corticosteroids, ATG, posttransplant cyclophosphamide), which differentially affect immune reconstitution after HSCT. Despite the heterogeneous conditioning and GVHD prophylactic regimens, the overall kinetics of immune reconstitution were broadly similar across patients, marked by early monocyte recovery and followed by expansion of lymphocyte subsets (Figure 1D). Further, CMV-associated immune responses appeared to mirror donor CMV serostatus rather than the specific immunosuppressive regimen administered. However, we cannot exclude the possibility that some phenomena observed between recipients reflect variation in immunosuppression within the complex post-HSCT clinical context. These findings should therefore be considered as hypothesis-generating and may not be readily generalizable to a broader allo-HSCT population. Validation in larger, more homogeneous cohorts stratified by clinical variables, such as immunosuppressive regimen, is warranted to corroborate these observations. Taken together, our findings highlight the heterogeneous immune processes behind the restoration of immunocompetency in HSCT recipients, with remarkable differences between recipients with CMVpos and CMVneg donors. Potent immunological responses were generated even in the absence of clinical CMV disease. CMV reactivation after HSCT remains associated with increased nonrelapse mortality and lower overall survival,3 and routine monitoring and preemptive treatment with antiviral agents like ganciclovir are standard clinical practice. However, antiviral therapy comes with the cost of significant myelotoxicity and impaired hematological reconstitution.36 Studies such as ours are the first step in characterizing markers that could be used to predict anti-CMV immunological responses and develop more targeted therapies. Conflict-of-interest disclosure: S.M. has received honoraria and research funding from Bristol Myers Squibb and Novartis; and research funding from Pfizer and Dren Bio (none related to this project). J.H. declares consultancy for Daiichi Sankyo (unrelated to this work). The remaining authors declare no competing financial interests.