Single-cell profiling identifies impaired adaptive NK cells expanded after HCMV reactivation in haploidentical HSCT

Haploidentical hematopoietic stem cell transplantation (h-HSCT) represents an efficient curative approach for patients affected by hematologic malignancies in which the reduced intensity conditioning induces a state of immunologic tolerance between donor and recipient. However, opportunistic viral infections greatly affect h-HSCT clinical outcomes. NK cells are the first lymphocytes that recover after transplant and provide a prompt defense against human cytomegalovirus (HCMV) infection/reactivation. By undertaking a longitudinal single-cell computational profiling of multiparametric flow cytometry, we show that HCMV accelerates NK cell immune reconstitution together with the expansion of CD158b1b2jpos/NKG2Aneg/NKG2Cpos/NKp30lo NK cells. The frequency of this subset correlates with HCMV viremia, further increases in recipients experiencing multiple episodes of viral reactivations, and persists for months after the infection. The transcriptional profile of FACS-sorted CD158b1b2jpos NK cells confirmed the ability of HCMV to deregulate NKG2C, NKG2A, and NKp30 gene expression, thus inducing the expansion of NK cells with adaptive traits. These NK cells are characterized by the downmodulation of several gene pathways associated with cell migration, the cell cycle, and effector-functions, as well as by a state of metabolic/cellular exhaustion. This profile reflects the functional impairments of adaptive NK cells to produce IFN-γ, a phenomenon also due to the viral-induced expression of lymphocyte-activation gene 3 (LAG-3) and programmed cell death protein 1 (PD-1) checkpoint inhibitors.


Introduction
Haploidentical hematopoietic stem cell transplantation (h-HSCT) is a life-saving therapeutic strategy to cure a wide range of hematologic malignancies (1,2). In particular, myeloablative (MA) and non-MA T cell replete h-HSCT protocols with posttransplant cyclophosphamide (PT-Cy) administration showed remarkably positive clinical outcomes in terms of overall survival and progression-free survival. However, the clinical benefits of h-HSCT are still hampered by life-threatening side effects including the occurrence of opportunistic viral infections as a consequence of the prolonged immune-deficiency after transplantation. In this scenario, acute and chronic human cytomegalovirus (HCMV) replication, either originating from a primary infection or from the reactivation of a preexistent latent virus in the graft or in the recipient, represents the most frequent adverse clinical condition, occurring in 35%-50% of patients who have undergone h-HSCT (3)(4)(5)(6)(7)(8).
Natural Killer (NK) cells are the first donor-derived lymphocytes that immune-reconstitute after h-HSCT, thus providing a prompt immune-surveillance against invading pathogens and tumor cells surviving despite conditioning regimens (9,10). Circulating NK cell subsets are defined by the surface expression of CD56 Haploidentical hematopoietic stem cell transplantation (h-HSCT) represents an efficient curative approach for patients affected by hematologic malignancies in which the reduced intensity conditioning induces a state of immunologic tolerance between donor and recipient. However, opportunistic viral infections greatly affect h-HSCT clinical outcomes. NK cells are the first lymphocytes that recover after transplant and provide a prompt defense against human cytomegalovirus (HCMV) infection/reactivation. By undertaking a longitudinal single-cell computational profiling of multiparametric flow cytometry, we show that HCMV accelerates NK cell immune reconstitution together with the expansion of CD158b1b2j pos /NKG2A neg /NKG2C pos / NKp30 lo NK cells. The frequency of this subset correlates with HCMV viremia, further increases in recipients experiencing multiple episodes of viral reactivations, and persists for months after the infection. The transcriptional profile of FACS-sorted CD158b1b2j pos NK cells confirmed the ability of HCMV to deregulate NKG2C, NKG2A, and NKp30 gene expression, thus inducing the expansion of NK cells with adaptive traits. These NK cells are characterized by the downmodulation of several gene pathways associated with cell migration, the cell cycle, and effector-functions, as well as by a state of metabolic/cellular exhaustion. This profile reflects the functional impairments of adaptive NK cells to produce IFN-γ, a phenomenon also due to the viral-induced expression of lymphocyteactivation gene 3 (LAG-3) and programmed cell death protein 1 (PD-1) checkpoint inhibitors.
JCI Insight 2021;6(12):e146973 https://doi.org/10.1172/jci.insight.146973 and CD16. CD56 bright /CD16 neg (CD56 br ) cells represent a regulatory subset accounting for about 10% of NK lymphocytes in peripheral blood (PB) able to secrete inflammatory cytokines and to regulate immune crosstalks. Terminally-differentiated CD56 dim /CD16 pos (CD56 dim ) NK cells represent the majority of these innate lymphocytes in PB exerting cytotoxic functions (11)(12)(13). In addition to these two conventional subsets, other NK cell subpopulations arise in response to pathologic stimuli. In this regard, HCMV replication promotes the expansion of CD56 neg /CD16 pos (CD56 neg ) NK cells that are otherwise almost undetectable under homeostatic conditions (6,14,15). CD56 neg NK cells were first reported as a highly impaired subset expanded in chronic HIV-1 and HCV infections (15)(16)(17)(18). Other studies later demonstrated that viruses greatly impact NK cell homeostasis and maturation by inducing the expansion of NK cells exhibiting adaptive traits and showing the strongest effector-functions (i.e., higher production of IFN-γ) when reencountering the same pathogen (19)(20)(21)(22). These adaptive NK cells have been extensively characterized in mice, where the binding of the Ly49H receptor to the murine cytomegalovirus glycoprotein m157 activates the NK cells to clear the virus (23). Unlike murine models, the precise receptor(s) binding HCMV antigens and the markers characterizing adaptive NK cells in humans are still being debated. Currently, human adaptive NK lymphocytes are defined as terminally-differentiated NKG2C pos cells lacking the expression of the inhibitory receptor NKG2A and expressing high levels of CD57 and killer immunoglobulin-like receptors (KIRs) (11,24). However, the appearance of adaptive NK cells does not necessary require the engagement of NKG2C, although this activating NK receptor (aNKR) exhibits specificity for the HCMV-encoded UL40 protein. Accordingly, the expansion of HCMV-induced KIR pos NK cells with adaptive traits has been also found in NKG2C-deficient subjects and in patients receiving cord blood grafts from NKG2C -/donors, suggesting that NKG2C is a marker rather than an essential mediator of this process. Hence, these findings indicate the presence of alternative NKRs and pathways involved in the generation of adaptive NK cells (25,26). In this regard, adaptive NK cells share with memory T cells an epigenetic reprogramming with genome-wide DNA methylation patterns responsible for higher IFN-γ production in response to HCMV infection (27)(28)(29). Moreover, and similar to memory T cells, adaptive NK cells can exhibit dysfunctional features, exhausted phenotypes, and increased levels of checkpoint inhibitors PD-1 and LAG-3 following chronic stimulation (24,(30)(31)(32)(33).
In a longitudinal setting, the present study characterizes the kinetics as well as the cellular and molecular features of NK cell subsets in response to in vivo HCMV infection/reactivation in recipients who underwent h-HSCT as therapy for hematologic malignancies.

Impact of HCMV infection/reactivation on the distribution and phenotype of immune-reconstituting NK cell subsets.
To investigate the impact of HCMV infection/reactivation on NK cell immune-reconstitution (IR) after h-HSCT, we longitudinally characterized the phenotype of viable CD14 neg /CD3 neg /Lineage neg lymphocytes within peripheral blood mononuclear cells (PBMCs) from both h-HSCT recipients and their relative HSC donors (Supplemental Figure 1; supplemental material available online with this article; https://doi.org/10.1172/jci. insight.146973DS1). To this end, h-HSCT recipients were stratified into two groups either experiencing (R) or not (NR) HCMV infection/reactivation, generally during the first two months after the transplant (Table 1  and ref. 3). We first observed that the absolute counts and frequencies of total NK cells did not significantly differ between R and NR recipients at any of the time-points analyzed ( Figure 1A). We and others have previously reported that donor-derived immune-reconstituting CD56 br and CD56 dim /CD16 neg (unCD56 dim ) NK cell subsets expand soon after h-HSCT and outnumber terminally-differentiated CD56 dim NK cells (10,34). We showed that HCMV infection/reactivation accelerated the maturation of NK cells by significantly decreasing the frequencies over time of CD56 br NK cells in R compared with NR. This phenomenon was counterbalanced by the HCMV-induced expansion, soon after HCMV infection/reactivation, of CD56 neg NK cells, a subset that is very low or undetectable in NR recipients ( Figure 1B).
To gain greater insight into the NK cell phenotypic features of NK cells associated with HCMV reactivation/infection, flow cytometry data were concatenated and analyzed with the PhenoGraph unsupervised clustering algorithm (35). The results obtained were visualized by using Uniform Manifold Approximation and Projection (UMAP) to simplify the visualization of marker expression at the single-cell level (Figure 2A).
This approach first confirmed that HCMV infection/reactivation modified the kinetics of NK cell subset IR and induced the expansion of CD56 neg NK cells, whose high frequencies persist for at least 1 year after h-HSCT (Figure 1, B  Through PhenoGraph analysis, we also identified 28 phenotypically distinct clusters of NK cells that were followed longitudinally before (1-2 months), soon after (3-4 months) and later after (8-12 months) the onset of this opportunistic viral infection. While the overall frequencies of the 28 NK cell clusters were similar between R and NR before HCMV infection/reactivation, the frequencies of the 5 distinct NK cell clusters (18,19,20,21,27) significantly increased early after the onset of this opportunistic infection at 3 to 4 months following h-HSCT and remained high even 1 year after the transplant ( Figure 2B and Figure 3A). These 5 clusters were mainly enriched within mature CD56 dim NK cells at all time points analyzed and expressed high levels of CD158b1b2j (KIR2DL2/2DL3/2DS2) and CD16, two surface markers of "licensed" human NK cells (11). Interestingly, the HCMV infection modified the phenotypic repertoires of mature NK cells belonging to the above-mentioned clusters by inducing the expression of NKG2C and by decreasing the surface levels of NKG2A, NKp30 and NKp46. The differential modulation of these NKRs have been associated with adaptive NK cells together with the increased expression of CD57 (11). Surprisingly, our data did not show any difference between R and NR in the surface levels of CD57 on NK cells comprised within these 5 clusters (Figure 3, B and C). Of note, our PhenoGraph analyses also showed that the expression of CD56 is decreased in R compared with NR. This phenomenon was due to the expansion of CD56 neg NK cells restricted to R and confirms our results obtained with manual gating and UMAP ( Figure 3C and Figure 1, B and C).
HCMV drives the expansion of long-lasting CD158b1b2j pos /NKG2A neg /NKG2C pos /NKp30 lo NK cells. After the identification of those NKRs differentially expressed on NK cells in response to HCMV infection/reactivation, we evaluated the kinetics of these receptors on total immune-reconstituting NK cells. Our results showed a statistically significant increase of CD158b1b2j and NKG2C together with a statistically significant decrease of NKp30, NKp46, and NKG2A starting from the third month after h-HSCT and soon after the onset of this opportunistic viral infection. Again, we did not observe any changes in the CD57 surface expression in R compared with NR. Interestingly, our data demonstrated that the amounts of NKp30 were significantly lower in R compared with NR even prior to clinically-recognized HCMV replication, and the degree of differential expression on NK cells between these two groups of recipients further increased following the onset of this opportunistic viral infection ( Figure 4A).
We then combined the kinetics of the above-mentioned NKRs in order to possibly identify a specific NK cell subset highly responsive to HCMV infection/reactivation. Although both NKp30 and NKp46 were significantly decreased on immune-reconstituting NK cells early after HCMV infection\reactivation, we focused our analysis only on NKp30 since this Natural Cytotoxic Receptor (NCR) showed the highest degree of differential expression between R and NR at 8-12 months after h-HSCT ( Figure 3, B and C). The manual gating on CD158b1b2j pos /NKG2A neg /NKG2C pos /NKp30 lo cells identified a potentially novel NK cell subset that arose soon after HCMV infection/reactivation, starting from the third month after h-HSCT and persisting at significantly higher levels in R compared with NR until 1 year after the transplant ( Figure 4B). As expected, CD158b1b2j pos /NKG2A neg /NKG2C pos /NKp30 lo NK cells were mostly comprised within the subset of mature CD56 dim and CD56 neg NK cells ( Figure 4C).
We then evaluated the clinical impact of HCMV viremia in the expansion of this NK cell subset, also taking into account that 9 out of 21 R recipients experienced multiple viral reactivation episodes (MR; Table 1). Our data showed that the frequency of CD158b1b2j pos /NKG2A neg /NKG2C pos /NKp30 lo NK cells positively correlated with HCMV viremia and was higher in MR compared with other R recipients (Figure 4D). Moreover, the percentages of this NK cell subset increased even more after every new HCMV reactivation event in MR, suggesting that these cells might remember and rapidly recall the previous viral challenges through a phenomenon of trained immunity ( Figure 4E).
Of note, each one of the 4 NKRs that characterizes this NK cell subset showed different kinetics during MR of HCMV. While the expressions of CD158b1b2j and NKG2C increased after each infection episode, surface levels of NKG2A and NKp30 significantly decreased after the second and third cycle of infection (Supplemental Figure 2).

CD158b1b2j pos NK cells in R patients show an adaptive transcriptional profile.
To assess if our newly disclosed NK cell subset was endowed with adaptive traits, we investigated its transcriptional profile by performing RNA-Seq. We FACS-sorted NK cells based on their expression of CD158b1b2j from recipients at 7-12 months after h-HSCT. We adopted this strategy to obtain a sufficient number of highly pure NK cells expressing CD158b1b2j that should identify those NK cell clusters highly enriched in R compared with NR ( Figure 3B and Figure 4A).
As expected, the analysis of NK cell subset distribution on FACS-sorted CD158b1b2j pos (KIR pos ) and CD158b1b2j neg (KIR neg ) NK cells demonstrated that KIR pos NK cells from R and NR had a lower frequency of CD56 br and unCD56 dim NK cells in comparison to KIR neg NK cells from R. Moreover, CD56 neg NK cells were mainly present among NK cells from R irrespective of the KIR pos and KIR neg status (Supplemental Figure 3).
The Principal Component Analysis (PCA) first revealed that KIR pos NK cells clustered separately from their KIR neg counterparts in R, thus indicating the existence of different gene expression profiles between them. Accordingly, 2,684 differentially expressed genes (DEGs) separated the two comparison groups (Supplemental We then focused our transcriptional analyses on NK cell metabolism given the recently reported key role of glucose-driven glycolysis and oxidative phosphorylation (OxPhos) in regulating NK cell cytotoxicity and generation of adaptive NK cells (38). Our results showed that KIR pos NK cells were enriched in pathways associated with mitochondrial respiration and ATP synthesis ( Figure 5C). Since OxPhos was the first enriched pathway of KIR pos NK cells within the HALLMARK database (NES = 3.8; FDR = 0.00), we analyzed the transcript levels of those genes included in this pathway. Succinate Dehydrogenase Complex Flavoprotein Subunit A (SDHA), involved in energy conversion and electron transport, was significantly upregulated in KIR pos NK cells. Moreover, similarly to long-living memory T cells, adaptive NK cells showed a preference for fatty acid oxidation (FAO) to fuel OxPhos (39,40). Indeed, CPT1A and TRAF6, which play a critical role in the development and the maintenance of the memory T population, were more highly expressed in KIR pos compared with KIR neg NK cells ( Figure 5D). Overall, this transcriptional profile indicates that KIR pos NK cells expanded in R are endowed with adaptive traits.
KIR pos NK cells from R patients show adaptive features upon restimulation. We then compared the transcriptional profiles of KIR pos NK cells between NR and R. The PCA results showed that KIR pos NK cells cluster differently in NR compared with R. Accordingly, 946 genes were significantly deregulated in the two groups, confirming that HCMV infection/reactivation affects the transcriptional profiles of NK cells (Supplemental Figure 4, C and D). The majority of DEGs belonged to cytokines-cytokine receptor interaction pathways and were enriched in NR (KEGG; NES = 2.08; FDR = 0.04). Again, we found a significant downregulation of IL2RB, IL18RAP, IL18R1, and IL12RB2 in R compared with NR (27). Furthermore, KIR pos NK cells from R downregulated several pathways relative to cell migration and genes associated with homing and chemotaxis (CXCR1, CXCR2, CCR1, and CMC1). KLRC1/NKG2A was one of the genes highly downregulated in R vs NR together with NCR3/NKp30, while KLRC2/NKG2C was upregulated in R versus NR (P value = 0.02 vs. P value adjusted [P-adj.] = 0.17). In contrast, the expression of B3GAT1/ CD57 was similar in KIR pos NK cells from NR and R ( Figure 6A and Supplemental Figure 5, A and B). Taken together, these data further confirmed our flow cytometry results ( Figure 3B). Our analysis also revealed a specific signature of KIR pos NK cells in R characterized by higher expression of CD2 coupled with a downmodulation of CD161, NKp80, and Siglec-7, similar to other clinical and experimental settings (25,26,41). As expected (27), genes coded for FcɛRγ, SYK, and PLZF were downregulated in R and these result have been further confirmed by semi-quantitative real-time PCR together with EWS-activated transcript 2 (EAT-2, also called SH2D1B) downregulation. As a result of this HCMV-induced epigenetic remodeling of IFN locus (29), we also found that IFNG was significantly upregulated in R compared with NR ( Figure 6A and Supplemental Figure 5C).
By using REACTOME and Gene Ontology (GO) databases, we confirmed that KIR pos NK cells in R are equipped to produce high amounts of IFN-γ and share pathways with activated T lymphocytes ( Figure  6B). Furthermore, the Weighted Gene Co-expression Network Analysis (WGCNA) dynamic tree cut on KIR pos NK cells from NR and R generated 8 distinct gene modules in a hierarchical clustering ( Figure 6C). We identified one module (depicted in blue) mostly related to features associated with of HCMV infection/ reactivation. In particular, the enrichment of signaling receptor activity (P-adj. = 0.0017) in GO together with MHC class II antigen presentation of REACTOME (P-adj. = 0.0022) databases further highlighted that KIR pos NK cells in R acquire traits of adaptive immunity. However, one of the main REACTOME-enriched pathways in the blue module is PD-1 signaling, thus indicating hallmarks of both cellular activation as well as exhaustion ( Figure 6D), similar to the T cell compartment (42)(43)(44).
Adaptive KIR pos NK cells in R patients are dysfunctional. Even though KIR pos NK cells in R hold adaptive traits, these overtrained cells also displayed a significant reduction in pathways related to mitochondrial respiration together with an enrichment of PD-1 signaling ( Figure 6D and Figure 7A). These features suggest that the sustained HCMV replication might be associated with an induced status of metabolic exhaustion and/or dysfunction (38). Indeed, the enrichment of cell-cycle pathways (REACTOME; NES = 2.05; FDR = 0.04) was restricted to KIR pos in NR, while in R they were characterized by a lower expression of the proliferation marker Ki67 both at the transcriptional (log 2 FC = -1.43) and cellular levels ( Figure 6A and Figure 7B). Our results further showed that KIR pos /NKG2C pos NK cells from R produce less IFN-γ than those from NR in response to target cells expressing NKG2C ligand ( Figure 7C).
In this context, the significant increase of PDCD1/PD-1, together with LAG3/LAG-3 expression in the KIR pos NK cells from R compared with NR, supported our working hypothesis of additional mechanisms of cell dysfunction associated with a higher expression of inhibitory checkpoints. We found a significant positive correlation between the transcript levels of PD-1 and LAG-3 that were expressed at their highest levels in R vs NR (Figure 7, D and E). Notably, PD-1 and LAG-3 are well-known for their important role in T lymphocytes where they are upregulated upon chronic stimulation and their blockade restores cell cytotoxicity (30,42). Therefore, we blocked either PD-1 or LAG-3 on NK cells cocultured with the HUVEC cell line expressing high levels of their ligands (Supplemental Figure 6 and ref. 45). Our results first confirmed that R-derived NK cells exhibited an impaired capacity to produce IFN-γ compared with NR-derived NK cells. The blockade of either PD-1 or LAG-3 induced a significant increase of IFN-γ production in R, which, however, did not reach the levels observed for NR ( Figure 7F). We also investigated the ability of FACS-sorted KIR pos NK cells from R and NR to control the HCMV infection in vitro. The results obtained demonstrated that KIR pos NK cells from NR have a greater ability to control the spread of HCMV in HUVEC cells. Furthermore, and in line with our previous data, the LAG-3/PD-1 blockade restore the ability of KIR pos NK cells from R to control the in vitro HCMV infection after 48 hours of coculture ( Figure 7G).
These results indicate that the dysfunction/exhaustion of "adaptive" KIR pos NK cells in R is mediated, at least in part, by the HCMV-induced expression of checkpoint inhibitors.
To assess the clinical relevance of our findings for h-HSCT recipients we stratified patients into two groups based on their individual frequencies of KIR pos NK cells ( Figure 4A). We thus set the threshold of KIR pos NK cells at 20%. Intriguingly, HCMV replication, and recurrent HCMV reactivation events in particular, occurred only in patients with a frequency of KIR pos NK cells higher than 20% (Table 2). We further calculate the cumulative incidence of acute and chronic graft-versus-host disease (GVHD) and the overall survival (OS). Six months cumulative incidence of grade II-IV acute GVHD was 48% versus 0% for patients whose KIR pos NK cell frequency was greater than 20% and less than 20%, respectively (P = 0.043, Figure 8A). Consistently, the 2-years cumulative incidence of chronic GVHD was 14% when KIR pos NK cells were greater than 20% versus 0% when KIR pos NK cell frequency was less than 20% (P = 0.196; Figure 8B). The 4-years OS was 61% and 71% for patients whose KIR pos NK cells were greater than 20% and less than 20%, respectively (P = 0.153, Supplemental Figure 7).

Discussion
The PT-Cy h-HSCT platform allows identification of a donor for almost every patient and induces immunologic tolerance between donor and recipient (46,47). However, due to the prolonged immunodeficiency after transplantation, the onset of opportunistic viral infections still significantly impairs the outcome of h-HSCT (2-5, 7, 8, 48). We show here that the occurrence of HCMV infection/reactivation in patients who underwent h-HSCT greatly impacts the homeostasis of NK cells by inducing the expansion of dysfunctional CD158b1b2j pos /NKG2A neg /NKG2C pos /NKp30 lo long-living lymphocytes endowed with adaptive traits and belonging to both CD56 dim and CD56 neg NK cell subsets.
The appearance of CD56 neg NK cells had been observed in response to several chronic viral infections in vivo, although the mechanisms of this phenomenon have never been fully recapitulated in vitro (15). Our longitudinal analysis from patients who underwent h-HSCT demonstrated that the occurrence of HCMV replication accelerates NK cell maturation by decreasing the frequencies of CD56 br NK cells and by inducing a significant expansion of CD56 neg NK cells. However, the onset of HCMV infection/reactivation does not affect the frequencies of immune-reconstituting CD56 dim NK cells, further supporting the hypothesis that CD56 neg NK cells likely originate from activated CD56 dim NK cells with which they share phenotypic similarities (17,20,(49)(50)(51).
Our unbiased computational approach at the single-cell level on flow cytometry data revealed that HCMV infection/reactivation deregulates the surface expression of several NKRs that, combined together, identify a potentially novel CD158b1b2j pos /NKG2A neg /NKG2C pos /NKp30 lo subset of mature NK cells. This population appears soon after the onset of this viral infection and persists for at least 1 year following the transplant. Surprisingly, and different from HCMV-seropositive healthy donors (HD), NK cells isolated from R do not have higher surface levels of CD57, a maturation marker widely associated with adaptive NK cells (11,21,24). These experimental findings indicate that the modulation of CD57 expression cannot be used in this in vivo h-HSCT clinical setting to track adaptive NK cells in response to HCMV infection/reactivation. We also show that CD158b1b2j pos /NKG2A neg /NKG2C pos /NKp30 lo NK cells belong to both CD56 dim and CD56 neg NK cells, further confirming the impact of this HCMV to accelerate NK cell terminal differentiation. In this context, the loss of CD56 might identify those "adaptive" NK cells that become impaired/exhausted in response to chronic inflammatory conditions and viral infections, as previously reported (15,17,20).
The expansion of CD158b1b2j pos /NKG2A neg /NKG2C pos /NKp30 lo NK cells directly correlates with HCMV viral loads and their frequencies are even higher in MR patients, thus indicating that viral rechallenges greatly influence NK cell homeostasis and promote the expansion of adaptive NK cells. Therefore, high frequencies of adaptive NK cells require increased levels of viral burdens to boost their responses against HCMV in immune-compromised patients during the first weeks following h-HSCT. Given the heterogeneity of the human adaptive NK cell phenotype, we then assessed the transcriptional profile of FACS-sorted KIR pos and KIR neg NK cells in R, considering that KIR pos NK cells are highly expanded in R even at 1 year after h-HSCT. The expression of KIRs as markers of licensed and adaptive NK cells in viral infections has been well demonstrated (11,21). We also compared the molecular fingerprints of KIR pos NK cells in R vs NR, since the latter cohort of recipients has lower but still adequate frequencies of circulating KIR pos NK cell. As expected (19,20), and in line with our flow cytometry results, KIR pos NK cells in R are characterized by the downregulation of KLRC1/NKG2A and NCR3/NKp30 and the upregulation of KLRC2/NKG2C when compared with their KIR neg counterparts. KIR pos NK cells also express higher levels of other KIRs, confirming that their higher degrees of maturation compared with KIR neg NK cells, also show a higher proportion of CD56 br and unCD56 dim NK cells.
Moreover, KIR pos NK cells show a more active metabolic status than KIR neg NK cells due to the enrichment of pathways associated with mitochondrial respiration. These results are consistent with recent studies uncovering the critical role of glycolysis and OxPhos in NK cell development, effector-functions, and generation of adaptive features (38). The upregulation of CTP1A and TRAF6 in KIR pos NK cells also suggests a preferential use of FAO to provide energy, exhibiting remarkable similarities to memory T cells (39,40,52). Finally, HCMV induces the downregulation of SIGLEC-7 in KIR pos NK cells, thus suggesting that the lack of this NKR might be associated with the acquisition of memory features (20).
The herein documented epigenetic reprogramming of KIR pos NK cells highlights once more the great impact of HCMV in shaping "adaptive traits" in human NK cells by inducing the deregulation of zinc finger proteins (i.e., ZBTB16, ZBTB32, ZBTB38) in KIR pos NK cells from R compared with both their KIR neg counterparts in R and KIR pos NK cells from NR. Of note, the reduction of ZBTB16/ PLZF is consistent with the consequent decrease of FCER1G, SYK, and cytokine receptors observed in KIR pos NK cells from R (27, 28). As a result of gene hypomethylation, KIR pos NK cells show a higher Although both flow cytometry data and transcriptional profiles clearly depicted the adaptive traits of KIR pos NK cells from R, their molecular fingerprints reveal a state of metabolic/cellular exhaustion that we did not observe in NR. Indeed, our data showed the downregulation in KIR pos NK cells from R of several pathways associated with cell migration and the cell cycle, while NK cell-mediated cytotoxicity pathways were enriched in KIR pos NK cells from NR. In line with these results, our functional experiments showed an impaired ability of adaptive KIR pos NK cells from R to produce IFN-γ compared with NR at 8-12 months after the transplant. In this context, a recent work showed that chronic stimulations of adaptive NK cells impairs their effector-functions and induces the expression of PD-1 and LAG-3 checkpoint inhibitors (33). In fact, and similar to what was reported for exhausted T cells (42), our results showed that KIR pos NK cells from R are characterized by increased levels of PDCD1 and LAG3 compared with their counterparts in NR or to KIR neg NK cells from R. Our masking experiments confirmed that the blocking of either PD-1 or LAG-3 increases the ability of NK cells from R to produce IFN-γ, although never reaching the levels of NK cells from NR, and to efficiently control HCMV infection in vitro. Taken together, these data highlight a remarkable dysfunction of adaptive NK cells due to HCMV-induced NK cell exhaustion and expression of inhibitory checkpoints. This functional impairment is reflected by a higher probability of experiencing a GVHD in patients with KIR pos NK cell expansion. Further investigations are required to understand how to properly expand more efficient and less exhausted adaptive NK cells in an in vivo human setting of viral rechallenge.
In summary, the present study sheds new light on the physiopathology of adaptive NK cells in response to chronic and multiple HCMV infections and paves the way to develop an immune-therapeutic and NK cell-based approach to improve the control of opportunistic viral infections in h-HSCT.

Methods
Patient recruitment. In this study, we enrolled 48 patients (Table 1) affected by hematologic malignancies and treated according to our published h-HSCT protocol (4). Patients were recruited at the Hematology and Bone Marrow Transplant Unit, Humanitas Cancer Center, Humanitas Research Hospital, Rozzano, Milan, Italy. Blood samples were collected from both donors and recipients before the transplantation, and from the recipients every month until 1 year after h-HSCT. The PBMCs were isolated from patients and relative HDs (54).
According to our institutional protocol, patients who had undergone h-HSCT were monitored for the HCMV infection/reactivation by assessing the HCMV viral load through real-time PCR (CMV R-GENE, Argene, Biomérieux) in the PB of both donor and recipients before and every week after the transplant to identify those patients experiencing HCMV infection/reactivation. Patients showing a viral load higher than 2,000 IU/mL, at least at one time point after the transplant, were considered as HCMV-reactivated and were treated with preemptive therapy (Foscarnet 90 mg/kg or Ganciclovir 5 mg/ kg and/or Valganciclovir, twice a day for 2 weeks) when the viral load exceeded 4,000 IU/mL.
Polychromatic flow cytometry and cell sorting. All the flow cytometry experiments were performed in batch on frozen cells to minimize variability. Frozen cells were thawed and stained as we previously described (10). The anti-human monoclonal antibodies (mAbs) used for the analysis of NK cells of both patients who had undergone h-HSCT and relative HDs are listed in Supplemental Table 1.
Samples were acquired at BD FACSymphony A5 flow cytometer for the analysis of NK cell phenotype, while FACS experiments were performed at BD FACS Aria III (both from BD Bioscience).
Unsupervised high-dimensional flow cytometry data analysis. Flow Cytometry Standard (FCS) 3.0 files were imported into FlowJo software (version 9.9.6., TreeStar), and analyzed by standard manual gating strategy in order to remove debris, doublets, and dead cells, as well as to identify NK cells (Supplemental Figure 1A). Two thousand NK cells per sample were imported in FlowJo (version 10.2.0), biexponentially transformed and exported for further analysis in Python (version 3.7.3) using a custom pipeline with PhenoGraph as methods for clustering (http://github.com/luglilab/Cytophenograph, where we modified the linux-community and the core.py script of the PhenoGraph package in order to fix the seed to "123456"). Each sample was labeled with a barcode for further identification, converted into comma separated (CSV) files and concatenated in a single matrix by using the "concat" function of the "pandas" package.
The K-value, indicating the number of nearest neighbors identified in the first iteration of the algorithm, was set at 45. The data were then reorganized and saved as new CSV files, 1 for each cluster, that were further analyzed in FlowJo (version 10.2.0) to determine the frequency of positive cells for each marker and visualized in a heatmap (Figure 2A). Clusters representing less than 0.5% were excluded from the following analysis. Subsequent metaclustering of frequencies was performed using the gplots R package (35). Finally, PhenoGraph clusters were visualized using the UMAP technique.
Total RNA was purified with the MicroRNAeasy KitTM with DNAse (Qiagen), according to the manufacturer's instructions and quantified by a Nanodrop 2000 (Thermo Fisher Scientific). Purified RNA was reverse-transcribed using a High Capacity cDNA Reverse transcription Kit (Thermo Fisher Scientific).
RNA-Seq. After RNA extraction, a RNA quality control was performed with the 4200 Tape Station system (Agilent). Only RNAs having an RNA Integrity Number (RIN) greater than 6 were used for library preparations. Total-RNA-Seq library preparation was performed starting from 0.5 ng of total-RNA with the SMART-Seq Stranded Kit (Clontech-Takara). Libraries obtained were qualitatively assessed by using TapeStation 4200 (Agilent) and quantified by Qubit Fluorimeter (Thermo Fisher Scientific). Afterward, they were multiplexed in equimolar pools and sequenced on a NextSeq-550 Illumina Platform generating at least 60 million 75bp-paired-end reads per sample.
RNA-Seq data are available at GEO under accession number GSE160362. RNA-Seq data analysis. After quality control, reads were aligned to the reference genome (GRCh38. p12) using the STAR aligner with default parameters (version 2.7.0). Gene-based read counts were then obtained using feature count (55) module and GENCODE v29.gtf annotation (56). The read counts were imported into R statistical software, and differential gene expression analysis was performed using the Deseq2 package (57). Raw read counts were normalized using the Relative Log Expression method (RLE), and genes with low counts were filtered out by setting the parameter "independence filtering" as True. P values were adjusted using the Benjamini-Hochberg method.
Differential gene expression analysis. Genes were considered differentially expressed with FDR less than 0.05, and the PCA plot was generated using the plotPCA function of the Deseq2 package.
Functional enrichment analysis was performed using GSEA (version 3.0) software (Broad Institute of MIT) and gene list-ranked based on log 2 -fold changes. The gene set enrichment analysis was conducted in JCI Insight 2021;6(12):e146973 https://doi.org/10.1172/jci.insight.146973 the preranked mode with scoring scheme "classic" and 1,000 permutations. The maximum gene set size was fixed at 5,000 genes and the minimum size was fixed at 10 genes. The gene signature was retrieved from the H (hallmark) and C2 (curated) collections (h.all.v6.2.symbols.gmt) of the Molecular Signatures Database (MSigDB v6.2). For GSEA only pathways with a |NES| greater than or equal to 1.8 and an FDR q value less than or equal to 0.05 were selected (58).
Gene correlation analysis and module detection were computed with the WGCNA R package (version 1.69; ref. 59).
The EndoGRO HUVECs (MilliporeSigma) were cultured in endothelial cell growth medium-2 (EGM-2) (Lonza) supplemented with 5% FBS, 1% Penicillin-Streptomycin, and 1% Ultra Glutamine in cell flasks precoated with Collagen I (Corning) at a density of 10 5 to 10 6 cells/mL in a humidified atmosphere at 37°C with 5% CO 2 . The HUVECs were treated with IFN-γ (100 ng/mL) for 3 days to induce the expression of PD-L1 and HLA-II and further used as target cells in functional assays.
After 4 hours of coculture, intracellular IFN-γ expression on NK cells was assessed by flow cytometry (Supplemental Table 1).
In vitro HCMV infections. Infection of HUVECs, previously incubated with IFN-γ for 3 days as detailed above, was performed at multiplicity of infection (MOI) of 1 by using the bacterial artificial chromosome (BAC) clone of the viral strain TB40/E expressing enhanced green fluorescent protein (EGFP; refs. 61, 62).
In parallel, KIR pos NK cells FACS-sorted from PBMCs of NR and R recipients at 8-12 months after h-HSCT and left overnight in RPMI complete medium supplemented with rhIL-2 (100 IU/mL) and rhIL-12 (100 IU/mL) at 37°C in 5% CO 2 . The next day, purified KIR pos NK cells (150,000 cells/mL) were incubated or not in the presence of α-PD-1 or α-LAG-3 blocking antibodies and cocultured for 48 hours with infected HUVECs in RPMI complete medium supplemented with rhIL-2 (200 IU/mL). The images were acquired to the DMi8 Wide-Field Microscope (Leica). Four representative fields for each well were analyzed and fluorescent-infected HUVECs were counted by using ImageJ software (NIH).
Statistics. Statistical analysis was performed using GraphPad PRISM (version 7.0) and R statistical software. An unpaired, 2-tailed Student's t test was used to compare the different variables among samples. A paired, 2-tailed Student's t test was used to compare the different variables within the same sample. Oneway ANOVA with Bonferroni correction was used to perform multiple comparisons. Results are shown as the mean ± standard deviation (SD) or as the mean ± SEM when repeated sampling was considered.
Categorical variables were expressed as numbers, and comparisons between groups were performed using a χ 2 test. Cumulative incidences of acute and chronic GVHD were constructed in the competing risks framework considering death without GVHD as competing events. The difference between cumulative incidence curves in the presence of a competing risk was tested using the Gray method. OS was estimated using the Kaplan-Meier method and the log-rank test was used for between-groups comparisons.
Study approval. This study has been approved by the Institutional Review Boards of the Clinical and Research Institute Humanitas (Approval 24/18). All patients and donors signed a written informed consent form in accordance with the Declaration of Helsinki.