Introduction

Autologous hematopoietic stem cell transplantation (autoHSCT) is a widespread procedure used to treat many different types of cancer, including hematological tumors in adults and solid tumors in children. It is a relatively common therapeutic tool to combat not only hematological malignancies, such as multiple myeloma (MM) or non-Hodgkin lymphoma (NHL), but also for some pediatric solid tumors, such as neuroblastoma, and might be a salvage option for other tumors like lymphomas and sarcomas1,2,3. Early immune reconstitution following autoHSCT is associated with a better outcome in a variety of cancers4,5,6. Day 15 absolute lymphocyte count (ALC-15) of ≥500 cells/µl after autoHSCT is an independent prognostic indicator in MM and NHL patients. Natural killer (NK) cells are a relevant lymphocyte subset in ALC-15 that affects the outcome following autoHSCT, as they are the first lymphocyte subset reaching normal circulating levels after HSCT5,7,8.

NK cells can directly kill target cells through different mechanisms, and they have a central role in the defense against virus-infected and cancer cells9. Human circulating NK cells have been classified into two major subsets with different functionalities, depending on the expression of CD56 and CD16: CD56brightCD16low/− (CD56bright) and CD56dimCD16+ (CD56dim). CD56bright NK cells produce large amounts of immunomodulatory cytokines and chemokines but have low cytotoxicity unless they are activated by cytokines. On the other hand, CD56dim NK cells are more cytotoxic but produce lower amounts of cytokines10,11. NK cells express activating and inhibitory receptors whose integrated signals will determine their response12. In this manner, they do not get activated when they recognize autologous determinants such as class I human leukocyte antigens (HLA-I) through inhibitory killer immunoglobulin-like receptors (KIRs) and CD94/NKG2A receptor. On the other hand, NK cells are activated when they interact with virus-infected or cancer cells, either by the direct detection of upregulated stress-induced self-molecules or via antibody-dependent cell-mediated cytotoxicity13,14. Given the potent anti-tumor effect of NK cells and their early recovery after autoHSCT, studying their biology and reconstitution in this setting is of utmost interest. Studies are scarce, and they are mostly focused on adult patients. They show that following autoHSCT, and shortly after leukocyte recovery, there is a redistribution of NK cell subsets. For example, in patients with MM and lymphoma, it has been described that there is an increased frequency of immature CD56brightNKG2A+ NK cells, expressing high levels of KIRs and CD57, early after transplant15. Another MM study showed a similar redistribution of NK cell populations characterized by a high proliferative rate and an increase in the frequency of both CD56bright NK cells as well as the most immature population within the CD56dim NK subset (CD57-NKG2A+)16. In general, NK cell function recovers early after autoHSCT15,16, and transcriptomic analyses have revealed profound changes related, among others, to cell cycle, DNA replication, and the mevalonate pathway17. Interestingly, following autoHSCT in MM patients, there is an expansion of a CD9+ decidual-like NK cell subset that is characterized by high granzyme B and perforin expression levels and exhibits some altered effector functions (degranulation and MIP1β production)17. Nevertheless, the factors responsible for this CD9+ NK cell subset expansion are not known.

Specific gene signatures have been associated with survival in cancer patients18. In addition, the number of NK cells and the frequency of several subsets are associated with overall and progression-free survival (PFS) in MM and NHL patients following an autoHSCT8,19. For example, it has been shown that MM patients with lower frequencies of the mature highly differentiated NKG2A-CD57+ NK cell subset after autoHSCT had a better outcome16. Other researchers have found that MM patients in long-term complete response after autoHSCT exhibit a higher frequency of NK cells expressing the inhibitory receptors KIR2DL1 and NKG2A and reduced numbers of NKp46+ NK cells20. Interestingly, as in allogeneic HSCT (alloHSCT), there could be a NK cell-mediated graft-versus-tumor (GvT) effect in autoHSCT, that is related to the inhibitory KIR-HLA-I receptor-ligand mismatch and affinity interactions between them21,22,23,24,25. Indeed, KIR and HLA-I genotypes have shown to have a significant influence on neuroblastoma patients undergoing autoHSCT and anti-GD2 treatment22,26.

Despite the advances in adults, in pediatric patients, there is still a substantial need to understand the complexity of the immune system reconstitution after autoHSCT. In this context, an in-depth characterization of the NK cell lymphocyte subset, exploring the biology and physiopathology of NK cells in pediatric patients, would be of great value. In the present study, we have found that following autoHSCT, there are significant changes in the expression of inhibitory and activating receptors, as well as in activation markers, together with a transient expansion of NK cells with a decidual-like phenotype. Furthermore, plasma levels of soluble molecules that are relevant for NK cell differentiation, survival, and function were analyzed. Changes in the levels of interleukin 15 (IL-15) in combination with transforming growth factor beta 1 (TGF-β1 or TGF-β hereafter) may explain the acquisition of the decidual-like phenotype. Also, analyses of maturation markers and KIR expression indicate that NK cell education and maturation are uncoupled processes. Altogether, our findings can be of relevance to understand the biology and physiopathology of NK cells during their reconstitution following autoHSCT and, therefore, help to design better therapies for children with cancer.

Methods

Patients’ characteristics and study design

Blood samples in EDTA-containing tubes from 13 children (male: 8; female: 5; age range: 3–11 years) diagnosed with different cancers and that received autoHSCT were collected for this study at Cruces and Donostia University Hospitals in the Basque Country. Patients’ clinical features are described in Table 1. Samples were obtained at five different time points: pre-transplantation (S1), after reaching leukocyte recovery: more than 1000 leukocytes/µl, usually around day 12 after autoHSCT (S2), 30 days after autoHSCT (S3), 100 days after autoHSCT (S4), and 180 days after autoHSCT (S5). Blood samples from 8 adult healthy donors (male: 5; female: 3; age range: 25–62 years) were also collected. Sample collection was carried out through the Basque Biobank for Research (https://www.biobancovasco.bioef.eus), which complies with the quality management, traceability, and biosecurity set out in the Spanish Law 14/2007 of Biomedical Research and in the Royal Decree 1716/2011. The study was approved by the Basque Ethics Committee for Clinical Research (BIO14/TP/003, PI+CES+INC-BIOEF 2017-03) and the Cruces University Hospital Ethics Committee (E23/78). All adult participants, along with the parents or legal guardians of pediatric patients, provided written and signed informed consent to participate in this study, in accordance with the Declaration of Helsinki.

Table 1 Patients’ clinical characteristics
Full size table

Sample preparation

Whole blood samples from cancer patients were collected to obtain plasma and PBMCs. For in vitro experiments, PBMCs were isolated from buffy coats and whole blood samples from eight adult healthy donors. Plasma was obtained by centrifugation, while PBMCs were enriched using density gradient centrifugation. The isolated PBMCs were cryopreserved in heat-inactivated fetal bovine serum (FBS) (GE Healthcare HyClone) supplemented with 10% dimethyl sulfoxide (Thermo Scientific) and stored in liquid nitrogen.

Cryopreserved PBMCs were thawed at 37 °C in a water bath and washed twice with RPMI 1640 medium with L-Glutamine (Lonza). After that, cells were incubated for 1 h at 37 °C and 5% CO2 with 10U DNase (Roche) in R10 medium (RPMI 1640 medium containing GlutaMAX, 10% FBS, and 1% Penicillin-Streptomycin, from Thermo Fisher Scientific). Following incubation, the cells were washed once, resuspended in NK cell medium (RPMI 1640 medium with GlutaMAX, 10% FBS, 1% P-S, 1% MEM Non-Essential Amino Acids Solution, and 1% Sodium Pyruvate, from Thermo Fisher Scientific), filtered through 70 µm cell strainers, and counted. The processed PBMCs were then used in flow cytometry or in vitro experiments.

NK cell absolute number quantification

Whole blood samples were used to quantify the absolute number of NK cells. NK cells were identified as CD45+, CD3−, CD56+ and/or CD16+ lymphocytes using the following clinical-grade fluorochrome-conjugated monoclonal antibodies (mAbs): FITC anti-CD16 (CLB/FcGran1), PE anti-CD56 (MY31), PerCP-Cy5.5 anti-CD3 (SK7), and V450 anti-CD45 (2D1), all from BD Biosciences. The absolute number of lymphocytes was obtained from the hemogram. The absolute number of NK cells per microliter of blood was calculated using the following formula: (percentage of NK cells within the lymphocyte gate × absolute number of lymphocytes)/100.

DNA extraction, HLA, and KIR genotyping

Following the manufacturer´s recommendations, DNA was obtained from PBMCs using the FlexiGen DNA kit (Qiagen). The first step is the addition of lysis buffer to each sample following the Qiagen FlexiGene DNA procedure flowchart. Briefly, cell nuclei and mitochondria were pelleted by centrifugation and resuspended in denaturation buffer containing QIAGEN Protease. Following protein digestion, DNA was precipitated by the addition of isopropanol, recovered by centrifugation, washed in 70% ethanol, and dried. DNA was resuspended in hydration buffer and stored at −20 °C for further use.

HLA-A, B, and C loci typing (necessary to discriminate the KIR ligands present in each patient) was performed using the following kits: LIFECODES HLA-A, B SSO Typing Kit and LIFECODES HLA-C eRES SSO Typing Kit (Lifecodes), all of them CE marked. Lifecodes HLA uses PCR-SSO methodology, subsequent data input on the Luminex® 200™ (Merck) analyzer, and results were examined using Matchit DNA software. HLA genotyping was performed at a level of resolution that allowed determining the KIR-binding epitope to distinguish Bw4 specificities and the HLA-C dimorphism at position 80 of the α1 helix. In situations of ambiguity when assigning alleles or to rule out more frequent null alleles, NGS typing was also performed. The NGSgo-MX6 multiplex amplification strategy was used, followed by sequencing on the Illumina Myseq platform and analysis by the NGS engine program.

KIR typing was performed by a PCR-SSP method (sequence-specific primers) using the KIR Ready gene kit (Inno-train Diagnostik GmbH). PCR products were amplified and subsequently resolved on agarose gels, and results were interpreted according to the manufacturer’s instructions. Based on KIR gene content, haplotype (A or B) was defined, and genotypes (AA or Bx, where x can be A or B) were grouped for each patient. The AA genotype is homozygous for inhibitory haplotype A (consisting of 3DL3, 2DL3, 2DP1, 2DL1, 3DP1, 2DL4, 3DL1, 2DS4, and 3DL2). Haplotype B or Bx genotype includes any combination of KIR other than the above.

Determination of cytokines in plasma

Plasma samples were obtained from blood samples and stored at −80 °C until they were required. For the measurement of IL-15, PDGF-DD, and PDGF-BB plasma levels, the human IL-15, PDGF-DD, or PDGF-BB Quantikine ELISA Kits (R&D Systems) were used, respectively, following the manufacturer’s recommendations. The optical density was determined using a Varioskan Flash fluorimeter (Thermo Fisher Scientific), and the standard curve and non-linear regression, and log-log line model were performed in GraphPad Prism software. For the measurement of TGF-β plasma levels, Luminex MILLIPLEX TGF-beta 1 Single Plex MAGNETIC Bead Kit (Merck) was used, following the manufacturer’s recommendations. Levels of TGF-β were quantified using Luminex® 200™ (Merck) and analyzed using xPONENT® software. For the determination of GDF-15 and placental growth factor (PlGF) plasma levels, Elecsys GDF-15 and Elecsys PlGF kits (Roche) were used, following the manufacturer’s recommendations. For quantification, the electrochemiluminescence was determined in a Cobas e 801 analytical unit (Roche) immunoassay analyzer.

NK cell culture with cytokines

For in vitro experiments, 1.5 × 106 PBMCs/mL were cultured in a 48-well plate in 1 mL of NK cell culture media (described above) with different combinations of IL-15 (10 ng/mL) and TGF-β (1–10 ng/mL) cytokines. We performed the following conditions: no cytokines (unstimulated), IL-15, 5 ng/mL TGF-β, IL-15 + 1 ng/mL TGF-β, IL-15 + 5 ng/mL TGF-β and IL-15 + 10 ng/mL TGF-β. Cells were cultured for 7 days, and the media was refreshed at day 4. For the scRNA-seq experiment, PBMCs were cultured in NK cell media without cytokines or with a combination of IL-15 (10 ng/mL) and TGF-β (5 ng/mL) for 4 days.

scRNA-seq

Four single-cell RNA libraries were constructed using a Chromium controller (10x Genomics, Pleasanton, CA, USA) as per the manufacturer’s instructions (10x Genomics Chromium Single Cell 3′ V3.1 chemistry). A total of 12 cDNA amplification cycles and 14 cycles of library amplification were performed. Sequencing was carried out using a Novaseq 6000 SP Reagent Kit v1.5 (Illumina, San Diego, CA, USA; 100 cycles). FASTQ sequence data were aligned to a custom hg38 (GRCh37, CellRanger reference genome version References- 2020-A, build GRCh38.p13) using CellRanger count (v7.2.0).

Gene counts were loaded into R (R-project.org/, v4.3.1) using Seurat (v5.0.1). Analysis was conducted with Seurat and extensive use of dplyr (v1.1.4) and tidyverse (v2.0.0) functionality. For each of the four libraries analyzed, we filtered poor-quality cells by requiring each cell to have at least 800 expressed genes (of at least one count), at most 40,000 reads (removing cell doublets), and at most 10% mitochondrial RNA (removing dying cells). We then proceed with the normalization, variable feature identification, scaling of the data, and principal component analysis (PCA) of each of the libraries. We next integrated the libraries with the CCA integration method using the IntegrateLayers function in Seurat. Uniform manifold approximation and projection (UMAP) dimensionality reductions were produced, and clusters were computed with the FindNeighbours function, followed by FindClusters (resolution of 0.6), resulting in 18 clusters. In addition, the phateR (v1.0.7) package was used to calculate 2-dimensional PHATE embeddings from the normalized data.

For the NK cell population, we further annotated cell type clusters of NK cells (clusters 7 and 9) by sub-setting the Seurat object to these cells only, and then repeating the procedure described above, but with only 20 dimensions for the UMAP and clustering. With this, NK cells were further divided into 9 clusters representing NK1, NK2, NK3, and NK4 cells, using expression patterns and marker gene lists. For each NK cell (sub)type, we compared gene expression levels between cells untreated and treated with IL-15 + TGF-β, using the FindMarkers function in Seurat on the normalized counts (Seurat RNA Assay). UMAP and PHATE cell embeddings were visualized using the function Dimplot in Seurat. Gene expression was visualized using Featureplots and dotplots in Seurat.

Flow cytometry

For flow cytometry analysis of NK cells, PBMCs were first stained with LIVE/DEAD Fixable Aqua Dead Cell Stain Kit (Invitrogen) reagent to exclude dead cells, following the manufacturer´s recommendations. Then, cells were washed with PBS containing 2.5% bovine serum albumin (BSA, Sigma-Aldrich) before performing surface receptor staining. For that, cells were incubated for 30 min on ice, in the dark, with the fluorochrome-conjugated mAbs summarized in Tables S1–S4 (for phenotypical analysis, three different panels were used, numbered from 1 to 3 in Tables S1–S3, and an additional panel for in vitro experiments in Table S4). After staining, cells were then washed again with PBS containing 2.5% BSA and resuspended in PBS. Samples were acquired in a LSR Fortessa X-20 flow cytometer (BD Bioscience).

Flow cytometry data analysis

Flow cytometry data were analyzed using FlowJo v10.8.1. software. Manual and automated analyses were performed. The following plug-ins were used: DownSample (1.1), UMAP, t-SNE, and FlowSOM (2.6). Briefly, for the automated analysis, events were first downsampled from the gate of interest (NK cells) across all samples using the DownSample plug-in and concatenated. For each donor, NK cells were downsampled to 157 cells in the case of the FlowSOM of Fig. 2 and 1510 cells in the FlowSOM of Fig. 6. Then, downsampled populations were concatenated for the analysis. FlowSOM was run using the indicated parameters in each figure.

Statistics and reproducibility

Statistical analyses were performed as described in the figure legends. GraphPad Prism v.9.3.1 was used for graphical representation and statistical analysis. Non-parametric Wilcoxon matched-pairs signed-rank test was used to determine significant differences among groups, and Mann–Whitney unpaired test was used to determine significant differences among non-paired data. The Rout test was used in the case of the HLA-DR marker in order to identify outliers in the CD56neg population.

Correlation plots between different variables were calculated and visualized as correlograms. The Pearson correlation coefficient was indicated by square size and heat scale. These analyses were performed with R (version 4.3.1): A language and environment for statistical computing (R Foundation for Statistical Computing, Vienna, Austria).

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Results

NK cell counts, subset frequencies, and maturation status after autoHSCT

NK cell phenotype changes during the immune system reconstitution after autoHSCT8,15,16,19. In this work we have studied the NK cell reconstitution in a cohort of 13 pediatric patients with cancer undergoing autoHSCT using blood samples taken at different time points: before autoHSCT (S1), after reaching leukocyte recovery (>1000 leukocytes/µl, usually around day 12 after autoHSCT, S2) and 30 days (S3), 100 days (S4), and 180 days (S5) after autoHSCT. Patients’ clinical characteristics are described in Table 1. After autoHSCT, absolute number of NK cells did not significantly change, although a reduction was noticed from S1 to S2, followed by a sustained increase until S5 (Fig. 1a). The percentage of NK cells within lymphocytes did not change during the reconstitution. To characterize NK cells, we defined them by the expression of NKp80 and CD56, and the negative expression of other cell markers (CD3, CD14, CD19, and CD123), as it is shown in Fig. S1. We also defined 3 different NK cell subsets based on the expression of the CD56 marker: CD56bright, CD56dim, and CD56neg27, and we observed that after autoHSCT, their percentage did not overly fluctuate (Fig. 1b).

Fig. 1: NK cell numbers, subsets, and maturation status after autoHSCT.
figure 1

Boxplot graphs showing the analysis of NK cells at the studied five time points: before autoHSCT (S1), after reaching leukocyte recovery (more than 1000 leukocytes/µl, usually around day 12 after autoHSCT) (S2), 30 days (S3), 100 days (S4), and 180 days (S5). a Absolute number of NK cells (left) and percentage of NK cells within lymphocytes (right). Each dot represents a patient (S1 n = 6; S2 n = 6; S3 n = 6; S4 n = 5; S5 n = 3). b Percentage of NK cell subsets based on CD56 expression, including CD56bright, CD56dim and CD56neg NK cells. c–e Analysis of NKG2A and CD57 receptors expression. c Percentage of NKG2A-expressing NK cells. d Percentage of CD57-expressing NK cells. e Percentage of NKG2A+CD57− (gray), NKG2A+CD57+ (yellow), NKG2A−CD57+ (green), NKG2A−CD57− (blue) cells within total NK cells, and CD56bright, CD56dim, and CD56neg NK cell subsets (left) and f representative contour plots showing the four analyzed populations (right). Boxplots show the median and 25–75th percentiles, and the whiskers denote the lowest and highest values. Each dot represents a patient (S1 n = 12; S2 n = 11; S3 n = 11; S4 n = 8; S5 n = 8 unless otherwise specified). The significance of the data was determined by comparing each sample with sample S2. *p < 0.05, **p < 0.01 (Wilcoxon matched-pairs signed-rank test). Exact p values are provided in a file named Supplementary Data 2, under Fig. 1.

Full size image

The frequency of NKG2A+ NK cells increased early after transplantation (from S1 to S2) (Fig. 1c). In contrast, the frequency of CD57+ NK cells increased at later time points (>100 days, S4 and S5) (Fig. 1d). Also, the co-expression of NKG2A and CD57 was analyzed after autoHSCT for all NK cell subsets. As expected, the results showed that, not only the most immature CD56bright subset were mostly NKG2A+CD57−, but also that these NKG2A+CD57− cells represent the most frequent subset within the total NK cells in these children (Fig. 1e). Overall, we observed that after autoHSCT there is a gradual decrease in the frequency of immature NK cells (NKG2A+CD57−), while the percentage of terminally differentiated (NKG2A−CD57+) NK cells increased. Nevertheless, the more immature subset (NKG2A+CD57−) is the predominant at every analyzed time point (Fig. 1e, f).

NK cell receptor repertoire is altered following autoHSCT

It was previously reported that genes associated with activation, cytotoxic functions, and several receptors were differentially expressed in adult MM patients after autoHSCT17. Hence, we decided to study the expression of these and other molecules in NK cell subsets in our pediatric cohort. The activating co-receptor and activation marker CD3828 is practically expressed by all NK cell subsets, and its expression increased early after autoHSCT (S2), while from day 30 (S3) onwards, the expression gradually decreased to the initial levels (Fig. 2a). Similarly, autoHSCT induces a transient increase of NK cells expressing HLA-DR and CD26 (Fig. 2a), which have been described as functionally more activated cells29,30. On the other hand, regarding the cytotoxicity-associated marker CD831, the transient increase at S2 was only observed in the CD56bright NK cell subset (Fig. 2a).

Fig. 2: NK cell receptor repertoire following autoHSCT.
figure 2

Analysis of NK cells in a cohort of pediatric patients before (S1) and following autoHSCT: after reaching leukocyte recovery (more than 1000 leukocytes/µl, usually around day 12 after autoHSCT) (S2), 30 days (S3), 100 days (S4), and 180 days (S5). a Boxplot graphs showing the median fluorescence intensity (MFI) of CD38 expression and the percentage of cells expressing HLA-DR, CD26, and CD8 within total NK cells and NK cell subsets based on CD56 expression, including CD56dim, CD56bright, and CD56neg NK cells. b MFI of CD31, LAIR-1 and CD55 expression. c Percentage of CD160+ cells and MFI of CD229 and 2B4 expression. Boxplots show the median and 25–75th percentiles, and the whiskers denote the lowest and highest values. Each dot represents a patient. A minimum of 10 cells were analyzed in each cell subset. S1 n = 12; S2 n = 11; S3 n = 11; S4 n = 9; S5 n = 8 for total, CD56bright and CD56dim NK cells. S1 n = 11; S2 n = 9; S3 n = 8; S4 n = 9; S5 n = 7 for CD56neg NK cells, except for the 2B4 marker (S1 n = 12; S2 n = 9; S3 n = 11; S4 n = 9; S5 n = 7). d UMAP projection of CD56+ NK cell populations (Pop) identified by the FlowSOM clustering tool for the specified markers, fluorescence intensity of each Pop as indicated in the column-scaled z-score, and bars graph showing the percentage of each Pop at the studied five time points (S1–S5). Results were compared with the S2 time point using the Wilcoxon matched-pairs signed-rank test. *p < 0.05, **p < 0.01 and ns no significant. Exact p values are provided in a file named Supplementary Data 2, under Fig. 2.

Full size image

In addition, results also showed that NK cells expressed higher levels of CD31 and LAIR-1 inhibitory receptors at S2 compared to other time points (Fig. 2b). CD55, a marker expressed at higher levels on CD56bright NK cells32, followed the same trend (Fig. 2b). However, regarding activating receptors, the percentage of CD160+ NK cells decreased at S2, as well as the expression of CD229 and 2B4 (also known as CD244) (Fig. 2c). Lastly, the expression of other receptors, such as Siglec-7, TIM3 and CD226 (DNAM-1) did not significantly change after transplantation (Fig. S2a).

We performed an unsupervised analysis using UMAP and FlowSOM clustering methodology (Fig. 2d). Among CD56+ NK cells, the analysis revealed 8 populations (or Pops) based on the expression of 10 cell surface markers (CD8, CD26, CD31, CD38, CD55, CD160, CD226, CD229, HLA-DR and LAIR-1). We observed a significant increase in the frequency of Pop 7 at S2 corresponding to NK cells expressing high levels of CD26, CD55, HLA-DR, CD38, CD31, and LAIR-1, and low levels of CD229 and CD160. We also observed a significant decrease of Pop 1 at S2 corresponding to NK cells which express high levels of CD229 and CD160, and low levels of CD26, CD55, HLA-DR, CD31, and LAIR-1. Altogether, these results indicate that, early after autoHSCT, NK cells are more activated and that their receptor repertoire is biased towards an inhibitory phenotype, as it was reflected by the transient increase in the expression of inhibitory receptors and the decrease of activating receptors.

To determine whether the previous data are clinically relevant, we performed correlation analyses between the expression of different markers and relapse in the cohort of pediatric patients undergoing an autoHSCT (Fig. S2b). We observed a negative correlation between relapse and percentage of CD56bright NK cells early after autoHSCT (S2) and, as expected, a positive correlation between relapse and percentage of CD56dim NK cells. This could indicate that patients with a higher frequency of mature (CD56dim) NK cells at S2 could have a worse outcome after autoHSCT. On the other hand, CD26 and CD55 expression levels at S2 were negatively associated with relapse. The relevance of these data is difficult to determine given the number and heterogeneity of the patients. Clearly, larger and more homogeneous cohorts of patients are needed to validate these results.

KIR expression was temporarily increased shortly after autoHSCT

KIRs regulate NK cell ability to recognize and kill tumor cells; their expression is associated with the outcome of patients with different types of cancer, including neuroblastoma26, and following HSCT, there are changes in their expression15,17. Therefore, we decided to study KIR expression in our cohort. Firstly, we analyzed the patients´ KIR haplotype (Table S5). Next, we analyzed KIR expression only in the patients who have the genes encoding the corresponding KIRs. We observed an increase in the percentage of KIR3DL1+ cells early after autoHSCT in total NK cells, and an increase of KIR2DL1 and KIR2DL2/L3/S2, only in the CD56bright NK cell subset (Fig. 3a). However, we did not see significant variations regarding KIR2DS4 after transplant (Fig. S3), a gene expressed only in three patients (Table S5).

Fig. 3: NK cell KIR expression is altered after autoHSCT, and NK cell education is not coupled with terminal differentiation
figure 3

a Boxplot graphs showing, at the studied five time points (S1–S5), the percentage of cells expressing KIR3DL1 (S1 n = 10; S2 n = 9; S3 n = 9; S4 n = 6; S5 n = 6), KIR2DL2/L3/S2 (S1 n = 12; S2 n = 11; S3 n = 11; S4 n = 8; S5 n = 8) and KIR2DL1 (S1 n = 11; S2 n = 10; S3 n = 10; S4 n = 7; S5 n = 7) within total NK cells and NK cell subsets based on CD56 expression, including CD56dim, CD56bright and CD56neg NK cells. Boxplots show the median and 25–75th percentiles, and the whiskers denote the lowest and highest values. The significance of the data was determined by comparing each sample with sample S2. Each dot represents a patient. *p < 0.05, **p < 0.01. No significance was not indicated (Wilcoxon matched-pairs signed-rank test). b–d Bar graphs showing the percentage of cells expressing 0 (green), 1 (gray), 2 (orange), or 3 (red) KIR (KIR2DL1, KIR2DL2/L3/S2, and KIR3DL1) at the studied five time points (S1–S5). Analysis of KIR combinations in total NK cells (b), CD56bright NK cells (c), and CD56dim NK cell subsets based on the expression of NKG2A and CD57 (d). Data are represented as the mean + SD. (S1 n = 12; S2 n = 11; S3 n = 11; S4 n = 8; S5 n = 8). Results were compared with the S2 time point using the Wilcoxon matched-pairs signed-rank test. Each dot represents a patient. *p < 0.05, **p < 0.01 and ns no significant. e, f Boxplot graphs showing the percentage of CD57+ cells at the studied time points on KIR- (0 KIR), 1 KIR or 2 KIR CD56dim NK cell subsets. Percentage of CD57+ cells in KIR-NKG2A+ or KIR-NKG2A- CD56dim NK cells at different time points (e). Percentage of CD57+ cells in educated (EDU+) or uneducated (EDU−) CD56dim NK cells with 1 or 2 KIR (f). Single KIR-positive cells were considered educated if the KIR is specific for the HLA class I haplotype in a given patient (Tables S5 and S6). Double KIR-positive cells were considered educated if they had at least one KIR that is specific for the HLA class I haplotype in that particular patient (Tables S5 and S6). Data are represented as the median and 25–75th percentiles, and the whiskers denote the lowest and highest values (S1 n = 12; S2 n = 11; S3 n = 11; S4 n = 8; S5 n = 8 were analyzed, but only data that followed the education criteria specified above were considered, and more than one value could be added for each patient because NK cells could be educated for more than one KIR). Results were compared between − and + samples for each time point using an unpaired Mann–Whitney test. No significance was not indicated. Each dot represents a patient. Exact p values are provided in a file named Supplementary Data 2, under Fig. 3.

Full size image

During NK cell differentiation, CD56dim NK cells gradually decrease NKG2A and increase CD57 and KIR expression33. Thus, we next analyzed within different NK cell subsets, the number of KIR (KIR3DL1, KIR2DL2/L3/S2, and KIR2DL1) that NK cells expressed. First, we observed a significant increase in NK cells expressing one or two KIR early after transplant (S2), followed by a gradual decrease reaching pre-transplant levels at day 180 (S5) (Fig. 3b). Regarding CD56bright NK cells, there was a significant decrease in KIR negative cells and an increase in one and two KIR expressing cells at S2 (Fig. 3c). Since we have previously seen that the maturation status is altered after autoHSCT (Fig. 1e), especially in CD56dim NK cells, we analyzed the frequencies of cells expressing different KIR combinations among different maturation stages. Immature CD56dim NK cells (NKG2A+CD57−) expressed lower KIR levels compared to mature (NKG2A−CD57+) NK cells (Fig. 3d). In addition, regardless of the maturation status, CD56dim NK cells had significantly fewer non-expressing KIR cells at S2 (Fig. 3d). In conclusion, there is a higher KIR expression in mature terminally differentiated CD56dimNKG2A-CD57+ NK cells and higher KIR expression early after autoHSCT at S2 in all NK cell subsets.

Finally, as we previously observed a reduction in the frequency of immature (NKG2A+CD57−) NK cells and an increment of mature (NKG2A−CD57+) NK cells later after transplant at S5 (Fig. 1e), we determined whether the process of education by KIRs and NKG2A can influence the expression of CD57, a marker of terminal differentiation, in our pediatric cohort, as it has been studied in alloHSCT33. The education status of these subsets was determined by analyzing KIR and HLA genotypes in every patient (Table S6). In this context, NK cells were considered as educated if their KIRs were specific for self-HLA-I molecules. In each patient, we determined the number of educated and uneducated CD56dim NK cells expressing 1 or 2 KIR, and we analyzed the expression of the CD57 receptor. In NK cells lacking KIRs, the expression of CD57 was compared between NKG2A+ and NKG2A− cells (Fig. 3e). Single KIR-positive NK cells were considered educated if the patient expressed the HLA-I ligand for this KIR, and double KIR-positive NK cells were considered educated if the patient expressed at least one HLA-I ligand. The results showed no significant differences in CD57 expression between educated and uneducated subsets after autoHSCT for any number of KIRs studied (Fig. 3f). We conclude that NK cell education and maturation are uncoupled processes during immune reconstitution following autoHSCT.

Transient acquisition of a decidual-like phenotype by NK cells early after autoHSCT

Next, we analyzed the expression of decidual NK cell markers in the pediatric cohort during immune reconstitution, since we have previously discovered that genes codifying CD9 and CD151, two receptors associated with decidual NK cells34,35, were overexpressed in adult MM patients after autoHSCT17. We observed a temporary increase in both CD9 and CD151 expressing NK cells at S2, decreasing to pre-transplant levels at S5 (Fig. 4a). This sharp increase is clearly evidenced at S2 where the double positive (CD9+CD151+) NK cell subset is 4 times higher than at S1, which is contrasting to the small increase observed for CD9+CD151− NK cells (Fig. 4b). This result is consistent among all NK cell subsets (CD56bright, CD56dim and to a less extent in CD56neg). In our cohort, we observed that CD151+ cells expressed lower levels of CD56 than CD151- cells (Fig. 4c). Additionally, we analyzed CD49a, another decidual NK cell marker34,36, and we observed low expression levels that, nevertheless, were significantly increased early after autoHSCT (Fig. 4d). Altogether, these results suggest that, following autoHSCT, NK cells acquire a decidual-like phenotype supporting previous results at the transcriptomic level17.

Fig. 4: Transient acquisition of decidual-like phenotype by NK cells early after autoHSCT.
figure 4

Boxplot graphs showing the analysis of total NK cells and NK cell subsets based on CD56 expression, including CD56dim, CD56bright, and CD56neg NK cells at the five studied time points (S1–S5). a Percentage of cells expressing CD9 (upper graph) and CD151 (lower graph) in total NK cells. b Percentage of CD9+CD151+ (gray) and CD9+CD151− (purple) within total NK cells and in the three NK cell subsets. c Median fluorescence intensity (MFI) of CD56 in CD9+CD151+, CD9+CD151−, CD9−CD151+, and CD9−CD151− in CD56+ NK cells. d MFI of CD49a in total NK cells. Data are represented as the median and 25–75th percentiles, and the whiskers denote the lowest and highest values. Each dot represents a patient (S1 n = 12; S2 n = 11; S3 n = 11; S4 n = 9; S5 n = 8). Results were compared with the S2 time point using the Wilcoxon matched-pairs signed-rank test. *p < 0.05, **p < 0.01 and ns no significant. Exact p values are provided in a file named Supplementary Data 2, under Fig. 4.

Full size image

Plasma cytokine profile following autoHSCT

We have hypothesized that the cytokine milieu that is present immediately after autoHSCT may be responsible for the acquisition of a decidual-like phenotype by NK cells. Therefore, to address this question, we have measured the plasma levels of molecules with important roles in NK cell biology and that may have an impact in the expansion of a decidual-like NK cell subset during autoHSCT. IL-15 is a cytokine with a relevant role in proliferation, development, cytotoxic activity, and cytokine production of NK cells37,38,39. Similar to what we and others previously described in adult patients16,40, we found that IL-15 plasma levels also significantly increased early after autoHSCT in our pediatric cohort (Fig. 5a). Interestingly, we found a positive correlation between IL-15 levels at S2 and the expression of HLA-DR, CD26, CD55 and CD38 (Fig. 5b). These markers also increased at S2, suggesting that increased levels of this cytokine are, at least partially, responsible for the activation phenotype of NK cells at S2.

Fig. 5: Cytokines plasma levels are altered early after autoHSCT.
figure 5

a Violin plots showing IL-15, TGF-β, GDF-15, PlGF, PDGF-BB, and PDGF-DD plasma levels in a cohort of pediatric patients before (S1) and after autoHSCT: after reaching leukocyte recovery (more than 1000 leukocytes/µl, usually around day 12 after autoHSCT) (S2) and 30 days (S3). Data are represented with the median (red). Each dot represents a patient (S1 n = 13; S2 n = 13; S3 n = 13). Results were compared with the S2 time point (blue) using the Wilcoxon matched-pairs signed-rank test. *p < 0.05, **p < 0.01, ***p < 0.001 and ns, no significant. b Correlogram showing Pearson correlation of the indicated flow cytometry data at S2 and IL-15, TGF-β, GDF-15, PlGF, PDGF-BB, and PDGF-DD plasma levels. Pearson correlation coefficient was indicated by square size (higher size means higher coefficient) and heat scale (positive correlation in red and negative correlation in blue): *p < 0.05, **p < 0.01. Exact p values are provided in a file named Supplementary Data 2, under Fig. 5.

Full size image

TGF-β has been shown to modulate NK cells’ receptor repertoire and to inhibit NK cell-mediated cytotoxicity, favoring immune evasion41,42,43,44,45. In our cohort, we observed that plasma levels of TGF-β slightly decreased, although it was not statistically significant (Fig. 5a). Additionally, a negative correlation was observed at S2 between plasma levels of TGF-β and the increased expression of CD26 and CD31 (Fig. 5b). On the other hand, we also analyzed the plasma levels of growth differentiation factor 15 (GDF-15), which has a role in the development of NK cell dysfunction during systemic inflammation46. We found that GDF-15 levels significantly increased at S2 (Fig. 5a) and that positively correlated with the frequency of CD9+ NK cells (Fig. 5b), a subset that is very elevated at S2.

PlGF is a pleiotropic cytokine which regulates the maturation of uterine NK cells and upregulates the expression of other angiogenic factors47. PlGF levels significantly increased at S2 (Fig. 5a). However, no significant association of PlGF levels with CD9+ NK cells, or any other NK cell subset, was observed (Fig. 5b). Platelet-derived growth factors (PDGFs) are relevant mitogens for many cell types48 and many tumors use them as an autocrine signal for proliferation49. NK cells express receptors for these molecules50 and PDGF-BB has been described to be associated with PFS in NHL patients who have undergone autoHSCT40. In our cohort, the plasma levels of PDGF-BB and PDGF-DD significantly decreased early after autoHSCT (Fig. 5a). Also, we observed a negative correlation at S2 between these PDGFs and the expression levels of CD31, as well as the frequencies of CD26+ and CD9+CD151+ NK cells (Fig. 5b), which were increased at this point. To sum up, early after autoHSCT, IL-15, GDF-15, and PlGF levels increase, PDGF-BB and PDGF-DD levels decrease, while the concentration of TGF-β in plasma does not change.

IL-15 and TGF-β cooperate to induce a decidual-like phenotype on NK cells

Given that others have shown that the combination of TGF-β with IL-15 induced the acquisition of decidual NK cell markers51,52, we have considered the possibility that a combination of cytokines may be required for the expansion of the decidual-like NK cell subset. Therefore, we performed in vitro experiments to determine whether the combination of these two cytokines recapitulates the NK cell phenotype observed after autoHSCT, including the decidual-like subset expansion, at S2. To study the effect of IL-15 and TGF-β at single-cell resolution, we performed a single-cell RNA sequencing experiment (scRNA-seq) on PBMC cells from a healthy donor cultured with and without IL-15 (10 ng/mL) + TGF-β (5 ng/mL). PCA and UMAP mapped these cells into 18 clusters representing the mayor immune cell populations, including B and T cells, monocytes, NK cells, and others (Fig. 6a). By further sub-clustering NK cells, we identified 4 different NK subpopulations, namely NK1, NK2, NK3, and NK4 (Fig. 6b). We followed the nomenclature described by Crinier et al53., and NK1 corresponded to CD56 dim and NK2 to CD56bright NK cells. NK1 cells were exclusively present on unstimulated NK cells, while NK2 cells were observed in both unstimulated and IL-15 + TGF-β-treated NK cells. On the other hand, the subsets that we have called NK3 and NK4 were almost exclusively observed in IL-15 + TGF-β-treated NK cells.

Fig. 6: Transcriptome of NK cells in vitro cultured with IL-15 and TGF-β.
figure 6

scRNA-seq-based clustering and cell type identification of 30,842 cells from unstimulated versus stimulated PBMCs shown by a UMAP embedding, colored by the predicted cell type (a) and cell cycle phase (d). Clustering and cell subtype identification of a subset of 1708 NK cells (unstimulated and stimulated) as shown by the UMAP (b, e) and PHATE embeddings (c, f), color coded by cell subtype (b, c) and cell cycle phase (e, f). UMAP embedding showing expression levels of relevant genes showing differentially significant expression in treated vs unstimulated cells. The panels show unstimulated (left) and stimulated (right) NK cells (g). Dot plot of relevant genes. The size of the dots is proportional to the percentage of cells that express a given gene within a cell type, while the color of the dot corresponds to the scaled average gene expression (h, i).

Full size image

We used a complementary method of dimensionality reduction, namely PHATE (Potential of Heat-diffusion for Affinity-based Transition Embedding), to further visualize the NK cells, and observed that the NK4 cluster was only distantly related to the cluster NK1, suggesting that cytokine stimulation has triggered a transcriptional program that significantly differs from the one observed in unstimulated NK cells (Fig. 6c). Determination of the cell cycle state based on the expression of G2/M and S phase markers, indicated that while the majority of NK4 cells were in the cycling phases (G2/M and S), only 49% and 64% of cells in the NK3 and NK2 clusters, respectively, were actively cycling. Virtually none were observed in the NK1 cluster (Fig. 6d–f). Finally, we analyzed the expression levels of genes that encoded cell surface markers that significantly changed early after autoHSCT (S2) in our cohort of pediatric patients. We observed significant upregulation of DPP4 (CD26), CD9 (CD9), ITGA1 (CD49a) and CD8A (CD8) in treated cells, and downregulation of CD160 (CD160) and CD244 (2B4) (Fig. 6g–i), consistent with what has been observed at the protein level at S2. The significant gene expression changes observed in CD38 and PECAM1 did not correlate with cell surface expression of CD38 and CD31, respectively, at S2. Genes encoding other cell surface receptors did not significantly change.

We also isolated PBMCs from several healthy donors and incubated them with IL-15 and TGF-β at increasing concentrations. We studied the following conditions: unstimulated, 5 ng/mL of TGF-β, 10 ng/mL of IL-15 without or with increasing concentrations of TGF-β (1, 5, or 10 ng/mL). NK cells were identified by flow cytometry as viable CD3-/CD14-/CD19-/CD123-/CD56+ cells and were analyzed to determine the differentially expressed receptors. Conventional supervised analyses were performed, and higher CD38, HLA-DR, CD55, CD56, and CD151 expression levels were observed when cells were incubated with IL-15 compared with the unstimulated condition (Fig. S4a–d). On the other hand, TGF-β alone induced a significant decrease in the expression of these markers when compared with unstimulated cells, except for CD151 (Fig. S4a–d). Nevertheless, TGF-β tended to down-regulate IL-15-induced expression of CD38, HLA-DR, CD55, and CD56, but not to the levels of unstimulated NK cells, suggesting that, at these concentrations, IL-15 has a more dominant effect than TGF-β. On the other hand, we observed a higher expression of decidual markers CD9 and CD49a in conditions containing both TGF-β and IL-15, suggesting that both cytokines are necessary in order to obtain high levels of these two markers (Fig. S4d). Moreover, we observed a significantly lower expression of CD160 and CD229 in response to TGF-β and IL-15 (Fig. S4e). The CD160 decrease was even higher when both cytokines were present, suggesting a synergistic effect, while this synergy was not observed when we looked at CD229 expression levels.

To gain more insight, we performed an unsupervised analysis using t-distributed stochastic neighbor embedding (t-SNE) and the FlowSOM clustering tool for total NK cells (Fig. 7). We observed an increase in Pops 0, 3, and 4 when cells were cultured with IL-15 and TGF-β in comparison with the unstimulated condition and single cytokine treatment. These Pops define CD9+CD38+CD56high NK cells that, specifically in Pops 0 and 4, also expressed CD55 and CD151. Pop 4, in particular, exhibited the highest expression of CD49a. Pop 1, with a CD9+CD151−CD56low phenotype, was almost exclusively present when NK cells were cultured only with TGF-β, and Pop 2, with a CD9-CD151-CD56low phenotype, is the most abundant in unstimulated or only TGF-β conditions. These results suggest that TGF-β is necessary for the expression of CD9 and that IL-15 is responsible for the increased expression of CD56, CD151, CD49a, CD38, and CD55. Within the CD9- NK cells, we found that Pop 5 and 6 were the most abundant when NK cells were cultured only in the presence of IL-15, and that the addition of TGF-β significantly decreased the frequency of these Pops. Furthermore, in several Pops, a TGF-β concentration-dependent effect was evident: Pops 0 and 3 increased with higher TGF-β concentration. Pop 7 is a very residual population (median of 0.53%) characterized by a very high expression of HLA-DR, which may suggest that these cells could be another ILC subset54. More studies are required to properly identify this very small subset. Altogether, these results suggest that IL-15 and TGF-β induce an activated and decidual-like phenotype on NK cells, which is characterized by higher expression of CD56, CD9, CD49a, CD151, CD55, CD38, and HLA-DR markers. We found these cells to be significantly expanded at S2, which correlates with very high levels of IL-15 and unchanged levels of TGF-β.

Fig. 7: IL-15 and TGF-β induce an NK cell decidual-like phenotype.
figure 7

Unsupervised flow cytometric analysis using t-SNE and FlowSOM clustering tools to analyze the indicated markers (CD229, CD55, CD160, CD49a, CD56, CD9, HLA-DR, CD38, and CD151). NK cells were cultured with no cytokines, IL-15, 5 ng/mL TGF-β, IL-15 + 1 ng/mL TGF-β, IL-15 + 5 ng/mL TGF-β, IL-15 + 10 ng/mL TGF-β. IL-15 concentration was 10 ng/mL (n = 7). a t-SNE projection of NK cell populations (Pop) identified by the FlowSOM clustering tool for the specified markers. b Fluorescence intensity of each Pop as indicated in the column-scaled z-score. c Bars graph showing the percentage of each Pop at the six studied conditions. Significance of data was determined using the Wilcoxon matched-pairs signed-rank test by comparing each sample with the condition IL-15 + 10 ng/mL TGF-β. *p < 0.05, **p < 0.01 and ns no significant. Exact p values are provided in a file named Supplementary Data 2, under Fig. 7.

Full size image

Discussion

In this work, we have studied the reconstitution of NK cells in pediatric patients with cancer who had undergone autoHSCT. We have analyzed NK cell subsets’ phenotype, plasma levels of cytokines that are relevant in NK cell function, and we have performed in vitro experiments (scRNA-seq and flow cytometry) to explore the mechanisms behind the expansion of NK cells with a decidual-like phenotype during immune system reconstitution following autoHSCT. The cell surface phenotype and the transcriptome of NK cells vary considerably after autoHSCT15,16,17 and, to the best of our knowledge, we have observed a previously undescribed change in NK cells phenotype towards a decidual-like and activated phenotype early after autoHSCT in pediatric patients. This activated and decidual-like phenotype is characterized by an increase of CD56, CD9, CD49a, CD151, CD38, HLA-DR, and CD55 expression shortly after autoHSCT. In addition, our in vitro experiments suggest that the combination of IL-15 and TGF-β induces, at least partially, this activated and decidual-like phenotype on NK cells.

It is known that TGF-β can induce the acquisition of a decidual-like phenotype in peripheral blood NK cells, characterized by the expression of CD955. But in our study, we did not observe a significant change in TGF-β levels following autoHSCT (Fig. 5a). However, other authors have described that the combination of TGF-β with IL-15 induces the acquisition of decidual NK cell markers such as CD9, CD49a, and CD15151,52,56. Therefore, we hypothesized that the increased levels of IL-15 (Fig. 5a), together with TGF-β, may contribute to the acquisition of the NK cell decidual-like phenotype. In addition, other cytokines may have a role in the expansion of decidual-like NK cells early after autoHSCT, as for example GDF-15 and PlGF increased plasma levels (Fig. 5a). Importantly, we observed that GDF-15 levels associate with CD9 expression at S2 (Fig. 5b), suggesting that this cytokine could also have a role in the decidual-like phenotype acquisition. Nevertheless, more studies are needed in order to determine the effect of GDF-15 and other cytokines, such as PlGF, on the expression of decidual markers such as CD9.

Several publications have shown an increased frequency of CD9+ circulating NK cells that associates with worse disease evolution57,58,59,60 suggesting that CD9 expression on NK cells may be used as a biomarker in cancer patients, although we were not able to find an association between relapse and frequency of these cells. We speculate that the expansion of CD9+ NK cells is independent of the malignancy for which patients receive an autoHSCT, because we have previously published the same phenomenon in MM patients early after autoHSCT17. Evidently, we need larger and more homogeneous cohorts in order to establish CD9+ NK cells as a possible biomarker that may predict the evolution of cancer patients undergoing an autoHSCT in childhood.

In addition to the decidual NK cell markers, the cytokine milieu may also have a significant role in the observed changes in the expression of other markers early after autoHSCT. For example, correlation analyses (Fig. 5b) and in vitro experiments (Figs. 7 and S4) showed that the increased expression of CD56, activation markers (HLA-DR and CD38) and receptors such as CD26 and CD55, could be, at least in part, a consequence of the high IL-15 levels at S2 (Figs. 5b and S4), as it is shown in other studies30,61,62,63. However, TGF-β tended to counteract the effect of IL-15 (Figs. 7 and S4). Remarkably, analysis of data from both scRNA-seq (Fig. 6) and flow cytometry (Figs. 7 and S4) experiments suggests that the combination of IL-15 and TGF-β has a relevant role in the expansion of decidual-like NK cells early after autoHSCT, at least partially. The two in vitro experimental approaches showed that the combination of these two cytokines is responsible for the increased expression of CD9 (CD9), CD151 (CD151), and CD49a (ITGA1), markers that are typical of decidual NK cells51,52,56. Other markers that changed their expression at S2, such as CD26, CD160, and CD229, followed a similar trend of expression after the in vitro culture with IL-15 and TGF-β.

As expected, the frequencies of CD57+ NK cells in our pediatric patient cohort were lower than those observed in the cohort of MM adults16. This could be owed to the fact that cytomegalovirus infection, more prevalent in adults than in children, induces an expansion of CD57+ NK cells64,65,66. In addition, we have observed that NK cells express a more immature phenotype early after transplantation at S2 and acquire a more mature phenotype 180 days after autoHSCT, as shown by the increase of the mature NKG2A-CD57+ cell subset at S5. This is in agreement with the model of NK cell development and with the decreased NKG2A and increased CD57 expression during NK cell maturation19,33. These mature NKG2A−CD57+ NK cells exhibited a higher KIR expression, and they showed a tendency to expand after autoHSCT at S2 (Fig. 3d). These results are in line with reported data from adults receiving an alloHSCT33. Importantly, we have also described for the first time that in autoHSCT, NK cell education and maturation are uncoupled processes, since CD57 expression is not influenced by KIR expression, as it was previously described in alloHSCT33,67.

The search and identification of prognostic biomarkers can be of great help to clinicians. We have observed a positive and significant correlation at S2 between the percentage of CD56dim NK cells and relapse (Fig. S2b), which express more CD57 than the CD56bright subset. These results are in agreement with data previously published by us and others in which low frequencies of circulating CD57+ and CD57+NKG2A− NK cells are associated with better clinical outcomes16,68. On the other hand, we have also observed that the increased expression levels of CD26 and CD55 at S2 seem to negatively associate with relapse (Fig. S2b). It is still unknown how these two receptors are involved in the evolution of autoHSCT and the underlying disease in pediatric patients. Some authors have found that high frequencies of transplanted CD26+ lymphocytes have a beneficial effect on NK cell recovery after autoHSCT in MM patients69. Conversely, others have shown that lower CD26 expression on different immune cell subtypes, including NK cells, of the donor stem cell harvest is associated with better survival in alloHSCT70. Regarding CD55, it has been described that, when expressed on NK cells, it makes them less effective at killing K562 targets71. Of note, it has been recently published that mouse NK cells present in the decidua were primarily the differentiated tissue resident NK cell subset, distinguished by their expression of CD5572. Definitely, it is required to explore if CD55 could be used as a human decidual (and decidual-like) NK cell marker. Nevertheless, these results have to be taken cautiously because our cohort is very small and heterogeneous. To validate these data, more studies, including a higher number of patients, are required.