A total of 23 adults underwent non-myeloablative haplo-HSCT at the National Institutes of Health (NIH) from March 2010 through September 2015 for SCD (21/23) and beta-thalassemia (2/23). One patient with SCD died <6 months post-HSCT and was not included in the study. 20 patients with SCD were evaluated for cytokines and immune cell subsets ( Supplementary Table S1 ). Patients were conditioned with alemtuzumab, 400 cGy total body irradiation, PT-Cy doses ranging from 0-100 mg/kg body weight in three dose dependent cohorts, (cohort 1: 0mg/kg body weight, cohort 2: 50mg/kg body weight and cohort 3: 100 mg/kg body weight). Sirolimus was loaded 1 day before transplant in cohort 1 and in the first 6 patients who received a transplant in cohort 2 and 1 day after PT-Cy in the remaining cohort 2 patients (day 4) and in all cohort 3 patients (day 5). A trough level of 10 to 15 ng/mL was targeted until 3 to 4 months posttransplant, and then the level was decreased to 10 to 12 ng/mL until 1 year posttransplant and then 5 to 10 ng/mL thereafter in engrafted patients (15). Donor engraftment was defined as sufficient donor chimerism (DMC≥20%) at PT-Day 180 and reversal of acute SCD complications. Immunophenotyping of the peripheral blood mononuclear cells (PBMCs) was performed in all available patient samples. The study was approved by the Institutional Review Board of the National Heart, Lung, and Blood Institute (NHLBI, ClinicalTrials.gov Identifier NCT00977691). All patients gave written informed consent. The study was monitored by an independent data and safety monitoring board.

Peripheral blood samples were collected at baseline (BSL) and serially at PT-Day 30, 60, 100, and 180. Blood samples were collected in EDTA tubes (Becton Dickinson, San Jose, CA, USA) and plasma stored at -80° C and PBMCs at -140° C until analysis. PBMCs were isolated using the Ficoll density gradient protocol. patients were grouped at each PT-time point based on their engraftment status [engrafted or rejected ( Supplementary Table S1 )] and DMC level [high DMC (HDMC) with ≥ 20% or low DMC (LDMC) with < 20%] ( Supplementary Table S2 ).

A multiplexed magnetic bead assay was employed to analyze 48 cytokines in plasma (Bio-Rad, Hercules, CA, USA). Two cytokines [transforming growth factor-b1 (TGF-b1) and B-cell-activating factor (BAFF)] were measured using an enzyme-linked immunosorbent assay (ELISA) based DuoSet kit (R&D, Minneapolis, MN, USA). All assays were performed according to the manufacturer’s instructions. Four cytokines [interleukin (IL)-1a, IL-12p40, monocyte-chemotactic protein (MCP)-3, and tumor necrosis factor-b (TNF-b)] had more than 75% of values below the lowest limit of detection (LLOD) and two cytokines [cutaneous T-cell-attracting chemokine (CTACK), stromal cell-derived factor-1a (SDF-1a)] failed standard curves. Therefore, we excluded these cytokines from the analysis ( Supplementary Table S3 ). Abbreviations for all the cytokines that are evaluated in this study are listed in Supplementary Data .

Based on the cytokine results, two panels ( Supplementary Tables S4A, B ) were designed to evaluate various regulatory and effector immune cell subsets ( Supplementary Table S5 ) by flow cytometry. Cell surface staining of PBMCs was performed as described with some modification (17). After thawing frozen vials, cells were suspended in a sterile complete medium. For surface staining, cells were stained in flow cytometry staining buffer (PBS, 2% heat-inactivated FBS), and prior to surface human antibody conjugates staining samples were treated with human FC block antibody. The immunophenotyping analysis was performed in two ways. The first analysis involved a comprehensive phenotyping of the following eight major immune cell subsets: (i) B cells: CD19⁺, (ii) CD8⁺ T cells: CD3⁺CD8⁺, (iii) regulatory T cells (Tregs): CD4⁺FoxP3⁺, (iv) effector CD4⁺ T cells: CD4⁺FoxP3^-, (v) natural killer (NK) cells: CD3^-CD56⁺, (vi) Monocytes: CD14⁺, (vii) dendritic cell (DC) subsets, plasmacytoid DCs (pDCs): lineage (CD3, CD19, CD56) (lin)^- HLA-DR⁺CD123⁺CD11c^- (18) and myeloid DCs (mDCs): lin^-HLA-DR⁺CD123^-CD11c⁺ (19), and (viii) myeloid-derived suppressor cell (MDSC) subsets (20, 21), early MDSCs (eMDSCs): lin^-HLA-DR^-CD11b⁺CD33⁺, monocytic MDSCs (mMDSCs): lin^-HLA-DR^-/lowCD14⁺CD15^-, polymorphonuclear MDSCs (PMN-MDSCs): lin^-HLA-DR^-/lowCD14^-CD15⁺CD11b⁺. Later more detailed analysis was performed to evaluate the following immune regulatory/effector cell types ( Supplementary Table S5 ): (i) Tregs: CD4⁺CD25⁺FoxP3⁺ (22); (ii) type 1 regulatory (Tr1) cells: CD4⁺FoxP3^-CD45RA^-LAG3⁺CD49b⁺ (23) (iii-v) eMDSCSs, mMDSCs, and PMN-MDSCs; (vi-vii) pDCs and mDCs; (viii) regulatory B cells (Bregs): CD19⁺CD24^hiCD38^hi (24) (ix) T helper (Th)1 cells: CD3⁺CD4⁺CD45RO⁺CXCR3⁺ (19), and (x) Th17 cells: CD3⁺CD4⁺CD45RO⁺CCR6⁺ (19). The gating strategies for these 10 subsets are described in Supplementary Figures S1–S4 . The gating strategy was adapted from the referenced articles indicating each cell type and validated by the NHLBI Flow Cytometry Core. The PBMCs were first stained with cell surface markers. Then FoxP3, LAG3, TGF-b1, IL-10, and IL-7 were stained intracellularly.

Mean, median, standard deviation (SD), minimum and maximum values of cytokine concentrations were calculated ( Supplementary Table S3 ). LLOD categories and logistic regression model details are described in Supplementary Data . Additionally, we used linear regression models to compare continuous cytokine concentrations between the engrafted and rejected groups at each time point. Spearman’s rank correlations were employed to examine the correlation between the different cytokines at each time point. Factor analysis was used to examine the relationships between the selected cytokines. The factors computed based on the BSL time point for all patients were categorized into quartiles and used as predictors in logistic regression models fit to all time points for all subjects, accounting for repeated measures for the same person over time in the variance computation. Random forests using continuous cytokine levels were implemented as additional sensitivity analysis. Missing values were excluded from the analyses.

The cellular flow cytometric data highlighting the immune reconstitution were analyzed by comparing the log10-transformed frequencies. Log10-transformed frequencies were used to compare differences between the engrafted versus rejected groups and HDMC versus LDMC groups using pairwise multiple t-tests at each time point. We calculated Spearman’s rank correlations between phenotypic frequencies of immune cell subsets at each time point. All tests were two-sided, and P <0.05 was considered statistically significant. Bonferroni corrections were applied to adjust for multiple testing. Analyses were performed using STATA software (version 14.2, StataCorp LLC., College Station, TX, USA), and graphs were generated using GraphPad Prism software (version 7 and 8).

The characteristics of the patients in the engrafted and rejected groups and their donors are described in Table 1A . The engrafted group comprised of an equal number of males and females (5), whereas the rejected group consisted of 7 males and 3 females. The mean age in the engrafted group was 34.4 ± 6.8 years and in the rejected group 34.2 ± 12.21 years. More donors were female in both engrafted and rejected groups, 7/10 (70%) and 8/10 (80%), respectively. There were no significant differences between the recipient’s or donor’s age, race, gender, and cell numbers infused between the two groups ( Table 1B ).

Among 44 cytokines evaluated, 23 with values over LLOD were further categorized into two groups: above or below the overall median of each cytokine. The remaining 21 cytokines were categorized into three groups: <LLOD, below the median, and above the median of detectable values ( Supplementary Table S3 ). We first assessed the association with engraftment for all 44 cytokines (Fisher’s exact P-values given in Table 2 ). The sample at PT-Day 60 revealed the lowest P-value difference between the engrafted and rejected groups. Fibroblast growth factor (FGF), granulocyte-macrophage colony-stimulating factor (GM-CSF), IL-12p70, IL-9, and macrophage inflammatory protein (MIP)-1a were associated with engraftment (Bonferroni-corrected P <0.001), whereas granulocyte colony-stimulating factor (G-CSF), interferon (IFN)-g, IL-10, IL-13, IL-17, IL-1b, IL-1RA, IL-4, macrophage migration inhibitory factor (MIF), TGF-b1, TNF-a, and vascular endothelial growth factor (VEGF) were associated with P <0.01, and growth-regulated protein (GRO-a), platelet-derived growth factor (PDGF)-BB with P =0.05 at PT-Day 60. At PT-Day 100, IL-7 was associated with engraftment (P <0.01), as were IL-2 and VEGF (P <0.05). MIF was associated with P <0.01, and IL-7 and TGF-b1 with P <0.05 at PT-Day 180. In contrast, IL-18 was associated with rejection at PT-Day 100 with P <0.005. Notably, IL-6 was associated with engraftment at BSL (P <0.017). The remaining markers did not show any associations.

Odds ratios (ORs) from logistic models using the categories of cytokine concentrations as ordinal variables and P values for all time points are given in Supplementary Table S6 . ORs of PDGF-BB, TGF-b1, and TNF-a were associated with engraftment with a P <0.05 at PT-Day 30. At PT-Day 60, ORs of FGF, GM-CSF, IL-9, and MIP-1a were associated with P <0.001, and G-CSF, IFN-g, IL-10, IL-17A, IL-1RA, IL-4, TGF-b1, and VEGF were associated with P <0.01; and GRO-a, IL-12p70, IL-7, MIF, and PDGF-BB with P <0.05. OR of IL-7 was associated with engraftment at PT-Day 100 with P <0.01, and IL-12p70, IL-13, IL-17, IL-2, IL-9, PDGF-BB, TNF-related apoptosis-inducing ligand (TRAIL), and VEGF were associated with P <0.05. At PT-Day 180, OR of MIF was associated with engraftment with P <0.01, whereas G-CSF, GRO-a, IL-10, IL-17A, IL-7, MIP-1a, PDGF-BB, TGF-b1, and TNF-a were associated with P <0.05. In contrast, the OR of IL-18 was associated with rejection at PT-Day 100 with P <0.01 and at PT-Day 180 with P <0.05. There were no significant differences between the two groups at BSL for any cytokines.

Results from linear regression models are presented in Supplementary Table S7 . As expected, all cytokine concentrations in plasma substantially dropped from their BSL levels after the HSCT ( Figure 1 ). Further, 18 cytokine concentrations were higher in the engrafted group from PT-Day 60 to 100. Only the concentration of IL-18 was higher in the rejected group. The remaining cytokines did not show statistically significant differences in concentrations between the two groups at any time. We thus found the most significant differences in cytokine levels between the engrafted and rejected groups at PT-Day 60, the time point around which secondary graft failure typically occurs.

We used factor analysis to describe the variability among the correlated cytokines in terms of a lower number of unobserved variables called “factors” that are linear combinations of the original cytokines. After removing highly correlated cytokines, we included the following ten cytokines in a factor analysis: GROa, G-CSF, IL-10, IL-17A, IL-18, IL-2, MIP-1a, PDGF-BB, TGFb1, and VEGF. We identified two factors as important, estimated factor loadings (i.e. the coefficients in the linear combination) based on the BSL levels, and computed factors for all time points. We then categorized the factors into quartiles and used them as predictors in logistic regression models. The first factor (IL-17A, IL-10, IL-7, G-CSF, IL-2, MIP-1a, VEGF, and TGFb1 contributed significantly) was strongly associated with engraftment with OR = 2.75 (95% CI of 1.40 to 5.38) whereas the second factor (GROa, and IL-18 contributed significantly) was not statistically significant ( Supplementary Table S8 ).

Immunophenotypic analysis of the patients’ immune cell repertoire comprising of B cells, CD8⁺ T cells, Tregs, effector CD4⁺ T cells, NK cells, monocytes, DCs (pDCs and mDCs), and MDSCs (eMDSCs, mMDSCs, and PMN-MDSCs) at BSL, PT-Days 30, 60, 100, and 180 were performed. The cellular frequencies of these cells are plotted ( Figures 2A –H ). A non-significant trend in the frequency of MDSCs was observed at PT-Day 60 (P >0.06). Since DMC is a critical factor in promoting allograft acceptance and treating SCD, we grouped the patients into HDMC (≥20%) and LDMC (<20%) at each time point and observed consistent PT-time point visual differences in DCs, and MDSCs between the engrafted and rejected patients between HDMC and LDMC patients ( Supplementary Figure S5 A–E ).

We evaluated the percentages of three different types of MDSCs (20, 21) in our patients and compared the frequencies of each type between the engrafted and rejected groups and HDMC and LDMC groups at each time point. We observed higher frequencies of eMDSCs in the HDMC group (P <0.003; Figure 3A and Supplementary Figure S6A ) and engrafted group (P <0.04; Figure 3B ) at PT-Day 30. We used distinguishable HLA to determine the source of eMDSCs in 9/20 patients ( Table 3 ). These nine patients had a total of 16 HDMC time points and 14 LDMC time points. The source of eMDSCs revealed the following patterns: in engrafted patient 225-19 up to 99% eMDSCs were donor-derived at all time points, wherein with rejected patients 225-10, 225-52, and 225-55, all eMDSCs were 100% recipient-derived in origin. Interestingly, in rejected patient 225-40, at day 30-PT, eMDSCs were 100% recipient-derived. At later time points, however, eMDSCs were 100% from the donor. Other patients maintained more mixed donor and recipient origins until at least day 180-PT. eMDSCs from both donor and recipient (mixed chimeric state) origins were observed at 15/16 HDMC time points as compared to only two LDMC time points ( Table 3 ; P< 0.00001 and Supplementary Figures S6B, C ).

Tregs are the most commonly observed cellular population in patients with immune tolerance (27), and they prevent acute graft versus host disease (GVHD) (28) following HSCT. We compared frequencies of Tregs between the engrafted and rejected patients and among HDMC and LDMC groups. While we did not find any significant differences after multiple testing correction, we noticed a trend towards increased frequencies of Tregs in the engrafted group at PT-Day 100, (P <0.04; Figures 4A, B ) and in the HDMC group at PT-Day 100 (P <0.09; Figure 4C ). The elevated frequencies of Tregs agree with our cytokine results, where we observed elevated plasma levels of IL-10 at PT-Day 60 (P <0.05), PT-Day 100 (P <0.01), and PT-Day 180 (P <0.05) in engrafted patients. We tracked the source of Tregs using distinguishable HLA in 9/20 patients. We did not observe statistically significant differences between mixed chimeric or non-chimeric Tregs in HDMC and LDMC groups ( Table 4 ).

We further calculated the percent change in the frequencies of Tregs from the BSL for each patient and compared the percent change between HDMC versus LDMC groups. We observed a higher Treg change in HDMC patients ( Figure 4D ). Although we observed no significant correlation between the frequencies of Tregs and percentages of DMC, frequencies of Tregs mirrored the DMC dynamics. This was observed in two of the patients who engrafted initially before they rejected their grafts at PT-Days 60 and 100 respectively. The frequencies of Tregs at these time points decreased close to BSL as opposed to one patient who maintained engraftment and high frequency of Tregs persisted ( Supplementary Figure S7A–C ). Further, we evaluated IL-10 and TGF-b1 producing Tregs and found a trend of higher IL-10 producing Tregs in the HDMC group at PT-Day 30 (P <0.02; Figure 4E ), however, TGF-b1 producing Tregs did not show any difference between groups (data are not shown).

Plasma cytokine data revealed higher levels of IL-17 in the engrafted patients at PT-Days 60 and 100 ( Figure 1 ). However, we did not observe any statistically significant difference in the frequencies of Th17 cells between engrafted and rejected patients and HDMC and LDMC groups (data are not shown). We also did not find a statistically significant difference in TGF-b1 and IL-10 producing Th17 cells between the HDMC and LDMC groups nor the frequencies of Bregs, pDCs, mDCs, mMDSCs, PMN-MDSCs, and Tr1 cells either between the engrafted and rejected or the HDMC and LDMC groups (data not shown).

Since Tregs have been associated with tolerance, we next performed the correlation analysis between eMDSCs and Tregs and observed a positive correlation between the frequencies of eMDSCs and Tregs at PT-Day 100 (r=0.72, P <0.0007; Figure 5A ). Importantly, Tregs at PT-Day 100 correlated positively with the eMDSCs at BSL (r=0.63, P <0.004; Figure 5B ). Tregs at PT-Day 60 tend to show positive correlation with eMDSCs at PT-Day 180 but the association was not significant after applying correction for multiple testing ( Supplementary Figure S8A ). We next tested the correlation of the frequencies of eMDSCs with percentages of DMC at all PT time points. We observed a trend towards a positive correlation between frequencies of eMDSCs at PT-Day 60 with the percentage of DMC at PT-Day 180, but the association could not stand the correction applied for multiple testing ( Supplementary Figure S8B ).

We next evaluated the number of patients at each post-transplantation time point who experienced graft failure with donor myeloid chimerism (DMC) levels below 20% as an indicator of graft failure based on Kaplan Meier estimates ( Supplementary Table S9 ). Of the total 20 patients, DMC levels decreased below 20% at PT-Day 30 in 3 patients. At PT-Day 60, 4 additional patients had their DMC below 20% and at PT-Day 100, 5 additional patients. Finally, DMC levels decreased below 20% in 6 additional patients at PT-Day 180, adding up to 11 patients with graft failure in total. The time to graft failure (DMC below 20%) is also plotted in Supplementary Figure S9 , that also shows numbers of subjects at risk for graft failure at each time point.