Heterogeneity of HSPCs is reduced in MDS

We first investigated in details the distribution of CD34+lin-HSPCs in a cohort of MDS and non-MDS samples (cohort #1 in Supplementary Table 1) using a panel of 19 antibodies (combination #1 in Supplementary Table 2) allowing the identification of CD34+HSPCs sub-populations among both CD38- and CD38+ fractions of cells. The gating strategy and the complete phenotype of sub-populations identified is provided in Supplementary Fig. 1 and Supplementary Table 3 respectively. As expected, CD71 and CD110 expression led to the identification of different cell fractions (F1, F2 and F3) among multipotent progenitors (MPPs) and common myeloid/megakaryocyte-erythroid progenitors (CMPMEPs), which were previously reported to have distinct differentiation potential [14]. CD34+lin- cells from 27 MDS BM samples (Supplementary Table 4) and from 9 non-MDS samples without cytopenia matched for age were clustered using the pipeline summarized in Fig. 1A. The repartition on the t-SNE plot of the 10 clusters identified was highly similar between the control samples, whereas 6 among 17 patients with low-risk MDS and almost all high-risk MDS samples demonstrated distinct t-SNE patterns characterized by an accumulation of cells in a reduced number of clusters (Fig. 1B and Supplementary Fig. 2).

Fig. 1: Deep immunophenotyping of HSPCs revealed a decreased heterogeneity of HSPCs in MDS samples.

BM samples are from cohort #1. A Scheme of the pipeline used for automated clustering of CD34+ cell sorted subpopulations with the Cytobank web-based software. B CD34+ clusters repartition on a t-SNE plot from 4 representative control samples (CTL) and 4 MDS samples (low risk (LRMDS), n = 2; high risk (HRMDS), n = 2). C Heat map representation of clusters abundance among CD34+lin- sorted cells. Each column represents the relative frequency of the 10 clusters identified by the t-SNE-FlowSOM workflow for each sample (rows). The dendrogram demonstrates how the different MDS and control samples grouped together. Note that samples LRMDS1 and HRMDS2 come from serial BM aspiration of the same patient (each several months apart) and demonstrate consistent properties. D Phenotypic cells characterization of clusters 8, 9 and 10 in samples CTL1, LRMDS7 and HRMDS8, respectively. E Abundance of cluster 8, 9 and 10 in control (n = 9) and MDS samples according their low (n = 17) or high (n = 10) risk status. Lines represent means ± s.d. Statistical significance was determined using unpaired two-tailed Mann–Whitney tests.

An automated hierarchical clustering of all samples according abundance of each clusters identified 3 groups of samples relatively homogeneous in terms of MDS or non-MDS conditions mainly according abundance of clusters 8, 9 and 10 (Fig. 1C). Interestingly, samples LRMDS1 and HRMDS2 which came from serial BM aspirations of a unique patient (12 months apart) demonstrated similar enrichment in clusters 8 and 9 at both low and high-risk stage, suggesting that abnormal HSPCs repartition pre-existed to disease progression. We identified the HSPCs sub-populations corresponding to cluster 8 as LMPPs, cluster 9 as CMPMEP F3 and cluster 10 as B/NK progenitors (Fig. 1D). The abundance of CMPMEP F3 and LMPPs were significantly modified in low- and/or high-risk MDS samples compared to control samples (Fig. 1E). B/NK progenitors and LMPP were respectively increased and decreased in low-risk compared to high-risk MDS (Fig. 1E).

Altogether, this indicates that a loss of HSPCs heterogeneity can be detected in most of MDS samples, including from early stage of these diseases.

Different profiles of HSPCs repartition emerge in MDS

We then applied a similar strategy of HSPCs identification with a 9 markers panel (combination #2 in Supplementary Table 2) to a large prospective cohort (cohort #2) of 184 fresh BM samples. The gating strategy and characteristics of MDS patients are described in Supplementary Fig. 3A, B and Table 1 respectively. We observed different types of abnormal HSPCs repartition in MDS samples such as accumulation of MLPs, LMPPs, MPP F2 (CD71 + CD110-), MPP F3 (CD71 + CD110 + ), GMPs, CMPMEP F2 (CD71 + CD110-), CMPMEP F3 (CD71 + CD110 + ) or decrease of GMPs and CMPMEP F3 (Supplementary Fig. 4A). Some sub-populations of HSPCs not observed in control samples such as CD34 + CD38-CD45Ra-CD10+ cells or MPP or CMPMEP CD110 + CD71- (called thereafter MPP F4 and CMPMEP F4) were also detected in MDS (Supplementary Fig. 4B), suggesting that the hematopoietic hierarchy could be impaired. We quantified the proportion of each CD34 + CD38+ and CD34 + CD38-HSPCs sub-populations detected in non-MDS samples without cytopenia (Supplementary Table 5) and this repartition was further used as a reference to analyze HSPCs distribution in BM samples from patients with cytopenia related or not to MDS through z-score calculation. Only few differences were observed in the HSPCs repartition in BM samples from patients with cytopenia unrelated to MDS (Fig. 2A), whereas HSPCs repartition was significantly more frequently abnormal in MDS patients (Fig. 2B) (7/260 (2.7%) vs 412/1703 (24.2%) abnormal z-score values <-2 or>2 respectively, p < 0.0001). GMPs, CMPMEP F4 and MPP F3 were the most frequently increased HSPCs sub-populations in MDS samples (Supplementary Fig. 4C). Hierarchical clustering identified distinct group of MDS samples with profiles of HSPCs repartition with increase of specific HSPCs sub-populations frequently correlated with IPSS-R stratification (Fig. 2B). Profiles with accumulation of LMPPs or MLPs were enriched in patients with high-risk MDS, whereas the accumulation of B/NK, GMPs or CMPMEP F3 was more frequently related to low-risk MDS (Fig. 2B). Accordingly, profiles with accumulation of LMPPs or B/NK progenitors were characterized by the highest and the lowest median blast cells count respectively (Supplementary Fig. 4E). Analysis of the dependency of each phenotypic pattern revealed association of not only LMPPs with GMPs, but also GMPs with MPP F3 and CMPMEP F4 with either MPP F3, CMPMEP F1 or GMP (Fig. 2C).

Table 1 Presenting features of patients with MDS from cohort #2 and cohort #3.Fig. 2: Identification of the different HSPCs repartition patterns in MDS.

BM samples are from cohort #2. A Heatmap visualization of z-scores computed for all HSPC sub-populations (rows) from BM samples collected in patients with cytopenia not related to MDS (n = 20) compared to non-MDS samples without cytopenia. Each column in the heatmap is an individual sample. B Hierarchical clustering heat map visualization of z-scores computed for all HSPC sub-populations (rows) from BM samples collected in patients with MDS (n = 131) with their low (very low/low), intermediate or high (high/very High) IPSS-R score compared to non-MDS samples without cytopenia. Each column in the heatmap is an individual sample. The dendrogram demonstrates how the different MDS samples grouped together. NA, IPSS-R not available due to karyotype failure. C The circos diagram visualizes the association of the different phenotypic pattern of CD38+ and CD38-HSPCs in MDS samples in which abnormal accumulation of a least two HSPCs sub-populations was detected (n = 70). D Mutational landscape of the 102 MDS samples according the phenotypic HSPCs pattern. Mutational VAF is coded; ice, 2–10%; sky, 10–20%; orchid, 20–30%; blueberry, >30%. If multiple mutations are present in a case, the highest VAF is shown. EM epigenetic modifiers, SF splicing factors, SIG signaling, TF transcription factors, TS tumor suppressors.

We also analyzed the genetic landscape of 102/131 MDS patients of the cohort #2 according to their phenotypic pattern of HSPCs repartition. Overall, 85/102 (83.3%) patients carried at least one somatic mutation (Table 1). Patients with an increase of B/NK progenitors or CMPMEP F3 or MPP F3 harbored the lowest number of mutations, whereas those with an increase of MPP F2, MLPs or LMPPs were associated with a higher number of mutations (Supplementary Fig. 4E). Interestingly, the profile with increased of B/NK progenitors was associated with DNMT3A mutations (5/10 patients, p = 0.0058), whereas SF3B1 mutations were more frequent in patients with increased of CMPMEP F3 (4/10 patients; p = 0.0482) (Fig. 2D). The profile with accumulation of CMPMEP F1 was also frequently associated with ASXL1 mutation (7/9 patients; p < 0.0001) and TET2 mutations were more frequently detected in patients with increase of GMPs (15/28 patients; p = 0.0217) (Fig. 2D).

Altogether, these results suggested the amplification of distinct HSPCs sub-populations in MDS and revealed phenotypic signatures associated with IPSS-R stratification and distinct genomic patterns of MDS.

MDS with decreased CD38 + HSPCs heterogeneity have a reduced entropy

To quantify the heterogeneity of HSPCs in the BM, we computed the Shannon entropy based on quantification of the proportion of each HSPCs subpopulations identified by MFC in samples from the cohort #2. Entropy of CD34 + CD38+ and CD34 + CD38-HSPCs was similar in non-MDS samples with or without cytopenia (Fig. 3A). Comparison of entropy values normalized to the maximum possible entropy level of these two fractions of cells showed a significantly lower CD38-HSPCs entropy, which reflects the lower phenotypic diversity of immature multipotent CD38- progenitors compared to more differentiated CD38+HSPCs (Fig. 3B). The entropy of CD38+HSPCs was significantly decreased in MDS samples and also drastically reduced in AML samples (Fig. 3C). Conversely, the CD38-HSPCs entropy was significantly increased in MDS samples (Fig. 3C), attesting the emergence and/or accumulation of one or more CD38-HSPCs sub-populations, creating a heterogeneity not observed in non-MDS samples. In AML samples, the level of CD38-HSPCs entropy remained low, mainly due to the full replacement of normal CD34 + CD38- HSPCs by abnormal CD34 + CD38- leukemic stem cells.

Fig. 3: Reduced HSPCs heterogeneity in MDS can be assessed by a decreased entropy of HSPCs.

BM samples are from cohort #2. A Entropy levels in both CD34 + /CD38- and CD34 + /CD38+ fractions of HSPCs in non MDS samples (CTL) without cytopenia (n = 24, left panel) or with cytopenia (n = 20, right panel). Lines represent means ± s.d. Statistical significance was determined using unpaired two-tailed Mann–Whitney tests. B Comparison of CD38+ and CD38-HSPCs entropy value in non MDS-samples with or without cytopenia. Given the different number of HSPCs sub-populations identified in the CD34 + CD38+ (n = 6) and the CD34 + CD38- (n = 7) fraction of HSPCs, entropy values were normalized to the maximum possible entropy rate (2.58 for CD38+HSPCs and 2.81 for CD38-HSPCs) to allow comparison of entropy rates in these two compartment of cells. Lines represent means ± s.d. Statistical significance was determined using Wilcoxon matched-pairs signed rank tests. C Comparison of CD38+ (left panel) and CD38- (right panel) HSPCs entropy levels in non-MDS samples (with or without cytopenia) (n = 44), in MDS samples (n = 131) and in AML samples (n = 9). Lines represent means ± s.d. Statistical significance was determined using unpaired two-tailed Mann–Whitney tests. D ROC curve for the CD38+HSPCs entropy in the whole cohort of patients with MDS (black curve, n = 131), or MDS low risk (very low/low IPSS-R; orange curve; n = 73) or MDS intermediate risk (blue curve; n = 17) or MDS high risk (high/very high IPSS-R; red curve; n = 33) compared to non-MDS samples with or without cytopenia (n = 44). The Youden index was defined for all points of the ROC curve and the maximum value of the index (152.54) was used as a criterion for selecting the optimum cutoff point of the CD38+HSPCs entropy (2.1) to identify patients with MDS. E Dynamic of CD38+HSPCs entropy during the follow-up of untreated low-risk MDS patients with stable disease (based on a stable BM blast cell count) (n = 24; median follow-up of 11.5 months ; range: 3–40 months); or with evolution attested by increased BM blast cell count (n = 7; median time between diagnosis of MDS and evolution of the disease: 12 months ; range : 9–26 months) and of MDS patients treated by HMA leading to stabilization of the disease based on the BM blast cell count (n = 10; median number of cures: 6 ; range : 2–12). Lines represent means ± s.d. Statistical significance was determined using Wilcoxon matched-pairs signed rank tests. Black arrows indicate the evolution of the mean value of CD38+HSPCs entropy between diagnosis (Diag) and follow-up (FU). Progression Free Survival curves (F) of MDS patients whatever the IPSS-R classification (n = 84) or (G) of patients with LR-MDS (defined as IPSS-R ≤ 3.5) (n = 61), according to the low or normal level of CD38+HSPCs entropy (z-score ≤-2 or >-2 compared to non-MDS samples without cytopenia).

Normal rates of CD38+ and CD38-HSPCs entropy in non-MDS samples without cytopenia were then used as a reference to analyze HSPCs entropy levels in patients with cytopenia related to MDS or not. Whereas CD38+HSPCs entropy was normal in all non-MDS samples with cytopenia, a decrease of CD38+HSPCs entropy (z-score ≤-2) was found in 47.3% of MDS cases (Supplementary Fig. 5A). Increase of CD38-HSPCs entropy (z-score ≥2) was also detected in MDS samples but to a significantly lower frequency (32.8% of cases; p = 0.0230) (Supplementary Fig. 5A). With a ROC analysis, we identified a CD38 + HSPC entropy value < 2.1 allowing the detection of MDS with 86.3% of specificity and 67.2% of sensitivity with higher AUC for intermediate and high-risk MDS (Fig. 3D). However, the increase of CD38-HSPCs entropy was less efficient to diagnose MDS (Supplementary Fig. 5B).

Altogether, these data indicated therefore that HSPCs entropy and more specifically the CD38+HSPCs entropy is a relevant parameter for the diagnosis of MDS.

Reduced CD38 + HSPCs entropy identified patients with specific MDS features

We then investigated whether CD38+HSPCs entropy correlated with clinical and biological features of MDS cases (Table 1). Age was similar in the two groups of MDS patients. A low level of CD38+HSPCs entropy was associated with lower neutrophils and platelets counts, deep anemia and pancytopenia (Table 1). The level of CD38+HSPCs entropy was inversely correlated with CD34+ cells quantified by MFC and with the BM blast count assessed by morphology (Supplementary Fig. 5C). Accordingly, low CD38+HSPCs entropy was associated with the MDS-IB entity of the WHO classification, a poor risk cytogenetic profile and a very high/high/intermediate risk IPSS-R (Table 1). We also analyzed the genetic landscape of MDS patients according to the level of the CD38+HSPCs entropy (Supplementary Fig. 6A). Abnormal CD38+HSPCs entropy was associated with increased proportion of mutated cases and a higher number of mutations (Table 1). Low level of CD38+HSPCs entropy was significantly more frequently associated with somatic mutations in genes encoding for transcription factors (p = 0.0144, Supplementary Fig. 6B). Among these genes, RUNX1 was the most frequently mutated in patients with low CD38+HSPCs entropy (p = 0.0109). ASXL1 (p = 0.0412) and DDX41 (p = 0.0228) mutations were also more frequently detected in patients with abnormal CD38+HSPCs entropy (Supplementary Fig. 6C). Interestingly, we detected in MDS patients harboring a somatic mutation with high variant allele frequency (VAF), a lower CD38+HSPCs entropy (Supplementary Fig. 6D), suggesting that evaluation of HSPCs entropy could reflect the clonal dominance observed in MDS. We also analyzed the association between the level of CD38+HSPCs entropy and patient restratification according to the IPSS-M computed based on these molecular data. The restratification of IPSS-R to IPSS-M was as expected (Supplementary Fig. 7A, B). Low CD38+HSPCs entropy was significantly associated with a moderate high/high/very high IPSS-M (Table 1). Interestingly when focusing on patients which were restratified upstaged, 13/31 (42%) of cases were characterized by low level of CD38+HSPCs entropy with higher proportion in the group of low, intermediate or high IPSS-R (Supplementary Fig. 7C). Furthermore, among patients with LR-MDS based on the IPSS-R ( ≤ 3.5), 5/7 (71%) of cases restratified in the group of HR-MDS based on the IPSS-M (>0) harbored a low level of CD38+HSPCs entropy (Supplementary Fig. 7D). These results suggest therefore that this parameter could allow the identification of some patients who will be restratified upstaged with the IPSS-M classification, especially those who will switch from a LR-MDS to a HR-MDS.

We also followed the level of CD38+HSPCs entropy during the course of the disease. In untreated patients with a stable low-risk MDS, the CD38+HSPCs entropy significantly decreased during follow-up and this was more pronounced in patients with disease progression (Fig. 3E). However, this decrease was no longer observed in patients with hypomethylating agents-mediated stabilization of the disease (Fig. 3E). Finally, we observed that MDS patients with an abnormal low level of CD38+HSPCs entropy computed at diagnosis had a significantly higher risk of progression and this was also true for the sub-group of patients with LR-MDS (Fig. 3F, G). Altogether, these results indicated that the detection of a loss of CD38+HSPCs heterogeneity through entropy calculation is an attractive biomarker for predicting the outcome of MDS patients, including from early stage of the disease and for monitoring their follow-up.

Validation of CD38 + HSPCs entropy as a diagnostic tool for MDS detection

To validate the importance of CD38+HSPCs entropy in the diagnosis of MDS, we took advantage of an independent cohort of 764 patients (cohort #3 in Supplementary Table 1 and Table 1). The panel markers (combination #3, Supplementary Table 2) and the analysis strategy (Supplementary Fig. 8) we used for these samples led to the identification of 6 different CD38+HSPCs subpopulations (Supplementary Table 3). We confirmed in this cohort that the level of CD38+HSPCs entropy was similar in non-MDS samples with or without cytopenia, and significantly decreased in MDS samples (Supplementary Fig. 9A). ROC curve evaluation confirmed the capacity of CD38+HPSCs entropy to diagnose MDS including for low-risk MDS (Supplementary Fig. 9B). Samples from non-MDS patients without cytopenia were then used to define the normal level of CD38 + HSPC entropy and calculation of a z-score < -2 identified 198 (46%) MDS patients with an abnormal low level of CD38+HSPCs entropy (Supplementary Fig. 9C). Similarly, the decreased CD38+HSPCs entropy level was associated with specific features of MDS such as deeper thrombocytopenia and neutropenia (Table 1). CD38 + HSPC entropy of MDS patients was closely related to the level of BM blasts with a negative linear correlation (Supplementary Fig. 9D). Consistent with this, low level of CD38+HSPCs entropy was also associated with the “MDS with increased blasts” entity of the WHO classification and with the high/very high-risk IPSS-R (Table 1).

These results confirmed therefore in this independent cohort, the capacity of CD38+HSPCs entropy calculation to identify MDS patients with a more advanced stage of their disease.

Low CD38 + HSPCs entropy is an independent negative prognostic marker for MDS patients

Finally, we wondered whether the CD38 + HSPCS entropy could be an independent prognostic parameter in MDS. We thus analyzed the follow-up of 584 MDS patients (median follow-up of 28.4 months) classified according their low or normal level of CD38+HSPCs entropy computed at diagnosis. The impact of the IPPS-R classification on LFS and OS was as expected (Supplementary Fig. 9E). Interestingly, the median PFS of patients with low CD38+HSPCs entropy was significantly decreased (14.9 months vs not reached (NR); Hazard Ratio [HR]:2.40, p < 0.0001) (Fig. 4A). The median LFS of patients with low CD38+HSPCs entropy was also significantly decreased (22.1 months vs NR; HR:2.62, p < 0.0001) (Fig. 4A). One-year (y) and 3-y LFS were respectively of 64% and 40% in the low entropy patient’s group and 83% and 72% for patients with normal entropy (Fig. 4A). The median OS of patients with a low CD38+HSPCs entropy was also significantly shorter (19.8 vs 33.2 months; HR:1.58, p = 0.0001) (Fig. 4A). One-y and 3-y OS were respectively 69% and 28% in low entropy patients and 78% and 49% in normal entropy patients (Fig. 4A). Interestingly, among the group of patients with LR-MDS (defined with IPSS-R ≤ 3.5), CD38+HSPCs entropy was still significantly associated to a shorter PFS (HR: 5.35; 95% confidence interval [95%CI]: 2.73–10.47, p < 0.0001), LFS (HR: 6.89; 95%CI: 2.98–15.91, p < 0.0001) and OS (HR: 2.37; 95%CI: 1.36–4.10, p = 0.0019) (Fig. 4B). Notably, LFS at 1 and 3 years were 96% and 93% respectively for MDS with normal entropy, compared to 82% and 57% for patients with low CD38+HSPCs entropy. In contrast, no or low significant differences of OS (p = 0.76), PFS (HR: 1.46; 95%CI: 1.07–2.00, p = 0.019) and LFS (HR: 1.59; 95%CI: 1.13–2.23, p = 0.0077) were observed between patients with low or normal level of CD38+HSPCs entropy among the group of patients with higher-risk MDS (HR-MDS) (Supplementary Fig. 10). In light of these results, we then tested whether the CD38+HSPCs entropy was an independent predicting factor for adverse outcome of these patients with LR-MDS. With a multivariate Fine and Gray analysis, low CD38+HSPCs entropy was found to be strongly independently associated with a shorter PFS (HR = 5.21; 95%CI: 2.58–10.51, p < 0.0001) and a shorter LFS (HR = 7.77; 95%CI: 2.81–21.45, p < 0.0001) (Fig. 4C). In multivariate Cox analysis, low CD38+HSPCs entropy was also significantly associated with a worse OS (HR = 2.03; 95%CI: 1.17–3.51, p = 0.012) (Fig. 4C).

Fig. 4: Low level of CD38 + HSPCs is an independent negative prognostic factor for LR-MDS.

BM samples are from cohort #3. Kaplan-Meier estimates of survival outcomes according to CD38+ HSPCs entropy level of (A) all patients (n = 584) or (B) of patients with LR-MDS (IPSS-R ≤ 3.5) (n = 213). C Multivariate analysis exploring factors that affect PFS, LFS and OS of LR-MDS patients (n = 213). Good or very good cytogenetic risk categorization was determined according the Comprehensive Cytogenetic Scoring System (CCSS) for MDS.

Altogether, these results confirmed that decreased CD38+HSPCs entropy is a strong predictor of adverse outcomes of LR-MDS independently of others parameters included in the IPSS-R risk stratification.