1. Computational and Systems Biology
  2. Genetics and Genomics

Uncovering perturbations in human hematopoiesis associated with healthy aging and myeloid malignancies at single-cell resolution

  1. Marina Ainciburu
  2. Teresa Ezponda
  3. Nerea Berastegui
  4. Ana Alfonso-Pierola
  5. Amaia Vilas-Zornoza
  6. Patxi San Martin-Uriz
  7. Diego Alignani
  8. Jose Lamo-Espinosa
  9. Mikel San-Julian
  10. Tamara Jiménez-Solas
  11. Felix Lopez
  12. Sandra Muntion
  13. Fermin Sanchez-Guijo
  14. Antonieta Molero
  15. Julia Montoro
  16. Guillermo Serrano
  17. Aintzane Diaz-Mazkiaran
  18. Miren Lasaga
  19. David Gomez-Cabrero
  20. Maria Diez-Campelo
  21. David Valcarcel
  22. Mikel Hernaez
  23. Juan P Romero  Is a corresponding author
  24. Felipe Prosper  Is a corresponding author
  1. Area de Hemato-Oncología, Centro de Investigación Médica Aplicada, Universidad de Navarra, Instituto de investigación sanitaria de Navarra (IDISNA), Spain
  2. Centro de Investigación Biomédica en Red de Cáncer, Spain
  3. Clinica Universidad de Navarra, Spain
  4. Flow Cytometry Core, Universidad de Navarra, Spain
  5. Hospital Universitario de Salamanca, Spain
  6. Red de Investigación Cooperativa en Terapia Celular TerCel, ISCIII., Spain
  7. Department of Hematology, Vall d'Hebron Hospital Universitari, Spain
  8. Computational Biology Program, Universidad de Navarra, Spain
  9. Translational Bioinformatics Unit, NavarraBiomed, Spain
  10. Biological & Environmental Sciences & Engineering Division, King Abdullah University of Science and Technology, Saudi Arabia
https://doi.org/10.7554/eLife.79363
Share this article
Cite this article
  1. Marina Ainciburu
  2. Teresa Ezponda
  3. Nerea Berastegui
  4. Ana Alfonso-Pierola
  5. Amaia Vilas-Zornoza
  6. Patxi San Martin-Uriz
  7. Diego Alignani
  8. Jose Lamo-Espinosa
  9. Mikel San-Julian
  10. Tamara Jiménez-Solas
  11. Felix Lopez
  12. Sandra Muntion
  13. Fermin Sanchez-Guijo
  14. Antonieta Molero
  15. Julia Montoro
  16. Guillermo Serrano
  17. Aintzane Diaz-Mazkiaran
  18. Miren Lasaga
  19. David Gomez-Cabrero
  20. Maria Diez-Campelo
  21. David Valcarcel
  22. Mikel Hernaez
  23. Juan P Romero
  24. Felipe Prosper
(2023)
Uncovering perturbations in human hematopoiesis associated with healthy aging and myeloid malignancies at single-cell resolution
eLife 12:e79363.
https://doi.org/10.7554/eLife.79363
  2. Download BibTeX
  3. Download .RIS

Abstract

Early hematopoiesis is a continuous process in which hematopoietic stem and progenitor cells (HSPCs) gradually differentiate toward specific lineages. Aging and myeloid malignant transformation are characterized by changes in the composition and regulation of HSPCs. In this study, we used single-cell RNA sequencing (scRNA-seq) to characterize an enriched population of human HSPCs obtained from young and elderly healthy individuals.

Based on their transcriptional profile, we identified changes in the proportions of progenitor compartments during aging, and differences in their functionality, as evidenced by gene set enrichment analysis. Trajectory inference revealed that altered gene expression dynamics accompanied cell differentiation, which could explain aging-associated changes in hematopoiesis. Next, we focused on key regulators of transcription by constructing gene regulatory networks (GRNs) and detected regulons that were specifically active in elderly individuals. Using previous findings in healthy cells as a reference, we analyzed scRNA-seq data obtained from patients with myelodysplastic syndrome (MDS) and detected specific alterations of the expression dynamics of genes involved in erythroid differentiation in all patients with MDS such as TRIB2. In addition, the comparison between transcriptional programs and GRNs regulating normal HSPCs and MDS HSPCs allowed identification of regulons that were specifically active in MDS cases such as SMAD1, HOXA6, POU2F2, and RUNX1 suggesting a role of these transcription factors (TFs) in the pathogenesis of the disease.

In summary, we demonstrate that the combination of single-cell technologies with computational analysis tools enable the study of a variety of cellular mechanisms involved in complex biological systems such as early hematopoiesis and can be used to dissect perturbed differentiation trajectories associated with perturbations such as aging and malignant transformation. Furthermore, the identification of abnormal regulatory mechanisms associated with myeloid malignancies could be exploited for personalized therapeutic approaches in individual patients.

Editor's evaluation

This study generated an important single-cell transcriptome dataset using young/aged hematopoietic stem/progenitor cells obtained from normal individuals and those with MDS. The new resource provides a convincing dataset to understand a unique transcriptional landscape in elderly individuals, compared to young individuals, proving the hematopoietic aging at a transcriptome level. This manuscript will be of interest to readers in the field of hematopoiesis and associated diseases, aging, and single-cell RNA sequencing.

https://doi.org/10.7554/eLife.79363.sa0

eLife digest

Our blood contains many different types of cells; red blood cells carry oxygen through the body, platelets help to stop bleeding and a variety of white blood cells fight infections. All of these critical components come from a pool of immature cells in bone marrow, which can develop and specialise into any of these. However, as we get older, these immature cells can accumulate damage, including mutations in specific genes. This increases the risk of diseases such as myelodysplastic syndromes (MDS), a type of cancer in which the cells cannot develop and the patient does not have enough healthy mature blood cells.

The changes in gene activity in the immature cells have previously been studied using samples from young and elderly people, as well as individuals with MDS. These studies examined large numbers of cells together, revealing differences between young and elderly people, and individuals with MDS. However, this does not describe how the different types alter their behaviour.

To address this, Ainciburu, Ezponda et al. used a technique called single-cell RNA sequencing to study the gene activity in individual immature blood cells. This revealed changes associated with maturation that may account for the different combinations of cell populations in younger and older people. The results confirmed findings from previous studies and suggested new genes involved in ageing or MDS. Ainciburu, Ezponda et al. used these results to create an analytical system that highlights gene activity differences in individual MDS patients that are independent of age-related changes.

These results provide new insights that could help further research into the development of MDS and the ageing process. In addition, scientists could study other diseases using this approach of analysing individual patients’ gene activity. In future, this could help to personalise clinical decisions on diagnosis and treatment.

Introduction

Mature blood and immune cells are generated by hematopoiesis, which is a well-characterized process that has been studied for more than a century (Jagannathan-Bogdan and Zon, 2013). Classically, hematopoiesis has been modeled as a stepwise differentiation process, at the core of which reside hematopoietic stem cells (HSCs), which are common precursors with self-renewal capacity. A tree-like hierarchy arises from HSCs, in which lineage commitment occurs at binary branching points, giving rise to functionally and phenotypically homogeneous progenitor populations (Laurenti and Göttgens, 2018; Notta et al., 2016). However, recent single-cell studies have questioned the validity of the classical model of hematopoiesis and have revealed heterogeneity within the HSC compartment and within progenitor populations that were previously considered homogeneous (Haas et al., 2018). Furthermore, it is difficult to establish boundaries between populations, which has led to the replacement of this concept by the idea of smooth transitions. Thus, early hematopoiesis is now viewed as a continuous landscape composed of undifferentiated HSPCs with a variable degree of priming toward specific lineages (lymphoid, myeloid, or erythroid) (Velten et al., 2017; Karamitros et al., 2018; Watcham et al., 2019; Buenrostro et al., 2018).

During aging, the hematopoietic system undergoes various changes. There is evidence of an aging-related increase in the relative number of HSCs, that presumably aims to compensate for a loss of repopulation ability (Dykstra et al., 2011; Pang et al., 2011). Despite this increase in the number of HSCs, the total bone marrow cellularity decreases (Ogawa et al., 2000). Additionally, a loss of lymphocyte production and a skewing toward myeloid differentiation have been demonstrated (Young et al., 2016). Moreover, recent studies have revealed an increase in the number of platelet primed HSCs (Grover et al., 2016). A general loss of immune function that affects both innate and adaptive immunity has also been associated with aging (Weiskopf et al., 2009). The molecular basis of these phenotypic changes includes an increased rate of random mutations in hematopoietic progenitors. Clonal hematopoiesis is a common trait of elderly individuals, in which blood cell subpopulations are clonally derived from a single HSC or progenitor with acquired mutations. Notably, some of these mutations are also associated with hematopoietic malignancies (Xie et al., 2014; Jaiswal et al., 2014; McKerrell et al., 2015; Genovese et al., 2014). Additionally, aging is also accompanied by transcriptional dysregulation. Gene expression analyses have revealed changes in cell cycle regulators (Kowalczyk et al., 2015), higher expression of myeloid signatures (Grover et al., 2016; Kowalczyk et al., 2015) and genes associated with leukemia (Rossi et al., 2005), stronger response to inflammatory stimuli (Martinez-Jimenez et al., 2017; Mann et al., 2018), and upregulation of pathways such as nuclear factor kappa beta (NF-κB) or tumor necrosis factor alpha (TNF-α) and downregulation of DNA repair (Chambers et al., 2007). Lastly, epigenetic modifications that are in line with the transcriptional lesions detected have been observed in HSPCs obtained from elderly individuals (Sun et al., 2014).

Aging is associated with a higher risk of developing myeloid malignancies (Zeidan et al., 2019; Deschler and Lübbert, 2006), suggesting a strong predisposition of hematopoietic cells from elderly individuals to lead to further alterations. Myelodysplastic syndromes (MDS) are among the main aging-related hematological disorders. MDS are characterized by ineffective hematopoiesis and predisposition to transformation into acute myeloid leukemia (Shallis et al., 2018), a highly aggressive neoplasm.

Collectively, previous evidence suggests the existence of a progressive decline in the hematopoietic system with age, which can result in disease if certain pathological events occur. HSC damage is the primary consequence of this process and its outcomes range from lineage bias to differentiation blockade and leukemic transformation (Chung and Park, 2017).

In this work we present a computational roadmap that can be used as a basis to fully exploit the potential of single-cell RNA sequencing (scRNA-seq) data. We applied a set of computational algorithms to assess cellular subpopulations, differentiation trajectories, and gene regulatory elements in the human hematopoietic system. This proposed methodology not only allows the characterization of healthy or control scenarios, but also the identification of specific perturbations. This is demonstrated by describing aging-dependent and pathological alterations in hematopoiesis.

Results

Transcriptional profiling of human young and elderly hematopoietic progenitor systems

To investigate the aging-dependent changes in the hematopoietic system, we performed scRNA-seq of bone marrow CD34+ cells obtained from five young (18–20 years of age) and three elderly (>65 years of age) healthy donors (Figure 1A). Briefly, single-cell libraries were prepared using a Chromium controller instrument (10× Genomics) and sequenced up to an average depth of 30,000 reads per cell. A total of 34,590 and 40,641 cells were profiled from young and elderly donors, respectively. We first constructed an integrated dataset with cells from both young and elderly donors using an integration procedure implemented in the Seurat R package (Stuart et al., 2019). Subsequently, we subjected the integrated dataset to quality control filtering and dimensionality reduction to obtain a reference map for visualization (Supplementary file 1). Secondary analyses including unsupervised clustering, identification of cluster markers, trajectory inference, and gene regulatory network (GRN) reconstruction were performed independently for each group.

Figure 1 with 4 supplements see all
Download asset Open asset
Transcriptional profiling of CD34+ cells from young and elderly healthy donors.

(A) CD34+ cells were obtained from bone marrow aspirates of young (n=5) and elderly (n=3) donors and subjected to single-cell RNA sequencing. (B) UMAP plot with young cells colored according to unsupervised clustering results (left) and elderly cells labeled using an in-house cell classifier (right). (C) Dot plot of cluster markers (adjusted p-value <0.05) for the different cellular subpopulations identified. Dot size represents the percentage of cells that express each marker, and color represents scaled expression values. (D) Bar plots showing the proportion of cells assigned to each cellular subpopulation for each donor independently. (E) Dot plot of enriched terms after performing gene set enrichment analysis (GSEA) for each identified cluster. Dot color represents the enriched group, size indicates the NES absolute value, and transparency indicates the adjusted p-value.

After integration, we extracted the cells obtained from young donors and subjected them to unsupervised clustering and differential expression analysis to identify cluster markers for manual cell-type annotation and labeling (Supplementary file 2). This analysis identified 14 cellular subpopulations comprising the landscape of HSCs, early progenitors: megakaryocyte-erythroid progenitors (MEPs), lympho-myeloid primed progenitors (LMPPs), common lymphoid progenitors (CLPs), and granulocyte-monocyte progenitors (GMPs), and cells already committed to specific lineages (pro-B cells, monocytic, erythroid, megakaryocytic, basophil and dendritic cell progenitors) (Figure 1B). Early progenitors were characterized by the expression of genes such as CRHBP, HOPX (initial lympho-myeloid differentiation), and PBX1 (initial megakaryocyte and erythroid lineage). More mature progenitors presented myeloid (MPO, CTSG, PRTN3, LYZ, and IRF8), lymphoid (DNTT, VPREB1, and EBF1), erythroid (HBD and CA1), and basophil (HDC and MS4A2)-specific markers (Figure 1C). Once cellular subpopulations from the young donors were identified, we labeled cells of elderly individuals using a classification method based on a logistic regression with elastic-net regularization (see Materials and methods). Regularized logistic regression has been previously applied to scRNA-seq data, due to its high interpretability and good performance with sparse input (Torang et al., 2019; Nguyen et al., 2018). Briefly, we created one classifier per cell type that receives as input the normalized gene by cell expression matrix and returns a per-cell probability to be a specific cell type. Using this method we classified the integrated elderly dataset to assign cell-type probabilities (Figure 1—figure supplement 1A). The performance of our classifier was determined by using a publicly available dataset containing 8176 human CD34+ progenitors with known identities (Granja et al., 2019). This dataset was labeled using our method and Seurat (Figure 1—figure supplement 1B) for benchmarking purposes. We obtained the proportion of cells classified within each cell type using both approaches (Figure 1—figure supplement 1C). Overall, both methods provided similar results. We observed the biggest differences when assessing progenitor cell types such as HSCs, MEPs, and LMPPs. Seurat classified 38% of HSCs in the reference as MEPs, compared to 12% by GLMnet. We also observed that 83.31% of LMPPs were correctly classified by GLMnet compared to 57.91% by Seurat. The lymphoid lineage showed the biggest proportion changes; however, this is related to the low number of B cells in the reference. Based on these results, we decided to use our method in the rest of the analyses.

When we applied the GLMnet classifier to the elderly individuals, we noted a significant increase in the proportion of HSCs and a reduction in the number of both committed lymphoid and myeloid lineages (adjusted p-value <0.05, Supplementary file 3) as well as an increase in the proportion of erythroid-committed cells in the overall proportion, both in the integrated and individual datasets (Figure 1D and Figure 1—figure supplement 2). We used flow activated cell sorting (FACS) as an orthogonal method and observed similar results for HSCs. However, changes in the proportion of GMPs and MEPs were less obvious than in the case of the transcriptomic analysis (Figure 1—figure supplement 3).

Next, we checked for functional differences between the young and elderly populations using gene set enrichment analysis (GSEA) after performing differential expression per cluster and using the genes overexpressed in each condition (young and elderly) within each cell type (Supplementary file 4) as an input to GSEA. Cells from elderly donors exhibited a generalized enrichment of pathways activated in response to external molecules and insults, such as TNF-α, transforming growth factor beta (TGF-β), hypoxia, or inflammation. Furthermore, the p53 pathway and apoptosis programs, which are activated by stressor stimuli, were also enriched in these elderly cells (Figure 1E). In particular, the TGF-β response and apoptosis genes, together with genes upregulated in response to ultraviolet radiation, were specially enriched in aging-associated HSCs. These results suggested the presence of an aging-related response to the known more inflammatory microenvironment that is present in elderly individuals (Leimkühler and Schneider, 2019). Shifts in the abovementioned stress-related pathways were associated with increased expression in the elderly of potential drivers of such pathways, including JUNB, FOSB, FOS, ID2, or DDIT4 (Figure 1—figure supplement 4). Hematopoietic progenitors from young donors were characterized by an increase in metabolism, including signatures related to glycolysis, fatty acid metabolism, and oxidative phosphorylation, in components of the respiratory chain (UQCR, NDUF, and COX), and the glycolytic enzymes PGK1 and TPI1. These findings point toward an increased metabolic activity in progenitors from young donors. We also found that cell cycle progression (E2F targets) and proliferation (MYC targets) pathways were enriched in multiple cell populations from young donors (Figure 1E), whereas MYC expression was generally downregulated in cells from elderly individuals, suggesting an aging-mediated decreased proliferative activity. DNA repair was also found to be more active in cells isolated from young individuals. This goes in agreement with the known predisposition of aging-associated progenitors to accumulate genetic lesions (Jaiswal and Ebert, 2019; Mohrin et al., 2010; Beerman et al., 2014).

Overall, we found that the proportion of cell types changed with age, observing a decrease in the most mature lymphoid and myelomonocytic compartments. Furthermore, GSEA revealed aging-related transcriptional alterations in all compartments, which suggests an altered biological behavior of HSPCs with age.

Trajectory inference revealed age-specific drivers along lineages

To study the distinct differentiation trajectories observed during hematopoiesis and, specifically, alterations during aging, we subjected the single-cell expression datasets to trajectory inference using STREAM (Chen et al., 2019). We first inferred the differentiation trajectories of the young donors to obtain a reference that could serve as a basis for comparison. These trajectories revealed a common starting point that comprised early progenitors (HSCs, LMPPs, and MEPs). This initial root reached the branching point that divides cellular differentiation into the three main branches that resemble the main hematopoietic lineages: the myeloid, erythroid, and lymphoid lineages (Figure 2—figure supplement 1A). The identity of these branches was confirmed by plotting the expression profiles of previously described genes, such as IRF8, GATA1, and EBF1 (Figure 2—figure supplement 1B). Then, in order to identify aging-dependent alterations, the elderly dataset was projected onto the reference trajectories using the approach described in STREAM (Figure 2—figure supplement 1C). We observed an increased proportion of HSCs in the root node, a reduced number of cells biased to both the myeloid and lymphoid lineages, and an increased number of progenitor cells committed to the erythroid compartment in the elderly individuals (Figure 2—figure supplement 1D).

Next, we aimed to identify shifts in the gene expression dynamics along differentiation trajectories. We applied the Palantir algorithm (Setty et al., 2019), with which we recovered a general pseudotime (Figure 2A) and differentiation potential (Figure 2B) for each cell. To measure the similarity between both methods, we performed a correlation analysis and noticed that the pseudotime values were strongly correlated (r=0.78) (Figure 2—figure supplement 1E). We observed significant differences in the distributions of differentiation potential and pseudotime, with aging-associated HSPCs accumulating greater differentiation potential and lower pseudotime (Figure 2—figure supplement 1F). This suggested an immature cell state, which was consistent with our previous analysis. We could also define six trajectories corresponding to six committed compartments: that is, erythroid, lymphoid, dendritic cells, monocytes, basophils, and megakaryocytes, and assigned each cell a probability of belonging to a specific branch (Figure 2C).

Figure 2 with 1 supplement see all
Download asset Open asset
Trajectory inference of the hematopoietic lineages at single-cell resolution.

(A) UMAP plots showing the results from applying Palantir algorithm to young and elderly cells. For both datasets, a hematopoietic stem cell (HSC) was established as initial state, based on UMAP coordinates. Final states were only indicated for the elderly dataset, as the UMAP nearest neighbors to the six young final points. Cells are colored by pseudotime and (B) differentiation potential. (C) Branch probabilities for each of the six differentiation paths retrieved. (D) Scatter plot of pseudotime vs. branch probabilities for the monocytic trajectory obtained using Palantir for young and elderly donors. Color represents the cellular subpopulation. (E) Heatmap of gene expression trends for dynamic genes along the monocytic trajectory in young and elderly donors. The columns are arranged according to pseudotime values, and the rows are grouped according to gene clustering results. A summary of enriched terms for the gene clusters in young donors is shown. (F) Expression trends in the comparison of young and elderly cells regarding the different genes involved in the monocytic trajectory (NS = not significant, *adjusted p-value <0.05, **adjusted p-value <0.01, ***adjusted p-value <0.001).

As an example of study of a particular trajectory, we focused on early monocytic differentiation, as it appeared to be impaired in the elderly donors. We plotted, in a per-cell basis, the associated pseudotime along the monocytic trajectory with the Palantir probability of reaching the final stage of the studied route (Figure 2D). This analysis revealed that HSCs and LMPPs from elderly donors had a higher probability of attaining the monocyte progenitor state, suggesting a stronger bias toward the monocytic compartment. However, GMPs displayed the opposite behavior which suggested that although aging-associated progenitors appeared to have a stronger initial bias toward this lineage, a large number of more advanced progenitor cells lost the capacity for monocytic differentiation. This result is in line with the aging-associated decrease in monocytic progenitors described above.

To determine the transcriptional lesions that potentially alter monocytic differentiation during aging, we computed gene expression trends and clustered them based on their dynamics along the monocytic differentiation trajectory, following the Palantir pipeline. We characterized young gene clusters by over-representation analysis. Terms related to the immune system and monocytic function were enriched as the trajectory progressed, such as antimicrobial response or regulation of the actin cytoskeleton in intermediate stages, and antigen processing and presentation, or interferon gamma signaling, in gene clusters expressed near the terminal state. On computing trends for the cells of elderly donors, we observed that the overall behavior was similar. Nevertheless, genes with altered expression patterns in the elderly donors were visible (Figure 2E). Several myeloid differentiation markers (CST7, PRTN3, MPO, CD74, CALR, and GNAS) were expressed at lower levels across monocytic differentiation in the elderly donors, suggesting a less differentiated state. Accordingly, genes characteristic of stem or very early progenitor cells (MYCT1, MLLT3, and ALDH1A1) showed higher levels of expression in cells from the elderly donors across differentiation. Genes related to stress and inflammation response (FOS, JUNB, TSC22D3, DUSP1, and DDIT4), or monocyte chemotaxis and extravasation (ANXA1, LGALS3, and JAML), exhibited higher expression levels at the start of differentiation in the elderly, with several genes showing decreased expression at the terminal state (Figure 2F), further reinforcing the idea of a loss of capacity for monocytic differentiation in cells from elderly donors.

These analyses of the differentiation trajectory provided data regarding the genes that may be involved in the aberrant differentiation of aging-associated HSCs. However, the regulation of these transcriptional programs cannot be inferred from this analysis.

GRNs guiding young and elderly hematopoiesis

To elucidate the regulatory mechanisms underlying healthy hematopoietic differentiation, we constructed lineage-specific GRNs using SCENIC (Van de Sande et al., 2020) for each of the datasets independently. Briefly, we obtained the set of activation regulons (transcription factor [TF] and their associated targets) for each of the cellular subpopulations and binarized its activity (on/off) in a per-cell basis using AUC values provided by SCENIC (through the AUCell algorithm). This approach enabled quantification of the proportion of cells that displayed an activated state for each regulon in each cluster. We then selected the top ranked regulons based on the regulon specificity score (RSS) (Figure 3—figure supplement 1A and B) to identify specific regulatory mechanisms per cell type. For the young donors, we obtained a series of GRNs guided by TFs known to be relevant for specific cell populations, such as HOXA9 in early progenitor cells, or CEBPG, PAX5, and GATA1 in the myeloid, lymphoid, and erythroid compartment, respectively (Figure 3A). For some regulons AUC values were highly correlated with the corresponding gene expression profiles of the guiding TF, whereas others such as CEBPG or TCF3 displayed a more specific dynamism of GRN activity that is not observed with gene expression. For each of the hematopoietic lineages, we observed sets of regulons that regulated the transcriptional state of specific cellular differentiation programs and also the presence of common targets among different TFs (Figure 3B). This observation indicates that a single TF might not only regulate cellular differentiation toward a specific route, but also contribute to other regulatory elements in other differentiation trajectories.

Figure 3 with 1 supplement see all
Download asset Open asset
Gene regulatory network reconstruction of hematopoietic cellular populations.

(A) (Left) Heatmap showing the proportion of cells per cluster that have an activated state for different regulons in young cells. (Right) UMAP plots with normalized expression and AUC values for specific transcription factors. (B) Gene regulatory network of the identified regulons for the hematopoietic system in young donors. Regulons were trimmed to include only targets with an importance score higher than the third quartile in each regulon. Node shape denotes gene-type identity, and color denotes cell population. Any target that can be assigned to multiple transcription factors is colored in gray. (C) (Left) Heatmap showing the proportion of cells per cluster that have an activated state for different regulons in elderly cells. (Right) UMAP plots with normalized expression and AUC values for specific transcription factors. (D) Gene regulatory network of the identified regulons for the hematopoietic system in elderly donors. Regulons were trimmed to include only the targets with an importance score higher than the third quantile in each regulon. Node shape denotes gene-type identity, and color denotes cell population. Any target that can be assigned to multiple transcription factors is colored in gray. (E) Bar plot with enriched gene ontology categories after over-representation analysis. Categories are grouped per cell type, and color denotes the enriched group. Bar length represents statistical significance of the enrichment, as -log10 p-value.

We performed the same analysis using the elderly dataset and observed that progenitor populations displayed a reduced proportion of cells with activated regulons. Specifically, HSCs showed a less active state of TFs, such as MECOM and GATA3, that were deemed specific to young donors (Figure 3C). In already committed cellular subpopulations, we observed similar proportions of activated cells between young and elderly donors, with the erythroid lineage showing the most similar profiles. However, progenitors (GMP, MEP, and CLP cells) from elderly donors showed a lower number of cells with specific regulons activated. Overall, we observed that the predicted regulatory network of elderly HSCs (Figure 3D) appeared as an isolated network compared to the young GRN. This finding could result in the loss of co-regulatory mechanisms in the elderly donors.

We then performed gene ontology enrichment analyses using as input the genes composing the highest ranked regulons in early progenitor compartments (HSC, LMPP, GMP, CLP, and MEP cells). HSCs from young donors displayed enriched terms related to the differentiation of hematopoietic lineages, such as myeloid cell differentiation, lymphocyte cell differentiation, and regulation of hematopoiesis, whereas HSCs from the elderly donors did not exhibit such enrichment (Figure 3E), suggesting a putative alteration of differentiation potential among aged HSCs. Although no increase in the percentage of LMPPs was observed in the elderly donors, this analysis indicated a clear enrichment in terms related to DNA replication in the elderly but not in the young donors, suggesting an aging-mediated alteration of the proliferative rate of this cell type. More mature progenitors also showed a youth-specific enrichment of differentiation processes such as lymphocyte differentiation, B-cell receptor signaling pathway, or T- and B-cell activation for CLPs, and erythrocyte differentiation and homeostasis for MEPs. This suggests that in aging-associated progenitor cells, relevant regulators of hematopoiesis undergo functional alterations that result in the loss of expression of genes that are required for proper hematopoietic differentiation. Collectively, these results show aging-dependent alterations in the GRNs that guide hematopoiesis, which may be associated with the diminished differentiation capabilities that early progenitors present in the elderly.

scRNA-seq analysis of MDS specimens reveals molecular lesions affecting normal hematopoietic differentiation

To further demonstrate the potential of the used computational methods in the prediction of the mechanisms underlying aberrant hematopoietic differentiation, we explored the transcriptional alterations that distinguish normal hematopoiesis during aging from abnormal hematopoiesis associated with the development of MDS. To this end, we performed scRNA-seq from bone marrow CD34+ cells obtained from four patients with MDS. In order to homogenize the group of study, and due to the great clinical and molecular heterogeneity of this disease, we focused our analyses in patients with a diagnosis of MDS with multilineage dysplasia.

We recovered 41,749 cells that passed the quality control filters (Supplementary file 5). All the analysis for the MDS samples was performed following the same methods used for the elderly individuals. For cell-type annotation, we predicted cell labels using our GLMnet classifier (Figure 4A). We noted the absence of lymphoid compartments, considerable reduction in the number of HSCs, and increase in LMPPs and GMPs subpopulations compared to healthy elderly individuals. The total percentage of erythroid compartments varied among patients, although in general it was similar to that of normal counterparts (Figure 4B); nevertheless, the proportion of specific erythroid populations varied, with most MDS patients showing less MEPs and more late erythroid progenitors.

Figure 4 with 1 supplement see all
Download asset Open asset
Computational analysis of pathological conditions, including myelodysplastic syndromes (MDS) and acute myeloid leukemia.

(A) UMAP plot of CD34+ cells from MDS (n=4). Cells are colored according to identity, as assessed using a previously described cell-type classification method. (B) Bar plots showing the proportion of cells assigned to each cellular subpopulation for each donor independently. Color denotes the cellular subpopulation. (C) Gene set enrichment analysis (GSEA) results after performing differential expression between MDS and elderly donors. Dot color represents enrichment direction, transparency the statistical significance, and size NES absolute value. (D) Expression trends in the comparison of healthy and pathological cells regarding the different genes involved in the erythroid trajectory (NS = not significant, *adjusted p-value <0.05, **adjusted p-value <0.01, ***adjusted p-value <0.001). (E) Heatmap showing the proportion of cells per cluster that had an activated state for different regulons in the four samples of patients with MDS among AML cells.

Next, we carried out GSEA of genes differentially expressed between healthy elderly donors and each of the patients for the detected subpopulations. Results showed potential aberrant functionality of MDS cells. The most evident alteration was the enrichment of MDS patients in genes related to interferon alpha and gamma response, which was more evident for cases MDS1 and MDS2. This observation is in accordance with previous reports demonstrating increased inflammatory signaling in the disease (Kim et al., 2015; Gañán-Gómez et al., 2015; Ivy and Brent Ferrell, 2018). Interestingly, genes associated with oxidative phosphorylation, a process that is very relevant for the metabolism of several types of cancer and that is considered as an emerging target (Ashton et al., 2018), were more prominent in MDS. MDS patients also showed alterations of hallmarks related to cell proliferation, including E2F targets, mitotic spindle, or G2/M checkpoint, suggesting an aberrant proliferative activity of hematopoietic progenitors in these patients (Figure 4C). We also performed GSEA using the young donors as controls, and again found enrichment of interferon response in MDS patients. However, we observed variability in other potentially altered pathways, highlighting the importance of using age-matched controls for comparisons (Figure 4—figure supplement 1A).

Most MDS patients are characterized by defects in erythropoiesis, showing dysplasia and/or citopenia of this lineage. Thus, as an example of the applicability of the trajectory inference analyses to identify MDS-related transcriptional alterations, we explored gene dynamics along the HSC-erythroid branch. We reconstructed the erythroid trajectory and computed gene trends along pseudotime using Palantir (Figure 4—figure supplement 1B). Genes that were specifically expressed by any of the progenitors belonging to this branch (HSCs, MEPs, and erythroid progenitors) were selected and clustered according to their expression pattern (Figure 4—figure supplement 1C). Although we did not detect broad changes in the transcriptional profiles among the conditions, we identified specific genes participating in erythroid differentiation that displayed different dynamics from both young and elderly healthy samples (Figure 4D). Examples of genes with altered dynamics included increased levels of genes with a negative role in erythropoiesis such as JUN and YBX1 (Bhullar and Sollars, 2011; Lee et al., 2014) and NME4, previously associated with poor prognosis in MDS (Kracmarova et al., 2008). Moreover, we also observed decreased expression across the trajectory of factors that promote erythroid differentiation, such as TRIB2, PHF6, and PDCD4 (Loontiens et al., 2020; Cho et al., 2015) and involved in erythroleukemia: PVT1 (Salehi and Sharifi, 2018). Interestingly, whereas some of the genes showed aberrant dynamics in all the patients analyzed (i.e. TRIB2), most of them were altered in a subset or in individual patients, reinforcing the heterogeneity of MDS at the molecular level. Despite these changes, we observed that erythroid specific markers such as ANK1, CA2, or AHSP followed trends similar to those of healthy HSPCs, which is in agreement with the apparently normal progression of differentiation at the progenitor stages. These results indicate that, although at very early progenitor stages some drivers of erythroid differentiation exhibit normal trends of expression, others show clear abnormalities that may manifest in more mature stages as erythroid differentiation is altered and may be responsible for the anemia and/or erythroid dysplasia that characterizes MDS patients.

Using SCENIC we analyzed the transcriptional programs regulating HSPCs and reconstructed GRNs by extracting those regulons that were specifically active in each of the cellular compartments (Supplementary file 6). We identified active GRNs that were guided by TFs characteristic of the different populations (Figure 4E and Figure 4—figure supplement 1D), and observed that a great number of the most prevalent regulons for each cell type were different from that of healthy donors. Interestingly, MDS cases showed regulons that were very active in most cell types, and that were not present in HSPCs from young or elderly samples, such as SMAD1 (MDS1); ATF3 and HOXA6 (MDS2); POU2F2, NR4A1, and HMG20B (MDS3); YEATS4, E2F1, and RUNX1 (MDS4). Interestingly, despite representing the same subtype, the individual MDS cases showed alterations of very specific regulons, demonstrating that the heterogeneity of the disease also takes place at the GRN level. Furthermore, some of the TFs guiding the regulons showing aberrant activity had previously been involved in the regulation of hematopoietic differentiation or the development of myeloid malignancies. For example, SMAD1 regulon was active in most cell types in patient MDS1, showing its highest activity in HSCs, LMPPs, and erythroid precursors. SMAD1 knockdown has been shown to promote erythropoiesis, suggesting that high activity of this factor may negatively impact erythroid differentiation (McReynolds et al., 2007). Furthermore, SMAD1 pathway has been shown to be active in a model of persisting LSCs, suggesting that this factor may be relevant in the development of myeloid malignancies (Lefort and Maguer-Satta, 2020). Patient MDS2 demonstrated aberrant high activity of the regulon guided by ATF3, which has been shown to drive cell cycle progression in AML, and of that of HOXA6, a TF which potentiates hematopoietic cell differentiation and self-renewal. Patient MDS3 presented a ubiquitous activity of the regulon guided by POU2F2/OCT2, a TF overexpressed in AML, in all the subpopulations analyzed. This patient also showed high activity of NR4A1, a factor that has been shown to specify a distinct subpopulation of quiescent myeloid-biased HSCs (Land et al., 2015), in HSCs and LMPPs. Moreover, we also detected that HMG20B, a known repressor of erythropoiesis (Esteghamat et al., 2011), was prominently active in several cell types, including erythroid precursors. Finally, patient MDS4 demonstrated aberrant high activity of the regulons guided by YEATS4, which is amplified in different tumors, and E2F1, a TF whose increased activity has been previously described in MDS. These results suggested that aberrant activity of specific regulons in MDS patients could drive aberrant gene expression and ultimately promote a myelodysplastic phenotype.

Collectively, the combination of computational methods and scRNA-seq demonstrate the power of these analyses to identify novel transcriptional alterations in a personalized manner, and therefore to help uncover the molecular heterogeneity of the disease. Moreover, the approaches used do not only have the power to characterize genes with altered expression across MDS hematopoietic differentiation or any other pathological condition, but also to identify deregulated GRNs that could act as master regulators in the disease, which could be exploited therapeutically in individual patients.

Discussion

In this study we report how the combination of computational tools can be used to generate high-resolution scRNA maps of human HSPCs and unravel changes associated with aging and disease. Although previous studies focused on other species (Mann et al., 2018; Flohr Svendsen et al., 2021) and other layers of information such as mutations (Jaiswal and Ebert, 2019), proteomics (Hennrich et al., 2018), and proteo-genomics (Triana et al., 2021), our study represents one of the first analyses that describes early human hematopoiesis based on its dynamic gene expression and transcriptional regulation by identifying changes that may be responsible for some of the phenotypic modifications observed during healthy aging. Applying this knowledge to HSPCs obtained from patients with MDS, we show how these analyses can help identify transcriptional alterations that play potential roles during disease development. We used MDS as a model to identify pathological alterations in the hematopoietic system, but we consider that the proposed methodology can be translated into other biological systems or pathological contexts.

Single-cell experiments are providing data at an unprecedented scale, which has allowed the identification of novel cell types and alterations that occur in biological systems. This increase in resolution has proceeded concurrently with the development of multiple computational tools that aspire to solve common problems arising with these technologies. An essential dilemma of this approach is how to establish reliable identities of the cells that will be used for subsequent analysis (Abdelaal et al., 2019). In this study, we generated a reference system based on the transcriptional profile of HSPCs from healthy young donors and used it to assign labels to cells from elderly healthy and pathological donors. We generated an in-house cell classifier that allowed us to reduce the effect of technical artifacts that can contribute to erroneous prediction of cell identity. This classifier was applied to cells that underwent the same sorting, library preparation, and sequencing procedures; thus, we expected to identify similar cell populations in each dataset. Additionally, when applied to external data, we found minor differences from the originally established labels, with the three lineages as well as the most immature states being precisely identified.

The results from our computational analysis go in line with previous studies, while also pointing toward genes and pathways of potential interest for further studies of hematopoiesis alterations. On one hand, we encountered an age-driven expansion of HSCs, as well as a decrease in the proportion of lymphoid precursors. Both events have been previously associated with aging (Dykstra et al., 2011; Dykstra et al., 2011; Pang et al., 2011; Pang et al., 2011; Young et al., 2016). Using functional analyses, we also detected an enrichment of pathways related to apoptosis and inflammatory conditions in early progenitors among elderly donors, which can be explained by the inflammatory microenvironment known to be present in elderly individuals (Leimkühler and Schneider, 2019). Conversely, cells from young donors showed highly active differentiation and proliferation profiles, reinforcing the idea of the higher differentiation capacity among young individuals (Chung and Park, 2017). On the other hand, trajectory analyses pointed toward abnormal transcriptional dynamics across monocytic development during aging. We observed that although the most immature clusters (HSCs and LMPPs) showed an increased monocytic potential of differentiation in the elderly, more mature precursor cells presented reduced potential. This could suggest an increased priming toward the monocytic lineage of early progenitors during aging. However, alterations at later stages could result in a loss of monocytic differentiation capacity and, accordingly, a delay in the expression of genes that characterize the monocytic lineage. In agreement with our results, an aging-associated skewing toward myeloid-biased HSCs and multipotent progenitor compartments has been described in mice (Elias et al., 2017). By focusing on GRNs, we observed that aged progenitor cell types showed a decrease in the activity of specific regulons, a lower degree of interaction between TFs and targets, and no enrichment of pathways involved in the cellular differentiation of specific lineages. Therefore, we were able to point toward regulatory factors guiding differentiation defects in the elderly. As computational methods used to predict regulatory mechanisms in single-cell RNA-seq datasets can provide false-positive and -negative predictions (Aibar et al., 2017; Pratapa et al., 2020), other types of assays, such as ATAC-seq, or functional assays could be used as a validation strategy.

We also shed light into the molecular pathogenesis of MDS. These syndromes are characterized by a significant phenotypic and genomic heterogeneity. In that context, our methodology could have a direct clinical application by promoting the identification of patient-specific transcriptional lesions with a potential involvement in the differentiation defects. Although previous studies focused on the transcriptional lesions of MDS exist (Hofmann et al., 2002; Pellagatti et al., 2006; Ueda et al., 2003; Miyazato et al., 2001; Im et al., 2018; Montalban-Bravo et al., 2020; Pellagatti et al., 2010), they were performed using bulk populations of mononuclear cells and, thus, provide a limited perspective of the pathology in these patients. The detailed analysis of erythroid differentiation in our patients with MDS led to the identification of genes with altered expression dynamics, which may play a key role in promoting dyserythropoiesis. Furthermore, GRN analysis revealed key regulons, whose activity could contribute to the phenotype of these cells. Further investigations will be needed to validate the roles of these regulators in the disease.

Overall, we propose the results of this work as an in silico basis for future, in-depth studies of early hematopoietic alterations. We performed a complex analysis, taking multiple approaches, that pointed toward genes and pathways potentially involved in aging and MDS. In addition, the methodology that we developed could be applied to fully analyze datasets set in other pathological scenarios.

Materials and methods

Sample collection

Request a detailed protocol

The samples and data from the patients included in the study were provided by the Biobank of the University of Navarra and were processed according to standard operating procedures. Patients and healthy donors provided informed consent, and the study was approved by the Clinical Research Ethics Committee of the Clinica Universidad de Navarra. Bone marrow aspirates were obtained from healthy controls (young individuals [n=5], median age, 20 years, range, 19–23 years) or patients undergoing orthopedic surgery (elderly donors [n=3], median age, 72 years, range, 61–84 years). Samples from newly diagnosed patients with MDS were obtained from the Clinica Universidad de Navarra and collaborating hospitals. The patient’s clinical characteristics are shown in Supplementary file 5.

Fluorescence-activated cell sorting

Request a detailed protocol

For the CD34+ cells purification, bone marrow mononuclear cells were isolated by Ficoll-Paque Plus (GE HealthCare) density gradient centrifugation and stained using CD34 (clone 8G12; BD Biosciences) CD64 (clone 10.1; BioLegend) CD19 (clone SJ25C1; BioLegend) CD10 (clone HI10A; BioLegend) CD3 (clone OKT3; BioLegend) CD36 (clone CLB-IVC7; Sanquin Plesmanlaan) CD61 (clone RUU-PL7F12; BD Biosciences) for 15 min at RT. CD34+ CD64- CD19- CD10- CD3- CD36+ CD61+ cells were then sorted in a BD FACSAria II (BD Biosciences). Purified CD34+ cells were directly used for scRNA-seq analysis.

scRNA library preparation

Request a detailed protocol

The transcriptome of the bone marrow CD34+ cells were examined using NEXTGEM Single Cell 3' Reagent Kits v3.1 (10× Genomics) according to the manufacturer’s instructions. Between 5000 and 17,000 cells, depending on the donor, were loaded at a concentration of 700–1200 cells/µL onto a Chromium Controller instrument (10× Genomics) to generate single-cell gel bead-in-emulsions (GEMs). In this step, each cell was encapsulated with primers containing a fixed Illumina Read 1 sequence, a cell identifying 16 bp 10× barcode, a 10 bp unique molecular identifier (UMI), and a poly-dT sequence. Upon cell lysis, reverse transcription yielded full-length, barcoded cDNA, which was then released from the GEMs, amplified using polymerase chain reaction, and purified using magnetic beads (SPRIselect, Beckman Coulter). Enzymatic fragmentation and size selection were used to optimize the cDNA size prior to library construction. Fragmented cDNA was then end-repaired, A-tailed, and ligated to Illumina adaptors. A final polymerase chain reaction amplification using barcoded primers was performed for sample indexing. Library quality control and quantification were performed using a Qubit 3.0 Fluorometer (Life Technologies) and an Agilent 4200 TapeStation System (Agilent), respectively. Sequencing was performed on a NextSeq500 instrument (Illumina) (Read1: 28 cycles; Read 55 cycles; i7 index: 8 cycles) at an average depth of 30,000 reads/cell.

scRNA preprocessing

Request a detailed protocol

The demultiplexing of raw base call files (BCL) was performed using the 10× software cellranger mkfastq. The generated FASTQ files were aligned to the GRCh38 version of the human genome, and count matrices were constructed using cellranger count. The default barcode filtering was performed at this step. To remove doublets, we plotted the distribution of total genes and UMIs detected per cell and established a customized superior threshold for each sample (Supplementary file 1). Additionally, we filtered out cells with >5% or 10% of counts landing in mitochondrial genes, as this is an indicator of dying cells. In the elderly and pathological samples, we detected a small number of cells in which <1% of the mitochondrial genes clustered together; thus, we excluded them from the analysis.

Data integration, clustering, and visualization

Request a detailed protocol

Samples from the same condition (young, elderly, or MDS) were integrated using the Seurat pipeline. Gene counts were divided by the total expression per cell, multiplied by a scaling factor of 10,000, and log transformed. Normalized counts were scaled across cells. The 2000 genes with highest variance were selected using the variance stabilizing transformation method. Next, we performed integration as described previously (Stuart et al., 2019) using 50 dimensions. After integration, we rescaled the data, regressed the cell cycle effect, and conducted principal component analysis (PCA). Based on the visual exploration of the scree plot, we selected the appropriate number of components to continue the analysis (Supplementary file 1).

Next, we performed unsupervised clustering on the young integrated data, using the algorithm implemented in Seurat. We constructed a shared nearest neighbor graph based on Euclidean distances in the PCA space with the chosen dimensionality and clustered the cells using the default Louvain algorithm. We tested several resolutions and assessed the results by calculating the average silhouette for each cluster. We determined the cluster markers using the Seurat function FindAllMarkers with the MAST method. Next, we annotated the clusters by manually inspecting the most specific markers and searching for curated markers in the literature.

To ease visualization, we embedded the cells from the young and elderly samples under the same coordinates. Thus, we integrated the young and elderly datasets using the pipeline described above. Using the first 30 principal components, we calculated the UMAP coordinates, as implemented in Seurat, which were then used in all the figures.

Classification model

Request a detailed protocol

We constructed a classification model to predict the cell identity of elderly and MDS single cells using the previously annotated young dataset as a reference. To this end, we performed logistic regression with elastic-net regularization in the glmnet R package. This approach is appropriate for sparse input data, and the elastic-net penalty allows flexible selection of variables.

We constructed an individual binary prediction model for each of the 14 identities established. To train and validate each model, we selected all the cells from the appropriate cluster and an equal number of random cells from the remaining clusters. This set was then randomly divided in two: 75% was destined to training the model and 25% was dedicated to validation. We also performed an initial feature selection by selecting the subset of highly variable genes present in both the training-validation data (young) and the data to be classified (elderly, MDS, or AML). Initially, we trained models with different α values (1, 0.75, 0.5, 0.25, and 0.1). For each model, we used 10-fold cross-validation, and the algorithm selected the optimal value for λ. Subsequently, we selected the most appropriate α value by evaluating the performance of the models on the validation dataset. If possible, we only retained the models that used between 20 and 150 variables. Among them, we selected the model with the maximum AUC and, in case of a tie, we successively searched for the minimum false-positive rate, minimum false-negative rate, and maximum number of variables. After the optimal model was selected, we used it to classify new data as positive or negative for a particular identity.

We repeated the steps from the creation of the training and validation sets to binary prediction, 10 times and stored the resulting predictions: both the class (positive or negative) and its associated probability of being positive. Finally, we averaged the 10 resulting probabilities for each binary model and assigned the identity that corresponded to the highest probability. In cases in which none of the identities were associated with a probability of >0.5, the cell was labeled as not assigned.

GSEA

Request a detailed protocol

We performed differential expression analysis of cells belonging to the same identity to identify differences between conditions. We applied the Seurat FindMarkers function to the log-normalized counts using the MAST method and the options logfc.threshold=0 and min.pct=0. In order to control for batch effect, we set patients as latent variables. In addition, we tested for all genes expressed in every sample. Only cell types with at least 25 cells per condition were used. We ranked the genes according to their average log fold change. We performed GSEA using the fgsea R package and tested for hallmark gene sets.

Trajectory analysis using STREAM

Request a detailed protocol

We extracted the differentiation branches in our data using STREAM. We then used the young batch-corrected matrix to establish the reference structure, onto which we then mapped the elderly cells. Using the STREAM pipeline, we computed the 2000 most variable genes and 20 principal components. Then, we performed nonlinear dimensionality reduction using modified locally linear embedding. We projected cells onto a three-dimensional space and set the algorithm to use the 30 nearest neighbors. An elastic principal graph was constructed by computing 20 clusters and was refined using the following parameters: epg_alpha = 0.02, epg_mu = 0.05, and epg_lambda = 0.01. Subsequently, mapping was performed using the default STREAM parameters. For this step, we used the elderly batch-corrected matrix.

Trajectory analysis using Palantir

Request a detailed protocol

Palantir was used to reconstruct hematopoietic lineages and recover gene expression dynamics along the lineages. This algorithm models differentiation and fate choice as a continuous probabilistic process. It orders cells along a global pseudotime, determines final states, and assigns each cell a probability to reach each terminal state.

We used the batch-corrected matrices to infer trajectories in young and elderly conditions. PCA was performed, and 30 diffusion components were calculated. For the young dataset, we selected 10 HSCs according to their position in the UMAP coordinates. We sequentially applied the Palantir algorithm using each of these cells as the initial state. Terminal states were stored for each run, and we retained those that appeared in more than five cases. If more than one terminal state corresponded to the same cell identity, we randomly selected one of them. We ran Palantir a final time adding the selected terminal states as prior information. For the elderly dataset, we selected the HSC with the highest probability, as determined by the GLMnet model, as the initial state. We then searched for cells with the highest similarity to the terminal states found in the young dataset. To this end, we calculated the nearest neighbor in the UMAP space. In the case of the MDS samples, we independently analyzed each patient. We used the normalized counts matrices, calculated 30 diffusion components, and set 10 eigen vectors to determine the multiscale space.

For each dataset, gene expression trends along pseudotime were calculated as described in Palantir using generalized additive models, with the addition of the branch probabilities as prior weights for the model.

GRNs

Request a detailed protocol

Python implementation of SCENIC was used for GRN inference. All required inputs were downloaded from https://resources.aertslab.org/cistarget/ and https://pyscenic.readthedocs.io/en/latest/installation.html. Raw UMI matrices were provided as input for the pyscenic algorithm, as described in https://pyscenic.readthedocs.io/en/latest/tutorial.html. RSS were obtained using the regulon_specificity_scores command.

AUC matrices were imported into R and batch-corrected using the removeBatchEffect implemented in the limma R package. Batch-corrected AUC matrices were binarized using the BinarizeAUC function implemented in the AUCell R package. The percentage of active regulons per cell type was determined by counting the number of cells with a value of 1 in the binarized AUC matrix and dividing it by the number of cells per cell type. The top five regulons per cell type were selected according to the RSS.

Networks were drawn in Cytoscape, and only edges with an importance value larger than the third quartile (per regulon) were retained.

We performed over-representation analysis for the different sets of genes that compose the top five specific regulons for HSC, LMPP, GMP, CLP, and MEP, both in data from young and elderly donors. We tested for the gene ontology biological processes gene sets.

Basic statistical analysis

Request a detailed protocol

We carried out pairwise comparison between pairs of proportions to test for significant differences in cell population proportions. We did so both among individuals from the same condition and between conditions. Two-sample Wilcoxon test was performed to test for differences in Palantir pseudotime, differentiation potential, and branch probabilities between young and elderly and per cell population. It was also used to find significant differences in gene expression trends across pseudotime between conditions. Multiple testing was addressed by adjusting p-values using the Bonferroni-Holm correction method. Results were considered significant when adjusted p-values <0.05.

Data availability

All the single cell RNA sequencing data is available at Gene Expression Omnibus under accession number GSE180298. The scripts needed to replicate the analysis are deposited on GitHub: https://github.com/mainciburu/scRNA-Hematopoiesis (mainciburu, 2023, copy archived at swh:1:rev:3e64802fb6497d396d74d9da02e9309432c8f82b).

The following data sets were generated
The following previously published data sets were used
    1. Granja JM
    (2019) NCBI Gene Expression Omnibus
    ID GSE139369. Single-cell, multi-omic analysis identifies regulatory programs in mixed phenotype acute leukemia.
    https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE139369

References

    1. Leimkühler NB
    2. Schneider RK
    (2019) Inflammatory bone marrow microenvironment
    Hematology. American Society of Hematology. Education Program 2019:294–302.
    https://doi.org/10.1182/hematology.2019000045

Decision letter

  1. Jiwon Shim
    Reviewing Editor; Hanyang University, Republic of Korea
  2. Utpal Banerjee
    Senior Editor; University of California, Los Angeles, United States
  3. Jong Kyoung Kim
    Reviewer; Pohang University of Science and Technology, Republic of Korea

Our editorial process produces two outputs: (i) public reviews designed to be posted alongside the preprint for the benefit of readers; (ii) feedback on the manuscript for the authors, including requests for revisions, shown below. We also include an acceptance summary that explains what the editors found interesting or important about the work.

Decision letter after peer review:

Thank you for submitting your article "Uncovering perturbations in human hematopoiesis associated with healthy aging and myeloid malignancies at single cell resolution" for consideration by eLife. Your article has been reviewed by 3 peer reviewers, one of whom is a member of our Board of Reviewing Editors, and the evaluation has been overseen by Utpal Banerjee as the Senior Editor. The following individual involved in the review of your submission has agreed to reveal their identity: Jong Kyoung Kim (Reviewer #2).

The reviewers have discussed their reviews with one another, and the Reviewing Editor has drafted this to help you prepare a revised submission.

Essential revisions:

1. Validate GLMnet and apply a consistent analysis platform throughout (e.g. STREAM or Palantir for pseudotime, manual annotation): All reviewers

2. Substantiate and clarify MDS analysis: (1) Same analytical strategy should be applied for young/old and MDS patients, (2) a detailed comparison between MDS-aging, (3) add clinical details: All reviewers

3. Additional supports required for the HSC population analysis, clarify young/elderly data analysis from the clonal hematopoiesis perspective: Reviewers 1 and 2

4. Consistent data description and additional validation for age-dependent cell type changes: Reviewers 1 and 3

5. Additional data/quantitation/explanation required for GRN analysis between young and old: Reviewers 2 and 3

Reviewer #1 (Recommendations for the authors):

1. In Figure 1b and d, elderly2 individual expresses a significantly higher HSC proportion than the other two with reductions in other populations, which can skew an entire landscape of elderly populations. Is there any other data supporting that the proportions of HSCs from elderly donors shown in Figures 1b,d are representative?

2. Regarding the above concern, what is the age distribution of elderly individuals? In a very recent study showing the clonal diversity of HSC/MPP cells in humans (for example, Mitchell et al., Nature 2022), it has been shown that the elderly over 65 yrs dramatically reduce the clonal diversity and this might be contributed to the biased HSC proportions shown in elderly2 data. Brief information, at least an age, of individuals needs to be provided as in sup table 5 for healthy donors, if possible, and discuss the extreme bias generated in elderly2 data.

3. In Figure 1e, the authors concluded that Myc is downregulated, and proliferative activity is decreased in cells of elderly individuals. However, some of the populations are rather expanded in elderly individuals; for example, the numbers of HSC or MEP are rather higher in elderly individuals. How would the authors explain such discrepancies?

4. As a nonexpert, it is not clear to me why Seurat or GLMnet-based labeling results in dissimilar proportions of cell populations. Would differing cutoffs or measurements of Seurat have given similar numbers to GLMnet? Or would it be possible to acquire numbers similar to Seurat or GLMnet with an alternative method? It would become a much-valued resource for the community if the data presented here, from young, and elderly to MDS, are analyzed in a consistent platform with a clear rationale.

5. In Figure 3e, the authors claim that HSCs from young donors show enriched terms related to differentiation of hematopoietic lineages while elderly donors do not display such an increase. However, it is not clear whether changes in the gene expression of HSCs from young donors are attributed to uniform alterations in HSC gene expressions or due to changes in the composition of HSC subsets.

1) The gene regulatory network of HSCs in elderly donors might have undergone a global change, but it is also possible that a landscape of HSC subsets (for example, long-term, short-term, or different subsets segregated by different niche interactions) could change, consequently leading to altered gene expressions.

2) Is it possible to subcategorize HSCs, for example, cells with high differentiation genes versus low, and compare them between young versus elderly?

3) Are HSCs in both young and elderly clear enough? It is possible that intermediate/committed cells, simultaneously holding stem cell characteristics, are mixed in HSC populations.

6. Even though this paper is a resource article, explanations of the MDS data are not clear enough and additional analysis may be required to better understand the disease.

1) Are the same cell types annotated when the same method used in young/elderly is applied to MDS patients?

2) It is reasonable to conclude that MDS cases show high heterogeneity of the GRN levels and each patient has specific regulons for the disease development. If so, do the four MDS patients show differential trajectories of HSC differentiation? And how are these single-cell landscapes from 4 patients associated with genetic mutations in each case (shown in sup table5)?

3) If MDS is a heterogenic disease, what would be a common idea, which can be used for future studies and therapeutics, extracted from single-cell RNA analyses?

Reviewer #2 (Recommendations for the authors):

I do not recommend this manuscript for publication form based on the following reasons:

1. The main claims derived from computational predictions were not well supported by the data presented and were not experimentally validated (Major points 1 to 6 of the Public Review).

2. I think if the authors address Major point 7 of the Public Review, the value of the single-cell dataset as a resource for understanding the age-associated cellular and molecular alterations during human hematopoiesis will be greatly improved. However, I agree that this would be beyond the scope of this manuscript.

If the manuscript is revised to address these concerns, I can reconsider my recommendation.

Reviewer #3 (Recommendations for the authors):

1. Section 'GRNs guiding young and elderly hematopoiesis' focuses on differences in transcription factor regulatory networks between young and old. Whilst an elegant analysis, it does not appear to add much more insight than the previous sections which identify expanded HSC and impaired differentiation in the elderly datasets. We would consider reducing this section and merging it with the previous one.

[Editors' note: further revisions were suggested prior to acceptance, as described below.]

Thank you for resubmitting your work entitled "Uncovering perturbations in human hematopoiesis associated with healthy aging and myeloid malignancies at single cell resolution" for further consideration by eLife. Your revised article has been evaluated by Utpal Banerjee (Senior Editor) and a Reviewing Editor.

The manuscript has been improved but there are some remaining issues that need to be addressed, as outlined below:

Although the authors addressed most of the concerns, some of the major comments require additional changes to warrant publication in eLife. Please find the concerns raised by reviewer 2, especially ones regarding the previous major comments 2 and 3.

Reviewer #1 (Recommendations for the authors):

The authors performed additional analyses and experiments and adequately addressed all of my major concerns.

Reviewer #2 (Recommendations for the authors):

The authors should specify or highlight changes in their manuscript and rebuttal. It is difficult for me to follow changes in this revised manuscript. The revised manuscript addressed some of my previous concerns but failed to address the following points:

1. Previous major comment 2 (cell-type composition changes): The newly added flow cytometry data did not support an expansion of MEPs and a reduction of GMPs in elderly individuals predicted by scRNA-seq analysis. The sentence at line 178-179 ("We used Flow Activated Cell Sorting (FACS) as an orthogonal method to support our findings (Figure 1—figure supplement 3) and observed similar results.") should be toned down accordingly.

2. Previous major comment 3 (STREAM and Palantir): I strongly disagree with the authors's opinion that mixing the results of two different methods in the same figure can be helpful for deciding which method is better suited to specific problems. Figure 2F and G can be equally well presented with pseudotime computed by STREAM as the authors showed that pseudotime values from two methods are highly correlated. To avoid any confusion and be consistent, the authors should not mix the results of two different methods in the same figure. The results generated by Palantir should be presented in a supplementary figure to demonstrate the robustness of pseudotime analysis.

3. Previous major comment 5: What does "independent network" mean?

4. Previous major comment 6: Even though this manuscript was submitted as a "Tools and Resources" article, the authors should demonstrate the robustness of their constructed GRNs. All research papers should convincingly show that the results and predictions presented in the manuscript are robust and consistent regardless of the category of the submitted manuscript. The benchmarking papers and other research papers have already shown that all methods for constructing GRNs from scRNA-seq data (including SCENIC) have an issue of false positive and negative predictions.

Reviewer #3 (Recommendations for the authors):

The authors comprehensively addressed the comments. I have no further concerns.

https://doi.org/10.7554/eLife.79363.sa1

Author response

Essential revisions:

1. Validate GLMnet and apply a consistent analysis platform throughout (e.g. STREAM or Palantir for pseudotime, manual annotation): All reviewers

2. Substantiate and clarify MDS analysis: (1) Same analytical strategy should be applied for young/old and MDS patients, (2) a detailed comparison between MDS-aging, (3) add clinical details: All reviewers

3. Additional supports required for the HSC population analysis, clarify young/elderly data analysis from the clonal hematopoiesis perspective: Reviewers 1 and 2

4. Consistent data description and additional validation for age-dependent cell type changes: Reviewers 1 and 3

5. Additional data/quantitation/explanation required for GRN analysis between young and old: Reviewers 2 and 3

In order to answer all the comments raised by the reviewers and the editor we have performed additional studies that can be summarized as follows:

– Used FACS data to confirm cell type proportion changes in young and elderly individuals.

– Perform and complete new analysis on young and elderly HSCs to confirm that this cell type can be classified using the same cell type markers, however the elderly cells display a more quiescent and less active cell type.

– Compared manual vs automatic cell type classification in MDS samples, to highlight how GLMnet accurately classifies cell types within different patients.

– We have updated the manuscript to clarifly the pipeline and workflow used in each of the analysis steps. We also improved data description to simplify interpretation of comparisons

Reviewer #1 (Recommendations for the authors):

1. In Figure 1b and d, elderly2 individual expresses a significantly higher HSC proportion than the other two with reductions in other populations, which can skew an entire landscape of elderly populations. Is there any other data supporting that the proportions of HSCs from elderly donors shown in Figures 1b,d are representative?

We thank the reviewer for the thoughtful question. In order to support the findings from the scRNAseq analysis, we have included Flow Activated Cell Sorting (FACS) plots in Figure 1—figure supplement 3. These plots show an increase in the number of highly enriched HSCs populations (sorted as CD34+ CD38- CD45RA- CD90+ cells) in elderly individuals compared to young donors. This has also been described in the literature and we have cited relevant papers in the introduction (1,2).

2. Regarding the above concern, what is the age distribution of elderly individuals? In a very recent study showing the clonal diversity of HSC/MPP cells in humans (for example, Mitchell et al., Nature 2022), it has been shown that the elderly over 65 yrs dramatically reduce the clonal diversity and this might be contributed to the biased HSC proportions shown in elderly2 data. Brief information, at least an age, of individuals needs to be provided as in sup table 5 for healthy donors, if possible, and discuss the extreme bias generated in elderly2 data.

In order to address this comment, we have updated Supplementary File 5 to include the age of the healthy donors. Regarding the bias in the number of HSCs in one of the elderly individuals, we suspect that this just correlates with a particular case, such as the one depicted in the Figure 1—figure supplement 3. As the collection of elderly samples was difficult and dependent on availability (samples were obtained from elderly individuals undergoing hip surgery), we decided to keep all the samples to maintain the number of individuals (3).

3. In Figure 1e, the authors concluded that Myc is downregulated, and proliferative activity is decreased in cells of elderly individuals. However, some of the populations are rather expanded in elderly individuals; for example, the numbers of HSC or MEP are rather higher in elderly individuals. How would the authors explain such discrepancies?

We thank the reviewer for this comment. In Figure 1—figure supplement 4 from the manuscript, we show significant downregulation of MYC in several subpopulations of elderly donors, including HSCs, LMPPs, MEPs and early erythroid progenitors. This goes hand in hand with our GSEA analysis (Manuscript Figure 1e), where we find enrichment of MYC target gene sets (V1 and V2) for young donors. To further confirm these results, we analyzed the MYC regulon activity, and we calculated scores to summarize the expression of MYC targets, using AUCell (Author response image 1). Once again, we observe higher activity of MYC in young donors.

The transcriptional program regulated by MYC plays a complex role in the balance between cell differentiation and self renewal. Its activity has unique consequences on the different hematopoietic lineages, as well as on HSC. In this last case, its overexpression promotes proliferation and differentiation (3). Thus, dormant HSC are characterized by low levels of MYC, which increase in active HSCs and later progenitors (4). In agreement with this, we have found an accumulation of quiescent HSCs in elderly donors (answer to question #5, Figure 5), which has also been observed in previous studies (5,6). This could explain the higher proliferation and MYC activity in HSC from young donors. Something similar could occur in other early progenitor subpopulations, such as MEP, promoting the accumulation of multipotent cells at that stage.

Overall, our hypothesis regarding the increased numbers of early progenitors (HSC, MEP) in elderly individuals arises from the lack of differentiation capabilities in elderly individuals. As cells are unable to undergo the expected transition (from progenitor state to a differentiated cell), the number of these progenitor cells increases. Compared with younger individuals, we see a higher proportion of HSC, but lower values for differentiated cells in elderly donors.

Author response image 1
Download asset Open asset
MYC activity in HSPC from young and elderly donors.

Violin plots showing the activity of MYC regulon obtained with SCENIC (top), and the expression of MYC target gene sets V1 (middle) and V2 (botton), summarized in a score calculated with AUCell.

4. As a nonexpert, it is not clear to me why Seurat or GLMnet-based labeling results in dissimilar proportions of cell populations. Would differing cutoffs or measurements of Seurat have given similar numbers to GLMnet? Or would it be possible to acquire numbers similar to Seurat or GLMnet with an alternative method? It would become a much-valued resource for the community if the data presented here, from young, and elderly to MDS, are analyzed in a consistent platform with a clear rationale.

The main reason the proportions differ is related to how each method works. The anchor-based approach used in Seurat identifies “anchors” or closely related cells between a given reference and a query. This is a generalized method that can be applied to transfer labels between any given reference and a dataset to be annotated.

Instead of developing a general method to classify any single cell experiment, we decided to focus on creating specific classification models for the cell types included in our integrated young donor reference. Our method is an embedded system that combines feature selection and building the classification model. The elastic-net penalization allows to select only a subset of features to avoid overfitting.

As all the samples that are analyzed in the manuscript (elderly donors and MDS patients) should include a similar subset of cell types, we decided to apply our GLMnet classifier.

The comparison between methods is described in the Figure 1—figure supplement 1 from the manuscript. We decided to use an external dataset (Granja, et. al) and assumed those cell types to be the ground truth. In the Figure 1—figure supplement 1 B and C, it can be seen that Seurat classifies a subset of HSCs as MEPs while GLMnet returns a more similar output (based on proportions) to the defined ground truth.

To further assess the raised issue, we decided to focus on HSCs that were classified as MEPs by Seurat and HSCs by GLMnet. Author response image 2 shows that the second highest score in such cells corresponds to HSCs.

Author response image 2
Download asset Open asset
Seurat classification scores.

Box-plot describing the distribution of seurat scores of cells classified as MEPs by Seurat and HSCs by GLMnet.

As it has already been described, one of the biggest challenges of single cell analyses is to transform a continuum system (such as cellular differentiation) to a discrete one. We believe that the results obtained with GLMnet, are closer to the ground truth for this specific project.

5. In Figure 3e, the authors claim that HSCs from young donors show enriched terms related to differentiation of hematopoietic lineages while elderly donors do not display such an increase. However, it is not clear whether changes in the gene expression of HSCs from young donors are attributed to uniform alterations in HSC gene expressions or due to changes in the composition of HSC subsets.

1) The gene regulatory network of HSCs in elderly donors might have undergone a global change, but it is also possible that a landscape of HSC subsets (for example, long-term, short-term, or different subsets segregated by different niche interactions) could change, consequently leading to altered gene expressions.

2) Is it possible to subcategorize HSCs, for example, cells with high differentiation genes versus low, and compare them between young versus elderly?

3) Are HSCs in both young and elderly clear enough? It is possible that intermediate/committed cells, simultaneously holding stem cell characteristics, are mixed in HSC populations.

We thank the reviewer for these questions. We have addressed all the comments related to HSCs in a similar approach, the analysis is described below:

We performed a new analysis focused on the HSC compartment, to study its heterogeneity and whether changes that take place with aging are observed throughout the whole compartment or can be associated with specific cell subsets.

First, we confirmed the suitability of the HSC classification that we established. To give additional evidence, we plotted the expression of canonical marker genes, exclusively in the HSC compartment, separated by individual donors. We found that every donor expressed the HSC markers, whereas the majority of them did not express markers of committed cell populations from any of the hematopoietic lineages (Author response image 3).

Author response image 3
Download asset Open asset
Expression of marker genes in the HSC compartment.

Dot plot depicting the normalized scaled expression of canonical marker genes by HSC of the 5 young and 3 elderly healthy donors. Marker genes are colored by the cell population they characterize. Dot color represents expression levels, and dot size represents the percentage of cells that express a gene.

Next, we chose an unsupervised approach to sub-cluster HSCs, allowing the main sources of variability among the cells to drive the process. We created two new datasets containing only HSCs, one for young donors and another one for elderly donors. We independently reintegrated each of them with Seurat, to remove donor-associated batch effects, and performed unsupervised clustering.

For the HSC coming from young donors, we chose a clustering resolution of 0.3, resulting in 4 clusters (Author response image 4a) homogeneously distributed across patients (Author response image 4b). Differential expression analysis returned few significant cluster specific markers (from 2 to 50), pointing towards the existence of high similarity between clusters. In the case of the HSC coming from elderly donors, we chose a resolution of 0.2, resulting in 6 clusters, three of them significantly smaller than the rest (Author response image 4a). Every cluster was present in the 3 donors, though we found more heterogeneity in the distribution among patients than in the young HSC (Author response image 4b). Differential expression analysis yielded >100 markers for clusters 3, 4 and 5 and <35 for clusters 0, 1 and 2.

Then, we explored whether any correspondence existed between young and elderly clusters, aiming to know if HSCs from both young and elderly cluster together due to similar biological reasons. We compared common marker genes for both sets of clusters and found the greatest overlap between the following pairs: cluster 0 from young and cluster 1 from elderly, cluster 2 from young and cluster 0 from elderly, cluster 3 from young and cluster 3 from elderly. We found that CDK6, an essential gene for HSC exit from quiescence, was a common marker for cluster 0 in young and cluster 1 in elderly (Author response image 4c top left), suggesting differences in HSC activation states among clusters. To further explore this, we scored cells for curated gene signatures characterizing quiescence (7) and proliferation (8). We observed that quiescence scores are higher in cluster 2 from young donors and cluster 0 from elderly donors compared to the rest (Author response image 4c middle left). In line with this, proliferation is especially high on a small subset of HSC in clusters 0 and 1 from young donors, a subset of HSC in cluster 1 from elderly and cluster 4 from elderly (Author response image 4c bottom left). When we assessed the cell cycle phase HSCs are in, following the Seurat pipeline, we found the great majority of them to be in G1, as expected. Cells with high proliferation scores correlate with S phase and, finally, cluster 4 from elderly donors was found to be specific to cells in G2/M phase (data not shown). Therefore, quiescence and proliferation appear to be the greatest effects driving HSC clustering.

In order to study differences in lineage priming, we also scored cells for gene signatures associated with HSC, GMP and MEP (9). We observed that HSC scores positively correlate with quiescence scores (Author response image 4c top right). On the contrary, GMP signature appears to be more active in cells with less quiescent state, and cluster 5 from elderly scored especially high for myeloid progenitors signature (Author response image 4c middle right). MEP genes did not show any cluster-related pattern of activation (Author response image 4c bottom right).

To sum up, we have found that the main driver for HSCs variability is their state of quiescence vs proliferation. Elderly donors present higher numbers of dormant HSCs, which could indeed be the cause for the lower expression of proliferation and metabolism related genes in this compartment. On the other hand, differences in lineage priming among HSCs, and age-associated changes in priming are not clear on this data.

Author response image 4
Download asset Open asset
HSC sub clustering.

(A) UMAP visualization of HSC from young (left) and elderly (right) donors subjected to re-integration and unsupervised clustering.

Cells are colored by clusters. (B) Bar plot showing the proportion of cells from each donor assigned to the different clusters. (C) UMAP plots for young (left) and elderly (right) HSC colored by the normalized expression of CDK6 (top left) and by the summarized expression of multiple gene signatures, quantified as scores calculated using the software AUCell.

We then focused on the HSC-specific regulons that resulted from our GRN analysis (Figure 3 from the manuscript). We aimed to explore whether differences between young and elderly donors in transcriptional programs activity are homogeneously distributed among HSC (Author response image 5). We assessed activity scores (AUC) across the HSC sub-clusters we generated with this analysis. We did not observe great differences in activity across clusters. For some transcriptional programs, such as PRDM16 or GATA3, we see slightly increased activity in the clusters corresponding to the most quiescent HSCs (cluster 2 in young and 0 in elderly), but this trend stays the same for both young and elderly.

Overall, we observe that the identified HSC population in both elderly and young donors have a common set of markers, showing subpopulations that seem to be driven by quiescence and proliferation (with higher quiescence scores in elderly). Further inspection of GRNs in these HSC clusters revealed no differences in regulation programs between conditions, suggesting that the observed patterns across conditions are equivalent.

Author response image 5
Download asset Open asset
Activity of HSC-specific regulons.

Violin plots showing activity scores for the top 5 HSC-specific regulons generated by SCENIC in HSC from both young (left) and elderly (right) donors, separated by sub-clusters. Color indicates sub-cluster.

6. Even though this paper is a resource article, explanations of the MDS data are not clear enough and additional analysis may be required to better understand the disease.

1) Are the same cell types annotated when the same method used in young/elderly is applied to MDS patients?

2) It is reasonable to conclude that MDS cases show high heterogeneity of the GRN levels and each patient has specific regulons for the disease development. If so, do the four MDS patients show differential trajectories of HSC differentiation? And how are these single-cell landscapes from 4 patients associated with genetic mutations in each case (shown in sup table5)?

3) If MDS is a heterogenic disease, what would be a common idea, which can be used for future studies and therapeutics, extracted from single-cell RNA analyses?

We thank the reviewer for this set of questions concerning the analysis of MDS samples.

Regarding the first point, cells in the 4 MDS datasets were annotated with the same method used for elderly: cell identities found in the healthy young HSPC pool, together with their gene expression profile, were used as reference to predict cell identities in the MDS datasets. To do so, we used the GLMnet-based method we developed. Cells in the MDS datasets were assigned the identity with highest probability among the ones calculated by the GLMnet binary classification models. If a minimum probability is not met for any identity, then a cell is classified as “not assigned” (see methods section for more details). This strategy was chosen as we expected the MDS samples to be composed by the same cell populations as the ones we identified in the healthy young samples. All of them underwent the same sample processing, FACS sorting, library preparation and sequencing processes. In addition, clinical data indicated that blast presence in the bone marrow of these patients was very low: between 0 and 4% (supp table 5). Therefore, this classification method assumes that the great majority of the cells belong to the established set of identities. At the same time, it allows for changes in the proportions of cell populations, as well as not to assign a label to cells that do not resemble any of the possible cell identities.

During this revision, we have performed additional validations for the classification, including clustering and manual annotation of the MDS samples. With this method, we have not found new cell types that were not present in our reference samples (detailed answer to Reviewer 2, first comment).

We have updated the manuscript to reflect the comment raised by the reviewer by adding the following phrase (Page 7, line 15): “All the analysis for the MDS samples was performed following the same methods used for the elderly individuals. For cell type annotation, we predicted cell labels using our GLMnet classifier”. We have emphasized that the same classification strategy used for elderly samples was used to classify MDS cells.

The second point made by the reviewer is of high interest. Exploring the effects of mutations in the transcriptional profile of patients could give key insight into the disease and contribute to the development of targeted treatments. In this study, however, we consider we have not enough samples to raise conclusions around this issue. Three out of four patients have mutations in genes of the spliceosome (ZRSR2 in MDS1 and U2AF1 in MDS2 and MDS3). MDS2 and MDS3 also share a mutation in the histone regulator ASXL1. However, MDS4 only has a record of a TET2 mutation at a very low VAF (2.6%). The sample size is not big enough to associate any of the alterations we observe to specific mutations.

Instead, we chose this set of patients because they all share MDS with multilineage dysplasia, resulting in more than one myeloid lineage affected by the disease. All of them have a deficiency of erythrocytes and/or platelets. This is the reason why we focussed on the megakaryocyte-erythroid differentiation trajectory. We could consistently reconstruct this path in the 4 MDS datasets, and calculate the gene expression trends among which we looked for alterations.

As mentioned by the reviewer, MDS is a heterogeneous disease that can display specific alterations per patient. As a resource article, we intend to highlight the use of single-cell RNA sequencing as a viable tool to analyze pathological cases on a case by case basis rather than in a collective approach. Any pathological sample can be used to follow our proposed approach and identify case-specific alterations.

Also, we have identified a set of common alterations shared between our pathological donors. Among these alterations, we describe the widespread enrichment of interferon α and γ response genes in MDS samples. This could point towards deregulated interactions with the microenvironment, where higher expression of interferon genes have been found (10), or towards an altered immune system. We also observed genes with altered expression trends along erythroid differentiation in multiple MDS samples, such as TRIB2 and PHF6.

Reviewer #3 (Recommendations for the authors):

1. Section 'GRNs guiding young and elderly hematopoiesis' focuses on differences in transcription factor regulatory networks between young and old. Whilst an elegant analysis, it does not appear to add much more insight than the previous sections which identify expanded HSC and impaired differentiation in the elderly datasets. We would consider reducing this section and merging it with the previous one.

We thank the reviewer for the comment regarding the GRN section of the manuscript. We consider that having an independent section for GRN analysis allows readers to better understand the corresponding section. As we use specific algorithms for the prediction or regulatory features, which are different from the trajectory analysis one, this layout avoids confusion between what can be obtained with each tool.

[Editors' note: further revisions were suggested prior to acceptance, as described below.]

Reviewer #2 (Recommendations for the authors):

The authors should specify or highlight changes in their manuscript and rebuttal. It is difficult for me to follow changes in this revised manuscript. The revised manuscript addressed some of my previous concerns but failed to address the following points:

1. Previous major comment 2 (cell-type composition changes): The newly added flow cytometry data did not support an expansion of MEPs and a reduction of GMPs in elderly individuals predicted by scRNA-seq analysis. The sentence at line 178-179 ("We used Flow Activated Cell Sorting (FACS) as an orthogonal method to support our findings (Figure 1—figure supplement 3) and observed similar results.") should be toned down accordingly.

We thank the reviewer for the comments. We have updated the manuscript and update the corresponding sentence (Page 4. line 27-29) as follows:

“We used Flow Activated Cell Sorting (FACS) as an orthogonal method and observed similar results for HSCs. However changes in the proportion of GMPs and MEPs were less obvious than in the case of the transcriptomic analysis (Figure 1—figure supplement 3)”

2. Previous major comment 3 (STREAM and Palantir): I strongly disagree with the authors's opinion that mixing the results of two different methods in the same figure can be helpful for deciding which method is better suited to specific problems. Figure 2F and G can be equally well presented with pseudotime computed by STREAM as the authors showed that pseudotime values from two methods are highly correlated. To avoid any confusion and be consistent, the authors should not mix the results of two different methods in the same figure. The results generated by Palantir should be presented in a supplementary figure to demonstrate the robustness of pseudotime analysis.

We thank the reviewer for the comments. We have updated Figure 2 to include all results obtained with Palantir and Figure 2—figure supplement with STREAM results. Also, we included the following sentence to emphasize that the results with both methods are strongly correlated (Page 5. line 24-26):

“To measure the similarity between both methods, we performed a correlation analysis and noticed that the pseudotime values were strongly correlated (r = 0.78) (Figure 2—figure supplement 1e).”

3. Previous major comment 5: What does "independent network" mean?

We thank the reviewer for the comment. We refer to the HSC GRN in elderly individuals as independent due to the lack of connection with other identified regulons in different cellular subpopulations. In contrast, the young HSC GRN shows a connection with the rest of elements in the network.

We have updated the corresponding section (Page 6. line 40) to better reflect the findings:

“Overall, we observed that the predicted regulatory network of elderly HSCs (Figure 3d) appeared as an isolated network compared to the young GRN.”

4. Previous major comment 6: Even though this manuscript was submitted as a "Tools and Resources" article, the authors should demonstrate the robustness of their constructed GRNs. All research papers should convincingly show that the results and predictions presented in the manuscript are robust and consistent regardless of the category of the submitted manuscript. The benchmarking papers and other research papers have already shown that all methods for constructing GRNs from scRNA-seq data (including SCENIC) have an issue of false positive and negative predictions.

We thank the reviewer for the comments and certainly agree with him that predicted GRN may need functional validation to determine their true biological value. In that sense, we have included a phrase in the discussion (Page 9. line 42-44) to reflect the fact that computational methods used to predict regulatory mechanisms can show false positive/negative results and that other strategies could be used as a validation strategy.

“As computational methods used to predict regulatory mechanisms in single cell RNA-seq datasets can provide false positive and negative predictions(58,59), other types of assays, such as ATAC-seq, or functional assays could be used as a validation strategy”

References

1. Dykstra B, Olthof S, Schreuder J, Ritsema M, de Haan G. Clonal analysis reveals multiple functional defects of aged murine hematopoietic stem cells. J Exp Med. 2011 Dec 19;208(13):2691–703.

2. Pang WW, Price EA, Sahoo D, Beerman I, Maloney WJ, Rossi DJ, et al. Human bone marrow hematopoietic stem cells are increased in frequency and myeloid-biased with age. Proc Natl Acad Sci USA. 2011 Dec 13;108(50):20012–7.

3. Delgado MD, León J. Myc roles in hematopoiesis and leukemia. Genes Cancer. 2010 Jun;1(6):605–16.

4. Cabezas-Wallscheid N, Buettner F, Sommerkamp P, Klimmeck D, Ladel L, Thalheimer FB, et al. Vitamin A-Retinoic Acid Signaling Regulates Hematopoietic Stem Cell Dormancy. Cell. 2017 May 18;169(5):807-823.e19.

5. Hérault L, Poplineau M, Mazuel A, Platet N, Remy É, Duprez E. Single-cell RNA-seq reveals a concomitant delay in differentiation and cell cycle of aged hematopoietic stem cells. BMC Biol. 2021 Feb 1;19(1):19.

6. Kowalczyk MS, Tirosh I, Heckl D, Rao TN, Dixit A, Haas BJ, et al. Single-cell RNA-seq reveals changes in cell cycle and differentiation programs upon aging of hematopoietic stem cells. Genome Res. 2015 Dec;25(12):1860–72.

7. Graham SM, Vass JK, Holyoake TL, Graham GJ. Transcriptional analysis of quiescent and proliferating CD34+ human hemopoietic cells from normal and chronic myeloid leukemia sources. Stem Cells. 2007 Dec;25(12):3111–20.

8. Venezia TA, Merchant AA, Ramos CA, Whitehouse NL, Young AS, Shaw CA, et al. Molecular signatures of proliferation and quiescence in hematopoietic stem cells. PLoS Biol. 2004 Oct;2(10):e301.

9. Laurenti E, Doulatov S, Zandi S, Plumb I, Chen J, April C, et al. The transcriptional architecture of early human hematopoiesis identifies multilevel control of lymphoid commitment. Nat Immunol. 2013 Jul;14(7):756–63.

10. Kim M, Hwang S, Park K, Kim SY, Lee YK, Lee DS. Increased expression of interferon signaling genes in the bone marrow microenvironment of myelodysplastic syndromes. PLoS ONE. 2015 Mar 24;10(3):e0120602.

11. Aibar S, González-Blas CB, Moerman T, Huynh-Thu VA, Imrichova H, Hulselmans G, et al. SCENIC: single-cell regulatory network inference and clustering. Nat Methods. 2017 Nov;14(11):1083–6.

12. Pratapa A, Jalihal AP, Law JN, Bharadwaj A, Murali TM. Benchmarking algorithms for gene regulatory network inference from single-cell transcriptomic data. Nat Methods. 2020 Feb;17(2):147–54.

13. Kuranda K, Vargaftig J, de la Rochere P, Dosquet C, Charron D, Bardin F, et al. Age-related changes in human hematopoietic stem/progenitor cells. Aging Cell. 2011 Jun;10(3):542–6.

14. Rossi DJ, Bryder D, Zahn JM, Ahlenius H, Sonu R, Wagers AJ, et al. Cell intrinsic alterations underlie hematopoietic stem cell aging. Proc Natl Acad Sci USA. 2005 Jun 28;102(26):9194–9.

15. Dinauer MC, Newburger PE, Borregaard N. Phagocyte System and Disorders of Granulopoiesis and Granulocyte Function. Nathan and Oski’s Hematology and Oncology of Infancy and Childhood [Internet]. 8th ed. 2015 [cited 2022 Oct 11]. p. 773–847. Available from: https://www.clinicalkey.com/#!/content/book/3-s2.0-B978145575414400022X

16. Domínguez Conde C, Xu C, Jarvis LB, Rainbow DB, Wells SB, Gomes T, et al. Cross-tissue immune cell analysis reveals tissue-specific features in humans. Science. 2022 May 13;376(6594):eabl5197.

17. Bapat A, Schippel N, Shi X, Jasbi P, Gu H, Kala M, et al. Hypoxia promotes erythroid differentiation through the development of progenitors and proerythroblasts. Exp Hematol. 2021 May;97:32-46.e35.

https://doi.org/10.7554/eLife.79363.sa2

Article and author information

Author details

  1. Marina Ainciburu

    1. Area de Hemato-Oncología, Centro de Investigación Médica Aplicada, Universidad de Navarra, Instituto de investigación sanitaria de Navarra (IDISNA), Pamplona, Spain
    2. Centro de Investigación Biomédica en Red de Cáncer, Madrid, Spain
    Contribution
    Conceptualization, Writing – original draft, Writing – review and editing, Computational analysis
    Contributed equally with
    Teresa Ezponda
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-6483-1901

  2. Teresa Ezponda

    1. Area de Hemato-Oncología, Centro de Investigación Médica Aplicada, Universidad de Navarra, Instituto de investigación sanitaria de Navarra (IDISNA), Pamplona, Spain
    2. Centro de Investigación Biomédica en Red de Cáncer, Madrid, Spain
    Contribution
    Conceptualization, Writing – original draft, Writing – review and editing
    Contributed equally with
    Marina Ainciburu
    Competing interests
    No competing interests declared

  3. Nerea Berastegui

    Area de Hemato-Oncología, Centro de Investigación Médica Aplicada, Universidad de Navarra, Instituto de investigación sanitaria de Navarra (IDISNA), Pamplona, Spain
    Contribution
    Resources, Data curation, Data acquisition
    Competing interests
    No competing interests declared

  4. Ana Alfonso-Pierola

    1. Centro de Investigación Biomédica en Red de Cáncer, Madrid, Spain
    2. Clinica Universidad de Navarra, Pamplona, Spain
    Contribution
    Resources, Data acquisition
    Competing interests
    No competing interests declared

  5. Amaia Vilas-Zornoza

    1. Area de Hemato-Oncología, Centro de Investigación Médica Aplicada, Universidad de Navarra, Instituto de investigación sanitaria de Navarra (IDISNA), Pamplona, Spain
    2. Centro de Investigación Biomédica en Red de Cáncer, Madrid, Spain
    Contribution
    Resources, Data acquisition
    Competing interests
    No competing interests declared

  6. Patxi San Martin-Uriz

    1. Area de Hemato-Oncología, Centro de Investigación Médica Aplicada, Universidad de Navarra, Instituto de investigación sanitaria de Navarra (IDISNA), Pamplona, Spain
    2. Centro de Investigación Biomédica en Red de Cáncer, Madrid, Spain
    Contribution
    Resources, Data acquisition
    Competing interests
    No competing interests declared

  7. Diego Alignani

    Flow Cytometry Core, Universidad de Navarra, Pamplona, Spain
    Contribution
    Resources, Data acquisition
    Competing interests
    No competing interests declared

  8. Jose Lamo-Espinosa

    Clinica Universidad de Navarra, Pamplona, Spain
    Contribution
    Resources, Data acquisition
    Competing interests
    No competing interests declared

  9. Mikel San-Julian

    Clinica Universidad de Navarra, Pamplona, Spain
    Contribution
    Resources, Data acquisition
    Competing interests
    No competing interests declared

  10. Tamara Jiménez-Solas

    Hospital Universitario de Salamanca, Salamanca, Spain
    Contribution
    Resources, Data acquisition
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-5894-2023

  11. Felix Lopez

    Hospital Universitario de Salamanca, Salamanca, Spain
    Contribution
    Resources, Data acquisition
    Competing interests
    No competing interests declared

  12. Sandra Muntion

    1. Hospital Universitario de Salamanca, Salamanca, Spain
    2. Red de Investigación Cooperativa en Terapia Celular TerCel, ISCIII., Madrid, Spain
    Contribution
    Resources, Data acquisition
    Competing interests
    No competing interests declared

  13. Fermin Sanchez-Guijo

    1. Hospital Universitario de Salamanca, Salamanca, Spain
    2. Red de Investigación Cooperativa en Terapia Celular TerCel, ISCIII., Madrid, Spain
    Contribution
    Resources, Data acquisition
    Competing interests
    No competing interests declared

  14. Antonieta Molero

    Department of Hematology, Vall d'Hebron Hospital Universitari, Barcelona, Spain
    Contribution
    Resources, Data acquisition
    Competing interests
    No competing interests declared

  15. Julia Montoro

    Department of Hematology, Vall d'Hebron Hospital Universitari, Barcelona, Spain
    Contribution
    Resources, Data acquisition
    Competing interests
    No competing interests declared

  16. Guillermo Serrano

    Computational Biology Program, Universidad de Navarra, Pamplona, Spain
    Contribution
    Computational analysis
    Competing interests
    No competing interests declared

  17. Aintzane Diaz-Mazkiaran

    1. Centro de Investigación Biomédica en Red de Cáncer, Madrid, Spain
    2. Computational Biology Program, Universidad de Navarra, Pamplona, Spain
    Contribution
    Computational analysis
    Competing interests
    No competing interests declared

  18. Miren Lasaga

    Translational Bioinformatics Unit, NavarraBiomed, Pamplona, Spain
    Contribution
    Computational analysis
    Competing interests
    No competing interests declared

  19. David Gomez-Cabrero

    1. Translational Bioinformatics Unit, NavarraBiomed, Pamplona, Spain
    2. Biological & Environmental Sciences & Engineering Division, King Abdullah University of Science and Technology, Thuwal, Saudi Arabia
    Contribution
    Computational analysis
    Competing interests
    No competing interests declared

  20. Maria Diez-Campelo

    Hospital Universitario de Salamanca, Salamanca, Spain
    Contribution
    Resources, Data acquisition
    Competing interests
    No competing interests declared

  21. David Valcarcel

    Department of Hematology, Vall d'Hebron Hospital Universitari, Barcelona, Spain
    Contribution
    Resources, Data acquisition
    Competing interests
    No competing interests declared

  22. Mikel Hernaez

    Computational Biology Program, Universidad de Navarra, Pamplona, Spain
    Contribution
    Computational analysis
    Competing interests
    No competing interests declared

  23. Juan P Romero

    1. Area de Hemato-Oncología, Centro de Investigación Médica Aplicada, Universidad de Navarra, Instituto de investigación sanitaria de Navarra (IDISNA), Pamplona, Spain
    2. Centro de Investigación Biomédica en Red de Cáncer, Madrid, Spain
    Contribution
    Conceptualization, Supervision, Writing – original draft, Writing – review and editing, Computational analysis
    Contributed equally with
    Felipe Prosper
    For correspondence
    jromeror@unav.es
    Competing interests
    Employed by 10x Genomics since February 2021; this employment had no bearing on this work
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-0441-4791

  24. Felipe Prosper

    1. Area de Hemato-Oncología, Centro de Investigación Médica Aplicada, Universidad de Navarra, Instituto de investigación sanitaria de Navarra (IDISNA), Pamplona, Spain
    2. Centro de Investigación Biomédica en Red de Cáncer, Madrid, Spain
    3. Clinica Universidad de Navarra, Pamplona, Spain
    4. Red de Investigación Cooperativa en Terapia Celular TerCel, ISCIII., Madrid, Spain
    Contribution
    Conceptualization, Supervision, Writing – original draft, Writing – review and editing
    Contributed equally with
    Juan P Romero
    For correspondence
    fprosper@unav.es
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-6115-8790

Funding

Instituto de Salud Carlos III

  • Marina Ainciburu
  • Teresa Ezponda
  • Nerea Berastegui
  • Ana Alfonso-Pierola
  • Amaia Vilas-Zornoza
  • Patxi San Martin-Uriz
  • Diego Alignani
  • Jose Lamo-Espinosa
  • Mikel San-Julian
  • Tamara Jiménez-Solas
  • Felix Lopez
  • Sandra Muntion
  • Fermin Sanchez-Guijo
  • Antonieta Molero
  • Julia Montoro
  • Guillermo Serrano
  • Aintzane Diaz-Mazkiaran
  • Miren Lasaga
  • David Gomez-Cabrero
  • Maria Diez-Campelo
  • David Valcarcel
  • Mikel Hernaez
  • Juan P Romero
  • Felipe Prosper

Federación Española de Enfermedades Raras (PI17/00701)

  • Marina Ainciburu
  • Teresa Ezponda
  • Nerea Berastegui
  • Ana Alfonso-Pierola
  • Amaia Vilas-Zornoza
  • Patxi San Martin-Uriz
  • Diego Alignani
  • Jose Lamo-Espinosa
  • Mikel San-Julian
  • Tamara Jiménez-Solas
  • Felix Lopez
  • Sandra Muntion
  • Fermin Sanchez-Guijo
  • Antonieta Molero
  • Julia Montoro
  • Guillermo Serrano
  • Aintzane Diaz-Mazkiaran
  • Miren Lasaga
  • David Gomez-Cabrero
  • Maria Diez-Campelo
  • David Valcarcel
  • Mikel Hernaez
  • Juan P Romero
  • Felipe Prosper

Federación Española de Enfermedades Raras (PI20/01308)

  • Marina Ainciburu
  • Teresa Ezponda
  • Nerea Berastegui
  • Ana Alfonso-Pierola
  • Amaia Vilas-Zornoza
  • Patxi San Martin-Uriz
  • Diego Alignani
  • Jose Lamo-Espinosa
  • Mikel San-Julian
  • Tamara Jiménez-Solas
  • Felix Lopez
  • Sandra Muntion
  • Fermin Sanchez-Guijo
  • Antonieta Molero
  • Julia Montoro
  • Guillermo Serrano
  • Aintzane Diaz-Mazkiaran
  • Miren Lasaga
  • David Gomez-Cabrero
  • Maria Diez-Campelo
  • David Valcarcel
  • Mikel Hernaez
  • Juan P Romero
  • Felipe Prosper

Federación Española de Enfermedades Raras (PI19/00726)

  • Marina Ainciburu
  • Teresa Ezponda
  • Nerea Berastegui
  • Ana Alfonso-Pierola
  • Amaia Vilas-Zornoza
  • Patxi San Martin-Uriz
  • Diego Alignani
  • Jose Lamo-Espinosa
  • Mikel San-Julian
  • Tamara Jiménez-Solas
  • Felix Lopez
  • Sandra Muntion
  • Fermin Sanchez-Guijo
  • Antonieta Molero
  • Julia Montoro
  • Guillermo Serrano
  • Aintzane Diaz-Mazkiaran
  • Miren Lasaga
  • David Gomez-Cabrero
  • Maria Diez-Campelo
  • David Valcarcel
  • Mikel Hernaez
  • Juan P Romero
  • Felipe Prosper

Centro de Investigación Biomédica en Red de Cáncer (CB16/12/00489)

  • Marina Ainciburu
  • Teresa Ezponda
  • Nerea Berastegui
  • Ana Alfonso-Pierola
  • Amaia Vilas-Zornoza
  • Patxi San Martin-Uriz
  • Diego Alignani
  • Jose Lamo-Espinosa
  • Mikel San-Julian
  • Tamara Jiménez-Solas
  • Felix Lopez
  • Sandra Muntion
  • Fermin Sanchez-Guijo
  • Antonieta Molero
  • Julia Montoro
  • Guillermo Serrano
  • Aintzane Diaz-Mazkiaran
  • Miren Lasaga
  • David Gomez-Cabrero
  • Maria Diez-Campelo
  • David Valcarcel
  • Mikel Hernaez
  • Juan P Romero
  • Felipe Prosper

Centro de Investigación Biomédica en Red de Cáncer (CB16/12/00225)

  • Marina Ainciburu
  • Teresa Ezponda
  • Nerea Berastegui
  • Ana Alfonso-Pierola
  • Amaia Vilas-Zornoza
  • Patxi San Martin-Uriz
  • Diego Alignani
  • Jose Lamo-Espinosa
  • Mikel San-Julian
  • Tamara Jiménez-Solas
  • Felix Lopez
  • Sandra Muntion
  • Fermin Sanchez-Guijo
  • Antonieta Molero
  • Julia Montoro
  • Guillermo Serrano
  • Aintzane Diaz-Mazkiaran
  • Miren Lasaga
  • David Gomez-Cabrero
  • Maria Diez-Campelo
  • David Valcarcel
  • Mikel Hernaez
  • Juan P Romero
  • Felipe Prosper

Gobierno de Navarra (ERAPerMed MEET-AML 0011-2750-2019-000001)

  • Marina Ainciburu
  • Teresa Ezponda
  • Nerea Berastegui
  • Ana Alfonso-Pierola
  • Amaia Vilas-Zornoza
  • Patxi San Martin-Uriz
  • Diego Alignani
  • Jose Lamo-Espinosa
  • Mikel San-Julian
  • Tamara Jiménez-Solas
  • Felix Lopez
  • Sandra Muntion
  • Fermin Sanchez-Guijo
  • Antonieta Molero
  • Julia Montoro
  • Guillermo Serrano
  • Aintzane Diaz-Mazkiaran
  • Miren Lasaga
  • David Gomez-Cabrero
  • Maria Diez-Campelo
  • David Valcarcel
  • Mikel Hernaez
  • Juan P Romero
  • Felipe Prosper

Gobierno de Navarra (DIANA)

  • Marina Ainciburu
  • Teresa Ezponda
  • Nerea Berastegui
  • Ana Alfonso-Pierola
  • Amaia Vilas-Zornoza
  • Patxi San Martin-Uriz
  • Diego Alignani
  • Jose Lamo-Espinosa
  • Mikel San-Julian
  • Tamara Jiménez-Solas
  • Felix Lopez
  • Sandra Muntion
  • Fermin Sanchez-Guijo
  • Antonieta Molero
  • Julia Montoro
  • Guillermo Serrano
  • Aintzane Diaz-Mazkiaran
  • Miren Lasaga
  • David Gomez-Cabrero
  • Maria Diez-Campelo
  • David Valcarcel
  • Mikel Hernaez
  • Juan P Romero
  • Felipe Prosper

"la Caixa" Foundation (GR-NET NORMAL-HIT HR20-00871)

  • Marina Ainciburu
  • Teresa Ezponda
  • Nerea Berastegui
  • Ana Alfonso-Pierola
  • Amaia Vilas-Zornoza
  • Patxi San Martin-Uriz
  • Diego Alignani
  • Jose Lamo-Espinosa
  • Mikel San-Julian
  • Tamara Jiménez-Solas
  • Felix Lopez
  • Sandra Muntion
  • Fermin Sanchez-Guijo
  • Antonieta Molero
  • Julia Montoro
  • Guillermo Serrano
  • Aintzane Diaz-Mazkiaran
  • Miren Lasaga
  • David Gomez-Cabrero
  • Maria Diez-Campelo
  • David Valcarcel
  • Mikel Hernaez
  • Juan P Romero
  • Felipe Prosper

Cancer Research UK (C355/A26819)

  • Marina Ainciburu
  • Teresa Ezponda
  • Nerea Berastegui
  • Ana Alfonso-Pierola
  • Amaia Vilas-Zornoza
  • Patxi San Martin-Uriz
  • Diego Alignani
  • Jose Lamo-Espinosa
  • Mikel San-Julian
  • Tamara Jiménez-Solas
  • Felix Lopez
  • Sandra Muntion
  • Fermin Sanchez-Guijo
  • Antonieta Molero
  • Julia Montoro
  • Guillermo Serrano
  • Aintzane Diaz-Mazkiaran
  • Miren Lasaga
  • David Gomez-Cabrero
  • Maria Diez-Campelo
  • David Valcarcel
  • Mikel Hernaez
  • Juan P Romero
  • Felipe Prosper

Fundación Científica Asociación Española Contra el Cáncer (Accelerator Award Program)

  • Marina Ainciburu
  • Teresa Ezponda
  • Nerea Berastegui
  • Ana Alfonso-Pierola
  • Amaia Vilas-Zornoza
  • Patxi San Martin-Uriz
  • Diego Alignani
  • Jose Lamo-Espinosa
  • Mikel San-Julian
  • Tamara Jiménez-Solas
  • Felix Lopez
  • Sandra Muntion
  • Fermin Sanchez-Guijo
  • Antonieta Molero
  • Julia Montoro
  • Guillermo Serrano
  • Aintzane Diaz-Mazkiaran
  • Miren Lasaga
  • David Gomez-Cabrero
  • Maria Diez-Campelo
  • David Valcarcel
  • Mikel Hernaez
  • Juan P Romero
  • Felipe Prosper

Associazione Italiana per la Ricerca sul Cancro (Accelerator Award Program)

  • Marina Ainciburu
  • Teresa Ezponda
  • Nerea Berastegui
  • Ana Alfonso-Pierola
  • Amaia Vilas-Zornoza
  • Patxi San Martin-Uriz
  • Diego Alignani
  • Jose Lamo-Espinosa
  • Mikel San-Julian
  • Tamara Jiménez-Solas
  • Felix Lopez
  • Sandra Muntion
  • Fermin Sanchez-Guijo
  • Antonieta Molero
  • Julia Montoro
  • Guillermo Serrano
  • Aintzane Diaz-Mazkiaran
  • Miren Lasaga
  • David Gomez-Cabrero
  • Maria Diez-Campelo
  • David Valcarcel
  • Mikel Hernaez
  • Juan P Romero
  • Felipe Prosper

Gobierno de Navarra (PhD fellowship 0011-0537-2019-000001)

  • Nerea Berastegui

Ministerio de Ciencia e Innovación (PhD fellowship FPU18/05488)

  • Marina Ainciburu

Fundación Científica Asociación Española Contra el Cáncer (Investigador AECC award)

  • Teresa Ezponda

H2020 Marie Skłodowska-Curie Actions (Grant Agreement No. 898356)

  • Mikel Hernaez

Gobierno de Navarra (AGATA 0011-1411-2020-000010/0011-1411-2020-000011)

  • Marina Ainciburu

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

We particularly acknowledge the patients for their participation and the Biobank of the University of Navarra for its collaboration. This work was supported by the Instituto de Salud Carlos III and co-finance by FEDER funds (PI17/00701, PI19/00726, and PI20/01308), CIBERONC (CB16/12/00489 and CB16/12/00225); Gobierno de Navarra (ERAPerMed MEET-AML 0011-2750-2019-000001; AGATA 0011-1411-2020-000010/0011-1411-2020-000011 and DIANA 0011-1411-2017-000028/0011-1411-2017-000029/0011-1411-2017-000030); Fundación La Caixa (GR-NET NORMAL-HIT HR20-00871); and Cancer Research UK (C355/A26819) and FC AECC and AIRC under the Accelerator Award Program. NB was supported by a PhD fellowship from Gobierno de Navarra (0011-0537-2019-000001); MA was supported by a PhD fellowship from Ministerio de Ciencia, Innovación y Universidades (FPU18/05488); TE was supported by an Investigador AECC award from the Fundación AECC. MH was supported by H2020 Marie S Curie IF Action, European Commission, Grant Agreement No. 898356

Ethics

The samples and data from the patients included in the study were provided by the Biobank of the University of Navarra and were processed according to standard operating procedures. Patients and healthy donors provided informed consent, together with consent for publication. The study was approved by the Clinical Research Ethics Committee of the Clinica Universidad de Navarra, following protocol # 2017.218.

Senior Editor

  1. Utpal Banerjee, University of California, Los Angeles, United States

Reviewing Editor

  1. Jiwon Shim, Hanyang University, Republic of Korea

Reviewer

  1. Jong Kyoung Kim, Pohang University of Science and Technology, Republic of Korea

Publication history

  1. Preprint posted: July 31, 2021 (view preprint)
  2. Received: April 8, 2022
  3. Accepted: January 10, 2023
  4. Accepted Manuscript published: January 11, 2023 (version 1)
  5. Version of Record published: February 7, 2023 (version 2)

Copyright

© 2023, Ainciburu, Ezponda et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 1,177
    Page views
  • 284
    Downloads
  • 0
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Open citations (links to open the citations from this article in various online reference manager services)

Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)

  1. Marina Ainciburu
  2. Teresa Ezponda
  3. Nerea Berastegui
  4. Ana Alfonso-Pierola
  5. Amaia Vilas-Zornoza
  6. Patxi San Martin-Uriz
  7. Diego Alignani
  8. Jose Lamo-Espinosa
  9. Mikel San-Julian
  10. Tamara Jiménez-Solas
  11. Felix Lopez
  12. Sandra Muntion
  13. Fermin Sanchez-Guijo
  14. Antonieta Molero
  15. Julia Montoro
  16. Guillermo Serrano
  17. Aintzane Diaz-Mazkiaran
  18. Miren Lasaga
  19. David Gomez-Cabrero
  20. Maria Diez-Campelo
  21. David Valcarcel
  22. Mikel Hernaez
  23. Juan P Romero
  24. Felipe Prosper
(2023)
Uncovering perturbations in human hematopoiesis associated with healthy aging and myeloid malignancies at single-cell resolution
eLife 12:e79363.
https://doi.org/10.7554/eLife.79363

Categories and tags

Research organism

Further reading

    1. Computational and Systems Biology
    2. Neuroscience

    Flexible control of representational dynamics in a disinhibition-based model of decision making

    Bo Shen, Kenway Louie, Paul W Glimcher
    Research Article

    Inhibition is crucial for brain function, regulating network activity by balancing excitation and implementing gain control. Recent evidence suggests that beyond simply inhibiting excitatory activity, inhibitory neurons can also shape circuit function through disinhibition. While disinhibitory circuit motifs have been implicated in cognitive processes including learning, attentional selection, and input gating, the role of disinhibition is largely unexplored in the study of decision-making. Here, we show that disinhibition provides a simple circuit motif for fast, dynamic control of network state and function. This dynamic control allows a disinhibition-based decision model to reproduce both value normalization and winner-take-all dynamics, the two central features of neurobiological decision-making captured in separate existing models with distinct circuit motifs. In addition, the disinhibition model exhibits flexible attractor dynamics consistent with different forms of persistent activity seen in working memory. Fitting the model to empirical data shows it captures well both the neurophysiological dynamics of value coding and psychometric choice behavior. Furthermore, the biological basis of disinhibition provides a simple mechanism for flexible top-down control of the network states, enabling the circuit to capture diverse task-dependent neural dynamics. These results suggest a biologically plausible unifying mechanism for decision-making and emphasize the importance of local disinhibition in neural processing.

    1. Computational and Systems Biology

    Resource allocation accounts for the large variability of rate-yield phenotypes across bacterial strains

    Valentina Baldazzi, Delphine Ropers ... Hidde de Jong
    Research Article

    Different strains of a microorganism growing in the same environment display a wide variety of growth rates and growth yields. We developed a coarse-grained model to test the hypothesis that different resource allocation strategies, corresponding to different compositions of the proteome, can account for the observed rate-yield variability. The model predictions were verified by means of a database of hundreds of published rate-yield and uptake-secretion phenotypes of Escherichia coli strains grown in standard laboratory conditions. We found a very good quantitative agreement between the range of predicted and observed growth rates, growth yields, and glucose uptake and acetate secretion rates. These results support the hypothesis that resource allocation is a major explanatory factor of the observed variability of growth rates and growth yields across different bacterial strains. An interesting prediction of our model, supported by the experimental data, is that high growth rates are not necessarily accompanied by low growth yields. The resource allocation strategies enabling high-rate, high-yield growth of E. coli lead to a higher saturation of enzymes and ribosomes, and thus to a more efficient utilization of proteomic resources. Our model thus contributes to a fundamental understanding of the quantitative relationship between rate and yield in E. coli and other microorganisms. It may also be useful for the rapid screening of strains in metabolic engineering and synthetic biology.

    1. Computational and Systems Biology
    2. Neuroscience

    Contrasting action and posture coding with hierarchical deep neural network models of proprioception

    Kai J Sandbrink, Pranav Mamidanna ... Alexander Mathis
    Research Article

    Biological motor control is versatile, efficient, and depends on proprioceptive feedback. Muscles are flexible and undergo continuous changes, requiring distributed adaptive control mechanisms that continuously account for the body's state. The canonical role of proprioception is representing the body state. We hypothesize that the proprioceptive system could also be critical for high-level tasks such as action recognition. To test this theory, we pursued a task-driven modeling approach, which allowed us to isolate the study of proprioception. We generated a large synthetic dataset of human arm trajectories tracing characters of the Latin alphabet in 3D space, together with muscle activities obtained from a musculoskeletal model and model-based muscle spindle activity. Next, we compared two classes of tasks: trajectory decoding and action recognition, which allowed us to train hierarchical models to decode either the position and velocity of the end-effector of one's posture or the character (action) identity from the spindle firing patterns. We found that artificial neural networks could robustly solve both tasks, and the networks'units show tuning properties similar to neurons in the primate somatosensory cortex and the brainstem. Remarkably, we found uniformly distributed directional selective units only with the action-recognition-trained models and not the trajectory-decoding-trained models. This suggests that proprioceptive encoding is additionally associated with higher-level functions such as action recognition and therefore provides new, experimentally testable hypotheses of how proprioception aids in adaptive motor control.