Abstract
Inflammatory cytokines perturb hematopoietic stem cell (HSC) homeostasis and modulate the fitness of neoplastic HSC clones in mouse models. However, the study of cytokines in human hematopoiesis is challenging due to the concerted activities of multiple cytokines across physiologic and pathologic processes. To overcome this limitation, we leveraged serial bone marrow samples from patients with CALR-mutated myeloproliferative neoplasms who were treated with recombinant interferon-alpha (IFNa). We interrogated baseline and IFNa-treated CD34+ stem and progenitor cells using single-cell multi-omics platforms that directly link, within the same cell, the mutation status, whole transcriptomes and immunophenotyping or chromatin accessibility. We identified a novel IFNa-induced inflammatory granulocytic progenitor defined by expression and activities of RFX2/3 and AP-1 transcription factors, with evidence supporting a direct differentiation from HSCs. On the other hand, IFNa also induced a significant B-lymphoid progenitor expansion and proliferation, associated with enhanced activities of PU.1 and its co- regulator TCF3, as well as decreased accessibility of megakaryocytic-erythroid transcription factor GATA1 binding sites in HSCs. In the neoplastic hematopoiesis, the lymphoid expansion was constrained by a preferential myeloid skewing of the mutated cells, linked with increased myeloid proliferation and enhanced CEBPA and GATA1 activities compared to wildtype cells. Further, IFNa caused a downregulation of the TNFa signaling pathway, with downregulation of NFKB and AP-1 transcription factors. Thus, IFNa simultaneously initiated both – pro-inflammatory and anti- inflammatory – cell states within the same hematopoiesis, and its phenotypic impact varied as a function of the underlying HSC state and mutation status.
Introduction
Inflammation perturbs hematopoietic stem cell (HSC) homeostasis. The isolated effects of individual inflammatory cytokines, including type 1 interferons (IFNa/b)1–3, type 2 IFN (IFNg)4–6, TNFa7, 8, and IL-19–11, have been well-documented in mouse hematopoietic development. These cytokines display overlapping features, such as inducing cell cycle entry of HSCs and their differentiation, frequently toward the granulo-monocytic lineage1, 4–6, 12–15. However, these cytokines also feature distinct roles in the inflammatory milieu. Recent reports, for example, link germline genetic defects in type 1 IFN with exaggerated or severe COVID-19 response associated with neutrophilia16, 17. However, deciphering the contributions of individual cytokines in human hematopoiesis is challenging due to the concerted activities of multiple cytokines in the setting of infection, inflammatory disorders or cancer. Moreover, HSC-enriched bone marrow biopsies are rare, as they are restricted to patients with suspected or diagnosed hematopoietic neoplasms or defects.
The use of recombinant IFNa for the treatment of patients with myeloproliferative neoplasms (MPN) therefore presents a unique opportunity to study the effects of isolated type 1 IFN in human HSCs. MPNs are often driven by a single somatic mutation in CALR, JAK2 or MPL, and thus provide an informative model of an early clonal process within the hematopoietic system18. Clinically, these patients present with an overproduction of one or more myeloid cell lineages. IFNa remains a highly effective therapy for MPN, frequently resulting in the normalization of the patients’ blood counts and even molecular response (i.e. reduction in variant allele frequencies), with evidence suggesting that IFNa may deplete clonal stem cells via induction of HSCs into cell cycle entry and exhaustion12, 18. However, the effects of IFNa on HSCs appear to be genotype- specific as IFNa frequently induces molecular response in patients with JAK2-mutated MPN but not in those with CALR-mutated MPN19. Thus, the molecular effects undergirding the clinical improvements of CALR-mutated MPN through IFNa therapy remain an open arena of investigations.
While MPNs provide a useful model to study the effects of somatic mutations on human hematopoiesis, the precise examination of IFNa effects on mutated versus wildtype blood formation is challenging because the mutated cells are not distinguishable from the admixed wildtype cells by cell surface markers. We therefore leveraged single-cell multi-omics platforms that detect the mutation status and whole transcriptomes with immunophenotyping or chromatin accessibility data, within the same thousands of cells. These innovative platforms build upon a method called Genotyping of Transcriptomes (GoT), in which we modified a high-throughput single-cell RNA-seq (scRNA-seq) platform to capture somatic genotype information within the same cells20. These methods allow us to overlay two hematopoietic differentiation landscapes— one mutated and the other wildtype—from the same individual, thus enabling a direct comparison between mutated and wildtype cells, before and after treatment. Focusing on CALR-mutated MPN, we applied these multi-modality single-cell methods to CD34+ hematopoietic stem and progenitor cells (HSPC) from serial bone marrow sampling from patients who were treated with IFNa for at least one year, enabling us to define the transcriptional and epigenetic alterations induced by IFNa in normal and neoplastic human hematopoiesis.
Results
GoT-IM captures cell identity, somatic genotyping and treatment status information for thousands of CD34+ cells from MPN patients treated with IFNa
To determine the in vivo effects of isolated IFNa on human normal and neoplastic hematopoiesis, we leveraged the GoT technology that simultaneously captures the mutation status and whole transcriptomes in thousands of single cells20. To overcome inter-patient variabilities in IFNa response, we applied GoT to FACS-isolated CD34+ cells from serial (i.e. baseline and treated) bone marrow from individuals who were diagnosed with CALR-mutated essential thrombocythemia (ET), a subtype of MPN that exhibits increased megakaryopoiesis and platelet counts21 (Fig. 1a). As these serial bone marrow specimens are exceedingly rare in clinical practice, we utilized cryopreserved specimens from our clinical trials MPN-RC-111 and -112 wherein patients were treated weekly with a pegylated form of IFNa22, 23 (Fig. 1a, n = 5 individuals, 4 baseline, 7 treated; additional five baseline samples were included from our previous work20; see Supplementary Table 1 for patient and sample information). We further incorporated cell hashing24 that enabled multiplexing baseline and IFNa-treated samples into the same scRNA-seq reactions via time-point specifying barcoded antibodies, to obviate technical batch effects between serial samples (Fig. 1b). We also advanced upon GoT by incorporating immunophenotyping (GoT-IM), through integration of Cellular Indexing of Transcriptomes and Epitopes by Sequencing (CITE-seq)25, to link cell identities to cell surface protein expression (Fig. 1b). GoT provided genotyping data for the canonical CALR frameshift mutations for 81.5% of CD34+ HSPCs (n = 40,668), consistent with our previously reported genotyping rates20. In this way, we obtained somatic genotyping, whole transcriptomes, immunophenotyping and treatment status for thousands of cells from the same GoT-IM experiment.
a. Representation of primary bone marrow samples at baseline and after IFNa treatment. b. Schematic of Genotyping of Transcriptomes with Immunophenotyping (GoT-IM), via CITE-seq and treatment status via Cell Hashing. MPN, myeloproliferative neoplasms; WT, wildtype; MUT, mutated. c. Uniform manifold approximation and projection (UMAP) of CD34+ cells (n = 49,891 cells) from MPN samples (n = 16 samples from 10 individuals), overlaid with cell type assignment. HSC, hematopoietic stem cells; IMP, immature myeloid progenitors; NP, neutrophilic progenitors; MP, monocytic progenitors; cDCP, classic dendritic progenitors; pDCP, plasmacytoid dendritic progenitors; MDP, monocytic dendritic progenitors; MLP, multipotent lymphoid progenitors; E/B/M, eosinophil/basophil/mast cells; MkP, megakaryocytic progenitors; EP, erythroid progenitors; MEP, megakaryocytic-erythroid progenitors. d. Violin plots showing normalized expression of HSC-defining protein and RNA markers. e. Volcano plot showing genes differentially expressed (DE) between IGPs (Cluster X) and IMPs identified via linear mixed modeling with/without cluster identity. Left panel: Volcano plot of DE genes; Genes highlighted in purple represent genes enriched in the MYC pathway and those in blue enriched in the TNFa signaling via NFKB; Boxes represent transcription factors (TF) of the AP-1 (blue), KLF (light green), and NR4A (purple) families and genes involved in emergency or stress-induced granulopoiesis (dark green). Right panel: Pre-ranked gene set enrichment analysis using the MSigDB Hallmark collection. IGP, inflammatory granulocytic progenitors. f. UMAP showing expression levels of representative TF that are upregulated in IGPs. g. Left panel: UMAP highlighting IGP, IMP and the HSC subclusters. Right panel: Dot plot showing expression levels of upregulated TFs in IGPs (vs. IMPs) h. DE genes between IGPs and HSC1 identified via linear mixed modeling with/without cluster identity. Only IFNa-treated cells were included in the DE analysis to remove the treatment variable altogether. Left panel: Volcano plot of DE genes; Genes highlighted in blue represent genes enriched in the TNFa signaling via NFKB and those in green enriched in G2M Checkpoint pathway; Boxes represent transcription factors (TF) of the AP-1 (blue) and KLF (light green) families. Right panel: Pre-ranked gene set enrichment analysis using the MSigDB Hallmark collection. i. UMAP of myeloid immune cells under COVID-19 infection displaying previously-assigned cell types (left panel) and projection of the IGP gene signature score (right panel). scRNA-seq data from an available dataset of immune cells from COVID-19 patients38. Neu, neutrophil; moDC, MD dendritic cell, rMa, Resident macrophage; nrMa, non- resident macrophage; MC, mast cell; MoD-Ma, MD macrophage.
Integration of CD34+ cells across baseline and IFNa treated samples reveals a novel inflammatory cell state
To determine the cell identities of the CD34+ HSPCs consistently across samples and treatment status, we analytically separated the cells from each timepoint from the same individual into individual datasets (based on the Cell Hashing data), and integrated across all of the 16 samples based on the canonical correlation analytic framework26 (n = 49,891 cells from 16 samples, Fig. 1c, Supplementary Fig. 1a,b). As single-cell gene expression provides high-resolution mapping of the HSPC identities, we clustered the cells based on gene expression data alone and annotated the clusters based on known canonical cell markers (Supplementary Fig 2a,b). To precisely identify the HSCs, we leveraged immunophenotyping via CITE-seq to identify the CD38-, CD45RA-, CD90+ HSCs (Fig. 1d, Supplementary Fig. 2c). Consistently, the RNA expression of the HSC gene marker, AVP, was also elevated in this population (Fig. 1d, Supplementary Fig. 2c, see Supplementary Table 2 for cell numbers). We observed the expected cell types, such as megakaryocytic progenitors (MkP), immature myeloid progenitors (IMP, consisting predominantly of phenotypically-defined common myeloid progenitors (CMP) and granulo- monocytic progenitors (GMP)27), and lymphoid progenitors (Pro-B, Large Pre-B, Small Pre-B cells). We also identified cell types not previously described in scRNA-seq datasets of human CD34+ cells. Namely, we identified monocytic dendritic progenitors (MDP) which were distinct from the uni-lineage monocytic (MP), classic dendritic (cDCP) or plasmacytoid dendritic progenitors (pDCP). They were assigned as such based on high SOX4 and RUNX3 expression (Supplementary Fig. 2a,b), which were characteristic of the MDP population in mice28. Transcriptionally, the MDPs were closely related to multipotent lymphoid progenitors (MLPs, Fig. 1c), consistent with data demonstrating that MLPs can give rise to monocytes and dendritic cells, along with lymphoid cells, in human29.
a Left panel: UMAP showing IFNa treatment time-points for a representative experiment that includes three time points from a patient (IFN03, n = 7329 cells). Right panel: Box plot showing transcriptional distance measurements between HSCs from each time point to HSCs from baseline (IFN03). b. Normalized cell type frequencies at baseline and IFNa treatment. Cells from each sample were down-sampled to the same number (n = 500 cells from each sample, n = 9 baseline samples, n = 5 treated samples). c. Box plot showing normalized cell type frequencies at baseline and IFNa treatment (n = 9 baseline samples, n = 5 treated samples). P-values from linear mixed model with/without treatment status. d. Fold change of IFNa versus baseline cell frequencies in individuals with serial sampling (n = 8 samples from 4 patients). P-values from linear mixed model with/without treatment status. e. Box plot showing normalized protein expression on HSCs before and after treatment with IFNa. P-values from linear mixed model with/without treatment status. f. Box plot showing cell frequencies of B-lymphoid progenitors and B-cells from bone marrow of patients with early stage MPN treated or not with IFNa (n = 9 and 47 samples, respectively), as determined by multiparametric flow cytometry. P-values from Wilcoxon rank sum test, two-sided. g. Platelet counts versus frequencies of MEPs and MkPs (n = 14 samples, P-value from F-test). h. Cell cycle gene expression in HSCs and progenitor cells (representative patient IFN01). i. Frequencies of cells in G2/M/S phase as assessed in panel h (n = 8 samples from 4 patients). P-values from linear mixed model with/without treatment status. j. Volcano plot showing DE genes between baseline and IFNa-treated HSCs via linear mixed modeling with/without treatment status. Genes highlighted in blue-green are enriched in the TNFa signaling via NFKB and those in blue enriched in the IFNa/g response, identified by pre-ranked gene set enrichment analysis using the MSigDB Hallmark collection; Boxes represent transcription factors (TF) of the AP-1 (blue) and NR4A (purple) families. k. Heatmap showing results of the pre-rank gene set enrichment analysis of genes DE before and after treatment with IFNa across HSC and progenitor subsets. Values show the sign of the normalized enrichment score (NES) multiply by -log10(Adjusted P-value). l. Pre-rank gene set enrichment analysis of DE genes before and after treatment with IFNa in cDCPs, showing downregulation of TNFa signaling via NFKB and the leading-edge genes. Genes highlighted in blue represent those upregulated in the IGP vs. IMP. m. Box plot showing HSC-specific IFNa-induced signature score in IFNa-treated HSCs and IGPs. Score calculated using upregulated or downregulated genes (left and right panels, respectively).
We also identified an unknown cluster (Cluster X), adjacent to the HSCs in the UMAP space (Fig. 1c), that resembled the IMPs immunophenotypically based on CD38mid, CD90lo, and CD45RAmid expression (Fig. 1d). To elucidate the identity of Cluster X, we performed differential expression analysis between Cluster X and IMPs and identified a striking upregulation of the immediate early response transcription factors (TF) of the AP-1 (JUN, FOS, JUNB, FOSB, ATF3, FOSL1, MAFF), KLF (KLF2, KLF4, KLF6), and NR4A (NR4A1, NR4A2) families (Fig. 1e left, Supplementary Fig. 2d, Supplementary Table 3, linear mixed model that explicitly models the effects of patient batch and treatment status, see online methods). We also observed a robust upregulation of RFX2 and RFX3 TFs, not well characterized in hematopoietic stem and progenitor cells (Fig. 1e left, Supplementary Fig. 2d, Supplementary Table 3). Upregulation of CEBPB and CEBPD (Fig. 1e left, Supplementary Fig. 2d, Supplementary Table 3), implicated in emergency granulopoiesis30 and granulopoiesis under cellular stress31 respectively, suggested a differentiation state into the granulocytic lineage. Interferon regulatory factor 1 (IRF1) was also upregulated (Fig. 1e left, Supplementary Table 3), indicating that the cluster was associated with IFNa treatment. Gene set enrichment analysis identified the upregulation of TNFa via NFKB pathway (Adj. P-val = 3.5 x 10-8, Hallmark), and downregulation of the MYC targets (Adj. P-val = 6.8 x 10-6, Hallmark, Fig. 1e right, Supplementary Table 4). In light of these findings, this novel cluster was named inflammatory granulocytic progenitor (IGP).
Strikingly, the distributions of these differentially upregulated TFs were highly enriched in the most quiescent HSCs with the highest overall expression of AVP and CD90 (HSC1, Fig. 1f,g, Supplementary Fig. 2c), driving the transcriptional similarities of the IGPs and HSC1 as revealed by their proximity on the UMAP space. To determine changes in gene expression that may drive the differentiation from HSC1 to IGPs, we compared IGPs to HSC1 and identified a reinforcement of the AP-1 and KLF family TF expression and downregulation of NR4A TFs (Fig. 1g,h, Supplementary Table 3,4; notably, only IFNa-treated cells were included in the differential expression analysis to remove the treatment variable altogether). NR4A1/2 have been demonstrated to maintain HSC quiescence32–34. Their downregulation relative to HSC1 is therefore consistent with the upregulation of cell cycle-related genes in the IGPs (Fig. 1h, Supplementary Table 3,4). MPO, a neutrophil-specific gene, and CSF3R, encoding the G-CSF receptor, were upregulated while the MHC class II genes (CD74, HLA-DRA, HLA-DPA1, HLA-DRB1, HLA- DPB1) were downregulated, further providing evidence for its differentiation into a neutrophilic lineage versus monocytic or dendritic cell lineages (Fig. 1h, Supplementary Table 3). Although RFX2/3 are not well characterized in hematopoiesis, other RFX members such as RFX1 and RFX8 have been demonstrated to regulate MHC class II expression35–37. Thus, the strong upregulation of RFX2/3 expression may be induced to rapidly downregulate MCH class II genes for granulocytic differentiation upon IFNa signaling. Of the interferon regulatory factors, IRF1 was upregulated in IGP compared to HSC1 (Fig. 1h, Supplementary Table 3; both comparison groups under IFNa treatment) and IMP (Fig. 1e, Supplementary Table 3), highlighting IRF1 as the key regulatory factor among other IFNa-induced TFs in IGP development. As only mononuclear cells were cryopreserved in these patient cohorts, precluding the assessment of neutrophils, we leveraged an available dataset of immune cells under inflammatory conditions38 (in this case, COVID-19 infection). We identified that the IGP-specific gene signature was highly specific to neutrophils, providing evidence against a general inflammatory signature and further supporting the predicted lineage fate (Fig. 1i). Furthermore, as the IGPs constitute a small minority of the progenitors (<1%), IGPs are not likely to contribute significantly to overall neutrophil pool during IFNa therapy, consistent with the absence of neutrophilia in patients treated with IFNa. Transcriptional similarity of IGPs to the most quiescent HSC1 population, especially with respect to the TF program, provides evidence for a novel route of granulocytic differentiation that bypasses the intermediate IMP states upon IFNa stimulation.
IFNa induces IGPs and lymphoid differentiation shift and proliferation
Next, we sought to determine the phenotypic and transcriptional impact of IFNa on hematopoiesis overall. As individual GoT-IM experiments represent sampling from the same individual at different time points, we clustered cells from each individual separately based on the gene expression data and identified that IFNa exerted strong transcriptional perturbations in the treated cells compared to baseline HSCs (Fig. 2a, Supplementary Fig. 3a). In an example case of patient IFN03 for whom three time points were available – (1) baseline, (2) treated for one year, and (3) treated for four years but off therapy for 3 weeks at time of collection – HSCs at year 1 displayed a striking transcriptional difference compared to baseline cells, whereas cells that have been off therapy for 3 weeks at year 4 were less distinct, consistent with a half-life of the pegylated-IFNa of ∼2 weeks (Fig. 2a). Projection of the cell identity assignments (as in Fig. 1c) revealed that the cells clustered based on cell identity as well as based on treatment status (Fig. 2a, Supplementary Fig. 3b). These findings contrasted with the more subtle transcriptional impact of CALR- mutations, since mutated and wildtype cells are co-mingled throughout differentiation20, 39. We previously demonstrated that the whole transcriptomic data alone could not distinguish mutated versus wildtype cells in early clonal processes such as CALR-mutated MPN20 (also confirmed in JAK2-mutated MPN40, 41 and DNMT3A-mutated clonal hematopoiesis42), necessitating the development of GoT. Given the strong transcriptional impact of IFNa, we focused on deciphering the effects of IFNa on the overall HSC population first, before dissecting the differential impact on mutated versus wildtype cells.
To determine how IFNa may impact the hematopoietic differentiation trajectories, we computed the proportion of stem and progenitor subsets within the CD34+ compartment and identified that the IGPs were indeed expanded upon IFNa treatment, consistent with the novel identification of this cell state (Fig. 2b,c). IFNa also induced a significant expansion of the lymphoid progenitors, especially large Pre-B and small Pre-B cells, and diminution of the megakaryocytic-erythroid lineage progenitors, namely MkPs, erythroid progenitors (EP), and megakaryocytic-erythroid progenitors (MEP; Fig. 2b,c; fold change of cell frequencies from serial samples in Fig. 2d). Corroborating a priming toward lymphoid differentiation, HSCs exhibited an increased protein expression of CD38 and CD45RA after IFNa exposure (Fig. 2e). Interestingly, while IFNa alters mouse HSC marker Sca-12 (rendering difficult the quantification of progenitor cell frequency changes by flow cytometric assessment in mice), IFNa did not alter CD90 expression after treatment in the HSPCs (Fig. 2e, Supplementary Fig. 3c).
To validate the generalizability of the lymphoid expansion in the HSPCs, we re-analyzed clinical multi-parametric flow cytometry data of bone marrow aspirates from patients with early-phase MPN with JAK2 or CALR mutations (n = 47 samples with no IFNa exposure; n = 9 samples with IFNa therapy). Indeed, we identified that the proportion of TdT+, CD19+ cells within CD34+ cells increased in IFNa-treated bone marrow compared to non-IFNa exposed samples (Fig. 2f). We also observed an increase in the proportion of CD34+, TdT+, CD19+ cells of total viable cells analyzed, as well as an increase in the CD34-, TdT-, CD19+ B-lymphocytes (Fig. 2f), providing evidence for an active expansion of the lymphoid progenitors, rather than simply a proportional increase due to diminution of the megakaryocytic-erythroid lineage cells. The major shifts in progenitor output suggested that the clinical improvement in the patient’s platelet count post-therapy (despite no significant reduction in the variant allele frequencies) may be due to the differentiation skewing away from the megakaryocytic lineage. Consistent with this hypothesis, the proportions of MEPs and MkPs in CD34+ cells could model the patient’s platelet counts (P = 0.0071, F-test, Fig. 2g), thus supporting a novel model of re-balancing the differentiation landscape as a mode of therapeutic efficacy of IFNa.
As IFNa has been shown to induce HSC and progenitors into cell cycle entry43, we posited that differential induction of progenitor subtypes into proliferation may underlie the shifts in differentiation. Therefore, we examined the gene signatures for proliferation in the context of cell identity shown to be an accurate assessment of proliferation state44. Consistent with previous reports in mice43, the proportions of HSCs expressing G2/M/S-phase genes were enhanced upon IFNa treatment (Fig. 2h,i). Surprisingly, we identified a global increase in proliferation across the progenitor lineages, including granulo-monocytic, erythroid, and lymphoid subsets (Fig. 2h,i). Nonetheless, the proportion of Pro-B cells in G2/M/S-phases demonstrated on average almost a 2- fold increase, compared to more modest increases in other lineages (Fig. 2i), consistent with the observed expansion of the lymphoid progenitors.
IFNa downregulates TNFa and TGFb signaling
To determine the rewiring of other transcriptional programs by IFNa, we performed a differential expression analysis between baseline and treated CD34+ cells, as a function of cell identity. We identified genes commonly regulated across multiple progenitor subsets upon IFNa administration, including the canonical IFNa genes, such as ISG15, IFITM3, IFI6 and EPSTI1 (Fig. 2j, Supplementary Table 5). In HSCs, we identified a downregulation of CXCR4 gene expression (Fig. 2j, Supplementary Table 5), a key regulator of HSC quiescence45, which may thus be involved in cell cycle entry of HSCs. Indeed, HSCs expressing high levels of CXCR4 gene were less likely to be in cell cycle and enriched for baseline cells (P < 2.2 x 10-22, Fisher exact test) and conversely, HSCs in the cell cycle did not express high CXCR4 gene expression (Supplementary Fig. 4a). In the MkPs, we identified a downregulation of CD9 and VWF, closely associated with MkP differentiation46–49, upon IFNa treatment (Supplementary Fig. 4b) which is consistent with the decrease in the proportion of MkP observed after IFNa treatment. We also observed a downregulation of TGFB1 (Supplementary Fig. 4b), that encodes the pro-fibrotic cytokine TGFb established as one of the main inducers of myelofibrosis in patients with fibrotic-phase of MPN50, 51. We and others have shown that MPN megakaryocytes indeed exhibit higher expression of TGFB120, 50, 51, and thus downregulation of TGFB1 by IFNa may be a key mechanism of improvement in fibrosis frequently observed in patients treated with IFNa.
Next, to determine differentially regulated pathways, we performed gene set enrichment analysis using the canonical Hallmark gene sets in the IFNa-treated versus baseline cells and identified upregulation of the IFNa signaling pathway across the cell subsets as expected (Fig. 2k, Supplementary Table 6). Interestingly, most cell types also upregulated genes within the IFNg pathway, but not the IGPs (Fig. 2k, Supplementary Table 6), consistent with a specific upregulation of type 1 interferon-associated IRF1 in the IGPs. The analysis also confirmed the upregulation cell cycle-related pathways (G2M check point and E2F targets) (Fig. 2k, Supplementary Table 6). Interestingly, MYC targets were also upregulated (Fig. 2k, Supplementary Table 6), thus corroborating a previous report demonstrating that IFNa upregulates MYC protein, which was thus postulated to regulate the cell cycle entry of HSCs in mice52. Consistent with downregulation of TGFB1 gene itself, gene set enrichment analysis between baseline and IFNa-treated cells revealed a global decrease in TGFb signaling pathway across multiple HSPC subsets, including downregulation of THBS1 and SERPINE1 (Fig. 2k, Supplementary Table 6). In a subset of the progenitor states, P53 and KRAS signaling were downregulated (Supplementary Table 6).
Moreover, in contrast to the pro-inflammatory state of the IFNa-associated IGPs, we observed a downregulation of pro-inflammatory pathways, including the TNFa signaling via NFKB and inflammatory response pathways, across most of the progenitor subtypes but particularly in the cDCPs (Fig. 2k-l, Supplementary Table 6). Downregulated genes in the TNFa pathway included NFKB1, VEGFA, MAP3K8, as well as IL1B, that encodes the pro-inflammatory cytokine IL-1b (Fig. 2l, Supplementary Table 6). Expression of IL1R1, TNFRSF1B and CXCL8 genes from the inflammatory response pathway were also downregulated (Supplementary Table 6). Previously, induction of IFNa in mice have yielded mixed results demonstrating either an upregulation2 or downregulation53 of TNFa. Our findings show that in human hematopoiesis, IFNa exerts an anti- inflammatory response associated with a downregulation of TNFa signaling and therefore may serve as another key mode of disease amelioration via downregulating MPN-associated inflammation. However, given the induction of the IGP differentiation, these findings suggest that IFNa is able to induce both a pro-inflammatory state via IGP differentiation and an anti- inflammatory state via differentiation shifts toward lymphoid development and downregulation of TNFa and NFKB activities – both occurring in parallel within the same individual’s hematopoiesis. In fact, we observed a significant overlap between the IFNa-downregulated genes in the TNFa signaling and the genes that are upregulated in the IGPs (Fig. 2l, Supplementary Table 6, P = 1.6 x 10-5, hypergeometric test). The inverse, i.e. the overlap of IFNa-upregulated genes and genes downregulated in IGPs, was also observed (P = 1.73 x 10-69, hypergeometric test, Supplementary Tables 3,6). Consistently, the HSC-specific IFNa-upregulated gene signature (data presented in Fig. 2j) was significantly downregulated in the IGPs compared to the HSCs while the IFNa- downregulated genes were upregulated in the IGPs, with both cell groups under the same IFNa treatment (Fig. 2m). Altogether these data indicated that IFNa can trigger opposing cell states within the same hematopoiesis.
GoT-IM reveals differential impact of IFNa on CALR-mutated versus wildtype hematopoiesis
Having determined the overall effects of IFNa on the CALR-mutated and wildtype cells in aggregate, we leveraged the genotyping capacity of GoT-IM to determine how IFNa may differentially impact neoplastic versus wildtype HSPCs. Consistent with our previous results20, the mutated and wildtype HSPCs were co-mingled throughout differentiation both at baseline and upon IFNa therapy (Fig. 3a, Supplementary Fig. 5a). To assess how IFNa may skew differentiation distinctly in the CALR-mutated versus wildtype HSPCs, we determined the progenitor subset frequencies between mutant and wildtype cells. As we had previously demonstrated, the mutated cells were enriched in the MkPs at baseline (Fig. 3b,c). After treatment, we observed the expansion of the lymphoid compartment in both wildtype and mutated cells (Fig. 3b,d). However, the lymphoid expansion was constrained in the mutated compartment, due to the relative expansion of the granulo-monocytic and megakaryocytic-erythroid progenitors compared to wildtype HSPCs (Fig. 3d). These findings provide direct evidence that IFNa exerts distinct phenotypic effects on HSPCs as a function of the mutation status in human.
a. UMAP of CD34+ cells (n = 49891cells) with mutation status highlighted for wildtype (WT; n = 17461), CALR-mutant (MUT; n = 23207) or unassigned (NA; n = 9223) cells. b. UMAP showing densities of mutated vs. wildtype cells at baseline and after IFNa treatment. c. Normalized cell frequencies of mutated (n = 15258) vs. wildtype (n = 10838) cells at baseline (n = 9 samples; cells down-sampled to 500 for each sample). d. Normalized cell frequencies of mutated (n = 7949) vs. wildtype (n = 6623) cells at IFNa-treatment (n = 5 samples). Left panel: Cells were down-sampled to 500 for each sample in the aggregate analysis. Right panel: Cell frequencies of wildtype vs. mutated cells by progenitor group. e. Box plot showing frequencies of cells in G2/M/S phase as assessed in Fig. 2i (n = 8 serial samples from 4 patients). P-values from linear mixed model with/without mutation status. f. Multiplex in situ fluorescence imaging (Vectra Polaris) of bone marrow biopsy sections from MPN patients treated (n = 4) or not (n = 5) with IFNa. Left panel: Representative images. Right panel: Bar plots showing frequencies of proliferating myeloid cells as assessed by Ki67 expression. P-values from linear mixed model with/without treatment status or mutation status g. Left panel: Schematic of clonal structure of MPN from patient IFN04. Right panel: Bar plot of normalized mutant cell frequencies across treatment time points. h. Volcano plot showing DE genes between single mutant and double mutant HSCs at baseline. DE genes identified using linear mixed modeling with/without treatment status. Genes highlighted in blue- green are enriched in the TNFa signaling via NFKB, those in red are enriched in the unfolded protein response and those in orange in the IFN-gamma response. Boxes represent TF of the AP- 1 (blue) and NR4A (purple) families. Enrichment identified by pre-ranked gene set enrichment analysis using the MSigDB Hallmark collection. i. Heatmap showing results of the pre-rank gene set enrichment analysis of genes DE between HSC clones at baseline. Values correspond to the sign of the normalized enrichment score (NES) times the -log10(Adjusted P-value). j. Box plot showing HSC-specific IFNa-induced signature score in HSC clones at baseline and after treatment. Score calculated using upregulated or downregulated genes (left and right panels, respectively).
Given that a primary effect of IFNa involves inducing robust cell cycle programs in HSPCs1, we hypothesized that the differential increase in proliferation in mutated versus wildtype hematopoietic progenitor cells may contribute to the differential expansion of the progenitor subsets in CALR-mutated versus wildtype cells. At baseline, the CALR-mutation induced proliferation in the HSCs and the myeloid cell lineages (i.e. the granulo-monocytic and megakaryocytic-erythroid lineages) but not in the lymphoid progenitors (Fig. 3e), consistent with our previous reporting20. Intriguingly, IFNa boosted the proliferation of mutated HSCs to a greater degree than the wildtype counterparts (Fig. 3e), with a corresponding decrease in CXCR4 gene expression that was more pronounced in the mutated cells (Supplementary Fig. 5b). Moreover, IFNa induced proliferation to a greater degree in the mutated cells compared to wildtype cells within the myeloid progenitor compartment, but not the lymphoid, that is, it was restricted to the progenitor subsets in which the CALR-mutation induced proliferation at baseline (Fig. 3e). Thus, the proliferative effects of IFNa appeared to be compounded by those of the underlying CALR mutation; in other words, the mutated cells were primed toward a greater proliferative response upon IFNa treatment. However, in contrast to the other myeloid progenitors, within the MkPs (wherein the CALR mutation exerts the greatest phenotypic effects due to the activation of the thrombopoietin receptor21, 54–57), the proliferative effects of IFNa were modest in the mutated cells compared to those at baseline (Fig. 3e), suggesting that the rate of cell cycle entry of the mutated cells was already near or at the point of saturation due to the CALR mutation. We orthogonally validated that IFNa induced greater proliferative rates in the myeloid progenitors by performing multiplexed in situ fluorescent imaging of bone marrow paraffin-embedded sections from patients with CALR mutations with or without IFNa treatment (n = 5 with no IFNa, n = 4 with IFNa, Fig. 3f, left). We determined the protein expression of CD34, CD117, CD38, Ki67 and mutant CALR to assess the frequencies of CD34+, CD38+, CD117+ myeloid progenitors that express Ki67, a gold standard of cell cycle entry58. Overall, agnostic to the mutation status, IFNa-exposed myeloid progenitors demonstrated a trending increase of proliferation even in this non-paired cohort (Supplementary Fig. 5c, P = 0.063, Wilcoxon rank sum). The availability of a CALR-mutant specific antibody59 enabled us to distinguish mutated from wildtype cells. We found that the frequency of myeloid cells in cell cycle was increased in the treated mutated versus wildtype progenitors (Fig. 3f, right). Overall, these findings provide evidence that the differential proliferation states as a function of the mutational status may contribute to the relative expansion of the myeloid versus lymphoid progenitors.
Impact of IFNa on clonal fitness of CALR-mutated HSC
As IFNa induces cell cycle entry of mutated HSCs to a greater extent than wildtype HSCs, it may be postulated that IFNa preferentially exhausts mutated HSCs for clonal remission, as was hypothesized to be the case for stem cells with JAK2 mutations60 and BCR-ABL translocations1. However, while JAK2-mutated MPN frequently displays clonal stem cell depletion upon IFNa treatment, CALR-mutated MPN rarely demonstrates molecular response19, indicating a genotype- specific response to inflammation. Consistent with these data, when we examined the clone size in the HSC compartment in the serial samples, we observed no significant shifts in the relative proportion, except in one patient IFN04 whose clone expanded significantly upon IFNa therapy (Supplementary Fig. 6a). Interestingly, we found that cells from IFN04 harbored another mutation in CALR: a single nucleotide variant in the CALR allele (M131I, predicted to impact protein structure61) which is trans to the MPN-causing frameshift mutation (Supplementary Fig. 6b). At baseline, the double mutant clone remained subclonal to the dominant clone with the single canonical CALR mutation, consistent with bulk sequencing data (Fig. 3g). Upon IFNa therapy, the double-mutated clone overtook the neoplasm and the overall stem cell population (Fig. 3g). We have previously demonstrated that CALR mutations enhance the unfolded protein response (UPR), specifically the pro-survival IRE1-XBP1 axis20, likely due to a heterozygous compromise of the critical chaperone protein encoded by CALR. Consistent with these data, we found that the additional insult to CALR activities in the double mutant clone induced an even greater UPR activation compared to single mutant HSCs (Fig. 3h,i, Supplementary Tables 7,8). Surprisingly, however, the predominant signatures of the double mutant HSCs were decreased TNFa and TGFb signaling and upregulation of IFN response genes which is the observed IFNa signature in single mutant HSCs compared to wildtype HSCs at baseline (Fig. 3h,i, Supplementary Tables 7,8). Intriguingly, the double mutant HSCs intrinsically exhibited high IFN signaling at baseline (Fig. 3h,i, Supplementary Tables 7,8), supporting a potential causal link between UPR and IFN activation62. Indeed, the double mutant HSCs exhibited high expression of the HSC-specific IFNa- upregulated gene signature (derived from data displayed in Fig. 2j) at baseline and even higher upon IFNa therapy compared to baseline (Fig. 3j, left). Similarly, the HSC-specific IFNa- downregulated gene signature was downregulated in the double mutant clone both at baseline and to a greater degree at treatment (Fig. 3j, right). These data further supported the concept that the degree of cellular response to IFNa varies upon the degree of priming toward that signal at baseline. Consistent with these data, unbiased clustering and dimensional reduction revealed that the double mutant clone at baseline clusters with the treated HSCs rather than with the other HSCs at baseline (Supplementary Fig. 6c). The striking similarity of the intrinsically-activated IFNa signaling in the double mutant cells at baseline to the extrinsic IFNa effects indicated that the predominant IFNa signaling signatures observed in IFNa-treated HSCs are largely direct rather than simply mediated by secondarily modulated cytokines (e.g. TNFa, TGFb). These findings also highlighted a genotype-specific modulation of HSC fitness by IFNa, specifically suggesting that IFNa selects for HSC clones with high IFNa signaling at baseline.
GoT-ATAC identifies transcription factor regulatory networks that govern IGP differentiation
Chromatin accessibility enables approximation of TF activity based on accessibility of the TF binding sites. Thus, in order to determine the regulatory networks that govern the IFNa-induced differentiation states, we expanded upon GoT to capture somatic mutation status, chromatin accessibility and whole transcriptomes by adapting the 10x Multiome platform that captures single-nuclei RNA-seq (snRNA-seq) and chromatin accessibility (snATAC-seq) to include somatic genotyping, i.e. GoT-ATAC (Fig. 4a). We applied GoT-ATAC to serial bone marrow CD34+ cells (n = 24,862 total; 12% of total cells with genotyping data (n=2,873 cells)) from the same cohort of IFNa-treated patient samples (n = 4 baseline, 3 IFNa-treated samples). Notably, genotyping efficiency was significantly lower in GoT-ATAC compared to GoT-IM, due to the limited transcript pool in the nucleus versus whole cell. As in GoT-IM, we incorporated time-point specifying barcoded antibodies24 to combine serial samples from the same individuals into a single experiment to remove technical batch effects (Fig. 4a, Supplementary Fig. 7a-d). We then identified the treatment status analytically and segregated the baseline and IFNa-treated samples to cluster the cells based on cell identity alone as we did for the GoT-IM data (Supplementary Fig. 7e). We clustered the cells based on the transcriptomic data and identified the cell states identified by GoT-IM, including the novel IGP cell state (Fig. 4b, Supplementary Fig. 8a,b, Supplementary Table 9). The activities of cell lineage specifying TF were consistent with the cell type assignments (Supplementary Fig. 8c-d). Furthermore, we integrated the cells based the snATAC-seq data, and identified the same stem and progenitor cell relationships, including the similarity of the IGPs to the quiescent HSCs, again providing evidence for a direct differentiation of IGPs from HSCs (Fig. 4c, Supplementary Fig. 9a).
a. Representation of primary bone marrow samples at baseline and after IFNa treatment. Schematic of GoT-ATAC. b. UMAP of CD34+ cells based on snRNA-seq data from MPN samples (n = 24,857 cells, 7 samples from 4 individuals), overlaid with cell type assignment. c. UMAP of CD34+ cells based on snATAC-seq data overlaid with cell type assignment n = 23,137 cells, 7 samples from 4 individuals). d. Motif enrichment and expression of Transcription Factors (TF) in the IGPs relative to HSCs from IFNa-treated samples. Gene expression derived from GoT-IM data. e. Chromatin accessibility tracks of regulatory regions of RFX3. f. Ranked TF motif enrichment of positively- regulating loci of the AP-1 members, relative to background peaks. g. Chromatin accessibility tracks of regulatory regions of HLA-DRA1 (bottom), distal region enriched with negatively- regulating loci (inset), and IGP-specific regulatory locus (top).
As the IGPs were characterized by a robust upregulation of TF gene expression, especially those of AP-1 and RFX2/3, we sought to determine whether the chromatin accessibility of their binding motifs was increased, as a reliable surrogate of TF activities. Strikingly, differential TF motif enrichment analysis between the IGPs and HSCs identified enhanced accessibility of the same TFs whose expression levels were elevated, including the motifs of the AP-1 family (FOS, JUN, JUND, JUNB, FOSL1, FOSL2), CEBPB/D, IRF1 and RFX2/3 (Fig. 4d, Supplementary Table 10). These data thus demonstrated an unusually high concordance of gene expression of the TFs and their target motif accessibility, suggesting a rapid induction of the transcriptional regulatory program. Enhanced chromatin accessibilities for small MAF factors (MAFK and MAFG) and their interacting partners NFE2L1/3 and BACH1/2 were also observed in the IGPs (Fig. 4d, Supplementary Table 10). The differential motif analysis further revealed an upregulation of STAT2 and the proinflammatory REL of the NFKB complex (Fig. 4d, Supplementary Table 10), consistent with the observed upregulation of the TNFa via NFKB signaling pathway (Fig. 1h). As RFX2/3 is not well characterized in hematopoietic stem and progenitor cells, we determined which TFs upregulated the expression of these TFs. We determined TF motifs present in the regulatory peaks that correlated with RFX3 gene expression (i.e. linked peaks analysis)63 and identified those of STAT2 and IRF1 (further supporting an IRF1-specific response) as well as FOSB and SPI1 (or PU.1), which were differentially upregulated at the gene expression level in IGPs (Fig. 4e, Supplementary Table 11). Similarly, the most significant regulatory peaks for RFX2 included motifs for PU.1, KLF, NR4A1/2 and AP-1 factors that were differentially expressed in the IGPs (Supplementary Fig. 9b, Supplementary Table 11). These data further supported the model in which HSCs with high expression of the AP-1 and other immediate early response factors were primed toward a robust upregulation of RFX2/3 gene expression upon induction of IRF1 and PU.1 activities with IFNa treatment. Further, to determine the regulatory networks that govern the robust upregulation of the AP-1 family TFs in the IGPs, we aggregated the significantly correlated peaks for the AP-1 TFs and assessed for motif enrichment. Among the top TF motifs were those of RFX2-4 (which have the same binding motifs), implicating RFX2/3 in the upregulation of the AP-1 members upon IGP differentiation (Fig. 4f, Supplementary Table 12). Thus, RFX2/3 and AP-1 TFs positively regulated the expression of one another, synergizing the IGP development. Finally, as other RFX members have been demonstrated to regulate MHC class II expression35, 64, we hypothesized that RFX2/3 may play a significant role in downregulating MCH class II genes in the IGPs. To test this, we determined the significantly linked peaks that negatively regulated HLA-DRA expression (Fig. 4g, bottom). We identified a distal regulatory region with four negatively regulating loci (Fig. 4g, inset, Supplementary Table 13). These peaks included motifs for IRF1 and STAT2 as well as immediate response factors including AP-1 and KLF families, but not RFX2/3 (Fig. 4g, inset, Supplementary Table 13). However, within the same negative regulatory region, we identified an IGP-specific peak that included the binding motif for RFX1-4, KLF factors and IRF1 (Fig. 4g, top, Supplementary Table 13). These findings identified the immediate response factors and IRF1/STAT2 as negative regulators of MHC class II genes across cell types, while RFX2/3 was highly specific for MHC class II downregulation during IGP differentiation. Notably, the motif RFX8, known to positively regulate MHC class II expression was present in the positively regulating peaks, but not in the negatively regulating loci (Supplementary Table 13). Genes positively regulated by RFX2/3 included MPO, consistent with neutrophilic differentiation, and genes involved in cell cycle entry, implicating RFX2/3 as key TFs in the overall regulatory network of IGPs (Supplementary Table 14).
PU.1 initiates IFNa-induced hematopoietic differentiation remodeling
Furthermore, the GoT-ATAC data confirmed the expansion of the lymphoid progenitors upon IFNa treatment (Supplementary Fig. 10). Thus, in order to determine the regulatory networks that governed the transcriptional rewiring and differentiation skewing induced by IFNa, we performed differential motif enrichment analysis between IFNa-treated and baseline HSCs. IFNa enhanced the activities of STAT2 and numerous interferon regulatory factors in HSCs, in contrast to the specific activity of IRF1 in IGPs (Fig. 5a, Supplementary Table 15). IFNa downregulated the motif accessibility of AP-1 TFs consistent with the downregulation of AP-1 TF gene expression associated with TNFa signaling via NFKB (Fig. 2j). Moreover, TWIST1 has been implicated in mediating the downregulation of TNFa upon type 1 IFN treatment53. Consistently, accessibilities of the TWIST1 motif were enhanced after treatment. Furthermore, the activity of TGFb induced factor homeobox 2 (TGIF2), which inhibits TGFb response genes, was also enhanced (Fig. 5a, Supplementary Table 15), highlighting TGIF2 as a key TF involved in the downregulation of the observed TGFb signaling after IFNa treatment. IFNa also downregulated the activities of cAMP responsive element binding (CREB) proteins (Fig. 5a, Supplementary Table 15), consistent with the observed downregulation of KRAS signaling (Fig. 2k).
a. Volcano plot showing TF motif differentially enriched in IFNa-treated versus baseline HSCs (ChromVar). P-values from Wilcoxon rank sum test with Benjamini-Hochberg FDR-correction. b. Heatmap showing synergy scores between TFs as assessed by measuring the excess variability of accessibility for peaks with both TF motifs (compared to peaks with one motif)83 c. TF motif accessibility across stem and progenitor subsets in lymphoid development between baseline and IFNa-treated cells. d. TF motif enrichment of CEBPA (left) and GATA1 (right) in mutated versus wildtype cells at baseline. e. Model of IFNa-induced remodeling of hematopoiesis. At baseline, MPN stem cells show a skewing toward the myeloid lineages. IFNa induces PU.1 activity that regulates downstream differentiation activities. In a minority population of HSCs with high RFX2/3 and AP-1 activities at baseline, PU.1 and IRF1 induce their differentiation into the IGPs, associated with high TNFa signaling. PU.1 also mediates skewing of hematopoiesis toward the lymphoid lineage via co-activity with TCF3 and toward the granulo-monocytic lineage, especially in the mutated cells, via CEBPA. Downregulation of TNF-a and TGFb signaling are also observed in the HSCs upon IFNa treatment.
Importantly, critical TFs involved in hematopoietic differentiation65 were differentially regulated. The motif accessibility of PU.1 and RUNX1, essential for early lymphoid and granulomonocytic differentiation65, was enhanced (Fig. 5a, Supplementary Table 15). Consistently, the activities of GATA1/2, demonstrated to be negatively regulated by PU.165, were downregulated as well as those of the other critical megakaryocytic-erythroid lineage factor, TAL166, 67, consistent with the differentiation away from the megakaryocytic-erythroid lineages upon IFNa treatment. Furthermore, the accessibilities of the motifs of the critical early B-lymphoid differentiation factors TCF3/4 were enhanced, highlighting TCF3/4 as TFs that govern the differentiation toward lymphoid progenitors by IFNa. CEBPA, essential for granulo-monocytic development68, 69, was also upregulated in its motif accessibility. Importantly, the IRFs, RUNX1, CEBPA and TCF3 have been demonstrated to co-regulate target gene expression with PU.170, suggesting that the induction of IRFs may have stabilized PU.1 activities, which in turn enhanced the activities of its co- regulating TFs, such as CEBPA and TCF3. To confirm these co-regulations in IFNa-induced hematopoiesis, we examined the synergistic activities of the different combinatorial TFs by measuring the excess variability of accessibility for peaks with both TF motifs (compared to peaks with one motif)71. Indeed, PU.1 exhibited high synergistic activities with the IRFs, RUNX1, CEBPA and TCF3, while displaying antagonism with GATA1 and TAL1 (Fig. 5b). These findings again highlighted PU.1 as the master regulator of IFNa-induced differentiation shifts, as was the case in IGP differentiation. To determine the activities of these TFs upon lymphoid differentiation, we assessed the motif accessibilities of the TFs across the early and late progenitors. The activities of PU.1, RUNX1 and CEBPA were enhanced upon IFNa signaling during early stages of hematopoiesis and diminished with lymphoid development, whereas the activities of TCF3 were pronounced in the latter stages of development (Fig. 5c).
To identify the TFs that may govern the observed myeloid skewing of the mutated versus wildtype hematopoietic development, we tested the differential motif enrichment of the IFNa-regulated lineage TFs (Fig. 5a) in mutated vs. wildtype HSCs and IMPs. We identified an upregulation of motif enrichment for CEBPA at baseline (Fig. 5d), suggesting that PU.1 activation upon IFNa, together with increased CEBPA activity, enhanced the granulo-monocytic differentiation in the mutated cells. Moreover, GATA1 activity was upregulated in the mutated MkPs and MEPs (Fig. 5d), highlighting GATA1 as the key TF that governed the preferential megakaryocytic-erythroid development.
Altogether, GoT-ATAC identified the key regulators that governed the IFNa-induced hematopoietic differentiation shifts, which varied as a function of the underlying mutational status. PU.1 was identified the master regulator that mediated the hematopoietic differentiation remodeling by IFNa in both the induction of the IGPs and lymphoid expansion (Fig. 5e). Thus, IFNa-induced PU.1 activities mediated both a pro-inflammatory state via IGP development and an anti-inflammatory state via lymphoid differentiation skewing, linked with opposing AP-1 activities, suggesting a strong dependency on the underlying HSC transcriptional and epigenetic state (Fig. 5e).
Discussion
Inflammation perturbs HSC homeostasis and differentiation across physiologic and pathologic conditions3, 14, 72–74. In mice, the contributions of the various inflammatory cytokines have been clarified by the administration of individual cytokines to mouse models4, 5, 13, 75–78. While inflammation also impacts human hematopoiesis across conditions such as ageing, cancer, autoimmune disorders and infections3, 72, 73, 79, a careful dissection of the contribution of individual cytokines in humans is challenging. Germline genetic disorders have provided clues into the distinct roles of individual cytokines. For example, germline defects in genes involved in the type 1 IFN signaling cascade have been linked to exaggerated or severe COVID-19 response16, 79, providing evidence for anti-inflammatory effects of type 1 IFN. Indeed, early studies of administering type 1 IFN to patients have suggested clinical benefit17, 80.
To directly examine the effects of type 1 IFN in human hematopoiesis, we identified a unique clinical setting in which patients with MPN receive recombinant IFNa as therapy. As patients with MPN often present with an early clonal and stable disease, serial bone marrow sampling enriched with stem and progenitor cells is rare. To overcome this limitation, we leveraged cryopreserved bone marrow samples at baseline and after IFNa treatment from our clinical trials. We applied our innovative single-cell multi-omics platforms to FACS-isolated CD34+ stem and progenitor cells to determine how IFNa impacts human hematopoiesis in normal and neoplastic blood development. We identified an HSC-specific IFNa signature that was defined by an upregulation of IFNa/g response genes (as expected) and downregulation of the TNFa and TGFb signaling pathways. In mouse studies, IFNa has been demonstrated to modulate TNFa expression13. However, different studies have yielded distinct results, either an upregulation2 or downregulation53 of TNFa signaling. Thus, our studies clarify that the predominant effect of IFNa in human HSPCs is the downregulation of TNFa and other proinflammatory pathways, consistent with the abovementioned genetic studies in patients with severe COVID-19 infections16, 79 and amelioration of the inflammatory state in MPN and multiple sclerosis upon type 1 interferon administration17, 19. Coherently, TNFa protein expression was demonstrated to be decreased in patients with MPN who received IFNa treatment81. Our single-cell multi-omics data revealed that the downregulation of AP-1 and NFKB transcription factors played a key role in downregulating the pro-inflammatory signaling. The pro-fibrotic TGFb signaling was also broadly downregulated across progenitor subsets, associated with the downregulation of the TGFB1 gene itself and upregulation of the TGFb signal-inhibiting TGIF2 activity. These findings were consistent with the observed downregulation of TGFb receptor (TGFBR1) in mice with IFNa administration2. In this way, the coordinated activities of IFNa downregulated two key cellular programs involved in the MPN-associated pathology and thus improved patient disease state.
Unexpectedly, we identified a novel progenitor state, we termed inflammatory granulocytic progenitors (IGP), that was induced by IFNa, specifically through IRF1. Immunophenotypically, IGPs were similar to the IMPs. Transcriptionally and epigenetically, they were most similar to a subset of the HSCs with the highest expression of CD90 and AVP (indicating a quiescent state) and with high gene expression and enhanced motif accessibility for the AP-1 members and RFX2/3 TFs. Upon IGP development, the AP-1 and RFX2/3 TFs and their activities were robustly upregulated, as key defining features of this novel progenitor state. The upregulation of PU.1, CEBPB and CEBPD expression and motif accessibility (as well as of their targets MPO and CSF3R), combined with a downregulation of the MHC class II genes indicated their lineage fate into the neutrophil lineage. The specificity of the IGP gene signature to neutrophils versus other myeloid lineages also supported this lineage fate. However, they were transcriptionally and epigenetically distinct from the IMPs and neutrophil progenitors, suggesting that they may arise directly from the quiescent HSCs, as a novel route of neutrophil formation. The functional and genetic lineage studies of this rare population are challenging, particularly since the in vitro effects of IFNa on HSCs completely differ from its in vivo effects. In vitro, IFNa induces cell death in the HSCs2 while it induces proliferation in vivo1. These findings suggest that IFNa effects on HSCs are dependent on the bone marrow microenvironment. Consistent with this inference, we identified a downregulation of CXCR4 linked with their changes in proliferation status upon IFNa treatment. Moreover, as scRNA-seq of HSPCs from mice treated with the IFNa-inducing polyI:C77 or infected with Mycobacterium tuberculosis82 (a model of IFNa signaling) did not reveal a comparable population, there is a possibility that the corresponding IGP population may not be present in mice.
Nevertheless, the identification of this novel IGP population highlighted an intriguing phenomenon in human HSCs. Induction of the IGP differentiation through upregulation of the pro- inflammatory AP-1 and REL, together with PU.1, indicated that IFNa is able to initiate this pro- inflammatory and overall anti-inflammatory cell states. Intriguingly, the IFNa-induced HSC transcriptional signature and the IGP differentiation program utilized an overlapping transcriptional program, but in the opposite direction, i.e. the IFNa-induced upregulated genes in HSCs were downregulated in the IGPs and vice versa. The AP-1 program was highlighted in our data as modulating this switch in response to IFNa. These findings overall highlighted non-genetic cell state heterogeneity of the HSC pool as a key mediator of distinct, even opposing, cellular responses to IFNa therapy. IFNa thus exposed the functional consequences of HSC heterogeneity. Another major discovery in our work was the differentiation remodeling of hematopoietic differentiation toward the lymphoid lineage by IFNa. While various differentiation skewing by IFNa has been reported in mice3, 60, 82, the interpretation of the results is complicated by the IFNa- induced alteration of the HSC-marker Sca-1 in mice (i.e. induction of Sca-1 in the CMPs, GMPs and MEPs, resulting in their inclusion within the Sca-1+, Kit+ HSC/multipotent progenitor compartment). As other inflammatory cytokines, such as TNFa, IL-1, and IFNg, have been demonstrated to induce granulo-monocytic differentiation6, 8, 9, 11, 76, IFNa presents as a unique cytokine among the inflammatory milieu to balance the granulo-monocytic differentiation with its activities on lymphoid differentiation as another mode of dampening the pro-inflammatory response.
The reshaping of the differentiation landscape by IFNa provided a novel model of therapeutic efficacy in patients with MPN. As MPN presents primarily as a defect in homeostatic hematopoietic development, due to an abnormal differentiation skewing and increased proliferation of one or more of the myeloid lineages, IFNa-induced lymphoid differentiation and proliferation may re-balance the differentiation landscape. Just as in the differentiation of the IGPs, PU.1 – the master regulator of granulo-monocytic and B-lymphoid lineages – emerged in our data as the initiating regulator of IFNa-induced differentiation remodeling. Intriguingly, IFNa reshapes the major bifurcation divide in MPN, that is, from the JAK2/STAT5-mediated bifurcation at the myeloid (i.e. granulo-monocytic and megakaryocytic-erythroid) versus lymphoid commitment, to the PU.1-mediated bifurcation at granulo-monocytic and lymphoid versus megakaryocytic- erythroid commitments (see model in Fig. 5e). Furthermore, the upregulation of PU.1 and its cooperating TFs, such as CEBPA and TCF3, indicated that the IFNa-induced shifts in the progenitor populations were due to an active differentiation re-programming, rather than due to a preferential cell death of a particular lineage.
Our single-cell multi-omics methods uniquely enabled us to decipher the distinct effects of IFNa on CALR-mutated versus wildtype HSPCs. Our data revealed that while lymphoid expansion was observed in both mutated and wildtype cells, the lymphoid differentiation was constrained by a pre-existing CALR-induced bias toward the myeloid lineages. The CALR mutation-induced proliferation of the myeloid lineages at baseline primed the mutated cells for an amplified response in the IFNa-induced proliferation. Further, our GoT-ATAC data revealed that the CALR-mutated HSPCs were primed toward the myeloid lineages via enhanced CEBPA and GATA1 activities at baseline. As PU.1 cooperates with CEBPA to induce the granulo-monocytic lineage and antagonizes GATA1 activities (thereby reducing megakaryocytic-erythroid lineage specifying programs), the upregulation of CEBPA and GATA1 activities in the mutated cells indicates a modulation of the PU.1 activities upon IFNa therapy and downstream phenotypic effects toward a greater myeloid expansion in the mutated versus wildtype cells. Altogether, these findings provided direct evidence in human that the extrinsic signaling by cytokines are constrained by the transcriptional rewiring from the underlying somatic mutation.
Evidence for genotype-specific cellular priming and response was further observed in serial sampling from an individual with a ‘double-hit’ in the CALR genes. While homozygous canonical mutations in CALR are quite rare and only observed in aggressive myeloid neoplasms with multiple genetic defects21, the presence of the subclone with an additional single nucleotide variant in the trans CALR allele provided further insights into IFNa signaling in HSCs. Surprisingly, the double mutant clone exhibited a strong HSC-specific IFNa signature compared to the wildtype and single mutant clone at baseline and were transcriptionally more similar to the IFNa-treated cells compared to baseline cells (in stark contrast to the other samples in which cells clustered based on treatment status). Three important inferences could be derived from this index case. First, the double mutant cells exhibited a greater IFNa-signature upon IFNa treatment compared to the treated wildtype and single-mutant HSCs, further supporting the idea that the degree of HSC- response to IFNa treatment is commensurate to the priming toward IFNa-induced signaling prior to exposure (e.g. proliferation, differentiation and IFNa-specific gene expression). Second, the double mutant clone exhibited a strong fitness advantage against non-neoplastic and single-mutant HSCs upon IFNa, by taking over both the neoplastic and the overall HSC population, suggesting that baseline type 1 IFN activity may enhance HSC fitness upon the IFNa challenge. Finally, in the double mutant cells, we observed not only the upregulation of the IFNa-positively induced genes, but also the reduction of the downregulated genes in the TNFa and TGFb pathways. These data demonstrated that the predominant effects of IFNa on HSCs is direct (and not secondary to other mediators). While the overall reduction of TNFa and TGFb production in the other bone marrow cells has likely reinforced the downregulation of these pathways in the HSCs, the strong downregulation of TNFa and TGFb pathways in the mutation (i.e. intrinsically)-induced IFN signaling demonstrates that the predominant effects are direct. These data are supported by an elegant experiment in mice1. In a chimeric mouse model consisting of mostly IFNa/b receptor (IFNAR) deficient cells, the minority of HSCs with intact IFNAR demonstrated a robust cell cycle entry upon IFNa administration, establishing that IFNa exerts profound effects directly on HSCs1. Overall, we identified a human HSC signature of IFNa response characterized by a re-modeling of the differentiation landscape and an intriguing phenomenon of opposing transcriptional programs triggered by the same extrinsic perturbation, as a function of the underlying HSC states. These data thus highlighted mediators of functionally significant stem cell heterogeneity that may be manipulated for therapeutic advantage.
Competing interests
D.A.L. has served as a consultant for Abbvie and Illumina, and is on the Scientific Advisory Board of Mission Bio and C2i Genomics; D.A.L. has received prior research funding from BMS and Illumina unrelated to the current manuscript.
Corresponding author
Anna S. Nam.
Correspondence and requests for materials should be addressed to A.S.N.
Supplementary Information
Supplementary Table 1. Summary of patients’ clinical history and mutation status for samples IFN01-IFN04 and IFN06-IFN08.
Supplementary Table 2 (related to Fig. 1c). Table of number of cells identified for each cell type in GoT-IM.
Supplementary Table 3 (related to Fig. 1e,h). Differential gene expression analysis between IGP versus IMP and IGP versus HSC1 cell type was performed via a linear mixed model.
Supplementary Table 4 (related to Fig. 1e,h). Gene set enrichment analysis of genes differentially expressed between IGPs versus IMPs, IGPs versus HSC1was performed against MSigDB Hallmark gene sets.
Supplementary Table 5 (related to Fig. 2j). Differential gene expression analysis between baseline and IFNa-treated HSPC subtypes was performed via a linear mixed model.
Supplementary Table 6 (related to Fig. 2k,l). Gene set enrichment analysis of genes differentially expressed between baseline and IFNa-treated HSPC subtypes was performed against MSigDB Hallmark gene sets.
Supplementary Table 7 (related to Fig. 3h). Differential gene expression analysis between single mutant and double mutant HSCs from IFN04 at baseline was performed via logistic regression framework.
Supplementary Table 8 (related to Fig. 3i). Gene set enrichment analysis of genes differentially expressed between HSC clones from IFN04 was performed against MSigDB Hallmark gene sets.
Supplementary Table 9 (related to Fig. 4b). Table of number of cells identified for each cell type in each sample using GoT-ATAC.
Supplementary Table 10 (related to Fig. 4d). Differential transcription factor motif enrichment analysis between IGPs and HSCs after IFNa treatment was performed using Wilcoxon rank sum test.
Supplementary Table 11 (related to Fig 4e). Motifs identified in peaks linked with RFX2/3 via motif scanning with FIMO (v5.4.1).
Supplementary Table 12 (related to Fig 4f). Overenrichment motif analysis for peaks linked with AP-1 genes using FindMotif function (Signac, v1.7.0).
Supplementary Table 13 (related to Fig 4g). Motifs identified in peaks linked with MHC class II genes via motif scanning with FIMO (v5.4.1).
Supplementary Table 14: Table with genes that are identified as being positively regulated by RFX2 and RFX3 using Gene-Peak cis-association .
Supplementary Table 15 (related to Fig. 5a). Differential transcription factor motif enrichment analysis between baseline and IFNa-treated HSCs was performed using Wilcoxon rank sum test.
Supplementary Table 16: List of antibodies used for CITE-seq.
Supplementary Table 17: Primer sequences used in GoT-IM and GoT-ATAC.
Methods
Patient samples
The study was approved by the local ethics committee and by the Institutional Review Board (IRB) of Weill Cornell Medicine. The study was conducted in accordance with the Declaration of Helsinki protocol, and all patients provided informed consent. Cryopreserved bone marrow mononuclear cells or peripheral blood mononuclear cells were obtained from patients with CALR- mutated essential thrombocythemia (ET) treated with weekly pegylated IFN-alfa2a during clinical trials MPN-RC-111 (NCT01259817)22 and MPN-RC-112 (NCT01258856)23 (n = 4 baseline and 7 treated samples collected from 5 individuals, see Supplementary Table 1 for clinical and sample information). Baseline ET samples from our previous work20 were included as additional baseline controls.
Cell preparation
Cryopreserved bone marrow mononuclear cells were thawed and stained using standard procedures with the surface antibody CD34-PE (clone AC136, dilution 1:50, Miltenyi Biotec) and DAPI (Sigma-Aldrich), according to manufacturer’s protocol. To eliminate experimental batch effects, cells were labelled simultaneously with hashing antibodies with timepoint-identifying barcodes as described15 using Hashtag Antibodies 1, 2 or 3 (TotalSeq-A, clone LNH-94, Biolegend) for GoT-IM. To link cell identities to expression of cell surface proteins, cells were also incubated with CITE-seq antibodies17 according to manufacturer protocol (TotalSeq-A, Biolegend, see Supplementary Table 16 for information on antibodies). Cells were subsequently sorted for DAPI−, CD34+ cells using BD Influx at the Weill Cornell Medicine flow cytometry core.
GoT-IM
In order to simultaneously capture genotyping data and whole transcriptomic data, Genotyping of Transcriptomes (GoT) was performed by adapting the 10x Genomics platform as previously described18. FACS-sorted CD34+ cells for each time point from the same individual were pooled, targeting a total of 10,000-15,000 cells. The standard 10x Genomics Chromium 3′ (v.3 or v.3.1 chemistry) was implemented according to the manufacturer’s recommendations up to the cDNA amplification step (10X Genomics, Pleasanton, CA). After cDNA amplification and SPRI cleanup, 10x scRNA-seq and ADT/HTO libraries were generated as recommended. A portion of the cDNA was used for somatic genotyping as previously described20. Briefly, to capture the somatic genotypes of cells, cDNA was amplified with a locus-specific amplification (10 PCR cycles), using the generic forward SI-PCR primer and the following locus-specific reverse primer for the CALR mutations (see Supplementary Table 17 for primer sequences). The amplified locus-specific cDNAs are then cleaned using SPRI purification to remove unincorporated primers. Finally, the targeted amplicon libraries are generated through a PCR performed with a generic forward PCR primer together with an RPI-x primer (Supplementary Table 17). The targeted amplicon libraries were spiked into the remainder of the gene expression and immunophenotyping libraries to be sequenced together on a NovaSeq (Illumina, San Diego, CA). The cycle settings were as follows: 28 cycles for Read 1, 90 cycles for Read 2, 8 cycles for i7 sample index and 8 cycles for i5 sample index.
GoT-IM scRNA-seq data processing, alignment, cell-type classification and clustering
For single-cell GoT-IM data from IFN01-IFN05 patient samples, the pooled scRNA-seq, CITE- seq and hashing libraries were processed with Cell Ranger (v6.1.1 and v6.1.2) using cellranger- multi pipeline (v1). The reads were aligned to the human genome GRCh38 with default parameters. The Seurat package26 (v4.1.0) was used to perform unbiased clustering of the CD34+ sorted cells from patient (v4.1.0). In brief, for individual datasets, cells with UMI > or < 3 standard deviations from the mean UMI or mitochondrial gene percentage > 10%, were filtered (Supplementary Figure 1a) . The HTO data was normalized with centered log-ratio (CLR) transformation and used to assign the time-points for each experiment24. The cells from each time point were analytically separated into individual datasets. These individual datasets, together with the baseline ET samples from our previous work20, were integrated and underwent batch- correction within Seurat, which implements canonical correlation analysis and the principles of mutual nearest neighbor84. Recommended settings were used for the integration (30 canonical correlation vectors for canonical correlation analysis in the FindIntegrationAnchors function and 30 principal components for the anchor weighting procedure in IntegrateData function). Principal component analysis was performed using variable genes using recommended settings (i.e. top 2000 variable genes using variance stabilizing transformation)84. The first statistically significant 30 principal components were used as inputs to the UMAP algorithm for cluster visualization85. Clusters were manually assigned based on differentially expressed genes using the FindAllMarkers function using default settings (using top 2000 variable genes, in a minimum of 10% of cells in either of the two comparison sets as input, and log-transformed fold change of 0.25 as the threshold, using Wilcoxon rank sum test). We identified 35 clusters in the integrated data, which were annotated according to canonical lineage markers identified previously in single-cell RNA- seq data of normal hematopoietic progenitor cells27. These clusters were collapsed into 15 main progenitor subsets based on expression of levels of these canonical markers (Supplementary Fig. 2a,b). The CITE-seq data was normalized using the CLR transformation, and used to identify the HSCs (Supplementary Fig. 2c)
Ironthrone-GoT for processing targeted amplicon sequences and mutation calling
Analysis of the GoT library was carried out as described previously20 using Ironthrone pipeline V2.142. Amplicon reads (Read 2) were screened for the presence of the primer sequence and the shared sequence (i.e. the expected sequence between the primer sequence and the mutation locus). Reads (Read 1) from GoT-IM experiments were also assessed for matching to the cell barcode list of the 10x dataset. A mismatch of 20% was allowed for all sequence matching steps. Only UMIs with at least 2 or more supporting reads were retained for final genotyping assignments, after the UMI collapse algorithm42, 86. Filtered cells were then genotyped as follows: cells with at least one mutant UMI were categorized as mutant cells whereas cells with no mutant UMI and at least one wildtype UMI were identified as wildtype.
Gene module scoring, differential expression and gene set enrichment analysis
For examining gene and gene module expression (e.g. HSC-specific IFNa-induced gene signature), the function AddModuleScore was used to calculate the relative expression of the genes for each cell within the Seurat package was calculated using the AddModuleScore function)84. To calculate the module expression of cell-cycle related genes, G2M phase and S phase marker genes were used as available in Seurat with CellCycleScoreing function. Briefly, control gene module expressions were calculated and subtracted from the average gene module expression of interest, as previously described87. All analyzed genes were classified based on average expression into 24 bins, and for each gene in the module, 100 control genes are randomly selected from the same expression bin as the gene of interest87. For statistical analysis, if cell types are being compared, cell type status was entered as the fixed effect and subjects as random effects in a linear mixed model. If the wildtype and mutated cells are being compared, genotype status was entered as the fixed effect and subjects as random effects in a linear mixed model. P-values were obtained by likelihood ratio tests of the full model with the fixed effect against the model without the fixed effect.
For differential gene expression testing within an individual (e.g. Fig. 3h) between two groups were calculated using FindMarkers function using the logistic regression framework. The tested genes included the top 2,000 variable genes from the CCA integration, which were filtered for those expressed in at least 10% of either group. In aggregated differential gene expression analysis, the two groups (e.g. treated versus baseline) via the linear mixed model framework. For each gene, treatment status was entered as the fixed effect and subjects as random effects. P-values were obtained by likelihood ratio tests of the full model with the fixed effect against the model without the fixed effect. Pathway enrichment was performed via a pre-ranked gene set enrichment approach (ranking based on the sign of the fold change * -log10(P-value)) using the msigdbr (v7.2.1) and fgsea (v1.12.0) R packages, using the canonical Hallmark pathway genes from MsigDB88.
GoT-ATAC
Cryopreserved bone marrow mononuclear cells were thawed and stained using standard procedures (30 min, 4°C) with the surface antibody CD34-PE (clone AC136, dilution 1:50, MACS) and DAPI (Sigma-Aldrich), according to manufacturer’s protocol. Cells were subsequently sorted for DAPI−, CD34+ cells using BD Influx at the Weill Cornell Medicine flow cytometry core. Nuclei were isolated from DAPI-, CD34+ cells according to 10x Genomics Demonstrated Low Cell Input Nuclei Isolation protocol. Lysis buffer was prepared following manufacturer’s recommendations and then split into aliquots for each serial sample. Either TotalSeq-A Anti-Nuclei Pore Complex Proteins Hashtag 9 or 10 Antibody (1uL at 1:5 dilution; BioLegend) was added to each aliquot of lysis buffer. Low-input nuclei isolation was otherwise performed following manufacturer’s recommendations. Subsequently, nuclei from each time point were counted and pooled together at approximately equal proportions. For IFN07, additional nuclei from the IFNa-treated sample were available to be run on a separate lane. Single-nucleus gene expression (GEX) and chromatin accessibility libraries were constructed from the pooled nuclei according to the Chromium Next GEM Single Cell Multiome User Guide (10X Genomics). Genotyping libraries targeting the CALR mutant transcripts were constructed from the remaining amplified cDNA, similar to the original GoT method. For each PCR, 12.5uL Kapa HiFI HotStart Ready Mix was mixed with 0.75uL of 10uM forward primer, 0.75uL of 10uM reverse primer and nuclease-free water for a total reaction volume of 25uL. In the first PCR, 3uL cDNA was re- amplified with Partial TSO and Partial Read 1 primers, using the following PCR condition: 98°C for 3min; 3 cycles of 98°C for 15sec, 67°C for 20sec and 72°C for 1min; 72°C for 1min. The re- amplified sampled was purified and concentrated via 0.7X SPRI cleanup, eluting it into 10uL Buffer EB. To pre-enrich the CALR mutation locus, a gene-specific PCR was performed with 3uL of cleaned re-amplified cDNA and Partial Read 1 and gene-specific primers (gene-specific, see Supplementary Table 17 for primer sequences). The following PCR condition was used: 98°C for 3min; 11 cycles of 98°C for 20sec, 60°C for 20sec, and 72°C for 2min; 72°C for 2min. After 0.7X SPRI cleanup, cDNA was eluted into 10uL Buffer EB. CALR locus-specific amplification was then performed with 3uL of cleaned gene-specific amplified cDNA and SI-PCR and loci- specific Primers, using the PCR condition: 98°C for 3min; 11 cycles of 98°C for 20sec, 60°C for 20sec, and 72°C for 2min; 72°C for 2min. A 0.7X SPRI cleanup was performed, and cDNA was eluted into 11uL Buffer EB. Finally, to construct the targeted amplicon library, loci-amplified cDNA was mixed with P5 Generic and RPxx indexing primers and amplified with the PCR condition: 98°C for 3min; 5 cycles of 98°C for 15sec, 60°C for 20sec, and 72°C for 1min; 72°C for 1min. The constructed library was cleaned via 0.8X SPRI cleanup and eluted into 12uL Buffer EB.
At the cDNA amplification stage of the Chromium Next GEM Single Cell Multiome protocol, supernatant from the 0.6X size selection was retained and was used to generate the hashing libraries (see Supplementary Table 17 for TruSeq DNA D7xx_s primers used) as per HTO protocol25 with the following modification. For the hashing library construction step, the PCR reaction was prepared with 0.65uL of 10uM SI-PCR primer, 0.65uL of 10uM TruSeq DNA D7xx_s primer, 11.25uL cleaned supernatant and 12.5 uL Kapa HiFI HotStart Ready Mix (Roche, Basel, Switzerland). Hashing, gene expression and genotyping libraries were pooled and sequenced together on a NovaSeq (Illumina) with cycle settings: 28 cycles for Read 1, 90 cycles for Read 2, 10 cycles for i7 sample index and 10 cycles for i5 sample index. The ATAC library was sequenced separately on a NovaSeq, with cycle settings: 50 cycles for Read 1 and Read 2, 8 cycles for i7 sample index and 24 cycles for i5 sample index.
Data preprocessing, alignment and cell type identification for GoT-ATAC
For single-nuclei GoT-ATAC data from IFN01 and IFN06-08 patient samples, 10x data were processed using Cell Ranger (v6.1.1 and v6.1.2). Multi-omic nuclear data for snATAC-seq and snRNA-seq were processed together with Cell Ranger Arc (v2.0.0). snRNA-seq data was also combined with cell hashing data (HTO) and run using the Cell Ranger Multi pipeline (v1). The reads were aligned to the human reference genome GRCh38. The downstream analysis of the processed data was performed using Seurat (v4.1.0)26 and Signac (v1.5.0)89 packages. Ironthrone- GoT protocol mentioned above was used to determine mutation calling within the GoT-ATAC assay. For the ATAC analysis, we called peaks on individual samples using MACS2 peak caller63. Gene annotations from EnsDb.Hsapiens.v75 and motifs annotations from Cis-BP90 for TF binding motifs were utilized. Cells with blacklist ratio >0.02, TSS enrichment <2 and nucleosome signal >4 were filtered out. For the RNA data, cells with UMI > or < 3 standard deviations from the mean UMI or mitochondrial gene percentage > 25%, were filtered. The nuclei hashing data processing was performed as for the GoT-IM data. As nuclear hashing is known to be noisy89 the nuclear hashing data was used in combination with cell clustering data, as the cells cluster based on treatment status. snRNA-seq data integration and and cell type assignment was performed as described. Integration via the ATAC-seq data was performed by normalizing the merged counts using first term frequency inverse document frequency (TFIDF) normalization with RunTFDIF followed by linear dimensional reduction using latent semantic indexing (LSI). The first 2:30 dimensions were retained and batch-correction was perfomed with runHarmony (Harmony, v0.1.0) which iteratively learns cell-specific linear correction function to account for batch effect.
Identification of distal regulatory elements with gene-peak cis-association
For each GoT-ATAC sample, we examined all ATAC peaks within ± 500kb of all annotated TSS to identify regulatory networks of genes using LinkPeaks function63. Pearson correlation between gene expression and accessibility of the peaks located in the window was calculated after correcting for bias arising from GC content, overall accessibility and peak size. Recommended settings were implemented (200 background peaks per peak with similar GC content and accessibility, P-value < 0.05 and min.cell = 10).
Motif enrichment analysis
Per-cell TF motif activity score (chromatin accessibility) was calculated by running chromVAR 1.18.071. We used the curated Cis-BP motif database90 which contains 1141 human TF motif PFMs. The function matchMotifs was first called to identify which peaks contain which motifs (P- value = 5 x 10-5). A set of background peaks that are similar to a peak in GC content and average accessibility was internally picked and used for normalizing the deviation scores. Deviation Z- scores, namely bias-corrected deviation z-scores in accessibility from the expected accessibility based on the average of all the cells, were then calculated for each TF motif and each cell.
To perform differential motif enrichment analysis, within a sample and on the deviation z-score computed by chromVAR, we applied the function FindMarkers in Signac (Wilcoxon Rank Sum test), where the average difference in z-score between the groups was calculated. For integrated data, we combined the P-values (Fisher’s method) and calculated weighted mean deviation score across individual samples. P-values were adjusted by the Benjamini-Hochberg method.
To find over-enriched motifs for a given genomic feature set, the FindMotifs function was used, accounting for accessibility and GC content bias by selecting 5000 accessible background peaks with similar GC content for each feature set. ChromVAR was used to compute the synergy between pairs of TF motifs, where synergy is defined as the excess variability of chromatin accessibility for peaks sharing both motifs compared to a random sub-sample of the same size of peaks with one motif. High synergy usually indicated that cooperative binding relationship between pairs of TFs. The function getAnnotationSynergy was called to calculate synergy scores.
Retrospective flow cytometry data analysis
Retrospective flow cytometry data analysis was performed in accordance with relevant guidelines, regulations and approval by the Institutional Review Board at Weill Cornell Medicine (IRB #1007011151). Patient flow cytometry data selected corresponded to patients with myeloproliferative neoplasms and which had the same antibody panel analyzed by flow cytometry. The antibody panel chosen for evaluation was a modified version of the EuroFlow AML/MDS tube #451, 52 that consisted of the following antibody-fluorophore pairs, in addition to forward scattering and side scattering pulse area and width measurements (FSC-A, FSC-H, SSC- A, and SSC-H): cytoplasmic TdT/FITC (clone HT-6, Agilent/Dako, cat. F7139), CD56/PE (clone C5.9, Cytognos, cat. CYT-56PE), CD34/PerCP-Cy5.5 (clone 8G12, BD Biosciences, cat. 347213), CD117/PE-Cy7 (clone 104D2D1, Beckman Coulter/Immunotech, cat. IM3698), CD7/APC (clone GP40 [Leu-9], Invitrogen, cat. 17-0079-42), Fixable Viability Stain 700 (BD Biosciences, cat. 564997), CD19/APC-H7, HLA-DR/Pacific Blue (clone SJ25C1, BD Biosciences, cat. 643078), and CD45/V500 (clone 2D1, BD Biosciences, cat. 347213). Data was collected using BD LSR II flow cytometers, with approximately 500,000 events collected per antibody panel per sample to generate the raw data FCS files.
Custom software was developed using python, FlowKit53 and umap-learn54–56 to detect the antibody panels that were used to generate each raw data FCS file and to determine the unused flow cytometer channels that should be disregarded using the self-contained metadata for each file. Subsamples from each FCS file were combined to make an “ensemble FCS file” that could be used to create the UMAP embedding that could be applied to all of the individual files. Each subsample consisted of the same number of randomly-selected flow cytometer events such that the combined total number of events was approximately 250,000 for each unique antibody panel. The various channels were normalized and processed using UMAP to calculate the normalization constants and UMAP embedding that were then applied to all FCS files of the given antibody panel. Of note, the UMAP calculations included the forward scatter height (FSC-H), side scatter height (SSC-H), and each of the defined fluorescence channels. The normalization factors and UMAP embedding were then applied to all of the individual files. Modified FCS files were created that included the UMAPs as additional channels for subsequent evaluation and gating using FlowJo software (v10.8.1, FlowJo LLC, Ashland, Oregon, USA).
Using FlowJo, appropriate gates based on the UMAP plot were determined using the ensemble FCS file. Additional standard gating was also performed using the original data channels (gating using the other channels is essential to determine the identities of the various cell clusters within the UMAP plots). UMAP gates were based on data after gating out doublets and non-viable cells via standard gating approaches. Once the UMAP gates were determined to adequate satisfaction (sufficient segregation of cell subpopulations and verified to encompass cells of the same or similar type), they were then applied to all of the FCS files of the given antibody panel. The FCS files were divided into an interferon-treated cohort (n = 9) and a non-interferon-treated cohort (n = 47). The ratios and absolute numbers of cells, as well as other summary statistics were then calculated, and the values exported as CSV files. Statistics included numbers of CD34+ blasts; CD19+, cTDT+, CD34+ lymphoid progenitors; CD19+ lymphocytes; CD19-negative lymphocytes/NK cells; and monocytes. Relevant distributions of cells for each cohort are plotted, and the Wilcoxon p-statistic was calculated for various compared distributions.
Multiplexed Immunofluorescence
Multiplexed immunofluorescence (mIF) was performed using the Opal system (Akoya Biosciences) by staining 4 micron-thick Bouin-fixed, paraffin-embedded whole-tissue sections from decalcified human bone marrow core biopsy specimens in a Bond RX automated tissue stainer (Leica Biosystems, Buffalo Grove, IL), as described previously91, 92. Briefly, tissue sections were first deparaffinized prior to EDTA-based antigen retrieval (Leica ER2 solution, 20min). A cyclical staining protocol was then performed, with horseradish peroxidase-mediated deposition of tyramide-Opal fluorophore constructs (Akoya Biosciences) in each cycle, with intervening application of heat, citrate-based epitope retrieval solution (Leica ER1), and Bond Wash Solution (Leica) to execute stripping of primary/secondary antibody complexes between staining cycles. Finally, 4’, 6-diamidino-2-phenylindole (Spectral DAPI, Akoya Biosciences) was applied per provided protocols to label nuclei. The following panel of primary antibody/fluorophore pairs was applied to all cases, in a sequential order as shown: 1) Opal 480/anti-mutant CALR (1:120, CAL2, Dianova), 2) Opal 520/anti-CD38 (1:50, 38C03 (SPC32), Invitrogen), 3) Opal 570/anti-CD117 (1:100, D3W6Y, Cell Signaling), 4) Opal 620/anti-TdT (1:8, SEN28, Invitrogen), 5) Opal 690/anti-CD34 (1:100, QBEND/10, Invitrogen), 6) Opal 780/anti-Ki67 (Ready-to-use, MM1, Leica). Slides were cover-slipped using ProLong™ Diamond Antifade Mountant (Invitrogen). Whole slide scans were subsequently obtained at 20X magnification using the Vectra Polaris Automated Quantitative Pathology Imaging System (Akoya Biosciences) to generate a collection of tiled images, which were subsequently spectrally unmixed in InForm (v2.4.8, Akoya Biosciences). Unmixed tiles were finally fused together in HALO (v3.3.2541.231, Indica Labs) to generate a multi-layered TIFF image file for each sample, which was used in downstream analyses.
Image Analysis with PathML
Vectra whole-slide images (WSIs) were digitized using digital whole-slide scanners and stored in tiff format. WSIs of bone marrow sections were captured (n = 9 samples). Each sample was stained based on 8 cell markers including DAPI, mutant-specifc CALR, CD38, CD117, TdT, CD34, Ki67, and auto-fluorescence. Using Fiji, the brightness/contrast is adjusted from the Image menu. To analyze the Fiji-preprocessed WSIs, a machine-learning based package namely pathML was applied to images93. The PathML framework is based on PyThrch and TensorFlow deep learning libraries to segment the images. For tiling, the Slide.setTileProperties() method was run for extracting and storing each tile into a single image file. WSI is read using the pyvips library94 within PathML. Next, the DenseNet CNN is trained based on the default parameters of pathML to detect the tissue, artefact and background regions from a robust dataset containing various tissues and species types93. PathML package provides a function for quantifying the images based on image intensity. Using that function, a counts matrix is generated per each cell marker based on the intensity of marker. To remove the noise from the counts matrix data, any cell with intensity less than 50 for DAPI and auto-fluorescence markers was excluded from the analysis. Next, the thresholds were obtained to find the positive and negative expression level for each cell marker. The thresholds were manually set for each marker based on examination by a board-certified hematopathologist.
Data Availability
Data generated in this study will be available upon publication.
Supplementary Figures
a. Box plots showing number of UMIs (left panel) and genes (right panel) detected per cell in sorted CD34+ hematopoietic progenitors from each patient after filtering based on Quality Control (QC) metrics. Cells profiled using GoT-IM (Genotyping of Transcriptomes integrated with immunophenotyping). b. UMAP of sorted CD34+ progenitors (n = 49891 cells) highlighting MPN samples (n = 10 individuals) (left panel) and treatment status (right panel) after integration with Seurat package.
a. Heatmap of top 10 differentially expressed genes for each progenitor cell type. Cells of each cell type were down-sampled to the same number (n = 100 cells per cluster) b. Dot-plot showing expression levels of cell type specific gene markers in each progenitor subset. c. UMAP of sorted CD34+ progenitors (n = 49891 cells) highlighting CD38, CD90 and CD45RA protein expression and AVP RNA expression. d. UMAP showing gene expression levels of differentially upregulated TFs in IGP compared to IMPs.
a. Left panel: Analytical time point assignment (data demultiplexing). Using the time point specifying barcodes, baseline and IFNs-treated cells were separated. Cells in which both barcodes are detected are considered as doublets and excluded. Right panel: UMAP of CD34+ cells from a representative experiment that includes two time points (IFN01, n = 5225 baseline cells, 3764 IFNa treated year 1 cells), highlighting treatment status. b. UMAP of CD34+ cells from patient IFN03 highlighting cell type assignments (n = 727 baseline cells, 1135 IFNa treated cells at year 1, 5467 IFNa treated cells at year 4). c. Box plots showing normalized CD90 protein expression levels in CD34+ progenitors before and after treatment with IFNa. P-values from linear mixed model with/without treatment status.
a. Expression of CXCR4 vs. cell cycle gene expression in HSCs before and after IFNa treatment. Pie charts show frequencies of baseline versus treated cells in cell cycle-low, CXCR4-high (n = 523 cells) and those in cell cycle-high, CXCR4- low (n = 573 cells) populations. b. Volcano plot showing DE genes between baseline and IFNa- treated MkPs. DE genes identified using linear mixed modeling with/without treatment status. Genes highlighted in orange are enriched in the TGFb signaling and those in blue enriched in the IFNa/g response. Enrichment based on pre-ranked gene set enrichment analysis (GSEA) using the MSigDB Hallmark collection.
a. UMAP of sorted CD34+ stem and progenitors at baseline and after IFNa treatment. CALR mutation status highlighted. Cells from each sample were down- sampled to the same number for each treatment status (n = 500 cells from each treatment status per sample) b. Box plot showing normalized CXCR4 expression levels in HSCs at baseline (n = 2655 wildtype; n = 2030 CALR-mutated) and after IFNa treatment (n = 1169 wildtype; n = 2655 CALR-mutated). P-values from linear mixed model with/without treatment status or genotype. c. Frequencies of proliferating myeloid cells before and after IFNa treatment, as assessed by Ki67 expression. P-values from Wilcoxon rank sum test.
a. Left panel: Normalized frequencies of HSCs at each time-point for IFN01 (n = 8989 cells), IFN02 (n = 5939 cells), IFN03 (n = 7329 cells) and IFN04 (n = 7964 cells). P-values from linear mixed model comparing different treatment time-points. Right panel: Normalized frequencies of each progenitor subset among WT, single MUT (Type 1 CALR) and double MUT (Type 1 and SNV mutations in CALR) cell populations at each timepoint for IFN04. b. Relative density of proportion of Type 1-mutant reads versus fraction of SNV-mutant reads in UMIs captured via GoT, showing mutual exclusivity. c. UMAP of IFN04 patient based on scRNA-seq data highlighting cell types, treatment time-points and mutation status (from left to right, n = 7964 cells).
a. Box plots showing number of UMIs (left) and genes (right) detected per cell in sorted CD34+ hematopoietic stem and progenitors from each patient after filtering based on Quality Control (QC) metrics. Cells profiled using GoT-ATAC. b. Density plot comparing percentage of snATAC fragments within peaks to the total number of fragments detected per sample. c. Distribution of mean TSS enrichment score at each position relative to the TSS per sample (n = 4 patients, 7 samples). d. Average distribution of fragment length per sample (n = 4 patients, 7 samples). e. UMAP of sorted CD34+ stem and progenitors (n = 24857 cells), with patient and treatment status highlighted (left and right panels, respectively).
a. Heatmap of top 10 differentially expressed genes for each progenitor cell type. Cells of each cell type were down-sampled to the same number (n = 100 cells per cluster). b. Dot-plot showing expression levels of cell type-specific gene markers in each progenitor subset. c. UMAP highlighting TF motif accessibility (n = 24857 cells). TF accessibility scores added with AddChromatinScore function in Signac. d. Heatmap showing cell type specific TF accessibility scores.
a. UMAP of CD34+ cells (n =23137 cells, 7 samples, 4 patients) based on snATAC- seq data overlaid with patient ID (left) and treatment time-point (right). b. Chromatin accessibility tracks of regulatory regions of RFX2 (bottom), distal region enriched with the two most significant positively regulating loci (top-left, top-right).
Left panel: Normalized cell frequencies of progenitor subsets at baseline and after IFNa treatment. Cells from each treatment status and individual were down-sampled to the same number (n = 100 cells per treatment status per sample). Right panel: Cell frequency distribution as in left panel for patients IFN01, IFN06 and IFN07 (n = 100 cells per treatment status).
Acknowledgments
The work was enabled by the Weill Cornell Epigenomics Core and Flow Cytometry Core. D.A.L. is supported by the Burroughs Wellcome Fund Career Award for Medical Scientists, Valle Scholar Award, and the National Institutes of Health Director’s New Innovator Award (DP2-CA239065). N.D. is supported by a F30 Predoctoral Fellowship from the NHLBI of the National Institutes of Health (F30HL156496) and by a Medical Scientist Training Program grant from the National Institute of General Medical Sciences of the National Institutes of Health under award number T32GM007739 to the Weill Cornell/Rockefeller/Sloan Kettering Tri-Institutional MD-PhD Program. A.S.N. is supported by the Burroughs Wellcome Fund Career Award for Medical Scientists, National Institutes of Health Director’s Early Independence Award (DP5 OD029619- 01) and Starr Cancer Consortium.
References
- 1.↵
- 2.↵
- 3.↵
- 4.↵
- 5.↵
- 6.↵
- 7.↵
- 8.↵
- 9.↵
- 10.
- 11.↵
- 12.↵
- 13.↵
- 14.↵
- 15.↵
- 16.↵
- 17.↵
- 18.↵
- 19.↵
- 20.↵
- 21.↵
- 22.↵
- 23.↵
- 24.↵
- 25.↵
- 26.↵
- 27.↵
- 28.↵
- 29.↵
- 30.↵
- 31.↵
- 32.↵
- 33.
- 34.↵
- 35.↵
- 36.
- 37.↵
- 38.↵
- 39.↵
- 40.↵
- 41.↵
- 42.↵
- 43.↵
- 44.↵
- 45.↵
- 46.↵
- 47.
- 48.
- 49.↵
- 50.↵
- 51.↵
- 52.↵
- 53.↵
- 54.↵
- 55.
- 56.↵
- 57.↵
- 58.↵
- 59.↵
- 60.↵
- 61.↵
- 62.↵
- 63.↵
- 64.↵
- 65.↵
- 66.↵
- 67.↵
- 68.↵
- 69.↵
- 70.↵
- 71.↵
- 72.↵
- 73.↵
- 74.↵
- 75.
- 76.↵
- 77.↵
- 78.↵
- 79.↵
- 80.↵
- 81.↵
- 82.↵
- 83.↵
- 84.↵
- 85.↵
- 86.↵
- 87.↵
- 88.↵
- 89.↵
- 90.↵
- 91.↵
- 92.↵
- 93.↵
- 94.↵
