Search

Supplementary Methods

Acute leukemias are systemic malignancies associated with a dire outcome. Because of low immunogenicity, leukemias display a remarkable ability to evade immune control and are often resistant to checkpoint blockade. Here, we discover that leukemia cells actively establish a suppressive environment to prevent immune attacks by co-opting a signaling axis that skews macrophages toward a tumor-promoting tissue repair phenotype, namely the GAS6/AXL axis. Using aggressive leukemia models, we demonstrate that ablation of the AXL receptor specifically in macrophages, or its ligand GAS6 in the environment, stimulates antileukemic immunity and elicits effective and lasting natural killer cell– and T cell–dependent immune response against naïve and treatment-resistant leukemia. Remarkably, AXL deficiency in macrophages also enables PD-1 checkpoint blockade in PD-1–refractory leukemias. Finally, we provide proof-of-concept that a clinical-grade AXL inhibitor can be used in combination with standard-of-care therapy to cure established leukemia, regardless of AXL expression in malignant cells.Significance:Alternatively primed myeloid cells predict negative outcome in leukemia. By demonstrating that leukemia cells actively evade immune control by engaging AXL receptor tyrosine kinase in macrophages and promoting their alternative priming, we identified a target which blockade, using a clinical-grade inhibitor, is vital to unleashing the therapeutic potential of myeloid-centered immunotherapy.This article is highlighted in the In This Issue feature, p. 2659

Supplementary Figures and Figure legends

The cellular mechanisms required to ensure homeostasis of the hematopoietic niche and the ability of this niche to support hematopoiesis upon stress remain elusive. We here identify Wnt5a in Osterix+ mesenchymal progenitor and stem cells (MSPCs) as a critical factor for niche-dependent hematopoiesis. Mice lacking Wnt5a in MSPCs suffer from stress-related bone marrow (BM) failure and increased mortality. Niche cells devoid of Wnt5a show defective actin stress fiber orientation due to an elevated activity of the small GTPase CDC42. This results in incorrect positioning of autophagosomes and lysosomes, thus reducing autophagy and increasing oxidative stress. In MSPCs from patients from BM failure states which share features of peripheral cytopenia and hypocellular BM, we find similar defects in actin stress fiber orientation, reduced and incorrect colocalization of autophagosomes and lysosomes, and CDC42 activation. Strikingly, a short pharmacological intervention to attenuate elevated CDC42 activation in vivo in mice prevents defective actin-anchored autophagy in MSPCs, salvages hematopoiesis and protects against lethal cytopenia upon stress. In summary, our study identifies Wnt5a as a restriction factor for niche homeostasis by affecting CDC42-regulated actin stress-fiber orientation and autophagy upon stress. Our data further imply a critical role for autophagy in MSPCs for adequate support of hematopoiesis by the niche upon stress and in human diseases characterized by peripheral cytopenias and hypocellular BM.

Clonal hematopoiesis (CH) is an age-related condition predisposing to blood cancer and cardiovascular disease (CVD). Murine models demonstrate CH-mediated altered immune function and proinflammation. Low-grade inflammation has been implicated in the pathogenesis of osteoarthritis (OA), the main indication for total hip arthroplasty (THA). THA-derived hip bones serve as a major source of healthy hematopoietic cells in experimental hematology. We prospectively investigated frequency and clinical associations of CH in 200 patients without known hematologic disease who were undergoing THA. Prevalence of CH was 50%, including 77 patients with CH of indeterminate potential (CHIP, defined as somatic variant allele frequencies [VAFs] ≥2%), and 23 patients harboring CH with lower mutation burden (VAF, 1% to 2%). Most commonly mutated genes were DNMT3A (29.5%), TET2 (15.0%), and ASXL1 (3.5%). CHIP is significantly associated with lower hemoglobin, higher mean corpuscular volume, previous or present malignant disease, and CVD. Strikingly, we observed a previously unreported association of CHIP with autoimmune diseases (AIDs; multivariable adjusted odds ratio, 6.6; 95% confidence interval, 1.7-30; P = .0081). These findings underscore the association between CH and inflammatory diseases. Our results have considerable relevance for managing patients with OA and AIDs or mild anemia and question the use of hip bone–derived cells as healthy experimental controls.

PurposeHypereosinophilia represents a heterogenous group of severe medical conditions characterized by elevated numbers of eosinophil granulocytes in peripheral blood, bone marrow or tissue. Treatment options for hypereosinophilia remain limited despite recent approaches including IL-5-targeted monoclonal antibodies and tyrosine kinase inhibitors.MethodsTo understand aberrant survival patterns and options for pharmacologic intervention, we characterized BCL-2-regulated apoptosis signaling by testing for BCL-2 family expression levels as well as pharmacologic inhibition using primary patient samples from diverse subtypes of hypereosinophilia (hypereosinophilic syndromen = 18, chronic eosinophilic leukemia not otherwise specifiedn = 9, lymphocyte-variant hypereosinophilian = 2, myeloproliferative neoplasm with eosinophilian = 2, eosinophilic granulomatosis with polyangiitisn = 11, reactive eosinophilian = 3).ResultsContrary to published literature, we found no difference in the levels of the lncRNAMorrbidand its targetBIM. Yet, we identified a near complete loss of expression of pro-apoptoticPUMAas well as a reduction in anti-apoptoticBCL-2. Accordingly,BCL-2inhibition using venetoclax failed to achieve cell death induction in eosinophil granulocytes and bone marrow mononuclear cells from patients with hypereosinophilia. In contrast,MCL1inhibition using S63845 specifically decreased the viability of bone marrow progenitor cells in patients with hypereosinophilia. In patients diagnosed with Chronic Eosinophilic Leukemia (CEL-NOS) or Myeloid and Lymphatic Neoplasia with hypereosinophilia (MLN-Eo) repression of survival was specifically powerful.ConclusionOur study shows thatMCL1inhibition might be a promising therapeutic option for hypereosinophilia patients specifically for CEL-NOS and MLN-Eo.

Clonal hematopoiesis (CH) is characterized by somatic mutations in blood cells of individuals without hematologic disease. While the mutational landscape of CH in peripheral blood (PB) has been well characterized, detailed analyses addressing its spatial and cellular distribution in the bone marrow (BM) compartment are sparse. We studied CH driver mutations in healthy individuals (n = 261) across different anatomical and cellular compartments. Variant allele frequencies were higher in BM than PB and positively correlated with the number of driver variants, yet remained stable during a median of 12 months of follow-up. In CH carriers undergoing simultaneous bilateral hip replacement, we detected ASXL1-mutant clones in one anatomical location but not the contralateral side, indicating intra-patient spatial heterogeneity. Analyses of lineage involvement in ASXL1-mutated CH showed enriched clonality in BM stem and myeloid progenitor cells, while lymphocytes were particularly involved in individuals carrying the c.1934dupG variant, indicating different ASXL1 mutations may have distinct lineage distribution patterns. Patients with overt myeloid malignancies showed higher mutation numbers and allele frequencies and a shifting mutation landscape, notably characterized by increasing prevalence of DNMT3A codon R882 variants. Collectively, our data provide novel insights into the genetics, evolution, and spatial and lineage-specific BM involvement of CH.

Convolutional neural networks (CNNs) excel as powerful tools for biomedical image classification. It is commonly assumed that training CNNs requires large amounts of annotated data. This is a bottleneck in many medical applications where annotation relies on expert knowledge. Here, we analyze the binary classification performance of a CNN on two independent cytomorphology datasets as a function of training set size. Specifically, we train a sequential model to discriminate non-malignant leukocytes from blast cells, whose appearance in the peripheral blood is a hallmark of leukemia. We systematically vary training set size, finding that tens of training images suffice for a binary classification with an ROC-AUC over 90%. Saliency maps and layer-wise relevance propagation visualizations suggest that the network learns to increasingly focus on nuclear structures of leukocytes as the number of training images is increased. A low dimensional tSNE representation reveals that while the two classes are separated already for a few training images, the distinction between the classes becomes clearer when more training images are used. To evaluate the performance in a multi-class problem, we annotated single-cell images from a acute lymphoblastic leukemia dataset into six different hematopoietic classes. Multi-class prediction suggests that also here few single-cell images suffice if differences between morphological classes are large enough. The incorporation of deep learning algorithms into clinical practice has the potential to reduce variability and cost, democratize usage of expertise, and allow for early detection of disease onset and relapse. Our approach evaluates the performance of a deep learning based cytology classifier with respect to size and complexity of the training data and the classification task.

Background: Clonal hematopoiesis (CH) describes the presence of genetic alterations and expansion of clonal cell populations in the peripheral blood (PB) of individuals without clinical manifestation of a hematologic malignancy. CH is a common, age-related state. However, individuals carrying CH are at greater risk for hematologic cancer. It has been shown that presence of TP53- and other high-risk mutations, variant allele frequency (VAF) >10% or multiple mutations in CH carriers may confer a higher risk of preleukemic development and transformation to myelodysplastic syndrome (MDS) or acute myeloid leukemia (AML). Nevertheless, further insights into the role of CH in clonal evolution towards leukemia and elaborating features that are predictive of leukemic progression are needed. Aims: We have recently shown that CH is common in individuals undergoing hip replacement surgery (Hecker, Hartmann et al., Blood 2021). Here, we compared mutational spectrum in these CH individuals to MDS and secondary AML (sAML) cohorts. Additionally, we analyzed spatial and lineage distribution of CH and longitudinally monitored CH and MDS clones over time. Methods: Samples from individuals without known hematologic disease undergoing hip replacement surgery (n=288 samples from 261 individuals), patients with MDS (n=92) or sAML (n=123) were screened for variants in 68 leukemia-associated genes using a targeted sequencing approach (VAF cut off, 1%). Follow-up (FU) PB samples were available for 21 individuals with CH and 16 untreated low-risk MDS patients, 6-24 months after screening. n=5 CH bone marrow (BM) samples carrying six ASXL1-mutations were sorted for seven different cell fractions. Results: At screening, variants were detected in 127/261 (49%) healthy individuals, 84/92 (91%) MDS and 117/123 (95%) sAML patients, with median VAFs of 2.7% (ranging from 1-44%), 18.8% (1.1-87%) and 37.1% (1-99%), respectively. Individuals with CH had a median number of 1 variant per individual, whereas median detected variants per patient increased with clonal evolution with 3 variants in the MDS and 4 variants per patient in the sAML cohort. CH, MDS and sAML showed entity-specific mutation profiles (Figure 1A). Most variants in CH affected epigenetic modifiers, while mutations in splicing factors, signaling pathways and transcription factors increased with clonal progression. During FU, untreated low-risk MDS patients more frequently gained additional mutations compared to CH individuals (7/16 vs 2/21, respectively; p=0.024). However, we did not observe significant changes in clone sizes over time. In 17 hip replacement individuals both femoral heads were removed simultaneously, allowing paired analysis of two different hematopoietic sites. CH prevalence in this subgroup was 70.6% (12/17). Ten individuals showed identical mutation patterns in BM obtained from the right and left femoral head, with little differences in VAFs (Table 1). In contrast, two other individuals showed significant differences in variant detection comparing one to the other side: ASXL1-mutations were only present in one hip sample, whereas all the other variants were detectable in both sides (p To further characterize ASXL1 mosaicism across hematopoietic lineages and differentiation stages, we sorted and sequenced n=5 CH BM samples carrying six ASXL1-mutations for seven different hematopoietic cell fractions. In all cases ASXL1-mutations were identified in CD34 +CD38 -CD90 + hematopoietic stem cells (HSC). VAFs of ASXL1-mutations were differentially distributed (p=0.0049, Kruskal Wallis test): detected VAFs were significantly higher in HSC and common myeloid progenitors (CMP) compared to VAFs in T-cells (p=0.045 and p=0.011, respectively; Dunn´s multiple comparison test; Figure 1B). Conclusions: CH, MDS and sAML show characteristic mutation profiles that remained stable over a 6-24-month period. Gains of additional variants and clonal expansion associated with disease progression. Cellular distribution analysis of ASXL1 CH variants revealed characteristic repartition patterns within the hematopoietic differentiation tree. Additionally, differences in variant detection between cellular BM compartments indicates spatial heterogeneity of CH clones warranting further investigation. J.S.H. and L.H. contributed equally. Figure 1 Figure 1. Disclosures Platzbecker: Geron: Honoraria; Takeda: Honoraria; AbbVie: Honoraria; Celgene/BMS: Honoraria; Janssen: Honoraria; Novartis: Honoraria. Goetze: BMS/Celgene: Other: Advisory Board, Research Funding; Abbvie: Other: Advisory Board.

Acute leukemias are systemic malignancies associated with a dire outcome. Because of low immunogenicity, leukemias display a remarkable ability to evade immune control and are often resistant to checkpoint blockade. Here, we discover that leukemia cells actively establish a suppressive environment to prevent immune attacks by co-opting a signaling axis that skews macrophages toward a tumor-promoting tissue repair phenotype, namely the GAS6/AXL axis. Using aggressive leukemia models, we demonstrate that ablation of the AXL receptor specifically in macrophages, or its ligand GAS6 in the environment, stimulates antileukemic immunity and elicits effective and lasting natural killer cell– and T cell–dependent immune response against naïve and treatment-resistant leukemia. Remarkably, AXL deficiency in macrophages also enables PD-1 checkpoint blockade in PD-1–refractory leukemias. Finally, we provide proof-of-concept that a clinical-grade AXL inhibitor can be used in combination with standard-of-care therapy to cure established leukemia, regardless of AXL expression in malignant cells. Significance: Alternatively primed myeloid cells predict negative outcome in leukemia. By demonstrating that leukemia cells actively evade immune control by engaging AXL receptor tyrosine kinase in macrophages and promoting their alternative priming, we identified a target which blockade, using a clinical-grade inhibitor, is vital to unleashing the therapeutic potential of myeloid-centered immunotherapy. This article is highlighted in the In This Issue feature, p. 2659

Summary Classically, hematopoietic stem cell (HSC) differentiation is assumed to occur via progenitor compartments of decreasing plasticity and increasing maturity in a specific, hierarchical manner. The classical hierarchy has been challenged in the past by alternative differentiation pathways. We abstracted experimental evidence into 10 differentiation hierarchies, each comprising 7 cell type compartments. By fitting ordinary differential equation models with realistic waiting time distributions to time-resolved data of differentiating HSCs from 10 healthy human donors, we identified plausible lineage hierarchies and rejected others. We found that, for most donors, the classical model of hematopoiesis is preferred. Surprisingly, multipotent lymphoid progenitor differentiation into granulocyte-monocyte progenitors is plausible in 90% of samples. An in silico analysis confirmed that, even for strong noise, the classical model can be identified robustly. Our computational approach infers differentiation hierarchies in a personalized fashion and can be used to gain insights into kinetic alterations of diseased hematopoiesis., Graphical abstract, Highlights • We assembled 10 lineage hierarchy models of human hematopoiesis • Multiparameter immunophenotyping determines HSC differentiation for 10 healthy donors • ODE fitting and model selection allows to identify plausible hierarchies • A simulation study confirms robustness of model selection for different noise levels, stem cells research; in silico biology; systems biology