Diversity in the bone marrow niche: Classic and novel strategies to uncover niche composition

Summary Our view on the role and composition of the bone marrow (BM) has dramatically changed over time from a simple nutrient for the bone to a highly complex multicellular tissue that sustains haematopoiesis. Among these cells, multipotent haematopoietic stem cells (HSCs), which are predominantly quiescent, possess unique self‐renewal capacity and multilineage differentiation potential and replenish all blood lineages to maintain lifelong haematopoiesis. Adult HSCs reside in specialised BM niches, which support their functions. Much effort has been put into deciphering HSC niches due to their potential clinical relevance. Multiple cell types have been implicated as HSC‐niche components including sinusoidal endothelium, perivascular stromal cells, macrophages, megakaryocytes, osteoblasts and sympathetic nerves. In this review we provide a historical perspective on how technical advances, from genetic mouse models to imaging and high‐throughput sequencing techniques, are unveiling the plethora of molecular cues and cellular components that shape the niche and regulate HSC functions.

depth to capture the small number of cells that are in cell-cell contact with the rare HSC population.
Thus, although a remarkable body of data has been produced, the exact composition of the niches that harbour and maintain HSCs and haematopoietic stem and progenitor cells (HSPCs) are still a matter of intense debate. In this review, we focus on the niches that support HSCs as the most extensively studied and on mouse as the most widespread mammalian model of haematopoiesis (Table 2). We provide a historical perspective on how our understanding of the structure and cellular composition of the BM niche has evolved over time based on the implementation of new approaches. We describe in detail these research strategies, their technical limitations and advantages, what was learnt from them and how new methods complement previous techniques to elucidate the cellular interactomes in the BM [1][2][3][4][5][6]15 (Table 1).

STE M CE L L S A N D N ICH E CE L L S: MOR E TH A N N EIGHBOU R S
In 1896 Artur Pappenheim 17 conceptualised the term stem cell describing a precursor cell capable of generating other mature F I G U R E 1 Schematic of the bone marrow (BM) microenvironment and its main regulatory components. The highly specialised BM niche supports and regulates the functions and dynamics of haematopoietic stem cells (HSCs) and other haematopoietic progenitors. Cumulative research indicates that arterioles, sinusoids and associated perivascular stromal cells play a key role in HSC maintenance. Besides endothelial cells, which produce stem cell factor (SCF) and CXC-chemokine ligand 12 (CXCL12), vascular niche components include mesenchymal stem cells (MSCs) such as periarteriolar NG2 + cells, Nestin high cells, MYH11 + cells, CXCL12-abundant reticular (CAR) cells, perisinusoidal LepR + cells and Nestin low cells. Of note, there is a significant overlap between these MSCs populations. Recently described Osteo-CAR and adipo-CAR cells produce the highest cytokine levels in the BM and preferentially localise close to arteriolar and sinusoidal niches, respectively. Sympathetic nerves, through the release of noradrenaline (NA), are also important regulators of the HSC niche. Non-myelinating Schwann cells produce transforming growth factor β (TGFβ) controlling HSC quiescence. The role of osteoblasts in HSC regulation is controversial but they may regulate lymphoid progenitors. Macrophages and megakaryocytes are mature haematopoietic cell types that control HSC behaviour via the release of TGFβ and CXCL4 respectively. Further studies are required to elucidate the role of adipocytes. Imaging studies of HSC subsets have demonstrated the existence of different bone marrow niches for platelet-myeloid-biased von Willebrand factor (VWF + HSCs and lymphoid biased (VWF − ) HSCs. In the figure, the endosteal and vascular niches (arteriolar and sinusoidal) are highlighted by changes in the colour of the background. For simplicity, only relevant HSC niche components are shown. Other structural components as different fibroblast-like cells with unknown roles as HSC niche components are not included.
T A B L E 1 Summary of the methods used to study the bone marrow niche Strategy Findings Advantages Disadvantages

References
Anatomical and histological studies Identified a variety of cell types, suggested the existence of stem cells and the importance of the marrow in blood production.
Information on tissue architecture. Limited functional and molecular information and difficulty to address the identity of specific cell types.

27-33
Transplantation of stromal components Defined the presence of a haematopoietic supportive microenvironments able to recruit and support host haematopoietic activity.
Functional information.
Lack of spatial information. Broad and non-specific side-effects affecting multiple molecular and cellular targets that complicate the interpretation of the results.

46,48-53
Expansion and depletion of candidate niche cells via genetic manipulation Explored the role of niche cells (e.g., CAR cells, megakaryocytes and macrophages) in a more specific fashion than drug treatments.
Molecular and functional information. Lack of specificity of most promoters complicates the identification of defined candidate niche cells. The use of various promoters is advised to achieve solid conclusions.
Molecular and functional information. Lack of spatial information. Promiscuous promoters target non-specific populations complicating conclusions.
The use of inducible models should be favoured over constitutive models to discern among embryonic and adult effects.

69-72
2D Imaging The use of HSC and stromal cell markers combined with genetic tracing allowed a simpler and specific identification of populations of interest.

Spatial information.
Conclusions based on limited number of cell layers. Lack of functional information. Biased by the selection of the markers used to define cell types. 8,9,41,47,69,77,79,81,82,84 3D Imaging Statistical analysis on the distribution of candidate niche cells and HSCs. Provided a comprehensive quantitative picture on the abundance of niche components.
Quantitative spatial and molecular information. Lack of functional information and limited by the number of antibodies that can be employed at one single time.
Spatial, cell behavioural and longitudinal information. Reduced molecular information and limited by the number of used markers (e.g., via genetic fluorescent labelling). 8,13,41,47,69,70,83,85,[89][90][91][92][93][94][95][96][97] Single-cell sequencing techniques Uncovered the diversity and heterogeneity of the BM and allowed to characterise the identity of different BM cells. types of blood cells. Early experiments during the atomic age demonstrated that lead-shielding the spleens of lethally irradiated mice prevented mortality. 18 Stem cells were proved later to be the critical protective factor. Till and McCulloch 19 reported that mice marrow cells injected in irradiated recipients could lead to the formation of colonies of proliferating cells with self-renewing abilities in their spleens (CFU-S). Noticing that CFU-S stem cells were less robust than the cells of the BM at reconstituting haematopoiesis in irradiated animals, Ray Schofield 20 formulated the niche hypothesis in which a stem cell is associated with other cells that determine its behaviour and fate. The identification of heterologous cells influencing stem/progenitor cells in mammals provided experimental evidence for this hypothesis. [21][22][23][24] Overall, BM stem cell niches can be defined as highly specialised and dynamic microenvironments that support HSPCs. Moreover, these niches integrate a variety of cues to efficiently respond to a plethora of insults, including infection and bleeding, to maintain tissue homeostasis throughout life. 4,5,10 This implies balancing stem cell differentiation and self-renewing decisions to generate the billions of blood cells required daily, while simultaneously preserving the stem cell population size and avoiding leukaemia development. 20

Anatomical and histological analyses: early studies
Numerous research methods have shaped our view on the role and composition of the HSC niches. Aristotle (384-322 BC) described the marrow as some 'sort of bone waste byproduct', 14 while Hippocrates (460-375 BC) and Galen  considered it the source of nutrients for the bone. 14,25 In the 18th century, Jacques-François-Marie Duverney and Charles Robin noticed that bone is formed before marrow during development and that not every adult bone harbours a marrow, leading them to consider the marrow as the vascular element of the bones. 14,25,26 In the 19th century, Ernst Neumann, Giulio Bizzozero (disciples of Rudolph Virchow) and William Osler described nucleated red blood cells, white blood cells and giant marrow cells (i.e. MKs) in the BM of humans and rabbits, [27][28][29] leading to the modern view of the BM as the site of adult haematopoiesis.
Anatomically, Xavier Bichat (1771-1802) had already described the presence of red marrow and yellow fatty marrow. 30,31 Neumann noted that in most bones the marrow changes from red marrow at birth to a yellow adult marrow and claimed that blood production is confined to the red marrow in the central bones. 32 Osler described the medulla of patients with leukaemia as similar to the 'matter in the core of an abscess' and the marrow in pernicious anaemia as comparable to the red marrow of a child. 14,29 Paul Ehrlich, using acid and basic aniline coal tar dyes, identified

References
Spatially resolved transcriptomics Locate transcriptionally profiled cells to particular areas previously dissected via laser-capture microdissection.
Molecular and spatial information from specific and defined subsets of marrow populations. Lack of information on proteomic profiles and cannot establish cell-cell interactions but only predicts them.

6,105-107
Fluorescent labelling of cells in spatial proximity Allowed labelling of neighbouring cells and the study of the cellular environment (e.g., in leukaemia, AML). Increased HSC numbers with no effect in the number of osteolineage cells, Nestin dim , CD51 + PDGFR-α + perivascular cells and Nestin high cells.
Reduction in the number of HSCs and repopulating units in the BM. Potential unspecific effects as expression of PF4 in HSCs cannot be ruled out.
Global loss of bone marrow cellularity, including lymphoid, myeloid and erythroid progenitors and HSCs. Specificity problems due to 'bystander killing'.
Loss of osteocalcin-positive osteoblasts and HSC mobilisation. Mafia transgene is also expressed in CD11b + and Ly-6G + myeloid cells.
Ablation of adipogenic and osteogenic differentiation potential of marrow cells and SCF and CXCL12 production in the BM. Reduction in HSCs number and cell size.
No effect in the numbers and function of HSCs in the BM. Nestin-Cre; Scf fl/− Scf ablation in mesenchymal stem cells.
Decreased cellularity in the bone marrow and spleen and depletion of HSCs.
Reduction in the number of HSCs and haematopoietic defects during embryonic development. This highlights the importance of using TAM-inducible Cre-ERT/Cre-ERT2 models to distinguish among embryonic and adult BM specific defects.

69
T A B L E 2 (Continued) numerous haematopoietic cell types and classified leucocytes based on the staining properties of their granules. 33 Remarkably, Neumann also proposed the presence of 'great lymphozyt stem cell' capable of both self-renewing and producing lymphocytes and erythroid cells in the BM. 34 Thus, early anatomical and histological studies demonstrated not only the presence of different anatomical types of BM but also identified a variety of cell types. Moreover, these pioneer studies cemented the importance of the marrow niche as the source for blood production and suggested the presence of stem cells.

Evaluation on the ability of niche cells to transfer haematopoietic niche activity in vivo
In the 1960s, transplantation of stromal tissues provided the first experimental evidence on the presence of a

Mouse strain Description Major findings References
Vav-Cre; Cxcl12 fl/fl Cxcl12 genetic ablation in haematopoietic cells.
No effect on the function and numbers of HSCs and HSPCs.
Modest reduction in the number of phenotypic and transplantable HSCs in the BM without increasing HSC numbers in the blood.
HSPC mobilisation and loss of B-lymphoid progenitors but normal HSC function.  haematopoietic microenvironment and on the role of niche factors. Particularly, subcutaneous marrow implantation revealed that stromal fibroblast-like precursors are able to support and reconstitute a haematopoietic microenvironment upon autologous and heterotopic transplantation. 23,24,35,36 Notably, co-transplantation of murine osteoblasts, but not dendritic cells, enhances multilineage engraftment of transplanted Lineage − haematopoietic progenitor cells in lethally irradiated mice. 37 Moreover, subcutaneous transplantation of human CD146 + multipotent subendothelial reticular BM cells into immunocompromised mice resulted in human-derived bone tissue, which was colonised by murine haematopoietic progenitor cells. 38 Similarly, foetal CD105 + Thy1.1 − skeletal bone progenitors transplanted in the adult mouse kidney capsule gave rise to donor-derived chondrocytes and recruited host haematopoietic activity, including long-term engrafting HSCs (LT-HSCs). 39 Similarly, subcutaneous and in renal capsule transplantations of matrix-embedded human and murine PDGFRα + CD51 + mesenchymal stem cells (MSC) rendered ectopic BM niches recruiting host-HSCs. 40,41 Interestingly, in the Steel-Dickie S1/S1 d mutant mice stem cell factor (SCF, also known as c-Kit ligand [KitL] and steel factor [SL]) levels are severely reduced and exhibit low HSPC numbers. 42,43 Transplantation of wild-type (wt) spleen stroma into spleens of non-irradiated S1/S1 d mice locally triggered host erythropoiesis and provided some of the first evidence on the role of SCF as a niche factor. 43

Selective expansion and depletion of candidate niche cells
Expansion and ablation of candidate niche cells via drug treatments and genetic manipulation followed by prospective evaluation on the numbers of HSC/HSPCs offered new means to functionally evaluate the role of these cells in the HSC niche. 21,22 Drug treatments

Expansion of candidate niche cells by drug treatments
Parathyroid hormone (PTH) administration in wt mice expands the number of osteoblasts, Nestin + MSCs and Lin − Sca1 + c-Kit + (LSK) HSPCs in the BM and improves survival in BM transplantation. 21,41 A primary role for the osteoblasts increasing LSK numbers was suggested; however, the multicellular effects of PTH makes it difficult to disentangle the role of those perturbed cells. Unfortunately, PTH administration after umbilical cord blood transplantation in human patients has shown no effect on blood count recovery in phase II clinical trials. 44 Conversely, in vivo administration of strontium, a bone anabolic agent, increases osteoblast number (although not N-cadherin + osteoblasts), bone volume, and trabecular thickness, but does not affect HSPC numbers. 45 Depletion of candidate niche cells by drug treatments Granulocyte colony-stimulating factor (G-CSF) administration mobilises HSPCs to the blood circulation through complex mechanisms involving various BM cell types. Importantly, G-CSF reduces chemokine ligand 12 (CXCL12) (i.e. stromal cell-derived factor-1 [SDF-1]) levels, 46 whose receptor, CXCR4, is highly expressed in HSCs. 47 The CXCL12-CXCR4 axis plays a critical role in HSC trafficking and niche retention, as illustrated by a severe reduction in BM-HSC numbers following Cxcr4 ablation in Mx1Cre-Cxcr4 flox/null mice. 47 Genetic and pharmacological disruption of the sympathetic system inhibits G-CSF-induced HSC mobilisation, 48 suggesting a regulatory role for this system on the HSC niche. Specifically, G-CSF increases the duration of sympathetic noradrenaline signals on β2and β3adrenergic receptors. In BM stromal cells, activation of these receptors downregulates CXCL12 expression promoting HSC release. 46,49,50 Additionally, G-CSF simultaneously depletes a population of endosteal macrophages (osteomacs, which normally support osteoblast) and induces G-CSF receptor-expressing BM leucocytes to suppress osteoblasts, 51,52 obscuring their specific role.
Administration of liposome-embedded clodronate, which specifically kills phagocytic cells as the only cells capable of engulfing liposomes, phenocopies G-CSF effects, including HSC mobilisation and loss of osteoblasts, supporting a role for macrophages in the HSC niche. 51 Treatments with anti-vascular endothelial (VE)-cadherin and vascular endothelial growth factor receptor 2 (VEGFR2) blocking monoclonal antibodies downregulate angiogenic NOTCH ligand expression in ECs and impair HSC engraftment in vivo, indicating a role for sinusoidal ECs in this context. 53 In contrast, zoledronate treatment, which severely decreases the number of endosteal osteoclasts, does not affect HSC mobilisation or numbers. 51 Overall, drug treatments often exhibit multiple cellular and molecular targets. Thus, it is not immediately possible to discern among direct versus indirect effects and their relevance in the BM niche. Genetic models aimed to specifically target cell types of interest are helping to elucidate the roles of particular BM-niche components.

Expansion of candidate niche cells by genetic manipulation
As with PTH treatments, expression of a constitutive active version of the PTH/PTH-related protein receptor (PPR) in osteoblasts (mCol2.3-PPR mice) increases the number of osteoblasts, which produce higher levels of NOTCH-ligand JAGGED-1 and trigger LSK expansion via NOTCH1 receptor activation, supporting a role for osteoblasts in the HSC niche. 21 Bone morphogenetic protein receptor, type IA (BMPRIA) expression inhibits osteoblastic lineage differentiation from mesenchymal progenitors. 54 Accordingly, blocking BMP signalling via Bmpr1a depletion in Mx1-Cre +/T ;Bmpr1a fl/fl mice results in increased numbers of osteoblasts and LSKs. 22 Additionally, transplantation of wt LSKs into Bmpr1adepleted recipient mice induces the expansion of wt transplanted LSKs, supporting a non-cell autonomous HSPC effect. 22 However, promiscuous expression of Mx1-Cre (e.g., in BM MSCs, Nestin + cells and perivascular cells) 16 obscures the specific cell type responsible for LSK expansion in this context.

Depletion of candidate niche cells by genetic manipulation ('suicide genes')
Other studies have investigated the role of candidate niche cells by genetically depleting osteoblasts, Nestin + MSC, CAR cells and MKs from the BM through the expression and activation of 'suicide genes' including the herpesvirus thymidine kinase (TK), the diphtheria toxin (DT) receptor (DTR) or genetically modified dimerisable Caspase genes (e.g. FK506-Fas, Mos-iCsp3) in the cells of interest (Table 2).
TK expression confers sensitivity to the initially nontoxic pro-drug ganciclovir (GCV). GCV phosphorylation by TK and subsequent phosphorylation yield triphosphate-GCV (the active metabolite), which incorporates into the DNA causing single-strand breaks and apoptosis. 55 GCV treatment of Col2.3-TK transgenic mice (TK under a osteoblast promoter [Rat-Col2.3]) results in conditional ablation of osteoblasts and a global loss of BM cellularity, including HSCs/HSPCs. 56 Importantly, not only cells expressing TK but neighbouring cells can undergo cell death by so-called 'bystander killing', 57 making this approach less specific than desired.
Macrophage Fas-induced apoptosis (Mafia) transgenic mice express FK506-FAS (a suicide fusion protein) under the c-fms 'macrophage-specific' promoter. Administration of AP20187 ligand induces dimerisation and activation of the suicide protein triggering FAS-mediated apoptosis in c-fms-expressing cells, leading to loss of osteocalcin + (Oc) osteoblasts and HSC mobilisation. 51 c-fms-Mafia is not expressed in osteoblasts, MSCs and ECs; however, CD11b + and Ly-6G + myeloid cells express it. Thus, a broader role of myeloid cells in HSC regulation cannot be formally excluded in this model. 51 DTR expression provides an analogous strategy for selective cell lineage depletion. DTR-expressing cells are sensitive to DT while wt murine cells are insensitive. 58 Different mouse lines have been genetically engineered to express DTR.
HSCs locate close to reticular CAR cells (which express high CXCL12 levels) 47 and to assess their role, CARs were ablated in Cxcl12-DTR-GFP mice (DTR knocked in the Cxcl12 locus) through DT treatment. 59 This abolished adipogenic and osteogenic differentiation potential of marrow cells and SCF and CXCL12 production in the BM and led to a marked reduction in HSC number 59 implicating adipoosteogenic CAR cells as part of the HSC niche.
In Cre-inducible DTR transgenic mouse strain (iDTR), 60 CRE mediated excision of a floxed transcriptional STOP cassette yields DTR expression. In iDTR; Cre-ERT2 double transgenic mice, combined administration of tamoxifen (TAM) and DT ablates CRE-ERT2-expressing cells. Nestin + MSCs depletion in Nes-CreERT2/iDTR mice halves the numbers of CD150 + CD48 − LSK LT-HSCs in the BM. Global BM and Lin − CD48 − cellularity were not affected in these mice, supporting a cell-specific effect of Nestin + cells on HSCs. 41 Ablation of MKs has led to conflicting results. MK ablation in Cxcl4-cre;iDTR mice yielded a substantial increase of CD105 + CD150 + HSCs and re-populating units, 61 with no effect in the BM cellularity suggesting a direct effect of MKs on HSC quiescence by CXCL4 secretion. 61 Nevertheless, MK depletion in Pf4-Cre;iDTR and Pf4-Cre;Mos-iCsp3 mice produced a significant reduction on HSCs and re-populating units in the BM. 62,63 In Pf4-Cre;Mos-iCsp3 mice, administration of AP20187 triggers Caspase-induced apoptosis via homodimerisation of iCSP3. 64 Importantly, platelet factor 4 (PF4) is reportedly expressed in HSCs, which could lead to undesired apoptosis of HSCs in Pf4-Cre;Mos-iCsp3 and Pf4-Cre;iDTR mice. 65 However, Pf4-Cre-lineage traced BM cells failed to reconstitute irradiated mice 63 questioning if HSCs express PF4.
Osteocyte depletion in dentin matrix protein-1 (DMP-1)-DTR transgenic mice yields a strain resistant to G-CSF-induced HSC mobilisation. 66 Likewise, klotho hypomorphic (kl/kl) mice, which display osteoporosis and a disrupted osteocyte network, exhibit a lack of HSC mobilisation in G-CSF treatments, supporting a role for osteocytes in regulating HSPC egress from the BM. The role of the osteocytes in the HSC niche may work indirectly through effects on osteoblasts and macrophages. 66 Overall, the lack of specificity of most promoters obscures the identification of defined candidate niche cells and advises the use of various promoters to infer consistent conclusions. Additionally, ablation of large numbers of cells in the BM may indirectly activate HSPCs to regain homeostasis. 7 Even in the event of cell-type-specific ablation, discerning if the effect on HSC numbers arises from direct or indirect perturbations on other niche components requires a detailed characterisation. 7

Inducible ablation of genes encoding niche factors
Conditional ablation of genes encoding critical niche factors such as Cxcl12 or Scf by CRE-mediated recombination in candidate niche cells followed by the evaluation of HSC numbers has been employed to unveil HSC niche components and their role in HSC regulation.
The c-Kit receptor (c-Kit)-SCF axis plays a critical role regulating quiescence 67 and self-renewal in HSCs, which express high c-Kit levels. 46,68 Conditional floxed Scf alleles (Scf fl ) allow Scf genetic deletion via CRE activity. 69 69 However, HSC numbers decrease following Scf depletion in ECs (Tie2-Cre;Scf fl/− mice) and perivascular stromal cells (Leptin receptor, Lepr-Cre;Scf fl/gfp mice), supporting the role of endothelial and Leptin + stromal cells as HSC-niche components. 69 Constitutive CRE activity can lead to haematopoietic defects during embryonic development (e.g., in Tie2-Cre; Scf fl/− embryos) precluding a proper interpretation of defects observed during adulthood. 69 Thus, the use of inducible TAM-regulated CRE activity (i.e., via Cre-ERT or Cre-ERT2) is advisable whenever possible.
As aforementioned, the CXCR4-CXCL12 axis is critical in HSC regulation.  71 Both Osx-Cre and Prx1-Cre transgenes drive CRE expression in mature osteoblasts, osteocytes and CARs but Prx1-Cre also targets CD45 − Lin − PDGFRα + SCA1 + Nestin − LepR − MSCs, which seem required for HSC and common lymphoid progenitors (CLP) maintenance. 71 Overall, Cxcl12 deletion in BM candidate niche cells suggests perivascular endothelial, LepR + stromal and Nestin − LepR − MSCs as HSC-niche components and that CLPs occupy an endosteal osteoblastic niche. 70,71 Accordingly, Cxcl12 deletion from endosteal osteoblasts (Col2.3-Cre; Cxcl12 fl/fl mice) diminishes CLP numbers and lymphoid reconstitution potential, albeit no effect on HSCs. 70 Additionally, ablation of Scf or Cxcl12 from sinusoidal (via Lepr-Cre mice), arteriolar (Ng2-Cre-ERT and Myh11-Cre-ERT2) or both (Ng2-Cre) perivascular niches show that NG2 + Nestin + arteriolar and LepR + Nestin low sinusoidal niches have a role in maintaining HSCs in the BM, and that depletion of Cxcl12 from LepR + sinusoidal cells also affects HSC location. 72 Moreover, genetic deletion of Ebf3 and Foxc1 transcription factors from all mesenchymal cells (via Prx1-Cre) or more specifically in CAR cells (Lepr-Cre mice) has revealed that these factors are essential to maintain the BM niches for HSCs. 73,74 Ebf3 and Foxc1 are expressed preferentially in CAR cells. Particularly, Ebf3 and Foxc1 (via Runx1 expression 75 ) prevent the differentiation of CAR cells into osteoblasts and adipocytes, respectively. 73,74 This supports CAR cells as specialised professional niche cells, whose specific features and identity are actively regulated. 76

Imaging techniques
Unveiling the cellular structure of the BM niche via imaging studies relies on the use of highly-specific HSC markers and of markers specific to the candidate niche cells so that colocation can be effectively assessed.

Haematopoietic stem cell markers
Initial studies showed that transplanted carboxyfluorescein succinimidyl ester (CFSE)-labelled Lineage − BM cells (a broad population of HSPCs) locate closer to the endosteum, while CFSE + Lineage + cells preferentially distribute around the central marrow. 77 Staining for LSK phenotype labels a mixed population of HSCs and HSPCs. 78 In vivo bromodeoxyuridine (BrdU) pulsing identifies more quiescent HSCs (i.e., CD45 + LSK BrdU − ), 22 which locate around N-cadherin + CD45 − osteoblastic cells in the BM endosteum. 22 The remarkable discovery of signalling lymphocyte attractant molecule (SLAM) markers with the ability to identify bona fide transplantable murine HSCs as about one in three CD150 + CD48 − LSK cells 79,80 enabled a simple antibody combination to precisely distinguish HSCs. 79 Two-dimensional (2D) microscopy, and other imaging approaches discussed below, support the presence of vascular niches where most Lin − CD48 − CD41 − CD150 + LT-HSCs locate to extraluminal perisinusoidal spaces and associate with sinusoidal endothelium and mesenchymal CAR cells 8,41,47,69,79,81,82 and a minor portion localise around arteria and arterioles vessels. 83,84 Three-dimensional (3D) whole-mount imaging Initial studies analysed HSC localisation taking into account single BM populations and lacked resolution at tissue level. 3D whole-mount imaging of optically cleared BM preparations coupled with simulations of randomly assigned positions has allowed for testing of the significance of the distribution of candidate niche cells to HSCs and compare them with a null distribution. 61,[85][86][87] 3D microscopy showed that Lineage − CD41 − CD48 − c-Kit + Sca1 + HSPCs preferentially localise in the endosteum interacting with sinusoidal and non-sinusoidal BM microvessels. 88 Importantly, imagedbased quantitative spatial analysis of BM tissues revealed that sinusoidal ECs (SECs) and CAR cells are ~30-fold more abundant than previously assumed by flow cytometry analyses. 86 This suggests that enzymatic and mechanical methods employed for tissue dissociation prior to flow cytometry analyses are not efficient in extracting every cell type, which can lead to confounding conclusions. 86 Moreover, high abundance of SEC and CAR cells makes them widely available in the BM for cell interactions. 86 Other 3D-microscopy studies have shown that CD41 + MKs are not randomly distributed to Lin − CD48 − CD41 − CD150 + HSCs in the BM sinusoids 61 and that quiescent HSCs associate with small endosteal arterioles ensheathed by NG2 + pericytes. 83 As the criteria employed to define a random distribution of dots and the methods used to test the statistical significance of cell location largely diverge among studies, this can be a source of variability and can lead to conflicting conclusions on cell interactions. 86 Intravital microscopy (IVM) Intravital microscopy combines high-resolution confocal microscopy and two-photon video imaging. It allows longitudinal in vivo studies of cellular dynamics including cell migration, division, death and cell-cell interactions. 13 IVM has exposed that LT-HSCs preferentially locate in the endosteum close to osteoblasts and near perivascular Nestin-GFP + following transplantation into immunoablated and immunocompromised mice. 8,41,[89][90][91] IVM has been particularly useful for longitudinal studies in leukaemia progression. 13 Recent studies on HSC motility in the BM combining IVM and the use of HSC genetic reporters are detailed below (see 'Haematopoietic stem cell genetic reporters').
Recently, quantitative 3D-microscopy studies have analysed the localisation of HSCs (labelled with different strategies) in relation to four simultaneous BM components in different bones. 85 In this report, α-catulin-GFP + HSCs and Mds1GFP + /Flt3Cre HSCs located close to sinusoidal CXCL12 + stromal cells and MK but not to bone, adipocyte or Schwann cells. 85 Additionally, dormant (non-dividing) HSCs, labelled as GFP-retaining c-Kit + cells in doxycycline (DOX)-chased SCL-tTA;H2BGFP mice, showed similar location to α-catulin-GFP + HSCs. 85 Importantly, HSC locations reflected the abundance of the analysed BM niche cell types rather than the presence of specific microenvironments within the analysed populations. 85 Notably, these niche cell types are much more frequent in the BM than previously assumed. 86 Remarkably, the transplantation of very large numbers of HSCs into non-myeloablated recipients strikingly demonstrated that donor HSCs are able to engraft and occupy niches distant from host HSCs without replacing host HSCs as visualised by DOX-chased Rosa26-M2-rtTA; TetOP-H2B-GFP labelled HSCs. 98,99 This further highlights the abundance of empty HSC niches available for engraftment.

Single-cell profiling of BM cells
To date, imaging techniques are still biased by preselection of antibodies and limited by how many BM niches can be simultaneously analysed. Mass cytometry, or cytometry by time of flight (CyTOF) recently unveiled 28 subsets of nonhaematopoietic cells in the BM during homeostasis. 100 This single-cell technique allows measurements of ~50 targets per cell and enables a detailed taxonomy of the BM niche; nevertheless, it is still restricted by the number and preselection of antibodies.
Broader implementation of single-cell RNA-sequencing (scRNAseq) procedures recently yielded the first transcriptional profiles of the BM at single-cell level. 2,3,6,15,101-103 scRNAseq provides an unbiased means with which to characterise BM cells with extraordinary precision. A total of 17 cellular subtypes were identified among nonhaematopoietic unfractionated cells (7AAD − Calcein AM + Ter119 − CD71 − Lin − ), comprising MSCs, osteolineages, chondrocytes, fibroblasts, ECs and pericytes. 3 Similarly, scRNAseq of VE-Cadh + endothelial, LepR + cells and COL2.3 + osteoblasts fractionated from Tom + CD45 low Ter119 low stromal cells (respectively from VECadh-Cre;LoxP-tdTomato, Lepr-Cre;LoxP-tdTomato and Col2.3-Cre;LoxP-tdTomato mice) identified two endothelial, four perivascular and three osteo-lineage clusters. 2 Within those subtypes some exhibited an HSC-regulatory gene profile (based on Scf and Cxcl12 expression) including LepR + MSCs derived osteolineage cells, fibroblasts and periendosteal ECs. 3 The detection of promiscuous expression of Lepr mRNA in multiple cell types, warns on the interpretation of data related to Lepr-Cre strains. 3 Similarly, scRNAseq data question the exact identity of classically defined Nestin + mesenchymal population. 2,3,6 Importantly, there are discrepancies in the expression pattern of the endogenous Nestin locus and marker genes. Particularly, the patterns of expression of CRE and GFP differ between Nestin-GFP and Nestin-Cre-ERT2 transgenic mice 83 and endogenous Nestin is not expressed in adult CAR/LepR + cells. 104 scRNAseq of tdTomato + mesenchymal lineage cells isolated from endosteal BM of Col2:tdTomato mice revealed a novel adipogenic Perilipin + population with key roles regulating marrow vasculature and bone formation. 103 Progressive depletion of abundant cell types in the BM (i.e., major immune populations and erythroid progenitors) followed by scRNAseq served to capture rare niche cellular components and exposed 32 cell clusters. 6 They encompass Schwann cells, smooth muscle cells, myofibroblasts, EC clusters (Sca1 + arterial and Emcn + sinusoidal ECs) and nine Pdgfra + mesenchymal populations (chondrocytes, osteoblasts, fibroblast-like populations, Ng2 + Nestin + MSCs, and two CAR clusters). 6 These two CAR populations, namely Adipo-CAR (similar to Lepr-Cre + cells) and Osteo-CAR cells, showed the highest cytokine levels among all BM cells. 6 The scRNAseq data lack spatial distribution information. Circumventing this, laser-capture microdissection coupled with sequencing (LCM-seq or spatial transcriptomics) of BM fixed sections allows assignment of cells to particular spatial locations. Particularly, Adipo-CARs preferentially locate to perisinusoidal endothelial areas, while Osteo-CARs to nonvascular regions and arteriolar endothelium. 6,105 An emerging challenge from scRNAseq databases is how to readily compare cell clusters identified by different laboratories. 102 Even more important is to establish the functional relevance of any of these novel cellular populations, which will need to rely on genetic-based approaches (e.g., genetic ablation of candidate cells).
To predict the likelihood of interaction among cells, various algorithms and databases (e.g., RNA magnet, CellPhone DB, NicheNet) have emerged based on the expression patterns of cell-surface receptors and their known surfaceexpressed ligands. 15,[105][106][107] Thus, scRNAseq data can be interrogated to expose cell-cell interactions and potential niche components.

Strategies to fluorescently label cells in cell proximity
Although progressive deletion of abundant populations enriches samples for less frequent cell populations, 6 rare populations may be missed. Particularly, rare HSCs (~20 000 total HSCs in an adult mouse 79 ) may interact with a very reduced number of bona fide niche cells, especially if HSCs show low motility. 95 Additionally, current spatial transcriptomics lack the ability to directly capture cell-cell interactions. Tackling this, a soluble lipid-permeable mCherry (sLP-mCherry) protein secreted by transduced cells and which can be absorbed by neighbouring cells, allows spatial location of the producer cells and prospective isolation and characterisation of niche cells (mCherry + ) within the bulk tissue. 108 This strategy was recently used to analyse the early niche in contact with LP-mCherry-expressing human acute myeloid leukaemia (AML) leukaemic cells xenografted in immunocompromised mice by the isolation and transcriptional profiling of mCherry + cells. 109 In this regard, Table 3 provides a summary of various changes in cellular composition in the BM niche in different malignant and non-malignant diseases and conditions. Of note, sLP-mCherry producing cells label cells in proximity but cannot distinguish between distant and direct physical interactions, and transient and stable contacts. 108,109 Unveiling the type of interactions among HSCs and niche components is likely vital to define bona fide cellular and molecular cues that regulate HSCs.

CONCLUSIONS A N D F U T U R E PE R SPEC TI V E
The development and implementation of new research techniques has dramatically changed our view of the BM from a source of nutrients for the bones to a highly specialised and complex tissue responsible for maintaining haematopoietic homeostasis.
Initial studies suggested osteoblasts as a major HSCniche component. 21,22,37 However, more recent studies based on: (I) genetic ablation of critical molecular niche factors (mostly Scf and Cxcl12) in candidate niche cells, (II) the use of more stringent HSC markers (i.e., SLAM markers, genetic HSC reporters) and (III) sophisticated imaging techniques 8,47,49,79 ; support two major HSC niches in the BM: (i) sinusoidal niches containing ECs, MKs and CAR cells and (ii) arteriolar niches encompassing ECs, NG2 + pericytes, CAR cells, sympathetic nerves and nonmyelinating Schwann cells. 1,4,6 Surprisingly, simultaneous imaging of HSCs and multiple BM components indicate that HSCs randomly localise within sinusoids, CXCL12 + stroma, and MKs. 85 Furthermore, 3D-quantitative microscopy indicates that these HSC niches are ~30-times more frequent than previously assumed. 86 Future research will determine the level of heterogeneity within the HSC niches and if some particular sub-compartments constitute specialised subniches.
In this regard, scRNAseq technologies are exposing an unappreciated cellular diversity in the BM; nevertheless, the functional relevance of most of these populations is still to be investigated. Next steps characterising the BM niche should provide a 'proteomic' perspective. Recently, proteogenomic techniques based on cellular indexing of transcriptomes and epitomes by sequencing (CITE-seq) coupled with scRNAseq have allowed mRNA and protein expression analyses at the single-cell level. 110,111 Additionally, multiplexed imaging techniques (e.g. CODEX 112 and IBEX CITEX 113 ) render multi-parameter high-resolution images in tissue sections and can help to phenotypically dissect the cellular complexity in the niche.
Unveiling the type of interactions among HSCs and niche components is critical to determine the cellular and molecular signals that regulate HSCs. Novel algorithms predicting the likelihood of cellular interactions are based on known ligand receptors but 'ignorant' for any unknown molecular interactors. Soluble lipid-permeable fluorescent proteins allow for identification of cells in cell proximity and enrichment for scarce cells that may not be captured by current scRNAseq techniques. 95,97,108,109 However, they do not provide information on the type of cellular interactions.

Experimental model-species References
Chronic myeloid leukaemia (CML) Drastic changes in the BM sinusoidal vasculature structure, increased microvascular density in CML patients. CXCL12 inhibits LSC expansion and maintains quiescence of TKI-resistant LSCs in CML. CXCL12 depletion from MSCs enhances TKI efficacy. CXCL12expressing MSCs are key for preserving TKI-resistant LSCs.
Human and mouse.

139-141
B-cell acute lymphoblastic leukaemia (B-ALL) B-ALL blasts activate osteoclasts leading to bone resorption. Increased microvessel density via VEGF production. MSCs can trigger chemoresistance in B-ALL thru VCAM-1 and NOTCH-related pathways. Increased Activin A levels inhibit CXCL12 production in MSCs. B-ALL cells are still able to migrate towards very low CXCL12 levels, while CD34 + human HSCs are not. B-ALL blasts induce MSCs to produce TGFβ promoting suppressive dendritic cells. In adult ALL, the adipocyte niche evolves during disease progression and following therapy and triggers a fate switch into quiescence in ALL cells promoting chemoresistance.
Bi-directional influence among B-ALL and BM niche supports B-ALL cells Human and mouse 142,143 Note: This table does not provide an exhaustive list on changes observed in the BM niche under each of the indicated diseases/conditions, but rather aims to illustrate the importance of the BM niche in disease by providing relevant findings in various conditions. Effects of ageing in the BM result from multiple factors. The changes observed during ageing contribute to the development of diseases more frequently observed in the elderly as MDS and AML. Thus, some changes in the BM are expected to be shared among conditions.
At this stage, only the implementation of novel unbiased methods (ideally modular and genetic) with the ability to identify and differentiate among frequent/stable cell-cell interactions versus transient and distant interactions in vivo will directly untangle the BM-niche components. These technologies will be extremely useful in revealing the microenvironments that support any other cell of interest in any tissue (including tumour chemotherapy-resistant cells). Neurobiology, with a long-standing interest in uncovering synaptic partners, has employed rabies viruses, optogenetics and split forms of GFP and CFP to investigate synaptic interactions. [114][115][116] The tropism of rabies viruses and nature of neurotransmitters make the first two approaches initially unsuitable to other tissues. However, split forms of GFP and CFP seem more amenable for a universal approach to detect cell-cell interactions. 115 In summary, four decades of intense technical development and biological studies have allowed remarkable advances in our understanding of the BM niche. The foreseeable implementation of novel approaches will identify critical factors required for HSC maintenance, selfrenewal and differentiation.

AU T HOR C ON T R I BU T ION S
Raúl Sánchez-Lanzas, Foteini Kalampalika and Miguel Ganuza wrote the manuscript.

AC K NOW L E D GE M E N T S
We thank the Centre for Haemato-Oncology at Barts Cancer Institute for critical discussions and reading of the manuscript. We are grateful to Victoria Godwin for English revision of the manuscript. This work was sup-

C ON F L IC T OF I N T E R E S T
Authors declare no conflicts of interest.

DATA AVA I L A BI L I T Y S TAT E M E N T
All the data reported here was gathered from published literature.