Integrins are non-covalent heterodimers of α and β chains. In mammals, 18 α and 8 β subunits form 24 different integrin heterodimers involved in embryonic development and maintenance of tissue homeostasis. α/β chain pairing and integrin interaction with ECM, cell surface molecules or soluble factors have been extensively reviewed in the past and will not be described in further details here (8–11).

One key property of integrins is that they can be expressed in inactive, activated or clustered state on the surface. The switch between inactive and active state results in increased ligand affinity as a consequence of inside-out or outside-in signalling. Integrin clustering further induces cytoskeleton rearrangement and enhanced cell signalling ( Figure 2 ).

Among α₄β₁, α₅β₁, α₆β₁, α₆β₄and α₉β₁ integrins that have been involved in interaction of HSC with bone marrow microenvironment (12–18), α₄β₁ is the most studied. The integrins α₄β₁ and α₅β₁ are activated by inside-out signalling that involves cytokines and divalent cations present in the bone marrow microenvironment, suggesting that they are essential for HSC retention in the bone marrow (19, 20). Accordingly, HSPC mobilization using G-CSF is correlated to decreased α₄ integrin expression (21) and deletion or inhibition of α₄β₁ integrin result in accumulation of HSC in the blood circulation (22–25). Similar results were obtained using antibody against VCAM-1, suggesting a central role of α₄β₁/VCAM-1 axis in HSC retention in the bone marrow (26). This is consistent with the finding that β₁ null HSC fail to engraft in irradiated recipient and that β₁ null HSC from chimeric embryos are unable to seed foetal liver (27, 28).

Along this line, β₇-deficient mice do not have defects in HSCs function (29), while interaction between α₄β₇ and MadCAM-1 (mucosal addressin cell adhesion molecule-1) accounts for half of the α₄-integrin mediated homing activity to the bone marrow (30, 31). Therefore, it seems that β₁ integrin heterodimers play a prominent role in bone marrow HSC retention as further supported by the fact that the dual α₉β₁/ α₄β₁ inhibitor BOP ((N-(benzenesulfonyl)-L-prolyl-L-O-(1-pyrrolidinylcarbonyl)tyrosine) induces a rapid mobilization of HSCs including those that are located in the endosteal region which bind thrombin-cleaved osteopontin with high affinity (32). This is also supported by the finding that patients treated with natalizumab, an anti-α₄ integrin antibody, present increased levels of circulating CD34⁺ progenitor cells associated with an higher migratory profile as compared to GM-CSF mobilization (33, 34).

Finally, it has recently been reported in zebrafish that VCAM-1⁺ patrolling macrophages can interact with HSCs in an α₄β₁ dependent manner and contribute to their retention in the niche (35). This study confirms earlier findings in mouse models showing that macrophages contribute to HSC retention within niches through integrin-mediated interactions (36–38).

The selectin family encompasses three members: E- (Endothelial), P- (Platelets) and L- (Leukocyte) selectins expressed by endothelial cells (E- and P- selectins), platelets (P-Selectin) and leukocytes (L-Selectin). They have been initially involved in the rolling of haematopoietic cells along vessels in flowing blood (39–41).

The minimal requirements for Ca^2+-dependent ligand binding to selectins are the tetra-saccharides Sialyl Lewis X (Sle^x) and Sialyl Lewis A (Sle^A) (42, 43). As reviewed elsewhere (44), Sle^x and Sle^A synthesis requires several enzymes including α (1–3)-fucosyltransferase activities as illustrated by defective selectin-dependent leukocyte trafficking in FucT-VII deficient mice (45). This is reminiscent of the phenotype of P-Selectin deficient mice that harbour elevated number of circulating neutrophils, loss of leukocyte rolling in mesenteric venules and delayed leukocyte recruitment in peritonitis model (46). In contrast, E-selectin deficient mice have no defect in neutrophils trafficking suggesting a compensatory mechanism mediated by P-selectin (47).

The study of double knockout mice for E- and P-selectin has revealed defect in haematopoiesis with increased extramedullary erythropoiesis and reduced haematopoietic progenitor cell homing in irradiated deficient mice upon transplantation (41, 48). However, such functions were mostly attributed to HSPC homing and it is only in 2012 that E-selectin was shown to mediate HSC proliferation at the expense of self-renewal (49). In contrast to E- and P- Selectin, early haematopoietic defects in L-Selectin-deficient mice have not been reported so far (50).

Cadherins are transmembrane glycoproteins characterized by tandemly repeated sequence motifs in their extracellular segments that allow homophilic interactions in a Ca²⁺ dependent manner (51). N-cadherin is not only expressed by neural cells but also by HSCs and spindle shaped osteoblastic cells lining the bones, called “Spindle-shaped N-cadherin⁺CD45^– Osteoblastic” (SNO) in the original publication. Because conditional inactivation of BMP receptor type IA (BMPRIA) led to expansion of both SNO and HSC, with asymmetric N-Cadherin distribution between SNO and HSC adjacent cells, it has been proposed that N-cadherin-mediated adhesion contributes to HSCs maintenance in endosteal niche (52). This concept was further supported by the fact that the knock-out of N-cadherin in LSK cells impairs long term engraftment in the bone marrow but not in the spleen (53). However, the latter demonstration used LSK cells, a compartment in which less than 20% of the cells are HSCs. Therefore, the function of N-cadherin mediated adhesion in HSC maintenance has been challenged in several studies. First, it was demonstrated that N-cadherin is not expressed on purified HSCs and that osteoblasts are dispensable for HSC maintenance (54). Second, the conditional deletion of N-cadherin in HSC using Mx1-Cre did not affect haematopoiesis, nor did its specific deletion in osteoblasts (55–57). Therefore, the controversial function of N-cadherin in HSC maintenance has been revisited in the light of the methodology used to study its function (engraftment versus knock-out) and with respect to heterogeneous expression of N-cadherin by HSC subsets (58, 59). This led to the most recent concept that N-cadherin mediated adhesion of HSC to BM stromal progenitor cells (BMSPC) may only be revealed during emergency haematopoiesis such as the one needed by “reserve” HSC to survive chemotherapy (60).

Several Ig Sf adhesion molecules such as ALCAM (CD166), ESAM, JAM-A or JAM-C are expressed by HSPCs and BM stromal or endothelial cells (61–64). Some others such as ICAM-1 or VCAM-1 are expressed in the BM microenvironment and interact with integrins expressed by HSPCs or contribute to more complex adhesive networks involving IgSf/Integrin as well as IgSf/IgSf interactions such as the JAM family members (65–68). Therefore, early haematopoietic defects reported for IgSf deficient animals have to be interpreted with caution unless specific conditional knock-out mouse models are combined with orthogonal methods such as long-term engraftment. Defects in early haematopoiesis following knockout have been reported for ALCAM, ESAM, VCAM-1, JAM-C, JAM-B and ICAM-1 ( Table 1 ).

CD44 is a class I transmembrane glycoprotein that does not belong to an adhesion molecular family and that interacts with ECM ligands such a as osteopontin, fibronectin or hyaluronan (HA). When CD44 is sialo-fucosylated and bears Sle^X glycan, it is called HCELL and interacts with E- and L-selectin (85, 86). In addition, several isoforms of CD44 are generated by alternative splicing and associated with different cellular processes (87). CD44 isoforms are widely expressed on AML cells and expression of the CD44-6v isoform has been associated with poor prognosis (88, 89). Functionally, CD44 has been involved in AML cell adhesion to bone marrow stromal cells (90, 91) and ligation of CD44 with HA or activating antibodies such as H90 has been shown to reverse differentiation blockage in AML cells (92). The same H90 activating antibody inhibited homing of AML-LSC to microenvironmental niches reducing the leukemic burden in a PDX setting. This was attributed to opposing effects of the H90 antibody which increases adhesion of normal CD34⁺CD38^- cells to HA but inhibits adhesion of CD34⁺CD38^- AML blasts to HA (93).

Overexpression of the integrins α_Mβ₂ (CD11b/Mac1), α_2, α₆ and α₄β₁ by AML cells has been associated with poor prognosis (94–96). Indeed, it has early been shown that both β₁ and β₂ integrin chains are necessary for AML blast adhesion to BM stromal cells (97).

Among the β₁ integrins, α₄β₁ seems to play the most prominent role through its interaction with fibronectin (FN) and VCAM-1. Interaction of integrin α₄β₁ with FN protects AML cells from chemotherapy and is associated with the maintenance of minimal residual disease (MRD). Treatment with a blocking antibody against α₄β₁ abrogates chemoresistance and MRD in mice (98). Similarly, integrin α₄β₁ interaction with VCAM-1 contributes to drug resistance by activating NF-κB pathway in BM stromal cells which is essential to promote chemoresistance in leukemic cells as demonstrated by inhibition of NF-kB signalling (99). This study illustrates the reciprocal crosstalk between LSC and stromal cells since NF-κB activation in stromal cells upregulates VCAM-1 which serves as a positive feedback loop for leukemic cell adhesion to stromal cells.

More recently, the interaction between the integrin α₂β₁ and collagen has been shown to confer doxorubicin chemoresistance via the inhibition of Rac-1 (100). This protective effect is reversed by anti-α₂β₁. Although these studies show the therapeutic potential of integrin inhibition in AML, they do not formally prove that LSC are more addicted to integrin-mediated adhesion than normal HSC. To find such differential adhesive cues, Ebert and collaborators have used results from pooled in vivo shRNA screens. They have found that the integrin α_vβ₃ is essential for leukemic initiation and maintenance but dispensable for normal HSPC activity (101). This was attributed to constitutive activation of Syk, a candidate therapeutic target in AML, that is phosphorylated upon engagement of surface receptors including not only α_vβ₃ integrin, but also β₂ integrins (102, 103). In summary, integrin signalling converging toward specific activation pathway such as NF-kB or Syk may represent attractive therapeutic targets.

E- and P-Selectins are constitutively expressed by bone marrow endothelial cells and play a role in HSPC rolling on micro vessels (39, 104, 105). However, they induce contrasting effects in HSPC upon interaction in vitro (86, 106–108). The study of early haematopoiesis in E-Selectin deficient mice (Sele ^-/-) has revealed that inhibition of E-Selectin in vivo increases dormancy and self-renewal of HSC (49). This is not mediated by the conventional ligands of E-Selectin since HSC isolated from mice deficient for P-selectin glycoprotein ligand-1 (Psgl-1 encoded by Selplg), HCELL (Cd44) or both do not present increased dormancy. In contrast, LSC of AML make a different selectin receptor usage that promotes AML cell survival. Indeed, leukemic cells present alterations in glycosylation which leads to expression of fucosylated ligands such as PSGL-1 (CD162) that activate PI3K/Akt survival pathway (109, 110). Even more interesting is the fact that inhibition of E-selectin interaction with its ligands using a glycomimetic stimulates proliferation of AML blast while dampening HSC cycling. Since these finding have been confirmed in preclinical mouse models, this led to the opening of phase II/III clinical trials combining inhibition of E-selectin with conventional chemotherapy in AML (NCT03616470, NCT03701308).

Most of the Ig Sf molecules expressed by normal HSC are also expressed by LSCs in AML, however only few of them allows enrichment of cells with leukemic initiating activity associated to poor prognosis. We have shown that JAM-C is expressed by a fraction of LSCs presenting high activation of Src kinase family and enriched for leukaemia initiating activity. Increased frequencies of JAM-C expressing cells identify AML patients with poor disease outcome (111). This has been confirmed in an independent study on a large cohort of AML patients (112, 113). The “CD34⁺ CD38^low CD123⁺ CD41^- JAM-C⁺” cells are enriched tenfold for LSCs as compared to cells lacking JAM-C expression within the same compartment suggesting that JAM-C may play a cell-autonomous signalling function at the transition between healthy HSC and LSC. This would be consistent with results showing that PDX or AML cell line engraftment of JAM-C-expressing cells is only partially dependent on JAM-B expression by recipient mice and with results showing that silencing JAM-C expression is sufficient to decrease Src family kinase activation (111). This could be due to promiscuous cis-interactions between JAM-C and the integrin α₄β₁ since JAM-B has been shown to bind α₄β₁ when interaction is facilitated by the simultaneous engagement with JAM-C (67).

NCAM1(CD56) is another Ig Sf molecules whose expression is correlated with poor overall survival in AML with t(8;21) (q22; q22) and highly expressed by LSC in mouse AML models using MLL-AF9 or Hoxa9-Meis1 as driver translocations (114). NCAM1 expression confers drug resistance to AML cells and knockdown of NCAM1 sensitizes blasts to genotoxic agents (115). This is likely due to constitutive activation of the MEK-ERK pathway, similar to what has been reported during neural development (116). These two examples pave the way for the use of Ig Sf molecule expression to stratify patients eligible to treatments targeting downstream signalling pathways such as Src or Mek/Erk.