CD44 and HCELL: Preventing Hematogenous Metastasis at Step 1

Abstract

Despite great strides in our knowledge of the genetic and epigenetic changes underlying malignancy, we have limited information on the molecular basis of metastasis. Over 90% of cancer deaths are caused by spread of tumor cells from a primary site to distant organs and tissues, highlighting the pressing need to define the molecular effectors of cancer metastasis. Mounting evidence suggests that circulating tumor cells home to specific tissues by hijacking the normal leukocyte trafficking mechanisms. Cancer cells characteristically express CD44, and there is increasing evidence that HCELL, a sialofucosylated glycoform of CD44, serves as the major selectin ligand on cancer cells, allowing interaction of tumor cells with endothelium, leukocytes, and platelets. Here, we review the structural biology of CD44 and of HCELL, and present current data on the function of these molecules in mediating organ-specific homing/metastasis of circulating tumor cells.

1. Introduction

Cancer deaths are generally caused by dissemination of cancer cells from a primary site to distant organs and tissues [1]. Despite intensive research into the pathobiology of cancer, our understanding of the metastatic process remains largely incomplete. Hematogenous metastasis is a complex, multi-step process in which cancer cells detach from the primary tumor, enter the blood circulation, avoid destruction by hemodynamic forces and the immune system, migrate to and extravasate into distant tissues, and, finally, seed and establish a secondary tumor [2]. Each of these sequential steps can be rate-limiting and failure at any step may prevent formation of a secondary lesion. Indeed, although millions of tumor cells may escape into the circulation every day, only a very small fraction successfully executes all the steps of the metastatic cascade and colonizes a distant organ [3–5]. The critical initiating event in target tissue infiltration involves the attachment of circulating tumor cells (CTCs) to vascular endothelium with sufficient strength to overcome the prevailing forces of blood flow. Since this “braking” process is, literally, the first step in tissue colonization, it is a focus of intense investigations aimed at yielding novel therapeutic strategies to contain and/or limit cancer dissemination.

The molecular effectors that mediate cancer metastasis are best known for their role(s) in directing normal leukocyte trafficking, such as to lymphoid organs and to sites of inflammation and tissue injury. The braking adhesive interactions of flowing cells onto the vascular endothelium consist of shear-resistant tethering and rolling of circulating cells onto the endothelial surface. This primary step is mediated principally by the selectin class of adhesion molecules. The selectins comprise a family of three lectins that bind sialofucosylated glycans on their respective ligands. One of these selectin ligands, known as HCELL (Hematopoietic Cell E/L-selectin Ligand), is a specialized sialofucosylated glycoform of CD44 that is characteristically expressed on human hematopoietic stem cells. HCELL is the most potent E-selectin and L-selectin ligand expressed on human cells [6–8]. Its “scaffold” protein, CD44, otherwise best known for serving as the principal receptor for hyaluronic acid (HA), is a multistructural and multifunctional molecule that is involved in a plethora of physiological and pathological processes, including cancer metastasis [9,10]. CD44 is prominently expressed on malignant cells; in fact, CD44 has been identified as a cancer stem cell marker, and, notably, various human malignancies express the HCELL glycoform. In this review, we focus on the structural biology of CD44 and HCELL and present recent insights into their role(s) in hematogenous metastasis.

2. The multi-step paradigm of cellular trafficking: the critical “roll” of Step 1

Recruitment of cells from the bloodstream into tissues is a highly regulated process that is central to multiple physiologic and pathologic processes, including inflammation, tissue repair, and metastasis. According to the multi-step paradigm of cellular trafficking (Fig. 1), extravasation is initiated by tethering and rolling interactions of blood-borne cells on the endothelial surface (step 1). Interaction of the rolling cells with locally produced chemokines subsequently results in integrin activation (step 2), which leads to firm adhesion (step 3) and finally, transmigration across the endothelium into the tissue (step 4).

Figure 1. The multi-step paradigm of cellular trafficking.

The critical first step for all vascular emigration is the deceleration of blood-borne cells on the endothelial surface, a process mediated by selectins and their ligands (step 1). The common minimal binding determinant for all three selectins is the tetrasaccharide sialyl-Lewis x (sLe^x). Rolling is necessary for the cells to ‘taste’ the local milieu, i.e., to interact with locally produced chemotactic stimuli. Successful chemokine signaling results in activation of integrins (step 2), which results in firm adhesion (step 3) and finally, transendothelial migration (step 4).

The critical first step for all vascular emigration is the deceleration of circulating cells on target endothelium to velocities well below the local blood flow rate. Cells first tether to the luminal side of the blood vessel wall, then start rolling and finally transition into a process called “slow rolling”. The most potent mediators of this physiologic deceleration process are the selectins, a family of three “C-type” lectins that bind to sialofucosylated proteins or lipids in a calcium-dependent manner. Though structurally related, the tissue distribution of selectins is distinct and largely non-overlapping. E-selectin (CD62E) is expressed exclusively on endothelial cells. In skin [11] and bone marrow microvasculature [12–14] its expression is constitutive. In most other tissues, however, E-selectin expression requires de novo synthesis following exposure to inflammatory stimuli, like TNF-α, IL-1β and LPS [15]. Following cytokine stimulation of cultured human umbilical vein endothelial cells (HUVECs), E-selectin surface expression typically peaks at 3–4 h and returns to baseline after 16–24 h [16].

In contrast to E-selectin, expression of P-selectin (CD62P), which can be found on both endothelial cells and platelets, does not require de novo synthesis. Though cytokines induce transcription of P-selectin RNA and subsequent protein production, P-selectin is stored in Weibel-Palade bodies of endothelial cells and in α-granules of platelets. As a result, P-selectin can be mobilized to the surface within seconds (platelets) to minutes (endothelial cells) in response to a variety of inflammatory mediators [15,17]. After 30–60 minutes, however, P-selectin is internalized again by endocytosis and a subset is recycled back into the Weibel-Palade bodies. This recycling is not observed for E-selectin, which is endocytosed as well, but is sorted from endosomes to lysosomes, where it is subsequently degraded [18]. In rodents, in addition to rapid cell surface mobilization of preformed P-selectin, de novo P-selectin synthesis can be induced by TNF-α and IL-1β with similar kinetics as seen for E-selectin [19]. Notably, on human endothelial cells, neither TNF-α nor IL-1β nor LPS can upregulate P-selectin expression [20]. However, the cytokines IL-4 and oncostatin M have been observed to induce prolonged P-selectin expression on HUVEC, an effect lasting up to 72 h [20].

L-selectin (CD62L) is well-recognized for its key function in directing lymphocyte trafficking to lymph nodes and is also involved in regulating migration of leukocytes to sites of inflammation. Apart from mediating endothelial adhesive interactions, L-selectin plays an important role in cell migration via secondary tethering [21], i.e., initiation of tethering and rolling of circulating leukocytes on each other through interaction with previously arrested leukocytes. This process allows leukocytes that do not express ligands for E- or P-selectin to home to sites of inflammation. L-selectin is expressed constitutively by hematopoietic stem cells, myeloid cells, naïve lymphocytes, central memory T cells and regulatory T cells, but effector memory T cells do not typically express L-selectin [22]. L-selectin expression is principally regulated by membrane shedding: following cell activation, membrane L-selectin levels are rapidly (within minutes) downregulated by proteolytic cleavage [23].

L-selectin predominantly mediates fast rolling (> 50 μm/s), engagement of P-selectin promotes rolling at 20–50 μm/s, and E-selectin mediates slow rolling at velocities typically below 10 μm/s [24–26]. Notably, though the selectins are the most potent mediators of step 1, several other molecules are able to mediate tethering and rolling adhesive interactions. Neutrophils sequestered in liver sinusoids typically don’t roll, but rather tether and then immediately transition to firm adhesion. This process has been shown to be independent of selectins [27] and to rely exclusively on leukocyte surface CD44 binding to endothelial HA deposits [28]. Besides CD44, integrins can support rolling (either alone [29] or in conjunction with selectins [30]) or, in some cases, can mediate rapid firm adhesion without prior rolling [31].

Rolling directly onto the endothelium is necessary for the cells to ‘taste’ the local milieu, i.e., to interact with chemotactic stimuli (principally chemokines) characteristically presented by glycosaminoglycans (GAGs) on the luminal and abluminal sides of the blood vessel [32–34]. According to the classical model of cellular trafficking, it is the expression of compatible combinations of chemokine receptors (on circulating cells) and chemokines (by target endothelium) that allows tissue-specific recruitment of selective subsets of circulating cells. Rolling cells that do not express the cognate receptor for the chemokines produced by the underlying tissue are unable to transition to firm adherence and subsequent transmigration into the tissue; for these cells, rolling is transient and they will eventually detach from the blood vessel wall and re-enter into the circulation.

Successful chemokine signaling results in a rapid (milliseconds) activation of integrins on the surface of rolling cells [35]. Chemokine-dependent inside-out signaling increases cell adhesiveness (avidity) through: (1) integrin conformational changes, leading to a high-affinity binding state; and (2) altering local integrin density, leading to increased valency [36]. Activated integrins subsequently bind to their endothelial ligands resulting in leukocyte arrest on the vascular endothelium. Integrin-dependent firm adhesion of leukocytes is mostly mediated by β₁-and β₂-integrins binding to respective endothelial ligands, typically VLA-4 (α4β1; CD49d/CD29) binding to VCAM-1, and LFA-1 (αLβ2; CD11a/CD18) binding to ICAM-1. Following firm adhesion, monocytes, neutrophils and T cells undergo β2-integrin-mediated intravascular crawling, seeking appropriate sites of transendothelial migration [32,37–39]. This crawling process, however, is not random. It has recently been shown that an intravascular chemokine gradient bound to endothelial heparan sulfate directs intraluminal crawling neutrophils toward transmigration loci [40]. The majority of emigrating cells transmigrate into the tissue via the paracellular route, i.e., through intercellular junctions. A small minority of leukocytes (4–15% in vitro [41]), however, prefers the transcellular route, i.e., through individual endothelial cells.

Though chemokine signaling imparts specificity to cellular recruitment, it is important to note that transendothelial migration may occur in absence of chemokine input. We recently reported that human mesenchymal stem cells (MSC), glycoengineered to express HCELL on their cell surface, efficiently home to and extravasate into bone marrow parenchyma [42]. The fact that the human MSCs used in these studies did not express the chemokine receptor CXCR4, which is essential for recruitment of hematopoietic stem/progenitor cells (HSPCs) to bone marrow, suggested the existence of a chemokine-independent pathway for transendothelial migration. Indeed, recent studies from our lab indicate that ligation of HCELL or CD44 (by E-selectin or HA, respectively) on the surface of human MSCs triggers VLA-4 activation and subsequent binding of to its ligands VCAM-1 and fibronectin. HCELL⁺ MSCs, but not HCELL⁻ MSCs, are able to transmigrate on E-selectin- and VCAM-1-expressing HUVEC monolayers, without chemokine signalling [43]. These findings indicate that some cell types, like human MSCs, can employ chemokine-independent effectors of cell trafficking, thereby achieving tissue-specific migration via a ‘step 2-bypass pathway’.

3. Selectin ligands

The common minimal binding determinant for all three selectins is the tetrasaccharide known as sialyl-Lewis x (sLe^x) and its isomer, sialyl-Lewis a (sLe^a). High resolution NMR spectroscopy has revealed that E-selectin binds sLe^x with the highest affinity; L- and P-selectin, respectively, display a 5- and 10-fold lower affinity [44]. Despite the low affinity of monovalent sLe^x-selectin interaction, high-avidity interactions can be generated by multivalent ligand presentation. In addition, further substitution of sLe^x with a sulfate ester on the 6-hydroxyl group of GlcNAc (i.e., 6-sulfo-sLe^x) significantly increases its affinity for L-selectin [45]. This modification is characteristic for L-selectin ligands expressed on endothelial cells such as high endothelial venules of mouse lymph nodes and human tonsils, e.g., CD34, GlyCAM-1, MadCAM-1, podocalyxin, and endomucin. Collectively, these selectin ligands are known as peripheral lymph node addressins (PNAd) [45]. The 6-sulfo-sLe^x glycotopes are typically presented on O-glycans, although they can be present on N-glycans as well [46]. On the major leukocyte ligand for P-selectin, P-selectin glycoprotein ligand-1 (PSGL-1), sulfation of specific tyrosine residues in close proximity to sLe^x-bearing glycans renders higher affinity binding to both P- and L-selectin without requirement for glycan sulfation [47,48].

Efficient selectin binding to its ligand(s) requires the carbohydrate moiety to be presented on either a protein (i.e., glycoprotein) or lipid (i.e., glycolipid) scaffold. The best characterized selectin ligands are glycoproteins that carry terminal sLe^x/a glycotopes on O- and/or N-glycans. However, despite the presence of hundreds of (glyco)proteins at the cell surface [49], only a handful function as selectin ligands. The reason for this remarkable specificity in sLe^x display is unknown. Indeed, CD44, a ubiquitously expressed protein, is decorated with sLe^x (thus, HCELL) on only a select subset of cells, and this remarkable specificity in HCELL expression is one of the most intriguing examples of CD44 pleiotropism.

4. CD44 Pleiotropism and Cancer Metastasis

CD44 is a class I integral membrane glycoprotein involved in many physiological and pathological processes. Though known best as a receptor for hyaluronic acid (HA), a linear polysaccharide composed of repeating units of N-acetylglucosamine and glucuronic acid [50], CD44 can interact with various other extracellular matrix components including chondroitin sulfate, fibrinogen, fibronectin, collagen and laminin. In addition, CD44 has been reported to bind mucosal vascular addressin, osteopontin, serglycin/gp600, and the MHC class II invariant chain [51]. The various adhesive capabilities of CD44 makes this molecule a hub for directing both cell-cell and cell-matrix interactions. This functional diversity is a reflection of the broad structural heterogeneity of CD44 which arises in part from the differential incorporation of nine alternatively spliced exons (10 in mice), and in part from a wide range of post-translational modifications, such as N- and O-glycosylation, and GAG substitutions (e.g., chondroitin sulfation). As a result, CD44 molecules range in molecular weight from 80 to over 200 kDa [10]. The smallest isoform, known as hematopoietic or standard CD44 (CD44s), is expressed on the surface of most mammalian cell types and contains none of the alternatively spliced exons. Although it has a theoretical molecular weight of ~37 kDa, it typically runs on SDS-PAGE gels at 80–95 kDa due to extensive glycosylation.

CD44 is structurally composed of an extracellular globular lectin-like domain, a membrane-proximal stem region, a transmembrane domain, and a cytoplasmic tail. Incorporation of splice peptide sequences, encoded by so-called ‘variant’ exons, occurs in the stem region. Besides encoding amino acids, some variant exons contain additional motifs for glycosylation (particularly O-glycan substitutions) and for GAG addition, each of which substantially increases protein molecular weight. For example, the entire CD44 variable region contains only four potential N-glycosylation sites, but has numerous O-glycosylation sites (e.g., 40% of the amino acids encoded by variant exon v9 are serine and threonine residues) [51]. As a result, CD44 variant isoforms (CD44v) can become quite large; one of the largest CD44 variants (CD44v3–10, also referred to as “epican”) is expressed on keratinocytes and has a molecular weight of ~230 kDa [52].

Numerous published reports have shown that CD44 is essential to many tumor cell activities. In addition, CD44 has been linked to the metastatic potential of a variety of cancers (comprehensively reviewed by Naor et al. [53]). Upregulation of CD44s and/or CD44v isoforms in malignant tissues compared to their benign counterparts has been shown for a variety of cancers. At the same time, however, contradictory observations of unchanged and even reduced CD44 expression in the same neoplastic disease have been reported as well [54–56]. Similar observations have been made regarding the association between expression of specific CD44 isoforms and tumor progression from benign to highly malignant stages. Some studies report a gradual increase in the expression of variant CD44 isoforms from early to late adenoma, and from non-metastatic to metastatic carcinoma, for example, in colon cancer [57–59] and in melanoma [60,61]. However, other researchers reported reduced levels of CD44 in the metastatic phase of colorectal cancer [62,63] or even a complete lack of correlation between disease stage and CD44 isoform expression [64,65]. These seemingly inconsistent findings are reflected in the value of CD44 expression as a prognostic marker for cancer. Not only has an unfavorable outcome been associated with either upregulation or downregulation of CD44 on tumor cells, different groups studying the same neoplastic disease have come to opposite conclusions, i.e., poor versus favorable prognosis for a specific cancer [53]. The reasons for these contradictory findings are not entirely clear, but it is believed that in many cases they are technical in nature and/or due to differences in experimental design [53]. Therefore, standardization of techniques and approaches to analyze CD44 on human cancers across laboratories will help obtain more consistent data from patient specimens.

4.1 Regulation of CD44 function by glycosylation

The activation state of CD44 is dictated by the cell in which it is expressed. Three activation states have been recognized: inactive, inducible and constitutively active [66]. Though CD44 is present on a wide variety of cell types, it is most often in an inactive form. However, cells can switch CD44 to an active binding state upon exposure to various stimuli. Resting B cells, for example, hardly bind HA despite high levels of CD44s on their cell surface. Following stimulation with IL-5, a subset of the lymphocytes binds HA to a high degree in a CD44-dependent manner. This switch to an active HA binding state requires hours to days and is associated with a decrease in molecular weight caused by N-glycosylation changes [67].

Whereas the overall N-glycosylation status of CD44 typically is inversely correlated with its capacity to bind HA (i.e., inactive CD44 is generally heavily N-glycosylated, and active CD44 tends to be poorly glycosylated [68,69]), results of studies on the effects of N-glycosylation on CD44 activity have been conflicting depending on cell type(s). In some cases (Chinese hamster ovary cells), enzymatic (PNGase F) and chemical (tunicamycin) de-N-glycosylation improves HA binding [69]; in other cases (e.g., in human lymphoma and melanoma cells), HA binding is prevented by such treatment [70]. In some cell lines (RAW 253 and L cells fibroblasts), treatment with deoxymannojirimycin (dMM), which inhibits further processing of high mannose-type N-glycans, has no effect on HA binding [68], while in other cell lines (an RAW 253 variant and T24 bladder carcinoma cells), dMM treatment enhances binding, suggesting an inhibitory effect of complex-type N-glycans [68,70].

It has been well established that neuraminidase treatment of cells or purified CD44-Ig fusion proteins increases their ability to bind HA [69,71,72]. Possibly, since both sialic acid and HA are negatively charged, sialylation of CD44 blocks HA binding via charge repulsion. Notably, peripheral blood monocytes express high levels of cell surface CD44 yet do not bind HA [73]. Upon activation with LPS, a functionally active HA-binding form of CD44 is induced by activation of a lysosomal sialidase [73–75]. Similarly, peripheral blood lymphocytes do not bind HA [73]. Following activation, however, T cells transiently switch CD44 to an active HA binding state [76]. In a mouse model for induced allergic asthma, airway accumulation of Th2 cells is critically dependent on CD44. It has recently been shown that activation of CD44 is dependent on the induction of Neu1 sialidase [77]. Although the latter is a lysosomal enzyme, it has been detected on the cell surface of both monocytes [78] and activated T cells [79]. In fact, mounting evidence suggests that desialylation is an important mechanism in many (patho)physiological processes, allowing cells to rapidly modulate the activation state of cell surface receptors and ligands [80–83].

Few studies have looked into the effect of specific glycosyltransferases on HA binding. B16 melanoma cells stably transfected to express β-1,4-N-acetylglucosaminyltransferase III (GnT-III) show an increased ability to bind HA. It has been suggested that addition of bisecting GlcNAc (via GnT-III) inhibits poly-LacNAc elongation and β-1,6-branching, resulting in fewer antennae that tend to be truncated [84,85]. CD44 from these cells is less heavily glycosylated (decreased molecular weight on SDS-PAGE), and is both hypogalactosylated and hyposialylated, again pointing towards inhibitory effects of sialylated complex-type N-glycans [86].

Human CD44 has six potential N-glycosylation sites (Asn25, Asn57, Asn100, Asn110, Asn120 and Asn255), the first five of which are located within the HA binding domain. Inactivation of individual N-glycosylation sites by site-directed mutagenesis has shown that N-glycans at Asn25 and Asn120 are involved in modulating HA binding since their inactivation converted an inducible HA binding cell line to constitutively active [71]. None of the mutations, however, had any effect on the HA binding ability of an active cell line, indicating that more subtle, cell-specific changes in N-glycosylation are sufficient to modulate HA binding by CD44. Nevertheless, elucidation of the 3D structure of the HA binding domain of CD44 has provided a mechanistic explanation of how glycans attached to Asn25 and Asn120 might regulate CD44 function. N-glycans at position 25 could directly impact HA binding by modulating CD44 affinity for its ligand; conversely, oligosacchardes attached to Asn120 might impact CD44 homo-oligomerization, as such regulating HA binding avidity [87,88].

O-glycosylation can enhance [72,89], decrease [90], or not influence binding to HA at all [68,91]. Bennett et al. showed that the reduced HA binding ability of CD44v8–v10 is due to the addition of O-glycans to the variable exons [90]. Several variably spliced CD44 domains are rich in serine and threonine residues (Ser/Thr content varies from 18% for exon v5 to 43% for exon v9), suggesting that O-glycosylation of these exons might play a role in modulating the lectin activity of CD44v isoforms.

TNF-α and IFN-γ stimulation of peripheral blood monocytes (but not T cells) increases the sulfation of CD44 and induces HA binding. A positive correlation between CD44s sulfation and HA binding has been shown, indicating that sulfation is yet another mechanism by which cells can regulate the activation state of CD44 [92]. Subsequent studies with cell lines demonstrated that TNF-α-induced sulfation occurs on both N- and O-linked glycans; at the same time, chondroitin sulfate addition was decreased [93]. More recently, it has been shown that chondroitin sulfate addition to serine 180 of CD44s negatively regulates HA binding by CD44-Ig fusion proteins [94] and mouse macrophages [95].

The seemingly contradictory results reported by different groups using different cell lines and various experimental approaches make it hard to formulate generally applicable principles for the role(s) of glycosylation in regulating CD44 activity. In addition, multiple studies have artificially altered CD44 post-translational modifications in an attempt to elucidate how they modulate HA binding. It remains unclear how such findings provide physiological insights since reagents commonly used in such studies (tunicamycin, dMM, benzyl-α-GalNAc) affect glycosylation of all surface glycoproteins and, as a consequence, potentially interfere, either directly or indirectly, with multiple physiological processes on the cell surface, not just involving CD44. In this regard, site-directed mutagenesis of potential glycosylation sites is a more elegant approach. However, this approach comes with its own limitations since the HA binding ability of CD44 mutants might be affected in an indirect way due to problems with protein folding, stability and secretion. As a result, many questions remain unanswered, and the full impact of CD44 glycan modifications in cancer metastasis remains to be established. For example, how do the glycomes of cells that have inactive, inducibly active, and constitutively active CD44 on their cell surface differ from each other? Or, similarly, what do cell types with the same CD44 activation state have in common? What array of glycans is present on CD44 from different cell types? What range of glycan structures is associated with each N-glycan site? Recent technological advances in high-throughput transcript analysis [96] and glycome profiling [97] may help identify the specific glycosyltransferases which regulate CD44 glycosylation and elucidate how the resulting oligosaccharide structures modulate HA binding.

5. Hematopoietic Cell E-/L-Selectin Ligand

The recognition of the HCELL glycoform of CD44 in studies performed in the 1990s and early 2000s added measurably to the already complex tapestry of glycosylation modifications of CD44 (reviewed in [98]). HCELL, a sialofucosylated CD44 glycoform of CD44 characteristically displayed on human HSPCs, functions as the most potent E- and L-selectin ligand found on human cells. Thus, not only is CD44 a lectin, it can also serve as a lectin ligand (i.e., a receptor for selectins). The latter function strikingly separates HCELL from CD44. Despite the fact that CD44s is modified with both O- and N-linked glycans, the sLe^x glycotopes that mediate the E- and L-selectin ligand activity of HCELL on HSPCs are displayed exclusively on N-glycans on standard CD44. It is presently unclear, though, to what extent each of the six potential N-glycosylation sites on CD44s contribute to the selectin ligand activity of HCELL, and current studies in our lab are directed to this question. Notably, in contrast to most other L-selectin ligands, binding of HCELL by this selectin is sulfation-independent and, consistent with N-glycan presentation of sLe^x, is resistant to O-sialoglycoprotease digestion [99,100].

Although CD44 is present on many cell types, expression of HCELL is natively restricted to human HSPCs and is not found on mouse HSPCs [7,101]. Conspicuously, G-CSF-mobilized peripheral blood leukocytes express high levels of HCELL [102]. HCELL has also been shown to be present at high levels on human malignant hematopoietic cells, including de novo acute myeloid leukemia (AML) cells and the AML-derived primitive human hematopoietic progenitor cell line KG1a [6,7,99]. More recent studies have shown that colon cancer cells [103–105], breast cancer cells [106] express an alternative form of HCELL. It is characterized by presentation of sialofucosylated glycotopes on O-glycans on variant CD44 isoforms. In line with the CD44 nomenclature currently in general use, HCELL activity on standard versus variant CD44 isoforms is referred to as HCELLs and HCELLv, respectively. Both HCELL isoforms differ from each other on two fundamental levels: protein backbone (CD44s versus CD44v) and glycosylation (the type of glycans that display the selectin binding determinants, i.e., N- versus O-linked oligosaccharides). As a result, both HCELL isoforms have a significantly different molecular weight: HCELLs migrates as a 90–100 kDa band on SDS-PAGE gels, while HCELLv from colon cancer cells typically has a molecular weight of about 150 kDa. Interestingly, colon cancer cells express HCELLv and CD44s, but not HCELLs [105]. Despite these differences, shear-based in vitro assays have shown that HCELLs and HCELLv are otherwise equally potent E- and L-selectin ligands [98,104].

6. CD44, HCELL, and the molecular basis of metastatic tissue tropism(s)

Some types of cancer show an organ-specific pattern of hematogenous metastasis. For breast cancer, the four leading metastatic sites are lung, liver, bone and brain [107]. Colorectal cancer commonly spreads to the liver, the lungs and, to a lesser extent, the bone [108]. In contrast, prostate tumor cells characteristically metastasize to bone, although the lungs and the liver are major target organs as well [109]. Melanoma, on the other hand, typically metastasizes to brain, lungs, and skin [110].

An issue of controversy in this regard has been whether the tissue tropism of certain cancers is caused by mechanical trapping within capillaries or by specific interactions with microvascular endothelial cells. The mechanical trapping theory, formulated by James Ewing in the 1920s, hypothesizes that circulatory patterns dictate the tissue distribution of metastases and that metastasis to distant organs essentially is a passive process [111]. Ewing formulated his theory to challenge the ‘seed and soil’ hypothesis published in 1889 by Stephen Paget, proposing that cancer cells (the seed) require a compatible microenvironment (the soil) provided by the target tissue in order to successfully form a secondary tumor [112]. In this view, metastasis is an active lodgment process that requires the expression of a compatible set of proteins by CTCs and the microvascular environment of the target distant organ. Our increasing understanding of tissue-specific patterns of metastasis based on endothelial expression of relevant effectors of cell migration has shifted attention away from the stochastic mechanical entrapment theory, pointing instead to an active homing model similar to that of leukocyte recruitment. Mounting in vitro and in vivo evidence suggests that cancer cells home to specific tissues by hijacking the normal leukocyte trafficking mechanisms by expressing selectin ligands, chemokine receptors and/or integrins on their cell surface.

Malignant transformation is associated with changes in the cellular glycosylation machinery, resulting in cell surface expression of tumor-associated carbohydrate antigens like sLe^x and sLe^a. The presence and level of sLe^x/a expression is often associated with cancer progression, metastatic spread and poor prognosis [113]. In fact, sLe^x and sLe^a were originally identified as tumor antigens [114]. As discussed earlier, the normal physiological role of these carbohydrate antigens is to mediate the earliest steps of leukocyte trafficking by interaction with their receptors, the selectins. Convincing data now suggests that, analogous to leukocyte recruitment to sites of inflammation, the observed tissue tropism of certain cancers is in part the result of CTC-endothelial cell interactions mediated by selectins and cognate selectin ligands expressed on cancer cells [5,115]. Mechanistically, tumor metastasis is facilitated by selectin-mediated interactions between cancer cells and endothelial cells, leukocytes and platelets (Fig. 2). Our emerging data hightlight the critical contributions of the step 1 effectors CD44 and HCELL in directing homing of CTCs to permissive metastatic sites.

Figure 2. The molecular basis of metastatic tissue tropism(s): CD44 and HCELL.

Tumor cells home to specific tissues by hijacking the normal leukocyte trafficking mechanisms by expressing selectin ligands on their cell surface. The principal selectin ligand on cancer cells, HCELL, allows CTCs to interact with E- and P-selectin on activated endothelium, L-selectin on leukocytes and P-selectin on platelets. Organs that do not constitutively express E-selectin, like the liver and the lungs, can be induced to do so by intravascular LPS (derived from the gut) and by pro-inflammatory cytokines (produced by the primary tumor or by cells from the immune system). In addition to E-selectin-mediated interactions, the CD44-HA axis is important for tumor metastasis to organs rich in HA, like the liver, lungs and bone marrow. Bone marrow-derived HSPCs, known to express both HCELL and L-selectin, home to pre-metastatic sites before the arrival of tumor cells and thus may play an important role in the establishment of the pre-metastatic niche.

6.1 HCELL is the principal protein scaffold for E-selectin ligands on cancer cells

To date, only a handful cancer-associated selectin ligands have been identified [103,106,116–120]. Interestingly, although PSGL-1 is one of the major selectin ligands found on leukocytes, it is often conspicuously absent on non-leukemic cancer cells [121–123]. In contrast, HCELL, which is typically only found on HSPCs, is now recognized as being one of the dominant selectin ligands on cancer cells. Leukemic blasts, particulary in acute myelogenous leukemia, express copious amounts of HCELLs. In contrast, malignant cells of solid cancers predominantly express HCELLv isoforms [103,105,106]. LS174T colon cancer cells express HCELLv isoforms composed of variant regions v3, v5, v7, v8 and v10; variant regions v4, v6 and v10 are present in a smaller percentage of the HCELLv isoforms [103]. On SDS-PAGE, these HCELLv isoforms migrate as a ~150 kDa band [103]. In contrast, metastatic MDA-MB-231 breast cancer cells express high levels of a ~170 kDa HCELLv4 isoform [106]. Upon down-regulation of HCELLv4 expression, MDB-MB-231 breast cancer cell adhesion and migration across TNF-α-activated HUVEC monolayers was significantly reduced, indicating that HCELLv serves as a major E-selectin ligand in mediating breast cancer metastasis. [106]. In both colon cancer and breast cancer cells, siRNA-mediated knock-down of CD44 significantly reduced binding to E-selectin [104,106].

On LS174T cells, the relevant sialofucosylated glycotopes of HCELLv are exclusively expressed on O-glycans. No information in this regard is available for breast cancer cells. However, based on the fact that variant CD44 isoforms are known to have a high Ser/Thr content, we suspect that O-glycans play an important, if not predominant, role in this type of cancer as well.

Accumulating evidence suggests that cancer stem cells (CSCs), which are thought to be essential for metastasis formation [124], are highly resistant to chemo- and radiotherapy, indicating that conventional anticancer strategies might fail to eradicate the cell subset that is crucial for tumorigenesis [125]. Therefore, CSC ablation represents a major therapeutic promise for clinical cancer therapy. This could potentially be achieved by targeting molecular markers specifically expressed by CSCs or expressed by both CSCs and non-CSC tumor cell populations [126]. In this regard, CD44 has been identified as a CSC marker for several human cancers, including colon cancer [127], breast cancer [128], and leukemia [129], all of which express HCELL. Since it is expressed on most normal cell types, though, therapeutic targeting of CD44 is may result in disruption of physiologic processes. The HCELLv glycoforms, on the other hand, are exclusively found on cancer cells and, therefore, represent a very promising therapeutic target. It remains to be formally demonstrated, however, whether CSCs express HCELLv.

6.2 E-selectin-dependent organotropism and metastasis

As noted above, E-selectin is constitutively expressed on bone marrow microvasculature. This explains, at least in part, the bone tropism of certain selectin ligand-expressing neoplasms, for example, breast cancer. However, with the exception of bone marrow, most organs that frequently host metastases (e.g., liver and lungs) express E-selectin only in an inducible way, typically as a result of inflammation. Nevertheless, convincing evidence supports an important role for E-selectin in metastasis to the liver and lungs. Studies in the 1990s showed that transgenic mice overexpressing E-selectin in the liver developed massive infiltrating liver tumors upon injection with sLe^a-positive B16 melanoma cells [130]. In contrast, normal mice injected with sLe^a-positive and negative B16 cells and transgenic mice injected with sLe^a-negative B16 cells developed lung tumors exclusively [130]. Inhibition of hepatic E-selectin expression has been shown to block metastasis to the liver [131]. Induction of E-selectin expression by treatment of mice with the pro-inflammatory cytokine IL-1 prior to injection with several sLe^x/a-positive colon carcinoma cell lines resulted in higher numbers of lung metastases, whereas incubation of the cancer cells with a soluble recombinant E-selectin prior to injection blocked metastasis [132]. Finally, injection of nude mice with an anti-sLe^a antibody prior to intravenous administration of colon cancer cells blocked metastasis to the lungs [133].

Endothelial beds in the liver and lungs are induced to express E-selectin by inflammatory cytokines, and it has recently become clear that inflammation is an important component of tumor progression and metastasis (see also section 6.4) [134]. Pro-inflammatory cytokines, like TNF-α and IL-1β, play an essential role in tumor progression and metastasis by upregulating the expression of E-selectin and other adhesion molecules, like VCAM-1, on endothelial cells. This is illustrated by the fact that TNF-α and IL-1β knock-out mice are resistant to liver metastasis [135–138]. Similarly, anti-TNF-α therapy strongly reduces both the number and size of liver metastases of pancreatic ductal adenocarcinoma [139]. Both TNF-α and IL-1β can be released by the primary tumor or by cells from the immune system. Prior to the arrival of CTCs, soluble factors secreted by the primary tumor prepare distant tissues for the arrival, engraftment and survival of metastatic tumor cells, a concept referred to as the pre-metastatic niche [140,141]. In fact, constitutive expression of pro-inflammatory factors (like TNF-α and IL-1β) from tumor microenvironment is a characteristic feature of several neoplastic diseases and has been demonstrated to promote metastasis [142,143]. For example, VEGF-A, TGF-β and TNF-α released by primary B16 melanoma tumors before metastasis to the lungs, have been shown to induce the expression of pro-inflammatory chemokines specifically in the lungs, which then leads to the recruitment of monocytes [144].

In addition to endogenous pro-inflammatory stimuli, exposure to LPS (lipopolysaccharide), a membrane component of gram negative bacteria, results in endothelial E-selectin surface expression. After every meal, small amounts of gut-derived LPS cross the intestinal lumen into the systemic circulation [145]. To minimize potential deleterious effects, intravascular LPS is rapidly cleared by circulating chylomicrons (i.e., large lipoprotein particles). Although chylomicron-bound LPS is rapidly inactivated by the liver [146], transient intravascular LPS results in the expression of pro-inflammatory cytokines (like TNF-α and IL-1β [147]), and itself activates endothelial cells, leukocytes and platelets (in humans) [148], and induces E-selectin and ICAM-1 expression in the liver and lungs [149,150]. In fact, it has been shown that chylomicrons alone induce the expression of E-selectin and VCAM-1on HUVEC to a similar extent as LPS [151]. This association between intestinal uptake of LPS and vascular E-selectin expression may explain why E-selectin-dependent metastases are so frequently found in the liver and the lungs, tissues that do not constitutively express E-selectin.

6.3 HA-dependent organotropism and metastasis

In addition to E-selectin-mediated interactions, the CD44-HA axis is essential to tumor metastasis to the liver, lungs and bone marrow. Hyaluronic acid is a major component of the extracellular matrix and is subject to dynamic regulation during inflammation [152]. Endothelial cells rapidly upregulate HA synthesis upon stimulation with TNF-α and IL-1β [153–155]. The liver and bone marrow, two major target organs for metastasis, are extremely rich in HA [28,156]. Neutrophils and HSPCs have been shown to depend on CD44-HA interactions for homing to liver and bone marrow [28,156]. In addition, convincing experimental evidence indicates that cancer metastasis can be inhibited in animals by targeting CD44. Transfection of the non-metastasizing rat pancreatic carcinoma cell line BSp73ASML with a CD44v4–v7 expression vector conferred metastatic potential to these cells upon injection in rats [157]. Conversely, anti-sense mediated CD44v6 downregulation decreased liver metastasis of HT29 colon cancer cells in nude mice following intrasplenic injection [158]. Multiple groups have demonstrated the ability of anti-CD44 antibodies to prevent or reduce metastasis [159,160]. Soluble versions of CD44 have been shown to efficiently inhibit hematogenous dissemination of breast cancer [161], lymphoma [162] and melanoma cells [163]. Finally, hyaluronidase injection of BALB/c mice inoculated with lymphoma cells transfected to express CD44v4–v10 prevented lymph node metastasis [164]. Taken together, these data show that, in addition to E-selectin-mediated interactions, the CD44-HA axis is contributory to tumor metastasis, particularly to liver, marrow and lung.

6.4 CTC adhesion to platelets and leukocytes

It is now well-established that inflammation plays a key role in many stages of tumor development, including metastasis [165]. Leukocytes and platelets promote growth of both primary tumors and metastatic lesions [5]. The involvement of L-selectin in this process has been demonstrated with function-blocking antibodies and L-selectin knock-out mice. L-selectin deficiency was shown to attenuate cancer metastasis, even in the absence of T- and B-lymphocytes, hinting at the involvement of granulocytes and/or monocytes in the metastatic process [166,167]. It is hypothesized that leukocytes help tumor cells transmigrate by inducing the transient upregulation of endothelial selectin ligands, thereby facilitating delivery to the (pre-)metastatic niche [167,168]. Recently, it has been shown that bone marrow-derived HSPCs home to pre-metastatic sites before the arrival of tumor cells and thus may play an important role in the establishment of the pre-metastatic niche [169]. Interestingly, in addition to HCELLs, HSPCs express L-selectin [6,7,170].

There is abundant evidence implicating platelets in the pathobiology of metastasis. Activated platelets are commonly observed in the circulation of cancer patients [171]. Platelet activation results in the release of a myriad of bioactive factors such as IL-8, IL-1β, CCL5, CD40L and histamine from the α-granules and dense granules [172], which are able to stimulate endothelial cells to upregulate cell surface expression of adhesion molecules and release chemokines to promote leukocyte recruitment and infiltration [172]. Moreover, there is substantial evidence that CTC-platelet interactions result in the formation of tumor microemboli that contribute to metastasis. Cancer cells express P-selectin ligands, including HCELLv species that bind to P-selectin [105]. In fact, HCELLv-mediated LS174T-platelet interactions have recently been shown under shear [173]. If true on primary colon cancer cells, this finding has major implications for the potential role of HCELL in cancer metastasis, since it not only allows CTCs to interact with endothelial cells but also with L-selectin-expressing leukocytes and with P-selectin-expressing platelets. Metastasis is attenuated by inhibition of P-selectin-mediated platelet-CTC interactions, either by using P-selectin knock-out mice, treatment with heparin, or selectively removing tumor cell P-selectin ligands [174,175]. It is believed that platelets form a protective cloak around CTCs. This platelet coating not only facilitates initial lodging of the resulting microemboli in blood vessels, but also protects the cancer cells from cytotoxic effector cells [175].

7. Concluding remarks

In this review we have, by design, focused on the pathobiology of step 1 effectors in metastasis. Yet, there is also evidence that step 2 effectors mediate metastatic patterns, e.g., chemokine CXCL12 interacting with CXCR4 expressed on cancer cells facilitate breast cancer dissemination to bone [176]. However, because engagement of step 2 effectors is, literally, downstream of the step 1 event, strategies to block tethering/rolling interactions would prevent chemokine-dependent processes. To this end, future studies will explore how modulating expression and/or activity of CD44 and of HCELL will impact metastatic organotropism. Beyond this important goal, the remarkable specificity in expression of HCELLv on cancer cells offers the exciting possibility that this structure could be exploited in creation of novel therapeutic reagents (e.g., HCELLv-specific mAb) to achieve cancer cell-selective cytotoxicity.

Highlights.

• HCELL, a sialofucosylated glycoform of CD44, is the most potent E-/L-selectin ligand on human cells

• Metastasis is facilitated by selectin-mediated interactions

• E-selectin facilitates metastasis to the liver, lungs and bone marrow

• HCELL is the principal protein scaffold for E-selectin ligands on cancer cells

• HCELL mediates interaction between cancer cells and endothelial cells, leukocytes and platelets

List of abbreviations

AML

acute myeloid leukemia

CD44s

standard CD44

CD44v

variant CD44

CSC

cancer stem cell

CTC

circulating tumor cell

dMM

deoxymannojirimycin

GAG

glycosaminoglycan

GnT-III

1,4-N-acetylglucosaminyltransferase III

GlcNAc

N-acetylglucosamine

HA

hyaluronic acid

HCELL

hematopoietic cell E-/L-selectin ligand

HUVEC

human umbilical vein endothelial cell

HSPC

hematopoietic stem and progenitor cell

MSC

mesenchymal stem cell

PNAd

peripheral lymph node addressin

PSGL-1

P-selectin glycoprotein ligand-1

sLe^a

sialyl-Lewis a

sLe^x

sialyl-Lewis x