Not just a marker: CD34 on human hematopoietic stem/progenitor cells dominates vascular selectin binding along with CD44

Hematopoietic stem cells (HSCs) are rare cells that are maintained throughout life (self-renewing). They produce hematopoietic progenitor cells that differentiate into every type of mature blood cell within a well-defined hierarchy. Among hematopoietic stem/progenitor cell (HSPC) markers, CD34 is well known for its unique expression on HSPCs. For this reason, it is used to enrich donor bone marrow (BM) with HSPCs prior to BM transplantation.¹  Although the role of CD34 as a marker of HSCs is under debate,^2,3  recent studies suggest the existence of a population of dormant human HSCs that are CD34 negative (CD34^neg) but become positive (CD34^pos) just prior to cell division.^4-8  Studying this negative population is challenging because a defined marker for its enrichment is still lacking and also because it demonstrates extremely poor homing and engraftment capabilities compared with its CD34^pos counterpart.^9-11  Studies of gene expression comparing lineage negative fractions of human peripheral blood HSPCs that either express the CD34 antigen or not imply that CD34^neg HSPC subsets are more kinetically and functionally dormant, whereas CD34 expression in CD34^pos HSPCs is related to cell cycle entry, metabolic activation, and HSPC mobilization and homing.^12-15  However, a detailed explanation of how CD34 contributes to CD34^pos HSPC engraftment into the BM remains unknown.

To date, the functional role of CD34 in migration has most clearly been understood in the context of recruitment of lymphocytes to specialized high endothelial venules^16-18  that line the secondary lymphoid organs. Naive T cells home to these lymphoid organs in a multistep process that involves initial tethering and rolling interactions with CD34 (along with other ligands with restricted expression to high endothelial venules, often referred to as peripheral node addressins) mediated by the L-selectin expressed on the migrating T cells.^16,17  In fact, ectopic expression of CD34 in murine T cells promoted their binding to human (but not mouse) BM stromal cells, suggesting that CD34 may bind a counterreceptor expressed on human BM endothelial cells to promote their homing.¹⁰  In support of this hypothesis, studies using CD34 knockout mice indicate that CD34 increases trafficking and migration of hematopoietic cells^11,19 ; however, the precise mechanism is still not fully understood.

Studies in both humans and mice indicate that E-selectin and P-selectin are constitutively expressed on BM endothelial cells,^20-22  and intravital studies have revealed that migration of HSPCs to BM occurs at specialized microvascular beds where E-selectin is expressed.²³  In another study, P-selectin–coated devices were shown to exhibit a sixfold enrichment of human CD34^pos HSPCs over anti-CD34 antibody-coated devices, implying the importance of P-selectin for binding HSPCs.^24,25  BM transplantation studies into lethally irradiated mice lacking both endothelial selectins revealed that these mice exhibited a substantial defect in HSPC homing and a reduced survival that was rescued following the expression of either E- or P-selectin.²⁶  These and several other independent lines of evidence have highlighted vascular-selectin–dependent interactions as central to the recruitment of HSPC to BM.^26-29  In the current study, we determine the link between CD34 expression and the concurrent hematopoietic activation that leads to its improved homing and whether these vascular selectins can explain the gap in our understanding of this process. We revealed a more defined role for CD34 as a vascular selectin ligand and showed that it has comparable affinity and functional performance to other selectin ligands on human HSPCs. These new findings add to our preexisting understanding of selectin ligand contributions toward hematopoietic cell migration in therapeutic settings.

Recombinant E-selectin chimeric protein (E-Ig) was used to immunoprecipitate ligands from a lysate of HSPC-enriched cells (as described in the supplemental Materials and methods), and the resulting samples were separated on 4% to 20% sodium dodecyl sulfate–polyacrylamide gel electrophoresis; protein bands were visualized by SimplyBlue SafeStain (Invitrogen). These bands were then reduced, alkylated, and digested in the gel with sequencing grade modified trypsin (Promega); the resultant peptides were extracted using buffer containing (5% acetonitrile [VWR], 95% water, and 0.1% formic acid [Sigma-Aldrich]). Following extraction, peptides were dried to ∼1 μL sample volumes using a speed vacuum, fractionated by nanoflow liquid chromatography, and analyzed using a LTQ Orbitrap mass spectrometer (all acquired from Thermo Scientific). Raw data were converted to Mascot generic format files and searched using the online Mascot database.

CD34 small interfering RNA (siRNA) was predesigned by Ambion silencer select (Applied Bioscience). For each nucleofection, 1 × 10⁷ cells were pretreated with bromelain (500 µg/mL, 20 minutes at 37°C), washed with phosphate-buffered saline, gently resuspended in SE buffer mix (Amaxa) containing 500 nM of CD34-targeting (referred to as CD34-knockdown [KD]) or nontargeting control (scrambled), and then pulsed with the program EO-100 using a 4D-Nucleofector system (Amaxa). CD34 expression in CD34-KD and scrambled nucleofected cells were monitored routinely after 48 to 72 hours to be eligible for experimental use. For the cell binding assay, 1 × 10⁶ CD34-KD or scrambled cells were suspended in Hanks balanced salt solution with 10 mM N-2-hydroxyethylpiperazine-N′-2-ethanesulfonic acid and 2 mM CaCl₂ and then perfused over an 80% confluent monolayer of Chinese hamster ovary cells expressing E-selectin (CHO-E) at 0.3 dyne/cm² for 60 seconds followed by stepwise increases every 15 seconds to a maximum of 4.2 dynes/cm², as previously described in Wiese et al.³⁰ 

Surface plasmon resonance (SPR) binding experiments were performed using the Biacore T-100 system (GE Healthcare) at 25°C. Immobilization of monoclonal antibodies (mAbs) on a carboxy-methylated-5 dextran sensor chip was performed using an amine coupling kit (BIAcore) as described in supplemental Materials and methods to achieve immobilization level of mAbs between 4000 and 8000 response units (RU). Binding experiments were performed at a flow rate of 20 µL/min in running buffer (1% Triton X-100, 50 mM NaCl, 50 mM Tris-base, and 1 mM CaCl₂) for E-selectin or running buffer (0.05% P20, 150 mM NaCl, 2 mM CaCl₂, 50 mM Tris-base pH 8.0) for P-selectin, unless otherwise stated. Whole cell lysates were prepared in buffer containing (1% Triton X-100 in 250 mM NaCl, 50 mM Tris-HCI; pH 7.4, 20 μg/mL phenylmethylsulfonyl fluoride, and protease inhibitor cocktail tablet [Roche]) as described in the supplemental Materials and methods. These lysates were perfused over immobilized mAb, washed with the respective running buffer for at least 3 minutes to remove multiligand complex contamination, and then a standard concentration of E-Ig (177 nM; diluted in the running buffer) was injected for 130 seconds, unless otherwise specified. To correct for the buffer bulk refractive index and nonspecific interactions with the surface and with the mAb, a control flow cell immobilized with an equal amount of the mAb-corresponding isotype was used. For more details about the surface regeneration procedure, glycoprotease experiments, and calculations used to determine the dissociation binding constant, K_D, the dissociation rate constant (k_off) of the mAb·captured-ligand complex, the apparent dissociation rate constant (k_off-apparent) of the interaction of mAb·captured-ligand with E-Ig from a single concentration-type binding study, or the apparent association rate constant from multiple concentration-type binding study (k_on-apparent), please refer to the supplemental Materials and methods.

Data are expressed as the means ± standard error. Statistical significance of differences between means was determined by 1-way analysis of variance. If means were shown to be significantly different, multiple comparisons were performed post hoc by the Tukey t test. Statistical significance was defined as P < .05.

Additional materials and methods and details of the cells, enzymes, and reagents used are outlined in supplemental Materials and methods.

Given the requisite expression of E-selectin on BM endothelial cells for HSPC migration and trafficking,²³  we sought to compare the ability of a recombinant E-selectin chimeric protein (E-Ig) to bind HSPC-enriched fractions either expressing CD34 (CD34^pos) or not (CD34^neg) that were isolated from human BM that was first depleted of lineage committed cells, including those cells expressing CD38 (Lin^neg CD38^neg). As expected, we found more CD34^neg than CD34^pos cells (15% vs 1%, respectively), which is in accordance with previous studies^5,31  (Figure 1A). Flow cytometric analysis comparing the Lin^negCD38^negCD34^pos to Lin^negCD38^negCD34^neg fractions showed higher E-Ig staining to the CD34^pos population than to the CD34^neg population (Figure 1B left panel). The positive population displayed a higher amount of E-selectin ligands (E-selLs) relative to the negative population after normalizing the slight differences in the amount of β-actin loaded as a control (Figure 1B middle panel). Note that CD34^pos-BM cells (isolated from mononuclear cells [MNCs] of human BM using only anti-CD34 microbeads alone; ie, not Lin^neg CD38^neg) express a range of E-selLs from ∼100 to 250 kDa, whereas the Lin^negCD38^negCD34^pos cells express ligands at a molecular weight (MW) just over 100 kDa primarily (Figure 1B middle panel). The protein concentration of each extract was normalized using a bicinchoninic acid assay and β-actin staining (Figure 1B middle panel). To determine whether the difference in E-Ig staining between CD34^pos and CD34^neg was due to differential expression of E-selectin protein ligands, the expression of the common E-selLs expressed on HSPCs, including CD44, PSGL-1, and CD43,^32-34  was analyzed using flow cytometry. As shown in Figure 1B right panel, CD44 was expressed on both populations. Western blot analysis of the whole-cell lysates further confirmed that although CD44 expression in the CD34^neg fraction was higher, it did not display HECA-452 reactivity (supplemental Figure 1) or E-selL activity (Figure 1B middle panel). Thus, the CD44 expressed on the CD34^neg fraction is not hematopoietic cell E-/L-selectin ligand (HCELL) because it does not display HECA-452 reactive CD44 that binds E-selectin, whereas that expressed on the CD34^pos fraction did exhibit HCELL characteristics (supplemental Figure 1).^32,35  Other known E-selLs (CD43 and PSGL-1) were expressed at low levels or absent (20% ± 4% vs 66.5% ± 24% for CD43 and 13% ± 4% vs 51% ± 11% for PSGL-1 positive cells in CD34^neg and CD34^pos populations, respectively; P < .05) (Figure 1B right panel). Overall, these results support that E-selectin binding is more pronounced on the CD34^pos fraction than on the CD34^neg fraction in part due to the differential expression of E-selLs between these 2 populations.

To fully elucidate all noncanonical E-selLs expressed on CD34^pos HSPC-enriched cells from the BM (Figure 1B middle panel), we used a mass spectrometry (MS)–based proteomics approach. Among the several hundred ligands recognized by MS with a P value < .01, our dataset verified the presence of CD44, CD43, and PSGL-1, 3 well-known HSPC E-selLs (Table 1).^32-34  Furthermore, MS data analysis identified CD34, a ligand previously identified on KG1a cells to bind E-selectin in a parallel plate flow setting,¹⁷  as a potential ligand on HSPCs. To validate the binding activity of CD34 to E-selectin, CD34 immunoprecipitates were prepared from lysates of CD34^pos fractions from normal umbilical cord blood (CD34^pos-UCB) and BM (CD34^pos-BM) as well as from acute myeloid leukemic (AML) cells (KG1a cell line and BM from AML patient [CD34^pos-AML]), all of which had been normalized for protein concentration. Western blots of these immunoprecipitates probed with E-Ig (1 µg/mL, n = 3) revealed a 120-kDa band in all samples tested, confirming that CD34 isolated from HSPC-enriched cells bound E-selectin (Figure 2A). Consistently, results from the reverse experiment verified that CD34 bound E-selectin (Figure 2B). MS analysis of CD34 immunoprecipitates from lysates of KG1a cells and CD34^pos human BM cells confirmed that the primary protein immunoprecipitated with the CD34 antibody is indeed CD34 without any contamination of other proteins (supplemental Table 1; supplemental Figure 2). Furthermore, an exhaustive immunoprecipitation with E-Ig (6 rounds) was performed on each of these cell types. Following the clearance of E-Ig reactive bands, the residual lysate was immunoprecipitated with the anti-CD34 mAb, QBend-10, to remove any leftover CD34 protein that was not removed by the E-Ig immunoprecipitations. Western blots were stained for CD34 and revealed the presence of a CD34 glycoform that was not bound by E-Ig in lysates from AML samples (Figure 2C); this glycoform was not present in lysates from normal cells. In addition, this glycoform lacked expression of sialyl Lewis x (sLe^x), which is essential for E-selectin binding (Figure 2D). Finally, confocal microscopy was used to further confirm the interaction of CD34 (CD34-mAb; red) and E-selectin (E-Ig; blue) on KG1a cells that were either pretreated with E-Ig for 1 hour or left untreated (Figure 2E). Using the Imaris Coloc analysis tool, we observed that CD34 appeared to colocalize with E-Ig (Figure 2E). Interestingly, CD34 appears to colocalize into clusters following a short exposure to E-Ig, suggesting that CD34 may be recruited to lipid rafts following exposure to E-Ig. Indeed, staining with lipid raft markers (CTB, green) indicated that E-Ig induced lipid raft formation and recruited CD34 to lipid rafts after E-Ig binding (Figure 2E). It should be noted that incubation of cells with E-Ig resulted in increased aggregation of cells (supplemental Figure 3). Furthermore, a confocal analysis following staining of CD34^posCD38^neg lineage-depleted BM cells from AML and normal BM with E-Ig (E-selectin), HECA-452 (sLe^x), and 8G12 (CD34 protein) revealed that CD34^pos AML cells from the BM (Figure 2F lower panels) express a CD34 glycoform that does not bind E-selectin or express sLe^x/a epitopes, whereas CD34^pos from normal BM (Figure 2F upper panels) counterparts do not; this is indicated by the yellow arrows in lower panels of Figure 2F. These data reinforce the data presented in Figure 2C-D, indicating that a unique glycoform of CD34 exists in AML samples that does not appear to be present in normal samples.

To examine whether native CD34 on human HSPC-enriched populations displayed functional E-selL activity in flow-based assays, we employed 3 approaches: the SPR-based binding assay developed in our laboratory,³⁶  the Stamper-Woodruff assay,³⁷  and the blot-rolling assay.^38,39  We compared CD34 binding to the well-established E-selL, CD44 (ie, HCELL).^32,35  The real-time binding feature of our SPR assay enabled us to isolate endogenously expressed CD34 and CD44 from human HSPC-enriched lysate and then measure their direct binding to recombinant E-Ig. Ligand-specific mAbs (4H11 for CD34 and Hermes-3 for CD44) or isotype controls (mouse immunoglobulin [MsIgG]) were first immobilized on a carboxy-methylated-5 sensor chip. Following the injection of the HSPC-enriched lysate, significant RU accumulated in the CD34 and CD44 flow cells, whereas only residual RUs were detected in the control flow cells that were either left blank (no mAb) or were immobilized with isotype control (MsIgG) (Figure 3A). It should be noted that the CD44 mAb used here captured a mixture of CD44 glycoforms³⁶  of which only a fraction bound E-selectin. Furthermore, the purity of the captured CD34 was assessed as described previously.³⁶  Briefly, following lysate injection, a 5-minute buffer-washing step was introduced in order to eliminate potential multiligand complexes. Captured proteins were recovered and then subjected to western blot analysis as detailed in supplemental Materials and methods. Immunostaining with CD34-mAb verified the presence of CD34 (Figure 3A inset), whereas immunostaining for the presence of PSGL-1, a highly expressed E-selL on hematopoietic cells,^33,34,36  verified its absence (Figure 3A inset). As estimated from Equation 1 in supplemental Materials and methods, we observed low ligand-capturing efficiency, likely attributed to the immobilization of the mAb rendering its binding site inaccessible, with only 9.5% and 20.2% for CD34 and CD44, respectively. The values used in these calculations for the maximum RU reached at the end of the lysate injection (RU_max) for CD34 and CD44 were 546 and 559, respectively (Figure 3A), and the apparent MWs for mAb, CD34, and CD44 were nearly 150, 120, and 80 kDa, respectively. Next, we examined the binding of CD34·4H11-mAb and CD44·Hermes-3-mAb complexes with E-selectin. Results in Figure 3A represent the raw data before subtracting the buffer bulk refractive index and residual nonspecific interactions using a flow cell with the isotype control, whereas Figure 3B shows the subtracted sensorgrams that were subsequently normalized based on the difference in the ratio of the captured CD34 and CD44 just before E-Ig injection. We observed an 8.6-fold faster dissociation rate constant (k_off) during the washing step for captured CD34 relative to CD44 from the mAb complex (Figure 3B). To assess the specificity of this interaction, we injected E-Ig (177 nM) in the presence of EDTA (5 mM) and confirmed that no interaction occurred in either complex. A marked increase in RU was detected when E-Ig was injected in the presence of Ca²⁺ (Figure 3B) with no or minimal binding to either isotype or blank controls (Figure 3A). The apparent dissociation rate constant (k_off-apparent), which consists of the dissociation of CD34 or CD34·E-Ig from 4H11-mAb as well as the dissociation of E-Ig from CD34·4H11-mAb, was 3.3 × 10⁻⁴ s⁻¹ (Figure 3B).

Immunoprecipitations of CD34 from both CD34^pos-BM cells and KG1a cells also supported the presence of adhesive interactions observed with CHO-E cells in Stamper-Woodruff assays displaying 14 ± 2 bound cells to CD34 isolated from CD34^pos-BM and 26 ± 2 bound cells to CD34 isolated from KG1a; no binding was observed in the presence of EDTA (Figure 3C). Moreover, using blot-rolling assays of HECA-452 stained blots of E-selL immunoprecipitates, we observed that CD34 supported significant rolling of CHO-E cells under physiological shear stress in addition to the well-known E-selLs (CD44, CD43, and PSGL-1) immunoprecipitated from the human KG1a cell lysate (Figure 3D). CD43 supported the least amount of CHO-E rolling of all the ligands tested. Specificity for E-selectin binding was confirmed by the elimination of binding in the presence of EDTA or by using mock-transfected CHO cells (CHO-M) (Figure 3D). Overall, these multiple binding studies revealed that CD34 is a relevant and functional E-selL that may cooperate with other E-selLs in directing human HSPCs to E-selectin expressing sites.

Next, we directly compared the binding affinities of E-selLs (CD44, CD34, CD43, and PSGL-1) expressed on KG1a cells by consecutive E-Ig injection at a physiological NaCl concentration (150 mM) using the SPR-based immunoprecipitation assay. During 5 minutes of washing at 20 µL/min, captured CD34 and CD43 continuously dissociated from their mAbs such that the amount of protein collected decreased by 20% to 25% (Figure 3B). This created an unreliable estimate of the dissociation binding constant (K_D). Therefore, the CD34 563-mAb and CD43·polyclonal-Ab complexes were covalently linked by including another run of amine coupling that comprised injecting a 1:1 mix of N-hydroxysuccinimide:1-ethyl-3-(3-dimethylaminopropyl) carbodiimide hydrochloride (NHS:EDC) for 350 seconds followed by deactivation with ethanolamine for 400 seconds at 5 µL/min. Using a steady-state model, K_D, k_off-apparent, and the apparent association rate constant (k_on-apparent) were calculated for E-selLs (Figure 4A; Table 2). We observed that CD34, CD44, and PSGL-1 bound to E-Ig similarly with K_D of 236.7 ± 38, 233 ± 9, and 259 ± 34 nM, respectively, values that are nearly half of that for CD43 (442 ± 47 nM) (Table 2). These results are consistent with the results of the blot-rolling assay, where the numbers of CHO-E cells rolling over CD43 immunoprecipitates were significantly reduced (P < .05; Figure 3D).

Assuming 1:1 stoichiometry, the percentage of active E-selLs that bound E-Ig was derived from the binding isotherms of their interactions (Figure 4A; Table 2) and was found to be 54% ± 17%, 23.0% ± 0.3%, 19% ± 2%, and 75% ± 3% for CD34, CD44, CD43, and PSGL-1, respectively. These percentages could, however, be upper estimates because we did not rule out the possibility that each ligand could bind >1 E-Ig molecule. Although E-Ig bound E-selLs with similar intrinsic binding behavior (comparable k_on-apparent and k_off-apparent), PSGL-1 and CD34 were found to express a significantly greater number of E-Ig binding sites than did CD44 or CD43, which could explain the rolling behavior of CHO-E over these ligands compared with rolling over PSGL-1 (P < .05 and P < .001, respectively; Figure 3D). These data suggest that PSGL-1 displayed the highest number of ligand active sites (P < .001 for CD44 and P < .01 for CD43) followed by CD34, CD44, and then CD43.

Further analysis of the E-selectin binding kinetics of CD34 extracted from primary cells was performed using CD34^pos-BM lysate. Because of the relatively low protein concentration in these cell lysates, limited amounts of CD34 (172 RU, data not shown) were captured compared with that from the KG1a cell lysate. However, E-Ig binding still displayed a dose-dependent response with a K_D of 261 nM (Figure 4B). After the last E-Ig injection, k_off-apparent was estimated to be 1.9 × 10⁻⁴ s⁻¹, which was very similar to that of the CD34 extracted from the KG1a lysate. Thus, k_on-apparent of 728 M⁻¹ s⁻¹ was considered a reasonable estimate for CD34. Overall, these results demonstrated relative similarities between the binding affinities of CD34 with cells isolated from either primary CD34^pos-BM or KG1a cell lysates, displaying slow on-/slow off-rate kinetics.

We used a parallel plate-rolling assay to directly measure the relative contribution of CD34 to the overall rolling behavior of HSPCs. Using a knockdown approach, we compared the ability of cells lacking CD34 (CD34-KD) and control cells (scrambled) to tether and roll over E-selectin-expressing cells. Changes in the phenotype and loss of the sLe^x epitope that are associated with native HSPCs knockdown cells³⁴  made it favorable to use the HSPC-like cell line KG1a. As measured by flow cytometry, siRNA nucleofection of KG1a cells resulted in a 50% reduction in the surface expression of CD34 relative to scrambled cells (geometric mean fluorescent intensity) (Figure 5A). Meanwhile, flow cytometry indicated that there were no changes in the expression of other E-selLs or in HECA-452 reactivity of CD34-KD cells compared with scrambled control, but a slight increase in E-Ig staining was observed in CD34-KD cells (Figure 5A). Western blots of equal amounts of loaded proteins revealed that CD34 expression was substantially reduced, correlating with a slight increase in both HECA-542 and E-Ig staining, especially at ∼100 to 150 kDa; this observation was not due to an increase in the expression of any of the measured E-selLs (Figure 5B). We observed that although the number of rolling cells after perfusing scrambled control or CD34-KD over CHO-E cell monolayers did not vary under any shear stresses tested (Figure 5C), CD34-KD cells rolled markedly faster than did the control, especially at or above 3 dyne/cm² (postcapillary venule shear; P < .05; Figure 5D). Specificity for E-selectin binding was confirmed by treatment of CHO-E cells with a function-blocking mAb (data not shown). Interestingly, transmission electron microscopic (TEM) images comparing CD34-KD and scrambled control KG1a cells revealed a profound reduction in microvilli type structures following CD34 knockdown (Figure 5E). Moreover, the cells showing this reduction in microvilli and plasma membrane convolutions did not present with any other changes in their ultrastructures: the endoplasmic reticulum was intact and not swollen; the mitochondria were comparable to those of control cells; and the nuclear envelope was intact. These results indicate that CD34 plays a significant role in slowing down the rate at which KG1a cells roll, although it does not affect the number of interactions between KG1a cells and CHO-E cells and suggests that this may be due, in part, to its role in regulating microvilli structures.

L-selectin is well known to bind sulfated sLe^x (6-sulfo-sLe^x) capped O- and N-glycans of the CD34 core protein.⁴⁰  We performed both SPR-based immunoprecipitation assay and western blot analysis to determine the glycan modifications necessary for CD34 to bind E-selectin following treatment with the glycosidases: O-sialoglycoprotein endopeptidase (OSGE; to remove O-glycans), Peptide:N-Glycosidase F (PNGase F; to remove N-glycans), or neuraminidase (to remove sialic acid). To perform a quantitative comparative analysis, we normalized the amount of mAb-captured ligand in the control to that in the treated based on their RUs prior to E-Ig injection, as described in the supplemental Materials and methods. Sensorgrams of E-Ig (354 nM) binding to antibody captured CD34 (CD34·4H11-mAb) from either neuraminidase-treated or control lysates showed that binding is eliminated following sialic acid digestion (Figure 6A). Western blot analysis of treated CD34 immunoprecipitates from KG1a or CD34^pos-UCB lysates verified SPR results (Figure 6B top panels). We evaluated the efficiency of neuraminidase to digest sialic acid using western blots and probing with the QBend-10 antibody (Class 2 mAb) to track the shift in the MW of CD34 from ∼120 kDa to 150 kDa (Figure 6B lower panels).^36,41  To determine whether the E-selectin binding epitope resides on N- and/or O-glycans, we treated the KG1a lysate with PNGase F or OSGE, respectively. We found that removing the N-glycans resulted in only a small reduction in RU_max (1.4-fold) and a similar dissociation constant (k_off-apparent) as the control lysate (Figure 6C). By contrast, O-glycan removal completely abolished the interaction of E-Ig with the CD34·4H11-mAb complex (Figure 6E); results were verified by western blot analysis (Figure 6D,F top panels). The glycoprotease-sensitive CD34-mAb, QBend-10, is sensitive to the removal of O- or N-linked glycans on CD34 and is therefore routinely used to detect the efficiency of glycan removal (Figure 6D,F lower panels).³⁶  These findings indicate that, in contrast to CD44 (ie, HCELL), which displays E-selL determinants on N-glycans and O-glycans,^32,36  CD34 presents sLe^x decorations mainly on O-glycans similar to CD43 and PSGL-1.^34,42 

E- and P-selectin are expressed on human BM endothelial cells.^20,21  To date, the only ligand known to bind all 3 selectins (E-/P- and L-selectin) on HSPCs is PSGL-1.^33,43,44  In order to determine if CD34 on HSPCs is a P-selectin ligand, recombinant P-selectin human immunoglobulin chimeric protein (P-Ig) was used to immunoprecipitate P-selectin ligands from KG1a or CD34^pos-BM lysates, and subsequently western blots were stained for CD34 (EP373Y and QBend-10-mAb) or for PSGL-1 (KPL-1-mAb). CD34 and PSGL-1 were pulled out using P-Ig, and no PSGL-1 was found within the CD34 immunoprecipitate (Figure 7A). In addition, reciprocal studies also confirmed that CD34 and PSGL-1 immunoprecipitated from lysates of KG1a and CD34^pos-BM bound P-selectin (Figure 7B). Intriguingly, 2 MW forms (between 120 kDa and 130 kDa) of CD34 appeared to be able to bind P-selectin in the HSPC-enriched lysates (Figure 7B); therefore, to determine whether E-selectin recognizes the same CD34 glycoform as P-selectin, we eluted E-selLs following E-Ig immunoprecipitation of KG1a lysate using EDTA and subsequently immunoprecipitated this elution with CD34-mAb. As shown in Figure 7C, the higher MW form of CD34 (∼130-kDa) is both an E- and a P-selectin ligand, denoting this MW glycoform of CD34 as a ligand that is able to bind both vascular selectins on HSPC-enriched cells. Furthermore, CD34 immunoprecipitated from CD34^pos-BM and KG1a cells supported CHO-P cell rolling (8 ± 2, compare with 41 ± 7 bound cells per field, respectively) using Stamper-Woodruff assays, whereas no binding was observed with the EDTA control (Figure 7D). Moreover, among the other E-selLs (CD34, CD44, CD43, and PSGL-1), only CD34 and PSGL-1 exhibited functional binding activity to CHO-P in blot-rolling assays (Figure 7E). Glycoprotease treatment of CD34 immunoprecipitates from KG1a cells revealed neither treatment with PNGase F nor with neuraminidase inhibited the binding of P-Ig to CD34, suggesting that N-glycans and sialic acid are not critical for P-selectin binding. Alternatively, OSGE treatment significantly reduced P-Ig staining, suggesting that O-glycans are essential for binding. To determine whether sulfation of CD34 is critical for P-selectin binding, we cultured KG1a cells in the presence or absence of chlorate, a metabolic inhibitor of the main sulfate donor on both tyrosine residues and glycoconjugates⁴⁵  prior to the preparation of cell lysates. As shown in Figure 7G, CD34 immunoprecipitates after chlorate treatment were not able to bind P-selectin compared with the control (left panel). Furthermore, CD34 immunoprecipitates treated with arylsulfatase, an enzyme that releases sulfates from tyrosine residues but not carbohydrates,^46,47  abrogated P-selectin binding to CD34 compared with the control (treated with buffer alone) (Figure 7G right panel). Complete cleavage of sulfated tyrosine residues was confirmed by the loss of anti-PSGL-1-mAb, KPL-1epitope (Figure 7G middle panels), which recognizes the tyrosine-sulfated motif without affecting the overall amount of PSGL-1 protein recognized by PL-2 (data not shown) or the CD34 protein level recognized by QBend-10-mAb following treatment (Figure 7D lower panels).⁴⁶  In summary, these data imply that unlike PSGL-1 binding to P-selectin, which requires sialylated glycans,^47,48  the binding of CD34 to P-selectin is not dependent on sialylation but does require O-glycans and tyrosine sulfation.

Finally, we sought to validate the binding kinetics of CD34 in comparison with PSGL-1 using our SPR-based immunoprecipitation assay. The amount of detergent and salt was increased in this binding measurement relative to those in E-Ig in order to minimize the nonspecific interactions of P-Ig with the istoype control (see supplemental Materials and methods). As shown in Figure 7H, P-selectin binding to PSGL-1 and CD34 was Ca²⁺ dependent because binding was abolished when injected in the presence of EDTA. The mean K_D of P-Ig binding to PSGL-1 was 372 ± 5 nM, whereas that of P-Ig to CD34 was 621 ± 4 nM (Figure 7H). This difference is attributed to a difference in their dissociation rates, k_off-apparent for CD34 = 3 × 10⁻⁴ ± 0.1 × 10⁻⁴ s⁻¹ and for PSGL-1 = 2 × 10⁻⁴ ± 0.3 × 10⁻⁴ s⁻¹.

For almost 30 years, the cell-surface sialomucin CD34 has been used as a marker to identify and enrich HSPCs in preparation for BM transplantation.¹  However, studies have revealed that the CD34^neg fraction of normal human BM is capable of differentiating into CD34^pos subsets that possess a more activated phenotype.^4-6,15  However, the CD34^neg fraction suffers from a profound impairment in migration after IV transplantation compared with the CD34^pos fraction.^{5,7,8,11,13,49-52}  Our flow cytometric and western blot data revealed a more pronounced E-selL activity on the CD34^pos subset of the Lin^negCD38^neg fraction than on the CD34^neg subset, suggesting that the enhanced migratory ability of CD34^pos cells is due to its possession of a specific set of functional E-selLs, required for proper migration and homing of HSPCs. In agreement with these results, a previous study illustrated that BM and fetal liver–derived CD34^pos subsets rolled over immobilized E-selectin with higher efficiency than CD34^neg subsets.⁵³  Furthermore, microarray data analysis of CD34^pos vs CD34^neg subsets revealed exclusive expression of PSGL-1 and CD43 in the positive subset, suggesting that these ligands may be responsible for mediating interactions with E-selectin in this subset.¹²  Note that the lack of the HCELL glycoform found here indicates that the higher expression of CD44 in the CD34^neg subset does not confer E-selectin binding activity, similar to that suggested previously.³²  A recent study also implied that distinct differences in other molecules critical to homing are also differentially expressed between these 2 populations of cells and may account for the inefficiencies in homing observed with CD34^neg subsets.¹³  These results also elicit the question of whether the appropriate glycosyltransferases/glycosidases and/or appropriate substrates for these enzymes are differentially expressed in these subpopulations of HSPCs. Such an understanding of this valuable population of cells will allow researchers to harness technologies^34,54-57  that improve the migration capacity of these cells with the ultimate goal of using these cells in clinical settings.

Among several E-selL candidates suggested by our MS analysis, CD34 surfaced as an attractive ligand. CD34 is a heavily sialyated O-glycosylated type 1 transmembrane glycoprotein that is negatively charged and is speculated to behave in an antiadhesive manner, much like its relative mucin CD43.^19,58,59  Using a number of biochemical and functional assays, we provide evidence that CD34 from human HSPCs binds E-selectin similar to the other well-described E-selLs, CD44 (ie, HCELL) and PSGL-1. However, unlike CD34, CD43 has been shown to contribute only modestly to the E-selectin interaction.^34,60  Here, we document individual E-selL (CD34, CD44, and PSGL-1) binding affinities expressed on human HSPCs in their native form and show that all ligands display similar dissociation binding constants (K_D) with slow on- and off-rate kinetics consistent with previously reported data from our laboratory.^36,60  Blot-rolling assay results confirmed that CD43 has a weaker binding K_D by 1.8- to 2.0-fold because of the higher rate by which CD43 dissociates from E-Ig. Furthermore, using a CD34 knockdown approach, CD34 was found to play a crucial role in slowing down the velocity of rolling cells at shear stresses ≥3 dyne/cm². Interestingly, although an increase in E-Ig staining was observed in cell lysates from CD34 knockdown cells, this did not result in an increase in the number of rolling cells. This may be explained based on the TEM imaging data of KG1a cells outlined in this study. Knockdown of CD34 conclusively showed that microvilli structures were dramatically decreased on KG1a cells, implicating the importance of CD34 and CD34 family members in microvilli formation.^18,59,61,62  Interestingly, many studies suggest that microvilli protrusions are critical for mediating interactions between KG1a cells in flow and selectins (or endothelial cells). The microvilli undergo extensions and may form tethers,^63-69  so if these structures are eliminated by the knockdown of CD34, this would likely explain the effect on cell rolling. Furthermore, these data are in agreement with previous flow-based studies that showed that CD34 isolated from lysates of KG1a cells bound CHO-E cells¹⁷  and that mouse thymocytes ectopically expressing CD34 specifically bound to human BM endothelial cells where control thymocytes not expressing CD34 did not.¹⁰ 

Both vascular selectins are required for human CD34^pos cell rolling and homing on BM microvessels, whereby defective rolling is only observed in E-/P-selectin double-knockout nonobese/severe combined immunodeficiency mice.²⁷  Indeed, P-selectin was found to significantly purify committed human HSPCs (CD34^posCD38^neg cells) from total BM MNCs.^24,25,70  Our data show that CD34 on HSPCs can function as an alternative P-selectin ligand. Analysis of the glycan requirements of CD34 binding to P-selectin underscore similar characteristic modifications of PSGL-1 binding to P-selectin with 1 key difference^47,48 : both CD34 and PSGL-1 depend on O-glycosylation and tyrosine sulfation but only PSGL-1 requires sialylation to mediate binding.⁷¹  However, the specific glycosylation profile needed for P-selectin recognition remains unknown. For example, P-selectin may bind to sialylated and nonsialylated forms of Le^x/a structures.⁷²  Also, binding of TIM-1 (T-cell immunoglobulin and mucin domain 1), a major P-selectin ligand that controls the rolling of activated T cells, requires α1-3 fucosylation and tyrosine sulfation for efficient binding but not sialylation.⁷³  On a similar note, CD24, a sialoglycoprotein highly expressed in neutrophils as well as at early stages of B-cell development, does not display the sLe^x epitope but does carry a HNK-1 sulfate-containing epitope and the O-glycans that are required for such binding.^73-75  In addition to the modifications for P-selectin binding discussed above, CD34 also stained positive for the HNK-1 epitope, suggesting that it may also be important in mediating the binding of CD34 to P-selectin (data not shown). The dissociation binding constants measured here for dimeric PSGL-1 and CD34 were 372 ± 5 and 621 ± 4 nM, respectively, which are in accordance with previous studies that reported a K_D of 320 ± 20 nM (with a k_off = 1.4 ± 0.1 s⁻¹ and k_on = 4.4 × 10⁶ M⁻¹ s⁻¹) for monomeric P-selectin binding to PSGL-1 (isolated from human neutrophils).⁷⁶  The injection of membrane-derived P-selectin, mainly composed of dimeric and oligomeric forms of P-selectin, resulted in slower on and off rates compared with monomeric P-selectin.⁷⁶  Our data therefore have a characteristically similar K_D but with a slow apparent dissociation rate constant (k_off-apparent ∼10 000-fold increase) and a slow apparent association rate constant (k_on-apparent ∼100-fold reduction). This difference in binding kinetics of monomeric and dimeric forms of P-selectin is supported by previous studies.^76,77 

Our results show that treatment of KG1a cells with E-selectin resulted in their aggregation (supplemental Figure 3), and CD34 appeared to cluster toward lipid rafts, as indicated by an enhanced CTB staining pattern and intensity following E-selectin treatment. A number of studies have suggested that, prior to activation, CD34 is homogeneously distributed over the entire cell surface of HSPCs. However, following activation of these cells (via fibronectin or ligation using anti-CD34 antibodies), CD34 redistributes to lipid rafts and in some cases is even polarized toward the uropod, suggesting a role in enhanced homotypic adhesion.^78-81  In fact, studies comparing full-length CD34, a truncated form of CD34 (where most of the cytoplasmic domain is removed), and a chimeric molecule where the cytoplasmic domain of CD34 is fused with the extracellular domain of a cytokine receptor that is not believed to have a role in adhesion, implicate the cytoplasmic domain in the mediation of signaling events causing increased adhesion.^4,79  This role of CD34 in homotypic cell adhesion is significantly abrogated through the tyrosine kinase inhibitor and integrin mAb blockers for LFA-1 and ICAM-1, suggesting a concomitant activation of the LFA-1/ICAM-1 pathway.⁷⁹  Thus, CD34 clustering following E-Ig treatment could enhance adhesion, downstream of selectin binding by activating LFA-1/ICAM-1 integrins and/or unmasking adhesiveness of integrins to one another or toward the endothelium.^19,79,82 

The CD34^pos population in AML is of special interest because the leukemic stem cells likely reside in this population.^83-85  Our work also reveals that a HECA-452 nonreactive form of CD34 that does not bind E-selectin was uniquely expressed on AML cells (CD34^pos AML cells and KG1a cells) but not on normal CD34^pos cells and could be considered a novel marker for this disease. Future studies using this glycoform of the protein as a target for the generation of mAbs may help to identify and target leukemic stem cells in the treatment of leukemia.^86,87  CD34 marker expression in AML is reported to be associated with poor prognosis and apoptosis-resistance markers.^88-90  Interestingly, CD34 fractions derived from AML samples with higher percentages of CD34 cells are more resistant to apoptosis than those derived from samples with a low CD34 percentage. In addition, apoptosis-related proteins are found to be differentially expressed among these populations,⁹¹  showing higher expression of antiapoptotic proteins such as Bcl-2 and Bcl-xL in CD34^pos vs CD34^neg subpopulations. It is appealing to consider that the special glycoform of CD34 identified in our work could be involved in mediating this anti-apoptotic behavior.

Despite CD34 being widely renowned as a marker on HSPCs, the data outlined here consider a likely function for CD34 in helping mediate the migration of HSPCs through the formation of microvilli structures and via direct interactions with vascular selectins.

The authors sincerely thank Samah Z. Gadhoum for her help and support throughout the project and Samir Hamdan for his time and fruitful discussions regarding the SPR analyses. The authors express their gratitude to Maryam Mih and Samar A. Rustum for their support in the management of the laboratory and also to Mohammad Imran Khan for his assistance with the confocal microscopy. Wei Xu at Imaging and Characterization core facility at King Abdullah University of Science and Technology (KAUST) was very helpful at providing analysis software and training on the confocal microscopes. The authors also thank Carolyn Unck from the KAUST Academic Writing Service for editing the manuscript. In addition, a special thanks to Aswini K. Panigrahi (MS assistance) and Alaguraj Dharmarajnadar (cell sorting on BD Influx) from the Bioscience Core Laboratory Facility at KAUST.

This research was supported by a KAUST Faculty Baseline Research Funding Program and by a Competitive Research Grant (CRG_R2_13_MERZ_KAUST_1) (J.S.M.).

Contribution: D.B.A. designed, performed, and analyzed experiments and wrote the manuscript; F.A.A., A.S.A.-A., H.M.J.A., and A.F.A. designed, performed, and analyzed experiments; H.M.J.A., P.B., and R.S. designed, performed, and analyzed the TEM imaging experiments; C.J.C. performed and analyzed the MS experiment; and J.S.M. conceived, designed, and analyzed experiments and wrote the manuscript.

Conflict-of-interest disclosure: The authors declare no competing financial interests.

The current affiliation for D.B.A. is Schepens Eye Research Institute, Massachusetts Eye & Ear, and Department of Ophthalmology, Harvard Medical School, Boston, MA.

Correspondence: Jasmeen S. Merzaban, Ibn Al Haytham Building, level 4, Office 4218, 4700 King Abdullah University of Science and Technology, Thuwal 23955-6900, Saudi Arabia; e-mail: jasmeen.merzaban@kaust.edu.sa.