Migration against the direction of flow is LFA-1-dependent in human hematopoietic stem and progenitor cells

Immune cell recruitment to sites of inflammation is mediated by a multistep process known as the leukocyte adhesion cascade. Initially, cells are slowed by selectin-mediated interactions, allowing for chemokine-induced activation of integrins and resulting firm adhesion, which eventually leads to transmigration across the endothelial layer (Ley et al., 2007; Springer, 1994). Hematopoietic stem and progenitor cells (HSPCs), which are a heterogeneous population of cells defined by the molecular expression of CD34 (CD34⁺), include long-term and short-term stem cells, multipotent progenitor cells and lineage-restricted myeloid and lymphoid progenitors (Carvalho et al., 2009). These cells reside in the bone marrow and give rise to all end-effector immune cells. HSPCs use the leukocyte adhesion cascade to home back to the bone marrow and re-establish normal hematopoiesis post transplantation (Appelbaum, 2007). In the first step of the adhesion cascade, human HSPCs bind and roll on all three (E-, P- and L-) selectins (Greenberg et al., 2000), but principally interact with E-selectin on the surface of the microvasculature through a specific sialyl-lewis X (sLe^X)-decorated glycoform of CD44 termed HCELL (Dimitroff et al., 2001a; Sackstein and Dimitroff, 2000). Adhesion can also be mediated by P-selectin glycoprotein ligand-1 (PSGL-1; also known as SELPLG) on the HSPC surface. Upon E-selectin-mediated braking, intracellular signaling cascades are activated in HSPCs by endothelium-bound CXCL12 through its cognate cell surface receptor CXCR4 (Peled et al., 1999b; Schweitzer et al., 1996). These cascades trigger cell-borne integrins, such as VLA-4 (α₄β₁) and LFA-1 (α_Lβ₂), to adopt high affinity states which lead to arrest on the apical surface of endothelium. Once stopped, cells can begin to migrate along surfaces presenting cellular adhesion molecules (CAMs) (Huttenlocher and Horwitz, 2011; Peled et al., 1999a), principally VCAM-1 and ICAM-1, which are distributed broadly (Schweitzer et al., 1996) but are specifically expressed at high levels on the bone marrow endothelium (Papayannopoulou et al., 1995). This tight binding and firm arrest then leads to subsequent diapedesis through the endothelium to reach the basal lamina. HSPCs must migrate through this space to reach the bone marrow (Sahin and Buitenhuis, 2012) and restart hematopoiesis.

Previous work by ourselves and others has demonstrated an interesting phenomena where both murine and human T lymphocytes can migrate efficiently upstream (against the direction of shear-flow) on surfaces presenting ICAM-1 and ICAM-2, while migrating downstream (with the direction of shear flow) on surfaces presenting VCAM-1 (Steiner et al., 2010). Other immune cell types, such as neutrophils, do not demonstrate upstream migration (Dominguez et al., 2015; Valignat et al., 2013). On surfaces presenting both VCAM-1 and ICAM-1, T cells also migrate upstream (Dominguez et al., 2015). Furthermore, the efficiency of the upstream migration is dependent on the shear rate, as increasing the shear rate results in increased upstream migration of T-cells on ICAM-1-presenting surfaces (Dominguez et al., 2015; Valignat et al., 2013). Since HSPCs have similar cell surface integrin receptor profiles to T-lymphocytes [such as VLA-4, LFA-1 and α_Xβ₂ (Chan and Watt, 2001)], it was natural to ask whether HSPCs display upstream migration on surfaces presenting ICAM-1. Like T-cells, HSPCs would have the capability to bind to both VCAM-1 and ICAM-1, and should be candidates for upstream migration due to their expression of β₂ integrins. HSPCs also express LPAM-1 (α₄β₇), which is the principal ligand for MAdCAM-1 (Berlin et al., 1993) which has been implicated as working in concert with VLA-4 in the homing of HSPCs to the bone marrow in mice (Katayama et al., 2004).

Our study sought to determine whether upstream migration under fluid flow could be seen in HSPCs. We used both a human cell line, KG1a cells, which have been previously used as a model for HSPCs (Dimitroff et al., 2001a,b; She et al., 2012), and primary bone marrow-derived HSPCs. Our results demonstrate that both KG1a cells and HSPCs migrate upstream in an LFA-1-dependent manner on surfaces presenting ICAM-1 or MAdCAM-1, and mixed surfaces containing ICAM-1, as well as on activated endothelial monolayers. These results imply that LFA-1 engagement is indispensable for the upstream direction of stem cells, which facilitates HSPC migration to specific sites to enter the bone marrow.

The KG1a cell line is a human hematopoietic cell line that expresses similar numbers of the surface marker CD34 to bone marrow-derived HSPCs (Fig. S1), making them an excellent model cell line for HSPCs. We performed flow cytometry to determine what cell-borne receptors for CAMs are expressed on the surface of KG1a cells and bone marrow HSPCs (Fig. 1). Flow cytometry revealed that KG1a cells express the α_L, α₄ and α₅ integrin subunits at high levels, have moderate expression of α_X, and show no expression of α_M (Fig. 1A). In addition, KG1a cells expressed high levels of β₁ and β₂ integrin subunits and low expression of β₇ (Fig. 1B). The primary bone marrow HSPCs expressed all of the α (Fig. 1C) and β integrin subunits (Fig. 1D), albeit at lower levels than KG1a cells. Taken together, KG1a cells and HSPCs express the integrin heterodimers LFA-1 (α_Lβ₂), α_Xβ₂, VLA-4 (α₄β₁), VLA-5 (α₅β₁) and LPAM-1 (α₄β₇), while being negative for MAC-1 (α_Mβ₂). Expression of these cell surface integrins suggests that KG1a cells have the ability to migrate on ICAM-1, VCAM-1 and MAdCAM-1 surfaces.

We measured the motility of KG1a cells on substrates coated with endothelial CAMs in the absence of flow at a fixed concentration of 2.5 µg/ml. In the traces presented in the figures tracking the path of migration, blue represents rightward migration (with the flow) and red indicates leftward migration (against the flow). Fig. 2 illustrates the motion of KG1a cells on VCAM-1 (Fig. 2A), ICAM-1 (Fig. 2B) and MAdCAM-1 (Fig. 2C) surfaces. The migration index (MI) fell between −0.05 and 0.05 for all three CAMs, indicating that in the absence of flow, there is no preference in direction of migration. The two-dimensional area KG1a cells explored through migration in the same period of time was different on each CAM surface and followed the trend VCAM-1<ICAM-1<MAdCAM-1. The persistence time (Fig. 2D) of KG1a cells on ICAM-1 (6.79±1.36 min; mean±s.e.m.) was slightly less than that of VCAM-1 (8.64±1.51 min), but both were significantly lower than on MAdCAM-1 (15.2±2.48 min). The speed (Fig. 2E) of KG1a cells on VCAM-1 (4.71±0.37 µm/min) was significantly slower than on ICAM-1 (7.44±1.18 µm/min), which was in turn significantly slower than on MAdCAM-1 (10.5±1.21 µm/min). The random motility coefficient (Fig. 2F), which is an overall measure of the motility, was also lower on VCAM-1 (90.4±8.1 µm²/min), than on ICAM-1 (175.1±29.4 µm²/min), which was less than on MAdCAM-1 (589.4±91.8 µm²/min). Thus, KG1a cells are motile on each CAM surface.

Next, we studied the motility of KG1a cells on ICAM-1, VCAM-1 and MAdCAM-1 under shear flow at both a lower shear rate (100 s⁻¹) and a higher shear rate (800 s⁻¹) (Fig. 3). On ICAM-1 surfaces, KG1a cells have a slight preference for upstream migration at low shear rates (Fig. 3A, middle) and a strong preference for upstream migration at high shear rates (Fig. 3A, right; Movie 1). The directional preference was quantified through the MI (Fig. 3D), which decreases (with more negative values indicating more persistent upstream motion) with increasing shear rate from −0.18±0.01 at 100 s⁻¹ to −0.37±0.04 at 800 s⁻¹ (mean±s.e.m.). On VCAM-1 surfaces, KG1a cells have a strong preference for downstream migration at both low (Fig. 3B, middle) and high shear rates (Fig. 3B, right; Movie 2). This preference is also seen in the MI (Fig. 3D), which increases with increasing shear rate from 0.6±0.03 at 100 s⁻¹ to 0.71±0.02 at 800 s⁻¹. Taken together, these data demonstrate that KG1a cells migrate more avidly against direction of flow with increasing shear rate on ICAM-1 surfaces, while migrating with the direction of flow on VCAM-1 surfaces. This phenotype has only been previously seen with T lymphocytes, so this observation has now been extended to an important new cell type.

MAdCAM-1 has not been previously studied for its propensity to direct directional migration under flow. On MAdCAM-1 surfaces, the cell tracks illustrate that the KG1a cells have a bi-directional motion at both low (Fig. 3C, middle) and high (Fig. 3C, right; Movie 3) shear rates. This bi-directional motion is caused by a distinct subpopulation of cells that migrate upstream, while the majority of cells migrate downstream. The MI (Fig. 3D) only slightly increases with increasing shear rate, from 0.22±0.05 at 100 s⁻¹ and to 0.29±0.05 at 800 s⁻¹. For all three CAMs, the speed and persistence time of migration showed no dependence on shear rate (Fig. S2).

To understand the role of integrins in the directional motion of KG1a cells under shear flow, we measured both the directional motion and amount of adhesion of KG1a cells on ICAM-1, VCAM-1 and MAdCAM-1 surfaces under high shear rates (800 s⁻¹) in the presence of blocking antibodies against specific cell-borne integrins (Fig. 4).

In the presence of an isotype control antibody, KG1a cells crawl upstream under shear flow on ICAM-1 surfaces (Fig. 4A, left; Movie 4). The binding to ICAM-1 was shown to be β₂ integrin dependent, as an antibody blocking the CD18 subunit reduced adhesion from 73.2±8.5 cells/mm² (for the isotype positive control; mean±s.e.m.) to 1.1±0.3 cells/mm² (Fig. 4D). By using antibodies, we further investigated the role of different β₂ integrin subunits in adhesion and migration under flow. With an antibody that blocks the CD11a subunit (α_L, LFA-1), migration switches from upstream to downstream (Fig. 4A, middle; Movie 5), while blocking the CD11c subunit (α_X) preserved upstream migration (Fig. 4A, right). Furthermore, while blocking CD11a produced only a marginal reduction in adhesion (62.2±10.9 cells/mm²), blocking the CD11c subunit almost completely removed adhesion to ICAM-1 (Fig. 4D). The reversal of migration through blocking CD11a was quantified by determining the MI (Fig. 4G), which switches from a negative value of −0.25±0.04 to a positive value of 0.33±0.03. Blocking CD11c does not affect the MI (−0.27±0.08). The MI could not be calculated when CD18 was blocked because so few cells bound to the surface. In summary, LFA-1 controls the upstream motion of KG1a cells on ICAM-1 while the α_Xβ₂ integrin is important for adhesion to the surface.

As seen from the cell tracks (Fig. 4B) and MI (Fig. 4H) on VCAM-1 surfaces, KG1a cells migrate downstream (left plot) in the isotype control and the direction of motion was unchanged for all blocking conditions. Binding to VCAM-1 surfaces was shown to be mostly β₁ integrin dependent, with a partial role for β₇ integrins, as blocking β₁ (CD29) reduced adhesion by ∼85% while blocking β₁ and β₇ in combination completely abolished adhesion. In addition, blocking the α₄ chain (CD49d) reduced adhesion ∼95%; this antibody would block the α chains of both VLA-4 (α₄β₁) and LPAM-1 (α₄β₇) (Fig. 4E). The MI (Fig. 4H) could not be calculated when β₁ and β₇ together or α₄ were blocked because so few cells were bound. Taken together, the data support the idea that VLA-4 controls the majority of adhesion of KG1a cells to VCAM-1 with the remainder controlled by LPAM-1.

KG1a cells retain their unique bi-directional motion on MAdCAM-1 surfaces upon blocking with an isotype control antibody (Fig. 4C, left). Blocking of L-selectin (CD62L, also known as SELL), a ligand for MAdCAM-1 which is expressed by KG1a cells (data not shown), did not affect either the adhesion to the surface (Fig. 4F) or the MI (Fig. 4I). Adhesion to MAdCAM-1 was β₇ integrin dependent, as blocking LPAM-1 abolished adhesion (Fig. 4F), preventing the calculation of the MI (Fig. 4I). Interestingly, blocking the CD11a chain (LFA-1) did not affect the adhesion to MAdCAM-1 (Fig. 4F) but instead changed the migration from bi-directional to completely downstream (Fig. 4C, middle and right panels, respectively). This change was also reflected in the MI (Fig. 4I), which increased from 0.27±0.02 to 0.74±0.04 upon blocking CD11a at a shear rate of 800 s⁻¹. Taken together, the data suggest that while LPAM-1 controls the adhesion, LFA-1 exclusively controls the upstream motion of KG1a cells on MAdCAM-1.

To determine whether KG1a cells were able to migrate upstream when more than one CAM was present (as the cell would encounter on the endothelial surface), we plated KG1a cells on 50:50 mixtures of both VCAM-1 and ICAM-1 (V₅₀+I₅₀), and ICAM-1 and MAdCAM-1 (I₅₀+M₅₀). As seen in the cell tracks (Fig. 5A,B), upstream migration was seen on both V₅₀+I₅₀ and I₅₀+M₅₀ surfaces, but with quantitative differences. The MI (Fig. 5C) and the percentage of cells migrating upstream (Fig. 5D) quantifies these differences, indicating that KG1a cells more readily migrate upstream on I₅₀+M₅₀ surfaces than on V₅₀+I₅₀ surfaces. At a shear rate of 800 s⁻¹, the overall MI is upstream on I₅₀+M₅₀ surfaces (−0.14) with 64% of cells migrating upstream, and downstream on V₅₀+I₅₀ surfaces (0.33) with only 32% of cells migrating upstream. Next, we demonstrated that blocking the α_L-integrin (CD11a) eliminates upstream migration on mixed surfaces, as illustrated by the lack of red traces after blocking CD11a on either V₅₀+I₅₀ and I₅₀+M₅₀ surfaces (Fig. 5A,B) and the corresponding increase in MI (Fig. 5C) from 0.33 to 0.62 on V₅₀+I₅₀ surfaces and from −0.14 to 0.32 on I₅₀+M₅₀ surfaces. The proportion of cells migrating upstream (Fig. 5D) was also reduced to 6.6% and 17.7% on V₅₀+I₅₀ surfaces and I₅₀+M₅₀ surfaces, respectively. Finally, we demonstrated that blocking the α₄ integrin (CD49d) enhances upstream migration on mixed surfaces, as illustrated by the abundance of red traces after blocking on either V₅₀+I₅₀ and I₅₀+M₅₀ surfaces (Fig. 5A,B) and the subsequent decrease in MI (Fig. 5C) from 0.33 to −0.28 on V₅₀+I₅₀ surfaces (with net migration switching from downstream to upstream) and from −0.14 to −0.33 on I₅₀+M₅₀ surfaces. When blocking the α₄ integrin, the proportion of cells migrating upstream (Fig. 5D) increased to 75.6% and 82.9% on V₅₀+I₅₀ and I₅₀+M₅₀ surfaces, respectively.

We next varied the ratio of VCAM-1 to ICAM-1 (Fig. 5E) or ICAM-1 to MAdCAM-1 (Fig. 5F) while maintaining a constant total concentration of adhesion molecules at 5 μg/ml. This was done to determine which CAM controlled the directional preference cell migration. On surfaces made from VCAM-1 and ICAM-1 (V_x+I_x, where x is the proportion of each molecule as a percentage), KG1a cells prefer to migrate downstream when even 10% VCAM-1 is present. The cell traces (Fig. S3A) and MI (Fig. 5E) show that on ICAM-1 surfaces (V₀+I₁₀₀), KG1a cells migrate against the direction of flow (MI=−0.30 at 800 s⁻¹), while on 10% VCAM-1 surfaces (V₁₀+I₉₀), KG1a cells migrate downstream, with a net MI of 0.15 at a shear rate of 800 s⁻¹. The preference for downstream migration increases with increasing VCAM-1 concentration (Fig. S3A) and shear rate (Fig. 5E).

Similarly, migration experiments on ICAM-1 to MAdCAM-1 (I_x+M_x, where x is the proportion of each molecule as a percentage) surfaces demonstrate that for KG1a cells, ICAM-1 dominates the directional preference in the absence of VCAM-1. The cell traces (Fig. S3B) and MI (Fig. 5F) show that while a purely MAdCAM-1 surface (I₀+M₁₀₀) supports bi-directional migration (MI=0.26 at 800 s⁻¹), introducing 25% ICAM-1 (I₂₅+M₇₅) leads to net upstream migration (MI=−0.08 at 800 s⁻¹). The MI then decreases (indicating increased upstream migration) with an increasing proportion of ICAM-1. For all ICAM-1 and MAdCAM-1 mixtures, increasing the shear rate results in more upstream motility.

Taken together, this data shows that KG1a cells can migrate against the direction of flow on surfaces of mixed composition, while the direction of motion under shear flow is determined by VCAM-1>ICAM-1>MAdCAM-1. Furthermore, LFA-1 is the critical receptor mediating upstream motion on surfaces of mixed composition.

We considered whether the behavior of the immortalized CD34⁺ KG1a cell line could be extended to primary CD34⁺ HSPCs from bone marrow (Fig. 6). Primary HSPCs migrate against the direction of flow on ICAM-1 surfaces (Fig. 6A, left panel; Movie 6). The direction of motility was reversed upon blocking the α_L integrin (Fig. 6A, right panel). The directionality was quantified with the MI (Fig. 6B), which decreases with increasing shear rate, from −0.05 at 100 s⁻¹ to −0.14 at 800 s⁻¹, indicating increased upstream migration with 60.8% and 65.2% of cells migrating upstream, respectively (Fig. 6C). Furthermore, the direction of motion can be switched to downstream by blocking LFA-1 (MI=0.38 at 100 s⁻¹ and 0.39 at 800 s⁻¹).

Fig. 6D illustrates that, on MAdCAM-1 surfaces, HSPCs exhibit the same bi-directional motion previously seen with KG1a cells (left plot). This bi-directional motion is illustrated by the MI (Fig. 6E), which only slightly increases with increasing shear rate, from 0.24 at 100 s⁻¹ to 0.30 at 800 s⁻¹ with ∼31% of isotype cells migrating upstream at both shear rates (Fig. 6F). Blocking LFA-1 again removes most upstream migration (right plot, Fig. 6D) as downstream motion can be induced by blocking the α_L integrin (MI=0.66 at 100 s⁻¹ and 0.79 at 800 s⁻¹) with only 2.1% and 2.6% of cells migrating upstream, respectively.

On VCAM-1 surfaces, HSPCs migrate downstream (Fig. 6G, left plot; Movie 7), and this is unchanged upon blocking the α_L integrin (right plot). Cells treated with isotype-matched control antibody migrate downstream with a MI of 0.69 at 100 s⁻¹ and 0.71 at 800 s⁻¹ (Fig. 6H). Blocking the α_L integrin in HSPCs did not significantly change the direction of migration on VCAM-1 (MI=0.60 at 100 s⁻¹ and 0.66 at 800 s⁻¹). No HSPCs were seen migrating upstream on VCAM-1 under any of the conditions. Taken together, this data demonstrates that HSPCs migrate with the direction of flow on VCAM-1, independent of shear rate.

Furthermore, on equal mixtures of VCAM-1 and ICAM-1 (Fig. 7A–C) and ICAM-1 and MAdCAM-1 (Fig. 7D–F), HSPCs demonstrate the same directional preferences as KG1a cells (Fig. 7A, left panel). Downstream motion was observed and ∼31% of cells migrated upstream (Fig. 7C) on VCAM-1 and ICAM-1 surfaces. Net upstream motion was observed on ICAM-1+MAdCAM-1 surfaces (Fig. 7D, left panel) with ∼60% of cells traveling upstream at both 100 s⁻¹ and 800 s⁻¹ (Fig. 7F). Blocking LFA-1 when HSPCs were on mixed VCAM-1+ICAM-1 surfaces (Fig. 7A, middle panel) or on ICAM-1+MAdCAM-1 surfaces (Fig. 7D, right panel) removes all upstream migration. Migration on a surface combining VCAM-1+ICAM-1 (Fig. 7B) gives an MI of 0.24 at 100 s⁻¹ and 0.29 at 800 s⁻¹ under isotype conditions, and the MI increases to 0.77 at 100 s⁻¹ and 0.76 at 800 s⁻¹ upon blocking LFA-1. In contrast, blocking VLA-4 on a surface with VCAM-1+ICAM-1 (Fig. 7A, right panel) enhances upstream migration to the point that the net migration switches from downstream to upstream, with >50% of cells traveling upstream. On surfaces of ICAM-1+MAdCAM-1, blocking LFA-1 changes the direction of migration from upstream to downstream. Migration on a surfaces made of ICAM-1+MAdCAM-1 (Fig. 7E) also switches direction upon blocking LFA-1; the MI changes from −0.12 at to 0.64 at 100 s⁻¹ and from −0.14 to 0.66 at 800 s⁻¹ upon blocking LFA-1.

In summary, these data demonstrate that primary bone marrow HSPCs, like KG1a cells, can migrate upstream on surfaces containing ICAM-1 and MAdCAM-1, and this upstream migration is dependent on LFA-1.

Considering that both KG1a cells and HSPCs were able to migrate upstream on recombinant CAM surfaces, we tested whether KG1a and primary HSPCs exhibited upstream motility on primary HUVEC monolayers (Fig. 8). KG1a cells (Fig. 8A, left panel; Movie 8) primarily migrate against the direction of flow, and all upstream motion could be suppressed upon blocking the α_L integrin (Fig. 8A, middle panel) or enhanced further by blocking the α₄ integrin (Fig. 8A, right panel). This preference is also seen in the MI (Fig. 8C) and the percentage of cells migrating upstream (Fig. 8D) as the overall MI is upstream (−0.10±0.03) and 57.4% of cells migrate upstream in the presence of negative control isotype-matched antibodies. Downstream migration could be induced (MI=0.52±0.05), with only 11.4% of cells migrating upstream, by blocking LFA-1. Blocking the α₄ integrin slightly increased upstream migration (MI=−0.19±0.01) with 71% of cells migrating upstream.

With HSPCs, blocking ICAM-1 or VCAM-1 on the HUVEC surface (Fig. S6) had similar effects to blocking the α_L integrin and the α₄ integrin, respectively. If ICAM-1 was blocked, only 8.9% of cells traveled upstream (MI=0.52±0.01). However, if VCAM-1 was blocked, 74.9% of cells traveled upstream (MI=−0.30±0.05). A substantial number of HSPCs were traveling upstream on HUVECs (Fig. 8B, left panel), although the upstream migration is not as pronounced as on molecularly pure surfaces. Like KG1a cells, all upstream motion could be removed upon blocking the α_L integrin (Fig. 8B, middle panel) or enhanced further (with KG1a cells) by blocking the α₄ integrin (Fig. 8B, right panel). The less pronounced, albeit still significant, upstream migration of HSPCs is also seen in the MI (Fig. 8C) and the percentage of cells migrating upstream (Fig. 8D), as the overall MI is downstream (0.25±0.04) and 34.7% of cells migrate upstream in the presence of negative control isotype-matched antibodies (as compared to a negative MI and >50% of cells migrating upstream with KG1a cells). Upon blocking LFA-1, the migration could be directed to almost completely downstream (MI=0.68±0.03) with only 3.8% of cells migrating upstream. Blocking the α₄ integrin significantly increased the upstream migration (MI=0.03±0.06) with 54.8% of cells migrating upstream. In summary, the data imply that upstream migration is a physiologically relevant phenomenon for primary bone marrow HSPCs, as they can migrate upstream on primary endothelial cells at shear rates seen in vivo.

Integrin interactions with endothelium-bound CAMs and subsequent migration under shear flow is one of the critical steps in the recruitment of immune cells to the sites of inflammation during the leukocyte adhesion cascade (Ley et al., 2007). HSPCs utilize this same pathway to migrate to the bone marrow microenvironment post transplantation where they re-establish hematopoiesis (Appelbaum, 2007; Sackstein, 2009). Much interest has been shown in understanding the trafficking of HSPCs in order to improve the ratio of cells injected that actually reach the bone marrow. In human stem cell transplants, a minimal dose of >2×10⁶ cells/kg is recommended for rapid hematopoietic recovery (absolute neutrophil count >500/mm³ for 3 consecutive days; Bensinger et al., 1995). With only 1×10⁸–2×10⁸ transplantable HSPCs recovered from a normal apheresis product, the number of cells available for transplantation from a single donor is limited. Therefore, a better understanding of the mechanisms involved in HSPC homing will presumably improve patient outcomes post transplantation by allowing clinicians to deliver more HSPCs and aid in faster hematopoietic recovery.

While the main investigations of HSPC homing have focused on the mechanisms of the initial selectin-mediated interactions (Robinson et al., 2012; Sackstein, 2012) or upregulating the chemokine-mediated ‘inside-out’ activation (Kahn et al., 2003; Kollet et al., 2002; Peled et al., 1999b) of the integrins to allow the cell to better contact the endothelium, we focused on the biomechanics of HSPC migration under shear flow while in contact with surfaces presenting the three principal endothelial ligands: ICAM-1, VCAM-1 and MAdCAM-1 (Luster et al., 2005). Recent observations by ourselves and others have demonstrated that both murine (Steiner et al., 2010) and human (Valignat et al., 2013) T lymphocytes, but not neutrophils (Valignat et al., 2013), are able to efficiently crawl against the direction of shear flow on ICAM-1 surfaces as well as on surfaces with mixtures of ICAM-1 and VCAM-1, but crawl with the direction of flow on VCAM-1 substrates (Dominguez et al., 2015; Steiner et al., 2010). Here, we demonstrate that HSPCs also display the ability to migrate upstream under flow. Both the primitive immortalized KG1a cell line and bone marrow-derived HSPCs, which displayed a remarkably similar integrin ligand profile to T lymphocytes, could support upstream migration on surfaces containing ICAM-1 while maintaining a preference for downstream migration on VCAM-1 surfaces. In addition, we demonstrated the ability of KG1a cells and HSPCs to migrate in almost equal numbers in both directions, or bi-directionally, on MAdCAM-1.

Interesting differences in cell morphology were seen during migration as cells are predominantly rounded on VCAM-1, while on the ICAM-1 surfaces, cells have a mix of well-rounded cells and cells that look more spread out (Movies 1 and 6 for ICAM-1; Movies 2 and 7 for VCAM-1). Several differences between VLA-4–VCAM-1 and LFA-1–ICAM-1 migration could account for this observation. These include the fact that VLA-4 can bind to VCAM-1 at basal levels without stimulation to either of its active forms, while LFA-1 cannot bind to ICAM-1 in this manner. This may allow KG1a cells to bind to VCAM-1 without any inside-out signaling, which may cause more spreading along the surface. Furthermore, since these assays are performed in the absence of chemokines, this may account for the difference in morphology as VCAM-1 may need chemokine stimulation to induce spreading in a more fibroblast-like manner. As for migration on MAdCAM-1 surfaces, it is true that very few KG1a cells adhere to the surface and this is likely explained by the relatively low levels of the β₇ integrin chain (as compared to β₁ and β₂ integrin chains). Only ∼10 cells/mm² adhere to the MadCAM-1 surface (Fig. 4F) as compared to >200 and 60 cells/mm² for VCAM-1 and ICAM-1 surfaces, respectively. As for the HSPCs, in general many fewer cells interacted with the surface per frame. This is most likely due to the lower expression levels of β₁, β₂ and α_L integrin in HSPCs as compared to that in KG1a cells.

Experiments performed on surfaces with mixtures of two CAMs were even more enlightening as they elucidated what the directional preferences of the HSPCs were when presented with multiple CAMs. When combining VCAM-1 and ICAM-1 or ICAM-1 and MAdCAM-1, KG1a cells and HSPCs showed a net overall upstream migration on ICAM-1+MAdCAM-1 surfaces and a net overall downstream migration on VCAM-1+ICAM-1 surfaces. If as little as 25% of the surface is ICAM-1 in ICAM-1+MAdCAM-1 mixtures, the cells migrate upstream. The effect of MAdCAM-1 on migration is not surprising, since both KG1a cells and HSPCs have a marginal expression of the MAdCAM-1 ligand LPAM-1 (α₄β₇) (Berlin et al., 1993) compared to LFA-1 which binds to ICAM-1. In addition, LPAM-1 can sustain binding to MAdCAM-1 in its closed form (Briskin et al., 1996), which may account for the majority of the downstream interactions seen in the bi-directional motion on pure MAdCAM-1 surfaces. While upstream migration is seen on VCAM-1+ICAM-1 surfaces, KG1a cells and HSPCs switch to a downstream MI when the mixture contains 10% of is VCAM-1. This is the opposite of what is seen for T lymphocyte migration on VCAM-1+ICAM-1 surfaces, as T cells prefer to travel upstream on VCAM-1+ICAM-1 surfaces when any ICAM-1 is added to the surface (Dominguez et al., 2015). This distinction likely results from a difference in how integrins respond to CAM ligation in the two cell types. For example, in HSPCs, VCAM-1 dominates over ICAM-1 and selectins (Papayannopoulou et al., 2001) for homing to the bone marrow, while in contrast T lymphocytes are in constant contact with ICAM-1 surfaces during migration (Hogg et al., 2004). The VCAM-1+ICAM-1 surface is also of particular interest because it recapitulates the composition of the primary CAMs that are present on endothelium upon activation by TNF and IL-β during inflammation (MacKay et al., 1993). In addition, the ability to increase upstream migration of HSPCs on surfaces of mixed composition by blocking the α₄ integrin is interesting and has implications for anti-VLA4 therapies such as Natalizumab (Schwab et al., 2015). Furthermore, since our laboratory and others have shown that upstream migration is preserved on endothelial cells (Steiner et al., 2010), which display both ICAM-1 and VCAM-1, the net downstream migration of KG1a cells and HSPCs seen on the I₅₀+V₅₀ surface may be due to some ligand that helps stabilize upstream migration that is lacking on the recombinant surface but preserved on the endothelial surface.

Our data also isolated LFA-1 as the critical cell-borne integrin controlling upstream migration. Blocking LFA-1 removed all upstream migration on surfaces made of the ICAM-1 or MAdCAM-1, and also surfaces that combine VCAM-1+ICAM-1 or ICAM-1+MAdCAM-1. In order to begin to understand how the cell reorganizes occurring during upstream and downstream migration, we performed fluorescence imaging to determine the localization of both the active forms (intermediate and high affinity conformations) of LFA-1 using the conformation-sensitive antibody KIM-127 (Fig. S4) and to image the F-actin network using KG1a cells transfected with GFP–Lifeact (Fig. S5). Our data demonstrate that the activated form of LFA-1 is found at the trailing edge and the uropod whenever the cell is crawling upstream (ICAM-1), whereas it is found in the lamellipod when moving cells are crawling downstream (as they do on VCAM-1 surfaces) (Fig. S4). Furthermore, the localization of F-actin was the the exact opposite, with GFP–Lifeact localized at the leading edge during upstream migration on ICAM-1 and at the trailing edge and the uropod during downstream migration on VCAM-1 surfaces (Fig. S5). Taken together, this points to critical differences in the integrin–actin complex during upstream and downstream migration, which warrants further study. Specifically, we are interested in how the reported crosstalk between the LFA-1 and VLA-4 integrins in both activating and inhibiting the function of each other (Chan et al., 2000; Chigaev and Sklar, 2012; Grönholm et al., 2016) plays a role in governing upstream versus downstream migration, the integrin activation states and trafficking to the bone marrow post transplantation in HSPCs.

A previous study has put forth the hypothesis that the upstream migration of the T lymphocytes is due to a passive steering mechanism of the uropod during shear flow (Valignat et al., 2014). The ‘wind-vane’ hypothesis was borne from the few changes in upstream migration seen when intracellular Ca²⁺ signaling and phosphoinositide 3-kinase (PI3K) were inhibited, and the ability to change the direction of migration by repositioning the uropod using beads that could change the hydrodynamic drag, or magnetic beads that could be used to manipulate the direction of migration when in a magnetic field. Our work further implicates the uropod as critical as the cytoskeletal organization of F-actin and distribution of active LFA-1 in the uropod is fundamentally different based on the direction of motion. However, it was also found previously that the uropod is weakly adherent in T cells (Valignat et al., 2014) and additional imaging by interference reflection microscopy needs to be performed to assess the intimacy of contact of the uropod in HSPCs.

Nevertheless, this paper illustrates that the fascinating phenomenon of upstream migration of trafficking immune cells under flow, which was first observed in T-cells, also occurs in HSPCs. As has been suggested by others, upstream migration after initial binding may be a natural mechanism by which cells home back to their site of activation after arrest. Depending on the levels of adhesion molecule expression on the endothelium (Beste et al., 2012), it may take several seconds and considerable distances before a cell arrests downstream of the stimulatory site. Thus, upstream migration under flow may be a more universal mechanism of fine-tuning homing than has been previously thought. The ability to recreate this phenomenon in an immortalized cell line such as KG1a cells, which capture the essential essence of HSPCs and are amenable to genetic manipulation, opens future possibilities for elucidating the molecular mechanisms of upstream migration under flow.

In conclusion, the fact that KG1a cells and primary HSPCs undergo LFA-1-mediated upstream migration, like T-lymphocytes, is fascinating. Since upstream migration has been demonstrated under biologically relevant shear flow conditions, with primary HSPCs, and in vitro on activated HUVEC monolayers, further study is warranted to determine whether the upstream migration occurs with HSPCs in vivo. Although a specific biological function for upstream migration has yet to be elucidated, one can speculate that this unique mode of motility is beneficial for HSPC homing to the bone marrow post transplantation and that it can be leveraged for clinical benefit.

All primary and cell lines were cultured as described previously (Lee et al., 2014; Mondal et al., 2014, 2016; Stolfa et al., 2016). Briefly, KG1a cells (ATCC) were cultured as described previously in Iscove's modified Eagle's medium (IMDM) containing 20% fetal bovine serum (FBS). Frozen aliquots of primary bone marrow HSPCs were obtained from the University of Pennsylvania Stem Cell and Xenograft Core. HSPCs were revived and cultured in Stemspan serum-free medium (Stem Cell Technologies, Vancouver, BC) overnight prior to experimentation (Dougher et al., 2017). Human umbilical vein endothelial cells (HUVECs) were maintained in EBM-2 growth medium with BulletKit supplement (Lonza, Wakersville, MD). For experimentation, HUVECs were seeded in a 35 mm×10 mm tissue culture treated dish (BD Falcon, Bedford, MA) and grown to confluence. HUVECs were then stimulated with IL-1β (BioLegend, San Diego, CA) for 4 h prior to experimentation.

Briefly, either TC treated 24-well plates (BD Falcon) for static experiments or 1 inch×3 inch plastic microscope slides (Fisher Scientific, Waltham, MA) for shear flow experiments were coated with 2 µg/ml Protein A/G (Biovision, Milpitas, CA) overnight at 4°C. Upon washing three times with 1× PBS, surfaces were treated with 0.2% Pluronic F127 (Sigma-Aldrich, St Louis, MO) for 1 h to block non-specific binding to the surface. After three additional 1× PBS washes, surfaces were the coated with a total concentration of 2.5 µg/ml of recombinant human ICAM-1 Fc, VCAM-1 Fc, MAdCAM-1 Fc (R&D Systems, Minneapolis, MN) or a mixture of these CAMs for 1 h. Human fibronectin surfaces (Sigma-Aldrich) were coated at a range of 2.5–100 µg/ml for 1 h at 37°C and then washed three times with 1× PBS. Surfaces were then treated with 0.2% pluronic acid F127 for 1 h and washed an additional three times with 1× PBS prior to experiments.

After the preparation of either fibronectin or CAM-coated 24-well plates, 500 µl of IMDM+20% FBS was added to each experimental well. KG1a cells were then seeded at a concentration 5000 cells/cm² and allowed to settle and attach to the surface for 30 min. The plate was then mounted on a microscope in a 5% CO₂ and 37°C environment for 15 min to allow for equilibration. Images were captured every minute for 30 min on a motorized stage and observed using a Nikon Eclipse TE300 phase-contrast microscope. Images were captured using a 10× objective.

Experiments were conducted as described previously (Buffone et al., 2013, 2017). Briefly, a parallel-plate flow chamber (GlycoTech, Gaithersburg, MD) was used with a prepared plastic slide functionalized with either ICAM-1 Fc, VCAM-1 Fc, MAdCAM-1 Fc or mixtures of the CAMs. The channel template was cut from 0.01-inch-thick Duralastic sheeting (Allied Biomedical, Ventura, CA). For each flow experiment, the template was placed over the prepared slide. The template and slide were placed in the bottom well of the flow chamber, and the top was secured with screws. The chamber was assembled under water to minimize the introduction of air. It was then mounted on the microscope in a 5% CO₂ and 37°C environment for 10 min to allow for equilibration. Before introduction of cells, the chamber was flushed with running medium (IMDM). A volume of 2 ml containing 10⁶ cells in running medium was introduced into the chamber and cells were allowed to attach for 30 min in which time most but not all migrating cells adopt a polarized morphology consisting of a lamellipod at the front and a uropod at the rear. These actively migrating cells were the cells included in the analysis. The overall morphology of cells varied as both KG1a cells and HSPCs are a heterogeneous population of cells but the majority of spherical and non-adherent cells were either washed away upon application of flow or demonstrated no motility and were not included in the analysis. Fluid flow was initiated by using a syringe pump (11 Plus, Harvard Apparatus, Holliston, MA) and volumetric flow rates were adjusted accordingly to correspond to desired shear rates. Shear rate was calculated using τ_w=(6 μQ)/(h²w) where μ is the fluid viscosity, Q is the volumetric flow rate, h is the channel height and w is the channel width. For this chamber, h=0.023 cm and w=0.1 cm. Images were captured every minute on a motorized stage of a Nikon Eclipse TE300 phase-contrast microscope. Images were captured using a 10× objective. The number of cells per mm² was calculated as the average of the number of cells migrating at six unique positions in the chamber and normalizing this average per mm².

Experiments were conducted under similar conditions to those in previous studies (MacKay and Hammer, 2016; Mondal et al., 2014). In brief, HUVECs (Lonza) were seeded and grown to confluence in EGM-2 medium with Bullet supplements (Lonza) on 35×10 mm tissue culture treated polystyrene plates (BD Falcon) and gown to near confluence at 37°C and 5% CO₂. Prior to experimentation, HUVECs were supplied with fresh EGM-2 medium supplemented with 10 ng/ml recombinant IL-1β (BioLegend) and allowed to incubate for a minimum of 4 h to ensure robust expression of E-selectin, ICAM-1 and VCAM-1 on the surface. To perfuse the KG1a cells and HSPCs across the HUVEC monolayer, we used a circular parallel-plate flow chamber (Glycotech 31-001) under vacuum, fitted with a rubber gasket to create a rectangular flow path (h=127μm, w=0.25 cm). The chamber was assembled under water to minimize the introduction of air. It was then mounted on the microscope in a 5% CO₂ and 37°C environment for 10 min to allow for equilibration. Before introduction of cells, the chamber was flushed with running medium (IMDM). A volume of 2 ml containing 10⁶ cells in running medium was introduced into the chamber and cells were allowed to attach for 30 min. From this point on, the assay was performed under similar conditions to the previous experiments on recombinant surfaces as described above.

Functional blocking antibodies HL111 (anti-CD11a/α_L integrin; cat. no 301214), 3.9 (anti-CD11c/α_X integrin; cat. no 301616), 9F10 (anti-CD49d/α₄ integrin; cat. no 103718), DREG-56 (anti-CD62L; cat. no 304812), HCD54 (anti-CD54/ICAM-1; cat. no 322704), HAE-1F (anti-CD62E/E-selectin; cat. no 336004), and FIB27 (anti-β₇ integrin; cat. no 121003) from BioLegend and HUTS-21 (anti-CD29/β₁ integrin; cat. no 556048), L130 (anti-CD18/β₂ integrin; cat. no 556084), and 51-10C9 (anti-CD106/VCAM-1; cat. no 556645) were obtained from BD Biosciences, San Jose, CA. 5×10⁵ KG1a cells or HSPCs were blocked with a final concentration of 50 μg/ml blocking antibody in 25 µl running medium and incubated for 30 min at 37°C and 5% CO₂ as described previously (MacKay and Hammer, 2016). Cells were then injected into flow chamber apparatus and allowed to adhere in the absence of flow for 30 min. Running buffer consisted of IMDM and a 1:10 final antibody blocking concentration. Cells were exposed to flow for 30 min before quantification of speed and MI.

Cell movement was tracked using the ImageJ plugin Manual Tracking as described previously (Dominguez et al., 2015). ImageJ and the plugin are both freely available through the NIH website (http://rsbweb.nih.gov/ij/). The centroid of the cell was considered to represent the cell position. Time-lapse microscopy was used and images were taken every minute. The result was a series of (x_i,y_i) positions with time for each cell. The net displacement during the ith minute increment, D_i, was calculated by determining the difference of the position at the beginning and end of the time step. The sum of total displacements (D_i,_accum) was used to calculate the cell speed over the entire experimental time course of 30 min. The MI was defined as the ratio of the difference between the initial and final x-displacement to total displacement where MI=(x_i,end–x_i,initial)/(D_i,accum). Values of the MI near −1, indicate that cells migrate in a straight trajectory against the direction of flow while values near +1, indicate migration in a straight trajectory in the direction of flow. When the MI is near 0, there is no preferred direction in migration, indicating random motility. Only single cells actively migrating along the surface which remained in the field of view for the entire experiment were included in the analysis. Dividing cells and clusters of cells in which the cells interacted with one another's path were not included in analysis, and neither were stationary cells or loosely adhered cells.

Immunofluorescence staining and flow cytometric analysis of cells were performed as described previously (Buffone et al., 2013). Cells (0.5×10⁶–1×10⁶) were washed in PBS twice before labeling. Samples were incubated with combinations of fluorescently labeled antibodies HI111 (anti-CD11a/LFA-1; cat. no 301206), ICRF44 (anti–CD11b/Mac-1; cat. no 301330), 3.9 (anti-CD11c; cat. no 301604), DREG-56 (anti-CD62L/L-selectin; cat. no 304608), 561 (anti-CD34; cat. no 343604), HB-7 (anti-CD38; cat. no 356606), 9F10 (anti-CD49d; cat. no 304316), NKI-SAM-1 (anti-CD49e; cat. no 328008), TS1/18 (anti-CD18/β₂-integrin; cat. no 302106), TS2/16 (anti-CD29/β₁-integrin; cat. no 303016), and FIB27 (anti-β₇-integrin; cat. no 121010) for 15 min before analysis. All antibodies were from Biolegend (San Diego, CA). Flow cytometric analysis was performed with a Calibur flow cytometer (Becton Dickinson Immunocytometry Systems, Franklin Lakes, NJ) and the FACS Diva software package and histograms were generated using the FlowJo software.

Parallel flow chamber assays were performed as described above and previously (Dominguez et al., 2015). Upon completion of experiments, the cells were fixed with 4% paraformaldehyde in PBS for 7 min that was flowed through at the experimental shear rate. The substrate was removed from the flow chamber apparatus and then blocked with 1% bovine serum albumin (BSA) in PBS for 30 min at room temperature. Cells were incubated with 1:1000 Alexa Fluor 568-labeled anti-LFA-1 intermediate affinity clone KIM-127 for 30 min at room temperature followed by DAPI for a subsequent 30 min. Cells were mounted in Fluoromount-G mounting medium (SouthernBiotech, Birmingham, AL) and examined by confocal microscopy (Leica SP5).

Recombinant lentivirus was produced as described previously (Buffone et al., 2013; Mondal et al., 2014) by transfecting 15 µg psPAX2 (Addgene plasmid #12260), 10 µg pMD2.G (Addgene plasmid #12259) and 20 µg pLX301 (Addgene plasmid #25895; Yang et al., 2011) modified to contain GFP–Lifeact into HEK239T (ATCC, Manassas, VA) cells through calcium phosphate precipitation. 5×10⁶ cells KG1a cells were mixed with harvested 100× lentiviral suspension, spun at 1200 g for 2 h, and re-suspended in fresh culture medium. After 48 h, GFP–Lifeact-positive cells were selected with 2.5 µg/ml puromycin (Sigma) and later verified by fluorescence microscopy.

Data are presented as mean±s.e.m. Testing for differences between means was determined with either Student's t-test or two-way ANOVA with post hoc comparisons in InStat 3 software (GraphPad, San Diego, CA). P<0.05 was considered significant. P-values are indicated in figure legends.

The authors would like to thank Dr Nathan Roy and Prof. Janis Burkhardt (Children's Hospital of Philadelphia) for the KG1a GFP–Lifeact cells and Prof. Donna Wall (The Hospital for Sick Children) for her helpful discussions regarding the clinical impacts of HSPC homing.