Analytical cell adhesion chromatography reveals impaired persistence of metastatic cell rolling adhesion to P-selectin

Circulating cell adhesion amidst the high shear force environment of the vasculature is initiated by selectins, the molecular ‘brakes’ that slow cells down relative to bulk flow through ligand recognition to facilitate paracrine signaling, firm adhesion and eventual transmigration into the adjacent tissue bed. Deletion of P-selectin, which is expressed by activated platelets and endothelial cells, dramatically attenuates metastatic cell seeding and tumor formation within the lung in experimental mouse models of human colon carcinoma metastasis (Borsig et al., 2001, 2002; Kim et al., 1998), suggesting that P-selectin is an important regulator of circulating tumor cell (CTC) adhesion to and extravasation from the vasculature. Therapeutic interventions disrupting the molecular recognition of P-selectin by CTCs therefore represent a rational means to prevent formation of disseminated metastatic tumor foci, which persists as the largest clinical hurdle in the management of advanced malignancies. However, P-selectin functions in numerous physiological homing processes including leukocyte recruitment during inflammation (Mayadas et al., 1993; Robinson et al., 1999) as well as platelet accumulation and aggregation at areas of vascular injury (Frenette et al., 1995). Broadly inhibiting P-selectin would therefore severely compromise patient immune function and wound healing responses. An ideal anti-metastasis intervention would instead selectively attenuate CTC but not normal circulating cell recognition of P-selectin.

To this end, in vitro screening has the potential to repurpose drugs developed in recent years for applications in the treatment of inflammatory conditions and ischemia-reperfusion injury (Lowe and Ward, 1997) to prevent CTC dissemination into systemic organs. A challenge posed in this application as opposed to other conventional drug targets, however, is that P-selectin-mediated recognition functionally contributes to metastasis under fluid flow rather than static conditions (McCarty et al., 2000). Therefore, as has been appropriately argued in the literature, data obtained using static (no flow) binding assays might not be relevant to the fluid dynamic environment of the vasculature. Another challenge is that selectin-mediated adhesion is highly heterogenous even within a clonal cell population (Aigner et al., 1998), necessitating large sample sizes. A system that uniformly subjects large numbers of whole cells to well-controlled shear flow conditions is thus required to evaluate the influence of therapeutic drug doses on the efficiency of sustained P-selectin adhesion. Such a platform would also reduce the number of animals used in laborious, expensive and time-prohibitive metastasis models to screen and dose-test drug candidates.

Previous efforts developed a parallel-plate flow chamber system for the separation of cells based on their rolling adhesion behavior (Greenberg and Hammer, 2001), a so-called ‘cell adhesion chromatography’ platform. This methodology exploits the differences in rolling adhesion, defined as the transient interaction between a cell in fluid flow and an immobilized adhesive substrate. In such a system where the velocity of the cell while mediating rolling adhesion is significantly lower than its velocity would be in the free flow stream immediately proximal to the surface, cell subpopulations can be enriched. The work which developed this methodology utilized a cell-free system to estimate how CD34⁺ cells can be enriched from a mixture of adult bone marrow cells on an L-selectin-functionalized substrate (Greenberg and Hammer, 2001) based on the differential rolling adhesion behavior of CD34⁺ versus CD34⁻ cells over L-selectin (Greenberg et al., 2000).

Based on these conceptual advances, but repurposed as an analytical rather than preparative chromatographic method, we report here the use of a microfluidic-based parallel-plate flow chamber device designed for use in conjunction with video microscopy to chromatographically interrogate adhesion efficiency of cells to P-selectin under physiological shear flow conditions as a novel drug screening platform. In order to achieve uniform cell–substrate contact of a pulse cell suspension input into a selectin-functionalized parallel-plate flow chamber, we designed a feature that enables settling to the chamber bottom of infused cells based on Stokes flow predictions. This simple modification increased the fraction of cells in contact with the substrate upon entry into the main chromatography channel to >95%, enabling the precise quantification of adhesion efficiencies to P-selectin under physiological levels of venular shear stress (∼1 dyn cm⁻²) (Konstantopoulos et al., 1998), mimicking conditions under which hematogenous metastasis principally occurs. By simultaneously monitoring individual cell rolling velocities and elution times, we unexpectedly observed that longer time- and distance-averaged velocities (determined by the cell elution time from the chamber) of cells do not necessarily correspond to their instantaneous velocities. Using this cell adhesion chromatography methodology, we define a new parameter termed ‘adhesion persistence’, which is conceptually consistent with migration persistence in the context of chemotaxis (Tranquillo et al., 1988) but instead describes the capacity of cells to resist the influence of shear flow and sustain rolling interactions with an adhesive substrate. Importantly, this is distinct from rolling velocity per se because we demonstrate that the adhesion persistence of rolling cell subtypes do not necessarily correspond with their instantaneous velocity. We report that metastatic cells exhibited marked deficiencies in adhesion persistence to P-selectin in shear flow relative to leukocyte-like cells, despite no significant differences in P-selectin ligand expression nor size between analyzed cell subtypes. This suggests that selectins separately regulate the strength versus persistence of cell adhesion, which might result in distinct homing behaviors during the processes of inflammation and wound healing compared with those in metastasis. We also provide evidence that this analytical adhesion chromatography system facilitates comparative dose-testing for the identification of P-selectin antagonists to selectively attenuate CTC but not leukocyte adhesion to P-selectin in shear fluid flow.

Within the circulatory system at the inflamed endothelium, activated endothelial cells upregulate surface expression of P-selectin to facilitate circulating cell capture from blood flow (Fig. 1A). For the in vitro investigation of adhesion dynamics of P-selectin-binding cells (Fig. 1B), parallel-plate flow chambers that induce a uniform level of shear stress are used in conjunction with video microscopy to interrogate cell adhesion to recombinant P-selectin-functionalized substrates (Fig. 1A). In conventional parallel-plate flow chambers, infused cells have a non-uniform distribution of y-axis positions upon entry into the chamber (Fig. 1C,D). Cell contact with the planar functionalized substrate occurs only after cell settling, which transpires at a rate proportional to the density difference between the perfusion medium and cells, as well as the flow rate (see Eqn 5 in the Materials and Methods). This means that within conventional channel lengths (∼2–5 cm), only a minority of cells are within a distance sufficient to support contact with the planar bottom substrate (Fig. 1E). In addition, any horizontal travel distance within the functionalized channel prior to complete cell settling results in non-uniform cell contact with the bottom substrate, complicating chromatographic analysis of cell adhesion behavior of the infused cell population.

To overcome these technical limitations, a non-functionalized settling feature was added upstream to the main parallel-plate channel (Fig. 1F,G) to increase the residence time of cells in the chamber. This ensures that cells flowing into the chamber as a pulse suspension input are able to settle to the bottom substrate before entering the main functionalized channel (Fig. 1G,H). This results in >95% cell contact with the substrate (calculated height based on observed velocity within 1.8× the predicted cell radius) after immediate entry into the functionalized main channel (Fig. 1H,I), as compared to only 40% cell contact achieved by conventional parallel-plate flow chambers (Fig. 1D,I). As a result, uniform cell contact is maintained throughout the entire length of the chamber (Fig. 1J), allowing legitimate chromatographic analysis of perfused cells over P-selectin substrates. This parallel plate design modification results in recovery of ∼75% of infused cells, as estimated from the number of detected cells in the field of view (FOV), as opposed to 40% recovery in chambers without an incorporated settling feature.

We used this adhesion chromatography platform to evaluate the adhesive behavior of various human cell lines with previously reported P-selectin-binding activity (Alves et al., 2008; Hanley et al., 2006; Napier et al., 2007) in a simulated shear flow environment. LS174T and Colo205 are human colon carcinoma cells with a high propensity to form lung and liver metastases when infused intravenously or after solid tumor outgrowth (Dallas et al., 2012). THP-1 and HL-60 are immortalized leukocytes that are commonly used as model monocytes and leukocytes (Alves et al., 2008; Hanley et al., 2006). We found that for all cell types used, a broad distribution of instantaneous velocities of cells perfused over P-selectin-functionalized substrates existed (Fig. 2A,B), including a substantial number of cells that travelled through the FOV at free flow velocity (i) and cells that were interacting in a characteristic rolling fashion (ii). Of the rolling cells (Fig. 2B), differences between the frequency of relatively fast (iii) versus slow (iv) rolling cells were found between each cell subtype. In particular, metastatic cells (LS174T and Colo205) interacted with P-selectin via slow rolling adhesion at lower frequencies relative to leukocyte-like (THP-1 and HL-60) cell subtypes.

Expressed as a proportion of total cells, metastatic and leukocyte-like cells mediated rolling interactions with P-selectin with similar frequencies at each shear stress evaluated (Fig. 2C,D). As expected, the median instantaneous velocity of rolling cells increased with shear stress for all cell subtypes, with metastatic cell subtypes demonstrating the highest median instantaneous velocities of rolling cells and THP-1 cells rolling on average with higher median instantaneous velocities relative to HL-60 cells (Fig. 2E). Of the rolling cell fraction, a markedly larger portion of leukocyte-like cells compared to metastatic cells exhibited instantaneous velocities characteristic of slow rolling (e.g. instantaneous velocity <20 µm s⁻¹) at all shear stress levels evaluated (Fig. 2F). These interactions were specific to cation-dependent P-selectin-mediated recognition of cell-surface-expressed selectin ligands given that co-infusion with 5 mM of the cation chelator EDTA resulted in a significant diminution of the frequency of adhesion and an increase in instantaneous velocities of rolling cells (Fig. 2C).

We also analyzed the elution time behavior of cell suspensions comprised of the human cell lines in a simulated shear flow environment. Based on measured instantaneous velocities in flow, a channel length (measured as starting after the settling feature to the imaging FOV) of 1.3 cm was required to elute >95% of the infused population of leukocyte-like cell subtypes within 120 min, presumably because of the high frequency of these cells that exhibit slow rolling behavior (Fig. 2F). Accordingly, chambers comprised of a settling feature upstream of a 1.3-cm long channel resulted in elution of the entire population of cells within the experimental timeframe during residence time distribution experiments (Fig. 3A). However, whereas leukocyte elution persisted throughout the course of experimentation, metastatic cancer cells eluted very quickly and at approximately one quarter the mean elution time of leukocytes (Fig. 3A), presumably due to their on average higher instantaneous velocities over P-selectin (Fig. 2E). In order to improve the resolution of residence time distributions obtained for metastatic cells using this chromatographic method, the chamber channel length was extended to 14 cm in order to delay the elution times of metastatic cells such that they occurred over the entire course of experimentation (Fig. 3B). The extension of the chamber had no impact on the average number of cells recovered nor on the measured median instantaneous velocity. Additionally, in both chamber lengths, co-infusion with 5 mM EDTA resulted in cells eluting as quickly from P-selectin-coated channels as non-functionalized (blank) channels (Fig. 3A,B), indicating that the delay in elution was the result of cation-dependent P-selectin-mediated cell adhesion.

Based on these measured elution times, the average velocity of individual cells during their transit through the chamber can be calculated using Eqn 1 and frequency distributions and/or medians compared across cell subtypes. Strikingly, the average velocities of LS174T and Colo205 cells were elevated approximately three- to eight-fold relative to THP-1 and HL-60 cells (Fig. 3C) at all shear stresses tested (Fig. 3D). The considerable dissimilarity in average velocities of metastatic versus leukocyte-like cells was not the result of subtle differences in the frequency of adhesive cells of each individual subtype, given that when accounting for the behavior of rolling cells only, this dramatic separation between the average velocity of metastatic versus leukocyte-like cells remained (Fig. 3E). Nor could these striking differences in average velocities between cell subtypes be solely attributed to the increase in instantaneous cell velocities exhibited by metastatic relative to leukocyte-like cells (Fig. 2E), which were increased by only 20–200%.

To investigate this discrepancy, we evaluated the instantaneous velocity distributions of eluting cells. We discovered that at any given elution time, cells from a clonal population exhibited different instantaneous velocities within the FOV (Fig. 4A), including instantaneous velocities indicative of both cells being in free flow (i) and mediating rolling adhesion (ii). This means that even for cells with identical average velocities (given their identical elution time), cells might not interact with P-selectin with the same instantaneous velocity (Fig. 4A). This is supported by the bimodal distribution of elution times predicted by observed instantaneous cell velocities compared to the monomodality of the observed elution profile (Fig. 4B).

When juxtaposing the measured average versus instantaneous velocities of rolling cells, we found that at any given instantaneous velocity, LS174T (Fig. 4C) and Colo205 (data not shown) cell subtypes exhibited a significantly higher average velocity relative to that of THP-1 (Fig. 4C) and HL-60 (data not shown) cell subtypes. This resulted in a slope of approximately seven for LS174T cells whereas THP-1 (Fig. 4C) and HL-60 (data not shown) were much closer to one. On a per cell basis, we next considered the ratio of average to instantaneous velocities, where a value of one implies that the instantaneous cell velocity persists throughout the channel and matches that of the average cell behavior. Cells exhibiting instantaneous velocities within the imaging FOV indicative of rolling adhesion (instantaneous velocities <250 μm s⁻¹ at 1.0 dyn cm⁻²) exhibited a ratio of average to instantaneous velocities substantially greater than one for metastatic LS174T and Colo205 cells subtypes but approached an average closer to one for THP-1 and HL-60 cells (Fig. 4D,F). This implies that THP-1 and to a lesser extent HL-60, but not LS174T nor Colo205, cells on average sustained rolling adhesion throughout the channel length. By contrast, LS174T and Colo205 cells interacting with P-selectin through rolling adhesion at the time of elution eluted much earlier than would be predicted from observed instantaneous velocities, indicating that these cells were instead not interacting with the P-selectin substrate for significant portions of the channel length. These data suggest that although metastatic cancer cells might have a capacity to interact with P-selectin at a given rolling velocity, rolling adhesion might not be sustained over prolonged times or distances. The small increase in v_avg/v_inst over a value of one also suggests the same to some extent for HL-60 cells.

The trend in v_avg/v_inst ratio of cells in free flow at the time of observation within the imaging FOV (>250 μm s⁻¹ at 1.0 dyn cm⁻²) was as follows: the ratio of average to instantaneous velocity approached a median of one for metastatic LS174T and Colo205 cells but was markedly less than one for THP-1 and HL-60 cells (Fig. 4E,F). This indicates that LS174T and Colo205 cells moving at free flow velocities at the time of elution travelled through the majority of the P-selectin-functionalized channel at that velocity, therefore mediating little to no adhesion. THP-1 and HL-60 cells moving at free flow velocities within the FOV, by contrast, eluted much later than would be predicted from observed instantaneous velocities, indicating that these cells were instead interacting with the P-selectin substrate for significant portions of the chamber length, but not within the imaging FOV.

To determine whether the differences in metrics of cell adhesivity to P-selectin could be directly attributed to differences in cell subtype densities of P-selectin ligand or sialyl-Lewis x/a (sLe^x/a) expression, flow cytometry was performed. We found that both metastatic and leukocyte-like cells stained prodigiously and to a relatively similar extent with recombinant P-selectin (Fig. 6A) and exhibited no notable differences in sLe^x/a expression (Fig. 6B). We also found only modest to subtle differences between diameters of each cell type in suspension, which ranged on average between 11 and 13 μm (Fig. 6C,D) and no appreciable differences in cell aspect ratios (Fig. 6D). Pearson correlation analysis also revealed no statistically significant relationships between cellular P-selectin ligand and sLe^x/a densities, diameter, as well as aspect ratio with metrics of P-selectin adhesivity that were measured for individual cell subtypes in flow including adhesion efficiency (the fraction of total cells that mediate rolling adhesion in flow), median instantaneous velocity, median average velocity and percentage binding time. These results indicate that neither adhesive ligand expression nor cell size or geometry could predict the results of functional P-selectin adhesion assays a priori.

We next evaluated the influence of P-selectin antagonist heparin on the capacity of metastatic cancer cells versus leukocytes to mediate and sustain adhesion to P-selectin in shear flow (Fig. 7A,B). We found that escalating concentrations of heparin resulted in a larger reduction in the frequency of malignant versus leukocyte-like cell rolling over P-selectin in shear flow at 1.0 dyn cm⁻² (Fig. 7C) and a very significant versus slight reduction in the frequency of slow rolling cells for LS174T and THP-1 cells, respectively (Fig. 7D).

In order to probe the molecular underpinnings of these differences, we incubated LS174T and THP-1 cells with P-selectin in the presence of increasing doses of heparin. Flow cytometry revealed that heparin interferes with P-selectin labeling in a dose-dependent manner for both cell types, with LS174T cells exhibiting a more dramatic attenuation in P-selectin staining with low doses of heparin than THP-1 cells (Fig. 7E), mirroring very closely the effect mediated by heparin on the frequency of cells rolling slowly over P-selectin (Fig. 7D). However, the percentage of P-selectin labeled cells (of the total population) significantly overestimated the attenuation in the total adhesion of (not just slow) rolling LS174T cells to P-selectin-functionalized planar substrates in flow (Fig. 7C), as exhibited by high ratio of LS174T cells exhibiting rolling adhesion to the number of cells staining positive for P-selectin at increasing doses of heparin (Fig. 7F).

Finally, using the percentage binding time as a newly defined metric of adhesion persistence, we determined that although low heparin doses significantly decrease the capacity of LS174T cells to sustain adhesion to P-selectin (Fig. 7G), THP-1 cell adhesion is only affected (and only modestly so) at the highest dose tested (Fig. 7H). These data indicate that this newly described cell adhesion chromatography platform and analytic methodology to quantify adhesion efficiency (based on frequency of rolling cells of total) and persistence (based on juxtaposing measured velocity versus elution time) can be used to measure the relative efficacy of agent dose in impeding P-selectin adhesion under physiological shear conditions.

P-selectin contributes to the hematogenous dissemination of CTCs by mediating interactions with host endothelial cells and platelets. For example, P-selectin expressed by the activated endothelium can facilitate CTC adhesion to the vessel wall (Konstantopoulos and Thomas, 2009). After entry into the bloodstream, CTCs also form complexes with platelets (Borsig et al., 2001, 2002), which masks CTCs from cytotoxic natural killer cells (Nieswandt et al., 1999) or bridges CTC adhesion to the vascular endothelium (Burdick and Konstantopoulos, 2004), thus assisting carcinoma escape from the circulation and seeding of metastatic foci. Given that genetic knockdown of P-selectin dramatically reduces metastasis (Borsig et al., 2001, 2002; Kim et al., 1998) interference with P-selectin recognition by CTCs is a rational therapeutic approach to attenuate cancer dissemination in patients with advanced malignancies.

The biophysical and biomechanical regulation of instantaneous selectin ligand recognition at either a whole-cell or purified single-molecule level has been widely interrogated (Alon et al., 1995; Fritz et al., 1998; Hanley et al., 2003, 2004; Marshall et al., 2003; McCarty et al., 2000) given that selectins facilitate high k_on and k_off rate interactions under hemodynamic force (Chen and Springer, 2001). However, the capacity of cells to sustain interactions with the vasculature during their transit through the circulatory system, which is mediated through selectins, influences their capacity to respond to chemokine signaling, firmly adhere to the endothelium (Diacovo et al., 1996; Veerman et al., 2007) and extravasate to take up residence in the surrounding tissue (Hanna and Etzioni, 2012). Thus, the lack of quantitative insight into the force regulation of selectin-mediated adhesion over longer time and distance scales limits the design of effective therapeutic interventions to selectively modulate cell homing in a cell-subtype-selective manner, such as in anti-metastasis applications.

Herein, we report the design (Fig. 1) and implementation of an analytical cell adhesion chromatography platform that enables the independent analysis of instantaneous versus longer distance and time scale adhesion behavior (Figs 2–4). This microfluidic platform allows for the quantification of cell adhesion to planar substrates functionalized with adhesive ligands in shear flow, thus recapitulating several of the salient features relevant to selectin-mediated cell adhesion in the vasculature. We have modified the conventional parallel-plate flow chamber configuration with a settling feature (Fig. 1F,G,K), which facilitates accurate quantification of the frequency of cells within an infused cell pulse that mediates rolling interactions, as determined by their measured instantaneous velocity within the video microscopy FOV. This approach offers the substantial advantage over other conventional parallel-plate flow chamber systems for the evaluation of cell adhesion efficiencies because it allows for the entire population of perfused cells to have contact with the substrate, in contrast to only ∼40% cell contact with typical chamber geometries. Indeed, although the number of interacting cells within a short experimental timeframe (typically ∼2–5 min) is a widely reported metric used for quantifying the extent of selectin-mediated adhesion, these levels will vary from system to system based on chamber geometry and imaging location. Furthermore, these experiments are typically performed under a constant perfusion of cell solution, complicating mass balance calculations (Munn et al., 1994).

Using this system with a pulse cell suspension input, we can recover the substantial majority of perfused cells, and thus quantify adhesion efficiencies in a manner that is reproducible and can be used to compare cell subtypes or treatment groups. Thus, rather then separating cell subpopulations based on rolling adhesion (Greenberg and Hammer, 2001), we utilize the chromatographic nature of this system to analyze the adhesive behavior of cells in flow. We find that metastatic human colon carcinoma LS174T and Colo205 cells interact with P-selectin under venous levels of shear stress at efficiencies on par with human leukocyte-like THP-1 and HL-60 cells (Fig. 2C,D). However, leukocytes interact with P-selectin through slow rolling interactions at higher frequencies than metastatic cancer cells (Fig. 2F), despite nearly identical levels of selectin ligand expression (Fig. 6A,B) and size features (Fig. 6C,D). These differences might result from P-selectin glycoprotein ligand-1 versus variant isoforms of CD44 (among others, see Napier et al., 2007) functioning as the predominant P-selectin ligands on THP-1 (van Genderen et al., 2006) and HL-60 (Norman et al., 1995) versus LS174T and Colo205 cells (Napier et al., 2007; Thomas et al., 2008), respectively, and their corresponding differences in bond tensile strength and unstressed off rate (Hanley et al., 2003; Raman et al., 2011). Additionally, although average per cell densities of selectin ligands are similar between cell types (Fig. 6B), whether P-selectin ligands expressed by tumor cells in suspension are clustered into microdomains on the cell surface is unknown. Thus, differences in the cell surface distribution of adhesion receptors, which influences the initiation of adhesive interactions but not rolling velocity nor the strength of established rolling adhesion (Stein et al., 1999), might contribute to differential persistence of rolling adhesion between metastatic and leukocyte-like cells. Other as-of-yet undefined biochemical (glycocalyx) and biophysical (cell deformability) differences might also play a role.

In addition to enabling uniform cell contact with the planar substrate to facilitate the quantification of cell adhesion efficiency in flow, our simple chamber design innovation homogenizes the channel length traveled by perfused cells within a pulse input. Given that all cells are in contact with the adhesive substrate for identical lengths in a given experiment using this configuration, the average velocity of a cell can be calculated based on its elution time (Fig. 1K). We found that although metastatic cells interact with P-selectin at similar frequencies to leukocyte-like cells (Fig. 2C,D), the longer distance- and time-averaged velocities of metastatic cells were dramatically higher relative to those of leukocytes (Fig. 3C–E). Given that metastatic LS174T and Colo205 colon carcinoma cells interact on average with P-selectin in shear flow at higher instantaneous velocities relative to THP-1 and HL-60 cells (Fig. 2E), these results would not be unexpected at first glance. However, the incongruent fold difference in instantanous versus average velocities between cell subtypes suggests that another more-dominant mechanism is present. Our finding of co-eluting cells exhibiting instantaneous velocities indicative of cells being in both rolling adhesion and free flow (Fig. 4A) suggests that an observed instantaneous velocity of a cell is not necessarily indicative of its capacity to mediate adhesion. Rather, instantaneous velocity describes the adhesive state of a cell (rolling versus non-interacting) and not the type (adhesive versus non-adhesive). This is best exemplified by our observation of cells with identical measured instantaneous velocities within the imaging FOV exhibiting significantly different average velocities depending on their subtype (Fig. 4C). Hence, despite individual cells exhibiting the same rolling adhesion characteristics over short time and length scales, their capacities to sustain that adhesion diverges between cell subtypes (Fig. 4C). This concept is furthermore supported by our findings that HL-60 cells on average exhibit slower instantaneous velocities (Fig. 2E) but higher average velocities (Fig. 3D) relative to THP-1 cells, corresponding to slightly lower mean percentage binding times (Fig. 5C).

Strikingly, for both the populations of cells rolling and in free flow within the imaging FOV, the median of the per cell ratio of average to instantaneous velocity is roughly tenfold higher for metastatic LS174T and Colo205 colon carcinoma cells than for THP-1 leukocytes. Furthermore, instantaneous velocity within the FOV appears to misconstrue the capacity of cells to sustain adhesive interactions, as this measure indicates too little adhesive interaction by THP-1 and HL-60 cells and too much adhesive interaction by LS174T and Colo205 (and to a much less extent HL-60) cells based on their respective elution time distributions. The percentage binding time metric, which quantifies rolling cell adhesion persistence in shear flow, might therefore represent a previously overlooked aspect of selectin-mediated adhesion that regulates homing activity throughout the circulation. It remains to be determined whether diminished adhesion persistence is a characteristic of P-selectin adhesion in other pathologic states.

The subtype-dependent differences in adhesivity metrics of cells to P-selectin in flow showed no statistically significant correlation with measured selectin ligand densities and size features (Fig. 6E). Although insignificant, cell size did demonstrate a stronger correlation with cell instantaneous and average velocities (Fig. 6E), which is expected given that larger cells are subjected to higher levels of hemodynamic force when interacting with a planar substrate in flow. Interestingly, a somewhat stronger correlation, which was still statistically insignificant, between cell size and percentage binding time was found (Fig. 6E). This suggests that a biophysical mechanism underpins this metric of adhesivity, a concept supported by the lower percentage binding time (as well as a higher v_avg/v_inst ratio) to P-selectin exhibited by HL-60 relative to THP-1 cells despite P-selectin glycoprotein ligand-1 functioning as the predominant P-selectin ligand on both cell subtypes (Norman et al., 1995; van Genderen et al., 2006). However, the underlying mechanism of this adhesive behavior remains to be elucidated.

Numerous agents have been or are in development for modifying selectin–ligand interactions in inflammatory and reperfusion injury states (Lowe and Ward, 1997). Given that P-selectin-mediated recognition is hypothesized to primarily contribute to metastasis and disease progression in the blood stream, these agents have the potential to be applied to use in treatment or management of disseminated cancers. A drug that has been proposed for this use is the antithrombotic agent heparin (Bendas and Borsig, 2012). It is a potent inhibitor of P-selectin binding at concentrations substantially lower than required for its anticoagulant activity (Koenig et al., 1998). Furthermore, in vivo studies have demonstrated considerable efficacy in the reduction of tumor metastasis by heparin and other sulfated compounds (Bendas and Borsig, 2012; Borsig et al., 2001). We found that a low dose of heparin significantly diminished the efficiency of adhesion, the proportion of slow rolling cells, and percent binding time of LS174T but not THP-1 cells to P-selectin (Fig. 7A–D,G,H). This differential response might arise due to the heparin-binding domain of P-selectin being more crucial to cell adhesion when mediated by the P-selectin ligands expressed by metastatic cells, as opposed to P-selectin glycoprotein ligand-1 expressed by leukocyte-like cells (Konstantopoulos and Thomas, 2009; Wang and Geng, 2003). Alternatively, LS174T cells might be more significantly affected by heparin-induced reductions in P-selectin levels available for adhesion due to the lower tensile strengths and higher unstressed off rates exhibited by whole LS174T cells (Hanley et al., 2003) or purified CD44v (Raman et al., 2011) relative to that of P-selectin glycoprotein ligand-1 binding to P-selectin. The long length of microvilli presenting P-selectin glycoprotein ligand-1 (Yago et al., 2002) might also confer a geometrical advantage over carcinoma-expressed selectin ligands for contact with P-selectin in the presence of heparin. These results suggest that heparin might provide robust anti-metastasis efficacy with only marginal effects on monocyte homing activity given the modest diminution in THP-1 cell adhesion efficiency and negligible effects on THP-1 instantaneous velocities and adhesion persistence behavior. Although flow cytometry also indicates a cell-subtype-dependent attenuation of P-selectin binding in the presence of heparin (Fig. 7E), it substantially underestimates the fraction of total cells that mediate rolling adhesion with P-selectin in flow (Fig. 7F). Taken together, these results indicate the potential of our cell adhesion chromatography platform for in vitro dose testing of pharmacologic agents with selective anti-metastasis activity and suggest that this platform has an advantage over single-variable molecular methods such as flow cytometry.

In conclusion, this cell adhesion chromatography-based analytical methodology facilitates the quantification of cell adhesion efficiencies under physiological fluid flow conditions. We demonstrate that this technique represents an effective platform to dose-test P-selectin antagonists and quantify their relative influence on cell-subtype-specific adhesion to P-selectin. Through the juxtaposition of the predicted rolling cell elution time based on instantaneous velocity with observed elution time, we find that metastatic colon carcinoma cells, but not leukocyte-like cells, exhibit a diminished capacity to sustain adhesion to P-selectin in shear flow. Strikingly, this reduced adhesion persistence is not solely reflected by cell rolling at higher instantaneous velocities. Taken together, these data support the notion that the aforementioned biophysical parameters are independently regulated at the cellular level. The analytical capabilities of cell adhesion chromatography will facilitate future work aimed at studying how this differential regulation might contribute to overall cell response to paracrine signaling and other factors regulating the multi-step adhesive cascade of cell homing, as well as identifying specific and targeted anti-metastatic therapies.

Cell culture reagents were purchased from Life Technologies. Heparin (1000 units ml⁻¹) was from SAGENT Pharmaceuticals. Anti-human-IgG (Fc specific) antibody and bovine serum albumin (BSA) were from Sigma-Aldrich. P-selectin-IgG Fc (P-selectin) was purchased from R&D Systems. HECA-452, Rat IgM isotype control and FITC-conjugated mouse anti-rat-IgM antibodies were purchased from BD Pharmingen (San Jose, CA, USA) and Cy3-conjugated anti-human IgG Fc specific antibody was purchased from Sigma-Aldrich. Polydimethylsiloxane (PDMS) base and curing agent were from Ellsworth Adhesives (Germantown, WI, USA).

The chamber was composed of a PDMS block with an open 2-mm wide by 100-μm deep channel either 1.3 or 14 cm long attached to a non-tissue culture treated polystyrene plate (Corning), forming a closed channel with a planar substrate. A settling feature, designed to enable complete cell settling to the chamber bottom before entry into the main channel was designed in a circular configuration with an inner and outer radius of 10.5 and 11.75 mm, respectively, and a 10.9-mm linear inlet region before a bifurcation at the 2.5-mm wide circular channel. The entire settling feature depth was 100 μm. The molds with a negative feature for the PDMS block were produced by micro-machining an aluminium block (Alloy 6013) with a micro-milling machine (OM 1-A, HAAS, Oxnard, CA, USA). Silicone elastomer base and curing agent at a ratio of 9:1 was poured into the molds and cured at 90°C for 3 to 4 h. After the curing process, a biopsy punch was used to make holes for an inlet and an outlet. Then, glass slides (VWR, Randor, PA, USA) were spin-coated (WS-400BZ-6NPP-LITE, Laurell, North Wales, PA, USA) with a mixture of PDMS elastomer base and curing agent at a ratio of 10:1. The bottom of the PDMS block, at the location of the channel, was stamped on the glass slides to apply uncured PDMS mixture. The PDMS block with its bottom covered with the uncured PDMS mixture (functioning as glue) was placed on a polystyrene plate. Care was taken to ensure that there were no air bubbles formed at the interface between the PDMS block and the polystyrene plate. Then, the whole assembly was treated in a 50°C oven overnight to completely cure the PDMS glue mixture, thus finalizing chamber bonding. After the curing process, the chamber was cooled to room temperature and either used immediately or functionalized with proteins for adhesion experiments.

All cell lines were obtained from the American Type Culture Collection. Human colorectal adenocarcinoma cell lines LS174T and Colo205 were cultured in Dulbecco's modified Eagle's medium supplemented with 10% heat-inactivated fetal bovine serum (FBS) and 1% antibiotic-antimycotic or RPMI 1640 supplemented with 10% heat-inactivated FBS and 1% antibiotic-antimycotic, respectively. Each line was subcultured when they reached 90% confluency. For perfusion and flow cytometry experiments, LS174T and Colo205 cells were harvested by mild trypsinization (0.25% Trypsin-EDTA at 37°C) and subsequently incubated in complete medium (at 10⁷ cells ml⁻¹) at 37°C for 2 h to allow regeneration of surface glycoproteins. In the case of Colo205 cells, the suspension cell fraction was first collected and subsequently combined with trypsin-harvested adherent cells. After regeneration, cells were centrifuged at 300 g for 5 min, washed with Dulbecco's phosphate-buffered saline (D-PBS) containing 0.9 mM Ca²⁺ and 0.5 mM Mg²⁺, and resuspended after centrifugation in 0.1% BSA in D-PBS containing Ca²⁺ and Mg²⁺ at a final concentration of 5×10⁵ cells ml⁻¹. Human monocytic THP-1 cells and human promyeloblast HL-60 cells were suspension cultured in RPMI 1640 supplemented with 10% heat-inactivated FBS, 1 mM sodium pyruvate and 10 mM 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid or RPMI 1640 supplemented with 10% heat-inactivated FBS and 1% penicillin-streptomycin, respectively. THP-1 and HL-60 cells were subcultured every 3 days at a dilution of 1:4 or 1:5. The density of THP-1 and HL-60 cells was maintained at between 10⁵ and 10⁶ cells ml⁻¹. For perfusion and flow cytometry experiments with these cell types, cell surface glycoprotein regeneration was not required. Collected cell suspensions were washed with D-PBS containing Ca²⁺ and Mg²⁺ and resuspended in 0.1% BSA in D-PBS containing Ca²⁺ and Mg²⁺ at 5×10⁵ cells ml⁻¹.

5-min videos of cell perfusion experiments performed at prescribed chamber locations were analyzed by using a custom MATLAB-based cell tracking software. Contrast was first applied to detect cell edges and the x and y positions of each detected cell were stored in the first frame. Cell position, shape and size were subsequently recorded until the cell could no longer be detected in the FOV. Tracking was constrained by considering less than half of the pixels of the previously detected cell diameter forward and a divergence angle 15° to the flow direction. Data was only used for cells tracked for more than five frames and debris smaller than 10 μm was excluded. Cell velocity, based on the total tracked distance and video frames per second, as well as the mean aspect ratio, x and y cell radius and projected area of each cell were calculated using these data. The metric y_vel/0.5(r_x+r_y) (the vertical location of the cell from the bottom of the chamber over an effective cell radius) was calculated from Eqn 1 and the mean x and y radius measured during the course of tracking each cell.

Videos from residence time distribution experiments were analyzed by modifying the OpenCV-based Traffic Flow Analyzer (https://github.com/telescope7/TrafficFlowAnalysis). Briefly, the background was subtracted using a mask weight of 0.0005, resulting in an approximate 80 s moving average. A blurring factor of 13 and a thresholding factor of 5 were used. The contour detection algorithm defines cells as >5 μm in diameter that are stored in the Moving Object database and compared to other previously tracked objects and mapped to the appropriate moving instance based on forward movement (using either a look-ahead window or an overlapping object boundary analysis). Once the object is no longer tracked or exits the FOV, the object data is read to file. Data were included in the cell velocity analysis if the tracking distance was >400 pixels, the cell diameter between 14 and 60 μm and the cell within 50 μm of the chamber bottom (as calculated based on its tracked velocity). Free-flow cell velocities were approximated as 529, 1059 and 1588 at 0.5, 1.0 and 1.5 dyn cm⁻², respectively. Cells were considered to be rolling if their velocities were approximately one quarter of the free flow cell velocity. The average velocity was calculated from the chamber length (1.3 or 14 cm) divided by elution time less the offset time (which accounts for the mean time cells spend in the settling feature before entering the main channel). For the percentage binding time calculations, only rolling cells are considered. Graphs labeled as ‘Total Cells’ or ‘Rolling Cells’ represent individual, pooled, or averaged cell velocity data from all observed cells or cells with instantaneous velocities less than the maximum rolling velocity threshold, respectively.

The flow velocity profile within the parallel-plate flow chamber was simulated using COMSOL (version 4.2.1.110, COMSOL Multiphysics, Stockholm, Sweden). The chamber was modeled in SolidWorks (version 2013 SP4.0, Dassault Systèmes SolidWorks Corporation, Waltham, MA, USA) and exported as a STL file to be imported into COMSOL. After importing the chamber model into COMSOL, the fluid material was chosen to be water (liquid), and the chamber flow profile solution model set to laminar flow. The boundary walls were chosen to be the outer surface of the liquid domain, the initial velocity of the flow was set to 0 m s⁻¹, and the location of the inlet was selected as a surface normal to the inlet circumference. The normal inflow average velocity to the inlet was set to 5.2×10⁻⁵ m s⁻¹, which was obtained by dividing a volumetric flow rate by an inlet area, and the outlet of the flow was selected to be a surface normal to the outlet circumference. Geometry meshing was performed (maximum and minimum element sizes were 343 μm and 22.4 μm and the maximum element growth rate was set to 1.08). The streamline profile for the settling feature was plotted by adding a streamline module. 100 streamline positions were defined at the inlet.

The main channel of the chamber was first incubated with anti-IgG (Fc specific) antibody solution made at desired concentration in D-PBS overnight in 4°C by infusing 50 µl of the solution from the outlet to just cover the main channel. The main channel was subsequently washed with D-PBS, blocked with 1% BSA in D-PBS for 1 h at room temperature, washed with D-PBS, and incubated with the desired concentration of P-selectin in D-PBS for 2 h at room temperature (all infused from the outlet). The main channel was subsequently washed with D-PBS, the whole chamber (settling feature included) blocked with 1% BSA for 1 h at room temperature (infused from the inlet), washed again with D-PBS (infused from the inlet), and stored at room temperature for use the same day. In select experiments, chambers were instead functionalized with BSA alone by overnight incubation with 1% BSA in D-PBS at 4°C. D-PBS used in all steps of the functionalization contained Ca²⁺ and Mg²⁺ except for the anti-IgG incubation step when cation-free D-PBS was used instead.

Cellular P-selectin ligand and sLe^x/a expression levels were measured via flow cytometry. P-selectin-IgG Fc or HECA-452 were premixed with Cy3-conjugated anti-human-IgG or FITC conjugated mouse anti-rat-IgM antibodies, respectively, for 1 h at room temperature. In select experiments, premixes also included 0.5, 5, or 25 units ml⁻¹ of heparin. Cells were incubated with the premixed antibody solutions at 5×10⁶ cells ml⁻¹ for 1 h on ice. Following incubation with antibody premixes, cells were washed, and resuspended in PBS for analysis of 10⁴ events on the Accuri C6 (BD Biosciences, San Jose, CA, USA) for P-selectin ligand expression or the LSRII for sLe^x/a expression (BD Biosciences). Specificity of P-selectin binding was confirmed by incubation of cells in P-selectin premixed with secondary antibody in 30 mM EDTA or incubation with the secondary antibody alone. Specificity of HECA-452 binding was confirmed by incubation of cells in the isotype control premix as well as incubation with the secondary antibody alone. Data was analyzed using FlowJo (Ashland, OR, USA).

Cell diameter measurements were acquired using a coulter counter (Multisizer III, Beckman Coulter).

Data were analyzed using RStudio (0.97.551, Boston, MA, USA) and plotted in Prism (version 5.0b, GraphPad Software). Statistical tests were performed in Prism by ANOVA followed by post-hoc Dunn tests (***P<0.001, **P<0.01 and *P<0.05).

We thank M. Thomas for technical assistance.