Convergent and Divergent Migratory Patterns of Human Neutrophils inside Microfluidic Mazes

Neutrophils are key cellular components of the innate immune response and characteristically migrate from the blood towards and throughout tissues. Their migratory process is complex, guided by multiple chemoattractants released from injured tissues and microbes. How neutrophils integrate the various signals in the tissue microenvironment and mount effective responses is not fully understood. Here, we employed microfluidic mazes that replicate features of interstitial spaces and chemoattractant gradients within tissues to analyze the migration patterns of human neutrophils. We find that neutrophils respond to LTB4 and fMLF gradients with highly directional migration patterns and converge towards the source of chemoattractant. We named this directed migration pattern convergent. Moreover, neutrophils respond to gradients of C5a and IL-8 with a low-directionality migration pattern and disperse within mazes. We named this alternative migration pattern divergent. Inhibitors of MAP kinase and PI-3 kinase signaling pathways do not alter either convergent or divergent migration patterns, but reduce the number of responding neutrophils. Overlapping gradients of chemoattractants conserve the convergent and divergent migration patterns corresponding to each chemoattractant and have additive effects on the number of neutrophils migrating. These results suggest that convergent and divergent neutrophil migration-patterns are the result of simultaneous activation of multiple signaling pathways.


Materials and Methods
Microfluidic devices. Microfluidic migration mazes were manufactured using standard microfabrication techniques. Briefly, two layers of negative photoresist (SU8, Microchem, Newton, MA), the first 10 µm thin, the second 50 µm thick, were patterned on a silicon wafer by sequentially employing two photolithography masks and photolithography processing cycles. The wafer with patterned photoresist was used as a mold to produce PDMS (Polydimethylsiloxane, Fisher Scientific, Fair Lawn, NJ) parts, which were then bonded irreversibly to standard glass slides (75 × 25 mm, Fisher Scientific). The device consists of an inlet and outlet connected to a main channel for cell-loading with an array of 14 side channel mazes, each ending in reservoirs designed to maintain chemoattractant gradients for more than 15 hours 21 . Cells can migrate from the main channel into the mazes of (350 µm by 550 µm), where they are confined inside 10 × 10 µm channels (cross-section). The distance between nodes was 50 µm. Cells that enter the mazes can be analyzed for their migration in the direction of the chemical gradients (along the y axis) as well as orthogonal exploration (along the x axis). The direct route from the cell loading area towards the chemokine source is a pathway of 650 µm. We analyzed the chemokine gradient using fluorescein 5(6)-isothiocyanate (MW = 1 kDa) and dextran labeled fluorescein (MW = 10 kDa, Thermo Fisher Scientific).

Neutrophil isolation and migration buffer.
Human peripheral blood samples from healthy volunteers, aged 18 years and older, were purchased from Research Blood Components, LLC. Peripheral blood was collected in 10  Neutrophil loading in microfluidic devices with mazes. Next, 10 µl neutrophil suspension in a density of 40 × 10 6 neutrophils/mL was loaded in the device and allowed to settle at 37 °C for a few minutes before imaging started. In experiments where the P38 MAP kinase pathway was inhibited, cells were incubated with SB203580 (10 μM, Cell Signaling Technology, Danvers, MA) 30-45 min prior to loading. To inhibit the PI3 kinase pathway, we used LY294002 (50 μM, Invitrogen, Frederick, MD). LTB4 intracellular synthesis was inhibited using MK866 (Abcam, Cambridge, MA).
Device setup for testing the roles of soluble and surface-bound chemoattractants. To test whether chemokines might be bound to the surfaces of our devices after washing, we employed a motility assay that consists of a single, straight, 500 × 70 µm PDMS channel bound to a glass-bottom well plate. The channel was primed with chemoattractant and fibrinogen in media, incubated at 37 °C for 15 mins, washed thoroughly with media, then the entire device covered in media. Fluorescence imaging of FITC-labelled IL-8 before and after the washing was employed to validate the efficiency of washing protocol for removing the chemoattractant from inside the channel, including the corners of the channel.
Neutrophils were loaded as a suspension in a density of 20 × 10 6 neutrophils/mL. After loading, neutrophils were allowed to settle at 37 °C for a few minutes before imaging started. The trajectories of the neutrophils were plotted with reference to their starting position. The velocity of neutrophils was calculated by automated cell tracking of cell movement (TrackMate, Image J) during time-lapse imaging.
Imaging and data analysis. Time-lapse microscopy was performed on a fully automated Nikon TiE microscope with bio-chamber heated to 37 °C, 5% CO 2 using NIS-Elements AR 3.10 software (Nikon Instruments Inc, Melville, NY). Images were acquired every 40 seconds for 3 hours, using a 10X objective. Cells were tracked manually using ImageJ software (NIH). A custom algorithm was then used to map each individual migration trajectory and determine the migratory speed, directionality and persistence of each moving neutrophil. Next, the traffic along each individual node was calculated. From these nodal maps, a "heat map" was generated in MATLAB (MathWorks, Natick, MA), which provides analysis of node passage by the neutrophils in each condition (and thereby area of maze coverage). The percentage of cells migrating was partially manually counted by determining the number of cells that were at T = 0 within 100 µm of the maze entrance. The migratory persistence was calculated by using the following formula: length of the maze/total distance travelled by the cells x the position of the cell when it stopped migrating as a relative number with entrance of maze being 0 and the end of the maze (reservoir) being 1. Nodal maps were visualized with a "heatmap" generated in Python (Python Software Foundation, Wilmington, DE). The bivariate Kernel Density Estimation (KDE) of cell position was plotted with Seaborn KDE plot using a Gaussian kernel.

Statistical analysis.
For most comparisons, statistical significance was determined using the student t test for normally distributed data and the Mann Whitney test for data that did not have normal distribution. Differences were considered statistically significant when p values were less than 0.05.

Results
A microfluidic maze replicates chemical and mechanical features of interstitial space. We designed a microfluidic device to study the migratory patterns of human neutrophils through orthogonal mazes, towards reservoirs of chemoattractants (Fig. 1A,B). The mazes mimic the confinement found in interstitial spaces by restricting the migration of neutrophils through channels with 10 × 10 µm cross-section 18 . Cells within the cell-loading chamber migrate towards sources of chemoattractant through mazes that are 550 µm long and 350 µm wide (Fig. 1C). Motile neutrophils are mechanically confined and their average migration speed is similar to that recorded in vivo 22 . We probed the evolution over time of the chemoattractant gradients for chemokines with a molecular weight less than 1 kDa (LTB4 and fMLF) and ~10 kDa (C5a and IL-8). We found that each gradient persists for more than 3 hours and is linear along the y-axis (Fig. 1D). On the x-axis, higher chemoattractant concentrations are present in the middle of the maze and lower concentration along the periphery (Fig. 1E). The migration trajectories of individual cells in the maze are composed of persistent migration segments along the straight channels 23 and discrete directional decisions at the nodes 19 . This decoupling of persistence and directional decisions allows for precise quantification and comparisons of migration patterns in the presence of a chemoattractant. Moreover, the device is ideal to differentiate a directed migration pattern (when cells traverse the maze along the long axis towards the chemokine reservoirs) from a random migration pattern (when cells explore the maze in the transversal direction).

Convergent migration patterns induced by fMLF and LTB4.
We analyzed human neutrophil responses to LTB4 and fMLF, two chemoattractants which induce a "fast and persistent" migration signature in straight channels 20 . Here, we find that both chemoattractants stimulated the migration of neutrophils into the mazes and their subsequent migration towards the chemoattractant reservoirs. LTB4 elicits a potent chemoattractant response when neutrophils are exposed to gradients ranging from 0.1-100 nM (Movie 1). After 3 hours, 24 ± 22% of the cells loaded into the device completely traverse the maze into the chemokine reservoir, while 7 ± 6% remain in the maze. This overall migration pattern was consistent with most migrating cells (73 ± 20%) moving in a directed manner towards the chemokine source (10 nM, Fig. 2A). The percentage of migrating cells increased from 6 ± 5% to 45 ± 28% in a concentration-dependent response to the chemoattractant concentration increase from 0.1 to 100 nM.
Similarly, fMLF (gradients of 0.1-100 nM) stimulated neutrophils in a manner such that they traverse mazes into the reservoirs (13 ± 7% in the reservoirs vs. 22 ± 16% remaining in the mazes, 10 nM). The fraction of cells that migrate towards the chemokine source was lower than that observed in response to LTB4 (42 ± 19%, p < 0.001) and was concentration-dependent from 0.1 to 100 nM. Notably, we observed that the highest neutrophil response occurred at a concentration of 10 nM. The number of neutrophils migrating decreased by ~10% in response to 100 nM (vs. 10 nM) LTB4 (Fig. 2B).
Analysis of individual trajectories shows unique patterns for each chemoattractant. Both LTB4 and fMLF stimulate migratory responses with high directionality towards the source (Fig. 3A,B). Analysis of individual cells shows that their average migratory speed was 27.4 ± 5.0 µm/min for LTB4 and 15.2 ± 4.4 µm/min for fMLF (Fig. 4A, p < 0.001). LTB4 also induces transmigration across the maze, within an average time of 37 ± 15 min, covering an average of 912 ± 227 µm during this period, while the shortest route is 650 µm (Fig. 4B). The Directional Persistence of migrating cells is 0.71 ± 0.11 (Fig. 4C). In contrast, neutrophils responding to fMLF do not always migrate along the direct and shortest route (Fig. 3B). The average distance traveled within the maze is 1484 ± 580 µm, the directional persistence is 0.42 ± 0.16, and the time to reach the reservoir is 93 ± 21 min (Fig. 4B,C). Divergent migration patterns induced by C5a and IL-8. We analyzed the migratory phenotypes in response to the chemokines C5a and IL-8, which we previously reported to induce bi-directional migration 20 . Although a significant number of neutrophils (17 ± 10% and 10 ± 7%) respond to chemokines in the device, a remarkably smaller number (0.2 ± 0.7% and 0.2 ± 0.8%) show directional migration towards C5a and IL-8, respectively (10 nM, Fig. 2C,D). Overall, we find that these two chemoattractants stimulate exploratory migration (Fig. 3), independent of their concentration. Different concentrations change the number of cells migrating, but  not the migration patterns of the cells (Fig. 2C,D). Interestingly, we found that the highest C5a concentrations (100 nM) inhibit migration compared to lower concentrations (Fig. 2C), consistent with a bell shaped curve of cell responses to chemoattractants 24 . Overall, the chemokine C5a (10 nM) stimulated exploratory neutrophil migration within the first 31 ± 17% of the maze, close to the entrance, and IL-8 (10 nM) stimulated exploratory patterns within the first 36 ± 11% of the maze (Fig. 4D, Movie 2). Neutrophils migrated a total distance of 1020 ± 760 and 690 ± 610 µm in response to C5a and IL-8, respectively (Fig. 4B). The average migration speed through mazes was 9.8 ± 7.9 µm/min for C5a and 3.2 ± 2.4 µm/min for IL-8, which is significantly slower than the velocity induced by LTB4 or fMLF, 27.4 ± 5.0 and 15.2 ± 4.4 µm/min, respectively (Fig. 4A, p < 0.001).

Chemokines do not bind to device surfaces and do not induce neutrophil motility. Fluorescence
imaging of FITC-labelled IL-8 before and after washing the straight channel devices showed that no fluorescent signal remained following washing compared to the media control. This suggests that FITC-IL-8 was not bound to the channel surfaces in any significant amounts (Fig. 5A). Tracking of neutrophil motility over 4 hours demonstrated that soluble IL-8 was sufficient to induce significant random migration of neutrophils, as indicated by increased migration velocity in the first hour of tracking (Fig. 5D). However, the motility of neutrophils in washed channels was indistinguishable from media control (Fig. 5B-D), suggesting that soluble IL-8 plays a dominant role in the neutrophil motility response. Similar changes in neutrophil motility were measured when comparing the response to soluble fMLF before and after washing the channel. Whereas soluble fMLF induces significant migration of human neutrophils, after the fMLF wash the motility of neutrophils was indistinguishable from the media control (Fig. 5C,D).

Exposure to collagen gels does not change neutrophil migration patterns in channels.
To determine if migratory patterns change in a true 3D environment, we filled the microchannels with a collagen suspension (rather than coating the channels with fibrinogen). Subsequently, we evaluated the migratory signature, migratory speed, and persistence in this environment. Neutrophil migration speed in response to LTB4 and C5a was similar when collagen was present inside the channel to that observed in fibrinogen coated channels. We measured only a small decrease in the migratory speed in response to LTB4, from 27.4 ± 5.0 to 24.2 ± 5.4 µm/ min (Fig. 6A, p < 0.05). The patterns of migration were similar towards LTB4 (persistent migration) and C5a (exploratory migration - Fig. 6C,D). Furthermore, neutrophils that migrated through collagen in response to C5a explored larger areas of maze compared to fibrinogen-coated channels, representing 44 ± 23% and 31 ± 17%, respectively (Fig. 6D, p < 0.05). These migratory phenotypes are comparable to those on bare glass (data not shown).

Combinations of chemokines show additive migration signatures.
We measured the effect of combinations of chemokines on human neutrophil migration through mazes and their ability to reach the reservoir. In response to combined LTB4 (100 nM) and C5a (1 nM) chemokines, there was an increase in the number of neutrophils entering the maze, from 15 ± 8% and 17 ± 12% with each chemokine alone, up to 34 ± 13% in response to combinations (Fig. 7A, Movie 3). Similarly, 31 ± 6% of neutrophils enter the reservoir in response to LTB4 (100 nM) vs. 0 ± 1% in response to C5a (1 nM). This is is comparable to the 32 ± 9% response in the presence of combined LTB4 and C5a (Fig. 7A). We observed similar changes in migratory neutrophil populations in response to combinations of low concentrations of LTB4 (1 nM) with high concentrations of C5a (100 nM, Fig. 7B, p < 0.01). Heat maps confirmed the additive responses. Superimposed migration patterns -inside the maze and reservoir -are consistent with the combined migratory signatures of LTB4 and C5a (Fig. 7E). We did not observe an additive effect during the neutrophil response to combinations of fMLF (100 nM) and IL-8 (1 nM) (Fig. 7C). Mixing lower concentrations of fMLF (1 nM) with higher concentrations of IL-8 (100 nM) enhanced directed migration towards the chemokine reservoir (Fig. 7D, p < 0.01). Distinct migratory patterns are not linked to any signaling pathway alone. To determine whether select signaling pathways are responsible for either divergent or convergent migratory behavior we used pharmacological inhibitors to selectively target the P38 MAP kinase (SB203580, 10 µM) or the PI-3 kinase (LY294002, 50 µM) pathways, as previously described 9 . We found that these concentrations of SB203580 (10 µM) and LY294002 (50 µM) used in this assay were nontoxic, and cells migrated in response to agonists.

Inhibition of LTB4 synthesis reduces the number of migrating neutrophils.
Neutrophils are known to secrete chemokines during inflammatory responses which may alter migration patterns. For example, neutrophils produce LTB4, which plays an important role in swarming 4,7 . Since we observe that LTB4 induces convergent migration, we explored whether LTB4 also potentiates this pattern of migration when other chemokines form the primary gradient. We tested the effects of LTB4 synthesis inhibitor MK866 on neutrophils migrating towards fMLF. Compared to our control experiments, inhibiting LTB4 decreases the percentage of cells migrating from 39 ± 12% to 22 ± 1% (Fig. 9A, p < 0.05). The migratory speed of the neutrophils did not change (Fig. 9B) However, there was a significant reduction in the total distance traveled by neutrophils before coming to a stop, from 1073 ± 340 to 741 ± 300 µm (Fig. 9C). The directional persistence of migrating neutrophils towards fMLF increases from 0.46 ± 0.16 to 0.63 ± 0.13 when LTB4 synthesis was inhibited (Fig. 9D).

Discussion
We employed microfluidic devices with 2D orthogonal mazes to quantify the migration patterns of human neutrophils towards various chemoattractants. The confinement of moving cells in small channels inside the devices allows the decoupling of the measurements of speed and persistence (along the straight channels), from directional decisions (at the nodes). This is unlike the migration on flat surfaces (e.g. micropipette, under-agarose, Zigmond chamber, Dunn assays), when neutrophils change continuously both their speed and their direction of migration. Instead, the migration through mazes takes place at constant speed 23 and the direction changes are discrete i.e. neutrophils migrate through one of four orthogonal channels at most nodes 25 . The particularities of migration through mazes helps decrease the noise of measurements and increase the precision of  26 . Thus, the maze assay enables unprecedented levels of precision for the measurements of speed, directionality, and persistence of the moving neutrophils and helps discriminate between migration signatures that could have not been distinguished by other assays.
Our studies provide insights into human neutrophil migration by showing that two patterns exist, divergent and convergent. These migration patterns may be relevant to the variety of roles that neutrophils play during inflammatory conditions. Rapid neutrophil convergence to sites of bacterial proliferation and their accumulation in swarms is consistent with the ability of neutrophils to accurately follow gradients of fMLF and LTB4 4,7,27 . The spreading of neutrophils through tissues, which favors the clearance of bacterial debris and enhances the surveillance of tissues under stress is consistent with the measured neutrophil divergent motility signature in the presence of C5a and IL-8. The fact that neutrophils retain both migration signatures in response to combined stimuli may be explained by the presence of subgroups of neutrophils. This possibility is supported by recent results documenting the presence of distinct neutrophil subpopulations during inflammation 28 and within cancer tissues 29 . An alternative explanation is that neutrophils exposed to combined stimuli make one stochastic choice between the stimuli and subsequently maintain one migration signature. This possibility is supported by observations of stochastic responses of neutrophils during exposure to repeated stimuli, when the time to respond and the subpopulation of neutrophils responding changes from one stimulus to the next 30 .
Several studies have focused on the mechanisms of how neutrophils prioritize competing cues 8,9 and building better tools to study this process 31 . Based on experiments where the migration was evaluated under conditions of opposing gradients, it was proposed that receptor desensitization plays an important role, or that certain chemokines stimulate different signaling pathways 9,32 . Complementing these early studies, we find that chemoattractant gradients in the same direction do not compete with one another and their effects are additive. The conditions we evaluate in this report may be relevant to in vivo situations, where neutrophils encounter overlapping gradients of chemokines that originate in the same local area. Our findings also suggest that the two previously proposed, mutually-exclusive signaling pathways may not adequately explain convergent and divergent migration patterns. We could not confirm that C5a elicits discrete signals through the P38 MAP kinase pathway to dominate over the PI-3 kinase pathway activated by LTB4, as suggested in previous studies using opposing gradients 9 . Instead, our data suggest that both the P38 MAP kinase and the PI-3 kinase pathway function in chemoattractant responses. Inhibiting either signaling path alone decreases the size of migratory populations, without altering the pattern of migration. The results of the under-agarose assay used in previous studies are dominated by the number of neutrophils moving, and thus the measured effect may be limited to the size of migratory populations. The under-agarose assay is less sensitive to the patterns of migration, which are the unambiguously decoupled from other changes in our maze assay.
An interesting finding is that in the setting of neutrophil stimulation with two chemotactic peptides (fMLF and IL-8), inhibition of either P38 MAP kinase or PI-3 kinase did not significantly reduce the number of migrating cells. However, when neutrophils encounter fMLF or IL-8 alone, inhibition of either signaling pathway results in significantly lower numbers of migratory cells. It is possible that when two different chemotactic molecules stimulate a cell simultaneously, there is enhanced threshold activity and/or crosstalk between the p38 MAP kinase and PI-3 kinase pathways to overcome the inhibition 33 . These findings again suggest that the two dominant migratory patterns are mechanistically complex and not just the result of discrete signaling through one pathway. The molecular mechanisms responsible for the two distinct migration patterns remain to be elucidated in future studies.
Both convergence and divergence patterns of migration appear to be relatively independent of the concentration of the chemoattractant. Exploratory patterns of neutrophil migration in response to IL-8 have been previously observed in vivo 34 . Although these migration patterns were proposed to be due to the degradation of IL-8 by matrix metalloproteinases, our study suggests that the exploratory migration patterns in response to IL-8 are intrinsic to the neutrophils and independent of the IL-8 concentration. Moreover, the convergent and divergent migration patterns observed inside the mazes hold true in a 3D environment, when the maze channels are filled with collagen. Our results suggest that the migration pattern and/or signatures are cell and chemokine intrinsic and they are not related to the physical microenvironment surrounding the cells. Finally, we examined the effect of paracrine signaling on migration phenotypes. We found that when LTB4 synthesis is inhibited during neutrophil migration towards gradients of fMLF; the number of migrating cells decrease, the migration paths are shorter, and migratory persistence is increased. These results are consistent with several reports 4,7,27,35 and suggest that paracrine LTB4 signaling plays an important role in recruiting cells to an inflammatory focus, potentiating the local inflammatory reaction.
Formation of haptotactic chemokine gradients is important for neutrophil migration in response to some chemokines but not others 36 . Although microfluidic devices have been developed specifically to dissect the differential contributions of superimposed soluble and haptotactic gradients to cell migration 37 , this was not possible using our current maze device design. Surface-bound IL-8 has been shown to induce random migration by human neutrophils 38,39 . A simpler device was therefore employed to compare the motility induced by soluble and bound gradients. These experiments helped verify that soluble gradients in our microfluidic devices dominate the neutrophil responses.
In conclusion, we developed a new set of tools that enable single cell resolution, quantitative measurements of neutrophil migration patterns in response to chemoattractants. We describe two distinct migration patterns, which we name divergent and convergent. Furthermore, we find that they are differentially induced by C5a or IL-8 and fMLF or LTB4, respectively. The new microfluidic tools will enable further studies on the requirements for efficient responses to infectious pathogens and/or could lead to new avenues of investigation on the restoration of tissue homeostasis after inflammation.