A Haptotaxis Assay for Neutrophils using Optical Patterning and a High-content Approach

Neutrophil recruitment guided by chemotactic cues is a central event in host defense against infection and tissue injury. While the mechanisms underlying neutrophil chemotaxis have been extensively studied, these are just recently being addressed by using high-content approaches or surface-bound chemotactic gradients (haptotaxis) in vitro. Here, we report a haptotaxis assay, based on the classic under-agarose assay, which combines an optical patterning technique to generate surface-bound formyl peptide gradients as well as an automated imaging and analysis of a large number of migration trajectories. We show that human neutrophils migrate on covalently-bound formyl-peptide gradients, which influence the speed and frequency of neutrophil penetration under the agarose. Analysis revealed that neutrophils migrating on surface-bound patterns accumulate in the region of the highest peptide concentration, thereby mimicking in vivo events. We propose the use of a chemotactic precision index, gyration tensors and neutrophil penetration rate for characterizing haptotaxis. This high-content assay provides a simple approach that can be applied for studying molecular mechanisms underlying haptotaxis on user-defined gradient shape.


Results
High-content framework. We have developed a framework for a high-content haptotaxis assay based on the well-established under-agarose assay (Fig. 1). In this approach, we control the camera, the light source and the motorised stage to acquire several movies in parallel and automatically track thousands of fluorescently labeled human primary neutrophils (Supplementary Videos S1 and S2). To avoid phototoxicity and photobleaching while accurately detecting and tracking cells, we carefully minimized the illumination and exposure during the assay. These adjustments were made by monitoring signs of toxicity such as morphological changes from amoeboid-like shape to round apoptotic cells (Supplementary Video S3).
Optical patterning. We modified the classic under-agarose assay, in which cells migrate toward soluble chemoattractant gradients, by micro-engineering substrate-bound gradients of a fluorescein-tagged formylated peptide (Fig. 1). LAPAP technology consists in varying the illumination power of a laser and moving the sample to tailor the geometry of the gradient produced by the photobleaching of the excited fluorophore 35 . We produced gradient of constant and continuous slopes to yield 4% and 8% difference in peptide concentration over the typical length of a cell ( Supplementary Fig. S1). At saturation, this corresponds to an estimated 2.4 × 10 4 molecules per cells (Supplementary Fig. S2 and Supplementary Methods). The pattern dimensions allow long and numerous cell tracks (Fig. 2a). After cells have migrated for two hours over the gradient, we detected a positive correlation between the cell density at any given position and the formyl peptide concentration (r = 0.97, p = 0.0078) ( Fig. 2b and c, and Supplementary Video S2). There was no such correlation in the control assay of non-formylated peptide patterns (r = 0.53, p = 0.3554) (Fig. 2b and c).
Detecting and Tracking neutrophils. We processed the acquired data by a multi-step image analysis approach. In the pre-processing phase, the exact geometry of the gradients was registered. The cell detection algorithm consisted of a feature enhancement step, increasing the signal, followed by an object detection step (Fig. 3). Feature enhancement was performed by first subtracting a background image produced by using a morphological opening operation and then applying a frequency filter. The object detection was done by localising the high intensity peaks. Our approach shows a sensitivity in detecting cells with ~80% accuracy.
Cell tracking was done using a method that dynamically adapts to local cell density 37 . On average, we detected 148,580 trajectory steps per experiment (Supplementary Table SI). Typical tracks extend on average for 109 µm before cell arrest or truncation by the tracking algorithm. This yields an average of 2133 tracks/assay with neutrophils migrating at a speed around 10-20 µm/min.

Chemotaxis indices.
Neutrophils migrating under-agarose behave differently in response to soluble or substrate-bound formyl peptides. Under both conditions, neutrophils tracks are first elongated; cells moving away from their original position (the cell seeding well). However, a high fraction of cells moving on surface-bound formyl peptide eventually slow down and accumulate in the regions of high peptide concentration, still exhibiting amoeboid shapes but moving randomly around a central position ( Fig. 4a and b, Supplementary Video S4). To describe this behaviour, we opted for an approach based on the gyration tensors of single cell track. Refer to Methods for a detailed explanation of the tensors. Intuitively, the gyration tensor provides a geometric representation of the distribution and shape of the coordinates defining a trajectory. The gyration radius (Rg) represents the spread of the cell locations, while the parameter A 2 indicates the degree to which the trajectory shape is linear. By studying the temporal evolution Rg (R g 2 N (t)) for each track, starting from the end of the track, we detected this dual behaviour (elongation followed by random walk) of cells migrating on substrate-bound gradient (Fig. 4c). Indeed, while cell positions spread monotonically in the classic under-agarose assay, cell positions plateau on substrate-bound gradients. As the graph shows the tracks from end to the start, the plateau (in green) is a very gentle slope lasting for at least the last half of the tracks. In the diffusion assay, the evolution of the slope is more linear and steeper. Thus, in haptotaxis assays, 31% of neutrophils eventually move randomly around an end position within 2 hours of migration as opposed to only 10% of the cells migrating along a soluble formyl peptide gradient (Fig. 4d). Truncating the standstill part of these tracks, we detected a positive correlation between the concentration of the peptide and the extracted centroid (Fig. 4e). Combining the results of 8 assays, we computed a Pearson correlation coefficient of 0.77 (p = 4.39 × 10 −9 ).
Rink et al. 32 recently suggested characterising chemotaxis by calculating the Chemotactic Precision Index (CPI) instead of the more common Forward Migration Index (FMI, also known simply as the Chemotactic Index or as the McCutcheon Index) [38][39][40] . They claim that CPI is a better indicator of chemotaxis because it reports simultaneously on three aspects of the cell movement: it increases as the cell moves directly toward an end point (directness), it increases if the movement is in the gradient direction (cos 2 φ, where φ is the angle between FMI and directness), and it decreases if the displacement diverts in other direction (1-|SMI|, where SMI stands for Side Migration Index). We tested the robustness of both FMI and CPI on data from of our haptotaxis assay as well as the classic under-agarose assay. In the latter assay, both indexes increased when neutrophils migrated towards a soluble fNLFNTK/FITC gradient versus the non-chemotactic control (Fig. 5). When cells migrated on surface-bound formyl peptides, since we oriented the gradient perpendicular to the radial axis of the seeding well, we expected that the intrinsic movement of cells moving radially away from the well should balance the migration in the gradient direction. Indeed, while the FMI, which considers only the forward force, was falsely elevated, the CPI was low, since both side and forward forces balanced each other. Speed and Orientation Dynamics. To investigate whether the correlation between cell density distribution and the gradient concentration was due to changes in cell speed (orthokinesis) or in turning angle (klinokinesis), we compared the distribution of speed and orientation of all trajectory steps. Regardless of their relative orientation to the gradients, neutrophils always moved faster in regions of high peptide concentration. In the classic under-agarose assays, the velocity was approximately 20% (n = 2) higher in the high fNLFNTK/FITC concentration regions than in the low ones (Fig. 6a). Although the correlation was subtler in the haptotaxis assays, neutrophils also migrated significantly faster on the high peptide concentration regions (mean value of 5.7%, n = 8) (Fig. 6b). This correlation was slope dependent, since the differences in speed were more pronounced for steeper slopes than gentler ones ( Supplementary Fig. S3). Likewise, in the classic chemotaxis assays, we detected a strong correlation between the speed and the orientation of the cell to the gradient (Fig. 6c). For the sake of simplicity, we considered the soluble gradient to be linear and we defined a trajectory step as oriented if it was within 45° of the gradient direction. A step was considered as opposed to the gradient if it was within the opposite 45° of the gradient direction. In the haptotaxis assays, we couldn't detect a correlation between the step speed and orientation (orthotaxis) (Fig. 6d).
In the absence of chemoattractant, analysing the distribution of cell orientation in time, we observed that neutrophils were preferentially oriented radially to the seeding well. This observation holds true especially at early time points in the migration assay, as neutrophils spread out more randomly with time ( Fig. 7a and b). We also measured the persistence of cell orientation in time by analysing pairs of step angle separated by different time lags. Figure 7c,d,e and f depicts the distribution of step angles classified as cells going forward or backward for 10 seconds, 1 minute or 10 minutes delays, respectively. A cell was defined as moving forward when its step was within [−45°, 45°] of the orthogonal direction of the well edge, pointing away from it. If a cell was initially going forward, it had a higher probability to be pointing in the same direction at a later time point, as indicated by the skewed distribution in 7c, d and e. On the other hand, if a cell pointed backward [135°, 225°], it had equal chances to be pointing backward or forward later on (Fig. 7f).
Penetration rate. To further characterize the correlation between neutrophil location and substrate-bound peptide concentration, we calculated the rate at which cells left the well to penetrate the under-agarose space. Comparing the number of tracks initiated on two 400 μm wide adjacent homogenous patterns of high (at saturation) or low concentration (20% of saturation) over 2 hours, we found that neutrophils exited the seeding well at a considerably faster pace when they moved over the high concentration pattern (Fig. 8 and Supplementary Video S5).

Discussion
We used an optical patterning technique to generate surface-bound formyl peptide gradients with precision of approximately 1 µm, determined by the diffraction limit 35 . Alignment of gradients to the edge of the cell seeding well and time-lapse imaging allowed the analysis of a large number of migrating neutrophils under diverse conditions. As opposed to the Boyden, Zigmond and Dunn chamber assays, optical patterning results in well-defined surface-bound gradients that are highly reproducible in shape. The stability of the gradient is ensured by the covalent peptidic cross-link to surface, thereby preventing the formation of a soluble gradient in close proximity to the surface, which might happen using the hydrogel stamping technique 32 . Although in vivo neutrophils are likely exposed to both immobilized and soluble gradients, temporal fluctuations can complicate interpretation of results from in vitro migration assays.
Our results show that, in addition to soluble or stamped formyl peptides, neutrophils sense and respond to covalently bound cues. The average velocities detected in the high concentration areas are larger than in the low concentration areas of the immobilized gradient. Such high velocities were observed until the cells change from haptotaxis to haptokinesis. At this stage, the trajectories of cells are random, of low velocity and centered on an end-point. The observed dual behaviour, mimics the in vivo situation where neutrophils stop as they reach their end target [41][42][43] . We propose using the squared radius of gyration to describe neutrophils accumulating at the end targets. This parameter is particularly suitable to assess the shape of a trajectory in time.
Our modification to the classic under-agarose assay allows a simple way to test arbitrary protein patterns. The slope ranges were chosen from previous studies that showed that neutrophils are sensitive to 1-10% differences in concentration 22,44 . Neutrophils exhibited more robust responses to steep gradients than gentle ones. These findings are consistent with a previous study, which reported a positive correlation between responses to CXCL8 and the slope of the gradient 45 . Furthermore, neutrophils were reported to preferentially migrate on a steeper gradient in bifurcating microfluidic channels containing different chemokine gradients 46 .
Neutrophils chemotaxis is governed by multiple cues generated by infectious agents and resident immune cells, including the neutrophils themselves. In our assay, we found that neutrophils tend to move away from the well, resulting in a bias in the forward direction even in the absence of a chemoattractant or when the gradient was perpendicular to the forward direction. This suggests that neutrophils might secrete chemokines in the seeding well, thereby confounding the influence of the immobilized gradients. While such chemorepulsive action has long been recognized, it was considered to have minimal impact at standard cell density in the under-agarose chemotaxis assay 19,47 . However, our results ( Fig. 7 and Supplementary Figures S4 and S5) indicate that at higher cell densities (such as that inside the well), neutrophil-secreted chemokines, such as CXCL8, might reach sufficiently high concentrations to evoke chemorepulsion via paracrine signalling 48 . Indeed, CXCL8 at low concentrations is chemoattractant for neutrophils, where as it becomes repulsive at concentrations higher than 1 µM 49 . This biased movement seems to differ from reverse migration, which has been attributed to desensitization to the chemoattractant 50 . Strong chemorepulsion may also explain why we failed to observe a clear preferential orientation of cell movement to the substrate-bound gradient oriented perpendicular to the well radius, regardless of the time of exit or the distance from the well (data not shown). A weak preferential orientation can be extrapolated from Supplementary Fig. S5e, where the cells are pulled sideward from their initial orientation at 10-minutes interval. However, in all other cases, they are oriented persistently in time. The chemorepulsion effect observed is also consistent with the lack of significant difference in CPI vs. vehicle control. The distinct responses to soluble and surface-bound formyl peptides can be explained by differences in the topographic distribution, association/ dissociation, dimerization or internalisation rate of the formyl peptide receptors, though these events remain to be investigated 51, 52 . Hoffman et al. 51    (assuming a cell size of 10 μm × 10 µm). For comparison, exposed to 50 nM fMLF in solution, for a volume of 10 × 10 × 10 μm, a cell would perceive 3.0 × 10 4 diffusing molecules. While this manuscript was under revision, Schwarz et al. 33 reported a similar LAPAP patterning gradient of CCL21 and the calculated number of molecules/ μm 2 was in the same order of magnitude as in our study.
We propose the penetration rate as an index describing the dynamics of the cells exiting the well and entering the under-agarose space. We explain the correlation between the neutrophil density and gradient concentration shown on Fig. 4 by the modulation of the penetration rate to the peptide concentration combined with cell accumulation at end points, as described by the gyration tensor.
In summary, we developed a novel high-content under-agarose assay for studying haptotaxis in vitro, where the geometry of the surface-bound chemical gradients can be controlled with micrometric precision. Our assay offers the advantage of studying 3D events in a 2D plane by confining neutrophils in space, thereby mimicking interstitial cell movement such as integrin-independent migration 34,46,53,54 . These assays are of particular interest when studying low adhering leukocytes such as neutrophils or dendritic cells 34 . While in this study we fabricated formyl peptide gradients bound to BSA-coated glass surface, LAPAP can be performed to immobilize other small molecules, like chemokines, to many matrices 33,34,36 . To successfully immobilize other cues than fMLF, one has to carefully design the binding moiety, taking into consideration the inherent adhesiveness of the molecule to the substrate and the proper presentation of the binding domain to the cell surface receptors 33,34 . Furthermore, LAPAP can produce continuous patterns, which could facilitate density-dependent cell spreading and/or adhesion studies. LAPAP represents an alternative to microcontact printing and recently has also been applied to microfluidic platforms 33 . Chemoattractants may have anionic or hydrophobic sites that are potential areas for surface binding 55 . Thus, extracellular matrix-bound CXCL8 55 , surface-bound PAF 56 , C5a 7 and exosomes presenting LTB4 gradients 57 were reported to evoke haptotaxis. fMLF can bind to albumin 58,59 , to neutral endopeptidase (EC 3.4.24.11) 60 and contribute to the association of neutrophils to fibrin and plasma clots 61, 62 , triggering migration. Since fibrin, albumin and neutral endopeptidase are present within inflamed or injured tissues 63 , it is plausible to assume that immobilization of fMLF does occur. Neutral endopeptidase cleaves formyl peptides and therefore is well-suited to modulate local levels. We are unaware of any direct demonstration of gradients of surface-bound chemoattractants extending from endothelial junctions through the intact to inflamed tissue in vivo. However, recruitment of neutrophils to inflamed tissues is likely regulated by subsequent chemoattractant gradients with formyl peptides being responsible for the final step of migration into the damaged area 3,4 . We believe our assay presents a haptotaxis design, where patterning fMLF can be used for the assessment of neutrophil migration as a proxy for pathology 64,65 . With our imaging and data processing system we can study a high number of cell trajectories, revealing previously unknown subtleties of haptotaxis. Thus, we showed that human neutrophils do respond to covalently-bound formyl peptide gradients. Neutrophils exit the well preferentially in the regions of high peptide concentration (penetration rate). In these regions, cells first speed up and suddenly stop migrating (although they continue to move randomly) and end up accumulating, passing from a haptotaxis behaviour to haptokinesis. This behaviour of neutrophils accumulating on surface-bound patterns mimics the in vivo situation where cells stop migrating after reaching their target. Our results also document a chemo-repulsive effect originating from the high cell concentration in the seeding well. Finally, as differences in neutrophil responses to soluble and surface-bound gradients are being increasingly recognized, our high-content assay provides a straightforward and inexpensive approach for studying haptotaxis and the underlying molecular mechanisms under conditions that mimic the in vivo situation.

Methods
Neutrophil isolation and staining. Neutrophils were isolated 66 from the venous blood of healthy volunteers who had denied taking any medication for at least 2 weeks. In brief, the isolation method is based on the standard Bøyum procedure 67 using lithium heparin as an anti-coagulant. On regular basis, we have monitor parameters indicating activation such as L-selectin shedding, CD11b upregulation, superoxide/ROS formation and found negligible or no activation during our isolation procedure. Neutrophils (5 × 10 6 cells/ml, purity >96%, viability >98%) in Hank's balanced salt solution supplemented with 10% autologous serum were stained with the dsDNA dye LDS751 (Molecular Probe) for 5 minutes at 37 °C 68 . The cells were then washed and resuspended in RPMI 1640 medium containing 5% FBS. 2.5 × 10 5 cells were used per experiment. All methods were performed according to the Maisonneuve-Rosemont Hospital guidelines and regulation. The Clinical Research Committee of the Maisonneuve-Rosemont Hospital approved the experimental protocols. The authors confirm that written informed consent was obtained from all subjects.
Optical patterning of substrate-bound gradients. We fabricated the haptotactic gradients optically, using a variation of LAPAP (Fig. 1a) 36,69 . In brief, formyl-Nle-Leu-Phe-Nle-Tyr-Lys coupled with FITC (fNLFNTK/FITC, Molecular Probe) or a custom-synthesized negative control non-formylated peptide of the same sequence (H 2 N-Nle-Leu-Phe-Nle-Tyr-Lys/FITC, New England Peptide, USA) at a concentration of 0.5 mg/ ml was incubated on BSA-coated glass bottom culture dishes (MatTek). The peptide was cross-linked by focusing a 473 nm laser (waist before the lens ~1 mm) on the glass surface with a 38 mm focal length lens. At this wavelength, the fluorophore photobleaches, liberating free-radicals that facilitate the adsorption of the peptide at the surface. We produced the concentration gradient by varying the power of the laser between 40 µw (minimal condition for peptide adsorption) and 400 µw (saturated peptide adsorption). We designed the shape of the pattern by moving the sample with a motorized stage at 0.1 mm/s controlled by a custom-made LabVIEW software (National Instruments, TX, USA). Under-agarose chemotaxis and haptotaxis assays. Low-melting agarose (UltraPure), solubilised in Earle's Balanced Salt Solution (EBSS, Gibco) and brought to a 3% final concentration in RPMI 1640 medium supplemented with 10% FBS (Gibco) was poured onto glass-bottom culture dishes to obtain a ~3 mm layer. Upon gelling, a cylindrical piece of agarose of 2 mm in diameter was removed with a chirurgical punch to seed neutrophils, corresponding to a volume of 10 µl. For the haptotaxis assays, the peptide gradients were patterned on the glass surface before pouring the gel. The cell seeding well was punched such as to expose a section of the gradient (Fig. 1b). For the classic under-agarose assays, chemoattractant or vehicle solution was added in wells punched 2 mm away from the central well. We used fNLFNTK/FITC diluted in RPMI 1640 (10% FBS) at 200 nM. This chemotaxin diffuses through the gel generating a concentration gradient (Fig. 1c).
Data acquisition. Typically, LDS751 dye-labeled neutrophils were imaged in epifluorescence every 5 seconds for approximately 2 hours with an excitation power of 300 µw (~1 nW per cell) during 100 ms, using a mCherry filter (excitation: 542-582 nm, emission: 603-683 nm). A custom-made LabVIEW program controlled a Retiga 2000R CCD camera attached to an inverted microscope (Olympus IX71, Japan) equipped with a 10x objective and a motorized stage. We obtained high throughput by multi-dimensional acquisition, programming the stage movement for sequential images at different locations of the sample. Throughout the assays, neutrophils were kept at 37 °C with 5% CO 2 using a stage top incubator (Tokai Hit, Japan). Fluorescent formyl peptide patterns were imaged with a FITC filter (excitation: 463-505 nm, emission: 515-565 nm) using 2 s exposition time.
Automated detection and tracking of neutrophils. We used MATLAB scripts (The MathWorks, MA, USA) for image analysis, neutrophil detection and tracking, as well as for describing cell trajectories and performing statistical analysis. The scripts are freely available upon request.
For each frame, we estimated the background intensity applying a morphological opening on the original image using a circular structuring element with twice the average diameter of neutrophils. We subtracted the background from the original image and enhanced the objects of neutrophil size by applying a band-pass filter. We defined the position of a cell at the highest peak within each region above a threshold 70 . For cell tracking, we employed an algorithm developed in-house 37 . Detection sensitivity. To assess the detection sensitivity, we acquired a set of data of 300 frames over 20 minutes, where cells where alternatively imaged either in bright field or fluorescence. The ground truth cell count for each frame was determined manually, using the bright field images data set. We calculated the sensitivity by dividing the true detected cell count by the ground true count.

Gyration tensor analysis.
We computed the gyration tensor to analyze cell migration dynamics. We analyzed the evolution of the gyration radius (R g ), which is derived from the gyration tensor, as a function of time 71 . The gyration tensor is defined as  are the Eigenvalues of T. Small R g values indicate a static period, since the cell positions are localized randomly around a same point. In contrast, during migration, the cell moves away from the starting position, yielding significantly larger R g values. We have defined a time-normalised squared radius of gyration R g 2 N (t), which is computed using Eqs 1 and 2 by calculating the time average only over time frames later than t as: and dividing it by the number of frames involved in the calculation. The moment where R g 2 N goes below a threshold determines when a cell reaches a static state.
Additionally, the parameter A 2 , derived from T, describes the elongation of the trajectory: Thus, an A 2 close to 1 represents elongated trajectories, whereas A 2 values around 0 indicate random walk trajectories. The time averages were computed as in Eq. 4.
We filtered the static end-segment of the tracks before calculating other parameters described afterwards. The thresholds for discarding end-segments were R g 2 N < 0.1 μm 2 /frame and A 2 < 0.6.
Scientific RepoRts | 7: 2869 | DOI:10.1038/s41598-017-02993-6 Chemotactic indices. Forward Migration Index (FMI), the Side Migration Index (SMI), the total length (l i ), the directness and the Chemotactic Precision Index were calculated as follows, using the terminology defined by Rink et al. 32   where φ is the angle between FMI and directness. To standardise the comparison between the classic chemotaxis assay in diffusion and the haptotaxis assay, we chose to define the forward direction as the radial axis of the well, regardless of the direction of the substrate-bound gradient. Mean cell speed within a trajectory and instant cell speed at any step point of a trajectory were calculated by dividing the total trajectory by the total time travelled (trajectory speed) and by dividing the length of each step by the time lag between frames (step speed). In order to evaluate the impact of the cell orientation on the speed, we also measure the angle between the direction of the immobilized gradient and the direction of each trajectory step.
Penetration rate. To assess the rate at which cells enter the agarose-glass interstice, we counted, at consecutive intervals, tracks initiated nearby the well edge (within 100 µm from the edge), lasting at least 25 seconds, and showing elongation (R g 2 > 40 μm 2 ). We normalized the rate by the area occupied by the pattern.
Statistical analysis. Data are presented as mean ± SD. Statistical comparisons were made by the Student's test or the Mann-Whitney U test. For correlations, we calculated Pearson correlation coefficients (r). p < 0.05 was considered to be statistically significant for all tests. For histograms, error bars are calculated based on the propagation of uncertainty using the square root of the number of events, assuming that each bin regions represent a Poisson distribution of number of events 72 .