Endothelial junctional membrane protrusions serve as hotspots for neutrophil transmigration

Upon inflammation, leukocytes rapidly transmigrate across the endothelium to enter the inflamed tissue. Evidence accumulates that leukocytes use preferred exit sites, alhough it is not yet clear how these hotspots in the endothelium are defined and how they are recognized by the leukocyte. Using lattice light sheet microscopy, we discovered that leukocytes prefer endothelial membrane protrusions at cell junctions for transmigration. Phenotypically, these junctional membrane protrusions are present in an asymmetric manner, meaning that one endothelial cell shows the protrusion and the adjacent one does not. Consequently, leukocytes cross the junction by migrating underneath the protruding endothelial cell. These protrusions depend on Rac1 activity and by using a photo-activatable Rac1 probe, we could artificially generate local exit-sites for leukocytes. Overall, we have discovered a new mechanism that uses local induced junctional membrane protrusions to facilitate/steer the leukocyte escape/exit from inflamed vessel walls.


INTRODUCTION 1
The current paradigm of leukocyte transendothelial migration (TEM) comprises 2 leukocyte rolling, arrest, crawling, firm adhesion and diapedesis (Alon & van Buul, 2017;3 Butcher, 1991; Muller, 2016;Nourshargh & Alon, 2014;Springer, 1994;Vestweber, 2015) . 4 The latter step occurs either through the endothelial junctions, known as the paracellular route 5 or through the endothelial cell body, called the transcellular route (Carman, 2009;Wittchen, 6 diapedesis by remodeling its actin cytoskeleton at the subcellular level. Thus, understanding 1 how actin-regulatory protein complexes orchestrate TEM under inflammation may thus be key 2 to discover the mechanisms of the endothelium that drive local leukocyte exit. 3 Our observations underscore that leukocyte diapedesis is not a random event but 4 occurs at predefined exit-sites. As TEM is a very rapid process (leukocytes cross the 5 endothelium within 120 seconds), we used lattice light sheet microcopy, allowing high-speed 6 imaging in 3 dimensions in time with high resolution but low toxicity to monitor TEM in detail. 7 We discovered a novel mechanism how the endothelium generates local exit-sites for 8 leukocytes to leave the vasculature. By asymmetrically inducing Rac1-and Arp3-dependent 9 apical membrane protrusions at cell-cell junction regions that express ACKR1, ICAM-1 and 10 PECAM-1, the endothelium guides leukocytes to the exit-side. Additionally, these protrusions 11 are also found in vivo, and local activation of Rac1, using a photo-activatable Rac1 probe drives 12 leukocyte exit on demand. Our work identified so-called endothelial junctional membrane 13 protrusions (JMPs), a novel molecular mechanism that allows local steering of leukocyte TEM 14 at the vascular level and offers new therapeutic targets to locally inhibit or enhance leukocyte 15 extravasation. 16 17 18 membranes. Figure 1G shows an example of such a JMP dynamics in time map and showed 1 that JMPs colocalized with VE-cadherin ( Figure 1H).      To study the JMP phenotype in more detail, we used LLSM imaging and found that VE-25 cadherin localized at the basolateral area of JMPs but is not present on the actual JMP ( Figure  26 3A). In contrast, junctional PECAM-1/CD31 was present on JMPs ( Figure 3B and S3A). Using 27 a 3D quantification algorithm, we found that PECAM-1 covered at least 50% of the surface of 28 the JMP ( Figure 3C). To examine whether JMPs are found in vivo and are correlated to TEM 29 events, we used the cremaster muscle as a well-established in vivo model. The cremaster 30 muscle of C57BL/6 mice were treated with TNF/IL1β and subsequently stained for PECAM-1 31 and neutrophil-specific coronin1 (Pick et al., 2017). We found that PECAM-1-rich membrane 32 protrusions can also be found in vivo. As with the in vitro data, PECAM-1 covered the apical 33 protrusions that surrounded adherent neutrophils in vivo ( Figure 3D). 3D Reconstruction of 34 these images showed the presence of PECAM-1 on the endothelial apical protrusions (     revealed that JMPs are rich in F-actin ( Figure 4A). To investigate if these structures are also inflamed cremaster muscle vessel morphology. As Lifeact-EGFP is mainly expressed in the 1 vascular endothelium and not in the leukocytes, allowing imaging of the endothelial actin 2 cytoskeleton with excellent contrast (Fraccaroli et al., 2012), we used these mice to study the 3 presence of luminal membrane protrusions. In vivo time lapse imaging showed actin-rich 4 vascular structures that protruded apically and were associated with TEM events ( Figure 4B). 5 As the Arp2/3 complex is involved in actin nucleation, branching and lamellipodia 6 formation (Goley & Welch, 2006), we hypothesized that its activity is required for the dynamics 7 of JMPs. Indeed, treatment of endothelial cells with the Arp2/3 inhibitor CK-666 reduced JMP 8 dynamics ( Figure 4C-D). Next, we studied the functional consequences of perturbing F-actin 9 branching on neutrophil TEM. However, the use of inhibitors can be problematic in assays that 10 include two different cell types: inhibitors may diffuse out of the endothelial cells and affect 11 migration motility of neutrophils under flow conditions. Therefore, we silenced Arp3 in 12 endothelial cells using shRNA ( Figure S4A-B). Indeed, silencing Arp3, with two independent 13 shRNAs, resulted in reduced number of neutrophils that crossed the endothelial monolayer 14 under flow conditions ( Figure 4E and S4C). To specifically quantify membrane dynamics in 15 Arp3-deficient endothelial cells, we co-expressed the Arp3 shRNA and mNeonGreen CAAX 16 from one plasmid, assuring that CAAX-expressing cells were indeed silenced for Arp3 ( Figure  17 S4D-E). Endothelial cells that were silenced for Arp3 showed reduced JMP dynamics ( Figure  18 4F and S4F). These data underscore the importance of actin nucleation by the Arp2/3 complex 19 for JMP formation and neutrophil TEM.        Interestingly, LLSM showed the presence of asymmetrical JMPs, meaning that one cell 30 generated the protrusion, while the neighboring cell did not ( Figure 6A and S6B). Using 31 scanning electron microscopy, we confirmed that JMPs displayed an asymmetric phenotype 32 at endothelial cell-cell junction regions ( Figure 6B and S6C). To quantify this, we analyzed 33 membrane dynamics at JMP regions using mosaic-expressing endothelial cells and found that 34 one endothelial cell showed higher JMP dynamics compared to the adjacent one, as indicated 35 by a ratio >1 ( Figure 6C). Thus, our data reveal that local JMPs have an asymmetric phenotype 36 at junction regions.
To further assess if local Rac1 activation can trigger the direction of neutrophil 1 transmigration across an asymmetric junction, we overexpressed the Rac-specific RhoGEF 2 Tiam1 in endothelial cells. Biochemical analysis showed that the active Tiam1 mutant, C1199, 3 activated endogenous Rac1 in endothelial cells ( Figure S6D). Additional immunofluorescence 4 imaging revealed that Tiam1 localized at junctions as determined by a perpendicular line scan 5 across the cell-cell junction indicated by VE-cadherin ( Figure S6E and S6F). Similar line scan 6 analysis showed that Tiam-C1199 recruited F-actin to these sites ( Figure   that display an asymmetric protrusion on one of the two neighboring endothelial cells. As a 20 result, neutrophils migrate from the top of an "inactive" cell to underneath the "active" protruding 21 endothelial cell. 22 Based on these results, we hypothesized that local activation of Rac1, i.e., one 23 endothelial cell but not the neighboring one, triggers asymmetric JMPs that function as a local 24 recognition site to drive diapedesis for neutrophils. To test this hypothesis, we used a      In summary, our data indicate that neutrophils preferred to exit the endothelium through 32 junctions that display JMPs in an asymmetric manner, meaning that one endothelial cell

DISCUSSION 1
Upon acute inflammation, neutrophils start to adhere to the inflamed inner layer of the 2 vessel wall, the endothelium, followed by crawling behavior. When crawling, neutrophils 3 appear to search for an optimal spot to cross the endothelium. Although there is consensus 4 that so-called TEM hotspots exist, and are in particularly recognized in vivo (Proebstl et al., defined and regulated. We found that neutrophils prefer specific, apical membrane structures 7 generated by endothelial cells, serving as exit-sites on inflamed endothelial monolayers. 8 As leukocytes cross the endothelial barrier within minutes, one requires high resolution 9 and high-speed imaging technology to be able to capture all details in 3 spatial dimensions in 10 time. To achieve this, we used lattice light sheet microscopy and discovered the existence of 11 endothelial membrane structures that protrude apically at junction regions and serve as local 12 recognition sites for crawling neutrophils. This discovery explains why neutrophils prefer one 13 junction over the other, namely the junction that displays an apical lamella that functionally 14 guides the crawling neutrophil through the junctional cleft underneath the endothelial layer. 15 Hence, such recognition sites may be considered as TEM hotspots. As the membrane with the help of ACKR1, to guide the neutrophils through. Our data add to this study by showing that JMPs express ACKR1, as well as ICAM-1 and PECAM-1. We propose that these guiding 1 molecules need JMPs for optimal exposure to the crawling leukocytes to initiate diapedesis. 2

Many different types of membrane structures have been described to emerge in 3 endothelial cells. Breslin and colleagues described local lamellipodia to be involved in 4
controlling the endothelial barrier function (Breslin et al., 2015). These structures depend on 5 Rac1 signaling and myosin-mediated local tension and show great similarities to JMPs. JMPs 6 are also regulated by the actin cytoskeleton to the same extent as regular protrusions are 7 induced. We show a prominent role for actin polymerization, Arp2/3-mediated branching and 8 involvement of the small Rho GTPase Rac1. When measuring Rac1 activity locally using a 9 FRET-based biosensor, we find Rac1 to be activated at junction regions of inflamed endothelial 10 cells, and to co-localize with JMPs. When locally activating Rac1, using a photo-activatable 11 Rac1 probe, we can drive leukocyte exit on demand. This is, to the best of our knowledge, the 12 first time that leukocyte extravasation can be triggered on command. For therapeutic potential, 13 not only can this mechanism now function as a target and pharmacological blockers can be 14 developed, but more importantly, it may lead to novel strategies to initiate leukocyte traffic to 15 parts of the body where we wish to have more immune cells present, e.g. for immune cell 16

therapies. 17
For the regulation aspect: it is important to note that JMPs do not seem to be induced 18 by leukocytes themselves, but rather through an intrinsic mechanism within the endothelium 19 in the presence or absence of crawling leukocytes, we did not detect any difference in JMP 23 activity. Also, the localization of the JMPs did not change when neutrophils were added, and 24 Rac1 activity did not change when leukocytes crawled on top of an endothelial cells. In addition, 25 we noticed that neutrophils did not per se choose the site of highest Rac1 activation in an 26 endothelial cell, presumably because the Rac1 activation itself cannot be sensed from a 27 distance. We hypothesize that a neutrophil continues to crawl over the endothelium until it 28 encounters a local JMP, which is associated with high Rac1 activity, which is then used for 29 diapedesis. We postulate that the regulation of JMPs happens in a stochastic manner, in line  As endothelial cells are constantly exposed to flow conditions, we expected JMPs to be 34 influenced by flow as well. To our surprise, we did not measure any differences in JMP  In summary, we have identified an endothelial membrane structure that supports local 10 exiting of crawling leukocytes. Our data show the presence of such structures in vivo as well. 11 If these structures are also involved and perturbed in diseased conditions is not known but is 12 an attractive hypothesis and would give new opportunities to target such pathologies. 13  Time-lapse images were recorded for 15-30 minutes on a widefield microscope with a 40x oil 17 objective at 37˚C and 5% CO2. Analysis was performed manually using Fiji. 18

Quantification modes of TEM 20
DIC and fluorescence images were recorded every 5 sec. To discriminate from the rolling 21 phase, a neutrophil was crawling when migration speed was below 10 um/5 sec. Implying that 22 its location in two subsequent frames is overlapping. We quantified the number of neutrophils 23 that undergo diapedesis at the first junction they encounter during the crawling phase. We also 24 distinguished between neutrophils that transmigrate at the first region they arrive at a junction 25 and neutrophils that crawled along the junction to their diapedesis site which was defined as 26 being more than 10 µm from the region they initially encountered that junction. Neutrophils that 27 turn around during diapedesis were scored as 'turn' when they were located on top of an 28 endothelial cell adjacent to the endothelial cell they arrived from the frame before diapedesis. 29 To determine whether neutrophils transmigrate at JMPs, we set a threshold at the JMP 30 dynamics map to 2x of the mean. The data to calculate this mean are shown in figure S7F. 31 When the diapedesis site had a value above this threshold, it was considered as a diapedesis 32 event at a JMP. These JMP maps were generated from the frames that were collected before 33 the neutrophil arrived at the diapedesis event, to exclude membrane dynamics because of the 34 interaction between the neutrophil and the endothelium. To measure the JMP value at the 35 diapedesis site, a region of interest around the diapedesis site was taken (region 1). To 36 measure the JMP value at junction region a neutrophil first encounters, a region of interest of 10 um was used (region 2). To get the relative MD increase, the difference between these two 1 values was divided by the value of region 2. To measure the relative JMP increase before and 2 after TEM, a region of interest around the diapedesis site was selected. Then the frames before 3 the neutrophils arrived at this site were used for calculating the JMP value (value 1) before 4 TEM. To exclude influences of subendothelial crawling of neutrophils, the 30 frames after a 5 neutrophil left from the diapedesis site were used for calculating the JMP value after TEM 6 (value 2). The difference between these two values was divided by value 1 to calculate the 7 relative JMP increase upon diapedesis.

Widefield imaging 20
Widefield images were recorded using a Zeiss Observer Z1 microscope using a 40x NA 1. seconds (only CAAX recording) or 4-7 seconds (also DIC and/or FRET recordings). 25

JMP maps 27
HUVECs expressing mNeonGreen-CAAX were imaged on a widefield microscope (Zeiss 28 Observer) with a 40x oil objective for at least 5 min. Fluorescence images were acquired every 29 2-7 sec, depending on the experiment and the need for DIC recordings. We developed an 30 image-based method for quantifying the dynamics in time series. It is based on the notion that 31 highly dynamic regions will show large fluctuations in intensity. This results in a high standard 32 deviation in the intensity over time. To extract this information, we generated a macro to 33 calculate, for each pixel, the standard deviation in the intensity over time. The standard 34 deviation is normalized by the average intensity and the image essentially depicts the 35 coefficient of variation. Since the intensity fluctuations may arise from sources other than 36 membrane dynamics, we implemented corrections for (i) bleaching and (ii) image drift, resulting in a membrane dynamic map that shows the membrane dynamics of a typical time lapse image 1 of endothelial cells. To focus only on the junctions, the cytoplasm of each cell was selected as 2 being 3 µm from the junctions and set to NaN. To measure junctional membrane dynamics in 3 a specific condition, the mean value of that image was taken. For each experiment, the data 4 was normalized to the condition with HUVEC, cultured on fibronectin-coated glass and treated 5 for 20h with TNFα. 6 7

Inhibitors 8
HUVECs expression mNeonGreen-CAAX to measure membrane dynamics were cultured on 9 coverslips. While imaging 50 µM of EHT1864 or 100 µM CK666 was added. Membrane 10 dynamics were determined 6 min after adding EHT1864 or 20 sec after adding CK666. Ratio 11 after versus before inhibitor treatment was calculated from the membrane dynamics values. Supplemental Methods and analyzed using Imaris software. 33 automatically using an Otsu global threshold, and a triangular mesh is created from the 1 volumetric image data using marching cubes isosurfacing. Local surface variation is measured 2 in the neighborhood of each vertex. JMPs are identified by a global threshold of surface 3 variation. Vertices with high surface variation are considered a JMP region. The threshold was 4 empirically determined and is the same across all datasets. This approach is effective in flat 5 regions of the cells but is sensitive to the thinner membrane around the nucleus of the cell. 6 Therefore, the nuclear regions were manually marked and ignored for JMP analysis. After 7 identification, JMP regions are separated from the cell mesh and capped to create a closed 8 volume. Area and total region volume are measured in every frame and collected for each 9 dataset. A coverage estimate of the JMP by PECAM or VE-Cadherin stains is also computed 10 by considering the distribution of staining across the JMP regions (details in the supplemental 11 methods). This pipeline was written in MATLAB R2019b (source code can be found on github 12 at: https://github.com/aicjanelia/visitor-van-buul). 13

Scanning Electron Microscopy 15
Samples were fixed in 4% paraformaldehyde and 1% glutaraldehyde for 1 hour at room 16 temperature and dehydrated using an ethanol series. To reduce sample surface tension, 17 samples were immersed in hexamethyldisilizane (Sigma-Aldrich) for 30 minutes and air dried. 18 Before imaging, samples were mounted on aluminum SEM stubs and sputter-coated with a 4 19 nm platinum-palladium layer using a Leica EM ACE600 sputter coater (Leica Microsystems, 20 Illinois, USA). Images were acquired at 2 kV using a Zeiss Sigma 300 SEM (Zeiss, Germany). 21

FRET imaging 23
Rac1 activation was measured using a DORA FRET-based Rac1 biosensor as described 24 before (Timmerman et al., 2015). Briefly, we used a Zeiss Observer Z1 microscope equipped 25 with a 40x NA 1.3 oil immersion objective, a HXP 120 V excitation light source, a Chroma 510 26 DCSP dichroic splitter, and two Hamamatsu ORCA-R2 digital CCD cameras to simultaneously 27 image Cerulean3 and mVenus emission. Data was analyzed using ImageJ software. ROI with 28 no cells presents throughout the movie was selected for background correction of Cerulean3 29 and mVenus image stacks. Alignment of Cerulean3 and mVenus was done using the 30 'MultiStackReg' plugin (http://rsb.info.nih.gov/ij/plugins/index.html). To reduce noise a smooth 31 filter was applied to both image stacks. To correct for bleed through, 0.62xCerulean was 32 subtracted from the mVenus signal for each frame to get the corrected mVenus stack, which 33 was divided by Cerulean to calculate the FRET ratio. To define regions of high Rac1 activation, 34 we selected the 1 um junction of the cell and put a threshold to approximately cover 50% of 35 the junction. Then diapedesis sites were scored for either high or low Rac1 activity. To measure ruffles. These ROIs were projected on the FRET ratio-image and the average FRET value of 1 such a JMP was calculated. Because the FRET signal is always increased at the junctions 2 compared to the cell body, we compared the JMP FRET value to the junction of the cell. For 3 this we selected a border of 3 um per cell and calculated the mean FRET value. 4 5 Photoactovatable-Rac1 probe 6 Flow experiment was performed as described above. For the control experiments cells were 7 kept in the dark until neutrophils transmigrated. To switch on photoactivatable Rac-1, channels 8 were illuminated with blue light from a HXP Light source for 10 seconds prior to neutrophil 9 injection into the perfusion system.

Identification of Cell Membrane Surface 3
The lattice light sheet used at the Advanced Imaging Center, collects images at a 31.8° angle 4 relative to the coverslip. The images are then "deskewed" to align each z-slice with one 5 another. In doing so, each z-slice is padded with zeros which can easily throw off any 6 automated segmentation algorithm. To overcome this challenge, a mask is created to 7 determine a valid image region for both the deskewed and deconvolved data. The mask is 8 generated by keeping all non-zero pixels that are identified across the first 50 frames of each 9 dataset. Each z-slice of the mask has hole-filling applied to make sure there are no internal 10 holes. Within the masked region, a global threshold is estimated for each cell membrane (Otsu, 11 1979). Threshold regions smaller than 0.1% of the masked region are discarded. Regions with 12 convex volume larger than 50% of the mask volume are also discarded. These constraints  To identify individual JMPs, each connected region is isolated into sub meshes. Each submesh 32 is closed using an approach similar to Liepa (Liepa, 2003). Once each submesh is capped, the 33 volumes and base areas are measured.

Coverage of JMP by PECAM/VE-Cadherin 1
A PECAM or VE-Cadherin stain value is computed for each vertex on the cell surface mesh. 2 The stain value at each vertex is calculated as the mean of the 8 voxels surrounding the vertex 3 in the PECAM or VE-Cadherin channel. Vertices are considered "stained" as long as their stain 4 value is above an Otsu threshold. 5 To estimate the distribution of stained vertices within a JMP, each vertex in the JMP is also 6 encoded with its minimum surface distance to the edge of the region. A normalized histogram 7 of stained vertices binned by distance from the edge is used to approximate the distribution of

Conflicts of Interest 19
The authors declare no conflict of interest.