Elongated neutrophil-derived structures are blood-borne microparticles formed by rolling neutrophils during sepsis

Neutrophils protect the host from invading bacteria by arriving at the site of infection and engaging in antibacterial secretory and phagocytic activities (Christoffersson and Phillipson, 2018; Gómez-Moreno et al., 2018; Soehnlein et al., 2017). Neutrophils mature in the bone marrow and travel in the blood circulation, surveilling for infectious or inflammatory signals (Ley et al., 2018; Nourshargh and Alon, 2014). Upon sensing such signals, the neutrophils roll and arrest on the endothelium and eventually migrate across the vessel wall into the inflamed tissue. Neutrophil rolling is enabled by intermittent bond formation between endothelial selectins and their ligands on the neutrophil (Ley et al., 2018; Nourshargh and Alon, 2014), with a contribution from β2 integrins on the neutrophils and their ligands on the endothelial cells (Ley et al., 2018). Rolling is stabilized by formation of tethers, which are long, ∼200-nm-thin structures extending from the adhered microvilli of rolling neutrophils (Marki et al., 2016, 2018; Sundd et al., 2012).

During this recruitment process, neutrophils undergo gradual activation and may already begin to discharge antibacterial compounds in membrane packages or freely, as suggested by the presence of primed neutrophils (Fine et al., 2019), neutrophil-derived extracellular vesicles (EVs; Timár et al., 2013), and neutrophil extracellular traps (NETs; Colón et al., 2019) in blood. Release of different EVs (Colombo et al., 2014; Maas et al., 2017; van Niel et al., 2018) includes 30–100-nm large exosomes, formed by multivesicular body fusion with the plasma membrane. The 200–500-nm large ectosomes (or microvesicles) bud off directly from the plasma membrane into the extracellular space, whereas large apoptotic bodies (up to several micrometers in diameter) bud off from apoptotic cells. Neutrophil-derived EVs are recognized by their submicron size and their neutrophil-specific surface marker, CD66b. The general EV surface markers phosphatidylserine (PS; Timár et al., 2013) and CD63 tetraspanin (TS; Genschmer et al., 2019) were also detected on neutrophil-derived EVs (Hong, 2018). Using imaging cytometry, Headland et al. (2014) found that healthy human blood plasma contains ∼2,500 EVs/µl, of which 12% are leukocyte derived (CD45⁺) and 3.6% are neutrophil derived (CD66b⁺). In culture, both suspended and adhered human neutrophils release EVs (Dalli et al., 2013; Headland et al., 2014; Timár et al., 2013; Lőrincz et al., 2015). Stimulation with inflammatory mediators increases the EV release rate and alters the EV composition (Headland et al., 2014; Timár et al., 2013; Johnson et al., 2014). Extravasated neutrophils also release compounds in migrasomes. These plasma membrane–covered microparticles detach from the neutrophil’s uropod as the neutrophil migrates across the vessel wall or in the tissue (Hyun et al., 2012; Lim et al., 2015). The main membrane-free release mechanism of neutrophils is degranulation, a special form of exocytosis, where the contents of neutrophil granules (intracellular vesicles) are released via granule fusion with the plasma membrane (Ley et al., 2018). Another membrane-free inflammatory mediator discharge mechanism is the release of NETs, nuclear or mitochondrial DNA structures decorated with cytoplasmic and granule proteins (Boeltz et al., 2019). Extensive NET release was found to have a disadvantageous proinflammatory and prothrombotic effect in various diseases (Jiménez-Alcázar et al., 2017; Sørensen and Borregaard, 2016). The importance of neutrophils and neutrophil-derived particles is illustrated by genetic defects. For example, inherited malfunction of neutrophil adhesion molecules (leukocyte adhesion deficiency syndromes) results in higher susceptibility to infection, because neutrophils poorly arrest and fail to reach pathogens (Ley et al., 2018). Patients with inherited degranulation malfunctions suffer from recurrent infections due to reduced antibacterial compound release (Ley et al., 2018).

Here, we report the discovery of a novel neutrophil-derived microparticle type, elongated neutrophil-derived structures (ENDS). ENDS derive from rolling neutrophils and contain the damage-associated molecular pattern S100A8–S100A9 complex. We describe ENDS formation, morphology, and molecular composition. We demonstrate that degrading ENDS release S100A8–S100A9 complex and that the number of ENDS is greatly increased in blood plasma of LPS-injected mice and septic patients.

We imaged in vivo neutrophil rolling in the mouse cremaster vasculature. In this model, the surgical preparation provides sufficient trauma to initiate neutrophil rolling and adhesion in venules (Kunkel et al., 1997). To increase neutrophil rolling and arrest, we induced the expression of endothelial E-selectin via an intrascrotal injection of TNF-α 2 h before imaging (Kunkel et al., 1997). To clearly demarcate the neutrophil plasma membrane, we injected Ly6G-AF647 antibody (Marki et al., 2018). This method reveals neutrophil tethers, which are not visible in vivo without fluorescent membrane labeling.

Slower-rolling neutrophils formed tethers that detached at the tether anchor point (where the tether is attached to the endothelial substrate; Sundd et al., 2012). However, some of the tether-forming neutrophils rolled at a faster rate, and their tethers broke along their length, sometimes multiple times, to form ENDS (Fig. 1, A and B; Video 1; and Video 2). In a venule with an estimated wall shear stress (WSS) of 9.6 dyn/cm², 4 of 69 rolling neutrophils (∼6%) formed 9 ENDS during 144 s (∼4 ENDS formed per minute; Video 2). In 27 venules of eight mice, 233 ENDS were visible, 81% of which remained immobile on the endothelium and 19% seemed to be sliding along the vessel wall (Fig. S1 A and Video 3). The median length of ENDS was 7 µm (2–112 µm); examples are shown in Fig. 1 C. 24% of ENDS showed a thicker, bead-like structure, typically at one end (Fig. 1 C and Fig. S1 B). ENDS were detectable in the circulation by intravital microscopy (Fig. S1 C) and in blood plasma isolated from mice after cremaster intravital microscopy (Fig. S1 D). These findings suggest that ENDS produced by rolling neutrophils in vivo can detach and appear in plasma.

To test whether human neutrophils can form ENDS, we imaged rolling of CD16-AF647 antibody–labeled human neutrophils on P-selectin or E-selectin substrates in flow chambers. We set the coating concentration to match the neutrophil rolling speed observed in mice in vivo, and we controlled the WSS with a syringe pump. Human neutrophils rolling on E-selectin at 40 dyn/cm² formed copious numbers of ENDS (Fig. 1 D). To investigate the dependence of ENDS formation on WSS, we developed an image-processing algorithm to segment ENDS and distinguish them from tethers, the structures still attached to neutrophils (Fig. 1 E and Video 4). We ramped up WSS in 10 dyn/cm² increments every 2 min. On both P- and E-selectin, ENDS started forming above 30 dyn/cm² and the formation accelerated with increasing WSS (Fig. 1 F, bottom). Equivalent experiments conducted on murine neutrophils showed similar results (Fig. 1 F). To test whether ENDS formation happens in the absence of antibody labeling of neutrophils, we rolled unlabeled human neutrophils under similar conditions, and after the neutrophils detached from the substrate, we perfused the flow chamber with CD16-AF647 antibody, revealing many ENDS (Fig. S1 E). Thus, ENDS formation is not antibody induced.

To study the mechanism of ENDS release into plasma, we coated the first half, but not the second half, of a flow chamber with P-selectin. After depositing ENDS on the P-selectin–coated surface, we increased shear stress. ENDS showed sliding behavior, similar to that observed in mice in vivo. When they reached the selectin-free part of the flow chamber, ENDS immediately detached (Fig. 1 G). In vivo, selectin expression is highest in venules <50 µm (Runnels et al., 2006; Norman et al., 2003). Thus, ENDS in vivo will eventually reach a selectin-poor endothelium, probably in a larger venule or small vein, and detach. During intravital microscopy, we observed several examples of sliding ENDS detaching from the vessel wall (Video 3). Both in vivo and in flow chambers, ENDS can become very long (maximal length of 112 and 423 µm, respectively; Fig. S1 F).

Our rolling experiments show that ENDS are preferentially released by fast-rolling neutrophils that show unstable rolling with jumps compared with non-ENDS-forming neutrophils in the same vessel. This suggests that the ENDS formation mechanism involves tether failure, because tethers, the precursors of ENDS, are load-bearing structures, and upon tether breakage, the rolling neutrophil shows a short acceleration (Marki et al., 2016, 2018; Ramachandran et al., 2004; Sundd et al., 2012). The number of tethers per neutrophil increases with increasing shear stress, suggesting that tethers stabilize neutrophil rolling (Ramachandran et al., 2004; Sundd et al., 2012; Marki et al., 2016, 2018). Various studies support the view that tether growth requires flow of plasma membrane into the tethers (Simunovic et al., 2017; Pospieszalska and Ley, 2009). This suggests that in stably rolling neutrophils, the tether growth rate keeps up with the rolling velocity (Marki et al., 2016, 2018). However, in unstably rolling neutrophils, the membrane supply cannot keep up with the tether-pulling rate. If the tether is strongly anchored to the substrate, the tether breaks between the cell and the anchor point. Another possibility could be that tether separation from the rolling neutrophil is mediated by an active molecular mechanism that could sever the tether.

To obtain high-resolution images of human ENDS, we isolated CD16-AF647–labeled ENDS from flow chambers, fixed them on a cover glass, and performed correlative confocal and scanning EM (SEM; Fig. 2 A and Fig. S1 A). By SEM, ENDS showed an apparent diameter of 77 ± 14 nm (average ± SD, n = 5). By comparing neutrophil diameter measured by SEM and by live confocal microscopy, we estimated the fixation-induced sample shrinkage to be 30%, which is in agreement with another report (Sundd et al., 2012). Thus, we estimate the true ENDS diameter to be ∼110 nm. The bead-like structures observed by confocal microscopy were resolved as branched, curled structures of unknown molecular composition by SEM (Fig. 2 A). To measure ENDS diameter with an alternative method, we measured human CD16-AF647 surface-labeled ENDS with stochastic optical reconstruction microscopy (STORM) superresolution microscopy. We measured five ENDS from two independent experiments (Fig. 2 B) by CD16 spot distribution across the cross section of ENDS (Fig. 2 C). Based on full width at half maximum, we found the average diameter of ENDS to be 115 nm, in excellent agreement with our SEM measurements.

Next, we investigated whether ENDS contain organelles. To this end, we produced human ENDS from neutrophils that were labeled for their surface with CD16-AF647 antibody and for different intracellular organelles. Labeling with CellTracker Orange CMRA, Mito Tracker Green, ER Tracker Green, or Hoechst showed that cytoplasm, mitochondria, ER, or DNA were readily labeled in neutrophils (positive control). ENDS only stained for cytoplasm (Fig. 2 D). We conclude that ENDS contain cytoplasm, but not mitochondria, ER, or DNA.

To investigate the molecular composition of ENDS, we used liquid chromatography–tandem mass spectrometry (LC-MS/MS) for an unbiased overview of the ENDS proteome. We generated ENDS in a flow chamber via rolling isolated human neutrophils. To minimize neutrophil contamination of the ENDS samples, we used low calcium (Ca²⁺/Mg²⁺-free PBS). Under these conditions, neutrophils cannot arrest but still form tethers by binding to P-selectin (coated at a higher density). The proteomes of three ENDS samples from different healthy donors were compared with the same amount of protein isolated from neutrophils. The total number of peptide spectrum matches (PSMs) detected in these samples showed modest variation (PSM count = 3,357 ± 336 [average ± SD]), showing that similar amounts of protein from each sample were analyzed. In ENDS, we identified 79 proteins (based on two or more PSMs and present in at least two of the three ENDS samples; Table 1 and Table S1). The 10 most abundant proteins in ENDS, which we also detected in similar amounts in whole neutrophil lysates, were lactotransferrin, albumin, actin (cytoplasmic 1), myeloperoxidase, S100A8, S100A9, α-enolase, GAPDH, annexin-A3, and catalase.

Lactotransferrin, myeloperoxidase, and albumin can be found in neutrophil granules (Rørvig et al., 2013), suggesting that ENDS may contain granules; however, by comparing the ENDS proteome to published proteomic datasets, we found that ENDS are more similar to neutrophil EVs (Timár et al., 2013) and NETs (Petretto et al., 2019) than to neutrophil granules (Rørvig et al., 2013) or neutrophil plasma membrane (Rørvig et al., 2013; Fig. S2 B). This appears to be in accordance with the ENDS formation mechanism, as ENDS are pinched off from the cell surface. Similar to ectosomes, they contain plasma membrane and cytosolic components. However, while ectosome formation requires assembly of molecular complexes at the plasma membrane, tethers, the precursors of ENDS, are pulled out from the plasma membrane. Adhesion molecules that anchor tethers are concentrated on the tips of microvilli (Bruehl et al., 1997). Most of the proteins common between ENDS and EVs belonged to the regulated exocytosis pathway. However, only ENDS contained certain proteins of the myeloid leukocyte activation and glycolytic metabolism pathways (Table S1). Thus, the molecular composition of ENDS is similar, but not identical, to EVs.

In neutrophils, we identified a similar number of proteins (76) as in ENDS (two or more PSMs; Table S1). Of these, 45 were present in both ENDS and neutrophils, and 31 were detected only in neutrophils. To obtain an overview of the pathways covered by the ENDS proteome, we conducted gene ontology analysis using Metascape and found that the leukocyte degranulation pathway was highly represented in ENDS (Fig. S2 C). On the other hand, in the neutrophil proteome, we found regulated exocytosis, humoral immune response, and phagocytosis to be highly enriched pathways (Fig. S2 C).

In flow chamber experiments, we observed that ENDS were long, elongated structures immediately after formation, but contracted and rounded up over the next 4 h in Ca²⁺/Mg²⁺ HBSS buffer at room temperature (Fig. 3 A). In some instances, we observed fragmentation of ENDS as well. Contracted ENDS had a twofold smaller surface area and significantly increased sphericity (Fig. 3 B). The initial elongated shape of ENDS is a consequence of their formation mechanism described above. Similar elongated microparticles of erythrocyte or platelet origin were detected with cryo-EM in EV isolates of healthy human blood (Yuana et al., 2013; Arraud et al., 2014). It is not clear how these structures maintain their elongated shape. Our observation suggest that ENDS can stay elongated for a limited time but eventually round up within a few hours after formation.

To test whether degrading ENDS lose their membrane integrity, we measured intracellular free calcium in ENDS labeled with the Ca²⁺ reporter dye Fluo4-AM. In fresh ENDS, intracellular free calcium concentration was low and greatly increased after exposure to ionomycin (Fig. 3 C), indicating that fresh ENDS were able to keep calcium out. ENDS stained with Fluo4-AM after 7-h incubation did not respond significantly to ionomycin (Fig. 3 C), suggesting that ENDS lose their membrane integrity over time. To test whether they release the abundant cytosolic protein S100A8–S100A9 complex, we prepared ENDS in a flow chamber and imaged the entire chamber and counted the ENDS. Following 1-h incubation during confocal scanning and 7- and 24-h incubation in a tissue culture incubator, ENDS supernatants contained increasing amounts of S100A8–S100A9 complex (Fig. 3 D). Our ENDS preparations contained a few contaminating neutrophils that did not detach during washing. To account for this, we corrected the supernatant S100A8–S100A9 complex for the count of contaminating neutrophils. The normalized data show that ENDS released at least four times more S100A8–S100A9 than neutrophils.

Sepsis is known to be associated with strong neutrophil activation in blood (Sônego et al., 2016). Thus, we reasoned that the number of ENDS might be elevated during sepsis. To test this hypothesis in an experimental model, we treated wild-type mice by i.p. injection of high-dose LPS of Escherichia coli (O5:B55 strain, 40 mg/kg; Soromou et al., 2014; Napier et al., 2019). With imaging cytometry, we detected 730 ± 199 ENDS/µl of plasma (average ± SD) at 2 h after LPS (Fig. 4 A) but no ENDS in plasma of control mice. To avoid any possible artifacts due to labeling with antibodies, we performed this experiment on Ly6G-cre-mT/mG mice (Muzumdar et al., 2007; Hasenberg et al., 2015) that expressed plasma membrane–anchored EGFP in neutrophils only. We bled the same mice before and 2 h after LPS injection, isolated blood plasma, and observed the ENDS count by confocal microscopy. Before LPS injection, we detected no ENDS in the blood plasma; however, 2 h after LPS injection, we detected 1.1 ± 0.2 ENDS per field of view (average ± SD; Fig. S2 D). Example images of these ENDS are shown in Fig. 4 B.

For quantitative analysis of ENDS and EVs in septic human blood, we used imaging flow cytometry on blood plasma from 12 septic patients and 7 healthy controls (Table S2). We used CD16 and CD66b as neutrophil markers, which were also present on human ENDS formed in a flow chamber (Fig. S2 E). Septic blood samples contained ∼100 times more neutrophil-derived CD16⁺ and CD66b⁺ microparticles (120,000 ± 32,000/µl [septic] versus 1,100 ± 200/µl [healthy; average ± SEM]; Fig. 4 C). With the same scanning technique, Headland et al. (2014) reported 100 neutrophil-derived EVs/µl healthy human blood plasma. The difference between these results could be explained by different plasma preparation protocols (Chandler, 2016). Headland et al. (2014) centrifuged the blood once at 300 g for 5 min and then at 10,000 g for 2 min. We centrifuged the blood once at 1,000 g for 5 min. Likely, centrifugation at higher g force may have removed some EVs from Headland et al.’s samples.

The dominant neutrophil-derived particle populations in healthy blood samples were negative for TSs and PS (50 ± 8% [average ± SEM]). The second-largest fraction was positive for TSs but negative for PS (40 ± 7%; Fig. 4 D). This distribution shifted in the septic samples, where the TS- and PS-negative particles became more abundant (72 ± 5%) at the expense of the TS-positive particles (17 ± 3%). Among TS and PS double-negative particles, ENDS with an aspect ratio <0.6 (considered elongated) accounted for 3.1 ± 1.2% in septic and 2.5 ± 1.6% in healthy individuals (average ± SEM; Fig. 4 E). Based on our microfluidic flow chamber studies, we know that the spherical (aspect ratio >0.6) objects double negative for TS and PS are most likely fragmented ENDS, because in vitro ENDS round up and fragment over time; after 4-h incubation, 40% of the degraded ENDS remain PS negative, and only few ENDS stain for TSs (Fig. S2 F). However, we do not have direct in vivo evidence for fragmentation. Example images of ENDS and fragmented ENDS captured with ImageStream are shown in Fig. 4 F. With confocal microscopy, we could confirm the presence of CD16-AF647–stained ENDS in septic plasma (Fig. 4 G). The septic samples contained on average ∼100-fold more ENDS (elongated only, all double negative for TSs and PS) compared with control plasma (595 ± 159/µl [septic] versus 5 ± 4/µl [healthy]; Fig. 4 H). The median length of the 1,277 ENDS that we detected in plasma of the 12 septic samples was 6 µm (range, 4–28.5 µm).

The ENDS count in septic plasma positively correlated with blood neutrophil counts (Fig. 4 I). In the septic samples, plasma S100A8–S100A9 complex levels significantly and positively correlated with fragmented ENDS, but not with intact ENDS counts (Fig. 4 J). Taken together with our in vitro observation that ENDS degrade over time and release S100A8–S100A9 complex, this suggests that degrading ENDS are a likely source of free extracellular S100A8–S100A9 complex in septic plasma.

Proteins S100A8 and S100A9, also known as MRP-8 and MRP-14 or calgranulin A and B (Pruenster et al., 2016), were among the most abundant proteins in ENDS (Table 1). These proteins are abundant in neutrophil cytosol, where they form heterodimers that buffer Ca²⁺ and regulate microtubule formation (Pruenster et al., 2016). Neutrophils release S100A8–S100A9 complexes following activation or during NETs release or necrosis (Pruenster et al., 2016). Outside of the cell, S100A8–S100A9 can bind TLR4 and elicit several effects. S100A8–S100A9 can activate immune cells (Ehrchen et al., 2009), but it can also induce endotoxin tolerance upon prolonged exposure (Austermann et al., 2014). In the presence of calcium, S100A8–S100A9 complexes form tetramers (S100A8–S100A9)₂ and lose their TLR4 binding affinity (Vogl et al., 2018). The S100A8–S100A9 complex is increased in the blood of septic or trauma patients (Austermann et al., 2014; Simm et al., 2016).

We found that degrading ENDS release S100A8–S100A9 into their environment. Our experiments show that ENDS initially maintain a low luminal calcium level and respond to ionomycin, suggesting that ENDS can keep calcium out and, at least for some time, can keep the S100A8–S100A9 complexes in. As long as S100A8–S100A9 complexes are in a low-calcium environment, they remain dimers that retain the ability to bind TLR4. This is reminiscent of migrating neutrophils in extravascular tissue, which were found to release LTB4 packaged into EVs (Majumdar et al., 2016) that extend the half-life of LTB4 and enable coordination of neutrophil swarming. Similarly, in the lungs of chronic obstructive pulmonary disease patients, neutrophil elastase is released on the surface of neutrophil-derived exosomes, which prevents neutrophil elastase inhibition by α-anti-tripsin (Genschmer et al., 2019). In virally infected trachea tissue of mice, migrasomes formed by migrating neutrophils gradually released CXCL12 over time, which formed a chemokine trail behind the neutrophil that guides the migration of lymphocytes (Lim et al., 2015).

Here, we report the discovery of ENDS, which are a new type of neutrophil-derived microparticle. ENDS form when tethers detach from rolling neutrophils. ENDS are initially elongated. Over time, they fragment and round up. Unlike the known neutrophil-derived EVs, ENDS do not express TSs and do not bind annexin-5. ENDS also differ from migrasomes, because ENDS form in the vessel lumen from microvilli of nonpolarized rolling neutrophils, whereas migrasomes form in the extravascular space from the uropod of polarized migrating neutrophils. Human blood plasma ENDS counts are increased ∼100-fold in sepsis. The ENDS fragment count positively correlates with the concentration of S100A8–S100A9 complex in septic blood.

The experimental animal procedures of this study were performed in accordance with guidelines approved by the Institutional Animal Care and Use Committee of La Jolla Institute (LJI) for Allergy and Immunology, which is an American Association for Accreditation of Laboratory Animal Care accredited facility. Cremaster intravital microscopy was performed on male C57BL/6J mice ≥8 wk as described earlier (Marki et al., 2018). Approximately 3 h before imaging, mice were administered an intrascrotal injection with 500 ng TNF (PeproTech) in 300 µl PBS. The femoral artery was cannulated with a PE-10 tube, and the cremaster muscle was exteriorized for imaging with a resonant scanner confocal microscope (Leica SP8; Leica Microsystems). To label the neutrophil surface, 10 min before imaging, 5 µl (2.5 µg) anti-Ly6G-Alexa Fluor 647 monoclonal antibody (1A8; BioLegend) was injected in 100 µl PBS via the femoral artery. To measure WSS, 0.5-µm fluorescent spheres (Spherotech) were injected via a femoral artery catheter, and the vessel was imaged at a high frame rate to resolve the displacement of the fastest beads. WSS was calculated with the following formula: WSS = (0.625 × 8 × 2.2 × v_max/diameter) × η, where the maximum flow velocity (v_max) was calculated from the fastest bead frame-to-frame displacement, vessel diameter was measured on transillumination images, and mouse blood plasma viscosity (η) at 37°C was estimated by another study (Windberger et al., 2003) to be 0.013 s × dyn/cm². At the end of the imaging session, blood was collected from the femoral artery catheter into a heparinized syringe. Plasma isolated via centrifugation at 1,000 g for 5 min was loaded into an ibidi flow chamber (#80161; ibidi) and scanned for the presence of ENDS with a Leica SP8 confocal microscope equipped with a resonant scanner and hybrid detectors through a Leica 25×/0.95 NA water-immersion objective (Leica Microsystems).

Human neutrophils were isolated with Polymorphoprep (Alere Technologies) from healthy donors of the LJI normal blood donation program. Human neutrophils (10⁶) were labeled for 10 min in 100 µl Ca²⁺/Mg²⁺ containing HBSS (GIBCO BRL, Thermo Fisher Scientific) supplemented with 1 µl (0.2 µg) anti-CD16-Alexa Fluor 647 antibody (3G8; BioLegend). Mouse neutrophils were isolated from the bone marrow of C57BL/6J mice ≥8 wk into Ca²⁺/Mg²⁺ containing HBSS. The isolate was filtered through a 40-µm nylon strainer (Biopioneer), mouse neutrophils (10⁶) were labeled for 10 min in 100 µl Ca²⁺/Mg²⁺ HBSS with 1 µl (0.5 µg) anti-Ly6G-AF647 antibody. ibidi flow chambers (#80161) were coated for 15 min with 1 µg/ml recombinant human or mouse P- or E-selectin-Fc chimera (R&D Systems) and blocked for 30 min with casein blocking solution (Thermo Fisher Scientific). A 10-ml syringe without a piston connected to the inflow orifice of the flow chamber was used as an inflow reservoir, and a 60-ml syringe connected to the outflow orifice was used to generate flow via suction with a syringe pump (PhD2000; Harvard Apparatus). 100 µl labeled neutrophil suspension was loaded into the flow chamber at the inflow point, and a gentle (0.1 ml/min) flow was applied to bring the cells into contact with the substrate. Once neutrophils covered the entire substrate, a perfusion ramp programed into the pump was started, and the cells were perfused with Ca²⁺/Mg²⁺ HBSS as follows (each perfusion lasted for 2 min): 1.9 ml/min (10 dyn/cm²), 3.8 ml/min (20 dyn/cm²), 5.7 ml/min (30 dyn/cm²), 7.6 ml/min (40 dyn/cm²), and 9.5 ml/min (50 dyn/cm²). Neutrophils were imaged during the perfusion with Zeiss LSM880 confocal microscope in Airyscan through 20× air 0.8 NA objective (Carl Zeiss Microscopy).

To analyze the records, a segmentation method was developed with Imaris 8.3 (Bitplane) and Fiji (Rueden et al., 2017; Schindelin et al., 2012; http://fiji.sc/) software. For quantification of ENDS formation, the raw records were surface modeled and the objects that were in connection with neutrophils were excluded based on their width. The dataset was cleaned up manually frame-by-frame via omitting the signals of fly-by and partially detected cells. The remaining objects were assumed to be ENDS. Eventually, count and total surface area of ENDS were measured. To analyze neutrophil rolling speed and count, a cell body mask was created with Fiji. “MinError” threshold and 1-pixel median filter were applied on the images. The outermost pixels of the objects were eroded to remove the masks of small or thin objects and then dilated with a few pixels to completely cover the cell body. This cell body mask was loaded into Imaris and, with the spot tracking function, the rolling speed and cell count were measured. To segment tethers, the raw record was surface modeled, and objects not in contact with neutrophils were omitted based on their width. From the resulting “neutrophil with tether” mask, the cell body mask was subtracted.

ENDS produced by human neutrophils were harvested from a flow chamber with Trypsin-EDTA (0.05%, 5 min). ENDS suspension was spun at 1,000 g in a swinging bucket centrifuge onto a graded coverglass taped to a 35-mm dish with a hole in the middle (MatTek dish without the central cover glass). Samples were fixed for 15 min with 1.5% glutaraldehyde in PBS and dehydrated in increasing concentrations of ethanol. ENDS were imaged by confocal microscopy (20× air 0.8 NA objective, LSM880; Carl Zeiss Microscopy). The cover glass was removed from the dish, critical point dried and heavy metal coated (Zhang et al., 2012), and imaged with Zeiss Sigma500 scanning electron microscope (Carl Zeiss Microscopy). The ENDS detected with confocal microscopy were located with SEM based on the grids. Fiji was used to align and overlay the confocal and SEM images.

Human neutrophils were isolated from blood of healthy donors as described above. One million human neutrophils were incubated for 10 min in 100 µl Ca²⁺/Mg²⁺-free PBS supplemented with 2 µl (1 µg) anti-CD18 antibody (clone 1B4; BioLegend) to block β2-integrins and prevent neutrophil arrest and 1 µl (0.2 µg) CD16-AF647 antibody to label neutrophil surface. The cells were rolled in an ibidi flow chamber (#80607, coated with 0.1 µg/ml human recombinant E-selectin and blocked with casein) under 50–100 dyn/cm² shear stress generated with perfusion of HBSS containing Ca²⁺/Mg²⁺. This resulted in ENDS formation and detachment of the vast majority of neutrophils. The ENDS were fixed via incubation of the chamber lumen for 1 h with 1.5% glutaraldehyde (EMS) PBS solution.

Chambers were washed with PBS and filled with STORM blinking buffer that was made freshly before application via mixing 890 µl buffer B (50 mM Tris, pH 8.0, 10 mM NaCl, and 10% glucose), 100 µl MEA buffer (1 M MEA [#30070; Sigma-Aldrich] in 0.25 N HCl), and 10 µl GLOX buffer (56 U/ml glucose oxidase [#G2133; Sigma-Aldrich], 340 U/ml catalase [#C40; Sigma-Aldrich], 10 mM Tris, pH 8.0, and 50 mM NaCl). Samples were imaged on a Nikon Ti microscope equipped with Nikon oil-immersion 1.49 NA 100× Apo total internal reflection microscopy fluorescence objective (Nikon Instruments) and an Andor Ixon3 Ultra DU897 electron multiplying couple-charged device camera (Oxford Instruments). Illumination angle was set to 80° and frame size was set to 256 × 256 pixels in the center of the field of view. After acquisition of a reference total internal reflection fluorescence microscopy image, the 647-nm laser output power was increased to induce blinking. 20,000 frames were collected over 12 min.

The x and y coordinates and lateral localization accuracy of the blinking spots were extracted with the Nikon STORM software. The obtained dataset was loaded into Imaris with the “Super Resolution Localization Data to Spot” function. Spots that clearly localized outside of ENDS and/or had lateral localization precision >20 nm were omitted. The remaining spots were converted into an image. Images of straight segments of each detected ENDS were imported into Fiji to find the rotation angle needed to align them to the y axis. The coordinates of the spots from each segment were then imported into MATLAB and rotated by the calculated angle using an affine transform. The segments were divided along the vertical y axis into 100-nm sections. In each section, the range of the x coordinates was calculated, and the localizations were shifted to center at 0 using the following formula: x_centered = x − x_min − (x_max − x_min)/2. The normalized x coordinates of localizations in all ENDS were plotted on a frequency histogram with a 5-nm bin size. The width at half maximum of this histogram was measured to determine the average diameter of ENDS.

Human neutrophils were isolated from blood of healthy donors as described above. One million human neutrophils were incubated for 15 min in 100 µl Ca²⁺/Mg²⁺-free PBS supplemented with 1:1,000 CMRA, 5:10,000 Mito Tracker Green, and 5:10,000 ER Tracker Green (Thermo Fisher Scientific). Suspension was diluted with Ca²⁺/Mg²⁺-free PBS to 500 µl and centrifuged for 5 min at 400 g. Cells were resuspended in 100 µl Ca²⁺/Mg²⁺-free PBS supplemented with 2 µl (1 µg) anti-CD18 antibody (clone 1B4; BioLegend) to block β2-integrins and prevent neutrophil arrest and 1 µl (0.2 µg) CD16-AF647 antibody to label neutrophil surface. After 10 min, the cells were rolled in an ibidi flow chamber (#80161, coated with 0.5 µg/ml human recombinant E-selectin and blocked with casein) under 50–100 dyn/cm² shear stress generated with perfusion of HBSS containing Ca²⁺/Mg²⁺. The perfusion was stopped before the last cells would detach from the substrate. For DNA imaging, after rolling, Hoechst 33342 (Thermo Fisher Scientific) diluted 1:1,000 in PBS was added into the flow chamber. Neutrophils and ENDS were imaged immediately without fixation on a Zeiss LSM 880 microscope with 20× air (NA = 0.8) or 63× oil-immersion (NA = 1.4) objective. Images were processed with Fiji software. Median filter 1 was applied, and the dynamic range was adjusted to omit the background signal.

To examine the protein composition, ENDS were produced in a flow chamber system as described above with minor modifications. Isolated human neutrophils were rolled in a P-selectin–coated ibidi flow chamber in Ca²⁺/Mg²⁺-free PBS. Lack of Ca²⁺ and Mg²⁺ prevented integrin-mediated neutrophil arrest but still allowed selectin mediated rolling, although at lower efficacy. To compensate for this, the flow chamber coating density was increased via coating with 10 µg/ml recombinant human P-selectin–Fc-chimera. Following application of a 4-min-long shear stress (ramp to 100 dyn/cm²), the neutrophils formed ENDS and detached from the substrate. ENDS were isolated from the flow chamber via perfusion with 0.5 mM EDTA and 10 s of vortexing. For each donor, ENDS isolate of three flow chambers were pooled together. The suspension was centrifuged at 400 g for 5 min to minimize any neutrophil contamination, and the supernatants were concentrated to ∼30 µl with an Amicon Ultra 0.5 ml 10k filter unit (Millipore-Sigma). The ENDS were lysed for 1 h at 4°C by adding RIPA lysis and extraction buffer (Thermo Fisher Scientific) supplemented with PMSF protease inhibitor (Thermo Fisher Scientific) and cOmplete protease inhibitor cocktail (Sigma-Aldrich) and stored at −20°C.

The proteomes of five samples (ENDS from donors 1–3, neutrophils from donor 3, and a coated empty chamber as background control) were determined using LC/MS-MS proteomics at the Systems Mass Spectrometry Core at the Georgia Institute of Technology. Proteins were digested as described (Wiśniewski et al., 2009) and analyzed by high-pressure nano-LC-MS/MS as previously described (Naing et al., 2018) with the following modifications. Trypsin was used as the protease, and the spectral dataset was analyzed with Mascot using a UniProt Reference Proteome database (downloaded on January 7, 2018, with 20,416 entries) combined with a database containing common laboratory contaminants (www.thegpm.org/cRAP). Following removal of common laboratory contaminants (e.g., keratin) and exogenously added proteins (e.g., trypsin from the digestion, P-selectin and BSA used in coating the plates), and hemoglobin, the total numbers of PSMs were used as crude estimates for relative abundance in subsequent analysis. The background PSM value was subtracted from the PSM value of each detected protein. Proteins, which had more than one PSM ≤0 out of the three donor values, were excluded from further analysis. The PSM values of the three donors were averaged, and proteins with an average of two or fewer PSMs were omitted from further analysis. Gene names were rendered to the detected proteins, and gene ontology analysis was performed with Metascape (http://metascape.org). The ENDS proteome was compared with published datasets of neutrophil-derived EVs (Timár et al., 2013), NETs (Petretto et al., 2019), and neutrophil granule and plasma membrane (Rørvig et al., 2013) proteomes by rendering gene names to every protein and dataset overlap analysis with Venny (Oliveros, 2007). Similarities were quantified via calculating the Jaccard similarity index based on the following formula: JSI = 100 × (count of shared proteins / count of shared and unshared proteins).

For the fragmentation test, 10⁶ human neutrophils were incubated for 10 min in 100 µl Ca²⁺/Mg²⁺ HBSS supplemented with blocking anti-CD18 antibody (clone 1B4; BioLegend) to block β2-integrins and prevent neutrophil arrest. The cells were rolled in an ibidi flow chamber (coated with 0.5 µg/ml human recombinant E-selectin-Fc chimera and blocked with casein) under 150 dyn/cm² shear stress generated with HBSS containing Ca²⁺/Mg²⁺. This resulted in ENDS formation and detachment of the vast majority of neutrophils. To label the ENDS, the chamber was incubated for 10 min with 100 µl Ca²⁺/Mg²⁺ HBSS supplemented with 1 µl CD16-AF647 antibody. Without washing out the antibody, every 5 min, confocal images of the ENDS were captured for 4 h. The records were analyzed in Fiji by applying “triangle” threshold and 1-pixel median filter and measuring the object count and total surface area covered by ENDS. For the permeability test, two chambers of ENDS were prepared shortly after each other in the same way as described above. One of the chambers was stained with Fluo4 (30-min incubation with 5 µM Fluo4) immediately and was imaged for CD16-AF647 and Fluo4 while being perfused with 14 µM ionomycin. 7 h later, the other chamber was labeled and imaged in the same way. Records were analyzed with Fiji, triangle threshold and 1-pixel median filter were applied on the images, a mask was created from the CD16-AF647 signal, and the Fluo4⁺ surface area was measured within that mask.

ENDS were prepared in the same way as for the fragmentation test (see above). To measure the neutrophil contamination of the ENDS samples and estimate ENDS quantity, the entire flow chamber was scanned in tile/z-stack mode with a confocal microscope (Zeiss LSM880 through 10× air objective; Fig. S1 E). The flow chamber filled with Ca²⁺/Mg²⁺ HBSS was placed into a tissue culture incubator. After 7 h, the chamber was washed through with 100 µl Ca²⁺/Mg²⁺ HBSS, and the effluent was collected. After 24 h, the supernatant was harvested via pressing air into the chamber with a syringe. The collected supernatants were immediately centrifuged at 22,000 g for 10 min and two 50-µl aliquots of each sample were placed at −80°C. To measure the S100A8–S100A9 released by whole neutrophils, we loaded with 10⁴ human neutrophils in 100 µl Ca²⁺/Mg²⁺ HBSS into wells of a 96-well plate coated with E-selectin and blocked with casein similarly to the flow chambers. After a 24-h incubation in a tissue culture incubator, the supernatant was collected and processed in the same way as the ENDS supernatants. S100A8 and S100A9 levels were measured by ELISA following the manufacturer instructions (BioLegend). The z-stack/tile scan images of the CD16-AF647–labeled chambers were analyzed with Fiji. Z-stacks were collapsed with maximum projection and were processed with threshold “Yen” and median filter 2. Objects with area between 50 and 150 µm² were counted as single cells. Objects with area >150 µm² were considered as clusters of neutrophils, and their area was divided with the average single-cell area, which was 86, 80, and 82 µm² for the three samples.

Mice were injected i.p. with 40 mg/kg LPS (E. coli, strain O55:B5; Sigma-Aldrich) that was dissolved in sterile PBS at 10 mg/ml concentration. This treatment induces severe septic shock to which the mice succumb within 24 h (Soromou et al., 2014).

For ENDS detection with antibody labeling and imaging cytometry, 33 noninjected and LPS-injected and 2-h incubated ∼4-mo-old wild-type C57BL/6J female mice were bled retro-orbitally. 70 µl blood was mixed with 70 µl PBS supplemented with heparin at 2%. 50 µl blood plasma isolated by 5-min centrifugation at 1,000 g was supplemented with 1 µl heparin and 0.5 µl (0.25 µg) Ly6G-FITC and Ly6G-AF647 antibodies (clone 1A8; BioLegend). After 15-min incubation at room temperature in the dark, 4 µl the suspension was added to 96 µl annexin-5 binding buffer, and the resulting 50-times-diluted plasma sample was loaded into a round-bottom 96-well plate for scanning with ImageStream (Amnis ImageStream X Mark II; Luminex). Please note that analyzing 50-times-diluted plasma can interfere with the hydrodynamic focusing of ImageStream and may result in unsuccessful experiments in which the stream of plasma does not pass under the detection zone. Each sample was analyzed for 20 min with ImageStream with a 40× objective at highest sensitivity, lowest speed, core flow diameter 6 µm, automatic bead removal and maximum laser powers. The records were analyzed with the IDEAS (Luminex). The compensation matrix was created based on healthy blood plasma of healthy mice stained with one antibody at the time. “Fill” mask was applied on each fluorescent channel. Flow speed gate 66–67 was applied to eliminate objects artificially elongated (smeared) by unstable flow. Events with a bright-field area signal >5 µm² were excluded from the analysis. Fluorescently labeled antibodies tend to aggregate in suspension; on ImageStream records, the resulting clumps have an appearance similar to EVs. To avoid counting antibody clumps as false-positive counts, only objects positive for Ly6G-FITC and Ly6G-AF647 were taken as neutrophil derived. Single-spot count gate was applied on the detected channels to exclude events with multiple objects in the field of view. ENDS were gated as objects with aspect ratio and length in the FITC channel >0.65 and 3 µm. ENDS concentration measured by ImageStream was multiplied by 50 to account for the plasma dilution.

For antibody-free ENDS detection with confocal scanning, mice with neutrophils expressing plasma membrane–targeted EGFP (Ly6G-cre-mT/mG mice) were bred by crossing a strain carrying mT/mG genetic construct (Muzumdar et al., 2007) with a strain carrying Cre-recombinase in the Ly6G locus (Hasenberg et al., 2015). The first-generation heterozygous offspring were used for the experiments. Three female littermate Ly6G-EGFP mice of almost 11 mo of age were bled retro-orbitally before and 2 h after LPS injection. During bleeding, the mice were anesthetized with isoflurane inhalation, and 35 µl blood was collected into a heparin-coated glass capillary. To prevent blood clot formation, the blood was mixed with 35 µl PBS containing 2% of sterile heparin stock (#504036; Fresenius Kabi). The blood was centrifuged for 5 min at 1,000 g, and 30 µl blood plasma was collected and loaded into an ibidi flow chamber (#80626; ibidi) and was scanned with the Leica SP8 confocal microscope equipped with a resonant scanner and 40× oil-immersion objective. With the LAS Navigator, an overview tile scan was captured with the Spiral function, 30 fields of view were randomly marked, and each field of view was manually scanned in XYT mode (settings: 1,024 × 128, 1.5 zoom, 3–4 Airy unit aperture, 8× line average, and 633 laser power at 20%). The representative images were filtered with median 1.

Human blood samples were collected under protocols approved by LJI and University of California, San Diego institutional review boards (IRBs). Healthy human blood samples were obtained from LJI normal blood donation program (IRB numbers VD-057 and BHR1-KL). Septic human blood specimens were obtained from the University of California, San Diego clinical laboratory (IRB number 130973X; “Identifying immune correlates of disease processes”). Patients were selected based on SIRS criteria (two or more of the following: hypo- or hyperthermia, tachycardia, hyperventilation, leukocytosis, or leukopenia; Singer et al., 2016) and diagnosis of infection upon hospitalization. Blood was drawn into EDTA vacutainers and stored for up to 3 d at 4°C without agitation. Blood plasma was isolated via centrifugation at 1,000 g for 5 min. Several 50 µl aliquots were frozen at −80°C for S100A8–S100A9 complex analysis. For confocal imaging, 50 µl of the plasma was supplemented with 1 µl anti-CD16-AF647 antibody. After a 15-min incubation at room temperature in the dark, the suspension was spun into the glass of MatTek dishes, and the sample was scanned with Zeiss780 confocal microscope through a 40× oil-immersion 1.3 NA objective (Carl Zeiss Microscopy). For scanning with ImageStream, 25 µl plasma was supplemented with 1 µl heparin, 24 µl annexin-5 binding buffer, 5 µl annexin-5-eFluor450 (Thermo Fisher Scientific), 2.5 µl (0.5 µg) anti-CD66b-FITC (G10F5), and 1 µl anti-CD9-PE (20 ng, HI9a), anti-CD63-PE (100 ng, H5C6), anti-CD81-PE (200 ng, 5A6), and anti-CD16-AF647 antibodies (BioLegend). After 30-min incubation at room temperature, the mixture was diluted 50-fold in annexin-5 binding buffer (123-fold final dilution of plasma) and loaded into a 96-well plate for automatic scanning. Data were acquired with a 40× objective at highest sensitivity, lowest speed, core flow diameter 6 µm, automatic bead removal, and maximum laser powers. The records were analyzed with the IDEAS (Luminex). Fill mask was applied on each fluorescent channel. Flow speed gate 66–67 was applied to eliminate objects artificially elongated (smeared) by unstable flow. Objects with bright-field area >8 µm² were excluded from analysis. Single-spot count gate was applied on the detected channels to exclude events with multiple objects in the field of view. To avoid false-positive counts due to antibody clumps, the neutrophil-derived objects were gated as objects double positive for CD66b and CD16. ENDS were gated as CD66b⁺CD16⁺[CD9,63,81]⁻annexin-5⁻ population, aspect ratio (CD16) <0.6, and area (CD16) >4. Fragmented ENDS were gated as CD66b⁺CD16⁺[CD9,63,81]⁻annexin-5⁻ population, aspect ratio (CD16) >0.6. The ENDS populations were manually cleaned up for events with multiple objects in the field of view. Object counts and ENDS length values were exported for further analysis. S100A8–S100A9 contents of the healthy and septic plasma samples were measured with S100A8–S100A9 ELISA (BioLegend). Before measurement, the plasma aliquots were centrifuged for 10 min at 18,000 g, and the supernatant was diluted 40 times.

Paired or unpaired two-tailed t tests with or without Welch’s correction and correlation analyses were performed in Prism software. Significance values are indicated in the figure legends (*, P ≤ 0.05; **, P ≤ 0.001).

Fig. S1 shows that rolling neutrophils form ENDS. Fig. S2 shows the ultrastructure and composition of ENDS. Video 1 shows ENDS formation in vivo. Video 2 shows the formation of multiple ENDS in the same vessel in vivo. Video 3 shows ENDS sliding and detaching from the venular wall in vivo. Video 4 shows ENDS formation in a flow chamber. Table S1 shows ENDS proteome analysis. Table S2 lists clinical data of septic patients.

We are thankful to Dr. Erzsebet Ligeti for her valuable suggestion on the experimental designs.

The Ley laboratory was supported by National Institutes of Health program project grants P01 HL078784 and P01 HL151433. A. Marki was supported by an American Heart Association postdoctoral fellowship (17POST33410940) and the Tullie and Rickey Families SPARK Award at La Jolla Institute for Immunology. H. Winkels was supported by Deutsche Forschungsgemeinschaft award GZ WI 4811/1-1. M. Orecchioni was supported by an American Heart Association postdoctoral fellowship 18POST34060251. Z. Fan was supported by National Institutes of Health grant R01HL145454 and American Heart Association postdoctoral fellowship and career development awards 16POST31160014 and 18CDA34110426. The purchase of the Zeiss LSM 880 microscope was made possible by National Institutes of Health grant S10OD021831 awarded to the Microscopy Core Facility at the La Jolla Institute.

Author contributions: K. Ley, A. Marki, K. Buscher, and Z. Fan designed the experiments. A. Marki set up and performed most of the experiments. K. Buscher, C. Lorenzini, and M. Orecchioni performed flow cytometry and flow chamber experiments. M. Meyer prepared the samples for proteomic analysis. R. Saigusa performed the S100A8–S100A9 ELISA measurements. Y.-T. Yeh performed the SEM experiments. N. Hartmann did flow chamber experiments. G.J. Golden developed the septic mouse assay. J.M. Dan and J. Bui provided the septic blood samples. W.B. Kiosses performed STORM imaging. R. Ganesan, S. McArdle, and H. Winkels performed data analysis. Z. Mikulski provided microscopy expertise. Y. Altman provided ImageStream expertise. D. Smalley performed the proteomics analysis. J. Roth provided S100A8–S100A9 expertise. A. Marki, V. Nizet, and K. Ley wrote the manuscript. M. Kronenberg, S. Chien, J.D. Esko, and K. Ley financed the study. K. Ley supervised the study.