Role for formin-like 1-dependent acto-myosin assembly in lipid droplet dynamics and lipid storage

Lipid droplets (LDs) are cellular organelles specialized in triacylglycerol (TG) storage undergoing homotypic clustering and fusion. In non-adipocytic cells with numerous LDs this is balanced by poorly understood droplet dissociation mechanisms. We identify non-muscle myosin IIa (NMIIa/MYH-9) and formin-like 1 (FMNL1) in the LD proteome. NMIIa and actin filaments concentrate around LDs, and form transient foci between dissociating LDs. NMIIa depletion results in decreased LD dissociations, enlarged LDs, decreased hydrolysis and increased storage of TGs. FMNL1 is required for actin assembly on LDs in vitro and for NMIIa recruitment to LDs in cells. We propose a novel acto-myosin structure regulating lipid storage: FMNL1-dependent assembly of myosin II-functionalized actin filaments on LDs facilitates their dissociation, thereby affecting LD surface-to-volume ratio and enzyme accessibility to TGs. In neutrophilic leucocytes from MYH9-related disease patients NMIIa inclusions are accompanied by increased lipid storage in droplets, suggesting that NMIIa dysfunction may contribute to lipid imbalance in man.

I n cells, lipid droplets (LDs) are highly dynamic, well-connected with other organelles and actively communicating with one another [1][2][3] . LDs undergo major morphological changes depending on the metabolic status of cells. In energy surplus, stimulation of triacylglycerol (TG) storage results in LD enlargement, apparently by both fusion and lipid transfer between adjacent LDs 4,5 , with the extreme phenotype being unilocular LDs in mature adipocytes. The homotypic fusion of LDs is regulated by Fsp27 enriched at LD contact sites and its activity is controlled by Rab8a 6 . Conversely, TG breakdown by lipolysis is accompanied by LD shrinkage, dissociation and dispersion 7 . This process involves modification of LD resident proteins to allow docking and activation of lipolytic enzymes to digest the lipid core 8 , and eventual dispersion of smaller LDs to the cell periphery in a microtubule-dependent fashion 7,9 . How the initial dissociation of adjacent LDs is achieved, is essentially unknown.
Actin and actin-binding proteins have been reported to be associated with LDs [10][11][12] . In murine macrophages, pharmacological disruption of the actin cytoskeleton reduces LD size and lipid storage 10 and in 3T3-L1 cells, inhibition of actin dynamics through cofilin-1 depletion disrupts adipogenesis and lipid storage 13 . At the organismal level, pharmacological perturbation of acto-myosin turnover in zebrafish affects the recruitment of LDs from the yolk-blastodisc to the animal pole 14 . However, whether LDs associate with specific actin filament structures and what functional role they may have, has not been defined.
Here, we provide evidence for the presence of a novel actomyosin structure that associates with LDs to control their dynamics. We show that the actin-binding protein formin-like 1 (FMNL1) facilitates the assembly of non-muscle myosin IIa (NMIIa) containing actin filament arrays on LDs. Importantly, depletion of NMIIa from human cells reduces LD dynamics and alters their clustering propensity, resulting in supersized LDs and reduced TG hydrolysis. Together, these data suggest that NMIIa assembling with actin in focal points around LDs assists in the dissociation of clustered LDs.

Results
NMIIa depletion results in enlarged LDs and increased TGs. We conducted a proteomic profiling of LD-associated proteins. This revealed several actin-binding proteins (Supplementary Table 1), including NMIIa (encoded by the MYH9 gene), a structural component of contractile actin filaments. NMIIa has earlier been identified in the LD proteome of several organisms [10][11][12] , but its functional role at LDs remains unknown.
To investigate the role of NMIIa at LDs, we silenced the protein from human osteosarcoma U2OS cells ( Supplementary Fig. 1a-d), a cell line commonly used for studying cytoskeletal dynamics. For systematic analysis of LDs in U2OS cells, we developed automated image analysis tools to quantify LD abundance, size and clustering ( Supplementary Fig. 2a). Under normal growth conditions, roughly 100 LDs were detected in these cells using fluorescent LD dyes, with small LDs (below 0.5 mm 2 ) representing the overwhelming majority ( Fig. 1a,b, Supplementary Fig 2b,c). Oleic acid administration for 7 h roughly doubled the number of LDs ( Supplementary  Fig. 2b), whereas their relative size distribution was not changed (Supplementary Fig. 2c). Interestingly, depletion of NMIIa resulted in a LD rearrangement with larger LDs being significantly increased (about sixfold for LDs between 0.5 and 1.0 mm 2 and ninefold for LDs in the range of 1.5-2 mm 2 ; Fig. 1a,b). A similar increase in LD size was observed upon oleic acid administration in NMIIa-depleted cells (Fig. 1c,d). Enhanced formation of large LDs was also observed with another NMIIa-targeting siRNA, suggesting that the effect is specific (Supplementary Fig. 2d).
To further assess the specificity of the LD enlargement upon NMIIa depletion, we generated U2OS cells stably overexpressing an siRNA-resistant Cherry-NMIIa, or Cherry alone as control (Fig. 1e,f). We found that the accumulation of enlarged LDs upon NMIIa depletion was rescued by Cherry-NMIIa overexpression in control conditions (Fig. 1e,f) as well as upon oleic acid loading ( Supplementary Fig. 2e). Conversely, NMIIa overexpression in control cells resulted in the reduction of large LDs (Fig. 1f). In addition, acute inhibition of NMIIa ATPase activity for 1 h with blebbistatin 15 showed a similar tendency toward enlarged LDs ( Supplementary Fig. 2f). Together, these data provide additional evidence that the LD enlargement is a specific effect of NMIIa loss.
We next explored whether depletion of NMIIa affects neutral lipid storage. We found that the larger size of LDs in NMIIa-depleted cells was accompanied by a significant increase in TG deposition compared to control cells (Fig. 1g). A similar effect was obtained upon NMIIa inhibition with blebbistatin (Fig. 1h). The increased TG content was due to impaired TG breakdown rather than increased TG synthesis, as in NMIIa-silenced cells, TG hydrolysis was reduced ( Fig. 1i) but TG synthesis was unaffected (Fig. 1j). This is in line with the idea that the lower surface-to-volume ratio in larger LDs reduces the accessibility of hydrolytic enzymes to droplets 16 .
Importantly, general disruption of NMIIa-positive stress fibres through inhibition of the small GTPase Rho in filopodia or silencing of tropomyosin 4 (ref. 17) did not increase the proportion of large LDs in either control or oleic acid loaded conditions ( Supplementary Fig. 3). These data argue that the observed LD phenotypes in NMIIa-depleted cells are not due to problems in stress fibre assembly, but rather that NMIIa contributes to LD dynamics through a more specific mechanism.
NMIIa localizes transiently to LD dissociation sites. We investigated the localisation of NMIIa at LDs using confocal and three-dimensional structured illumination microscopy (3D-SIM). To facilitate this, we employed mild saponin permeabilization of cells prior to fixation, to reduce the cytoplasmic pool of NMIIa ( Supplementary Fig. 1e,f). We found that endogenous NMIIa localized to focal points around LDs visualized by the LD marker peptide HPOS 18 -Cherry stably expressed in U2OS cells (Fig. 2a,b), whereas the periodic labelling pattern of NMIIa along stress fibres was preserved (Fig. 2b). The LDassociated NMIIa patches often located between two neighbouring LDs or between clustered LDs (Fig. 2a,b, Supplementary Fig. 1g-k). On average, LDs were associated with three NMIIa patches and approximately one patch per LD-LD contact site ( Supplementary  Fig. 1g,h). Also filamentous actin (F-actin) was found at LDs and LD-LD contact sites and a subset of this actin colocalized with endogenous NMIIa at LDs (Fig. 2b, Supplementary Fig. 1i). For more detailed investigation of NMIIa localisation at LDs we used immuno-electron microscopy. This recapitulated, in addition to the expected periodic NMIIa immunoreactivity along stress fibres, NMIIa on the surface of individual LDs and at LD-LD contact sites (Fig. 2c).
To monitor NMIIa dynamics at LDs in living cells, we used Airyscan superresolution microscopy of GFP-NMIIa-expressing cells. This revealed distinct GFP-NMIIa patches transiently concentrating between clustered LDs (Fig. 2d . In many cases, transient NMIIa accumulation between two LDs was followed by their dissociation (Fig. 2d, Supplementary Movie 1). Plotting of NMIIa and LD intensity profiles of dissociating LDs before NMIIa accumulation, upon NMIIa enrichment and after LD segregation, showed a transient increase in NMIIa fluorescence during the partitioning of LDs (Fig. 2e, Supplementary Fig. 4b). Furthermore, we observed transient accumulation of fluorescently tagged actin between dissociating LDs ( Supplementary Fig. 4c,d and Supplementary Movie 3), suggesting that NMIIa and actin filaments cooperate in the dissociation of LDs.
NMIIa affects LD dissociation dynamics and LD clustering. To address if the loss of NMIIa affects LD dynamics, we monitored the association kinetics of clustered LDs, that is, LDs that were associated with each other. Coherent anti-stokes Raman scattering (CARS) microscopy was employed for label-free imaging of LDs in live cells (Fig. 3a, Supplementary Fig. 2g). Owing to the similar effects of NMIIa depletion and blebbistatin on LDs, the motor activity of myosin II appears functionally important, and other myosin isoforms, including NMIIb, seem not to fully compensate for the loss of NMIIa.
To obtain further quantitative information on the altered LD association characteristics upon NMIIa depletion, we extended the automated LD detection pipeline with a cluster analysis tool (Fig. 3c). This tool considers LDs to be clustered if the distance between droplet boundaries is smaller or equal to two pixels (0.266 mm), enabling the detection of clusters with a variable LD number and quantification of average LD sizes as well as LD size heterogeneity within clusters. In control cells loaded with oleic acid, the majority of LD clusters comprised two droplets, followed by a decreasing abundance of clusters containing more droplets (Fig. 3d). Interestingly, NMIIa-depleted cells loaded with oleic acid contained overall less-clustered LDs, with most pronounced reductions for clusters of three and four LDs (Fig. 3d). In parallel, the mean size of clustered LDs increased by B75% upon NMIIa silencing (Fig. 3e) and their size heterogeneity increased, as evidenced by greater than threefold increase in the s.d. of the size of clustered LDs (Fig. 3f,g). The observed changes in the size of clustered LDs were due to loss of NMIIa, as they were reversed in cells overexpressing the siRNA-resistant Cherry-NMIIa (Fig. 3h,i). These results indicate that NMIIa depletion results in reduced motility of associated LDs and a larger size heterogeneity between associated LDs. Moderate alterations in the balance of LD dissociation/re-association may over time result in prominent LD size changes, with bigger LDs deriving from coalescence of neighbouring LDs, as suggested by the observed fusion between immotile, clustered LDs ( Supplementary Fig. 4e,f, Supplementary Movie 7). FMNL1 localizes to LD dissociation sites. Our proteomic analysis also revealed the actin regulatory protein FMNL1 among the LD-associated proteins (Supplementary Table 1). FMNL1 has several activities, such as stimulation of actin polymerisation and bundling of actin filaments 19,20 , providing a candidate to drive the formation of actin filaments at LDs. We therefore investigated the localisation of endogenous FMNL1 in U2OS cells using Airyscan and 3D-SIM imaging. FMNL1 antibodies revealed punctate cytoplasmic as well as perinuclear immunoreactivity, in agreement with previous studies 21 . To reduce cytoplasmic immunoreactivity, cells were saponin permeabilized pre-fixation, similarly as for NMIIa staining ( Supplementary Fig. 5d,e). This highlighted punctate FMNL1 labelling on the surface of LDs (Fig. 4a,b). We found on average three FMNL1 patches associated per LD and of these approximately half localized to LD-LD contact sites ( Supplementary Fig. 5f,g). Interestingly, FMNL1 patches at LDs were closely apposed to NMIIa dots, as visualized by myosin light chain staining (Fig. 4c, Supplementary  Fig. 5h). Live cell Airyscan video microscopy demonstrated that GFP-FMNL1 localized transiently to LD dissociation sites (Fig. 4d, Supplementary Movie 8). Of the 155 LD dissociation events quantified, over 60% showed FMNL1 foci between the dissociating LDs ( Fig. 4f) and in nearly all of them, FMNL1 was present already before the separation of LDs occurred (Fig. 4g). When a similar analysis was performed from live cell Airyscan videos of GFP-NMIIa expressing cells, we observed about half of the LD dissociation sites to contain NMIIa accumulations (Fig. 4e,f; Supplementary Movie 9). Interestingly, NMIIa seemed to be recruited to the dissociation sites after FMNL1, as in B1/3 of the events the NMIIa accumulation appeared once LDs had started to separate from each other (Fig. 4g). We also noted that after dissociation some LDs underwent reassociation and that FMNL1 could stay associated with LDs during such events (Supplementary Fig. 6 Supplementary Movies 10 and 11).
To investigate the interrelationship of FMNL1 and actin upon LD dissociation, we performed triple-colour live cell Airyscan imaging of GFP-FMNL1, BFP-Lifeact and LDs in a region below the nucleus, to minimize cytoplasmic background (Fig. 5a). This revealed a transient focal accumulation of GFP-FMNL1 in a LD cluster, followed by dynamic accrual of BFP-Lifeact between dissociating LDs (Fig. 5b,c; Supplementary Movie 12). We also monitored the accumulation of a larger FMNL1 patch apparently driving the separation of neighbouring LDs ( Fig. 5d-f; Supplementary Movie 13). This enabled focal intensity measurements of FMNL1 and Lifeact, further substantiating the idea that an increase in FMNL1 intensity preceeded that of Lifeact between LDs (Fig. 5f).
FMNL1 drives actin assembly on LDs and recruits NMIIa to LDs. Together, these data suggest that FMNL1 can generate actin filaments on LDs, which in turn may recruit NMIIa. In support of this idea, we found a major reduction of NMIIa in LD fractions of FMNL1-silenced cells (   LDs. Finally, FMNL1 depletion resulted in the formation of enlarged LDs analogously to NMIIa depletion ( Supplementary  Fig. 8a). This was observed despite the fact that FMNL1 depletion also inhibited fatty acid uptake, thereby lowering TG storage ( Supplementary Fig. 8b,c). Together, these data suggest that FMNL1 mediates LD localisation of NMIIa, possibly by generating specific actin filaments to which NMIIa binds.
To provide more direct evidence for actin filament assembly on LDs and the potential role of FMNL1 in this process, we designed an in vitro assay, which allows the monitoring of actin polymerisation on isolated LDs using total internal reflection fluorescence (TIRF) microscopy. LDs were isolated from oleic acid-loaded U2OS cells, labelled with LipidTOX green, and then mixed with rhodamine actin and unlabelled actin in a TIRF reaction chamber (Fig. 6d). Actin filament assembly was monitored by video TIRF microscopy (Fig. 6d,e). We found that under these conditions B6% of isolated LDs associated with actin filaments and displayed time-dependent growth of actin filaments from them ( Fig. 6e-g). That only a small fraction of LDs displayed actin nucleation may result from partial dissociation of FMNL1 from LDs during isolation or absence of soluble FMNL1-interacting proteins in the in vitro system. Thus, the process may be more efficient in living cells. Addition of the formin inhibitor SMIFH2 (ref. 22) or use of LDs isolated from FMNL1-depleted cells in the in vitro assay blunted actin filament association with LDs (Fig. 6g).   sensorineural hearing loss, renal failure and presenile cataract, the pathogenesis of which are incompletely understood 24 . To investigate whether cells from patients with MYH9-RD display a LD-related phenotype, we isolated peripheral blood neutrophilic leucocytes from two patients with MYH9-RD carrying a point mutation in the C-terminal non-helical tail of NMIIa (S. Ryhänen, personal communication) and two healthy individuals. These patients present with marked macrothrombocytopenia and several family members have been diagnosed with hearing loss 25 , in accordance with the site of the mutation 23 . By NMIIa immunostaining, the diagnostic NMIIa inclusion bodies of the disease were evident and the overall NMIIa immunoreactivity was decreased in the patient neutrophils (Fig. 7a,b). Under basal conditions (fasting blood samples, no lipid loading in vitro), both control and patient neutrophils displayed few LDs (approximately seven LDs per cell) (Fig. 7c,d).
Remarkably, upon 1 h of oleic acid loading in vitro, LD accumulation was more pronounced in the patients' neutrophils, resulting in roughly twofold more LDs compared with controls ( Fig. 7d,e). In parallel, there was a tendency for the enlargement of LDs in the patients' cells, especially in the bigger LD size classes (Fig. 7e). Thus, the LD phenotypes in the patient neutrophils appeared partly reminiscent of those observed in NMIIa-silenced U2OS cells. Together, these findings provide the first evidence that NMIIa dysfunction in man causes a lipid imbalance, possibly contributing to the pathogenesis of MYH9-RD. For instance, neutrophil LDs can modulate the immune response, as impaired LD lipolysis was found to be accompanied by reduced lipid mediator release 26 .

Discussion
Actin-binding proteins are prominent constituents of the LD proteome but their specific roles on LDs are not well understood.
In this study, we describe a functional interplay of the actin-polymerising factor FMNL1 and NMIIa, a force generating protein, in LD dynamics. NMIIa is generally known as a major constituent of contractile stress fibres, regulating cell Recent evidence indicates that NMIIa can also act on specialized actin filament structures, for instance, during mitochondrial fission 28 . We found that NMIIa transiently accumulates at sites where LDs separate from one another. LDs are known to undergo homotypic associations and dissociations, resulting in the formation of dynamic LD clusters. Our data show that NMIIa has a major impact on cellular LDs, with NMIIa depletion resulting in enlarged LDs and increased TG storage owing to impaired lipid breakdown. We postulate that impaired dissociation/prolonged association of LDs enhances their coalescence, either via direct fusion ( Supplementary Fig. 4e,f; Supplementary Movie 7) and/or via lipid transfer 5 . This would result in an increased size heterogeneity of LDs and a net consumption of LDs within a cluster, with a concomitant reduction of clusters. Indeed, NMIIa silencing reduced LD dissociations. In parallel, the LD clustering propensities were altered: there were fewer LD clusters and in these big LDs often associated with small ones. This is strikingly different from control cells where similarly sized LDs formed clusters.
NMIIa requires a preformed actin filament to exert its function. But how could an actin filament be generated at LD dissociation sites? We demonstrate that FMNL1, a protein with actin polymerisation activity, localizes to LD dissociation sites. It can thus provide a seed for acto-myosin assembly, as depletion of FMNL1 abolishes NMIIa association with LDs. Furthermore, we found that isolated LDs have the ability to induce actin filament assembly in vitro. This activity was dependent on FMNL1, as pharmacological inhibition of formins or depletion of FMNL1-blunted actin filament assembly on isolated LDs. Based on these observations, we suggest that FMNL1 localizes to LD dissociation sites to nucleate actin filaments that are functionalized by NMIIa, providing a force to assist in LD dissociation. This idea is supported by the temporal association of the actin-binding proteins with LDs, with FMNL1 accumulation typically occurring between closely apposed LDs before dissociation, potentially defining LD dissociation sites. Although most dissociating LDs also harbour NMIIa, it may also be recruited to the site once LDs are moving apart. Considering that LDs also undergo rapid reassociations, an alternative, although not mutually exclusive explanation is that FMNL1, in concert with actin and NMIIa, prevents the reassociation and fusion of LDs. Both mechanisms, promotion of LD dissociation and prevention of LD reassociation, highlight the importance of LD dynamics in controlling the coalescence and surface to core ratio of LDs, and thereby the accessibility of modifying enzymes to core TGs. Cell culture and transfection. U2OS cells kindly provided by Marikki Laiho (John Hopkins University, USA) were cultured in Dulbecco's Modified Eagle's Medium (DMEM) with 15% FBS, supplemented with penicillin/streptomycin (100 U ml À 1 each) and L-glutamine (2 mM). Stable U2OS Cherry-NMIIa, Cherry-NMIIa-siRes and Cherry-HPOS cell lines were selected with 500-600 mg G418 per ml culture medium. SiRNA treatments were performed for 48 or 72 h using Hiperfect (Qiagen) or RNAiMax (Thermo Fisher) using reverse transfection with 50 nM siRNA. Oleic acid loading was performed during the siRNA treatments. Plasmids were transfected using Lipofectamine LTX (Thermo Fisher) with the PLUS (Thermo Fisher) reagent. Cells were tested for Mycoplasma infection.
The treatments were combined for LD isolation to maximize the yield. Cells were washed with cold phosphate buffered saline (PBS) and resuspended in 3.5 ml hypotonic lysis buffer (HLM; 20 mM Tris pH 7.4, 1 m M ethylenediaminetetraacetic acid), and incubated on ice for 10 min. Cells were disrupted by repetitive passaging through a 25 g needle, 3 ml of cell suspension was transferred to a Beckmann polyallomer tube (14 Â 95 mm) (No. 331374) and mixed with 1.5 ml 60% sucrose solution, overlayed with 4 ml 5% sucrose solution and 4 ml HLM Buffer. Samples were centrifuged at 28,000 g with a SW40 rotor for 2 h. The LD fraction was recovered by tube slicing and subjected to liquid chromatography-mass spectrometry (LC-MS/MS). For LC-MS/MS, the samples were prepared as follows: the cysteine bonds were reduced with 5 mM TCEP (Sigma-Aldrich), alkylated with 10 mM iodoacetamide (Sigma-Aldrich), and the proteins were digested to peptides by trypsin (Promega, Madison, WI). After overnight incubation at 37°C, samples were quenched with 10% TFA, purified with C18 Micro SpinColumns (Harvard Apparatus, Holliston, MA) and re-dissolved in 30 ml 0.1% TFA, 1% CH3CN. The LC-MS/MS analysis was carried out on an EASY-nLCII nanoflow system (Thermo Scientific) connected to a Velos Pro-Orbitrap Elite hybrid mass spectrometer (Thermo Scientific) using the Xcalibur version 2.7.1. The peptides were separated using 10 cm analytical column (Thermo Fisher Scientific) with 60 min gradient ranging from 5 to 35% buffer B (0.1% formic acid in 98% acetonitrile and 2% water), followed by a 5 min gradient from 35-80% buffer B and 10 min gradient from 80-100% buffer B at a flow rate of 300 nl min À 1 . Proteome Discoverer software (Thermo Scientific) together with SEQUEST search engine was used for peak extraction and subsequent peptide and protein identification. Error tolerances on the precursor and fragment ions were 15 ppm and 0.6 Da, respectively. Database searches were limited to fully tryptic peptides with maximum one missed cleavage. For peptide identification the FDR was set to o5%.
LD isolation and actin assembly on isolated LDs. U2OS cells (6-7 10 cm dishes) were treated overnight with 400 mM oleic acid in culture medium. EDTA in HLM buffer was replaced with EGTA. LD isolation was performed as described above and LD fractions were retrieved from the top with a 1 ml pipette. In vitro TIRF imaging was performed as previously described 33 , but muscle actin was substituted with non-muscle actin (Cytoskeleton Inc.), prepared according to the manufacturer's instructions, and non-muscle rhodamine actin (Cytoskeleton Inc.) was used for labelling of the filaments. Videos were obtained with 0.15 mM unlabelled non-muscle actin, 0.065 mM rhodamine non-muscle actin and 0.43 mM profilin concentrations. Profilin was purified from cow spleen as previously described 34 SMIFH2 formin inhibitor 22 (Sigma-Aldrich) was used at 100 mM. TIRF imaging was performed in mPEG-Silane (MW 5k)-coated chambers with a Nikon Eclipse Ti-E N-STORM microscope, equipped with Andor iXon þ 885 EMCCD camera and Â 100 Apo TIRF oil objective NA 1.49, a 561 nm laser line was used for visualisation of Rhodamine actin. Actin filament polymerisation was followed in 0.1 s intervals for 15 min after addition of the reaction into the imaging chamber.
LD preparation for western blot analysis. U2OS Cherry-HPOS cells were seeded in 2-4 10 cm dishes and transfected with siControl or siFMNL1. After 24 h cells were incubated with 400 mM oleic acid in control medium for an additional 24 h. Cells were washed 2 Â with Dulbecco's phosphate-buffered saline (DPBS) and then scraped into 5 ml DPBS. Cells were pelleted by centrifugation at 730 g for 5 min and resuspended in 900 ml HLM buffer. Cells were broken by repetitive passaging through a 25 g needle, and 750 ml of the suspension was mixed with 750 ml HLM buffer with 60% sucrose, overlayed with 1 ml HLM with 5% sucrose followed by 1.5 ml HLM buffer. LDs were floated by centrifugation at 28,000 g for 2 h in a SW60 rotor. Of the LD fraction, 10% was removed for lipid determination and the rest was subjected to acetone precipitation with 10 Â volume of À 75°C acetone. Samples were incubated at À 20°C degree overnight and centrifuged at 3,300 g for 100 min. Proteins were resuspended in 2 Â loading buffer by sonication and subjected to western blot analysis. Original scans of the respective western blots shown in Fig. 6a can be found in Supplementary Fig. 9.
Automated LD analysis. U2OS cells were cultured on coverslips or 96 well Screenstar plates (Greiner) and incubated with or without oleic acid for 7 h. Cells were treated with 2 mM CellTracker Red for 30 min to 1 h in culture medium with or without oleic acid, washed 3 Â with PBS and fixed with 4% paraformaldehyde in 250 mM HEPES, 1000 mM CaCl, 100 mM MgCl, pH 7.4. Cells were washed 3 Â with PBS and stained with 1 mM 4 0 ,6-diamidino-2-phenylindole (DAPI) and 100 ng ml À 1 LD540 (or 1:1,000 LipidTox green) in PBS for 30 min at room temperature (RT), washed 3 Â with PBS and mounted with PBS and epoxy glue. For every treatment 10 image stacks were acquired for DAPI, LD and cell tracker channels using a Nikon Eclipse Ti-E inverted microscope equipped with a 60 Â PlanApo VC oil objective NA 1.40 (or 40 Â Planfluor objective with NA 0.75 and 1.5 zoom). Image stacks were automatically deconvolved using the Huygens batch processing application (Scientific Volume Imaging) and deconvolved image stacks were transformed into maximum projection images using custom MatLab scripts. Cell segmentation and LD detection was performed with CellProfiler 1.0.5122 (CellProfiler.org) 35 in a hierarchical manner. First, cell nuclei with a typical diameter between 100 and 250 pixels were detected in DAPI images based on the Otsu adaptive thresholding method. Touching nuclei were separated by build in intensity methods. Second, the cytoplasm was detected in CellTracker Red images using intensity propagation based on the Otsu global thresholding method using the identified nuclei from the first step as a seed point. Third, LDs were detected using a combination of a A-trous spot detection method with Otsu adaptive thresholding, enabling the detection of small and large LDs separately. The detected LDs were further processed with custom Matlab scripts, which corrected for large LDs misclassified as several small LDs. Furthermore, the custom Matlab scripts contained an algorithm to identify clustered LDs. Data visualisation was done using Excel and Origin 8.6.
Lipid extraction, analysis and TG hydrolysis. Lipid extraction was performed as described previously 36 . Cells were washed 3 Â with PBS and scraped into 2% sodium chloride and lipids were extracted with chloroform methanol in a 2:1 ratio. Solvents were evaporated and lipids dissolved in 2:1 chloroform: methanol for application to thin layer chromatography (TLC Quantification of NMIIa and FMNL1 localisation at LDs. Confocal image stacks of U2OS Cherry-HPOS cells stained with anti-FMNL1 or anti-NMIIa antibodies were deconvolved using Huygens deconvolution software. NMIIa and FMNL1 patches with a preset minimum intensity were counted at LDs with a diameter of Z1 mm by inspecting each individual section of a z-stack. On average, 10 LDs were counted per cell. NMIIa or FMNL1 patches, which associated with two neighbouring LDs were counted as patches at LD contact sites. Live cell imaging. U2OS cells were seeded in LabTEK chamber slides coated with 5 mg human fibronectin (Roche Diagnostics). Cells were treated with 200-400 mM oleic acid for 8 h or overnight. Cells were imaged in fresh culture medium or culture medium containing oleic acid including 10 mM HEPES pH 7.3. LDs were stained either with LipidTOX green or LipidTOX deep red. Live cell microscopy was performed with a Leica TCS SP8 confocal microscope or a LSM880 confocal microscope with an Airyscan module. Live cell microscopy for fluorescently tagged NMIIa at LDs was performed in nine independent experiments; GFP-FMNL1 at LDs in four independent experiments; GFP-FMNL1 with BFP-Lifeact at LDs in six independent experiments and BFP-Lifeact/BFP-actin at LDs in four independent experiments. LD dynamics with blebbistatin was assessed in four independent experiments and with cytochalasin D in one experiment.
CARS microscopy. CARS microscopy is a non-linear label-free imaging method especially well suited for LD imaging 37,38 . CARS video microscopy was done with non-labelled cells by adjusting pump and stokes laser to the vibration frequency of C-H groups (2845 cm À 1 ). An infrared corrected 25 Â objective (HCX IR APO L, NA 0,95) was used for CARS image acquisition. Images were acquired every 2 s over a time course of 5 min. Live cell CARS microscopy for LD association dynamics was assessed in three independent experiments (Fig. 3a,b, Supplementary Movie 4).
Neutrophil isolation and lipid loading. Freshly isolated peripheral blood from MYH9-RD patients and from healthy individuals was used for neutrophil isolation using Polymorphprep (Axis-Shield) and contaminating erythrocytes were lysed using 155 mM NH 4 Cl, 12 mM NaHCO 3 , 0.1 mM EDTA. Neutrophils were resuspended in Hank's balanced salt solution without calcium and magnesium and adhered to poly-D-lysine-coated cover glasses for 15 min or transferred to 24 well plates with poly-D-lysine-coated cover glasses and incubated with 10 mM oleic acid for 1 h at 37°C. Neutrophils were fixed with 4% paraformaldehyde and subjected to NMIIa or LD stainings. Neutrophils were imaged with a Nikon Eclipse Ti-E N-STORM microscope, equipped with a 100 Â Apo TIRF oil objective NA 1.49. LD images were automatically deconvolved using Huygens software (Scientific Volume Imaging). Maximum intensity projections were generated using MatLAB and thresholded LD images were quantified using the ImageJ analyse particles tool. The MYH9-RD patients provided a written informed consent in accordance with the Declaration of Helsinki.
Data analysis. Data are presented as mean±s.e.m. Statistical significance was calculated with students t-test (two tailed) using Microscoft Excel.
Data availability. The data that support the findings of this study are available in the article, including Supplementary Files, or are available from the authors upon reasonable request.