The organization of leukotriene biosynthesis on the nuclear envelope revealed by single molecule localization microscopy and computational analyses

The initial steps in the synthesis of leukotrienes are the translocation of 5-lipoxygenase (5-LO) to the nuclear envelope and its subsequent association with its scaffold protein 5-lipoxygenase-activating protein (FLAP). A major gap in our understanding of this process is the knowledge of how the organization of 5-LO and FLAP on the nuclear envelope regulates leukotriene synthesis. We combined single molecule localization microscopy with Clus-DoC cluster analysis, and also a novel unbiased cluster analysis to analyze changes in the relationships between 5-LO and FLAP in response to activation of RBL-2H3 cells to generate leukotriene C4. We identified the time-dependent reorganization of both 5-LO and FLAP into higher-order assemblies or clusters in response to cell activation via the IgE receptor. Clus-DoC analysis identified a subset of these clusters with a high degree of interaction between 5-LO and FLAP that specifically correlates with the time course of LTC4 synthesis, strongly suggesting their role in the initiation of leukotriene biosynthesis.


Introduction
All cells must integrate and transduce multiple extracellular signals to achieve an appropriate functional response. In mast cells, the classic pathway of cell activation in response to an allergen is initiated by antigen binding to allergen-specific IgE antibodies coating mast cells via the IgE receptor (FcεR1; FcERI; UniProtKB: P12319) [1][2][3]. Antigen binding triggers the aggregation of FcεR1 receptors, activating their downstream pathways by recruiting a series of kinases to the cytoplasmic tail, and prolongs its presence on the plasma membrane [2,[4][5][6][7]. The kinase cascade sets in motion multiple processes including degranulation and the synthesis of leukotriene B 4 (LTB 4 ) [8,9] and LTC 4 [10,11]. Activated mast cells predominantly make LTC 4 [10,11] and to a a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 determine 5-LO and FLAP organization on the nuclear envelope, two different computational analyses were applied following image acquisition (Fig 1). Tab-delimited text files (.txt) containing the localization list from each image set were output for each approach. We employed a clustering and colocalization algorithm, Clus-DoC [31] to two-color dSTORM data (purple). Clus-DoC assigns a degree of colocalization (DoC) score to each localization and determines percent colocalization, as well as cluster properties including the number of clusters in a region of interest (ROI), cluster area, and density. We also calculated other parameters such as the percent of interacting localizations inside clusters. In a second, supportive approach, we developed a novel unbiased clustering analysis based on a variable bandwidth mean-shift algorithm for conventional STORM data (orange). This analysis makes no assumptions about cluster size and is especially valuable for clusters curved around a membrane (S1 Fig).

Relationship of 5-LO and FLAP on the nuclear envelope
RBL-2H3 cells were primed with anti-TNP IgE and then stimulated with TNP-BSA for 2, 5, 7, and 10 min. Total media concentrations of LTC 4 at each time point were measured, with significant LTC 4 accumulating by 5 min and reaching peak levels by 7 min (Fig 2A). Cells were Initially, a .txt file with tab-delimited x/y coordinates of all localizations from both channels was generated. To analyze two-color dSTORM (purple), FLAP localizations were used to define a region of interest (ROI) around the nuclear envelope, which was also applied to 5-LO within the Clus-DoC user interface. Degree of colocalization (DoC) scores were calculated for each localization. DBSCAN detected clusters and defines cluster contours, yielding cluster maps and cluster properties. For conventional STORM (orange), FLAP localizations were used to define an ROI around the nuclear envelope, or in the perinuclear region and nucleus for 5-LO. Within unbiased cluster analysis, cluster maps with cluster properties are determined. Data from both methods were combined to produce a model of membrane reorganization of 5-LO and FLAP.

5-LO and FLAP colocalize on the nuclear envelope
Colocalization of 5-LO and FLAP in dSTORM images of activated mast cells at NT, and 2, 5, 7, and 10 min post-activation was determined using Clus-DoC [31]. Analysis was restricted to the nuclear envelope by manually drawn regions of interest (ROIs) based on FLAP localization. Fig 3A shows both the localization maps (left panels) and colocalization maps colored by DoC score, -1 (anticorrelated) to 1 (correlated) (right panels). The representative cells shown are identical to those shown in Fig 2B. The increase in orange and red localizations at 5 and 7 min represent increased colocalization between 5-LO and FLAP (Fig 3A, right panels). At 5 and 7 min a two-fold increase in the percent of 5-LO molecules that were colocalized with FLAP was detected compared with NT ( Fig 3D, left). No such difference was detected comparing FLAP to 5-LO due to high levels of FLAP compared with 5-LO (Fig 3D, right). These times correspond with maximal observed LTC 4 synthesis (Fig 2A).

The formation of higher order assemblies of 5LO and FLAP
If higher order assemblies of 5-LO and FLAP play a regulatory role in LT synthesis, their formation and disassembly on the nuclear envelope should be correlated with LTC 4 production. We employed the Clus-DoC algorithm to test this hypothesis. The algorithm measured the colocalization of the localizations and then sorted them into clusters or excluded them as outliers. The algorithm then calculated the characteristics of the identified clusters, allowing changes in molecular organization to be detected. Cluster maps of 5-LO and FLAP for the cells analyzed in Figs 2 and 3 are shown in S3 Fig; cluster contours are indicated by the black outline and points outside of clusters are shown in gray. Two levels of cluster limits were applied as modifications of Clus-DoC. First, a threshold of >5 localizations for 5-LO or >10 localizations for FLAP were used to define a true cluster. A higher threshold was chosen for FLAP because it functions as a homotrimer [24], increasing the chance that identified clusters contain at least 3 functional FLAP molecules. A second level was used to distinguish clusters with no colocalization (no interaction clusters (NIC); white bars), a high degree of colocalization (high interaction clusters (HIC); black bars) denoting at least 5 localizations with a DoC score of � 0.4, and all other clusters (low interaction clusters (LIC); gray bars). The properties of clusters were compared between HIC, LIC, and NIC at different time points (Fig 4). The number of HIC per ROI increased from 0.3 at NT and 2.3 at 2 min to 6.4 and 7.8 at 5 and 7 min, respectively, and returned to 0.6 at 10 min ( Fig 4A). Because there were <3 ROIs containing HIC at NT, 2 and 10 min, we excluded them from further analysis. The appearance and disappearance of HIC correlates with peak LTC 4 synthesis, suggesting that the formation of HIC is required for LTA 4 synthesis and that disassembly of these clusters is a critical step in the termination of synthesis.

Properties of HIC link them to LT synthesis
Changes in six separate properties of HIC are temporally correlated with LTC 4 synthesis. First, their area is approximately 2-fold larger than LIC and NIC at 5 and 10 min after activation, respectively ( Fig 4B). Similarly, the average number of 5-LO localizations in HIC are 10-20 fold higher than those in NIC and LIC (Fig 4B). In contrast, number of 5-LO localizations in NIC and LIC remained unchanged over time (Fig 4C). The number of FLAP molecules in HIC was approximately 3-fold higher (5 min) and 2-fold higher (7 min) than in LIC and NIC, which did not change over the time course of the experiment (Fig 4D).  The relative density of 5-LO in clusters remained unchanged within NIC and LIC over time. Relative density is calculated as the average of local density within 20 nm of each localization in the cluster, divided by the average cluster density, providing a measure of the distribution of localizations within a cluster. However, the relative density of 5-LO in HIC was elevated 2-fold at 5 min and 15-fold greater than that of NIC and LIC at 7 min ( Fig 4E). Interestingly, FLAP relative densities for NIC were 2-fold lower at 5 and 7 min than in LIC or HIC ( Fig 4F).
We next calculated the percent of all 5-LO and FLAP interacting localizations that are in clusters. At 2, 5 and 7 min after activation, the vast majority of 5-LO molecules were found in either LIC or HIC ( Fig 4F). This was true for FLAP at all time points (Fig 4G). Taken together, these data link the formation and disassembly of HIC to LTC 4 formation.

The dynamic organization of 5-LO analyzed by conventional STORM
Conventional STORM experiments in combination with unbiased cluster analysis revealed similar cluster properties to those revealed using Clus-DoC. The cells were prepared identically but stained only for 5-LO and imaged by STORM. Because conventional STORM was used in these experiments, FLAP localizations were not available to define the nucleus in the same cell. Therefore, the whole nucleus and perinuclear area was included in the ROI. RBL-2H3 cells were analyzed at 0, 2, 5, and 10 min after cell priming and activation. 5-LO localizations in representative cells acquired by STORM are shown in Fig 5A (top panels). Unbiased cluster analysis grouped localizations into clusters of >3 localizations. The organization of 5-LO was not uniform across the nucleus and the region that included the nuclear envelope, both in size and shape ( Using 95% confidence intervals (CI), the mean number of localizations associated with clusters (pooled across all cells of a condition and time point) of 5-LO increased from 32 to 38, 2 min after the addition of antigen and peaked at 5 min, having 45 localizations per cluster (a 25% increase). After 10 min, the mean number of localizations dwindled to 22, which is lower than unstimulated conditions (0 min), suggesting complex disassembly (Fig 5B). The mean size of 5-LO cluster areas was also analyzed using 95% CIs. In non-activated cells, there were small numbers of 5-LO associated with the nuclear envelope and a correspondingly low number of localizations per cluster were detected in unstimulated conditions. The mean area of 5-LO clusters identified on the nuclear envelope was~2.1×10 4 nm 2 (Fig 5C). At 2 min following cell activation, the mean cluster area climbed to 3.4×10 4 nm 2 and then to 4×10 4 nm 2 at 5 min and 10 min (Fig 5C). The mean cluster density declined over time, decreasing by 50% at 5 min and returning to control levels at 10 min post-activation (Fig 5D).
To determine whether 5-LO clusters of a specific size, area, or density are assembled after cell activation, we employed weight-normalized histograms of the ROIs in Fig 5. There was a progressive increase in the frequency of the clusters containing the highest numbers of ANOVA with Tukey's post-hoc multiple comparisons test was performed to determine significance among interaction groups where � p < 0.05, �� p < 0.005 and ���� p < 0.0001; within timepoint, HIC-NIC or HIC-LIC indicated by black brackets, LIC-NIC by orange line; between timepoints by blue line. Bars   localizations (greater than 250 localizations or more) at 2 and 5 min stimulation, then decreasing at 10 min (S4 Fig). The organizational shift from a high frequency of clusters in control cells containing few localizations per cluster (less than 100) to significantly more localizations per cluster (greater than 250) occurred at 2 and 5 min after cell activation, concurrent with LT production (Fig 2A). Stimulating cells for 10 min resulted in the same pattern of localization frequency as control. The mean number of 5-LO localizations per cluster was discernably higher following cell activation at 2 and 5 min compared to control (S4 Fig, inset). Following stimulation, the frequency of clusters with larger areas increased over time (S4 Fig, inset), while the densities of clusters remained constant (S4 Fig, inset).
The formation of higher order FLAP assemblies was detected using conventional STORM in combination with unbiased cluster analysis. FLAP is primarily localized to the nuclear envelope and to a lesser extent on the ER membrane. Because we were interested in analyzing FLAP distribution in its primary localization, we used FLAP localizations to define the nuclear envelope for unbiased cluster analysis. We first probed whether FLAP would reorganize within the nuclear envelope following IgE priming and antigen activation. RBL-2H3 cells were primed and activated for 0 or 7 min. Fig 6A shows the STORM and cluster analyses images with convex hulls (middle panels) and numbers per cluster (lower panels, pseudocolor legend below). Weight-normalized histograms show that distribution of clusters shifts to the left after addition of antigen, whereas cells only primed with IgE contain a higher frequency of clusters with more localizations (Fig 6B). There was an increase in clusters containing between 250 and 450 localizations corresponding to~83-150 trimers when cells were primed and activated (Fig 6B). A similar pattern was observed for cluster areas. Cells that were only primed with IgE contained a greater number of clusters with larger cluster areas compared to cells that were primed and activated with antigen ( Fig 6C). The frequency profile for cluster density was marginally different between IgE-primed and IgE-antigen activated cells (Fig 6D). The mean number of localizations per cluster decreased approximately 25% following addition of antigen ( Fig  6B, inset) and the mean cluster area decreased from 1.4×10 5 nm 2 to 0.8×10 5 nm 2 following antigen stimulation (Fig 6C, inset). In parallel, the mean density of clusters increased~25% from 1.7x10 -3 localizations/nm 2 to~2.4x10 -3 localizations/nm 2 after antigen stimulation, suggesting that following antigen, FLAP clusters become more compact.
Since the interaction of 5-LO with FLAP is dependent on the association of AA with FLAP [24], we tested whether the organization of 5-LO into higher order assembles was also dependent on AA (S5 Fig). Cells were primed with IgE and then stimulated by the addition of antigen for 7 min in the presence or absence of the cPLA 2 inhibitor (Inh), or the presence or absence of the FLAP inhibitor MK886. When the release of AA was blocked (S5A- S5C Fig,  insets), there was no change in distributions for the number, area or density cluster properties of 5-LO localizations (S5A-S5C Fig, insets). As shown in weight-normalized histograms, the frequency of these clusters was unchanged (S5A- S5C Fig). When MK886 was added the properties of clusters changed. The mean number of localizations per cluster and cluster area both increased (S5A and S5B Fig insets), while the cluster density decreased (S5 Fig, inset). There was a higher frequency of clusters containing 100 or more localizations per cluster in cells pretreated with MK886 compared to primed and activated cells without inhibitor (S5 Fig). A similar pattern was seen in cluster area where the frequency of clusters with larger areas was higher in cells pre-treated with MK886 ( S5 Fig). Though the mean density per cluster decreased, the distribution was similar to that with cells treated with IgE and antigen alone and cells pretreated with MK886 ( S5 Fig). In the presence of the cPLA 2 Inh or the FLAP Inh following priming with IgE and antigen, the number of FLAP localizations per cluster decreased (S5 Fig, inset). The shifts in the frequency distribution of clusters for each inhibitor were similar, with fewer localizations in each cluster compared to cells only primed and activated (S5 Fig). The mean cluster area slightly decreased when cells were exposed to the cPLA 2 inhibitor during priming and antigen activation (S5 Fig, inset). MK886 had no effect on FLAP cluster area following cell priming and antigen activation (S5 Fig). Localization frequencies were left shifted when cells were exposed to either cPLA 2 inhibitor or FLAP inhibitor. This was a slight, but notable decrease in the frequency of clusters containing 300 localizations or more (Fig 4D). Cluster density was unchanged in the presence of cPLA 2 or FLAP inhibitors (S5 Fig).

Discussion
Analysis of two color dSTORM data using the Clus-DoC algorithm allowed us to define three classes of clusters: NIC, LIC, and HIC, and provided strong evidence linking HIC to the synthesis of LTs (Figs 1-4). The peak changes in the number of HIC and their properties occurred at 5 and 7 minutes, times of maximal LTC 4 generation. These changes include increases in area, the average number of 5-LO and FLAP localizations per cluster, and the percent of 5-LO and FLAP in HIC interacting with each other (Fig 4). As with 5-LO, the number of FLAP localizations in HIC increased at 5 and 7 min. Furthermore, almost all FLAP molecules that interact are in LIC or HIC. Interestingly, compared to LIC, the relative density of FLAP in HIC is lower. Using conventional STORM, the same trends were observed for the area and number of 5-LO localizations, whereas there was a modest decrease in average density (localizations per unit area) at 2 and 5 min post-activation, which may be attributed to the use of the whole nucleus as an ROI. The FLAP localization data obtained with conventional STORM identifies global changes in properties of clusters, but highlights the benefits of twocolor dSTORM paired with clustering algorithms that consider both molecules of interest in characterizing higher order assemblies. Despite the limitations of conventional STORM, a role of FLAP in regulating higher order assemblies of 5-LO is supported by experiments with MK886, which occupies the extended AA binding site, in which the number of 5-LO molecules and area of 5-LO in clusters is increased. In contrast, cPLA 2 Inh had no effect on organization, suggesting that overall re-organization of 5-LO or FLAP is independent of AA.
The number of localizations can be considered an estimate of the number of molecules in a cluster. Determining the absolute number of molecules from STORM data is difficult due to the stochastic blinking behavior of the fluorophores [32], but relative assessments should be less affected. In two-color dSTORM experiments the number of localizations per HIC was 20-30 for 5-LO and 40-60 for FLAP at 5 and 7 min. Because FLAP functions as a trimer [24], this data provides a rough estimate of a near-stoichiometric 5-LO:FLAP ratio of 1:1 to 2:1 in HIC. At 5 and 7 min post-activation, the peak of LT synthesis, a significant fraction of interacting localizations for both 5-LO and FLAP are localized in HIC. Together, these data strongly indicate a role for clustering in mediating LT synthesis.
The critical regulatory role for higher order assemblies, typified by signalosomes, is the new paradigm which must be considered when evaluating all membrane signaling processes [26][27][28][29][30]. These studies were based upon data obtained with crystallography and electron microscopy (EM) [21]. When combined with algorithms such as Clus-DoC, two-color dSTORM is a nanoscale SMLM approach that provides the ability to test for higher order organization while maintaining cellular organization. SMLM combined with cluster analysis has allowed the identification of the segregation of clusters of active and inactive integrin molecules on membranes [33]. iPALM has successfully elucidated the role of integrin configuration during adhesion [34]. Previous work by our laboratory using fluorescence lifetime imaging microscopy and biochemical crosslinking identified the interaction of 5-LO and FLAP [22,23,35]. This work was supported by subsequent studies, including those using overexpression approaches and/or non-physiological stimuli to probe the relationship between the two molecules [36][37][38]. None of these earlier approaches have addressed the critical concept of membrane organization at nanoscale and were limited by the technologies employed.
Overall, our work links higher order assemblies of 5-LO and FLAP to the initiation of LT synthesis on the nuclear envelope. Fig 7 shows our proposed model: At steady state (NT), small numbers of clusters (NIC; light gray outline) of 5-LO (green) or FLAP (red) exist on the nuclear membrane, few with 5-LO associated with FLAP (LIC; dark gray outline). By 2 min after activation, there is an increase in both total number of 5-LO and FLAP in NIC and LIC. Between 5 and 7 min post-activation, clusters with extensive interaction between 5-LO and FLAP (HIC; black outline) have formed. By 10 min, no HIC remain. Our work illustrates the power of using two-color dSTORM combined with algorithms such as Clus-DoC to analyze data, allowing us to unveil relationships not obtainable by conventional STORM and cluster analysis.
Characterizing the mechanisms of higher order assembly and disassembly of biosynthetic complexes is ultimately essential to understanding the initiation of LT synthesis and the generation of other eicosanoids and products of eicosapentaenoic acid and docosahexanoic acid.

LTC 4 synthesis assay
To analyze LTC 4 generation, an enzyme immunoassay kit for LTC 4 (Cayman Chemical) was used per manufacturer's instructions. Culture media was removed at desired times and stored at -80˚C until use. Assembly of leukotriene biosynthetic enzymes on nuclear envelope

Stochastic optical reconstruction microscopy
Preparation of secondary antibodies. To produce activator-reporter tandem dye pair antibodies for conventional STORM, Donkey anti-rabbit and donkey anti-goat affinity purified secondary antibodies (H+L chains) (Jackson Laboratory) were conjugated to Cy3 (Jackson ImmunoResearch) (activator) and Alexa Fluor 647 (AF647, Life Technologies) (reporter). For the conjugation reactions, 65 μg of the secondary antibody was reacted with 3 μg of the activator dye (Cy3) and 1.5 μg of the reporter dye (AF647) in 103 mM carbonic buffer for 2 hours at room temperature in the dark.
To produce reporter antibodies for direct STORM, donkey anti-goat and donkey antimouse affinity purified secondary antibodies (H+L chains) (Jackson Laboratory) were conjugated to ATTO 488 (Millapore Sigma) or AF647 dyes, both with carboxylic acid succinimidyl ester moieties. For the conjugation reactions, 240 μg of the secondary antibody was reacted with 6 μg of dye in 56 mM carbonic buffer for 2 hours at room temperature.
After the reactions, the antibodies were separated from unconjugated dye by gravity filtration through Sephadex G-25 DNA grade size exclusion columns (GE Healthcare) by visual detection. Antibody and dye concentrations were determined using a NanoDrop spectrometer (Thermo Fisher) to record absorbance at 280 nm and at the dye absorbance peak. Antibody concentration was calculated by subtracting the contribution of each dye to absorbance at 280 nm using correction factors provided by the dye manufacturers (ATTO488: 0.1, Cy3B: 0.09, AF647: 0.03) and the molar extinction coefficient of the antibody (210,000).
Two-color dSTORM. After activation, cells were fixed and prepared for superresolution microscopy [39]. The following day they were probed with antibodies to FLAP (Novus, IMG 3160, 1:100) and to 5-LO (BD Biosciences, 610695, RRID: AB_398018). Secondary antibodies were applied at 3 μg/mL for 1 h. The imaging buffer containing 100 mM 2-mercaptoethanolamine (MEA) and 1% (v/v) GLOX was used to promote photoswitching and reduce photobleaching [39]. An inverted Nikon Ti2 Eclipse STORM 5.0 system with Perfect Focus focal plane lock was used for image acquisition. This system contains a NSTORM quadband filter, and 405, 488, 561, and 647 nm lasers and was equipped with an HP APO TIRF AC 100x/1.49 NA oil objective and ORCA-Flash4.0 SCI CMOS PLUS camera (Hamamatsu Photonics). 15,000 frames for each dye were collected at 30 ms exposure time. Localizations were identified with NIS Elements 5.0 (Nikon Instruments) and exported as tabdelimited text files.

Clus-DoC analysis of single molecule localizations
To avoid prejudicial selection of membrane regions for analysis, we used localization data prior to cluster analysis to specify ROIs; thereby restricting our analysis to localizations that represent molecules on the nuclear envelope. EM studies demonstrated the localization of 5-LO and FLAP in the nuclear envelope after cell activation [21]. We employed Clus-DoC [31], which quantifies colocalization of individual proteins and molecules (localizations) and cluster properties. Clus-DoC allows for the user to define the number and DoC threshold.

Unbiased cluster analysis
We implemented a modified form of the variable bandwidth mean shift (VBMS) algorithm with automatic bandwidth selection previously described [40], using the diagonal bandwidth matrix estimation method. The final bandwidth used (250 nM) was then selected from these estimates automatically, and is allowed to vary from point to point, thereby adapting to variations in scale and structure between and within datasets.
Our modifications were primarily designed to improve analysis time. Running all test bandwidths in parallel rather than in serial decreased execution time over 10-fold. Storing the STORM data in kd-trees and calculating kernel updates using points within only 4x the bandwidth of the current mean provided faster mean shift direction estimates without sacrificing accuracy. The data was preprocessed with the DBSCAN algorithm [41], with the distance cutoff set based on the 99.9 th percentile of nearest-neighbor distances within the dataset. Regions which were determined to be independent during this conservative DBSCAN pre-clustering stage were processed independently in parallel. Because the VBMS algorithm scales nonlinearly with the number of points, this optimization further increased processing efficiency beyond that provided by parallelization. Single isolated points were also removed at this stage, providing another~10x increase in processing speed.

Statistics
EIA was analyzed using one-way ANOVA followed by Bonferroni multiple comparison posttest where p < 0.05 was considered significant. Unbiased cluster property histograms were first tested for normality using Kolmogorov-Smirnov and D'Agostino-Pearson omnibus tests. After passing the normality tests (p > 0.5), the data were tested for differences between samples by one-way ANOVA followed by Bonferroni multiple comparison post-test where p < 0.05 was considered significant. A Welch's unpaired t-test (unequal variances) was performed to determine to significance where two conditions were compared. All statistical tests were performed using Graphpad Prism 7 (Graphpad Software).