Septins organize endoplasmic reticulum-plasma membrane junctions for STIM1-ORAI1 calcium signalling

ORAI1 Ca2+ channels in the plasma membrane (PM) are gated by STIM1 at endoplasmic reticulum (ER)-PM junctions to effect store-dependent Ca2+ entry into cells, but little is known about how local STIM-ORAI signalling at junctions is coordinated with overall cellular architecture. Filamentous septins can specify cytoskeletal rearrangements and have been found recently to modulate STIM-ORAI signalling. Here we show by super-resolution imaging of ORAI1, STIM1, and septin 4 in living cells that septins facilitate Ca2+ signalling indirectly. Septin 4 does not colocalize preferentially with ORAI1 in resting or stimulated cells, assemble stably at ER-PM junctions, or specify a boundary that directs or confines ORAI1 to junctions. Rather, ORAI1 is recruited to junctions solely through interaction with STIM proteins, while septins regulate the number of ER-PM junctions and enhance STIM1-ORAI1 interactions within junctions. Thus septins communicate with STIM1 and ORAI1 through protein or lipid intermediaries, and are favorably positioned to coordinate Ca2+ signalling with rearrangements in cellular architecture.

ER-PM junctions are small-their typical diameter is less than 200 nm-and so conventional light microscopy lacks the resolution to investigate changes in ORAI1 recruitment to junctions in living cells. Here we have applied single-molecule localization microscopy and single-molecule tracking of ORAI1, septin 4, and STIM1, with the initial goal of investigating whether septins 4/5 specify a corral for ORAI1 at ER-PM junctions. Our results rule out this simple possibility, but delineate several features of PM organization downstream of septins 4/5 that reinforce physiological STIM-ORAI Ca 2+ signalling. These include the local density of ER-PM junctions, the internal specialization of junctions for STIM-ORAI signalling, and possibly the local modulation of ORAI1 concentration. The links that we demonstrate from septins to cellular Ca 2+ signalling could help to explain how cells coordinate their local Ca 2+ responses during physiological signalling with changes in cytoskeletal and PM organization.

Results
Single-molecule localization of ORAI1 channels in living cells. We used PATagRFP-ORAI1 and photoactivated localization microscopy (PALM) to examine the distribution and movements of ORAI1 molecules in HeLa cells, at rest and after ER Ca 2+ store depletion, with an experimentally determined localization uncertainty ~70 nm. The observations of store-depleted cells were made at least 5 min after addition of modified Ringer's solution containing 1 μM thapsigargin (TG) and 1 mM EGTA, when ORAI1 clusters had formed. STIM1-PAGFP in the TIRF layer served as a marker for near-plasma membrane ER in resting cells and for the more spatially restricted ER-plasma membrane contacts after store depletion.
ORAI1 was widely distributed across the footprint of resting cells [ Fig. 1a-d; Supplementary Fig. 1]. There was no enrichment of ORAI1 above random in the subset of pixels where STIM1 was localized. (To assess enrichment, we have normalized the observed colocalization to the overall density of ORAI1 localizations in the region analyzed [ Fig. 1c, legend]. Normalized colocalization values consistently greater than 1 indicate enrichment, whereas values consistently less than 1 indicate exclusion.) After TG treatment, ORAI1 was recruited to STIM1-marked puncta, with median enrichment above random in the subset of STIM1 pixels ~3, and the median distance from ORAI1 to STIM1 territories was reduced. The normalized colocalization values for stimulated cells were uniformly above those for resting cells. Note that ORAI1 enrichment over random is not the same as the ratio of ORAI1 density at junctions to ORAI1 density outside junctions. We have not attempted to make the latter estimate because, as explained in the next paragraph, ORAI1 scored as 'nonjunctional' is likely to include some junctional ORAI1.
The overlap of ORAI1 and STIM1 clusters after TG was imperfect [ Fig. 1b]. This result does not imply that ORAI1 clustering occurs without STIM1, because coexpressed STIM1-PATagRFP and STIM1-PAGFP examined in the same way showed similarly imprecise overlap [ Supplementary Fig. 2]. Both biological and technical considerations probably contributed to the imperfect overlap. On the biological side, ER-PM junctions are not static at 37 C, and small local redeployments of individual junctions between the acquisitions of STIM1 and ORAI1 images, which were sequential, would affect apparent overlap. On the technical side, ThunderSTORM localization software may fail to map all STIM1 or ORAI1 molecules in areas of very high density, a shortcoming it shares with other localization algorithms; and, conversely, limited sampling may lead to false negative pixels at loci where STIM1 or ORAI1 are sparsely distributed.
We examined the effect of TG on ORAI1 localization in siSeptin4/5-treated cells. ORAI1 was still measurably recruited to junctions marked by STIM1 in these cells, but recruitment was less pronounced than in control cells, with median enrichment above random ~2, and an apparent small decrease in the median ORAI1-STIM1 distance was not statistically significant [ Fig. 1e-h]. The findings replicated lower-resolution observations by conventional TIRF microscopy 10 [1][2][3][4] for quantitative analysis of ORAI1 movements. One frequently used method of analysis is the mean squared displacement (MSD) plot [ Fig. 1o]. MSD plots of ORAI1 channel movements mirrored the observations from ORAI1 localizations in Fig. 1a-h. ORAI1 trajectories in control cells were, on average, more confined after TG, as indicated by the early flattening of the MSD curve. TG-induced confinement was less pronounced in siSept-in4/5-treated cells, but reconstitution with RNAi-resistant septin 4 restored it, firmly linking the confinement to a direct or indirect effect of septins.
MSD analysis can be particularly informative regarding diffusive movements. Since the ORAI1 MSD plots exhibited curvature within the first 200 ms, indicative of spatial restrictions on ORAI1 diffusion on that time scale, we could not take the usual approach of estimating an average diffusion coefficient D for each condition as a linear fit to r 2 values over several steps. Instead, we applied two alternative methods to estimate average D: from single-step r 2 values, and from the slope of the MSD plot between 50 ms and 100 ms [ Table 1]. Corrections for localization uncertainty and for blurring due to particle movement within each frame have to be introduced in the single-step calculation [29][30][31][32]. No corrections are required in the 50ms-100 ms slope calculation, since localization error and blurring make the same average contributions to the 50 ms and 100 ms r 2 values, and therefore the corrections cancel upon subtraction. The two independent estimates of the ORAI1 diffusion coefficient agreed for all conditions observed [ Table 1].
The quantitative analysis bore out the qualitative conclusions drawn above. The average D value in resting HeLa cells, ~0.11 μm 2 s −1 [ Table 1], was similar to the value measured in HEK293 cells 33  www.nature.com/scientificreports www.nature.com/scientificreports/ an approximate 50% reduction in average D. ORAI1 mobility in resting siSeptin 4/5-treated cells resembled ORAI1 mobility in resting control cells, with average diffusion coefficient ~0.10 μm 2 s −1 , but diffusion was on average much less affected by calcium store depletion. TG-triggered slowing of ORAI1 was restored by reconstitution of siSeptin4/5-treated cells with RNAi-resistant septin 4.
The effects of siSeptin4/5 and of septin 4 reconstitution were equally evident in histograms depicting the 50-ms, 100-ms, or 150-ms step distances observed in the raw data [ Fig. 1p for the 50-ms interval; Supplementary  Fig. 3 for all three time intervals]. This alternative view offers richer detail than the averaged diffusion coefficients or the averaged displacements of an MSD plot. Collectively, the data of Fig. 1 establish that the slowed diffusion of ORAI1 channels and their confinement to ER-PM junctions after store depletion are compromised by a reduction of septins 4/5 in the cell. Septin 4/5 depletion has no effect on STIM1 trajectories in STIM-ORAI coexpressing cells. We followed STIM1-PATagRFP trajectories similarly, in cells coexpressing ORAI1 to maintain consistency in the experimental conditions [ Fig. 2]. The trajectories in resting cells had a pattern consistent with STIM1 diffusion within ER tubules located in the TIRF layer. An average D value calculated from single-step r 2 values, 0.16 μm 2 s −1 [ Table 2], was in line with an estimate from single-molecule measurement in HEK293 cells and with earlier estimates from FRAP measurements [33][34][35] . The flattening of the MSD plot at longer times probably reflects movement of the more mobile STIM1 molecules out of the TIRF layer 33 . STIM1 diffusion was slowed and STIM1 trajectories became more confined after store depletion. Although the STIM1 MSD plot for TG-treated cells did not reach a clear plateau within 500 ms, its shape suggested an equivalent confinement radius less than 250 nm [ Supplementary Fig. 4], in line with the expected dimensions of ER-PM junctions in HeLa cells. In contrast to the marked effect of septin 4/5 depletion on ORAI1 movements, septin 4/5 depletion did not alter STIM1 diffusion appreciably either in resting cells or after TG treatment. This observation assumed importance in the later analysis of ORAI1 retention at junctions. We mapped the plasma membrane distribution of PAGFP-septin 4 relative to ER-PM junctions, using PATagRFP-MAPPER, an engineered marker of close ER-PM contacts 38 , and STIM1-PATagRFP as alternative markers for ER-PM junctions. Septin 4 territories were largely distinct from junctions both in resting cells and after TG treatment [ Fig. 3]. This separation was visible qualitatively in both conditions for MAPPER, and quantitatively the median normalized colocalization of MAPPER with septin 4 was appreciably less than 1 [ The median normalized colocalization of STIM1 with septin 4 decreased from ~1 in resting cells, indicating a random distribution of STIM1 relative to septin 4, to ~0.5 following store depletion. The median distance from STIM1 or MAPPER to the nearest septin 4 territory increased after store depletion [ Fig. 3d,h], probably reflecting the previously observed reorganization of septins around junctions 10 in addition to the relocalization of STIM1. The separation of septin 4 from junctions was clear in views across entire cell footprints [ Supplementary Fig. 6]. Thus, strikingly, a substantial patchwork of structures incorporating septin 4 and a large number of ER-PM junctions are interleaved at the plasma membrane, but the two occupy disjoint areas of the membrane.

Septins 4/5 do not direct ORAI1 to ER-PM junctions.
The sharp separation between septin 4 territories and ER-PM junctions raised the possibility that exclusion of ORAI1 from septin territories could assist in directing ORAI1 to junctions. As a first approach to this question, we examined the localization of ORAI1 relative to PAGFP-septin 4 [ Fig. 4a-d]. ORAI was indifferent to septin 4 territories in resting cells (median normalized colocalization ~1), ruling out constitutive ORAI1 exclusion from the territories, but redistributed away from septin 4 after calcium store depletion (median normalized colocalization ~0.8). Of course, given the results of the previous section, ORAI1 might be expected to move away from septin 4 simply because a fraction of ORAI1 becomes confined at junctions after store depletion. Therefore, we constructed an ORAI1 channel, ORAI1-LSLD, that does not interact with STIM1 by incorporating the previously reported mutations L273S and L276D, either of which alone abrogates the STIM1-ORAI1 interaction 39,40 . In measurements similar to those made for wildtype ORAI1, ORAI1-LSLD did not relocalize away from septin 4 following store depletion [ Fig. 4e www.nature.com/scientificreports www.nature.com/scientificreports/ that wildtype ORAI1 segregates from septin 4 only due to its STIM1-dependent partial confinement at junctions after store depletion.
Further consideration of our ORAI1-LSLD localization data and data from the single ORAI1(L273S) and ORAI1(L276D) mutants 39,40 supports an additional strong conclusion. Provided the LSLD mutations affect only the interaction with STIM proteins, the fact that ORAI1-LSLD is not preferentially localized at junctions after store depletion [ Fig. 4i-l; Supplementary Movies 11,12] rules out any recruitment of ORAI1 channels to junctions by mechanisms other than their interaction with STIM proteins. The argument applies to any hypothesized STIM-independent mechanism, whether it involves septins or not.

Septins 4/5 enhance ORAI1 retention at junctions.
In a different mechanism compatible with the evidence that STIM-ORAI interaction is required for ORAI1 recruitment to junctions, septins 4/5 might enhance the retention of ORAI1 channels at junctions after their STIM-dependent clustering. We used modified inverse FRAP to look directly at the retention of GFP-ORAI1 at individual ER-plasma membrane junctions of stimulated cells. Following cluster assembly, the region surrounding an individual ORAI1 cluster was continuously illuminated at high laser intensity, except during image acquisition, so that any ORAI1 leaving the cluster would be bleached and any ORAI1 newly added to the cluster would be nonfluorescent [ Fig. 5a]. Under these conditions, most of the junctional ORAI1 signal disappeared over a period of tens of seconds in both control and septin 4/5-depleted cells [ Fig. 5b,c]. However, the fluorescence loss was more rapid in septin-depleted cells, and left a lower fraction of relatively stable signal at times from 50-80 s after the onset of bleaching laser illumination. The residual signal at 10 s and 80 s in control cells was 50% and 16%, respectively, whereas the residual signal at 10 s and 80 s in siSeptin4/5 cells was 30% and 9%. In control experiments, ORAI1 loss from regions of comparable size in resting cells was very rapid and was unaltered by septin depletion [Fig. 5c].
Part of the explanation of the inverse FRAP results is that a smaller fraction of junctional ORAI1 is engaged with STIM1 when septins 4/5 have been depleted. This is evident in higher ORAI1 mobility in septin 4/5-depleted cells than in control cells, in trajectories starting within STIM1-marked junctions [ Fig. 5d]. The average ORAI1 diffusion coefficient at junctions after TG, calculated as in Table 1 from single-step r 2 values [see Methods], was 0.080 μm 2 s −1 in siSeptin4/5-treated cells, 0.047 μm 2 s −1 in siControl cells. Since STIM1 visible in the TIRF layer after TG is essentially confined to junctions, STIM1 mobility under these conditions is adequately measured from overall MSD plots, where single-step D estimates were in the range 0.038-0.041 μm 2 s −1 [ Table 2]. Thus siSept-in4/5-treated cells show an especially marked disparity between ORAI1 and STIM1 diffusion. With the simplifying assumption that ORAI1 has only two diffusional states, as free ORAI1 or as STIM1-bound ORAI1, and taking the diffusion parameters for ORAI1-LSLD from single-step measurements at junctions [ Fig. 5e] as representative of the diffusion of free wildtype ORAI1, we estimated the respective fractions of ORAI1 bound as 84% and 56% [see Methods]. This analysis shows that septins 4/5 enhance the retention of ORAI1 at individual ER-PM junctions of store-depleted cells, and do so mainly by promoting the effective engagement of ORAI1 with STIM1.  Table 3; Supplementary Movies 17,18]. We conclude that septin-dependent barriers are in place across the cell footprint after TG treatment of control cells. The local barriers are not restricted to junctions, but they could nevertheless reduce the escape of free ORAI1 from junctions and favor rebinding of ORAI1 to STIM proteins under conditions where there is net exchange of ORAI1 out of the junctions.

TG/EGTA treatment increases PM barriers to ORAI1 diffusion.
In parallel, we examined the excursion distances of wildtype ORAI1 trajectories at junctions and outside junctions after TG [Fig. 6d,e]. Although the experiment was intended only to check that we could see a difference in the distances with wildtype ORAI1, and was not explicitly designed for comparisons between wildtype ORAI1 and ORAI1-LSLD, it does offer a further perspective on ORAI1 confinement at junctions. The comparison of WT siSeptin to LSLD siSeptin, on junction, is a measure of purely STIM-dependent confinement. And the progressive decrease in excursion values d max for LSLD siSeptin > LSLD siControl > WT siControl, on junction, indicates that both septins and STIM proteins contribute tangibly to confinement. www.nature.com/scientificreports www.nature.com/scientificreports/ The constraints on ORAI1 excursion at junctions in Fig. 6d-comparing control cells on-and off-junctionmay appear unexpectedly modest. This can be attributed primarily to local septin-dependent constraints on ORAI1 movement outside junctions, which are also evident with ORAI1-LSLD in Fig. 6b. However, another contributing factor could be that a fraction of junctional trajectories were incorrectly identified as nonjunctional [see above in the section 'Single-molecule localization of ORAI1 channels in living cells'].

STIM1 junctional area is reduced in siSeptin 4/5-treated cells.
Finally, to assess whether septins 4/5 affect maintenance of the junctions themselves, we quantitated ER-PM junction numbers and areas in living cells under control and siSeptin4/5 conditions using STIM1 in store-depleted cells as a junctional marker. Recently published methods based on Voronoi tessellation have the advantage that they can analyze clustering directly from PALM localization data, thus avoiding the loss of resolution that comes with an intermediate step of converting localization data to an inferred image 41,42 . We conducted tessellation analysis of STIM1 localization data in living cells after TG using ClusterViSu 42 [Fig. 7a,b]. The number of STIM1 molecules localized per cluster was unchanged in siSeptin4/5-treated cells compared to control cells, and cluster dimensions-expressed as equivalent diameter-were only modestly altered. However, the density of STIM1-marked clusters per μm 2 in the image was significantly reduced by siSeptin4/5 treatment, and the median fraction of the cell footprint occupied by clusters was reduced by 23%, from 0.031 to 0.024.
The decrease in fractional junctional area was further confirmed by a separate ImageJ particle analysis, which estimated a reduction of 31%, from 0.059 to 0.041 [ Fig. 7c,d]. ImageJ identified the same clusters as ClusterViSu in a given data set, but, perhaps related to the loss of resolution inherent in the processing steps, ImageJ tended to include a somewhat larger area within an individual assigned cluster, and also showed less tendency to subdivide contiguous areas into multiple clusters. It is unclear whether the finer subdivision performed by ClusterViSu  www.nature.com/scientificreports www.nature.com/scientificreports/ corresponds to real substructure in the biological data sets. When we simulated data by placing localizations randomly within idealized 200-nm diameter or 400 nm-diameter circular junctions, ClusterViSu as implemented performed well on the 200-nm junctions but found apparent substructure in the 400-nm junctions where none was built into the simulated data [ Supplementary Fig. 9]. The issue is tangential to our use of ClusterViSu, though, since both methods of clustering analysis supported the conclusion that there is a decrease in the number of clusters and the area occupied by STIM1 per μm 2 in the cell footprint. Therefore, in addition to enhancing the retention of ORAI1 after it has arrived at an ER-PM junction, septins 4/5 also support the maintenance of a normal complement of ER-PM junctions where STIM1 and ORAI1 can interact.

Discussion
Our findings highlight facets of PM organization that underlie cellular STIM-ORAI Ca 2+ signalling, and identify control points that could coordinate regional changes in Ca 2+ signalling in cells [ Fig. 8]. The most prominent finding is that normal HeLa cells have a larger number of STIM1-competent junctional sites than septin4/5-depleted cells. Septins are known to organize and stabilize subcellular structures, and typically are visibly associated with those sites, for example with the mitotic spindle and the metaphase plate 43,44 and with actin stress fibers 45 . In contrast, septin 4-taken as a marker for the class of septin filaments relevant to STIM-ORAI signalling-is not stably present at ER-PM junctions. The latter observation suggests that septins act through intermediaries to render the PM receptive to junction formation, to position ER tubules for junction formation, or to slow junction disassembly. The most plausible intermediaries would be cytoskeletal proteins 28 , given that cortical actin supports junction formation or stability 46 and microtubules guide the continual reshaping of the ER [47][48][49] . The RNAi experiments here allowed us to infer global septin4/5-dependent processes that contribute to basal maintenance of junctions in cultured cells, but in vivo the same processes would be suited to coordinate more localized changes in STIM-ORAI signalling during the dynamic remodelling of ER-PM junctions that accompanies Ca 2+ signalling, cell movement, and developmental changes in cellular architecture [50][51][52] .
A second intriguing finding is that STIM1 engages ORAI1 more effectively at ER-PM junctions in normal HeLa cells than in septin4/5-depleted cells. Individual junctions in control and septin4/5-depleted cells are comparable in size and recruit equivalent amounts of STIM1 after store depletion, and we showed previously that STIM1-ORAI1 gating-reflected in NFAT nuclear import triggered by the soluble STIM1 cytoplasmic domainis unimpaired by septin4/5 depletion 10   www.nature.com/scientificreports www.nature.com/scientificreports/ evidence of substructure within ER-PM junctions, with distinct junctional nanodomains dedicated to the assembly and the disassembly of STIM-ORAI complexes 53 . How and when septins speak to the internal structure of junctions is an important area for future investigation. One possibility is that a message is delivered when the  www.nature.com/scientificreports www.nature.com/scientificreports/ junctions form, and that the junctions in normal HeLa cells are already poised for constructive STIM-ORAI engagement in their basal state. An alternative is that septin filaments or their associated proteins prime the junction for STIM-ORAI engagement when septins rearrange near junctions early in the response to ER Ca 2+ store depletion 10 . The involvement of septins 4/5 once again raises the possibility that alterations in junctional signalling efficiency could be coordinated regionally and even very locally during physiological processes in vivo.
Finally, the generalized slowing of ORAI1 diffusion after global store depletion raises the possibility that Ca 2+ signalling might be modulated locally or regionally by controlling ORAI1 diffusion itself. During physiological signalling that is restricted to only one part of the cell surface, ORAI1 diffusion might well be slowed within a zone of influence of a single store-depleted ER-PM junction or within a limited region in which ER stores have been depleted. Under these conditions ORAI1 diffusion into the zone or region would be greater than its slower diffusion out of the region. Assuming a localized change in D similar to what we have observed globally with TG, the ORAI1 concentration in the region would rise with time to a level 25-30% over the concentration outside the region, favoring increased STIM-ORAI interaction at junctions in the region if ORAI1 is otherwise limiting. From the diffusion equation, ORAI1 might be expected to accumulate from surrounding regions over a distance of one to a few micrometers on a time scale of a few seconds up to tens of seconds, a spatial and temporal range suited to modulating many localized cellular processes. For example, the lamellipodia of migrating endothelial cells exhibit small fluctuating Ca 2+ signals on these distance and time scales and exhibit polarization of other Ca 2+ signalling proteins toward the leading edge 54,55 . This polarization includes locally increased levels of STIM1 in the ER 55 , offering a precedent for a mechanism that modulates signalling by increasing local ORAI1 concentration. www.nature.com/scientificreports www.nature.com/scientificreports/ Our single-molecule localization and tracking data establish that septins 4/5 at native levels facilitate STIM-ORAI Ca 2+ signalling in several specific ways that do not require the continuing presence of septin structures at ER-PM junctions. Thus STIM-ORAI signalling is responsive to local or regional features imprinted by the septins. The shared dependence of cellular architecture, PM organization, and STIM-ORAI signalling on septins is likely to facilitate regional coordination of Ca 2+ signalling with other signalling pathways and cellular processes.
siRNA oligonucleotides were obtained from the Dharmacon Human siGenome collection (v2007 or 2010), with the exception of siControl, which was custom synthesized by Dharmacon. Sequences are listed in Supplementary Table 1      Cells were seeded on poly-d-lysine treated Nunc ™ Lab-Tek ™ II Chambered Coverglass (Thermofisher Scientific) the following day. Briefly, Chambered Coverglass was coated with 100 μl per well of 0.1 mg/ml poly-d-lysine (Sigma-Aldrich) at 37 °C for 10 min followed by two washes with cold PBS. Experiments were performed 24 hours after cell seeding. Thapsigargin (TG) was purchased from Calbiochem. All other chemicals were purchased from Sigma.
For siRNA transfections, HeLa cells were reverse transfected with 25-50 nM siRNA using Mirus TransIT-X2 (Mirus Bio), with co-transfected fluorescence expression plasmid, according to the manufacturer's instructions. siRNA-transfected cells were placed on coverglass after 48 hours. Cells were treated with siSeptin4/5 as indicated in the text. Cells were treated with the siRNA targeting native ORAI1 in all experiments where ORAI1-LSLD was expressed, in order to avoid assembly of heteromultimeric ORAI1-LSLD/ORAI1 channels containing wildtype ORAI1 subunits.  Lower, Expanded view of a single junction after store depletion. ORAI1 has been recruited effectively and is engaged with STIM1. (b) siSeptin4/5-treated cells expressing STIM1 and wildtype ORAI1. The density of STIM1-competent ER-PM contacts is reduced. STIM1 still comes to the junctions, but ORAI1 is recruited less efficiently, and a lower fraction of junctional ORAI1 is engaged with STIM1. (c) Control cells expressing STIM1 and ORAI1-LSLD. STIM1 relocalizes to the junctions after store depletion, but ORAI1-LSLD is not recruited. This finding is the clearest evidence that septins and their associated proteins do not direct or confine ORAI1 to junctions, and implies the stronger conclusion that ORAI1 channels are recruited to ER-PM junctions solely by their interactions with STIM proteins. (2019) 9:10839 | https://doi.org/10.1038/s41598-019-46862-w www.nature.com/scientificreports www.nature.com/scientificreports/ Single particle tracking and photoactivated localization microscopy (sptPALM and PALM) were performed with a previously described microscope configuration 56 . Briefly, a Nikon Ti-E inverted microscope platform with a 100 × objective (CFI Apo TIRF oil immersion NA 1.49, Nikon) imaged HeLa cells held at 37 °C in a Tokai Hit Stage Top Incubator. Images were recorded with an Andor iXon 860E EMCCD camera and kept in focus with the Nikon Perfect Focus System. The excitation path 488 nm and 568 nm lasers (Genesis MX, Coherent) were air coupled with the 405 nm activation laser (CUBE, Coherent) and manually tuned for Total Internal Reflection Fluorescence (TIRF) excitation. PALM and sptPALM experiments were recorded with MetaMorph imaging software (Molecular Devices) at 20 Hz in streaming acquisition mode, in sequential acquisition periods of 25 s (500 frames) for the green/PAGFP channel, 25 s for the red/PATagRFP channel, and again 25 s for the green/PAGFP channel.
Image analysis. Localizations. PALM localizations were generated with the ImageJ plugin ThunderSTORM. Images were filtered with a B-spline wavelet filter (order 3, scale 1.3-1.5), approximate molecular localizations assigned by a local maximum estimate with 8-connected neighborhoods, and sub-pixel localizations determined by weighted least-squares fitting to an integrated Gaussian PSF model over a 5 pixel radius with initial sigma value 1.3 pixels. Individual red-channel and green-channel PALM data were drift corrected to 100-nm TetraSpeck ™ Microspheres (ThermoFisher Scientific), and the red and green channels were brought into register using the same fiducial markers. PALM localizations were visualized by the ' Averaged Shifted Histogram' command with lateral shift value 2 and magnification chosen to give a pixel size of 10 nm. The output representation was a 32-bit image with pixel size 10 nm.
Normalized colocalization. The colocalization and distance analyses were performed in MATLAB (MathWorks, Natick, MA). For colocalization analysis, output images from the PALM analysis were converted to binary images, and Normalized Colocalization in Green (NCG) was calculated as = . NCG (colocalized pixels/green pixels)/(red pixels/total pixels) Territories and median distances. PAGFP-septin 4 (or STIM1-PAGFP) PALM data from the two 25-s acquisitions in the green channel were merged, and filtered to remove localizations that did not fall within 100 nm of at least five other septin 4 (or STIM1) localizations. The binary green-channel image of the merged data defined the septin 4 (or STIM1) territory within the ROI, and was used to calculate the initial fractional coverage of red pixels in the corresponding binary red-channel image as = .
Fractional coverage of red pixels (colocalized pixels/total red pixels) The green territory was then dilated stepwise by three pixels at a time [ Supplementary Fig. 10], recalculating the fractional coverage of red pixels at each step, until the fraction was equal to or greater than 50%. The total dilation at this point, in pixels, was converted to a median distance in nanometers for each region of interest (ROI) analyzed.
Trajectories. Single-molecule trajectories were analyzed with DiaTrack 3.05 software 57 . Image stacks were background-subtracted with the built-in 'Subtract Background' command and the data smoothed with a Gaussian filter (1.05 pixels). Subpixel localizations were determined by fitting to a Gaussian PSF model of half-width-half-max value 1.2 pixels, as described previously 56 . Particle selection ensured that molecules were sufficiently contrasted from their surroundings by using a value of 1.05 to remove blurred objects, and by applying user-determined intensity thresholds to remove objects that were too dim or too bright to be considered single fluorophores. After frame-by-frame single-molecule subpixel localization, trajectories were determined allowing a maximum displacement of three pixels (480 nm) between 50-ms frames and requiring a minimum trajectory lifetime of 3 frames, and drift corrected to 100-nm TetraSpeck ™ Microspheres (ThermoFisher Scientific).
Resulting track files were exported and processed in Igor Pro 6.2 (WaveMetrics).
MSD plots. The mean squared displacement (MSD) curve for each experimental condition was determined by averaging the squared displacement over all occurrences of the corresponding time intervals in all trajectories under that condition. The inclusion of overlapping steps in the calculation has been faulted, since it results in a correlated data set 32 , but this method is nonetheless commonly used to derive the MSD plot. (Note that the issue of overlapping steps does not arise in the single-step D estimates described below.) Error bars at each point represented the standard error of the mean. Corresponding ORAI1 displacement distributions were plotted for the 50, 100, and 150 ms time points for each experimental condition.
Diffusion coefficients. Average diffusion coefficients D for each protein and experimental condition were estimated in two ways. Estimate D 1 was calculated from the mean squared displacement for a 50-ms interval (MSD 1 ), corrected as required for the localization uncertainty estimated from STIM1 and ORAI1 fixed-sample controls 30 and for the apparent contraction introduced by continuous data acquisition 29,31,32 . The basic equation relating D to expected mean squared displacement <r 2 > and time τ is The correction for localization uncertainty 30  We estimated the contribution of localization error from STIM1 and ORAI1 'trajectories' in fixed cells. Cells were fixed in 4% paraformaldehyde in phosphate-buffered saline for ten minutes at room temperature, washed, and imaged under the same experimental conditions as for living cells. The observed MSD values for a 50-ms interval in fixed cells (MSD 1,fixed ) were 0.0086 ± 0.0002 μm 2 for STIM1 (from 15,350 trajectories) and 0.0085 ± 0.0002 μm 2 for ORAI1 (from 11,436 trajectories). Thus 2σ localization 2 ≈ 0.0085 μm 2 , corresponding to a two-dimensional localization error σ = 65 nm, or σ x = σ y = 46 nm.
The second correction is necessary because the tracked molecules are moving during continuous data acquisition. Fluorescence originating from intermediate positions along a step that starts at point X and arrives at point Y tends to make the step look shorter than the actual distance from X to Y. The correction factor when fluorescence data are collected continuously during a single step is multiplication by 1.5 [refs 29,31,32 ]. Hence, incorporating both corrections, and for a 50-ms step, Inverse fluorescence recovery after photobleaching (iFRAP). Modified iFRAP experiments were performed on HeLa cells 24-48 h after transfection with 1-2 μg of a plasmid encoding GFP-ORAI1 and an equal amount of a plasmid encoding unlabelled STIM1. Cells were imaged on a Zeiss LSM 800 confocal system with a 63x oil plan-apo 1.4 NA objective and a 488-nm laser. A selected region of each cell was imaged before TG treatment to obtain iFRAP data for nonjunctional ORAI1, and a selected ORAI1 cluster in a different region of the same cell was analyzed after TG treatment to obtain iFRAP data for junctional ORAI1. An area bracketing the selected ROI was bleached with laser illumination at high power, bringing the fluorescence signal close to background in the illuminated area, and bleaching laser illumination was continued during the subsequent 90 s, except during image collection at 1 s intervals. This procedure eliminated any signal increase in the clusters from newly recruited GFP-ORAI1, and allowed us to measure fluorescence loss due to GFP-ORAI1 exit from the ROI. A photobleaching control was run in parallel in each case, but no correction for photobleaching was required. Fluorescence data were recorded with Zeiss software, and statistical analysis was performed in Excel. Mean values (normalized and background corrected) were plotted from 3 independent experiments for each condition.
Analyzing trajectories relative to ER-PM junctions. Trajectories at and away from junctions. For these analyses, both PAGFP and PATagRFP were localized and the localizations drift-corrected using DiaTrack 3.05 software. Localizations and trajectories were exported to MATLAB (MathWorks, Natick, MA), with each data set represented as a 2048 × 2048 array of 10-nm pixels, and the PAGFP array was aligned to the PATagRFP array via the fiducial TetraSpeck ™ Microspheres. STIM1-marked ER-PM junctions were defined as regions where STIM1-PAGFP localized and clustered after TG treatment. STIM1 localizations that did not fall within 100 nm of at least five other STIM1 molecules were discarded, similarly to the density filtering performed in ThunderSTORM, described above. Gaps in ER-PM junction clusters were filled in with a 5-pixel (50-nm) dilation of each localization retained after density filtering. PATagRFP-ORAI1 tracks were then classed as "on junction" if the first localization in the track occurred in a STIM1-marked junction, and "off junction" if all localizations in the track were outside STIM1-marked junctions.
Excursion distances. Only tracks consisting of 6-25 steps were used for the analysis of excursion distances. The maximum excursion d max was defined as the maximum displacement from the first localization to the furthest localization within the same track [ Fig. 6a]. A total of 1500 tracks from each treatment group were randomly chosen and the d max distributions were compared between conditions.

Diffusion coefficients at junctions and STIM-ORAI engagement.
For each on-junction ORAI1 trajectory, the squared displacement (r 1 2 ) of the first 50-ms step was calculated based on the coordinates of the first and second localizations, ], except that MSD 1 was replaced by the averaged single-step r 2 value. The fraction of junctional ORAI1 engaged with STIM1 was estimated based on the simplifying assumption that ORAI1 has only two diffusional states, as free ORAI1 (D given by the value for ORAI1-LSLD in siControl-or siSeptin4/5-treated cells after TG, as appropriate) or as STIM1-bound ORAI1 (D given by the measured STIM1 D value). Then www.nature.com/scientificreports www.nature.com/scientificreports/ The single-step D values for ORAI1-LSLD on-junction were 0.0840 μm 2 s −1 (siControl/TG) and 0.1298 μm 2 s −1 (siSeptin/TG). Since STIM1 diffusion was not affected by siSeptin4/5 treatment, the average of STIM1 D values for siControl/TG and siSeptin/TG conditions, 0.0394 μm 2 s −1 , was used. STIM1 clustering analysis. Clustering of PALM localizations were analyzed with ClusterVisu Voronoi tessellation segmentation software 42 . The localization file from ThunderSTORM was imported, and at least three smaller ROIs (~16-20 µm 2 ) were chosen per cell to increase the analysis speed. ClusterViSu uses Monte Carlo simulations-here set at 100 iterations-to determine the expected Voronoi polygon area distribution for a random data set consisting of the same number of localizations in the same ROI, in order to define a threshold value of cell area below which individual Voronoi cells are accepted as part of a nonrandom cluster. A Voronoi density map with 10-nm pixel size was constructed on this basis. Watershed segmentation was applied to this map, and clusters less than 10 pixels in size were excluded. Statistics on cluster dimensions, localizations per cluster, and cluster density per unit area were collected and analyzed in Excel.
As an independent check, clusters were analyzed using ImageJ ' Analyze Particles' to determine cluster sizes, localizations per cluster, and cluster density per unit area. The same ROIs were used and localization images reconstructed similarly with a 10-nm pixel size.
Cluster simulation. Randomized clusters were generated using a custom program written in MATLAB (MathWorks, Natick, MA). First, 17 cluster centers were placed randomly in a 4 μm x 4 μm region of interest. Then either 32 or 100 localizations, each with 70-nm localization uncertainty, were placed randomly in a 200-nm or 400-nm circular disc surrounding each cluster center, and background noise localizations amounting to 0.7% of pixels in the ROI was added randomly. The numbers of localizations per cluster, localization uncertainty, and level of background localizations were chosen to be representative of the actual data in our experiments. The generated clusters were then imported to ClusterVisu for analysis as described above.
statistics. Statistical analysis was carried out in MATLAB. Comparisons in Figs 1, 3, 4, 5 and 6 were by t-test, with equal or unequal variance as determined by an F-test. The box plot whiskers represent 1 SD from the mean. Comparisons in Fig. 7 were by Student's t-test, carried out in Excel, and the box plots were drawn in Igor Pro 6.2. Box-plot whiskers represent the 10%-90% range of observed values. In the MSD plots and the intensity decay curves of the iFRAP experiments, error bars indicate SEM.

Data Availability
Data are available from the corresponding author upon request.

Code Availability
Custom MATLAB codes are available from Dr Chen Zhang or the corresponding author upon request.