In situ structure of actin remodeling during glucose-stimulated insulin secretion using cryo-electron tomography

Actin mediates insulin secretion from the pancreatic β-cell through a remodeling process. Previous studies have been hampered by limited resolution, providing an ambiguous depiction of actin remodeling as a process that begins with depolymerization into actin monomers, followed by repolymerization into actin filaments. Here, we report the in situ structure of actin remodeling in INS-1E β-cells during glucose-stimulated insulin secretion at nanoscale resolution. We demonstrate that actin remodeling occurs at the cell periphery rather than in the cell interior. The actin filament network at the cell periphery exhibits three marked differences after remodeling compared to those under basal conditions. First, approximately 12%of actin filaments reorient, their angle changing from 0–45° to 45–90° relative to the plasma membrane. Second, the actin filament network remains predominantly as cell-stabilizing bundles but partially reconfigures into a less compact arrangement. Third, actin filaments anchored to the plasma membrane reorganize from a “netlike” to a “blooming” architecture, featuring radial projections emanating from their anchor points. Remodeling precedes the transport of insulin secretory granulesto the plasma membrane and their release from it. Furthermore, the density of actin filaments and microtubules around insulin secretory granules is lowered after remodeling compared to the basal conditions, as expected for the subsequent granule transport and release. Finally, actin filaments and microtubules are more densely packed than under basal conditions. These findings advance our structural and functional understanding of actin remodeling during glucose-stimulated insulin secretion in pancreatic β-cells.


Introduction
Actin laments are essential for a variety of cellular functions, including maintaining the cell structure, driving cell migration and division, and governing the movement of subcellular components, such as mitochondria, lysosomes, and insulin secretory granules (ISGs) 1,2 . Cellular signaling pathways, such as Rho GTPase pathway and PI3K/Akt pathway, cause the remodeling of actin laments 3,4 . An early 2D visualization of actin remodeling, starting with a parallel alignment and ending in an "unidenti able dense mass", was achieved in vitro for chemically xed β-cells using double-contrasting electron microscopy in 1972 5 . Since then, actin remodeling has been studied at an even lower resolution by uorescence imaging with arti cial uorescence labeling and by western blot analysis, revealing that actin laments rst depolymerize into actin monomers (G-actin) and then repolymerize into actin to the plasma membrane (PM) and the release of their contents into the extracellular space 8- 10,14 . Under basal conditions, actin laments act as a barrier parallel to the PM to block ISGs from approaching the PM 8, 10 . In the rst phase, actin laments depolymerize into actin monomers, allowing the rapid release of ISGs in RRP 8, 9 . In the second phase, actin laments repolymerize into a new actin network that facilitates ISGs to transport to and release from the PM 8, 9,13 . Nevertheless, quantifying the differences between the actin lament network before and after remodeling is imperative to understand the mechanisms underlying its blocking or facilitating role during GSIS.
During GSIS, the transportation of ISGs rst involves long-range movement driven by kinesins on microtubules (MTs), with the subsequent handoff to myosin V for the short-range movement on actin laments to the PM 15,16 . Using uorescence microscopy, MTs have been reported to negatively regulate insulin secretion at the cell periphery under basal conditions 17 ; it was proposed that direct interaction between actin laments and MTs promotes the transport and release of ISGs in the second phase 10 . Due to the limited resolution of uorescence microscopy, the structural basis of interactions among ISGs, actin laments, and MTs during GSIS remains to be determined.
In this work, we used a multimodal imaging approach to visualize and quantify actin remodeling at both the periphery and interior of the cell under basal conditions as well as during the rst and second phases of insulin secretion in rat INS-1E β-cells. First, we quanti ed changes in actin meshwork organization during GSIS using total internal re ection uorescence (TIRF) microscopy and structured illumination microscopy (SIM). Next, we employed cryo-electron tomography (cryo-ET) to reveal the structure of actin remodeling at the nanoscale and variations in the volume and orientation of actin laments during GSIS 18-20 . In addition, we quanti ed the arrangement of peripheral actin laments before and after remodeling by their distances, and analyzed the alignment of peripheral actin laments anchored to the PM via their distances and angles. Lastly, we characterized the interactions among ISGs, actin laments, and MTs at the cell periphery by their distances. Based on these quantitative analyses, we establish the model for actin remodeling at the cell periphery, providing a more detailed and accurate depiction of changes in the architecture, alignment, and interaction of actin laments with ISGs and MTs during GSIS in pancreatic β-cells.

Multimodal visualization of actin remodeling during glucose-stimulated insulin secretion
We imaged the clonal INS-1E cells, a rat β-cell widely used to mimic the biphasic insulin secretion in βcells in response to glucose stimulation [21][22][23] . We began by multimodal imaging to visualize actin remodeling during GSIS. We rst employed TIRF microscopy (x and y resolution of ~120 nm) to capture the biphasic insulin secretion response over 40 minutes of glucose stimulation (Extended Data Fig. 1 and Supplementary Video 1). The rst phase occurred during the rst 10 minutes after glucose stimulation, while the second phase occurred between 10-40 minutes, as we recorded the intensity of the NPY-mCherry marker for ISGs over time. Actin stress bers were observed at the basal membrane, maintaining the cell structure. We then performed a western blot experiment to detect the G-/F-actin ratio in the cell. We observed a signi cant increase in the ratio of the monomeric to lamentous actin states during the rst phase, followed by a signi cant decrease during the second phase, indicating the remodeling of actin laments during each phase (Extended Data Fig. 2).
Next, we performed SIM (x and y resolution of ~30 nm) to assess the actin meshwork organization around ISGs in different subsections of the β-cell. We visualized the projections of actin and ISGs along the z-axis of a 300 nm-thick cell section near the basal membrane (Fig. 1a-c). We observed a number of bundles in the basal conditions (44% ± 6%, bundles area by total actin area) and the second phase (30% ± 3%), but few in the rst phase of GSIS (18% ± 1%). Compared to the basal conditions, actin intensity signi cantly decreases in the rst phase of insulin secretion, followed by an increase in the second phase ( Fig. 1a-f). These changes are in agreement with changes in the G-/F-actin ratio measured by western blot (Extended Data Fig. 2). In addition, they are consistent with similar measurements by western blot and immuno uorescence techniques in previous studies 24,25 . To quantify the structural properties of actin meshwork organization during GSIS, we selected a few subsections of actin surrounding ISGs rather than other areas where actin forms bundles in the cell (Extended Data Fig. 3). After skeletonizing the actin meshwork based on the intensity of the actin uorescence label 26-28 (Methods, Fig. 1g), we quanti ed the length of actin meshwork and identi ed network junctions as connections in the actin meshwork. The actin meshwork is signi cantly shorter in the rst phase (515 ± 53 nm) and signi cantly longer in the second phase (2435 ± 167 nm) compared to the basal conditions (1432 ± 116 nm), in line with the western blot measurements ( Fig. 1h and Extended Data Fig. 2). Notably, actin forms a more complex network with an increased number of junctions in the second phase (9.4 ± 0.7) compared to the basal conditions (6.5 ± 0.4) and the rst phase (4.7 ± 0.5) (Fig. 1i). In summary, these results characterize actin remodeling near the basal membrane during GSIS, including depolymerization in the rst phase as well as repolymerization into a longer and more complex actin meshwork in the second phase.
Furthermore, we used cryo-ET to examine actin remodeling during GSIS at nanometer resolution, and quantify changes in architecture, alignment, and interaction of actin laments with other subcellular components. We collected tomograms of the cell periphery (0-6 μm from the PM) and the cell interior (0-3 μm from the nuclear membrane) (Supplementary Table 2). The cell periphery of INS-1E β-cells is onlỹ 250 nm thick, and vitri ed cells can be imaged directly using cryo-ET. To access the cell interior, vitri ed INS-1E β-cells were rst milled using a cryo-focused ion beam (cryo-FIB) machine to obtain lamellas with a thickness of approx. 150 ± 50 nm suitable for cryo-ET data collection 29,30 (Extended Data Fig. 3). In total, we collected 42 tomograms for 34 cells, including 24 tomograms at the cell periphery and 18 tomograms at the cell interior under basal conditions (2.8 Glu -30 min), in the rst phase (16.7 Glu -5 min), and the second phase of insulin secretion (16.7 Glu -30 min) (Supplementary Table 1).
In tomograms of the cell periphery, we distinguished various types of subcellular components, including actin laments as single brous laments 31,32 ; MTs as rigid tubular structures 33,34 ; mature and immature ISGs as membrane enclosed vesicles with and without dense core, respectively 35 ; lysosomes as monolayer vesicles with several types of contents in the lumen 36 ; and ribosomes ( Fig. 2a-c). We labeled these components via automated segmentation followed by manual re nement (Fig. 2d-f and Supplementary Video 2-4). Speci cally, we applied the cylinder cross-correlation to segment actin laments and MTs 37 , tomosegmemtv to segment membranous organelles such as ISGs and lysosomes 38 , and template matching to map ribosomes 39 . We de ne the segmentation unit as 60 nm for actin laments and 100 nm for MTs.
We observed changes in the amounts of actin laments and MTs in the tomograms of the cell periphery under different conditions ( Fig. 2a-f), in accordance with previous uorescence studies 25,40 . Speci cally, both the volume ratio of actin laments to the tomogram and the volume ratio of MTs to the tomogram signi cantly decrease in the rst phase, indicating a nearly complete depolymerization. The volume ratio of actin laments at the cell periphery after repolymerization does not change signi cantly compared to the basal conditions, in contrast to observations from randomly selected subsections of the cell in uorescence images acquired above (Fig. 2g, h and Fig. 1h). This analysis provides a direct in situ quanti cation of remodeling of actin laments and MTs at the cell periphery.
Actin laments maintain cell structure by forming bundles parallel to the PM at the cell periphery 41 . They also regulate the transport and release of ISGs through a remodeling process, as proposed in previous studies using uorescence microscopy 8, 13 . We computed changes in the orientation of individual actin laments relative to the PM (in the XY plane of a tomogram). Under basal conditions, 97% of the actin laments are oriented to the PM at the angles of 0-45° (ie, parallel orientation). In the second phase, the amount of actin laments oriented to the PM at the angles of 45-90° (ie, quasi-orthogonal orientation) increases from 3% to 15% (Fig. 2i); the amount of actin laments with angles of 60-90° to the PM increases from 1% to 11%. This quanti cation of the reorientation after remodeling allows us to analyze various features of actin laments as a function of their orientations to the PM under different conditions (below).
In tomograms of the cell interior, we distinguished different types of subcellular components, including actin laments, MTs, endoplasmic reticulum (ER) as a continuous monolayer membrane 42,43 , mitochondria as a double layer membrane structure with a at outer membrane and a curved interior membrane 35,44 , nuclear membrane 45 and ribosomes 46 (Fig. 3a-c). We labeled these components using the same methods as for the tomograms of the cell periphery ( Fig. 3d-f and Supplementary Video 5-7). In the cell interior, we observed no substantial difference in either the amounts of actin laments and MTs or the orientations of actin laments relative to the PM under different conditions ( Fig. 3g-i). Thus, we con rmed that the remodeling process does not appear to occur in the cell interior. The number of actin laments is insu cient to demonstrate statistically signi cant differences between the different conditions (data not shown). Similarly, there are no statistically signi cant differences between distances spanned by the subcellular components studied here (data not shown).
In summary, peripheral actin laments undergo a nearly complete depolymerization in the rst phase and then a repolymerization during GSIS. Most importantly, more actin laments are oriented quasiorthogonally to the PM after remodeling than before remodeling (3% and 15% for the basal conditions and the second phases, respectively). In contrast, actin laments in the cell interior do not undergo a remodeling process. Therefore, we proceed with the analysis of the peripheral tomograms only.
The architecture of the actin lament network before and after remodeling Tomograms of the cell periphery show a change in the architecture of the actin lament network during remodeling, despite a similar volume of actin laments (Fig. 4a-d and Fig. 2g). We computed the Actin-Actin distance as a function of the actin laments angles (Methods). Most actin laments form bundles, presumably to ful ll their primary function of maintaining cell structure, with an angle of 0-15° and a distance of 12-13 nm, as found in lopodia and stress bers in epithelial cells 32 (Fig. 4e, f). However, we observe fewer bundles in the second phase compared to basal conditions (yellow-to-red circled region in Fig. 4e, f). In addition, actin laments oriented parallel to the PM maintain the same distance distribution after remodeling, whereas actin laments oriented quasi-orthogonally to the PM show a shift to longer distances in the second phase compared to the basal conditions ( Fig. 4g, h). These results indicate that actin laments mostly retain their bundled formation but partially recon gure into a less compact arrangement that may be required to transport subcellular components from the cell periphery to the PM.
We manually labeled the PM and considered actin laments within 60 nm of the PM as being anchored to the PM ( Fig. 5a-d). The distance is de ned as the shortest distance between the end-points of each actin lament and the PM; this distance threshold takes into account the length of the actin lament segmentation unit (60 nm). Under basal conditions and in the second phase, the height of the cell periphery (190.5 ± 16.9 nm and 258.4 ± 31.3 nm) remained constant, as well as the number and percentage of anchored actin laments (263 ± 55 and 183 ± 68; 90% ± 4% and 71% ± 8%), as determined by a t-test (Extended Data Fig. 6). We calculated the angles between neighboring actin lament vectors that were anchored to the PM and mapped these angles against the orientations of the anchored actin laments relative to the PM ( Fig. 5e-h, Methods). The lament vector is de ned as extending from the end-point near the PM to the end-point farther from it; neighboring lament vectors are de ned as those anchored to the same PM whose end-points near the PM are within twice the length of the actin lament segmentation unit. Anchored actin laments are primarily oriented parallel to the PM at the angles of 0-15° with the highest and next highest peaks of the lament alignment at the angles of 0-30° and 150-180° ( Fig. 5e-f). This indicates a parallel alignment of the anchored actin laments both with the PM and with each other (ie, bundles). Under basal conditions, actin laments oriented quasi-orthogonally to the PM have the highest peak at 60-90°, suggesting a "netlike" architecture, as expected to act as a barrier that blocks ISGs from approaching the PM (Fig. 5g). In the second phase, actin laments oriented parallel to the PM show a lower probability at angles of 60-90° compared to basal conditions (Fig. 5f, h). Actin laments oriented quasi-orthogonally to the PM show a clear shift to smaller angles in the second phase compared with basal conditions (Fig. 5f, h). These ndings suggest that actin laments reorient quasiorthogonally to the PM and parallel to each other after remodeling compared to basal conditions. This reorganization changes the actin lament network from a "netlike" to a "blooming" architecture, where radial projections emanate from anchor points at the PM (Fig. 5b, d). The "blooming" architecture is thought to act as transportation tracks to the PM, which is expected to facilitate the transport and release of ISGs at the PM. These results offer the rst structural evidence for the previous hypothesis of actin laments blocking and facilitating ISG release at the PM before and after remodeling, respectively 5,8,10 .
Interaction of ISGs, actin laments, and MTs at the cell periphery Previous studies demonstrated that actin laments interact with MTs to facilitate insulin secretion 10 . Speci cally, the transport of ISGs occurs along MTs from the cell interior to the cell periphery, then along actin laments from the cell periphery to the PM 10 . In tomograms of the cell periphery, we observed a few ISGs in vicinity of both actin laments (Fig. 6a-d) and MTs (Fig. 6e-h). In addition, we observed several ISGs particularly close to the reoriented actin laments in the second phase, suggesting a potential association with the transport process (Fig. 6b, d). These observations led us to quantitatively analyze the shortest distances between ISGs and actin laments as well as between ISGs and MTs, as follows.
To characterize interactions between ISGs and actin laments, we focused on variations in their shortest distances within 200 nm during GSIS. The shortest distance is calculated as the distance between the resampled points of actin laments (at 4 nm intervals) and the nearest ISG surface (Methods). This distance threshold is estimated by considering the length of the actin lament segmentation unit (60 nm) and the lengths of myosin V and other unknown associated proteins (approximately 55 nm for the structurally de ned part of myosin V 16 and an additional tolerance of 85 nm to account for the uncertainty resulting from the unknown structure of 268 residues of myosin V and unknown associated proteins). The probability for the shortest distance to be within 200 nm decreases in the second phase compared with basal conditions (Fig. 6i). We mapped the shortest distances along different orientations of actin laments relative to the PM. Actin laments oriented parallel to the PM at orientations less than 15° have a larger shortest distance to ISGs in the second phase (130-200 nm) compared with basal conditions (70-200 nm) (cyan in Fig. 6j, k). These results indicate a further spatial arrangement of actin laments relative to ISGs after remodeling compared with basal conditions, which is expected for the subsequent transport and release of ISGs.
The critical role of MT-dependent transport of newly generated ISGs under glucose stimulation has been well established 10,47 . MTs have also been proposed to negatively regulate insulin secretion at the cell periphery under basal conditions, based on a uorescence microscopy observation of restricted ISG movements during MT stabilization by drugs 17 . Here, we calculated the MT-ISG distance as the distance between the resampled MTs (at 4 nm intervals) and the nearest ISG surface (Methods). We observe the highest peak region starting at 20 nm and centered at 200 nm under basal conditions (Fig. 6l), with a signi cant shift to longer distances in the second phase. This observation provides a structural perspective at the nanoscale on the previously proposed negative regulation of MTs for ISG release under basal conditions 17 . Speci cally, MTs "trap" instead of "transport" ISGs, because the transportation requires Kinesin-1, which has a size of approx. 60 nm. Moreover, this effect is signi cantly reduced in the second phase.
Lastly, for each actin lament, we computed the distance to the closest MT (shortest MT distance, Methods) to investigate their interactions (Fig. 7). The total volume of actin laments and MTs remains constant during GSIS. Actin laments oriented parallel to the PM at the angles of 0-15° tend to be closer to MTs, as indicated by their shortest MT distances in a narrower range in the second phase (20-300 nm) compared to basal conditions (20-800 nm; red to green in Fig.7e, f). Actin laments oriented quasiorthogonally to the PM also tend to be located closer to MTs during the second phase compared to basal conditions. In summary, actin laments parallel and quasi-orthogonal to the PM are both closer to MTs during the second phase than under basal conditions.

Discussion
Actin laments are known to regulate ISG transport and release under glucose stimulation. Because of the limited resolution of uorescence microscopy used in previous studies, actin remodeling has been characterized primarily as a "strong distribution" under basal conditions, a "diminished amount" during the rst phase, and a "recovery" during the second phase 48 . Here, we applied multimodal imaging to map INS-1E β-cells under basal conditions, as well as in the rst and second phases of insulin secretion. At the whole-cell level, we captured the biphasic insulin secretion and actin meshwork organization using TIRF (Extended Data Fig. 1 and Video 1), measured the G-/F-actin ratio by western blot (Extended Data Fig. 2) and revealed changes in the length and number of junctions in the actin meshwork in randomly selected subsections of the β-cells by SIM (Fig. 1). At the subcellular level, we mapped actin remodeling at the periphery and interior of β-cells by cryo-ET ( Fig. 2 and Fig. 3). Our work provides the rst in situ structure of actin remodeling at the nanoscale, as well as a quantitative analysis of changes in the architecture, alignment and interaction of the actin laments during GSIS. We demonstrate that actin remodeling occurs at the cell periphery but not in the cell interior. We establish the model for actin remodeling at the cell periphery as follows.
Under basal conditions (Fig. 8, left panel), actin laments are mostly oriented parallel to the PM at the angles of 0-45°, forming bundles that maintain the cell structure (Fig. 4). Actin laments anchored to the PM form a "netlike" architecture that presumably prevents ISGs from approaching the PM (Fig. 5). MTs adopt a close conformation surrounding ISGs, potentially limiting their movement (Fig. 6). Thus, we present a structural perspective on how actin laments and MTs act as barriers to block the transport and release of ISGs under basal conditions 13,17,49 . In the rst phase (Fig. 8, middle panel), actin laments are nearly completely depolymerized, MTs and ISGs are almost absent, indicating rapid insulin secretion. In the second phase (Fig. 8, right panel), most actin laments remain organized in bundles, presumably to ful ll their primary function of maintaining cell structure. Actin laments recon gure into a new network after repolymerization, leading to three major structural rearrangements that potentially facilitate the transport and release of ISGs from the PM: (i) approximately 12% of actin laments reorient themselves quasi-orthogonally to the PM at the angles of 45-90° (Fig. 2); (ii) the actin lament network mostly remains as cell-stabilizing bundles but partially recon gures its architecture into a less compact arrangement (Fig. 4); (iii) actin laments anchored at the PM reorganize from a "netlike" to a "blooming" architecture, which is apparently required for transporting ISGs from the cell periphery to the PM (Fig. 5).
The rearrangements of the actin lament network serve the functional role of regulating insulin secretion. They precede the transport of ISGs by their reorientations quasi-orthogonal to the PM and less compact packing as well as the release of ISGs by the formation of a "blooming" architecture of actin laments anchored to the PM. Additionally, our analysis reveals changes in interactions among actin laments, MTs, and ISGs during GSIS. Previous uorescence studies have led to a proposal that both actin laments and MTs regulate ISG transport and release by demonstrating limited ISG movements under chemical stabilization of either actin laments or MTs 13,17 . Here, we show that actin laments and MTs are further away from ISGs in the second phase compared to the basal conditions (Fig. 6), while being closer to each other (Fig. 7), as expected for the subsequent transport and release of ISGs.
We now discuss the limitations of our work. Another future focus should be on the secretion sites near the PM, to gain deeper insights into the secretion mechanism. Indeed, mapping the interactions between actin laments and other organelles throughout the entire β-cell would provide valuable insights into their importance for ISG positioning, transport and release 51 . Third, we analyzed the structure of actin laments anchored to the PM using a distance threshold. It is known that actin laments are anchored to the PM via physical connections with focal adhesion complexes at the PM 8 . As an example, gold particle labeling can be used to detect these complexes in order to reveal interactions between focal adhesions and actin laments during remodeling, as well as to better understand the molecular mechanisms involved in insulin secretion.
In summary, our study presents the structure and quantitative analysis of actin remodeling at the nanoscale in both the periphery and interior of β-cells, including changes in the architecture of the actin lament network, the alignments of actin laments and the interaction among actin laments, MTs, and ISGs during GSIS. These analyses contribute to our understanding of actin remodeling and its role in the regulation of biphasic insulin secretion in pancreatic β-cells.
To achieve different cellular phenotypes under different conditions, INS-1E β-cells were exposed for 30 min in KREBS solution containing 2.8 mM glucose as the basal conditions, then changed into the KREBS solution containing 16.7 mM glucose and kept for 5 min to simulate the rst phase of GSIS, and kept for 30 min to simulate the second phase of GSIS. cells/cm 2 . The cells were grown on the grids for 48 hours, and then the grids were blotted from the backside for 10s using a Vitrobot Mark IV (Thermo Fisher Scienti c) with 10% humidity, and plungefrozen in ethane liquid. Subsequently, the vitri ed samples were transferred into cryo-EM boxes and stored in liquid nitrogen before FIB-milling or cryo-ET data collection.

Live-cell uorescence imaging
To record the INS-1E live-cell images during the time of GSIS, cells were grown on 35 mm glass-bottom dishes (ibid), transfected with plasmids to label ISGs and actin laments, and stimulated with glucose as described above. The movies were acquired using wide-eld TIRF mode under light microscope system at 37°C in 5% CO 2 . This light Microscopy was performed on a Nikon Ti2-E with TIRF equipped with a Prime

Western blot
The ratio of G-and F-actin was performed following the manufacturer's protocols (Cytoskeleton). Protein samples were separated by SDS-PAGE and transferred to PVDF membrane (Thermo Fisher) with known amounts of actin used for quantitation in G-/F-actin ratio experiments. Membranes were blocked in 5% skim milk powder in tris buffered saline (TBS, 50 mM Tris pH 7.5, 150 mM NaCl) with 0.1% Tween 20 (Merck). Incubate with the primary antibody (rabbit) provided in the kit overnight at 4 °C. Then incubate with the secondary antibody (Abcam, anti-rabbit, 1:5000). Finally, the membrane was exposed after incubation in a dark environment with ECL substrate for 1 min (Bio-rad). subsections were subjected to a cascade of processing using Image J (v1.53f51), rstly the actin is ltered using the "Threshold-default" tool, then skeletonized using the "skeletonize" tool, ultimately, the information of actins was determined using "Analyze skeleton" tool, including the total length and the network junction. Based on "Analyze skeleton", we here only considered meaningful actins whose length is larger than one pixel for the following measurement 26 . Analysis was repeated for six independent cell images under each condition.

Cryo-FIB milling
Cryo-FIB milling was carried out following similar procedures as previously described 53 , using dualbeam FIB-SEM microscopes Aquilos 2 Cryo FIB (Thermo Fisher Scienti c). Grids holding frozen cells were clipped into autogrid support rings with a cutout region to facilitate shallow-angle cryo-FIB milling. The Autogrids were mounted onto the cryo-FIB AutoGrid shuttle and transferred to the cryo-stage at liquid nitrogen temperatures. Before the milling procedure, grids were sputter-coated with platinum for the 30s (10 mA) and subsequently sputter-coated with organometallic platinum using the gas injection system (GIS, Thermo Fisher Scienti c) for 8 s to improve conductivity and remove artifacts. Samples were milled by gallium ion beam at 30 kV with a stage tilting angle of 17-19° to generate 10-12 μm wide lamellas, the initial rough milling was done under 0.5 nA high currents, then the current was gradually decreased in stepwise manner to 30 pA for ne milling and nal polishing. Electron beam at 2 kV/13 pA or 5 kV/25 pA was used for SEM imaging during the milling process. In total, 18 lamellas from randomly chosen cells were used for the following cryo-ET data collection.

Cryo-ET acquisition and reconstruction
The grids holding either frozen INS-1E β-cell or lamella samples produced by FIB-milling were loaded in a Titan Krios G3 or G4 TEM (Thermo Fisher Scienti c). The Titan G3 TEM was equipped with a 300-kV eld-emission gun, a post-column energy lter (Gatan), and a 5760 x 4092 K3 Summit direct electron detector (Gatan), operated using SerialEM 54 . Low-magni cation images were captured at 3600×. Highmagni cation tilt series were recorded in counting mode at 26000x (calibrated pixel size of 0.3353 nm).
The Titan Krios G4 TEM (Thermo Fisher Scienti c) was equipped with a 300-kV C-FEG eld-emission gun, a post-column energy lter (Selectris), and a 4096 x 4096 direct electron detector (Falcon4), operated using Tomography (Thermo Fisher Scienti c). Low-magni cation images were captured at 5600×. Highmagni cation tilt series were recorded in counting mode at 42000x (calibrated pixel size of 0.3028 nm).
For cell periphery samples, tilt series were collected with ± 60° tilt range, 2° step. 24 tilt-series, collected from 16 different cells were collected in total. For lamella samples, a similar scheme covering 120° with 2° increments starting from ~ 10° compensating the pre-tilt of lamella was applied. 18 tilt-series, collected from 18 different cells were collected in total. All 42 tomograms were recorded at a total dose of 110-140 e-/Å 2 using a dose-symmetric tilt scheme 55 and a target defocus range of -3 to -7 µm (Supplementary   Tables 1-2).
Data preprocessing including motion correction by MotionCor2 56 and the dose-ltering step 57 was performed using the TOMOMAN package which could execute on MATLAB (2019b) software (https://github.com/williamnwan/TOMOMAN). Tilt-series were aligned by patch tracking in IMOD 58 and reconstructed to 4x binned tomograms (with a pixel size of 13.4 Å and 12.1 Å for cell periphery and lamella, respectively) using weighted-back projection. Tomograms were then denoised by the cryoCARE algorithm (https://github.com/juglab/cryoCARE_T2T) for better segmentation and visualization.

Filament and membrane segmentation
Page 14/28 The above mentioned 4× binned tomograms were used for the segmentation. Correlative volumes of membrane positions were detected and generated automatically using tomosegmemtv 38 and imported to Amira software (Thermo Fisher Scienti c) for manual re nement and segmentation. Actin laments and MTs were traced automatically in Amira software using an automated segmentation algorithm, which adopted a cylinder as a template 37 and implemented in the X-Tracing extension. The cylindrical templates were generated with a length of 60 nm or 100 nm and diameters of 8 nm or 15 nm for the actin laments or MTs, respectively. After the automated segmentation algorithm, the segmented results were manually checked by the tomogram intensity map. The identi cation of the PM positions within the tomogram is achieved through a manual segmentation process, whereby both intracellular and extracellular regions are recognized to create a PM mask.

Data analysis
Data analysis was performed using in-house Python scripts using coordinates of actin laments and MTs, PM mask volumes, and ISG mask volumes, which were exported from Amira software. First, we adjusted the coordinates based on the offset of each tomogram. Next, we resampled actin laments and MTs with 4 nm intervals; this interval has been previously shown to provide a reasonable t in other analyses 32 . The Actin-Actin distance represents the distance between the centerline of two actin lament segments, which was calculated as the Euclidean distance between each resampled point on the actin lament and its nearest resampled points on all other actin laments (Fig. 4). The actin laments angles represent the relative orientations between each two actin laments. This angle was calculated as the angle between the vectors formed by two end-points of the actin lament and the vectors formed by two end-points of all other laments (Fig. 4). The anchored actin lament angles represent the relative orientations between two nearby actin laments anchored to the PM. For each anchored actin lament, the neighboring laments are identi ed if: 1) they were anchored to the same PM, and 2) the distance between their end-points near the PM was less than 120 nm, which is twice the length of the actin lament segmentation unit (60 nm). The angle was then calculated between each anchored lament vector extending from the end-point near the PM to the end-point farther from it and all other nearby anchored lament vectors (Fig. 5). The Actin-ISG distance was calculated as the Euclidean distances of each resampled point of actin laments to the nearest ISG surface (Fig. 6). The MT-ISG distance was calculated in the same way as the Actin-ISG distance (Fig. 6). The shortest MT distance of actin lament was calculated as the Euclidean distance between each resampled point on the actin lament and its nearest resampled points on all MTs (Fig. 7).
All statistical analysis was undertaken using GraphPad Prism (v9.4.1) and OriginPro (v9.8.0.200). Statistical signi cance tests for the total length and network junction of actin meshwork, the analysis of both cell periphery and interior of the cell ratio of actin lament-vol/tomo-vol and MT-vol/tomo-vol, and orientation of actin laments at the cell interior were calculated using one-way ANOVA test, followed by Tukey's multiple comparisons tests. Statistical signi cance tests for the orientation of peripheral actin laments, the average cell height, the number of anchored actin laments, and the percentage of anchored actin laments by total actin laments were calculated using an F-test and t-test. Values are reported as the mean values. The error bars represent SEMs unless otherwise stated.
Declarations 57. Grant, T. & Grigorieff, N. Measuring the optimal exposure for single particle cryo-EM using a 2.6 Å reconstruction of rotavirus VP6. Elife 4, e06980 (2015).    views. e, f Density map of the distance between actin laments as a function of the angle between actin laments before (e) and after (f) remodeling. g Histogram of the distances between actin laments whose orientation relative to the PM is less than 45° before and after remodeling. Bin size = 10. h Histogram of the distances between actin laments whose orientation relative to the PM is greater than or equal to 45° before and after remodeling. Bin size = 10. Each point on the density map re ects the corresponding density values by calculating kernel density. Quantitative analysis of PM-anchored actin laments at the cell periphery. a, c 3D visualization of actin laments (anchored: violet, not anchored: green) before (a) and after (c) remodeling. b, d Zoomed-in views of the actin laments highlighted in a and c, represented as a function of their orientation relative to the PM in a blue-to-red color map, before (b) and after (d) remodeling. e, f Density map of the distance between PM-anchored actin laments as a function of their orientation relative to the PM before (e) and after (f) remodeling. g Histogram of the angle between PM-anchored actin laments whose orientation is less than 45° relative to the PM before and after remodeling. Bin size = 15. h Histogram of the angle between PM-anchored actin laments whose orientation relative to the PM is greater than or equal to 45°b efore and after remodeling. Bin size = 15. Each point on the density map re ects the corresponding density values by calculating kernel density.

Figure 6
Quanti cation of the interaction between actin laments and MTs with ISGs. a, b ISGs (gray) and actin laments (same color code as in Fig. 4) visualized under basal conditions (a) and in the second phase

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.