Hypoxia induces stress fiber formation in adipocytes in the early stage of obesity

In obese adipose tissue (AT), hypertrophic expansion of adipocytes is not matched by new vessel formation, leading to AT hypoxia. As a result, hypoxia inducible factor-1⍺ (HIF-1⍺) accumulates in adipocytes inducing a transcriptional program that upregulates profibrotic genes and biosynthetic enzymes such as lysyl oxidase (LOX) synthesis. This excess synthesis and crosslinking of extracellular matrix (ECM) components cause AT fibrosis. Although fibrosis is a hallmark of obese AT, the role of fibroblasts, cells known to regulate fibrosis in other fibrosis-prone tissues, is not well studied. Here we have developed an in vitro model of AT to study adipocyte-fibroblast crosstalk in a hypoxic environment. Further, this in vitro model was used to investigate the effect of hypoxia on adipocyte mechanical properties via ras homolog gene family member A (RhoA)/Rho-associated coiled-coil kinases (ROCK) signaling pathways. We confirmed that hypoxia creates a diseased phenotype by inhibiting adipocyte maturation and inducing actin stress fiber formation facilitated by myocardin-related transcription factor A (MRTF-A/MKL1) nuclear translocation. This work presents new potential therapeutic targets for obesity by improving adipocyte maturation and limiting mechanical stress in obese AT.

. Experimental schematic and workflow. Human mesenchymal stem cells (hMSCs) were expanded in 2D until confluence. To differentiate to adipocytes, adipogenic induction media was added for 7 days. To prepare the 3D constructs, adipocytes were encapsulated in type 1 collagen hydrogels (2 mg/mL) alone or with fibroblasts in a 50:1 ratio with a final seeding density of 8 million/mL. The constructs were maintained in adipogenic maintenance media. Constructs were kept in normoxia to mimic healthy adipose tissue (AT) or in hypoxia to mimic the early stage of obesity. Hypoxia was chemically induced by 1 mM DMOG addition to the media or by culturing in 1% O 2 . After 7 days, AT constructs were sacrificed for endpoint analyses. www.nature.com/scientificreports/ 1% oxygen in a hypoxia chamber. Hypoxia was confirmed by western blot, where HIF-1⍺ was present in both hypoxic conditions. Further, HIF-1⍺ target genes, LEP and vascular endothelial growth factor (VEGF) were affected in hypoxia (Fig. 2a,g, Supplementary Fig. S1) 28 . There were no differences in adipocyte features between chemically induced hypoxia via DMOG and 1% oxygen groups ( Supplementary Fig. S2). Subsequent data sets are shown for constructs in 1% oxygen as a more physiologically relevant model of hypoxia over chemically induced hypoxia. Adipocyte differentiation and lipid droplet accumulation were confirmed through gene expression and morphological characterization (Fig. 2b,c,e,f, representative confocal images of the mono-culture constructs maintained in 1% oxygen). Hypoxia did not significantly affect cell viability (Fig. 2d). Culturing in 1% oxygen resulted in significantly smaller lipid droplet size and number and cell area (Fig. 2e,f). When exposed to hypoxia, adipocyte specific genes, adiponectin (ADIPOQ), peroxisome proliferator-activated receptor gamma (PPARG), and fatty acid binding protein 4 (FABP4) were significantly downregulated, whereas leptin (LEP) was upregulated (Fig. 2g). The presence of fibroblasts did not affect adipocyte maturation (Fig. 2g,h, Supplementary  Fig. S8). It is known that alpha integrin expression shifts from α5 in preadipocytes to α6 in mature adipocytes during adipogenesis . Since decreased adipocyte maturation in hypoxia was observed, integrin α6 (ITGA6) and integrin α5 (ITGA5) gene expression were quantified. ITGA6 was significantly downregulated in the hypoxia condition, while no significant changes were seen for ITGA5 and integrin β1 (ITGB1) (Fig. 2h, Supplementary  Fig. S2). Altogether, these data demonstrate that, hypoxia reduces hallmarks of adipocyte maturation through downregulating PPARG, ADIPOQ, increasing LEP expression, and results in smaller and fewer lipid droplets. Furthermore, ITGA6 gene expression is significantly downregulated in less mature adipocytes (Fig. 2i).

Figure 2.
Hypoxia develops a diseased phenotype by inhibiting adipocyte maturation. (a) Hypoxia was confirmed by western blot, HIF-1⍺ was present in hypoxic conditions (cropped blot is shown for the monoculture condition, full length blots are presented in Supplementary Fig. S1). (b) Cross-section of AT constructs, fluorescent confocal microscopy images were taken at least 20 µm above the glass coverslip. (c) Fluorescent confocal microscopy images of the mono-culture AT constructs in normoxia or 1% oxygen after 7 days stained for Propidium Iodide (PI) to quantify cell viability (scale bar-100 µm). (d) Hypoxia did not affect cell viability. (e) Adipocyte area was significantly decreased in 1% oxygen, data are presented for the mono-culture condition (n = 4 biological replicates, 14-20 cells per replicate). (f) Lipid droplet size and average lipid droplet number per adipocyte were significantly decreased in 1% oxygen, data are presented for the mono-culture condition (n = 3-4 biological replicates). (g) Adipocyte gene expression after 7 days, PPARG , ADIPOQ, and FABP4 were significantly downregulated, and LEP was significantly upregulated in 1% oxygen (n = 3-8 biological replicates). (h) ITGA6 was significantly downregulated in 1% oxygen (n = 4-8 biological replicates). (i) Hypoxia reduces hallmarks of adipocyte maturation through the downregulation of PPARG, ADIPOQ and FABP4 expression, increasing LEP expression and results in smaller and fewer lipid droplets. ITGA6 gene expression is significantly downregulated in less mature adipocytes. Data are presented as means ± SEM. Comparisons between groups and statistical analysis were performed using two-way ANOVA with Tukey post hoc test, unpaired t-test or Mann-Whitney test with two-tailed p-values (*p < 0.05, **p < 0.01, ****p < 0.0001).  29 . To further investigate the effect of hypoxia on integrins and their interactions with ECM, we assessed gene expression for the ECM proteins that bind to integrins α6 and α5, LAMA4, and FN, respectively. FN expression was significantly upregulated, and matrix metalloproteinase-2 (MMP-2), known to degrade fibronectin, was significantly downregulated in the chemically induced hypoxia, but not in 1% oxygen. In addition, LOX expression, known to increase deposition and assembly of fibronectin fibrils 30,31 , was significantly upregulated in both hypoxic conditions ( Fig. 3a,b, Supplementary Fig. S5). Fluorescent confocal microscopy of the constructs for fibronectin revealed that adipocytes can assemble fibronectin into a fibrillar structure in both hypoxic conditions, while a punctuate fibronectin organization was observed in normoxia. However, the fibrillar structure was more prominent in the chemically induced hypoxia group (Fig. 3c,d, Supplementary Fig. S3, representative confocal images of the mono-culture constructs). These results suggest that hypoxia increases FN and LOX gene expression and promotes a fibrillar fibronectin structure (Fig. 3e).
Hypoxia induces stress fiber formation through MKL1 nuclear translocation. Fibronectin fibrillar assembly is known to be mechano-regulated via the cell-mediated contractile forces 32 . These contractile forces are partly related to increased actin stress fiber formation within the cells. Dysfunctional adipocytes have more filamentous actin as they interact with their fibrotic environment. Therefore, we hypothesized that distinct fibronectin matrix assembly and lack of maturation observed in hypoxia (Figs. 2, 3) are related to actin cytoskeletal remodeling. To investigate the effect of hypoxia on actin cytoskeletal remodeling, we examined stress fiber formation in adipocytes. In this model, upon exposure to hypoxia, actin alpha 2 (ACTA2), a gene that encodes actin stress fibers (α-SMA), was significantly upregulated (Fig. 4a,b, Supplementary Fig. S4). The presence of actin stress fibers was further confirmed through confocal images and protein expression (Fig. 4c,d, representative confocal images of the mono-culture constructs maintained in 1% oxygen, Supplementary Fig. S5). To further determine the mechanical pathway that results in enhanced α-SMA expression in hypoxia, MKL1, a transcriptional factor translocated to the nucleus during activation of the RhoA/ROCK pathway, and regulates α-SMA expression was explored 33,34 . In normoxia, MKL1 expression was greater in the cytoplasm, in contrast to 1% oxygen, where MKL1 expression was greater in the nucleus. (Fig. 4e, representative confocal images of the mono-culture constructs maintained in 1% oxygen, Supplementary Fig. S6). The nuclear to cytoplasmic distribution ratio for MKL1 was significantly greater in 1% oxygen (Fig. 4f). The greater nuclear to cytoplasmic ratio in 1% oxygen was further confirmed through protein expression (Fig. 4g). Together, these data support that hypoxia induces actin stress fiber formation in adipocytes with MKL1 nuclear translocation (Fig. 4h). Comparisons between groups and statistical analysis were performed using two-way ANOVA with Tukey post hoc test (*p < 0.05, **p < 0.01). www.nature.com/scientificreports/

Y27 treatment partly attenuates hypoxia effects by inhibiting MKL1 nuclear translocation.
To determine the molecular pathway that supports stress fiber formation and MKL1 translocation in hypoxia, the RhoA/ROCK mechanical pathway was explored by exposure to Y-27632 (Y27), a biochemical inhibitor of ROCK. After Y27 treatment, the α-SMA signal was present, but the defined stress fiber morphology was not visible, confirming that inhibiting ROCK disrupts stress fiber formation ( Fig. 5a,b,e, representative confocal images of the mono-culture constructs maintained in 1% oxygen). Furthermore, the inhibition of ROCK by Y27 led to an upregulation in adipogenic related genes, PPARG and ADIPOQ (Fig. 5c). Y27 treatment results in decreased MKL1 expression in the nucleus as well as nuclear to cytoplasm distribution ratio, confirming the possible role of ROCK in MKL1 translocation to the nucleus. MKL1 protein expression further confirmed decreased nucleus to cytoplasmic ratio after Y27 treatment (Fig. 5d,f,g, Supplementary Fig. S6). Overall, we demonstrated that RhoA/ROCK signaling pathway regulates actin stress fiber formation and MKL1 nuclear translocation within adipocytes in hypoxia (Fig. 5h).

Discussion
This study sought to develop an in vitro AT model to study adipocyte-fibroblast crosstalk in healthy (normoxia) and diseased (hypoxia) states. Further, we used this model to investigate the effect of hypoxia as the early stage of obesity on RhoA/ROCK signaling pathway in adipocytes. Multiple cell types in AT contribute to ECM synthesis during fibrosis, such as adipocyte progenitors as well as adipocyte crosstalk with other cells, including macrophages [35][36][37][38] . Fibroblasts and activated fibroblasts (myofibroblasts) are sources of ECM production in other fibrosis-prone tissues; therefore, we co-cultured adipocytes with fibroblasts to determine if they played a role in the observed AT diseased phenotype. The ratio of adipocytes to fibroblasts (50:1) was selected to match the ratio (f) Nuclear to cytoplasmic ratio for MKL1 was significantly increased in 1% oxygen (n = 5 biological replicates, 20 cells per replicate). (g) MKL1 nuclear (N) to cytoplasmic (C) protein ratio was significantly increased in 1% oxygen, arrows point to the quantified MKL1 protein bands (data are presented for the mono-culture condition, n = 3 biological replicates, cropped blot is shown, full length blot is presented in Supplementary Fig. S6). (h) Hypoxia induces actin stress fiber formation via MKL1 nuclear translocation. Data are presented as means ± SEM. Comparisons between groups and statistical analysis were performed using two-way ANOVA with Tukey post hoc test or unpaired Mann-Whitney test with two-tailed p-values (*p < 0.05, **p < 0.01, ***p < 0.001). www.nature.com/scientificreports/ found in human AT 39 . We did not observe any significant differences between mono-and co-culture conditions in the assays performed. Further, our results indicated that fibroblasts did not go under myofibroblastic activation under hypoxic conditions. This confirms previous studies demonstrating that hypoxia can inhibit the activation of fibroblasts through inhibition of pathways that lead to α-SMA activation, like SMAD independent pathways and RhoA GTPase dependent actin polymerization ( Supplementary Fig. S7) [40][41][42] . In this crosstalk model, fibroblasts were not activated directly through hypoxia or via adipocyte crosstalk. Therefore, we concluded that hypoxia mainly affects adipocytes through a decrease in their adipogenic regulation (i.e., PPARG) and downstream changes to their mechano-sensitive pathways while other fibrotic stimuli such as transforming growth factor-β (TGF-β) and platelet-derived growth factor (PDGF) may be needed to trigger fibroblast activation 43 .
In the current study, we focused on hypoxia as an early feature of an obese AT. Our results demonstrated that hypoxia creates a diseased model by affecting both adipocytes and their ECM. Hypoxia inhibited maturation was confirmed through smaller and fewer lipid droplets, smaller adipocyte size, and downregulation of PPARG and ADIPOQ, and upregulation of LEP. PPARG is the master regulator of adipogenesis, and it is known to upregulate other adipocyte related genes, such as adiponectin (ADIPOQ). Similarly, leptin is also an adipokine and a hallmark of adipogenesis. However, leptin is abnormally increased in obesity. Therefore, taken together, downregulation of PPARG and ADIPOQ and upregulation of LEP expression in the constructs exposed to hypoxia indicates a diseased adipocyte phenotype.
Adipocyte differentiation and maturation are, in part, regulated through integrin expression, where integrin α5β1 attachment to fibronectin is replaced by α6β1 attachment to laminin during adipogenesis [44][45][46] . We found ITGA6 was significantly downregulated in hypoxia, which supports the earlier results that adipocyte maturation is reduced in hypoxia. While we did not observe transcriptional level changes for ITGA5, there are several routes to post-transcriptional integrin α5 regulation, including changes to integrin activation state or shuttling to the (c) Adipogenic (ADIPOQ, PPARG ) genes were partly rescued towards a normoxic phenotype after Y27 treatment (n = 4-8 biological replicates, data are normalized to mono-culture samples maintained in 20% oxygen). (d) Fluorescent confocal microscopy images of the mono-culture AT constructs and the orthogonal view in 1% oxygen before and after Y27 treatment, arrows point to MKL1 signal localization (scale bar-20 µm, insets-10 µm). (e) Cross-section of AT constructs, fluorescent confocal microscopy images were taken at least 20 µm above the glass coverslip. (f) The nucleus to cytoplasm ratio for MKL1 was significantly decreased after Y27 treatment (n = 5 biological replicates, 20 cells per replicate). (g) MKL1 nuclear to cytoplasmic protein ratio in 1% oxygen and after Y27 treatment, arrows point to the quantified MKL1 protein bands (n = 3 biological replicates, cropped blot is shown, full length blot is presented in Supplementary Fig. S6). (h) Y27 treatment disrupts actin stress fibers due to MKL1 nuclear translocation inhibition, and this process is regulated via RhoA/ ROCK signaling pathway. Data are presented as means ± SEM. Comparisons between groups and statistical analysis were performed using unpaired Mann-Whitney test with two-tailed p-values (*p < 0.05, **p < 0.01). www.nature.com/scientificreports/ cell surface. Interestingly, here we noted differences in the hypoxia groups, where in chemically induced hypoxia, fibronectin (FN), the α5β1 integrin binding partner, was upregulated, and at the protein level was organized in a more fibrillar manner. Further, the enzyme which degrades fibronectin, MMP-2, was downregulated in chemically induced hypoxia. Taken together, this shows that adipocytes in hypoxia are further inhibited in maturation by increased FN expression and organization with limited matrix remodeling by MMP-2, while also limiting the a6 attachments to their ECM and potentially favoring existing α5 integrins. Determining the α5 integrin activity of adipocytes in hypoxia may increase our understanding of the temporal and functional effects regulating FN organization and adipocyte mechanical pathways. This study has shown that the methods used to generate hypoxia can lead to differences in outcomes. Given that hypoxia is often confirmed by HIF-1⍺ expression by immunoblot, it was important to note that while both chemically induced hypoxia and 1% oxygen led to similar accumulation of HIF-1⍺ (Fig. 2a). However, the use of a hypoxia chamber to induce hypoxia may be a less stable hypoxic environment, as the constructs are intermittently briefly exposed to normoxia during media exchanges, while chemically induced hypoxia groups are not.
Adipogenesis and adipocyte maturation are dependent on actin cytoskeletal remodeling. This remodeling process involves stress fiber disruption and cortical actin formation to facilitate lipid accumulation 16 . The inhibited maturation we observed in hypoxia is partly related to increased stress fiber formation in adipocytes exposed to hypoxic conditions. In obesity, adipocytes become metabolically dysfunctional due to a fibrotic phenotype, including enhanced expression of cytoskeletal, ECM, and focal adhesion genes 47,48 . MKL1, a mechanoresponsive transcription factor, is a known regulator of myofibroblast differentiation that controls the expression of cytoskeletal and ECM genes such as α-SMA (ACTA2); some collagens, as well as PPARG 40,41 . MKL1 is prohibited from translocating into the nucleus within the healthy adipocytes due to active actin remodeling (polymerization) and interaction with globular (G)-actin [49][50][51][52] . However, when MKL1 is translocated to the nucleus due to RhoA/ROCK activity, it can decrease PPARG and increase α-SMA (ACTA2) expression, similar to the results observed in hypoxia (Figs. 2g, 4a).
A contractile cell phenotype is known to contribute to fibronectin organization. Previous studies have demonstrated that increased α-SMA and MKL1 expression in myofibroblast enabled those cells to assemble the extrinsic fibronectin to a fibrillar matrix 32,[53][54][55] . This agrees with our findings, where in constructs exposed to hypoxia, we observed increased contractile gene expression, ACTA2, which enables these cells to assemble fibronectin molecules into a fibrillar structure due to these applied contractile forces (Figs. 3c, 4a).
Overall, this study illustrates the role of hypoxia as an early hallmark of obesity in creating a diseased environment not only by affecting adipocyte maturation but also their mechanical properties. Hypoxia affects mechanical properties by inducing stress fiber formation resulting from MKL1 translocation to the nucleus. Therefore, hypoxia and its downstream mechanical pathways, including MKL1 that is regulated via RhoA/ROCK signaling pathway, can be used as a potential target for treating the early stage of obesity (Fig. 6).
Polydimethylsiloxane cast preparation. Polydimethylsiloxane (PDMS; Sylgard 184, Dow Corning) mixture, the weight ratio of base to curing agent at 10:1, was cast at 100 °C oven for 45 min to fabricate 1 mm thick sheets. The Cured PDMS was cut into squares, and 5 mm holes were made using a 5 mm biopsy punch (AcuPunch). To attach PDMS to the glass coverslip, the surface of both was activated by plasma ashing (PE25W, PlasmaEtc) for 1 min. The assembled devices were then treated with 0.01% (v/v) poly-L-lysine (PLL, Sigma-Aldrich) for 1 h, rinsed thoroughly in deionized (DI) water, and then treated with 1% (v/v) glutaraldehyde (Electron Microscopy Sciences) for 30 min to enhance collagen adhesion. Devices were then rinsed thoroughly in DI water again and ultraviolet (UV) sterilized for 15 min. www.nature.com/scientificreports/ bonate (NaHCO 3 , Fisher Scientific), 10X Medium199 (Sigma-Aldrich) and water to the final concentration of 2 mg/mL, while maintained on ice. The neutralized collagen solution was mixed with either adipocytes or fibroblasts alone (mono-culture) or the mixture of two cell types (co-culture) in a 50:1 ratio with a final seeding density of 8 million/mL. Collagen droplets with a volume of 30 μL were dispensed into the wells created on the PDMS/glass coverslip device, followed by 30 min of incubation at 37ºC to allow for collagen crosslinking. Fresh maintenance media was added to the wells post-crosslinking and was replenished every 2-3 days. Constructs were kept either in normoxic conditions or, after 24 h, switched to hypoxic conditions. Hypoxia was induced chemically with the addition of 1 mM HIF-Hydroxylase inhibitor (dimethylglycine, DMOG) (EMD Millipore) dissolved in dimethyl sulfoxide (DMSO, Millipore Sigma) or by placing constructs in 1% oxygen within a hypoxia chamber (Billups-Rothenberg, Inc.) (Fig. 1).
To reduce cytoskeletal tension, Y-27632 (Y27), a pharmacologic ROCK inhibitor (Tocris Biosciences) dissolved in DMSO, was added to the maintenance media every 48 h, at a final concentration of 10 µM.
Gene expression. RNA was collected from 2-3 pooled constructs per condition using Trizol (Ambion) and stored at -80ºC until needed. Chloroform (Acros Organics) phase separation technique was used to isolate RNA, and the RNA was precipitated in 50% isopropanol (Fisher Chemical). After that, it was washed twice with 75% ethanol and resuspended in diethylpyrocarbonate (DEPC)-treated water (Invitrogen). Total RNA reverse transcription and Real-time PCR (RT-PCR) was performed with High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems) and Power Up SYBR Green Master Mix (Applied Biosystems), respectively. Primers were custom designed (Integrated DNA Technologies). Samples were run in triplicate using an Eppendorf Mastercycler and StepOnePlus PCR system (Applied Biosystems). Fold change in relative gene expression levels was determined using the 2 -ΔΔCt method.
Protein expression. Protein extracts from constructs were obtained by either lysing the cells in 2X laemmeli sample protein buffer (Bio-Rad) supplemented with 1 mM Dithiothreitol (DTT, Sigma-Aldrich) or by radioimmunoprecipitation assay (RIPA) buffer supplemented with phosphatase inhibitor cocktail that contains 250 mM sodium fluoride, 50 mM sodium orthovanadate, 50 mM Sodium pyrophosphate decahydrate, and 50 mM β-glycerophosphate (Sigma-Aldrich). To separate nuclear and cytoplasmic fractions, AT constructs were incubated in hypotonic buffer (20 mM Tris-HCl, 10 mM NaCl, 3 mM MgCl 2 ), vortexed and subjected to centrifugation (1699 rcf) and the supernatant was collected as a cytoplasmic fraction. The remaining pellet was resuspended in a cell extraction buffer, vortexed in 10 min intervals for 30 min and subjected to centrifugation (20,817 rcf) and the supernatant was collected as the nuclear fraction. www.nature.com/scientificreports/ Proteins were separated by gel electrophoresis using 8% or 12% SDS-PAGE gel and then transferred onto a nitrocellulose membrane (Immobilon-P). After blocking and probing with antibodies, the detection was performed using horseradish peroxidase (HRP, Bio-Rad) conjugated anti-rabbit secondary antibody (1:3000, Cat# 170-6515, RRID:AB_11125142) and enhanced chemiluminescence reagent (Bio-Rad), and the signal was visualized using C-DiGit Blot Scanner.
Fluorescent confocal microscopy and image analysis. Images were acquired with a laser scanning confocal mode of the hybrid confocal-multiphoton microscope (Olympus FluoView FV1200) using 10X (UPLXAPO10X, 0.4 NA, Olympus) or 30X (UPLSAPO30XSIR, 1.05 NA, Olympus) oil immersion inverted objectives with four main laser units (405, 488, 543, and 635 nm.), and four photomultiplier (PMT) detectors. Z-stacks (thickness ~ 50 µm, step-size: 5 µm, 20 µs/pixel, 1024 × 1024) were captured at least 20 µm above the glass coverslip to ensure visualized cells were in a 3D environment and not in contact with the glass coverslip surface (Figs. 2b, 3d, 4b, and 5e). Brightness and contrast were applied equally to the experimental conditions when appropriate to improve signal and to reduce background.
Acquired images were imported to ImageJ for further analysis. Lipid droplet diameter and number were quantified by an ImageJ Plugin, "MRI_Lipid Droplets Tool" (http:// dev. mri. cnrs. fr/ proje cts/ imagej-macros/ wiki/ Lipid_ Dropl ets_ Tool). Adipocyte area was calculated by specifying adipocyte area using the area selection tool in ImageJ, and quantifying the area using the measurement tool. Since the adipocyte seeding ratio is high (8 million/mL), to ensure that one cell at the time was measured, image analysis was performed slice by slice such that selected cells are not overlapping. To clarify cell boundaries, we have counterstained the samples with phalloidin to visualize actin. To quantify MKL1, slices from the stacks were z-projected using the Max Intensity method. The nucleus to cytoplasm ratio for MKL1 was calculated using the "RawIntDen" measurement tool, and the intensity value was normalized to the nucleus and cytoplasm area. To quantify the α-SMA area, the α-SMA channel was converted to a binary display, and the percentage area of the signal in the field of view (FOV) was calculated with the measurement tool. Collagen cluster thickness was quantified by using "BoneJ" macro in ImageJ. In brief, BoneJ macro fits spheres along the collagen fibers after a threshold is applied to the image and calculates the thickness as the average largest circle that can fit within the fibers 57,58 . A thermal map for fibronectin was created by pseudo coloring the fluorescent confocal images with thermal lookup table (LUT) in ImageJ (Supplementary Fig. S3).
Statistics. GraphPad Prism 8 (GraphPad Software Inc.) software was used to perform all statistical analyses. All experiments were conducted with at least three biological replicates and 15-20 cells per biological replicates for image quantification. The distribution of each data set was analyzed, and the D' Agostino-Pearson test (α = 0.05) was performed to test for normality. Statistical comparisons between two experimental groups were performed using two-tailed Student's t-test or Mann-Whitney test for non-normally distributed data when appropriate. Comparisons among more groups were performed using two-way analysis of variance (ANOVA) with Tukey post hoc testing. All graphs are presented as mean ± standard error of the mean (SEM) unless otherwise stated. Significance was determined according to *p < 0.05, **p ≤ 0.01, ***p ≤ 0.001, and ****p ≤ 0.0001.

Data availability
The datasets generated and analyzed during the current study are available from the corresponding author upon reasonable request.