Ca²⁺-driven PDIA6 biomolecular condensation ensures proinsulin folding

Abstract

The endoplasmic reticulum (ER) plays crucial roles in maintaining protein quality control and regulating dynamic Ca²⁺ storage in eukaryotic cells. However, the proteostasis system involved in ER-mediated protein quality control has not been fully characterized. Here we show that Ca²⁺ triggers the condensation of PDIA6, an ER-resident disulfide isomerase and molecular chaperone, into quality control granules. In contrast to the condensation mechanism observed for proteins containing low-complexity domains, our results indicate that transient but specific electrostatic interactions occur between the first and the third folded thioredoxin-like domains of PDIA6. We further show that the PDIA6 condensates recruit proinsulin, thereby accelerating the oxidative proinsulin folding and suppressing the proinsulin aggregation inside quality control granules, essential for secretion of insulin.

Lee et al. show that Ca²⁺ triggers condensates enriched with PDIA6, an ER-resident disulfide isomerase and chaperone, along with other protein disulfide isomerase family proteins and some chaperones that in turn enhance folding of proinsulin.

Main

Cells have evolved membrane-bound or membraneless compartments to isolate biochemical and biophysical reactions^1,2. Nearly one-third of nascent polypeptide chains enter the endoplasmic reticulum (ER) in which a proteostasis network, comprising molecular chaperones and oxidoreductases, guides the correct folding of a wide variety of client proteins^3–6. During this process, the proteostasis network prevents aggregation and promotes post-translational modifications such as disulfide bond formation and glycosylation⁷. The ER is also involved in dynamic storage of Ca²⁺ (refs. ^8,9), a second messenger in cells, and alterations in Ca²⁺ concentrations in the ER cause ER stress, which can ultimately lead to apoptosis¹⁰. The functions of many ER-resident chaperones and oxidoreductases, such as the members of the protein disulfide isomerase (PDI) family, are modulated by fluctuations in Ca²⁺ concentrations within the ER¹¹. Ca²⁺-driven proteostasis in the ER, however, remains incompletely understood. Members of the PDI family are key components of the ER proteostasis system^12,13. These proteins are composed of multiple thioredoxin (Trx)-like domains, which have evolved distinct structures and activities^14–19, resulting in functional diversity²⁰. One key feature is the ability of several PDI family members to not only bind to Ca²⁺ (refs. ^15,21) but also to form complex reversible oligomeric states that regulate their functions^14,22.

These observations led us to investigate the possibility that PDI family members engage in even more complex phase behaviour and form condensates that have active roles in ER proteostasis. Our results reveal that Ca²⁺-driven condensates of PDIA6 recruit proinsulin, thereby accelerating the oxidative proinsulin folding and suppressing the proinsulin aggregation inside quality control granules, essential for secretion of insulin.

Results

Ca²⁺ induces phase separation of PDIA6

To determine whether members of the PDI family can self-assemble into condensates in a Ca²⁺-driven process, we measured the diameters of proteins and condensates under increasing Ca²⁺ concentrations using dynamic light scattering (Fig. 1a and Extended Data Fig. 1a). Although almost no changes were detectable in the average diameter of highly purified PDIA1, PDIA3, PDIA4, PDIA10 and PDIA15, a Ca²⁺ concentration-dependent increase in the average diameter was observed for PDIA6 (Fig. 1a). Following the addition of EDTA, the average diameter of oligomeric PDIA6 returned to that of PDIA6 dimers (Extended Data Fig. 1b). At PDIA6 concentrations exceeding 5 μM in the presence of >2.5 mM Ca²⁺, differential interference contrast (DIC) microscopy revealed numerous spherical droplets of up to 50 μm in diameter, the formation of which was induced by Ca²⁺ from the bulk solution (Fig. 1b,c and Extended Data Fig. 1c). Notably, condensates with a diameter of ≥50 μm were not observed at lower (<5 μM) PDIA6 concentrations. Measurement of fluorescence recovery after photobleaching (FRAP) revealed that fluorescence recovered in droplets containing mCherry–PDIA6 with a half-time (t_1/2) of about 19 s (Fig. 1d). Three-dimensional (3D) holographic imaging of the refractive index (RI) demonstrated that smaller droplets coalesced to form larger spherical droplets over time (Extended Data Fig. 2 and Supplementary Videos 1 and 2)²³, whereas both liquid-like droplet fusion and droplet growth were inhibited by 10 mM NaCl (Fig. 1e and Extended Data Fig. 1d,e). Droplet formation was most efficient below pH 7.4 and impaired above pH 7.6 (Fig. 1f and Extended Data Fig. 1f), the physiological pH range within the ER. Thus, PDIA6 droplets can coalesce into relatively large droplets (such as those with a 50 μm diameter) in vitro in the presence of Ca²⁺, although the size of droplets in cells may become smaller depending on the salt concentration and pH inside the cells.

Fig. 1. Ca²⁺ drives PDIA6 condensation.

a, Mean diameter of ensemble particles in solution (Z-average) of PDIA1, PDIA3, PDIA4, PDIA6, PDIA10 and PDIA15 under various Ca²⁺ concentrations. The values are the mean ± s.d. of three independent experiments. b, Liquid droplets observed by DIC microscopy when 50 μM PDIA6 and 4 mM CaCl₂ were mixed at pH 7.2. This experiment was replicated three times independently. c, PDIA6 phase diagram obtained by DIC microscopy when 5–100 μM PDIA6 and 0.5–10 mM CaCl₂ were mixed at pH 7.2. Dominant PDIA6 states at varying protein and Ca²⁺ concentrations are indicated by symbols: black circles, dispersed state; black triangles, condensed state. The dashed black line represents the critical droplet concentration. Three independent experiments were performed. d, Confocal fluorescence images of PDIA6 droplets before and after photobleaching. The white arrowhead indicates the laser irradiation area. Rapid recovery of mCherry–PDIA6 fluorescence after photobleaching (left). Increases in the normalized fluorescence intensity of mCherry–PDIA6 after photobleaching (five replicates; right). The fluorescence recovery t_1/2 was calculated from the normalized fluorescence intensity of the five replicates. e, Liquid droplets observed by DIC microscopy when 50 μM PDIA6 and 4 mM CaCl₂ were mixed with (right) or without (left) NaCl. This experiment was replicated three times independently. f, Liquid droplets observed by DIC microscopy when 50 μM PDIA6 and 4 mM CaCl₂ were mixed in solutions with different pH values. This experiment was replicated three times independently. g, Time course of representative two-dimensional (2D) RI distribution (top), and bright-field (middle) and fluorescence images (bottom) of FUS and PDIA6 droplets monitored by 3D holographic imaging (green, ThT fluorescence; three independent experiments). [Ca²⁺], Ca²⁺ concentration; BF, bright field; FI, fluorescence image; [PDIA6], PDIA6 concentration.

Source data

Extended Data Fig. 1. Ca²⁺-driven PDIA6 droplet formation.

a, Purity of the recombinant PDI family proteins. Representative data from three independent biological experiments are shown. b, Z-average (the mean diameter of ensemble particles in solution) of Ca²⁺-driven PDIA6 droplets with/without EDTA. The values are the mean ± s.d. of three independent experiments. c, Liquid droplets observed by DIC microscopy when 5–100 μM PDIA6 and 0–7.5 mM CaCl₂ were mixed at pH 7.2. This experiment was performed three times independently. d, Critical concentration of PDIA6 for droplet formation. The values are the mean ± s.d. of three independent experiments. e, Critical concentration of PDIA6 for droplet formation in the presence of 10 mM NaCl. The values are the mean ± s.d. of three independent experiments. f, Z-average of PDIA6 under Ca²⁺ at various pH values. The values are the mean ± s.d. of three independent experiments.

Source data

Extended Data Fig. 2. Real-time monitoring of PDIA6 condensate formation.

a, Real-time images of PDIA6 (50 μM) condensate formation after addition of 4 mM CaCl₂. b, Real-time imaging of fusion between the two droplets. The data were obtained by 3D holographic imaging of the refractive index (RI). These experiments were performed three times independently.

We next investigated whether amyloid formation takes place within the liquid-like PDIA6 droplets. This question is relevant as it is well-established that other proteins—including, for example, fused in sarcoma (FUS), an intrinsically disordered protein with low-complexity sequences associated with amyotrophic lateral sclerosis—can form amyloid-like assemblies within liquid-like condensates^24,25. We found that although liquid-like droplets of PDIA6 and FUS have heterogeneous internal RIs due to their high fluidity, in contrast to FUS, PDIA6 condensates did not exhibit thioflavin T (ThT) fluorescence after 44 min (Fig. 1g), indicating that they have a persistent liquid-like nature (Fig. 1g).

An examination of the endogenous localization of PDIA6 in U2OS cells revealed that it co-localized with calnexin (CNX), an ER marker, with some endogenous PDIA6 forming puncta (PDIA6 foci; Fig. 2a). In FRAP experiments using U2OS cells, fluorescence recovered with a t_1/2 of about 2.3 s after photobleaching of mCherry–PDIA6 foci (Fig. 2b). To explore the behaviour of PDIA6 in response to changes in Ca²⁺ levels in the ER, we investigated the effects of thapsigargin and A23187, which deplete Ca²⁺ in the ER²⁶. U2OS cells overexpressing mCherry–PDIA6 were treated with 0.3 μM thapsigargin or 3 μM A23187 for 1 h, rinsed with PBS and then cultured for 0, 2 or 4 h. Staining with Mag-Fluo4 AM, which detects Ca²⁺ in the ER, revealed that thapsigargin and A23187 reduced the Ca²⁺ concentrations in the ER but Ca²⁺ reinflux into the ER was observed after 2 h (Fig. 2c and Extended Data Fig. 3a). After Ca²⁺ reinflux into the ER, PDIA6 foci appeared at significant levels at 2 h and PDIA6 co-localized with CNX and the binding immunoglobulin protein (BiP, also known as GRP78; Fig. 2d–g and Extended Data Fig. 2b–d). CNX is known to be a lectin chaperone in the ER, and BiP is a chaperone of the Hsp70 family that is also located in the ER. Consistent with a previous report showing that BiP physiologically interacts with PDIA6 (ref. ²⁷), the co-localization of BiP and PDIA6 foci suggests that BiP condenses in these foci. In response to Ca²⁺ reinflux into the ER, the number of PDIA6 foci was also increased at 2 h (Fig. 2h). Foci formation by PDIA6 was reversed by thapsigargin treatment (Fig. 2i), as revealed in FRAP experiments of mCherry–PDIA6 foci, similar to the steady-state foci (Fig. 2b). Together, these results indicated that PDIA6 can reversibly regulate its oligomeric state in response to Ca²⁺ in vitro and in cells.

Fig. 2. Formation of Ca²⁺-driven PDIA6 foci in U2OS cells.

a, U2OS cells were stained for endogenous PDIA6 and CNX. b, Rapid recovery of the fluorescence of mCherry–PDIA6 foci after photobleaching in U2OS cells. The graph shows the recovery in normalized mCherry–PDIA6 fluorescence for ten foci after photobleaching, obtained from three biologically independent experiments. The fluorescence recovery t_1/2 was calculated from the normalized fluorescence intensity of three replicates. c, U2OS cells stably expressing mCherry–PDIA6 were stained with Mag-Fluo4 AM and treated with thapsigargin. Three independent biological experiments were performed. d, U2OS cells were treated with thapsigargin. After thapsigargin treatment, the cells were stained for PDIA6 and CNX. Five independent biological experiments were performed. e, U2OS cells stably expressing mCherry–PDIA6 were treated with thapsigargin. After thapsigargin treatment, the cells were stained for mCherry and BiP. f, U2OS cells stably expressing mCherry–PDIA6 were treated with thapsigargin. After thapsigargin treatment, the cells were stained for mCherry and CNX. Magnified views of the dashed boxes are provided (right). a,e,f, Representative images from five independent biological experiments are shown. g, Following treatment with thapsigargin or A23187, the signal intensities of mCherry–PDIA6 and CNX were statistically analysed using line scan data (n = 6 individual experiments). h, Average number of PDIA6 foci in a cell after thapsigargin (control, n = 27 cells, 759 total foci; 0 h, n = 31 cells, 269 total foci; 2 h, n = 24 cells, 950 total foci; 4 h, n = 29 cells, 926 total foci) and A23187 treatment (control, n = 19 cells, 509 total foci; 0 h, n = 25 cells, 212 total foci; 2 h, n = 19 cells, 559 total foci; 4 h, n = 20 cells, 545 total foci) pooled from five independent biological experiments. Box plots show the median (centre line), the 25th and 75th percentiles (bounds of box) and the minimum and maximum values (whiskers). g,h, Statistical significance was examined using a one-way analysis of variance (ANOVA) with Tukey’s honest significant difference post-hoc test; the test was two-sided. *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001; NS, not significant. i, Following thapsigargin treatment, mCherry–PDIA6 foci in U2OS cells were analysed using FRAP. Normalized fluorescence intensity curves are shown for ten (thapsigargin) and seven (dimethylsulfoxide, DMSO) foci obtained from three biologically independent experiments. The fluorescence recovery t_1/2 was calculated from the mean of these three independent biological replicates. b,g,i, Data are presented as the mean ± s.d. TG, thapsigargin.

Source data

Extended Data Fig. 3. Formation of Ca²⁺-driven PDIA6 foci in U2OS cells.

a, U2OS cells stably expressing mCherry–PDIA6 were stained with Mag-Fluo4 AM and treated with A23187. Representative data from three independent biological experiments are shown. b, U2OS cells were treated with A23187. After A23187 treatment, the cells were stained for PDIA6 and CNX. Five independent biological experiments were performed. c, U2OS cells stably expressing mCherry–PDIA6 were treated with A23187. After TG treatment, the cells were stained for mCherry and GRP78/BiP. Representative images from five independent biological experiments are shown. d, U2OS cells stably expressing mCherry–PDIA6 were treated with A23187. After A23187 treatment, the cells were stained for mCherry and CNX. The dashed boxes on the left are shown in more detail on the right. Representative data from five independent biological experiments are shown.

Biophysical and structural characterization of Ca²⁺ binding to PDIA6

The concentration of NaCl in cells can vary between cell types and depends on cytosolic fluctuations in Ca²⁺ and osmotic stress^28–31. Given that the Ca²⁺-driven condensation of PDIA6 is sensitive to NaCl concentration (Fig. 1e and Extended Data Fig. 1d,e), we examined the effects of NaCl on the interaction between Ca²⁺ and PDIA6. To explore the thermodynamics of Ca²⁺ binding to PDIA6, we monitored heat changes during Ca²⁺ binding to PDIA6 at 0 and 100 mM NaCl using isothermal titration calorimetry (ITC), which measures intermolecular interactions, including those leading to protein aggregation^32–36 (Extended Data Fig. 4a,b). Both ITC thermograms displayed upwards-pointing peaks, indicating that binding of Ca²⁺ to PDIA6 is an endothermic reaction (Extended Data Fig. 4a,b). An increase in the NaCl concentration from 0 to 100 mM reduced the magnitude of the peaks. Further fitting analyses of the binding isotherms revealed the thermodynamic parameters for PDIA6:Ca²⁺ complex formation. As the NaCl concentration increased from 0 to 100 mM, the unfavourable enthalpy changes (ΔH > 0) decreased from 0.65 ± 0.11 to 0.08 ± 0.01 kcal mol⁻¹ and the favourable entropy changes (TΔS > 0) decreased from 5.51 ± 0.19 to 4.26 ± 0.12 kcal mol⁻¹ (Extended Data Fig. 4b). As a result, the binding affinity between PDIA6 and Ca²⁺ weakened, with an increase in the dissociation constant (K_d) from 271.0 ± 33.3 to 857.0 ± 153.0 μM and a decrease in the change in free energy (ΔG) from −4.86 ± 0.07 to −4.18 ± 0.11 kcal mol⁻¹. Notably, the stoichiometry (n) of Ca²⁺ binding to PDIA6 increased from 4.2 ± 0.6 to 18.0 ± 1.8, which is likely to be due to a reduction in electrostatic interactions between Ca²⁺ and PDIA6. The decrease in electrostatic interactions may open additional binding sites for Ca²⁺ on PDIA6 that were previously inaccessible due to strong electrostatic repulsion. Overall, the binding of Ca²⁺ to PDIA6 is purely entropically driven, probably due to the displacement of water molecules following Ca²⁺ binding^35,36. The notable suppression of the interaction by NaCl further emphasizes the importance of electrostatic forces for PDIA6 binding to Ca²⁺ (Extended Data Figs. 1e and 4c).

Extended Data Fig. 4. Biophysical characterization of PDIA6 droplets.

a, Effects of NaCl on the interaction between Ca²⁺ and PDIA6. ITC data (upper panel) and binding isotherm data (lower panel) for the titration of Ca²⁺ against PDIA6 with and without 100 mM NaCl. The solid line in the binding isotherm represents the best fit curve based on a model with one set of binding sites. The experiments were independently repeated three times with reproducible results. b, Thermodynamic parameters of Ca²⁺ binding to PDIA6. The error value of each parameter represents the fitting error. c, Effects of NaCl on Ca²⁺-driven droplet formation. Bright-field observation of PDIA6 droplets observed by 3D holographic imaging when 50 μM PDIA6 and 4 mM CaCl₂ were mixed at various NaCl concentrations. This experiment was performed three times independently. d,e, Structural analyses of PDIA6 by NMR (related to Fig. 3). ¹H–¹⁵N TROSY-HSQC spectra (d) and ¹H–¹³C HMQC spectra (e) of [U-²H¹⁵N; Ala-¹³CH₃; Met-¹³CH₃; Ile-δ1-¹³CH₃; Leu, Val-¹³CH₃/¹³CH₃]-labelled full-length PDIA6. f, ¹H–¹⁵N TROSY-HSQC (left panel) and ¹H–¹³C HMQC spectra (right panel) of [U-¹⁵N; Ala-¹³CH₃; Met-¹³CH₃; Ile-δ1-¹³CH₃; Leu, Val-¹³CH₃/¹³CH₃]-labelled PDIA6 a⁰ (orange), a (green), and b (grey) domains. g, Electrostatic surface of each PDIA6 domains.

Source data

Measurement of the Ca²⁺ concentration in the supernatant after centrifugation of PDIA6 droplets revealed that centrifugation reduced the Ca²⁺ concentration in the supernatant from 4 to 3.6 mM (Fig. 3a) at the critical droplet concentration of the PDIA6 (Extended Data Fig. 1d). The stoichiometry of Ca²⁺ binding to PDIA6 was found to be n = 8, which was determined by dividing the decrease in Ca²⁺ concentration (400 μM) by the protein concentration (50 μM).

Fig. 3. Ca²⁺ binding and Ca²⁺-induced phase separation of PDIA6.

a, Concentration of Ca²⁺ outside PDIA6 droplets. Values are the mean ± s.d. of three independent experiments. b, PDIA6 is composed of two redox-active domains (a⁰ and a) at the N terminus and a redox-inactive domain b at the C terminus. PDIA6 forms a homodimer via a unique Val-Leu adhesive motif contained in a⁰. c–e, Evaluation of Ca²⁺ binding sites on PDIA6 using NMR. c, ¹H–¹⁵N HSQC spectra of the three ¹⁵N-labelled domains of PDIA6 (left, a⁰ domain; middle, a domain; right, b domain) obtained in the absence (orange, a⁰ domain; green, a domain; grey, b domain) and presence (red) of 20 mM CaCl₂. Insets: expanded views of the regions outlined by dashed boxes in the spectra. d, Graphical representation of the CSPs between the resonances of domains a⁰ (left), a (middle) and b (right). The solid and dotted lines indicate the average values and average values plus standard deviations, respectively. e, CSP mapping of PDIA6. The residues with remarkable CSPs after Ca²⁺ binding are indicated as blue spheres.

Source data

PDIA6 is composed of two amino-terminal redox-active Trx-like domains (a⁰ and a) and a carboxy-terminal redox-inactive Trx-like domain (b). PDIA6 forms a homodimer via a unique Val-Leu adhesive motif contained in a⁰ and all domains are located to form various extended conformation, thereby conferring a flexible nature to the protein in solution (Fig. 3b)¹⁵. The sites responsible for Ca²⁺ binding were characterized by performing nuclear magnetic resonance (NMR) analysis of the PDIA6 domains (Fig. 3c and Extended Data Fig. 4d–g). NMR measurements were performed in the presence of 200 mM NaCl to discriminate the Ca²⁺ binding to PDIA6 from the interaction among PDIA6 assembly. ¹H–¹⁵N heteronuclear single quantum coherence (HSQC) spectra of PDIA6 showed that Ca²⁺ induced notable perturbations in the resonances from all three domains (Extended Data Fig. 5a–c). Chemical shift perturbation (CSP) mapping revealed that Ca²⁺ binds to multiple specific sites on PDIA6, including an Asp and Glu-rich segment at the C terminus of b, an acidic patch in a⁰ consisting of Asp27, Asp28 and Glu31, and an acidic patch in a consisting of Glu166, Glu199 and Asp223. Notably, in the a⁰ domain, which is essential for dimerization, CSPs were also detected in peripheral α-helix 4 region, which contains the Leu-Val adhesion motif responsible for dimer formation, suggesting this characteristic motif in the a⁰ domain is critical for both Ca²⁺ binding to PDIA6 and driving liquid-like condensation of PDIA6 (Fig. 3d,e)¹⁵.

Extended Data Fig. 5. NMR spectra of each domain of PDIA6 by CaCl₂ titration.

a–c, ¹H–¹⁵N heteronuclear single quantum coherence (HSQC) spectra of the three ¹⁵N-labelled domains of PDIA6 a⁰ (a), a (b), and b (c) obtained in the absence (black) and presence of 10 (blue) or 20 mM (purple) CaCl₂.

Both a⁰ dimerization and a⁰–b interaction are critical for PDIA6 phase separation

To further determine whether each Trx-like domain is essential for the Ca²⁺-driven PDIA6 condensates, DIC microscopy was performed on individual Trx-like domains. It is noteworthy that although it has been pointed out that the linker between a⁰ and a is important for the PDIA6 condensates³⁷, we used a⁰ that does not contain a linker between a⁰ and a. The results showed that although the a⁰ domain formed slightly smaller droplets than full-length PDIA6 after Ca²⁺ was added, none of the other domains formed droplets (Extended Data Fig. 6a). Residual concentration analysis, gel shift assays and microscopy analyses using 3D imaging that constructs bright-field contrast and the RI distribution were also performed using individual a⁰, a and b domains. Each domain was added pairwise to another domain in the presence Ca²⁺ and incubated for 30 min. After centrifugation, the protein concentration in the supernatant was determined. The results showed that only the a⁰ and b combination decreased the protein concentration (red circles in Fig. 4a) in the supernatant compared with the theoretical concentration (black circles in Fig. 4a) and resulted in a precipitate after centrifugation. Under the same pre-centrifugation conditions, microscopy analyses showed that only the a⁰ and b combination resulted in droplet formation, whereas the a and b combination resulted in the formation of slightly smaller droplets (Extended Data Fig. 6b). As observed with full-length PDIA6 (Extended Data Fig. 4c), Ca²⁺-induced phase separation between the isolated a⁰ and b domains was suppressed by an increase in NaCl concentration (Extended Data Fig. 6c), indicating that transient but specific electrostatic interactions between the a⁰ and b domains are essential for Ca²⁺-induced phase separation.

Extended Data Fig. 6. The second Trx-like domain a is not essential for PDIA6 droplet formation.

a, Phase separating ability of each PDIA6 domain. Liquid droplets observed by DIC microscopy when 50 μM full-length PDIA6 or PDIA6 domains and 4 mM CaCl₂ were mixed. Three independent experiments were performed. b, Liquid droplets observed by bright-field (BF) microscopy when a⁰, a, or b domain containing Ca²⁺ was mixed with b, a, or a⁰. Three independent experiments were performed. c, Liquid droplets observed by BF microscopy when a⁰ or b domain containing Ca²⁺ was mixed with b or a⁰ in the presence of different NaCl concentrations. Three independent experiments were performed. d, Quantification of residual domains after each domain was mixed pairwise together. a⁰ or b domain containing Ca²⁺ was added to another domain, incubated for 30 min, and centrifuged, and the supernatant was separated by SDS‒PAGE for quantitative analysis. Three independent experiments were performed. Data are presented as mean ± s.d. e, Liquid droplets observed by BF microscopy when a⁰A5 (monomeric mutant) or b domain containing Ca²⁺ was mixed with b or a⁰A5. Three independent experiments were performed.

Source data

Fig. 4. Mechanism of PDIA6 droplet formation.

a, The a⁰–b interaction is critical for PDIA6 condensation. Residual concentrations obtained when the domains were mixed pairwise together. The black and red open circles indicate the theoretical and measured values, respectively. The values are the mean ± s.d. of three independent experiments. b, Dimerization of a⁰ is critical for PDIA6 condensation. Close-up view of dimeric motif within a⁰ (top). Representative 2D RI distribution when the a⁰ and b domains (bottom left) or monomeric mutated a⁰A5 and b domains (bottom right) are mixed, monitored by 3D holographic imaging (three independent experiments). BF, bright-field images. c,d, NMR investigation of the interaction between PDIA6 molecules during droplet formation. c, ¹H–¹⁵N transverse relaxation optimized spectroscopy-HSQC (left) and ¹H–¹³C heteronuclear multiple quantum coherence (HMQC; right) spectra of [U-²H¹⁵N; Ala-¹³CH₃; Met-¹³CH₃; Ile-δ1-¹³CH₃; Leu, Val-¹³CH₃/¹³CH₃]-labelled full-length PDIA6 in the absence and presence of 4 mM CaCl₂. d, ¹H–¹⁵N HSQC spectra of ¹⁵N-labelled a⁰ (top left) in the absence (orange) and presence (red) of b and ¹⁵N-labelled b (top right) in the absence (grey) and presence (red) of a⁰. Insets: expanded views of the regions outlined by dashed boxes in the spectra. The intensity ratios between the resonances of ¹⁵N-labelled a⁰ (bottom left) and b (bottom right) in the absence and presence of the other domain are provided. The solid and dotted lines indicate the average and average values, minus the s.d., respectively. Residues with large changes in intensity are indicated as blue spheres in the structure of the a⁰ (left) and b (right) domains. The error bar of intensity ratio (ΔR) was estimated from signal-to-noise ratios using the following equation: $\mathrm{\Delta }R=R\sqrt{{\text{SNR}}_{\text{ref}}^{-2}+{\text{SNR}}_{\text{tit}}^{-2}}$ where R is the intensity ratio, and ${\text{SNR}}_{\text{ref}}^{-2}$ and ${\text{SNR}}_{\text{tit}}^{-2}$ are the signal-to-noise ratios of the peak on the reference spectrum and the titrated spectrum, respectively. e, Conformational change of PDIA6 induced by condensation. Arrival time distributions of 19+ charged ions of PDIA6 dimer before (top) and after (bottom) condensation observed in the IM–MS experiments. The arrival times of PDIA6 dimer with 0, 4 and 8 Ca²⁺ ions before and after condensation are indicated in the plots.

Source data

To further detect the decrease in content of a⁰ and b domains in the supernatant after centrifugation, a⁰ and b domains in the supernatant after centrifugation were quantified by SDS–PAGE (Extended Data Fig. 6d). When domain b was added in excess and the a⁰ amount was decreased, the amount of b decreased to approximately 60% after centrifugation (Extended Data Fig. 6d). On the other hand, the amounts of a⁰ and b domains decreased in tandem in a a⁰ domain dose-dependent manner, demonstrating that a sufficient amount of the a⁰ domain is essential for Ca²⁺-induced phase separation. Furthermore, 3D holographic imaging of the RI revealed that loss of the dimerization motif in the first a⁰ domain suppressed droplet formation, indicating that the dimerization motif regulates Ca²⁺-induced phase separation (Fig. 4b and Extended Data Fig. 6e).

Next, NMR spectra of full-length PDIA6 in the absence and presence of Ca²⁺ were analysed to obtain structural information on the PDIA6 droplet. The NMR measurements were performed in the absence of NaCl, a condition that favours droplet formation. The addition of Ca²⁺ markedly broadened PDIA6 resonances and resulted in the disappearance of most of the dispersed resonances (Fig. 4c). This can be attributed to the resonance broadening induced by PDIA6:Ca²⁺ interactions as well as by the reduced mobility induced by oligomerization. To evaluate the specific interactions between a⁰ and b domains that are essential for Ca²⁺-induced phase separation, NMR was used to study the interaction between the a⁰ and b domains in the presence of 20 mM CaCl₂ and 200 mM NaCl. The addition of a⁰ to isotopically labelled b led to notable perturbations in the resonances of the b domain around its N-terminal α helix (Fig. 4d), which suggests that b uses a specific region to interact with a⁰. On the other hand, the addition of b to isotopically labelled a⁰ resulted in fewer notable perturbations from multiple locations within the domain, suggesting that a⁰ undergoes multivalent and non-specific interactions with b (Fig. 4d).

To detect Ca²⁺ binding and determine phase separation status by native-mass spectrometry (MS), electrospray ionization (ESI) mass spectra of PDIA6 in 30 or 60 mM ammonium acetate (NH₄OAc) were measured in the presence of 250 μM CaCl₂. Droplet formation occurred in 60 mM NH₄OAc but was inhibited in 30 mM NH₄OAc (Extended Data Fig. 7a). Consistent with a previous report of PDIA6 dimerization in solution¹⁵, native-MS detected 17+, 18+, 19+ and 20+ ions of PDIA6 homodimers (Extended Data Fig. 7b). Each m/z value is equal to (M + n × 1.008)/n, where 1.008 is the mass of a hydrogen atom, so M can be calculated from the m/z values of ions with at least two consecutive charge states³⁸. Dimer peak broadening was attributed to inefficient desalting and/or desolvation due to the presence of 250 μM calcium ion. Furthermore, when coupled with ion mobility MS (IM–MS), native-MS can provide insights into protein conformation³⁹. Dimeric PDIA6 in the 19+ charge state exhibited a shorter arrival time in IM–MS in the Ca²⁺-bound state than in the Ca²⁺-free state (Fig. 4e), suggesting that the binding of Ca²⁺ to dimeric PDIA6 induced its conformational compaction. On the other hand, the opposite trend was observed for PDIA6 under phase separation conditions: Ca²⁺-bound dimeric PDIA6 in the 19+ charge state exhibited a longer arrival time in IM–MS than Ca²⁺-free dimeric PDIA6. Together, these findings indicate that PDIA6 homodimers in droplets assume more extended conformations than those of soluble PDIA6 homodimers to achieve weak and multivalent intermolecular interactions between PDIA6 molecules.

Extended Data Fig. 7. Condensate images and ESI mass spectra of PDIA6 in NH₄OAc with 250 μM CaCl₂.

a, Condensates observed by bright-field (BF) microscopy when 5 μM PDIA6 and 250 μM CaCl₂ were mixed in NH₄OAc. Three independent experiments were performed. b, ESI mass spectra of 5 μM PDIA6 with 250 μM CaCl₂ in NH₄OAc.

Recruitment of other PDI family members and clients into PDIA6 condensates

Given that many PDI family members are ubiquitously expressed in the ER, we investigated whether other PDI family members are incorporated into PDIA6 droplets. Using the fluorescence intensity of GFP-labelled PDI family members, we found that among the PDI family members, PDIA3 was most efficiently concentrated in PDIA6 droplets (Fig. 5a and Extended Data Fig. 8a,b). To determine the effect of each PDI on PDIA6 condensation, we examined RI changes inside PDIA6 droplets after the addition of other untagged PDI family members using 3D holographic imaging. Consistent with the fluorescence microscopy results (Fig. 5a), the RI values of PDIA6 droplets increased to 1.4 in a PDIA3 dose-dependent manner, revealing selective condensation between PDI family members (Fig. 5b,c). To further investigate whether PDIA3 co-localized with PDIA6 foci in U2OS cells, we performed co-immunostaining of PDIA3 and PDIA6. The fluorescence intensity of PDIA3 was higher in areas with PDIA6 foci, indicating intracellular co-localization of PDIA3 and PDIA6 foci (Fig. 5d). The RI values inside PDIA6 droplets were decreased in a PDIA4 dose-dependent manner and became sparse (Fig. 5b,c and Extended Data Fig. 8c). Thus, PDIA4 could conceivably act to extract substrates from inside PDIA6 condensates or it may control the formation of PDIA6 condensates.

Fig. 5. Condensation of PDIA3 in PDIA6 droplets.

a, Condensation of the green fluorescent protein (GFP)-tagged PDI family members in PDIA6 droplets. This experiment was performed three times independently. Statistical significance was examined using a one-way ANOVA with Tukey’s honest significant difference post-hoc test; the test was two-sided. ****P < 0.0001. b, Changes in the RI inside PDIA6 droplets with increasing concentrations of different PDI family members. Representative 2D RI distributions are indicated. Data were analysed for 198 particles for PDIA6 only, 217 particles for 1 µM PDIA1, 210 particles for 5 µM PDIA1, 212 particles for 10 µM PDIA1, 215 particles for 1 µM PDIA3, 220 particles for 5 µM PDIA3, 212 particles for 10 µM PDIA3, 215 particles for 1 µM PDIA4, 219 particles for 5 µM PDIA4, 212 particles for 10 µM PDIA4, 214 particles for 1 µM PDIA10, 212 particles for 5 µM PDIA10, 217 particles for 10 µM PDIA10, 217 particles for 1 µM PDIA15, 218 particles for 5 µM PDIA15 and 226 particles for 10 µM PDIA15 pooled from three independent replicates. a,b, Data are presented as the mean ± s.d. c, Representative images of the 2D RI distribution of PDIA6 droplets with untagged PDI family members, monitored by 3D holographic imaging from the same dataset analysed in Fig. 5b. d, Co-localization of endogenous PDIA3 and PDIA6 in U2OS cells (top). The fluorescence intensity (bottom) was analysed against the yellow line (top). The merged figure shows the fluorescence intensities of mCherry–PDIA6 and PDIA3 as a solid line, which means that their respective fluorescence intensities overlap (bottom). This experiment was performed three times independently.

Source data

Extended Data Fig. 8. Selective condensations of PDI family members into PDIA6 droplets.

a, Condensation of GFP-tagged PDI family members into PDIA6 droplets. Representative images of bright-field (BF) and fluorescence microscopy of PDIA6 droplets with GFP-tagged PDI family members in the uptake assay. Three independent experiments were performed. b, Signal-enhanced images of Extended Data Fig. 8a. c, Condensation of PDI family members into PDIA6 droplets. The average refractive index (RI) inside the droplet and droplet radius were calculated by enclosing the RI image in a circle. Data were analysed for 198 particles for PDIA6 only, 217 particles for 1 µM PDIA1, 210 particles for 5 µM PDIA1, 212 particles for 10 µM PDIA1, 215 particles for 1 µM PDIA3, 220 particles for 5 µM PDIA3, 212 particles for 10 µM PDIA3, 215 particles for 1 µM PDIA4, 219 particles for 5 µM PDIA4, 212 particles for 10 µM PDIA4, 214 particles for 1 µM PDIA10, 212 particles for 5 µM PDIA10, 217 particles for 10 µM PDIA10, 217 particles for 1 µM PDIA15, 218 particles for 5 µM PDIA15 and 226 particles for 10 µM PDIA15, pooled from three independent replicates.

Source data

PDIA6 condensates as an enhanced protein quality control granule

Given that PDIA6 is known to participate in insulin production and glucose-stimulated insulin secretion in insulin-producing mouse cells^40,41, the mechanism by which PDIA6 contributes to the homeostasis of insulin, as a physiological client, was investigated in the light of the new findings in this study showing that PDIA6 forms liquid-like droplets. The PDIA6 A5 mutant monomerizes¹⁵. This mutant exhibited a reduced rate of droplet formation and an abnormal droplet shape after a 30-min incubation (Fig. 6a). To determine the effect of PDIA6 on exogenously expressed insulin secretion, we used HCT116 PDIA6-knockout (KO) cells transfected with proinsulin-Gluc⁴² and PDIA6 WT or PDIA6 A5 mutant. Luminescence measurements of the cell culture media revealed marginal insulin secretion in the PDIA6-KO cells but insulin secretion was increased by more than tenfold following the reintroduction of wild-type (WT) PDIA6, restoring it to WT levels (Extended Data Fig. 9a). In contrast, the reintroduction of PDIA6 A5, which inhibits and impairs droplet formation, partially restored insulin secretion (Extended Data Fig. 9a). When the effects of PDIA6 WT or its A5 mutant on the insulin secretion were normalized to the reintroduced PDIA6 expression level (Extended Data Fig. 9b), the PDIA6 A5 mutant was predominantly less effective in promoting insulin secretion than the WT (Fig. 6b). Considering the similar fluidity inside the droplets at steady state and stress recovery from Ca²⁺ depletion (Fig. 2), we also quantified insulin secretion following thapsigargin treatment. As in the thapsigargin untreated condition, the insulin secretagogue effect was reduced in the PDIA6 A5 mutant compared with the WT (Fig. 6b). Western blot analysis revealed that the PDIA6 A5 mutant increased the level of detergent-insoluble proinsulin, which is likely to be responsible for the observed reduction in secretion (Fig. 6c,d). We further determined the effect of Ca²⁺ depletion alone in the ER on client folding. When MIN6 cells were depleted of Ca²⁺ using thapsigargin, endogenous insulin secretion was dramatically reduced (Fig. 6e), indicating that the expression level of droplet-forming PDIA6 WT as well as the Ca²⁺ concentration in the ER are essential for insulin production.

Fig. 6. Impaired PDIA6 droplet formation inhibits insulin secretion.

a, Time course of impaired PDIA6 droplet formation. PDIA6 A5 is a dimer-deficient mutant. This experiment was performed three times independently. b, Exogenously expressed insulin secretion was assessed in HCT116 PDIA6-KO cells transfected with proinsulin-Gaussia luciferase (hINS-Gluc) along with either PDIA6 WT or the PDIA6 A5 mutant construct. The cells were treated with or without 0.3 µM thapsigargin for 1 h at 37 °C. Following thapsigargin treatment, the amount of insulin-Gluc secreted into the medium during the subsequent 2-h incubation period was quantified by measuring luminescence (n = 3 biologically independent experiments). The specific Gluc signal induced by PDIA6 WT or its A5 mutant (shown in Extended Data Fig. 9a) was normalized to the PDIA6 WT or A5 protein levels, which were quantified by western blotting (shown in Extended Data Fig. 9b). Statistical significance was assessed using a two-tailed unpaired Student’s t-test (n = 3). c, Intracellular proinsulin-Gluc levels in NP-40-soluble (left) and -insoluble (right) fractions were analysed by western blotting in three biologically independent experiments. d, The band intensities in c were quantified by densitometry. Statistical significance was assessed using one-way repeated-measures ANOVA, followed by Dunnett’s post-hoc test (n = 3). P < 0.05 was considered statistically significant; the exact P value for this comparison was P = 0.03. Each biological replicate (n) is shown in a different colour. e, MIN6 cells were treated with 1 µM thapsigargin for 16 h. The amount of insulin secreted within 6 h after thapsigargin treatment was then quantified using ELISA and is expressed as a percentage of the amount secreted by the DMSO-treated control cells. Statistical significance was assessed using a two-tailed paired Student’s t-test (n = 6 independent biological replicates). b,d,e, Data are presented as the mean ± s.d. ***P = 0.0001 and ****P < 0.0001. TG, thapsigargin.

Source data

Extended Data Fig. 9. The expression of exogenous PDIA6 in HCT116 PDIA6-KO cells.

a, The specific Gluc signal induced by PDIA6 WT or PDIA6 A5 mutant was normalized to the PDIA6 WT or A5 protein levels, which were quantified by western blotting (shown in Extended Data Fig. 9b). Statistical significance was assessed using one-way ANOVA, followed by Dunnett’s post-hoc test (indicated in red) or a two-tailed unpaired t-test (indicated in black). Data are presented as mean ± s.d. (n = 3 independent replicates). **P < 0.01, ***P < 0.001, ****P < 0.0001. b, After collecting the medium for the Gluc assay, cells were lysed for western blotting to verify the expression of exogenous PDIA6 constructs using an antibody to PDIA6. Ponceau S staining was performed as a loading control.

Source data

To elucidate the enzymatic and chaperone functions of PDIA6 condensates, we assessed the effect of PDIA6 condensates on proinsulin folding in vitro. Dye-labelled proinsulin without disulfide bonds was found inside PDIA6 droplets (Fig. 7a), implying that the oxidative folding of proinsulin can take place inside PDIA6 droplets. To further clarify the enzymatic activity of PDIA6 inside droplets, we performed experiments with 20 μM PDIA6 and 20 μM proinsulin with/without 3 mM Ca²⁺ at the phase separation point in the phase diagram (Extended Data Fig. 1c). During the early folding step (up to 60 s), the oxidative proinsulin folding in PDIA6 condensates proceeded faster than that in dispersed PDIA6 (Fig. 7b and Extended Data Fig. 10). Kinetic analysis of native proinsulin formation indicated an increase in PDIA6 activity of approximately threefold in droplets compared with dispersed PDIA6 activity (Fig. 7c,d), which resulted in acceleration of oxidative proinsulin folding inside the PDIA6 condensates.

Fig. 7. PDIA6 quality control granules promote oxidative folding and inhibit aggregation of proinsulin.

a, Condensation of fluorescently labelled proinsulin into PDIA6 droplets. In labelled proinsulin, five Cys residues (but not Cys8) were replaced with Ser and Cys8 was selectively fluorescently labelled. This experiment was performed three times independently. BF, bright field. b, High-performance liquid chromatography profiles of oxidative proinsulin folding using 20 μM PDIA6 and 20 μM proinsulin with (right) or without 3 mM Ca²⁺ (left). The oxidative folding reaction of proinsulin was quenched with 7.0 mg ml⁻¹ 2-aminoethyl methanethiosulfonate, a selective thiol functional group modification reagent, at the selected time points. Three independent experiments were performed. c, Quantification of native proinsulin, determined from three biologically independent experiments. d, The folding kinetics were determined from the formation rate of native proinsulin. The rate constants were determined from three independent experiments. e, Amyloid fibril formation of all-Cys/Ser proinsulin with or without PDIA6 droplets detected by ThT (three independent experiments); a.u., arbitrary units. d,e, Data are presented as the mean ± s.d. f,g, AFM (f) and TEM (g) images obtained with or without PDIA6 droplet. These experiments were replicated three times independently. h, Proposed model of PDIA6 condensation as a protein quality control mechanism.

Source data

Extended Data Fig. 10. PDIA6 droplets catalyse oxidative proinsulin folding.

HPLC profiles of oxidative proinsulin folding using 20 μM PDIA6 and 20 μΜ proinsulin with (red) or without 3 mM Ca²⁺ (black). The oxidative folding reaction of proinsulin was quenched with 2-aminoethyl methanethiosulfonate (7.0 mg ml⁻¹), a selective thiol functional group modification reagent, at the selected time points. Three independent experiments were performed. R and N represent reduced/denatured and natively folded proinsulin respectively.

Assuming that disulfide bonds had not yet been introduced into proinsulin in the ER, we investigated the effect of PDIA6 droplets on the misfolding pathway of proinsulin. In the absence of PDIA6 droplets, all-Cys/Ser proinsulin, which cannot form disulfide bonds, was found to be highly prone to aggregation, as monitored by ThT fluorescence. PDIA6 droplets suppressed proinsulin aggregation at pH 7.4, the physiological pH in the ER (Fig. 7e). Atomic force microscopy (AFM) and transmission electron microscopy (TEM) images clearly showed proinsulin amyloid fibrils in the absence of PDIA6 droplets (Fig. 7f,g). These results indicate that PDIA6 droplets have a chaperone function, that is, to act as quality control granules, which strongly inhibits protein aggregation and amyloid fibril formation. Ca²⁺-driven PDIA6 droplets incorporate clients to increase the local concentration of these proteins, resulting in an increase in client folding efficiency. Together, these results show that Ca²⁺-driven PDIA6 droplets exhibit greater catalytic activity for the oxidative proinsulin folding process than the dispersed state of PDIA6 and additionally inhibit proinsulin amyloid-like aggregation to ensure correct proinsulin folding (Fig. 7h).

Discussion

We report that PDIA6 undergoes Ca²⁺-induced phase separation and that the resulting condensates act as quality control granules for the correct folding of proinsulin. Although the Trx-like domains of PDIA6 are highly soluble, our results suggest that Ca²⁺ binding leads to the liquid-like condensation of Trx-like domains. Unlike the droplet formation mechanism that involves low-complexity domains⁴³, transient but specific electrostatic interactions between the first redox-active Trx-like domain and the third redox-inactive Trx-like domain of PDIA6 play a pivotal role in its phase separation. All PDI family members have at least one conserved Trx-like domain. Most of these domains accommodate a redox-active CxxC motif that catalyses the introduction or isomerization of disulfide bonds of client proteins. On the other hand, at least 13 members of the PDI family contain one or more redox-inactive Trx-like domains, and the b′ domain of PDIA1 has been extensively studied with respect to client protein recognition⁴⁴. Our findings indicate that specific interactions occur between the redox-inactive b and redox-active a⁰ domains via electrostatic forces, which endows the b domain with the ability to drive Ca²⁺-dependent phase separation. Dimeric a⁰ serves as a scaffold for constructing a multivalent intermolecular interaction network with b via Ca²⁺. This result provides important insight into the involvement of redox-inactive Trx-like domains as well as redox-active ones in the enzymatic catalysis performed by members of the PDI family.

Molecular chaperones have already been reported to regulate liquid-like condensation. For example, Hsp70 is incorporated into the nucleolus⁴⁵ and TAR DNA-binding protein droplets⁴⁶ to inhibit solid-like aggregation, and Hsp27 and Kapβ2 modulate FUS phase separations^47,48. Here we report that PDIA6 droplets can selectively recruit folding assistants, such as PDIA3, BiP and CNX (Figs. 2 and 5). Because PDIA6 binds to BiP²⁷ and PDIA3 is known to catalyse oxidative folding of glycosylated clients via the CNX cycle⁵, PDIA6 condensates not only directly promote oxidative client folding (Fig. 7) but also serve as a scaffold to assemble key folding factors in this compartmentalization, thereby facilitating efficient client folding. Providing support for a functional role for such protein quality control granules, PDIA6 puncta are present in the ER under steady-state conditions and increase in number during stress recovery after Ca²⁺ depletion. In addition, PDIA6 has been shown to inactivate activated IRE1α⁴⁹, an unfolded protein response sensor. Thus, PDIA6 may contribute to the restoration of ER homeostasis by preventing the accumulation of misfolded client proteins, which can disrupt proteostasis and overall ER function. Further studies should seek to clarify whether recruited proteins such as PDIA3, CNX and BiP are co-localized in condensates within the ER or just in the ER or ER-derived vesicles as a whole, thereby helping to elucidate the detailed mechanism of protein quality control granules in the ER.

In conclusion, our study reveals that PDIA6 can form droplets through liquid-like condensation and that such condensates can act as protein quality control granules. Our results contribute to our knowledge of a range of Ca²⁺-mediated proteostasis processes that are dependent on PDI family condensates and add PDIA6 quality control granules to the growing number of components involved in proteostasis. We anticipate that further studies on these quality control granules will help uncover the molecular origins of ER-related misfolding diseases such as type II diabetes.

Methods

Recombinant protein expression and purification

Complementary DNA encoding human PDIs (PDIA1, PDIA3, PDIA4, PDIA6, PDIA10 and PDIA15) without the N-terminal signal sequence were subcloned and inserted into the NdeI and BamHI sites of the pET15b vector (Novagen). The plasmids encoded a 6-His tag at the N terminus of the proteins. PDIA6 domains (a⁰, a and b) and mutants (a⁰A5, b-ND) were constructed using a PrimeSTAR mutagenesis basal kit (Takara Bio). All PDIs used in this study were overexpressed in the Escherichia coli strain BL21(DE3) and purified by ion-exchange and size-exclusion chromatography^{14–16,50,51}.

Proinsulin, subcloned and inserted into the pET17b vector (Novagen), was overexpressed in the E. coli BL21(DE3), harvested by centrifugation, solubilized from inclusion bodies and purified by reversed-phase high-performance liquid chromatography⁵².

For the isotopically labelled proteins used in the NMR studies, E. coli cells were cultured in minimal (M9) medium at 37 °C in the presence of 100 mg l⁻¹ ampicillin. ¹³C/¹⁵N-labelled protein samples used for resonance assignment were cultured in M9 medium supplemented with ¹⁵NH₄Cl (1 g l⁻¹; CIL) and ¹H–¹³C-glucose (2 g l⁻¹; ISOTEC). For the isotopic labelling of Leu and Val methyl groups, α-ketobutyric acid (50 mg l⁻¹) and α-ketoisovaleric acid (85 mg l⁻¹) were added to the medium 1 h before isopropyl β-D-1-thiogalactopyranoside was added^53–57. To label the Met and Ala methyl groups, [¹³CH₃] Met (50 mg l⁻¹) and [²H,¹³CH₃] Ala (50 mg l⁻¹) were added to the medium 1 h before the addition of isopropyl β-D-1-thiogalactopyranoside^53–57. M9 medium containing 99.9% ²H₂O was used.

Dynamic light scattering measurements

PDIA1, PDIA3, PDIA4, PDIA6, PDIA10 or PDIA15 (50 µM) were mixed with 0–3 mM CaCl₂ in 50 mM HEPES-NaOH (pH 7.2) at room temperature. Average Z-values of all of the mixtures were measured using a Zetasizer Nano instrument (Malvern Instruments).

Quantification of protein concentrations after liquid-like condensation

PDIA6 (5–100 µM) with or without 10 mM NaCl was incubated with 0–10 mM CaCl₂ in 50 mM HEPES-NaOH (pH 7.2) at room temperature for 30 min. All samples were centrifuged for 5 min at 21,500g and 25 °C, after which the concentrations of proteins in the supernatant were measured by monitoring the absorbance at 280 nm.

Refractive index measurements

PDIA6 droplet formation was initiated by adding 4 mM CaCl₂ to 50 µM PDIA6 in 50 mM HEPES-NaOH (pH 7.2) at room temperature. PDIA6 droplets were observed using a holotomography microscope equipped with a laser-induced fluorescence system (HT-2H; Tomocube). The RI and fluorescence images showing the xy cross-section of the droplet centre. The average RI inside the droplet and the droplet radii were calculated by enclosing the RI image in a circle using the TomoStudio version 3.2.8 software (Tomocube) and plotted using the Igor Pro version 6.36 software (WaveMetrics). The statistical significance of differences was examined using a one-way ANOVA with Tukey’s honestly significant difference post-hoc testing. All statistical tests were performed using the KaleidaGraph version 4.5.1 software (Synergy Software) at a significance level of α = 0.05.

Ca²⁺ concentration outside droplets

PDIA6 (50 µM) was incubated with 0.5 or 4 mM CaCl₂ in 50 mM HEPES-NaOH (pH 7.2) at room temperature for 30 min. The solution was centrifuged at 1,400g and 25 °C for 30 min. The Ca²⁺ concentration in the supernatant was measured by monitoring the absorbance at 612 nm using a QuantiChromTM calcium assay kit (DICA-500; BioAssay Systems).

Residual protein concentration determination

First, 50 µM PDIA6 a⁰, a and b domains were incubated with 4 mM CaCl₂ in 50 mM HEPES-NaOH (pH 7.2) at 25 °C. After 10 min, 0, 5, 10, 25, 50, 62.5, 75 or 100 µM PDIA6 a⁰, a, and b domains were added and the solution was incubated at 25 °C for 30 min. All samples were centrifuged at 21,500g and 25 °C for 5 min. The total concentration of proteins in the supernatant was measured using a BCA kit (Nakarai Tesque). For the quantification of each PDIA6 domain in the supernatant, all supernatants were separated by non-reducing SDS‒PAGE. The gel images were captured using a ChemiDoc Touch imaging system (Bio-Rad) and the intensity of each band was quantified using ImageJ/Fiji.

Condensation of PDI family members and proinsulin

Droplet formation was performed by mixing 50 µM PDIA6 and 4 mM CaCl₂ in 50 mM HEPES-NaOH (pH 7.2) at room temperature. After 1 min, 1, 5 or 10 µM PDIA1, PDIA3, PDIA4, PDIA10 or PDIA15 was added to the PDIA6 droplet solution. The RI inside the PDIA6 droplets was monitored using a holotomography microscope.

For proinsulin, 50 µM PDIA6 was mixed with 4 mM CaCl₂ in 50 mM HEPES-NaOH (pH 7.2) at room temperature. After 1 min, DY-605-maleimide-modified proinsulin was added to the PDIA6 droplet solution. Bright-field and fluorescence images were obtained using a holotomographic microscope equipped with a laser-induced fluorescence system. The DY-605-maleimide-modification method was the same as that used for ATTO-532-maleimide-modification described previously¹⁴.

Size analysis of liquid-like condensates using microscopy

Droplet formation was triggered by adding 4 mM CaCl₂ to PDIA6 solutions containing 50 µM PDIA6 in 50 mM HEPES-NaOH (pH 7.2) at room temperature with or without 10 mM NaCl, and PDIA6 droplets were observed within 30 min using a confocal microscope (FV1000, Olympus). For the pH-change experiments, 50 µM PDIA6 and 4 mM CaCl₂ were mixed in 50 mM HEPES-NaOH at different pH values (pH 6.8–8.0) and PDIA6 droplets were observed after 10 min using a confocal microscope (LSM 800, Carl Zeiss). To observe impaired PDIA6 droplet formation, 50 µM PDIA6 WT and A5 were mixed with 4 mM CaCl₂ in HEPES-NaOH (pH 7.2) at room temperature and the reaction mixtures were monitored by a confocal microscope (LSM 800) for 30 min. Using ImageJ/Fiji⁵⁸, the number, radius and area of each droplet in the field of view were analysed, assuming that the droplets were perfect spheres. The area occupancy rate was calculated by summing the ratios of the area of each droplet to the total area of the field of view.

Fluorescence recovery after photobleaching

Droplet formation was triggered by adding 4 mM CaCl₂ to PDIA6 solutions containing 50 µM PDIA6, 4 µM mCherry–PDIA6 and 50 mM HEPES-NaOH (pH 7.2) at room temperature, and PDIA6 droplets were observed with the 559 nm laser line of a confocal microscope (FV1000, Olympus). A specific spot in the PDIA6 droplet was bleached at 80% transmission for 0.1 s and time-lapse images before and after photobleaching were collected (0.5 s frame rate, 300 frames). The fluorescence intensity of the region of interest was then calculated using the FV10-ASW software (Olympus). Images were processed using the ImageJ/Fiji⁵⁸ and GNU Image Manipulation Program software (http://gimp.org). The fluorescence intensity before photobleaching was set to 100%, that immediately after photobleaching was set to 0%, and half the fluorescence recovery was calculated from the normalized fluorescence intensity using Microsoft Excel based on previous studies^59–61.

Cell cultures

U2OS (ATCC, catalogue number HTB-96), HCT116 (ATCC, catalogue number CCL-247) and MIN6 cells (provided by J. Miyazaki, The University of Osaka) cells were cultured at 37 °C with 5% CO₂ in Dulbecco’s modified Eagle’s medium (DMEM) supplemented with 10% fetal bovine serum (FBS). The U2OS and HCT116 cell lines were authenticated by the ATCC using short-tandem-repeat profiling. The MIN6 cell line was originally established and authenticated by J. Miyazaki and has been widely used in the field but was not independently authenticated by short-tandem-repeat profiling in this study. All cell lines used in this study were tested for mycoplasma contamination by PCR-based assays and found to be negative. U2OS/TR_mCherry–PDIA6 cells were generated using a Flp-In System (Thermo Fisher Scientific). First, we generated the doxycycline-induced U2OS/FRT stable cell Line U2OS/FRT/TR, which expresses the tetracycline repressor (TR) from pcDNA6/TR-IRES-puro⁶². The pcDNA6/TR-IRES-puro plasmids were transfected into U2OS/FRT cells using Lipofectamine 3000 (Thermo Fisher Scientific), the cells⁶³ were selected with 2 μg ml⁻¹ puromycin and single clones of U2OS/FRT/TR cells were isolated. U2OS/TR_mCherry–PDIA6 cells were prepared using U2OS/FRT/TR isolated by the Flp-In System and cultured with 5% CO₂ in DMEM containing 10% FBS supplemented with 2 μg ml⁻¹ puromycin and 200 μg ml⁻¹ hygromycin B. Doxycycline-induced expression of mCherry–PDIA6 was attained through culture in a medium supplemented with 15 ng ml⁻¹ doxycycline (Wako) for 24 h. The ER Ca²⁺ transporter inhibition experiments were performed in culture medium supplemented with thapsigargin (0.3 μM; Cayman Chemical) and the Ca²⁺ ionophore A23187 (3 μM; Sigma-Aldrich) for 1 h. After release from the ER stress inducer, the cells were rinsed with PBS and cultured for 0, 2 or 4 h.

Immunofluorescence

Cells were seeded onto round 12-mm-diameter coverslips (Matsunami) and fixed with 4% paraformaldehyde in PBS for 10 min at room temperature. The fixed cells were permeabilized with 0.5% Triton X-100 in PBS for 15 min, rinsed and then blocked with 1% BSA in PBS containing 0.1% Tween-20 (PBST) for 1 h at room temperature. The slides were incubated overnight at 4 °C with primary antibodies (diluted in PBST containing 1% BSA). Primary antibodies to PDIA6 (1:1,000; Proteintech, 18233-1-AP), CNX (1:1,000; MBL, M178-3), mCherry (1:1,000; Proteintech, 26765-1-AP; 1:1,000; Proteintech, 68088-1-Ig), BiP (1:1,000; Abcam, ab21685) and PDIA3 (1:1,000; Proteintech, 15967-1-AP) were used. Unbound antibodies were removed by three 10-min washes with PBST. The slides were then incubated with goat anti-mouse Alexa Fluor Plus 488 (1:1,000; Thermo Fisher Scientific, A32723; Thermo Fisher Scientific, A11034) or goat anti-rabbit Alexa Fluor Plus 594 (1:1,000; Thermo Fisher Scientific, A32740; Thermo Fisher Scientific, A11032) for 1 h at room temperature, washed and mounted with Fluoro-KEEPER antifade reagent (Nacalai Tesque). Immunostained cells were examined using a confocal laser scanning microscope (FV1000D, Olympus or Zeiss LSM 800). Each data series was processed with fixed parameters to enable comparison of the signal intensities.

Measurement of Ca²⁺ concentrations of the ER

The cells were seeded onto eight-well chamber slides (ibidi), rinsed with PBS and then incubated for 1 h with 6 μM Mag-Fluo4, AM (Thermo Fisher Scientific), which was prepared in FluoroBrite DMEM medium (Thermo Fisher Scientific) supplemented with L-Glu (Nacalai Tesque), 0.04% Pluronic F-127 (BIOTIUM) and 1.25 mM probenecid (TCI). The Mag-Fluo4 and AM were removed by three washes with PBS. The cells were then incubated with FluoroBrite DMEM supplemented with 4 mM L-Glu, 10% FBS and 1.25 mM probenecid. Following treatment with ER stress inducers and release, the cells were examined using a confocal laser scanning microscope (FV1000D, Olympus).

CRISPR–Cas9 method to generate KO cell lines

To construct PDIA6-KO HCT116 cells, using px330-PuroR⁶⁴ as a template, the DNA oligonucleotides 5′-CGGTGTTTCGTCCTTTCCACAAGATATATA-3′ and 5′-AAGGACGAAACACCGACTTCTCGGTTGAAATTCGAGTTTTAGAGCTAGAAATAGCAAG-3′ or 5′- AAGGACGAAACACCGAACACCATACTGACCTCCTAGTTTTAGAGCTAGAAATAGCAAG-3′ were utilized for inverse PCR to express guide RNAs targeting exons 2 and 4, respectively, of the PDIA6 gene. The PCR products were assembled by NEBuilder (New England Biolabs). HCT116 (ATCC, catalogue number CCL-247) cells were co-transfected with these two plasmids using polyethylenimine Max and screened for puromycin (0.5 μg ml⁻¹) resistance.

Quantification of insulin secretion in HCT116 cells

Exogenous insulin secretion was measured using the proinsulin-Gaussia luciferase (Gluc) system⁴². Subconfluent HCT116 PDIA6-KO cells cultured in 24-well plates were transfected with pJNC-hINS-Gluc (RDB19845, obtained from RIKEN DNA Bank) and pcDNA3.1 WT PDIA6 or its variant using Polyethylenimine (PEI) max (Polysciences Inc.). Two days post transfection, the cells were treated with 0.3 µM thapsigargin for 1 h at 37 °C. Following thapsigargin washout, 500 µl cell culture medium (DMEM supplemented with 10% FBS) was added to each well and the cells were incubated for 2 h at 37 °C. The medium was then collected and centrifuged at 775g for 2 min to remove cellular debris. The supernatant containing secreted insulin-Gluc was transferred to 1.5 ml tubes. Quantification of secreted insulin-Gluc was performed by measuring the luminescence activity of Gluc. For this, 10 µl supernatant was transferred to white 96-well plates (Thermo Scientific, 136101) and 50 µl PBS containing 5 ng µl⁻¹ coelenterazine (CAS 55779-48-1, Santa Cruz Biotechnology, sc-205904) was added to each well. Immediately after adding the coelenterazine, luminescence was measured for 0.7 s using a Nivo multimode plate reader equipped with a dispenser (PerkinElmer).

Western blotting

HCT116 PDIA6-KO cells transfected with Proinsulin-Gluc and either PDIA6 WT or A5 mutant constructs were solubilized in a RIPA buffer supplemented with 1 mM phenylmethylsulfonyl fluoride (FUJIFILM Wako Pure Chemical Corporation), 5 µg ml⁻¹ leupeptin (FUJIFILM Wako Pure Chemical Corporation), and 5 µg ml⁻¹ pepstatin A (Peptide Institute, Inc.) two days post transfection. The cell lysates were subjected to western blot analysis as previously described⁶⁵.

To assess the solubility of proinsulin, HCT116 PDIA6-KO cells transfected with Proinsulin-Gluc and either PDIA6 WT or A5 mutant constructs were lysed in NP-40 buffer (1% NP-40, 50 mM Tris–HCl pH 8.0, 150 mM NaCl, 10 mM N-ethylmaleimide, 20 µM MG-132, 5 µg ml⁻¹ leupeptin, 5 µg ml⁻¹ pepstatin A and 1 mM phenylmethylsulfonyl fluoride). The NP-40-insoluble pellet was subsequently resuspended in 1% SDS buffer (1% SDS, 50 mM Tris–HCl pH 8.0, 150 mM NaCl, 10 mM N-ethylmaleimide, 20 µM MG-132, 5 µg ml⁻¹ leupeptin, 5 µg ml⁻¹ pepstatin A and 1 mM phenylmethylsulfonyl fluoride) and incubated at room temperature for 10 min. The mixture was homogenized by passing it through a 27 G needle, followed by centrifugation for 10 min at 12,396g and room temperature. The resulting supernatant was collected as the NP-40-insoluble fraction.

Equal amounts of proteins from cell lysates were subjected to SDS–PAGE and the gels were transferred onto nitrocellulose membranes using the Mini Trans-blot cell system (Bio-Rad). Following this, the membranes were blocked with 5% skim milk or 5% BSA and incubated with anti-PDIA6 (18233-AP, Proteintech) or anti-insulin (L6B10, 8138, Cell Signaling Technology) and horseradish peroxidase-conjugated anti-rabbit IgG (711-035-152, Jackson Immuno Research). Subsequently, the membranes were subjected to three washes with 0.1% Tween in PBS. Proteins on the western blots were visualized using ImmunoStar Zeta (FUJIFILM Wako Pure Chemical Corporation) or SuperSignal West Pico PLUS Chemiluminescent Substrate (Thermo Fisher Scientific) and analysed using the FUSION Chemiluminescence Imaging System (Vilber Bio Imaging).

FRAP of PDIA6 foci in cells

Cells were cultured on eight-well chamber slides (ibidi). FRAP experiments were performed on a Zeiss LSM 800 confocal laser scanning microscope equipped with a ×40, 1.3 numerical aperture oil-immersion objective. PDIA6 foci were bleached with 100% laser intensity at 561 nm. The fluorescence intensity was recorded for ten frames before bleaching and the recovery of fluorescence was recorded for 60 frames at 1 frame s⁻¹. The mean fluorescence intensity within the region of interest was calculated using the ZEN blue software (Zeiss) and half the fluorescence recovery was calculated from the normalized fluorescence intensity using Microsoft Excel.

NMR measurements

For the NMR experiments, isotopically labelled full-length PDIA6 was prepared at a concentration of 50 µM in 50 mM HEPES (pH 7.2) and 10% ²H₂O in the absence and presence of 4 mM CaCl₂. To obtain NMR spectra of isotopically labelled PDIA6 domains (a⁰, a and b), the protein was prepared at a concentration of 100 µM in 50 mM HEPES (pH 7.2), 200 mM NaCl and 10% ²H₂O in the absence and presence of 20 mM CaCl₂. For domain titration, 100 µM isotopically labelled a⁰ or b in the presence of 20 mM CaCl₂ was mixed with 200 µM unlabelled b or a⁰ in the same buffer solution.

NMR spectra were recorded on a Bruker AVANCE NEO 800 MHz NMR spectrometer equipped with a cryogenic probe (CPTCI(F)), Bruker AVANCE III HD 600 MHz NMR instrument equipped with a TBI probe, Bruker AVANCE III HD 500 MHz NMR instrument equipped with a BBO probe and Agilent UNITY INOVA 600 MHz NMR instrument equipped with a TR5 probe. All NMR spectra were processed using the NMRPipe program⁶⁶ and TopSpin, and data analysis was performed using Olivia (https://github.com/yokochi47/Olivia), SPARKY or POKY⁶⁷.

Resonance assignments for the b domain were performed by standard spectral sets using ¹³C/¹⁵N-labelled samples. Resonance assignments for a⁰ and a domains were performed using band-selective optimized flip angle short transient (SOFAST) HMQC-nuclear Overhauser effect spectroscopy (NOESY)⁶⁸, and by referring to BMRB data (11110 and 11116) and a previous study ¹⁵ using [U-¹⁵N; Ala-¹³CH₃; Met-¹³CH₃; Ile-δ1-¹³CH₃; Leu, Val-¹³CH₃/¹³CH₃]-labelled samples. The amide signals for each domain were assigned except for Ser35 and Ser50 in the a⁰ domain; the a⁰–a linker region (Gly143–Val164); Asp178, Gly191 and Asp253 in the a domain; and Ala396, Trp412, Asp413 and Gly414 in the b domain. All methyl signals of each domain were assigned. Except for Ala361, Ala396 and the single-sided methyl groups at Leu45 and Leu427, all the full-length methyl groups were assigned by the SOFAST-HMQC-NOESY spectral set according to each domain assignment.

Native-MS

ESI–MS and ESI–IM–MS experiments were performed on a SYNAPT G2 HDMS (Waters) instrument equipped with a nano ESI source with travelling wave ion mobility. The samples were deposited in borosilicate capillaries with an inner diameter of approximately 2 µm (prepared in-house), and a 0.127-mm-diameter Pt wire (Sigma-Aldrich) was placed in the capillary. The experimental parameters for ESI–MS were as follows: capillary voltage, 0.6 kV; source temperature, 70 °C; sampling cone voltage, 10 V; trap collision energy, 20 V; argon gas flow rate, 3.0 ml min⁻¹; and backing pressure in the source region, 5.5–5.6 mbar. For ESI–IM–MS, the capillary voltage was set to 0.7 kV, the ion mobility wave velocity was set to 800 m s⁻¹, the ion mobility wave height was set to 38 V and the gas pressure of the ion mobility cell was 2.94–2.92 mbar.

Detection of protein uptake into PDIA6 droplets using fluorescence microscopy

Droplet formation was triggered by the addition of 4 mM CaCl₂ to PDIA6 solutions containing 50 µM PDIA6, 5 µM GFP or GFP-fused PDI family proteins and 50 mM HEPES-NaOH (pH 7.2) at room temperature; the GFP or GFP-fused PDI family members in PDIA6 droplets were observed with the 473 nm laser line of a confocal microscope (FV1200, Olympus). Confocal fluorescence images of the proteins in the PDIA6 droplets were obtained 30–40 min after droplet formation. The images were processed and fluorescence intensity was extracted using ImageJ/Fiji⁵⁸. The fluorescence intensity of GFP–PDIA6 in PDIA6 droplets was set to 100% and the relative fluorescence intensity was calculated using Microsoft Excel.

Insulin ELISA

The insulin secreted from MIN6 cells (provided by J. Miyazaki, The University of Osaka) was quantified using a mouse insulin ELISA kit (Mercodia, 10-1247-01) according to the manufacturer’s protocol. Briefly, MIN6 cells cultured on 24-well plates⁶⁹ were treated with 1 µM thapsigargin (Sigma-Aldrich, T9033) for 16–24 h. The cells were then washed three times with serum-free DMEM and incubated in 0.5 ml serum-free DMEM for 6 h at 37 °C. The cell culture medium was collected and used for ELISA after brief centrifugation to remove detached cells.

Isothermal titration calorimetry

To determine the binding affinity of Ca²⁺ for PDIA6, ITC experiments were carried out with a VP-ITC instrument (Malvern Panalytical). PDIA6 and CaCl₂ were dissolved in 50 mM HEPES (pH 7.2) containing 0 or 100 mM NaCl. The solutions were degassed for 3 min before being loaded into the ITC instrument. The concentrations of PDIA6 in the cell and CaCl₂ in the syringe were 50 μM and 10 mM, respectively. Titration experiments consisting of a total of 25 injections were performed. The injection volume was 2 μl for the first injection to minimize the effects of bubbles and 10 μl for the remaining injections. The temperature and stirring speed were maintained at 25 °C and 307 rpm, respectively. An initial delay of 600 s was applied and the reference power was set to 10 μCal s⁻¹. The CaCl₂ solution was titrated into the same buffer for the heat of dilution. ITC thermograms and binding isotherms were created after the dilution heat was subtracted. The binding isotherms were fitted using the one-set-of-site binding model in the MicroCal PEAQ-ITC analysis software version 1.41 (Malvern Panalytical).

ThT fluorescence assay

The kinetics of amyloid formation in proinsulin were monitored using a ThT fluorescence assay at 37 °C. To prepare the stock solution, proinsulin powder was dissolved in 50 mM carbonate bicarbonate buffer (pH 10) with 1 M urea and then centrifuged at 15,000g and 25 °C to remove insoluble material. The concentration of proinsulin was determined by ultraviolet light absorbance at 280 nm with a molar extinction coefficient of 5,560 M⁻¹ cm⁻¹ and adjusted to a final stock concentration of 60 μM. Subsequently, 3 μM proinsulin solutions in the absence or presence of 50 μM PDIA6 were prepared in 20 mM HEPES (pH 7.2) containing 4 mM CaCl₂ and 5 μM ThT. The ThT fluorescence assay was conducted using a Synergy H1 microplate reader (BioTek Instruments). We loaded each sample (150 μl per well) in triplicate into a 96-well full-area plate (Corning) and affixed a sealing film (PowerSeal Cristal View, Greiner-Bio-One) to prevent sample evaporation. All samples were continuously shaken at 807 cycles per minute with 3 mm stainless steel beads. The fluorescence intensity of ThT was recorded from the top of the plate at excitation and emission wavelengths of 445 and 490 nm, respectively.

The kinetic parameters of amyloid formation were determined by fitting the ThT emission curves using

$$Y=y_i+m_{\text{i}}t+\frac{y_{\text{f}}+m_{\text{f}}t}{1+\exp \left[-k\left(t-t_0\right)\right]}$$

where y_i + m_it and y_f + m_ft represent the initial and final baseline values, respectively^70–73, t₀ denotes the half-life at which the ThT fluorescence intensity reaches 50% of the maximum amplitude and k represents the elongation rate constant. We calculated the lag time using the following relationship: lag time = t₀ − 2(1/k). The average and error values of the lag time and elongation rate constant were calculated for three separate samples in a single set.

AFM measurement

A sample solution volume of 10 μl was applied to a freshly cleaved mica plate and incubated for 10 min. Each sample was gently rinsed twice with 20 μl double-deionized water. The remaining water was carefully removed using filter paper and compressed air. An AFM Park NX10 (Park Systems Corp.) microscope was used to obtain AFM images.

TEM imaging

Glow-discharged grids (collodion-coated copper TEM grids; Nisshin EM Co.) were treated with 5 μl of the resultant samples for 1 min. Excess sample was removed with filter paper. Next, 5 μl of 2% (wt/vol) uranyl acetate solution was added to the grids for staining. After 30 s, the remaining solution was blotted dry using filter paper. The grids were then left to dry in air at room temperature. A JEM-1400 Plus TEM (JEOL Ltd) microscope operated at 120 kV was used to obtain TEM images.

Proinsulin folding assay

To estimate the activity of PDIA6 inside droplets, we performed experiments with 20 μM PDIA6 and 20 μM proinsulin in 50 mM HEPES-NaOH (pH 7.2) containing 0.5 mM oxidized form of glutathione and 0.1 mM oxidized form of glutathione with or without 3 mM CaCl₂ at 25 °C. The oxidative folding reaction of proinsulin was quenched with 7.0 mg ml⁻¹ 2-aminoethyl methanethiosulfonate, a selective thiol functional group modification reagent, at the selected time points. Subsequently, the folding intermediates of proinsulin were analysed by reversed-phase high-performance liquid chromatography and matrix-assisted laser desorption/ionization (MALDI) with a time-of-flight MS without centrifugation^52,74. Folding kinetics were determined from the formation rate of native proinsulin. The rate constants were calculated by single exponential curve fitting using the IGOR Pro6 software. The experiments were independently repeated at least three independent experiments (mean ± s.d.).

Statistics and reproducibility

No statistical methods were used to pre-determine sample sizes. Sample sizes were chosen on the basis of previous studies and are similar to those generally employed in the field. All experiments were repeated independently at least three times with similar results, unless otherwise stated in the figure legends. No data were excluded from the analyses. Data distribution was assumed to be normal but this was not formally tested.

Statistical analyses were performed using KaleidaGraph version 10.4.2 (Synergy Software) and GraphPad Prism version 4.5.1 (GraphPad Software). For comparisons between two groups, two-tailed unpaired or paired Student’s t-tests were used, as indicated in the figure legends. For comparisons among multiple groups, one-way ANOVA, followed by Tukey’s honestly significant difference post-hoc test was applied. For repeated-measures designs, one-way repeated-measures ANOVA with Dunnett’s post-hoc test were used. The significance level was set at α = 0.05 and exact P values are reported where available. Data are presented as the mean ± s.d. unless otherwise stated. Box plots show the median (centre line), the 25th and 75th percentiles (bounds of the box) and the minimum and maximum values (whiskers).

Data collection and analysis were not randomized. The investigators were not blinded to allocation during experiments and outcome assessment.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Online content

Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at 10.1038/s41556-025-01794-8.

Author contributions

Conceptualization: Y.-H.L., T.S. and M.O. Methodology: K.A., T.O., E.M., S.N., H.A., and S.A. Investigation: M.M., Y.H. and H.K. (NMR); S. Kanemura, M.W., S. Kajimoto, T.N., K.S., K. Ishii, K.B., T.K. and A.T. (HT and biochemical assays); M.H. (PDI family purification); Y. Lin, Y. Li, H.Y. and Y.-H.L. (fluorescence assay, ITC, TEM and AFM); T. Mannen, M.W. and K. Iuchi (cultured cell assay); T.K. (oxidative folding); S.O. and Y.K. (insulin secretion assay) and M.T. (native-MS); M.O. (first discovery of PDIA6 droplet). Funding acquisition: Y.-H.L., T.S., T.M. and M.O. Project administration: Y.-H.L., T.S. and M.O. Supervision: Y.-H.L., T.S. and M.O. Writing (original draft): M.O. Writing (review and editing): Y.-H.L., T.S., M.V., K. Inaba and M.O.

Peer review

Peer review information

Nature Cell Biology thanks the anonymous reviewers for their contribution to the peer review of this work.

Data availability

Source data are provided with this paper. All other data supporting the findings of this study are available from the corresponding author on reasonable request.

Competing interests

E.M. is a CEO of Molmir, Inc. The remaining authors declare no competing interests. A patent (Japanese Patent number NO. 7194403; PDIA6 droplet formation method) has been registered by M.O., M.M., S. Kanemura, T.S. and K. Inaba.

Extended data

is available for this paper at 10.1038/s41556-025-01794-8.

Supplementary information

The online version contains supplementary material available at 10.1038/s41556-025-01794-8.