Disordered regions in the IRE1α ER lumenal domain mediate its stress-induced clustering

To investigate whether membrane association influences IRE1α LD clustering, we reconstituted the system in vitro using purified human IRE1α LD tethered to supported lipid bilayers (SLBs). SLBs are constituted of planar membranes formed on solid surfaces which are widely used as membrane-mimics (Fig. 1A,B).

We reconstituted SLBs composed primarily of 1-palmitoyl-2-oleoyl-glycero-3-phosphocholine (98.92 mol% POPC). We used 1 mol% nickel–nitrilotriacetic acid (Ni-NTA) lipids to tether mCherry-IRE1α LD-10His to SLBs through its C-terminal 10His tag, which allows placing the entire LD of IRE1α (aa 24-443) in the topologically correct orientation (Fig. 1B). To monitor SLB integrity and fluidity, we used 0.08 mol% Atto488 labeled 1,2-Dipalmitoyl-sn-glycero-3-phosphoethanolamine (DPPE). Fluorescence recovery after photobleaching (FRAP) experiments of mCherry-IRE1α LD-10His demonstrated that mCherry-IRE1α LD-10His diffused on the SLB surface (Figs. 1C, left and EV1A; Appendix Table S1). Similarly, FRAP of Atto488 labeled DPPE lipids confirmed that the membrane was fluid (Fig. EV1A; Appendix Table S1).

To mimic the crowding of the ER environment in our in vitro assays, we used the molecular-crowding agent polyethylene glycol 8000 (PEG) (Kuznetsova et al, 2014). We monitored the behavior of mCherry-IRE1α LD-10His on the membrane surface via total internal reflection fluorescence (TIRF) microscopy and FRAP experiments at various PEG concentrations (Figs. 1C, right, F and EV1B,C). We calculated the diffusion coefficient from fluorescence recovery half-life times as previously described (Axelrod et al, 1976)^,(Soumpasis, 1983). Interestingly, increasing the PEG concentration in solution gradually decreased the mobile fraction and diffusion rates of membrane-bound mCherry-IRE1α LD-10His from 0.18 µm²/s without PEG to 0.02 µm²/s in presence of 11% (w/v) PEG (Fig. EV1A,B). In the presence of 10% PEG, mCherry-IRE1α LD-10His displayed a diffuse fluorescence signal (Fig. 1F), while 11% PEG induced the formation of large mCherry-IRE1α LD-10His clusters on the SLB (Fig. 1C, right, F; Movie EV1; Appendix Table S2). In the presence of 11% PEG, both mCherry-10His control and Atto488-labeled DPPE retained their dynamic behavior confirming that the integrity of the SLB was not compromised and clustering is specific to IRE1α LD (Fig. EV1C–F; Appendix Table S2). Under those conditions, mCherry-IRE1α LD-10His clusters formed and fused over time (Fig. 1D; Movie EV2). Yet, FRAP experiments showed that photobleached IRE1α LD clusters did not recover even after 300 s (Figs. 1E and EV1B,C). Instead, we observed a slight increase in mCherry-IRE1α LD-10His fluorescence at the periphery of the clusters (Fig. 1E) indicating that membrane-tethered IRE1α LD assembles into stable clusters driven by molecular crowding. To test whether membrane-tethered IRE1α LD clusters are not just aggregates, we performed wash-out experiments in which we removed the crowding agent from the well. Removal of PEG led to the disappearance of IRE1α LD clusters back to a diffuse fluorescence signal (Fig. EV1G). Importantly, clusters could reform by adding 11% PEG, indicating that they are dynamic and reversible. Altogether, we found that IRE1α LD forms stable but reversible clusters on synthetic membranes. Notably, our data are in agreement with the FRAP experiments performed with IRE1α in cells indicating that IRE1α LD reconstituted on membranes recapitulates the physical properties of IRE1α assemblies in cells (Belyy et al, 2020).

IRE1α’s LD binds unfolded peptides that are enriched in arginine, aromatic and hydrophobic residues as a means of recognizing aberrant protein conformations (Gardner and Walter, 2011; Karagoz et al, 2017). We next tested whether binding of model unfolded polypeptides would enhance IRE1α LD clustering on SLBs. We used peptides that we had previously shown to interact with IRE1α LD (Karagoz et al, 2017). The binding peptides with the highest affinity were derived from Myelin Protein Zero (MPZ) referred to as MPZ derivatives. MPZ1N is a 12mer peptide with a single binding site for IRE1α LD and binds IRE1α LD with an approximate affinity of 20 µM (Fig. EV1H,I). MPZ1N-2X consists of two MPZ1N 12mers arranged in tandem, and it binds IRE1α LD with 2 µM affinity due to avidity (Karagoz et al, 2017) (Fig. EV1I). As a control, we mutated arginine residues in MPZ1N-2X to impair its interaction with IRE1α LD, yielding MPZ1N-2X-RD (Karagoz et al, 2017). Using fluorescence anisotropy experiments, we confirmed the MPZ1N-2X-RD interaction with IRE1α LD is largely impaired (Fig. EV1I).

In the stressed ER, IRE1α LD clustering may be initiated by specific interactions of IRE1α LD with un/misfolded proteins, and a bulk increase in molecular crowding due to blocked secretion of un/misfolded proteins. Therefore, we next tested whether IRE1α LD’s interactions with model unfolded polypeptides would decrease the threshold for its clustering in the presence of a crowding agent. We found that incubation with peptides reduced the effective concentration of PEG required to drive the clustering of mCherry-IRE1α LD-10His (Figs. 1F–H and EV1J). This increased propensity was specific, as incubation of mCherry-IRE1α LD-10His with the mutant peptide MPZ1N-2X-RD did not impact its clustering (Fig. 1I). Importantly, FRAP experiments revealed that the peptides did not impair SLB integrity (Figs. 1F–I and EV1K,L; Appendix Fig. 1 and Appendix Table S2). In sum, we succeeded in reconstituting ligand-enhanced IRE1α LD clustering on synthetic membranes from minimal components, thus recapitulating a critical step of the UPR.

Our data suggested that IRE1α LD tethered to synthetic membranes forms stable clusters due to restricted conformational freedom on planar surfaces. This model predicts that IRE1α LD clusters formed in solution should exhibit a more dynamic behavior when compared to those formed on SLBs. To test this prediction, we monitored IRE1α LD clustering in solution by differential interference contrast (DIC) microscopy. The systematic analyses of protein concentrations and buffer conditions by DIC showed that in the presence of 6% PEG, 12.5 µM IRE1α LD formed droplets in solution. IRE1α LD droplets resembled biomolecular condensates formed through liquid-liquid phase separation (LLPS). Both the number and size of the condensates increased at higher protein concentrations (Figs. 2A and EV2A,B). IRE1α LD formed condensates in the presence of other crowding agents (17% Ficoll400) suggesting that molecular crowding induces LLPS of IRE1α LD (Fig. EV2C). IRE1α LD condensates displayed dynamic and liquid-like behavior in solution, as evidenced by fusion events (Fig. 2A,B; Movie EV3). FRAP experiments confirmed the liquid-like nature of IRE1α LD condensates and revealed that IRE1α LD molecules exchanged in and out of the condensates with a mobile fraction of 82% (t_1/2 = 169 s); Fig. 2C; Movie EV2). It has been observed that if a protein goes through LLPS, the liquid droplets will wet the glass surface, whereas hydrogels or less dynamic condensates do not wet solid surfaces or change shape (Wang et al, 2019). IRE1α LD condensates wetted the bottom of the glass surface in a time-dependent manner, confirming their liquid-like properties (Fig. EV2D). Neither IRE1α LD^D123P mutant, which is impaired in dimerization (Zhou et al, 2006), nor the mCherry control formed condensates (Fig. EV2E). This suggested that D123 is required for the formation of larger IRE1α LD clusters in solution. Altogether, our data revealed that in solution IRE1α LD forms dynamic condensates upon molecular crowding. These data suggested that tethering IRE1α LD to membranes leads to the stabilization of IRE1α LD assemblies. It is plausible that the restriction of IRE1α LD’s degree of freedom, or membrane-induced structural rearrangements stabilize interfaces, which are crucial for its clustering. Consequently, this could drive the formation of long-lived IRE1α LD assemblies on membranes.

We next characterized the impact of unfolded polypeptides on the formation and dynamics of IRE1α LD condensates. Fluorescein labeled MPZ1N-2X efficiently partitioned into preformed IRE1α LD condensates, revealing that they recruit client proteins (Fig. EV2F). Instead, the Fluorescein-MPZ1N-2X-RD control peptide was not enriched in the condensates (Fig. EV2G). In an experimental condition where IRE1α LD barely formed condensates (Fig. 2D, left panel), its incubation with stoichiometric amounts of model unfolded peptides led to the formation of large condensates (Figs. 2D and EV2H). Instead, the control peptide MPZ1N-2X-RD did not impact IRE1α LD phase separation (Figs. 2D, right panel and EV2H). Incubation with MPZ1N did not lead to LLPS of IRE1α LD^D123P indicating that dimer formation is necessary for peptide induced LLPS (Fig. EV2I). Importantly, model unfolded polypeptides did not undergo phase separation under those conditions (Fig. EV2J,K). This data indicated that specific interactions with unfolded polypeptides facilitate IRE1α LD phase separation. We next assessed whether unfolded polypeptide-binding would impact the dynamics of IRE1α LD assemblies within condensates. FRAP experiments showed that, while MPZ1N did not significantly affect the half-time recovery of IRE1α LD after photobleaching, binding of the MPZ1N-2X peptide led to an increase in the recovery time of IRE1α LD (Figs. 2C and EV2L,M; Appendix Table S3). These data revealed that a peptide with a single binding site shifts IRE1α LD to a conformation that favors its clustering consistent with previous findings (Karagoz et al, 2017). MPZ1N-2X, which has two binding sites, can nucleate clusters by bridging IRE1α LD molecules and further stabilize IRE1α LD assemblies. We speculate that binding of unfolded polypeptides with various stoichiometry and biochemical properties could modulate the dynamics of IRE1α assemblies, ultimately influencing UPR signaling in cells.

The generation of biomolecular condensates by IRE1α LD in solution prompted us to ask which molecular interactions might explain this behavior. The formation of biomolecular condensates is often controlled by disordered regions in proteins (Hyman and Brangwynne, 2011). IRE1α LD comprises a mostly folded N-terminal motif (aa 24–307) joined to the transmembrane helix by a disordered region (aa 307–443) (Fig. 3A; Appendix Fig. 2A,B) (Erdos and Dosztanyi, 2020). In the crystal structure of IRE1α cLD (aa 24–390, pdb: 2hz6 (Zhou et al, 2006)), several segments (i.e. (aa) 131–152, 307–358, and 369–390) are not resolved due to their flexibility (Fig. 3B; Appendix Fig. 2A,B). We refer to the disordered regions in IRE1α LD as Disordered Region 1 (DR1, aa 131–152), Disordered Region 2 (DR2, aa 307–358), Disordered Region 3, (DR3, aa 369–389), and the linker region (aa 390–443), respectively. Here, we employed molecular dynamics (MD) simulations to characterize their conformation and interaction.

Atomistic MD simulations of the IRE1α cLD dimer (residues aa 29–368) revealed that DR1 and DR2 remain highly disordered during a 1 µs long simulation, not adopting any distinct secondary structures (Fig. 3C). DR2 was the most flexible part of the dimer. These data are in line with published hydrogen-deuterium exchange experiments (Amin-Wetzel et al, 2019). We then performed coarse-grained MD simulations to test whether the disordered regions might self-associate (Fig. 3D,E). We observed that DR1 did not form clusters in a 20 µs-long simulation (Appendix Fig. 2C). Instead, DR2 and the linker region readily clustered after 1 µs of simulation. The clusters were highly dynamic, and we observed reversible association of single polypeptide chains. To test the potential of heterologous associations, we simulated a system containing DR1 and DR2 and another system comprising DR2 and the linker region (Appendix Fig. 2C,D). DR2 clusters did not interact with DR1 segments, which remained free in solution, while DR2 and the linker formed well-mixed clusters. These data suggested that DR2 and the linker have the potential to form protein condensates.

We next investigated which specific interactions may drive the disordered regions to cluster. In protein condensates, the contacts formed inside a single polypeptide chain often resemble the ones formed across different polypeptide chains (Tesei et al, 2021), indicating that the same interactions promote the internal and intermolecular organization. Therefore, we computed contact maps for interactions formed within single polypeptide chains and across different polypeptide chains in the clusters. Indeed, 1-D plots derived from the contact matrices, where we summed up the contributions from all possible interactions of a single residue in the simulations, confirmed this feature. Interactions within and between DR2 and the linker region were mainly formed by the charged and aromatic residues (Asp, Lys, Phe) (Appendix Fig. 3A–D). In DR2, Asp328 and Lys349 formed the most probable contacts, suggesting an important role in cluster formation (Appendix Fig. 3A–D). The contact analysis showed that distinct regions in the disordered segments in IRE1α LD have the propensity to form low-affinity transient interactions driven by aromatic and charged residues. In summary, MD simulations revealed the biochemical potential of the disordered segments in IRE1α’s LD in driving its LLPS.

As MD simulations predicted that DR2 and the linker region form clusters in isolation, we next tested their LLPS potential through DIC microscopy with the purified constructs. We found that the core lumenal domain (cLD aa 24–389), which lacks the disordered linker region, formed condensates (Fig. 3F, right). These data revealed that the linker region is not necessary for the formation of IRE1α LD condensates. IRE1α cLD rapidly formed condensates at lower protein and PEG concentrations compared to IRE1α LD (Fig. EV2A,B vs Fig. EV3A,B). While the size of IRE1α cLD condensates was highly comparable to that of IRE1α LD, DIC microscopy revealed that IRE1α cLD (aa 24–389) formed structures resembling beads on a string suggesting that they are incapable of fusion (Figs. 3F, right and EV3C). IRE1α cLD condensates accumulated on the glass slide without wetting the surface (Fig. EV3D). FRAP experiments showed that IRE1α cLD recovered after t_1/2 = 281.5 s and displayed a 27.6% mobile fraction confirming that cLD condensates are less dynamic in comparison to IRE1α LD condensates (Fig. 3G; Appendix Table S4). In support of the low mobile fraction of IRE1α cLD shown by the FRAP data, mCherry-tagged IRE1α LD partitioned into preformed IRE1α LD condensates after 5 min, in contrast mCherry tagged-cLD failed to do so (Fig. EV3E). Altogether, our data showed that IRE1α cLD formed stable condensates, indicating that the linker segment (aa 390–443) modulates both the propensity to coalesce and IRE1α LD associations in condensates. Together with the MD simulations, these data suggest that the linker region forms transient intra- and intermolecular contacts with the disordered segments. We anticipate that the linker binding may compete for low-affinity interactions that are involved in IRE1α LD self-assembly leading to an increase in its dynamics and clustering threshold. Altogether, our data reveal that disordered segments potentially regulate a critical switch in UPR signaling.

We next screened for mutants in the disordered segments that may regulate IRE1α LD’s clustering. For these experiments, we used LLPS assays in solution to rapidly screen for mutants that impair IRE1α LD self-assembly. As IRE1α cLD could readily form condensates in solution, we introduced mutations in DR2 and DR3 of IRE1α LD. Based on the MD simulations and published work on LLPS, we chose regions enriched in hydrophobic, aromatic or charged sequences, which might form intermolecular contacts to nucleate phase separation (Choi et al, 2020). Specifically, we mutated 3- or 4-residue stretches to glycine-serine residues, which often form dynamic segments acting as spacers in biomolecular condensates (Fig. 4A).

All the mutants were biochemically stable upon purification, allowing us to study their clustering behavior by DIC microscopy. The IRE1α LD ³¹²TLPL³¹⁵ mutant formed smaller condensates with slower kinetics that failed to fuse efficiently suggesting that this region is important for condensate formation. A mutation in the ³²⁰QTDG³²³ segment did not impair phase separation (Figs. 4B and EV4A), whereas, impairing electrostatic interactions predicted by the MD simulations in ³⁴⁶LKSK³⁴⁹ largely abolished phase separation (Figs. 4B and EV4B). Yet, mutating the neighboring segment (³⁵⁰NKLN³⁵³) only slightly impacted LLPS (Fig. 4B). Complementarily, mutating a segment enriched in hydrophobic and aromatic residues (IRE1α LD ³⁵²LNYL³⁵⁵) impaired the formation of IRE1α LD condensates and introducing a mutation to the following segment ³⁵⁴YLR³⁵⁶ substantially compromised LLPS (Fig. EV4C). In contrast, the IRE1α LD ³⁷³TKML³⁷⁶ mutant in DR3 did not impact condensate formation (Fig. 4B). These data revealed that the ³⁵⁴YL³⁵⁵ region forms a hot spot for molecular interactions driving IRE1α LD clustering and both electrostatic and hydrophobic interactions contribute to LLPS of IRE1α LD.

Notably, ³⁵⁴YL³⁵⁵ resides near ³⁵⁹WLLI³⁶², whose mutation to GSGS impairs the formation of IRE1α LD oligomers (Karagoz et al, 2017; Belyy et al, 2022). DIC microscopy indicated that the IRE1α LD ³⁵⁹WLLI³⁶² mutant underwent LLPS, even under conditions in which wild type IRE1α LD barely formed condensates (Figs. 4B and EV4D). These results revealed that disrupting the canonical oligomerization interface did not hinder phase separation of IRE1α LD and, moreover, suggested that oligomers are distinct from condensates (Figs. 4B and EV4D). These results motivated us to interrogate the oligomerization behavior of the IRE1α LD ³¹²TLPL³¹⁵ and ³⁵²LNYL³⁵⁵ mutants using orthogonal methods. To this end, we performed analytical ultracentrifugation sedimentation velocity (AUC-SV) experiments to determine whether IRE1α LD mutants could form high-order oligomers. These experiments revealed that, similar to previous observations on IRE1α cLD (Karagoz et al, 2017), IRE1α LD was found in equilibrium of dimers and high-order oligomers at 25 µM. Strikingly, under those conditions, IRE1α LD ³⁵²LNYL³⁵⁵ and ³¹²TLPL³¹⁵ mutants only formed dimers. These data revealed that ³⁵²LNYL³⁵⁵ and ³¹²TLPL³¹⁵ regions are important for the formation of high-order IRE1α LD oligomers (Fig. 4C).

Tethering a protein to model membranes might impact its clustering propensity compared to in solution due to the restricted conformational freedom or due to limited diffusion to two dimensions. Therefore, we next assessed whether the ³⁵²LNYL³⁵⁵ and ³¹²TLPL³¹⁵ mutants could form clusters on SLBs. Similar to wild type IRE1α LD, we tethered mCherry-tagged ³¹²TLPL³¹⁵ and ³⁵²LNYL³⁵⁵ mutants via a 10His-tag on their C-termini to SLBs containing 1% Ni-NTA lipids. To determine the molecular mass and the oligomeric status of mCherry-IRE1α LD-10His and its mutants on SLBs, we used mass photometry (Foley et al, 2021; Steiert et al, 2022). Mass photometry determines the protein mass by measuring light scattering of single molecules (Sonn-Segev et al, 2020). We performed mass photometry analyses of mCherry-IRE1α LD-10His and its mutants at protein concentrations with optimal particle density to enable single-molecule measurements (between 30 and 100 nM). Up to 50 nM, mCherry-IRE1α LD-10His and its mutants (D123P, ³¹²TLPL³¹, ³⁵²LNYL³⁵⁵) predominantly formed monomers. At 75 nM, the mCherry-IRE1α LD-10His and ³¹²TLPL³¹, ³⁵²LNYL³⁵⁵ started forming dimers on SLBs (Fig. 5A). In contrast, the D123P mutant was mainly monomeric at this condition (Mw= 95 ± 37 kDa) (Fig. 5A). At 100 nM, the mCherry-IRE1α LD-10His and the ³¹²TLPL³¹ mutant were mainly dimeric (Mw wild Type= 135 ± 67 kDa and Mw TLPL = 157 ± 73 kDa), but showed a broad size distribution spanning higher molecular masses, suggesting that the proteins begin to form oligomers at this concentration (Fig. 5A). Notably, at 100 nM, the ³⁵²LNYL³⁵⁵ mutant showed a narrower size distribution with the predicted mass of dimers (Mw= 147 ± 57 kDa). These data were consistent with the in solution LLPS assays, which showed that the ³⁵²LNYL³⁵⁵ mutant was more impaired in condensate formation compared to the ³¹²TLPL³¹ mutant. In agreement with published work (Zhou et al, 2006), at a concentration of 100 nM in solution, mCherry-IRE1α LD-10His and all mutants were monomeric (Fig. 5B). Overall, mass photometry analyses revealed that restriction of these proteins on membranes increased their propensity to self-associate compared to the in solution experiments.

We next tested the clustering efficiency of the mutants upon molecular crowding at various protein concentrations ranging from 100 to 200 nM, including lower concentrations (100 nM) where they mainly formed dimers determined by the mass photometry analyses. In the absence of crowding agent, similar to mCherry-IRE1α LD-10His, the ³¹²TLPL³¹⁵ and ³⁵²LNYL³⁵⁵ mutants displayed diffuse signal and dynamic behavior on SLBs (Fig. EV4E–G). At 100 nM, upon molecular crowding (11% PEG), mCherry-IRE1α LD-10His and its mutants (³¹²TLPL³¹⁵ and ³⁵²LNYL³⁵⁵) started to assemble into clusters shown by TIRF imaging (Figs. 5C and EV4H–J). However, the clusters became more prominent in both intensity and number at higher protein concentrations (Figs. 5C and EV4H–J). Intensity plot analyses of the mCherry signal of the TIRF images revealed that at 150 nM, the mCherry-IRE1α LD-10His displayed a higher number of high intensity clusters compared to the ³¹²TLPL³¹⁵ and ³⁵²LNYL³⁵⁵ mutants suggesting that the wild type is more efficient in cluster formation compared to the mutants at this concentration (Fig. EV4H–J). Importantly, at 200 nM in the presence of 11% PEG, the mutants formed clusters with similar intensity as the wild type IRE1α LD (Figs. 5C and 4H–J, bottom row). Multivalent low-affinity interactions have been shown to drive LLPS of biological systems (Alberti and Dormann, 2019). Our data indicate that on model membranes in the presence of molecular crowding and high protein concentrations, IRE1α LD self-association is favored. Under those conditions, alternative low-affinity interaction networks could form and drive clustering of IRE1α LD mutants in contrast to the experiments conducted in solution. Under those conditions, the monomerization mutant D123P did not efficiently assemble into clusters, further underling that dimerization is important for building valency on membranes similar to what we observed in solution (Figs. 5D and EV4K). Altogether, our data converge on a model in which IRE1α LD dimers interact with each other in various conformations. These interactions are facilitated by the DRs driving low-affinity contacts with no fixed valence to allow the assembly of IRE1α LD into stable clusters on membranes. We propose that the transient low-affinity interactions are crucial in bringing IRE1α molecules in close proximity to drive the formation of active IRE1α assemblies.

IRE1α forms foci in cells experiencing ER stress (Li et al, 2010; Ricci et al, 2019; Belyy et al, 2020; Tran et al, 2021). To assess the role of the disordered segments in IRE1α LD clustering in cells, we established stable cell lines expressing wild type human IRE1α or IRE1α mutants with impaired (IRE1α LD ³¹²TLPL³¹⁵ and ³⁵²LNYL³⁵⁵ mutants) or enhanced (IRE1α cLD Δlinker) clustering, as determined by our in vitro assays (Figs. 3F and 4B). Moreover, we generated cell lines expressing the previously established IRE1α LD dimerization (D123P) and oligomerization (³⁵⁹WLLI³⁶²) mutants (Zhou et al, 2006; Kitai et al, 2013; Karagoz et al, 2017; Belyy et al, 2022). We introduced doxycycline-inducible transgenes encoding mNeonGreen (mNG) tagged variants of IRE1α into mouse embryonic fibroblasts (MEFs) deficient for both isoforms of IRE1 (IRE1 α^-/- and IRE1β^-/-) and monitored IRE1α clustering by fluorescence microscopy. We introduced the mNG tag into IRE1α’s cytoplasmic flexible linker (Li et al, 2010; Karagoz et al, 2017; Belyy et al, 2020). In the absence of doxycycline, cells expressed low levels of IRE1α due to the inherent leakiness of the doxycycline-inducible system (Appendix Fig. 4A–E). Under these conditions, the expression levels of IRE1α-mNG and its mutant variants were similar to the level of endogenous IRE1α observed in wild type MEFs as assessed by Western blot analysis (Appendix Fig. 4D,E). When we treated the cells with the ER stress inducing drug tunicamycin, cells carrying IRE1α-mNG showed a modest reduction in XBP1 mRNA splicing activity compared to wild type control MEFs, suggesting that the mNG-tag slightly impairs its activity (Appendix Fig. 4F).

The size of IRE1α clusters in the cell depends on the protein concentration (Li et al, 2010; Belyy et al, 2022; Gomez-Puerta et al, 2022). While in most tissues endogenous IRE1α clusters are generally too small to overcome the diffraction limit of light, IRE1α forms microscopically visible clusters observed as distinct foci when it is ectopically expressed to levels 2–20 times over endogenous protein levels (Li et al, 2010, Karagoz et al, 2017; Ricci et al, 2019; Belyy et al, 2020; Tran et al, 2021; Belyy et al, 2022; Gomez-Puerta et al, 2022). To visualize IRE1α clustering in mammalian cells with confocal microscopy, we overexpressed IRE1α in MEFs in a controlled manner using a doxycycline-inducible promoter. IRE1α-mNG expression levels increased linearly with doxycycline concentration in the range of 25–400 nM (Appendix Fig. 4C). Treatment with 400 nM doxycycline led to expression levels of IRE1α-mNG and its mutants roughly 30-fold over endogenous IRE1α in wild type MEFs (Appendix Fig. 4C,D,G). Confocal microscopy experiments in which we monitored IRE1α cluster formation indicated that at 400 nM doxycycline, but not 100 nM, cells expressing wild type IRE1α-mNG formed microscopically visible foci in a stress-dependent manner (Figs. 6A and EV5A,C). Consistent with published work, at 400 nM Dox, IRE1α-mNG showed low constitutive XBP1 mRNA splicing activity due to overexpression, which further increased upon treatment with tunicamycin (Fig. 6B) (Li et al, 2010).

Consistent with published work, the IRE1α-WLLI-mNG mutant did not form foci under ER stress conditions and displayed a diffuse signal in the ER under overexpression conditions (Fig. EV5C) (Karagoz et al, 2017; Belyy et al, 2022). Under the same conditions, the IRE1α-TLPL-mNG and IRE1α-LNYL-mNG mutants failed to form foci at comparable expression levels as the wild type IRE1α-mNG (Fig. 6A; Appendix Fig. 4A,D). These results indicate that the TLPL and LNYL mutants are defective in clustering in cells compared to wild type IRE1α-mNG. These results substantiated our findings that disordered segments in the LD are important for IRE1α assembly. In stark contrast, IRE1α cLD-mNG (Δlinker), which lacks the linker region, formed clusters constitutively in the absence of stress (Fig. 6A), or even upon induction of its expression with a lower (100 nM) doxycycline concentration (Fig. EV5A). This suggested that the clustering threshold is lower for IRE1α cLD-mNG consistent with the results we obtained in our in vitro experiments. Moreover, the clusters formed by IRE1α cLD-mNG (Δlinker) were smaller compared to the IRE1α-mNG. We speculate that the formation of highly rigid condensates by IRE1α cLD observed in vitro might lead to impaired fusion and consequently the formation of smaller clusters observed in cells. Taken together, these results corroborate the role of disordered segments in IRE1α’s LD in regulating IRE1α’s self-association into high-order assemblies.

Next, we investigated whether the mutants would impact IRE1α activity monitored by its ability to splice XBP1 mRNA. Semi-quantitative polymerase chain reaction (PCR) and quantitative real-time PCR (qRT-PCR) analyses revealed that IRE1α LNYL-mNG mutant exhibited impaired XBP1 mRNA splicing activity compared to wild type IRE1α-mNG, indicating that the interfaces formed by the LNYL region in IRE1α LD play an important role in forming active IRE1α assemblies in cells (Fig. 6B,C). Instead, the IRE1α TLPL-mNG splicing activity was slightly but consistently diminished, where the TLPL mutant did not reach the same splicing efficiency as IRE1α-mNG (Belyy et al, 2022; Gomez-Puerta et al, 2022) (Figs. 6B,C and EV5B; Dataset EV1). Those results did not depend on IRE1α-mNG expression levels as we got similar results upon treatment with 25 nM doxycycline, where IRE1α-mNG and the mutants were expressed around 3-fold over the endogenous IRE1α level (Fig. EV5B; Appendix Fig. 4C,E). These data suggest that in cells, the mutation in the interface formed by the ³¹²TLPL³¹⁵ segment can be largely compensated for by the formation of other protein interfaces, consistent with our in vitro results. Notably, IRE1α cLD-mNG, which constitutively formed foci in the absence of stress, displayed high constitutive XBP1 mRNA splicing activity independent of IRE1α-mNG expression levels (Figs. 6B,C and EV5B; Appendix Fig. 4B,D,G; Dataset EV1). Under those conditions, the dimerization mutant IRE1α-D123P-mNG was inactive regardless of its expression level (Figs. 6D and EV5D) (Kitai et al, 2013). Similarly, the IRE1α-WLLI-mNG mutant was impaired in XBP1 mRNA splicing when expressed near endogenous levels (25 nM doxycycline) supporting previous work (Karagoz et al, 2017). At higher expression levels (400 nM doxycycline), the WLLI mutant showed comparable activity to the LNYL mutant (Figs. 6D and EV5D). At these expression levels, the CHOP mRNA levels were highly similar suggesting that, under acute ER stress conditions, cells expressing IRE1α-mNG mutants did not show increased apoptotic potential through PERK hyperactivation (Fig. EV5E; Dataset EV1). Altogether, these results indicate that disordered segments in the IRE1α LD mediate the formation of signaling competent IRE1α assemblies in cells.

All constructs were constructed in pET-47 b(+) vector. hIRE1α LD mutants ³¹²TLPL³¹⁵-GSGS, ³⁵⁰NKLN³⁵³-GSGS, and ³⁵²LNYL³⁵⁵-GSGS were based on the T274C variant of hIRE1α LD (Cysteines are substituted by Alanins, Threonine aa274 was substituted by Cysteine). All other mutants were based on the WT hIRE1α LD. We could not observe any differences between hIRE1α LD WT and T274C in our assays. ³²⁰QTDG³²³-GSGS and ³⁷³TKML³⁷⁶-GSGS were constructed through site-directed mutagenesis, with subsequent blunt end ligation. For mCherry tagged proteins, an N-terminal mCherry sequence and C-terminal 10HisTag was used.

hIRE1α LD expression and purification were adapted from published protocols (Karagoz et al, 2017). In brief, Escherichia coli strain BL21DE3* RIPL was grown with the respective antibiotics in Luria Broth at 37 °C until OD₆₀₀ = 0.6–0.8. The protein expression was induced with 400 µM IPTG for hIRE1α variants without and 1 mM for variants with mCherry at 20 °C and grew overnight. Before lysis and after each purification step, 1× Roche cOmplete Protease Inhibitor Cocktail was added to the cells or fractions containing protein. Cells were harvested and lysed (50 mM HEPES pH 7.2–7.4, 400 mM NaCl, 20 mM Imidazol, 5 mM β-mercaptoethanol, 0 to 10% Glycerol) in an Avestin EmulsiFlex-C3 cell disruptor at 16,000 psi. The lysate was spun at 30,700 × g for 45 min. The supernatant was applied to a 5 ml His-TRAP column (GE Healthcare) and eluted with a gradient of 20 mM to 500 mM imidazole. The eluate was diluted with 50 mM HEPES pH 7.2–7.4 (in the absence or in the presence of 5 or 10% Glycerol, 5 mM β-mercaptoethanol) to a concentration of 50 mM NaCl, to apply it to a HiTRAP Q HP (5 ml, GE Healthcare) anion exchange column. The protein was eluted with a linear gradient from 50 mM to 1 M NaCl. To remove the His tag from hIRE1α LD without mCherry, the protein was incubated with 3C Precision protease at a ratio of 50 to 1 over night at 4 °C. The protein was loaded to a His-TRAP column before it was further purified on a Superdex 200 10/300 gel filtration column (25/50 mM HEPES pH 7.2–7.4, 150 mM NaCl, 5 mM DTT; for mCherry proteins: 25 mM HEPES pH 7.4, 150 mM KCl, 10 mM MgCl₂, 5 mM DTT). The Expasy ProtParam tool (http://web.expasy.org/protparam/) was used to determine the extinction coefficient at 280 nm to get the final protein concentration. All the expression plasmids for the proteins are available upon request.

The protocol was adapted from (Bakalar et al, 2018; Cell (Bakalar et al, 2018)).

The phase separation behavior of hIRE1α LD protein variants in presence of PEG-8000 (Sigma-Aldrich, P2139), 40% (w/w) PEG-8000 (Sigma-Aldrich, P1458) or Ficoll-400 (Sigma-Aldrich, F4375) were observed via DIC microscopy on a Zeiss Axio Observer inverted microscope. Images were acquired at room temperature with a Plan-Apochromat 63×/1.4 Oil DIC RMS objective and CoolSnapHQ2 or Hamamatsu ORCA-Flash4.0 LT+ Digital CMOS camera controlled by Visitron and Zeiss systems, respectively. Therefore, glass wells (Greiner Bio-One 96 Well SensoPlate™) were pretreated with 1% (w/v) Pluronic® F-127 (PF127, Sigma-Aldrich) for 2 h at room temperature. After three wash steps (150 mM NaCl, 1 M HEPES pH 7.3 (Molecular Biology, Fisher BioReagents™), phase separation of the protein of interest was induced. Hence, the protein was mixed with equal volumes of PEG containing buffer (25 mM HEPES pH 7.3, 150 mM NaCl, 4 mM DTT, 20 mM MgCl₂, 2-times of final PEG or Ficoll concentration depending on the condition) in a final volume of 50 µl. For peptide experiments, 24.5 µM hIRE1α LD and 0.5 µM mCherry-hIRE1α LD-10His were preincubated with the respective peptide for 30 min on ice before phase separation was induced via PEG. The final protein and PEG concentration and incubation time is indicated in the Figure legends. Image processing was performed in ImageJ and Adobe Photshop® adjusting the brightness, contrast and sharpness of the images.

Phase separation was induced as described in the “In solution phase separation” section. For FRAP experiments hIRE1α LD was mixed with 2% of the corresponding mCherry tagged protein at a concentration of 25 µM, phase separation was induced in a test tube for 30 min in the presence of 6% PEG in the well. Experiments were performed on a Zeiss Axio Observer inverted microscope equipped with a Yokogawa CSU-X1-A1 Nipkow spinning disc unit (Visitron Systems; pinhole diameter 50 μm, spacing 253 μm), sCMOS camera (Pco.edge 4.2) and a Plan-Apochromat 63x/1.4 Oil DIC objective. Images were conducted every 5 s for a time course of 10 min with 80% HX, 50 ms exposure and 10% 561 nm laser intensity exposed for 100 ms. Per condition, 3 condensates were bleached after 2 frames with 100% 561 nm laser power for 10 ms per pixel. Image processing was performed in ImageJ selecting the FRAP ROI and two ROIs of the same size within a non FRAPed condensate for bleaching correction and an area without condensate for background correction for every FRAPed condensate. The background value was subtracted from the FRAP and bleaching value, followed by calculating the bleaching factor to correct the FRAP values leading to the final normalization. The normalized FRAP values were fitted to the One-phase association in PRISM.

The protein structure of the human IRE1α core Lumenal Domain (cLD) dimer was obtained from the Protein Data Bank (www.rcsb.org (Berman et al, 2000), PDB ID: 2HZ6) (Zhou et al, 2006). We added the missing residues (66–70, 89–90, 11–115, 131–152, 308–357, 369–443) as unfolded loops using UCSF Chimera (version 1.15, (Pettersen et al, 2004).

From this model, the regions DR1, DR2 and linker were extracted as isolated peptides and individually mapped to coarse-grained representation.

The peptides obtained were:

DR1, 131 – LTGEKQQTLSSAFADSLSPSTS -152;

DR2, aa 307-VPRGSTLPLLEGPQTDGVTIGDKGESVITPSTDVKFDPGLKSKNKLNYLRNY- 358;

Linker region, aa 369 -LSASTKMLERFPNNLPKHRENVIPADSEKKSFEEVINLVDQTSENAPTTVSRDVEEKPAHAPARPEAPVDSMLKD - 443.

The conversion of the all-atom models into Martini 3 (Souza et al, 2021) coarse-grained models and the setup of the simulation systems were performed using the tools martinize2 (https://github.com/marrink-lab/vermouth-martinize) and insane.py (Wassenaar et al, 2015) and gromacs/2020.5 tools (gmx insert-molecules). The termini were neutralized and the side chain fix was applied to prevent unrealistic side chain orientations as proposed in (Herzog et al, 2016).

We set up systems containing two disordered regions’ peptides by randomly inserting 16 copies of each region in a 30 × 30 × 30 nm³ simulation box. We obtained a system containing DR1 and DR2 and a system containing DR2 and the linker region. We solvated the systems with Martini water molecules and chloride and sodium ions, corresponding to a salt concentration of 150 mM.

After a first energy minimization we equilibrated the system. First, we ran a 10 ps-long simulation using a 1 fs time step and restraining the position of protein backbone beads by using harmonic potentials with force-constants of 1000 kJ mol⁻¹ nm⁻². Afterwards, we ran another 2.1 ns without restraints using a 30 fs time step and a final equilibration of 21 ns. After the equilibration, we ran MD simulations using 20 fs time step. The temperature in the simulation box was controlled by a velocity rescale thermostat (Bussi et al, 2007) (reference temperature T_ref = 300 K, coupling time constant tau_T = 1 ps). The Parrinello-Rahman barostat (Parrinello and Rahman, 1981) (reference pressure p_ref = 1 bar; coupling time constant τ_p = 24 ps) was used for the last equilibration step and for the production run.

Coarse-grained molecular dynamics simulations were performed using with the Martini 3.0 forcefield (Souza et al, 2021) and the GROMACS 2020.5 software (Lindahl et al, 2019).

We set up individual simulations for each region (DR1, DR2 and linker) in two different settings, namely containing two peptides or 33 peptides. For the two peptides’ simulations, we determined the dimensions of the box by setting a 2 nm distance between periodic images in a cubic box. In the latter simulation setting we randomly inserted 33 peptide copies in a 30 × 30 × 30 nm³ simulation box to obtain a protein concentration of 2 mM. We solvated the systems with Martini water molecules and chloride and sodium ions, corresponding to a salt concentration of 150 mM.

We analyzed the contacts formed over time among the peptide chains in these simulations. Initially, we computed the contact map between all beads of all peptides at each frame thanks to the Python package Contact Map Explorer (https://github.com/dwhswenson/contact_map, version 0.7.0). Two Martini beads were considered in contact if nearer than 0.5 nm. In the simulations containing 33 copies, we considered which peptide chains are interacting to create a network representation of the clusters at each frame from which we could determine which is the central chain of the cluster. Then we counted all the contacts between beads of the central chain and beads of its neighboring chains at each frame and we averaged over the number of chains interacting with the central one at each frame.

We obtained a matrix of dimensions (Number of beads per chain, Number of beads per chain) and we convert it to dimensions (Number of residues per chain, Number of residues per chain) by retaining the maximum score present between all the beads of a pair of residues. The contact matrices were computed in a similar way for systems of two peptides. In these simulations we considered all the interactions happening between the two chains, removing the notion of a central chain. We produced 1D-projections of the contact maps by summing up all the contribution for a specific residue in the final contact matrices for simulations of two or 33 copies.

For the establishment of stable cell lines in Mouse Embryonic Fibroblasts, a vector with a Tet-On doxycycline-inducible TRE3G promoter was utilized. TRE3G-P2A-eBFP2-PGK-puroSTOP-IRES-rtTA3 (kind gift from Gijs Versteeg) was cut using restriction enzymes BsrGI-HF and BamHI (New England Biolabs). hIRE1α signal sequence with the transmembrane domain (amino acids 1–469) 3XFlag and 6XHis tag and hIRE1α kinase-RNase domain (amino acids 470–977) were amplified frompShuttle-CMV-TO_hsIRE1-3F6H-GFP-LKR-K36.3. Additionally, due to its higher stability, the GFP tag, upstream of the kinase domain, was replaced with mNeonGreen through Gibson Assembly. The mutations were introduced into IRE1α N-terminal part (amino acids 1–469) or into kinase-RNase domain (amino acids 470–977) via PCR and the mutated fragment were used for Gibson assembly as described above. For the ³⁵⁹WLLI³⁶² and D123P construct, overlap extension PCR and site-directed mutagenesis of the wild type lumenal domain plasmid TRE3G-hsIRE1-3F6H-mNeonGreen-LKR-K36.3-PGK was used. All the constructs, except for the core LD, encode for the full-length lumenal sequence of hIRE1α (amino acids 1–469). The lumenal boundaries for the cLD include the core sequence (amino acids 24–389), as well as, a short region proximal to the transmembrane domain (amino acids 434–443), which was shown to be essential for the interaction with the Sec61 translocon (Plumb et al, 2015).

All transfections were performed by mixing DNA and Polyethylenimine (PEI, Polysciences, 23966) in a 1:3 ratio (μg DNA/μg PEI) in DMEM without supplements. Plasmids for the transfection were purified using an endotoxin-free Plasmid Kit (Qiagen). Transfection was performed using 1100 ng of total DNA (500 ng transfer plasmid, 500 ng pCMVR8.74 Addgene plasmid # 22036, 100 ng pCMV-VSV-G Addgene plasmid # 8454) The day before transfection, 2*10⁵ HEK293T HiEx packaging cells were seeded in 6-well plates in fully supplemented media. The following day, the above-described transfection mixture was added dropwise to the cells. Subsequently, cells were incubated for 48 h.

Following the 48 h incubation period, the viral supernatant was sterile filtered with a syringe. The day before transduction 1*10⁵ MEF (IRE1α^−/−/IRE1β^−/−) were seeded in a six-well plates in fully supplemented media. For the transduction, the virus was mixed with fully supplemented DMEM (Sigma-Aldrich, D6429) and 8 μg/ml Polybrene (Sigma-Aldrich, TR-1003-G) at 1:50 (v/v). After a 48 h incubation period, cell lines were expanded to 10–15 cm dishes. Protein expression, for subsequent Fluorescence Activated Cell Sorting (FACS), was induced with 400 nM of Doxycycline for 24 h. Cells were sorted in yield mode using BD FACSAria II or BD FACSMelody, gated for low and high-expression cells. The high-expression cells were resorted in stringent mode, following the same procedure. The second FACS sorted high population of the first FACS sorted high population was used for characterization. All the stable cell lines generated in this work are available upon request. The cell lines were not authenticated and tested for mycoplasma contamination.

IRE1 double-knockout Mouse Embryonic Fibroblasts (MEF) (IRE1 α^−/−/IRE1β^−/−) reconstituted with a doxycycline-inducible hIRE1α-mNG (or mutants) were seeded at a density of 20000 cells in a µ-Slide eight-well dish (ibidi) 1 day before the experiment. IRE1α expression was induced for 24 h by adding 400 nM doxycycline. Cells were stressed with 5 µg/ml Tunicamycin for 4 h. The experiment was stopped by washing with cold PBS and fixation with 4% paraformaldehyde for 7 min. After two more washes with PBS, the cells were incubated for 1 h in blocking buffer (PBS, 10% FBS, 1% Saponin) followed by primary antibody incubation overnight at 4 °C (Calnexin, Abcam ab22595 at a dilution of 1:200). After washing twice with wash buffer (PBS, 10% FBS) the secondary antibody (Alexa Fluor 594 goat anti-rabbit, Invitrogen A11037 at a dilution of 1:1000) was incubated for 1 h at room temperature. After three additional wash steps, the sample was imaged in PBS on a Zeiss LSM 980 inverse point scanning confocal microscope with a Plan-Apochromat 63x/1.4 Oil DIC, WD 0.19 mm objective. The microscope is operated by the Zeiss ZEN 3.3 microscope software. mNG and Atto549 were excited by the 488 nm and 561 nm laser diodes of the microscope, respectively. Image processing was performed in ImageJ.

Treated MEFs at a confluency of 80% were collected in RIPA buffer. The protein concentration was determined by a bicinchoninic acid assay using a commercially available kit. In all, 10–15 μg protein of the lysate in sample buffer was loaded after denaturation for 10 min at 95 °C on a 10% sodium dodecyl sulfate gel. The proteins were wet transferred from the gel to a nitrocellulose membrane in transfer buffer (25 mM Tris, 192 mM glycine, 20% (v/v) ethanol, pH 8.3) for 120 min at 110 V. The proteins on the membrane were stained with Ponceau S for 5 min followed by blocking in 5% milk for 1 h at room temperature. The primary antibody was applied in 2.5% milk for 1 h at room temperature or overnight at 4 °C. The membranes were washed five times in TBST for 5 min before the secondary antibody in 2.5% BSA (anti-rabbit IgG (H + L), HRP Conjugate, Promega W401B at a dilution of 1:10,000) was added and incubated for 1 h at room temperature. After five 5 min TBST wash steps, the chemiluminescence substrate for the horseradish peroxidase was applied using a commercially available kit. The membranes were imaged using a ChemiDoc system and analyzed with the Image Lab software of Bio-Rad.

Primary antibodies