RHBDL4 protects against endoplasmic reticulum stress by regulating the morphology and distribution of ER sheets

In metazoans, the architecture of the endoplasmic reticulum (ER) differs between cell types, and undergoes major changes through the cell cycle and according to physiological needs. Although much is known about how the different ER morphologies are generated and maintained, especially the ER tubules, how context dependent changes in ER shape and distribution are regulated and the factors involved are less characterized. Here, we show that RHBDL4, an ER-resident rhomboid protease, modulates the shape and distribution of the ER, especially under conditions that require rapid changes in the ER sheet distribution, including ER stress. RHBDL4 interacts with CLIMP-63, a protein involved in ER sheet stabilisation, and with the cytoskeleton. Mice lacking RHBDL4 are sensitive to ER stress and develop liver steatosis, a phenotype associated with unresolved ER stress. Our data introduce a new physiological role of RHBDL4 and also imply that this function does not require its enzymatic activity.


Introduction
The ER is the largest membrane-bound organelle of eukaryotic cells, comprising the nuclear envelope and peripheral ER. In metazoans, the peripheral ER spreads throughout the cytoplasm as a network of interconnected flat sheets and tubules; the flat sheets are present mostly around the nucleus while the tubules interconnect at the level of three-way junctions generating a reticular network towards the cell periphery 1, 2 . The two distinct ER morphologies accommodate different functions.
The sheets are studded with ribosomes and form the rough ER, the site of import, folding and quality control of secreted and transmembrane proteins, and are abundant in professional secretory cells 3,4 . The tubules mostly lack ribosomes and form the smooth ER, associated with lipid synthesis and Ca 2+ signalling/homeostasis, abundant in steroid secreting cell [3][4][5][6] . The ER tubules also engage in contacts with other membrane-bound, as well as membraneless, organelles, and a role is emerging for the ER in regulating the biogenesis and/or the dynamics of organelles with which it shares contact sites 7,8 . Several proteins that shape the ER have been discovered. Reticulons and REEPs generate the high membrane curvature characteristic of ER tubules, while Atlastins mediate the homotypic fusion of ER membranes [9][10][11][12][13][14] . ER sheets have a more complex structure and less is known about how they are generated and maintained. Reticulons generate the curvature at sheet edges, and several components were shown to stabilise the flat sheets, including the coiled-coil proteins CLIMP-63, p180 and kinectin, the attached polysomes engaged in protein translation, as well as the actin cytoskeleton through myosin 1c 4,15,16 . None of these factors, however, are essential for the existence of ER sheets, suggesting additional structural and regulatory complexity. Beyond its morphology, the factors that contribute to the positioning of the ER within the cell are also not well established; again, more is known about tubules than sheets. ER tubules are often situated along microtubules, and their growth and movement in the cells rely largely on this interaction [17][18][19] . ER-sheet proteins CLIMP-63 and p180 can also bind to microtubules, and cytoskeleton depolymerisation was shown to change ER sheet distribution 15,20,21 , although the mechanisms involved in positioning and the dynamics of the ER sheets remain poorly understood. The ER network varies in volume, shape and distribution between cell types and it undergoes reorganisation through the cell cycle, differentiation, and according to physiological needs 3, 22, 23 , implying the need for regulation. The abundance of ER-shaping proteins is a determining factor of the sheet to tubule ratio: high levels of reticulons and REEP favour the abundance of tubules over the ER sheets, while CLIMP-63 has opposite effect 4,24 . Overall, though, rather little is known about the regulation of the rapid and complex changes in ER morphology and distribution in different cellular contexts.
One challenge to ER function, and implicitly to its morphology and distribution, is ER stress. The homeostasis of protein folding in the ER is essential for healthy cells and organisms, and is maintained by a balance between protein biosynthesis and folding on one hand, and export and degradation on the other. Conditions that perturb this homeostasis, generically referred to as ER stress, lead to the accumulation of unfolded proteins in the ER lumen and the activation of the unfolded protein response (UPR). The UPR is an integrated signalling network from the ER to cytoplasm and nucleus, which counteracts the effects of ER stress by upregulating ER chaperones, folding enzymes, and components of the ER-associated protein degradation (ERAD) machinery, while also reducing protein translation 25,26 . ER stress also causes significant changes in the size, morphology and distribution of the ER in mammalian and yeast cells 27,28 .
RHBDL4 (aka RHBDD1) is a member of the rhomboid intramembrane serine proteases; it is localised in the ER and is upregulated by ER stress 29,30 . Although its overall core function remains mysterious, several roles have been assigned to RHBDL4. Its increased expression has been correlated with cancer progression 31,32 . RHBDL4 has also been shown to activate EGFR signalling by promoting the secretion in extracellular microvesicles of the growth factor TGFα 33,34 . And, in a more classic protease function, RHBDL4 is reported to cleave amyloid precursor protein (APP) 35 . The most well characterized RHBDL4 function, and the one most apparently relevant to its ER localization, is a function in cleaving and promoting the degradation of various unstable proteins or orphan subunits of protein complexes 29,36 , implying that RHBDL4 participates in some forms of ERAD.
Here, we describe an unexpected role of RHBDL4 in regulating the shape and distribution of the ER. By gain and loss of function experiments, we found that the levels of RHBDL4 modulate the shape and distribution of the ER, particularly under ER stress and when the cytoskeleton is disrupted, conditions that require rapid, dynamic changes in ER sheet distribution. We report that RHBDL4 interacts with CLIMP-63 and it potentially associates with the cytoskeleton. Unexpectedly, this role of RHBDL4 appears to be independent of its enzymatic activity, implying an additional, non-proteolytic function of this intramembrane protease. Finally, we also show that RHBDL4 is essential for resolving acute ER stress in mice.

The levels of RHBDL4 affect the ER organisation
Overexpression of RHBDL4 disrupts ER morphology RHBDL4 is an ER-resident protein 29 . While doing transient transfections and immunofluorescence, we noticed that overexpressed RHBDL4 staining looked unusual in many cells. To characterise this within the context of ER organisation, we used CLIMP-63 as a marker for ER sheets and RTN4 for the whole ER -reticulons reside in ER tubules and the edges of the ER sheets 4 . When expressed at low levels, RHBDL4 showed a perinuclear localisation, characteristic of ER sheets, similar to that of CLIMP-63 (Fig. 1a, second column, the cell indicated by arrowhead). At higher expression levels, RHBDL4 distributed widely into the ER and induced the appearance of punctate structures. Some of the RTN4 was recruited to the puncta, while CLIMP-63 redistributed widely towards the cell periphery into reticular structures (Fig. 1a, second and third column, the cells indicated by arrow).
Very high expression levels of RHBDL4 led to a massive disruption in the appearance of the ER (Fig. 1a, fourth column), especially at the level of ER sheets, as shown by CLIMP-63 distribution: part of CLIMP-63 was redistributed into tubules toward the cell periphery, while the rest was confined into compact structures around the nucleus.
Overexpression of a RHBDL4 mutant that had the catalytic serine replaced by an alanine -RHBDL4 S144A 37 -showed similar results: increasing levels of RHBDL4 S144A led to increased disruption of ER architecture (Fig. 1b), suggesting that the proteolytic activity of RHBDL4 is not required in this context. This disruption of ER morphology was specific to RHBDL4: the overexpression of another rhomboid protease, RHBDL3, which is structurally very similar to RHBDL4 did not induce changes in the ER morphology, even when expressed at very high levels (Fig. 1S).
To make this a more rigorous control for ER-localised RHBDL4, RHBDL3, which normally moves through the ER to its primary plasma membrane location 30 , was artificially retained in the ER by the addition of a KDEL retention sequence at its Cterminus 38 .

RHBDL4 KO affects the ER sheet distribution
Given the effect of RHBDL4 overexpression on the ER, we hypothesized that RHBDL4 plays a role in the ER shape and/or distribution under physiological conditions. To test this, we used two different RHBDL4 KO cell lines -mouse embryonic fibroblasts (MEFs) and human HeLa cells -and compared their ER distribution to that of their wild-type (WT) counterparts.
To confirm this, we quantified the signal distribution using CellProfiler-3.0.0 39 . The area between the nucleus and plasma membrane was divided into ten bins (Fig. 2b) and the mean fractional intensity at a given radius was calculated as fraction of total intensity normalized by the fraction of pixels at that radius. Indeed, there was no difference between WT and RHBDL4 KO MEFs (Fig. 2c). The shape of the ER needs to change in response to varying demands on its function. As RHBDL4 was shown to be upregulated in response to ER stress 29 , we asked whether RHBDL4 is important for the ER shape changes that occur during ER stress. We challenged the cells with tunicamycin, an inhibitor of N-linked glycosylation that leads to accumulation of misfolded proteins in the ER and therefore induces ER stress 40,41 .
In WT MEFs, after 24 hours of tunicamycin treatment, CLIMP-63 redistributed dramatically throughout the cytoplasm as compared to the control (Fig. 2d, upper row), implying a major reorganisation of ER sheets. In contrast, CLIMP-63 redistribution was abnormal in RHDBL4 KO MEFs, with 28.76% of cells showing a strong phenotype, with a "collapse" of CLIMP-63 signal around the nucleus (Fig. 2d, lower row, and quantified in Fig. 2e). This phenotype was significantly rescued by transient transfection of RHBDL4 WT (Fig. 2e), indicating that the effect was specific to loss of RHBDL4. We could not detect rescue when using the RHBDL4 S144A mutant (data not shown) but this result is hard to interpret because expression of this mutant, even at low levels, induces acute ER stress on its own 29 (Fig. S2, lanes 3 and 4 compared to lanes 7 and 8). It is therefore likely that its expression combined with tunicamycin treatment induced very high levels of ER stress in RHBDL4 KO cells, preventing a rescue of ER morphology.
In WT HeLa cells, the distribution of CLIMP-63 and RTN4 followed their expected localisation: CLIMP-63 was more concentrated within the innermost bins, while RTN4 was distributed more widely towards the plasma membrane (Fig 3a, upper row, and quantified in Fig. S3). The ER was more compacted around the nucleus in the RHBDL4 KO cells when compared to the WT, especially when looking at CLIMP-63 (Fig. 3a, lower row, and quantified in Fig. 3b and 3c). Transient transfection of the RHBDL4 KO cells with either WT or the catalytically inactive RHBDL4 rescued the phenotype (Fig. 3d), supporting the conclusion that the role of RHBDL4 in ER sheet distribution is not dependent on its proteolytic activity.
As in MEFs, tunicamycin treatment of HeLa cells led to a redistribution of CLIMP-63 throughout the cytoplasm in WT cells, whereas, in RHDBL4 KO, a subset of cells showed a pronounced phenotype of more concentrated perinuclear staining and a non-homogeneous distribution towards the cell periphery (Fig. 3e, 3f).
Altogether our results in MEFs and HeLa cells show that RHBDL4 is important for ER sheet distribution and that the effect was more pronounced during ER stress.

RHBDL4 KO cells
The role of RHBDL4 in ER shape/distribution could be direct or indirect, via other ER-shaping proteins. Also, the phenotypes we described in RHBDL4 KO cells could be due to a defective folding/secretory function of the ER, or failure to activate the UPR and cope with the ER stress. Analysis of MEFs and HeLa cells by western blot showed no obvious differences in the levels of RTN4, ATL1 or CLIMP-63 between WT and RHBDL4 KO, neither in control nor under induced ER stress conditions ( Fig.   4a and Fig. S4a, respectively). This implies that RHBDL4 does not affect the levels of these ER-shaping proteins.
We next asked whether RHBDL4 affects the ER stress response. Under ER stress, to cope with the folding load, cells increase their protein folding capacity by activating the UPR 42 . Two of the chaperones that are upregulated most in response to ER stress are BiP and GRP94 43 . As shown in Figure 4b  cells. Failure to restore protein homeostasis in the ER results in activation of apoptosis. We assayed a central mediator of apoptosis, the pro-apoptotic transcription factor CHOP 44 , and found similar levels in WT and RHBDL4 KO MEFs ( Fig. 4b). We conclude that the ER stress response is not significantly affected in these RHBDL4 KO cells.
To assess the general secretory function of the ER in KO cells, we examined the cellular location of Nicastrin, a component of the γ-secretase complex, which is found in two forms: one immature, ER-localised, detected as a lower molecular weight band in western blots, and one mature, post-Golgi, detected as a higher molecular weight form 45 . We found no differences in the levels of post-Golgi Nicastrin, neither in control nor under ER stress conditions ( Fig. 4b and Fig S4b).
The efficiency of tunicamycin treatment was similar in the WT and RHBDL4 KO cells, as shown by the appearance of non-glycosylated species of Nicastrin. This implies that the overall secretory function of RHBDL4 KO cells is not significantly compromised. Taking all these results together, we conclude that the role of RHBDL4 in shaping the ER under stress conditions is not a secondary effect of a role in homeostasis of ER shaping proteins, response to ER stress, or general secretion.

RHBDL4 localises to the ER sheets and interacts with CLIMP-63
The phenotypes we observed so far indicate that RHBDL4 levels have an impact on the ER sheets, so we wanted to determine the localization of endogenous RHBDL4 within the ER. Several commercial antibodies as well as one produced by our group failed to detect the endogenous RHBDL4 by immunofluorescence. We therefore used a biochemical approach to examine endogenous RHBDL4 localisation. We isolated microsomes from mouse liver as well as from HeLa cells, and used differential centrifugation to separate the rough and smooth ER on a sucrose cushion 46,47 . As sketched in Figure 5a, left, following ultracentrifugation the rough ER (ER sheets studded with ribosomes) sediments through the 1.3 M sucrose cushion, while the smooth ER (ER tubules) segregates at the interface between the two sucrose concentrations; the material in between is a rough/smooth, or intermediate, ER 5 . For HeLa cells, due to a lower amount of material, the smooth ER was not readily visible and, therefore, we analysed all fractions starting one millilitre above the interface and finishing with the pellet, as indicated in Figure 5a, right side.
In agreement with their published localisation, by this fractionation method we found ATL1 located in the smooth ER, RTN4 in both smooth and rough ER, while Calnexin (CNX) -an ER-resident chaperone -and CLIMP-63 were specific to the rough ER ( Fig. 5b and 5c), 4, 48 , confirming a good separation. In both HeLa cells and mouse liver, RHBDL4 distributed predominantly to the rough ER ( Fig. 5b and 5c).
Upon ER stress there was no obvious change in RHBDL4 distribution in mouse liver, and a somewhat increased level in rough ER in HeLa cells. We obtained similar results in MEFs (Fig. S5).
These results show that endogenous RHBDL4 is localised predominantly within the ER sheets, which is consistent with the ER disruption phenotypes we observed.
Next, we asked whether we could detect interactions between RHBDL4 and ER shaping proteins. In co-immunoprecipitation experiments, we found that endogenous RHBDL4 physically interacts with CLIMP-63 (Fig. 5d). Significantly, this interaction was also found in our previous large-scale interaction screen of the RHBDL4 interactome -CLIMP-63 aka CKAP4 was one of the reproducible hits in HeLa and HEK 293 cells 49 . We could not detect an interaction between RHBDL4 and RTN4 or ATL1, or between RHBDL4 and CNX, another ER sheet membrane protein (not shown). The interaction between endogenous RHBDL4 and CLIMP-63 was recapitulated using overexpressed, tagged proteins: CLIMP-63 HA pulled down RHBDL4 Myc and, in turn, RHBDL4 HA pulled down CLIMP-63 Myc (Fig. 5e, lane 7 and 8, respectively). Both, RHBDL4 WT and RHBDL4 S144A co-immunoprecipitated with CLIMP-63 ( Fig. 5f, lanes 9 and 10), indicating that the interaction with CLIMP-63 did not depend on the RHBDL4 active site.

RHBDL4 associates with the cytoskeleton
CLIMP-63 interacts with microtubules 20 , and given the impact of RHBDL4 overexpression and KO on the shape and distribution of the ER, we asked whether RHBDL4 might also interact with the cytoskeleton, which provides a functional scaffold for shaping and distributing the ER.
Using an approach expected to preserve protein-cytoskeleton interactions, we separated the cytoskeletal fraction from HeLa cells 50 . During the fractionation, the actin cytoskeleton is preserved, as well as the cold-stable microtubules often described as the ones that interact with other proteins 51 (Fig. 6, the lower two rows show actin and tubulin). Strikingly, most of RHBDL4 and CLIMP-63 segregated into the cytoskeletal fraction while RTN4 and ATL1 did not (Fig. 6a,  Microtubule depolymerization results in rapid reorganisation of the ER network 17,52,53 . We used this to investigate further a role for RHBDL4 in ER dynamics and distribution in the context of ER-cytoskeleton interaction. We exposed WT and RHBDL4 KO MEFs and HeLa cells to nocodazole, to disrupt the microtubules, and analysed the resulting changes in ER sheet morphology and distribution. In WT cells, both MEFs and HeLa, the nocodazole treatment led to wide spreading of CLIMP-63 signal throughout the cytoplasm, confirming that the ER sheets underwent a major reorganisation (Fig. 7c, third row and Fig. S7a). In RHBDL4 KO, CLIMP-63 redistribution was less pronounced and, in many cells there were peripheral areas devoid of ER sheets (Fig. 7c, fourth row and Fig S7a). Quantification of this CLIMP-63 spreading phenotype confirmed the difference in the peripheral distribution between WT and RHBDL4 KO cells (Fig. 7d, f and Fig. S7b). These results showed that, as when ER stress was induced by tunicamycin treatment, in the absence of RHBDL4, microtubule-dependent ER sheet dynamics are defective.

RHBDL4 protects against ER stress in vivo
To study the role of RHBDL4 in vivo we created RHBDL4 KO mice. We detected no obvious phenotype under normal conditions but the fact that RHBDL4 KO MEFs only showed a phenotype when challenged with tunicamycin (or nocodazole) prompted us to examine the effect of ER stress in KO mice.
We used intraperitoneal injection of a sublethal dose of tunicamycin (1 μg/g body weight) 54, 55 to induce ER stress, and followed the mice over three days. The first difference we noticed between WT and RHBDL4 KO mice was in weight loss: at day three of treatment the RHBDL4 KO mice had lost significantly more weight than WT animals (Fig. 8a), suggesting that the former were more severely affected by induced ER stress. Liver is the mouse tissue with the strongest response to tunicamycin-induced ER stress 56 and one of the tissues with highest RHBDL4 expression (Fig. 8b). Therefore, it was the tissue of choice to look for further differences between the WT and RHBDL4 KO mice. Upon dissection, we also noticed a striking difference in the appearance of WT and RHBDL4 KO livers at three days of treatment, with the latter being pale in colour, a sign of liver steatosis.
Oil Red O staining of liver sections confirmed this; after three days of treatment, the RHBDL4 KO mice accumulated dramatically elevated levels of neutral lipids in their livers when compared to WT (Fig. 8c). At day one, there was no apparent difference between the two genotypes (Fig. S8).
To check whether the failure of RHBDL4 KO mice to cope with ER stress induced by tunicamycin was due to a deficient activation of the UPR, we measured the upregulation of UPR targets BiP and GRP94, as well as another marker of UPR activation, splicing of the X-box binding protein-1 (XBP1). In response to ER stress, XBP1 mRNA undergoes splicing and is translated into an active transcription factor.
The transcriptional program triggered by XBP1 leads to ER expansion, upregulation of ERAD and secretory pathway components 57 . We found no differences between WT and RHBDL4 KO mice (Fig. 8d), indicating that RHBDL4 KO mice do not have a defect in UPR activation. Nevertheless, three days after tunicamycin treatment, the ER stress was unresolved in RHBDL4 KO mice, as indicated by the persistence of the pro-apoptotic transcription factor CHOP (Fig. 8e). RHBDL4 expression itself was upregulated by the tunicamycin treatment, consistent with a role for it in coping to ER stress (Fig. 8f).
For a better insight into the phenotype of RHBDL4 KO mice under ER stress, we performed transmission electron microscopy (TEM) on liver sections and looked at the appearance of the ER. Under control conditions, the ER looked similar in WT and RHBDL4 KO livers, with stacks of rough ER packed around the mitochondria (Fig. 8g, left column, blue arrows). Twenty-four hours after tunicamycin treatment, in both genotypes the rough ER lost the appearance of organised stacks, became dilated and distributed broadly throughout the cytoplasm, with only scarce ER sheets around mitochondria (Fig. 8g, middle column). By 72 hours after tunicamycin treatment, a clear difference between the genotypes became clear: in WT livers the ER had started to recover into structures of stacks around the mitochondria, while in the RHBDL4 KO the rough ER remained dilated, filling the cytoplasm, and large lipid droplets were accumulated (Fig. 8g, right column). These TEM data further demonstrate a major defect in the resolution of ER stress in the absence of RHBDL4.

Discussion
In metazoans, the ER has a complex architecture with different ER morphologies hosting different functions. The ER architecture differs between cell types and undergoes changes during the cell cycle and according to physiological needs.
There is significant knowledge about how the ER tubules are generated and maintained, but not as much is known about the ER sheets. How do rapid changes in shape and distribution of the organelle take place and what are the regulators; what factors contribute to the positioning of the ER within the cell, and how does the ER interact with the cytoskeleton? We report here that RHBDL4 plays a role in ER sheet dynamics during ER stress, that it interacts with CLIMP-63 and the cytoskeleton, and that it is essential for coping with ER stress in mice.
We have shown that endogenous RHBDL4 resides primarily within the ER sheets. Consistent with this, the phenotypes induced by its overexpression or knockout were mostly at the level of ER sheets, especially during conditions that require their reorganisation/redistribution -ER stress or microtubule depolymerisation. In response to ER stress, mammalian and yeast ER undergoes significant changes in size, morphology and distribution 27,28,57 . In both cell lines we used in this study, human HeLa cells and mouse embryonic fibroblasts, RHBDL4 knockout led to defective redistribution of ER sheets. At the same time, the function of the ER, in terms of secretion, and the activation of the UPR, were similar in WT and RHBDL4 KO cells, suggesting that RHBDL4 participates in adapting ER shape to the cellular context rather than to the ER function itself. Overall, the common theme that emerges from our results is that RHBDL4 has a role in ER sheet dynamics; in its absence, the ER sheets were more compact around the nucleus and more stable to disassembly and redistribution, while moderate overexpression had the opposite effect.
While the reticular network of ER tubules can be reconstituted in vitro using a small number of proteins -reticulons, REEPs and Atlastins 10, 11 -the ER sheets have a more complex structure, and it is not clear how they are generated and maintained. None of the components shown to stabilize ER sheets is essential, and homo-oligomers via its luminal domain, which limit its movement within the plane of membrane 58 . Also, CLIMP-63 is thought to form intermembrane bridges to stabilize the ER sheets 4, 59 . One possibility is that RHBDL4 interferes with the formation of CLIMP-63 homo-oligomers or intraluminal bridges to allow ER sheets to be dynamic when needed. We emphasise that this is a speculative idea at this stage but overall our data suggest that, while CLIMP-63 plays a structural role in ER sheets, RHBDL4 appears to be involved in their dynamics, playing a regulatory role.
To date, there is no reported substrate for RHBDL4 that could explain this role and none of the ER shaping proteins we checked appeared to be cleaved by RHBDL4. Indeed, the effects we describe here using overexpression were similar for RHBDL4 WT and the catalytically inactive mutant RHBDL4 S144A: the effect on ER and microtubules, the segregation into the cytoskeletal fraction and the interaction with CLIMP-63. Moreover, both RHBDL4 WT and RHBDL4 S144A rescued partially the phenotype in RHBDL4 KO HeLa in control conditions. This leads us to conclude that the function we describe in ER shape and dynamics is independent of the catalytic function of this rhomboid protease. This conclusion is made more complex by the fact that expression of RHBDL4 S144A induces ER stress, making the Pharmacological ER stress induces transient lipid droplet (LD) accumulation in mouse liver and, when combined with a defective UPR signalling, it leads to hepatic steatosis. Genetic disruption of any of the three UPR branches or of protein quality control results in failure to cope with the ER stress, and hepatic steatosis 28, 55 , most likely caused by a failure to upregulate the necessary folding machinery. When challenged with tunicamycin, RHBDL4 KO mice showed a phenotype similar to that described for UPR deficient mice: significant weight loss and liver steatosis, characterised by massive accumulation of LDs. However, in RHBDL4 KO mice, UPR activation was normal (as shown by XBP1 splicing, and BiP and GRP94 upregulation), suggesting a different cause for the failure to cope with the ER stress.
Three days after tunicamycin treatment, in WT livers, ER sheets began to reassemble around the mitochondria, whereas in RHBDL4 KO livers, the ER remained dilated and scattered through the cytoplasm, and large LDs were common.
Although ER stress results in LD accumulation, and the disruption of LD biogenesis induces the UPR 65 , the role of LDs in ER stress response remains unclear. Our data further emphasise the relationship between ER stress and LD accumulation but further work is needed to uncover the factors and mechanisms involved in ER sheet reassembly and LD clearing after ER stress: our work suggests that RHBDL4 participates in this stress resolution. Notably, some proteins with well defined roles in ER shape -for example, Atlastin, REEP1 and Spastin -are known to affect LD biology [66][67][68] , but none of these roles have been studied in the context of ER stress.
In conclusion, while the relevant molecular and cellular mechanisms remain to be discovered, we have uncovered a physiological role for RHBDL4 in regulating the morphology and distribution of ER sheets. Our work has also shown that this previously unrecognised RHBDL4 function is essential in cells for coping with conditions that require dynamic reorganisation of ER sheets, and in mice for resolving the pathology associated with ER stress.

RHBDL4 KO HeLa cells
The RHBDL4 KO HeLa cells were generated using the CRISPR/Cas9 system. The

Tunicamycin and Nocodazole treatment
To induce ER stress, the cell culture medium of ~70% confluent cells was replaced with fresh medium containing tunicamycin (0.1 µg/ml for MEFs and 5 µg/ml for HeLa, final concentration, T7765; Sigma-Aldrich), and the cells were incubated at 37˚C, 5% CO2, for 24 hours. For microtubule depolymerisation experiments, the cell culture medium of ~70% confluent cells was replaced with fresh medium containing 10 μM Nocodazole (M1404; Sigma-Aldrich) and cells were incubated at 37˚C, 5% CO2, for 15 minutes. In both cases, control cells were treated with DMSO.

Immunoprecipitation
All experiments were performed on ~80% confluent cells. For endogenous RHBDL4, the immunoprecipitation was done using the Pierce™ Crosslink Magnetic IP/Co-IP Kit (88805; Thermo Scientific), according to the manufacturer's protocol. The cells were washed twice with ice-cold phosphate-buffered saline (PBS) and lysed in lysis buffer (25mM Tris, 150mM NaCl, 1mM EDTA, 1% NP40, 5% glycerol, pH 7.4) containing EDTA-free protease inhibitor cocktail (11873580001; Roche). Lysates were centrifuged at 10,000g for 10 min and the post-nuclear supernatants were used for subsequent immunoprecipitation. Cell lysates were incubated with a rabbit anti-

Cytoskeleton Enrichment
The cytoskeletal fraction was separated using the ProteoExtract® Cytoskeleton

Microsome separation and ER fractionation
Microsomes were separated from MEFs, HeLa cells or mouse liver using the protocol described by Song et al. (2006) 47

Image analysis and quantification
For figures, the brightness/contrast of images were adjusted in Fiji, using the same settings across images belonging to the same experiment. Quantification of signal intensity was done using CellProfiller -3.0.0 39 . The area between the nucleus and plasma membrane was divided into ten bins and the mean fractional intensity at a given radius was calculated as fraction of total intensity normalized by the fraction of pixels at that radius. For cell segmentation, the nuclei were used as primary objects and merged images of Phalloidin or tubulin (for whole cell staining) and CLIMP-63 (ER staining) were used to identify the whole cell area. The cells that were not segmented correctly were manually removed from the analysis.

Reverse transcriptase polymerase chain reaction (RT-PCR)
To detect the XBP1 splicing, the RNA from mouse liver was isolated using RNeasy

Statistical analysis
The error bars in all figures represent the 95% confidence interval. The statistical significance was calculated using two-sided t-test and two-sided Mann-Whitney-Wilcoxon test.