The vast scope of the transcriptional profile of UPR target genes previously suggested that the UPR leads to a comprehensive remodeling of the secretory pathway, allowing cells to adjust their ER protein folding and secretory activities according to need. The transcription factor XBP1, the metozoan ortholog of Hac1, was shown in mammalian cells to induce an expansion of the ER [67
]. Here we show that in yeast, a similar organelle expansion occurs, with the volume of the ER increasing at least 5-fold upon UPR induction. It seems logical for a cell to expand both the machinery and the space dedicated to protein folding to meet the needs of a new physiological state in which proteins stay longer in the ER until they are properly folded or committed to degradation. Proliferating the ER reduces the concentration of unfolded protein, thereby preventing aggregation and giving more time to properly fold proteins or to degrade folding failures. To our surprise, we discovered that an ER-selective UPR-induced form of autophagy, ER-phagy, is activated and is required for cells to survive under conditions of severe ER stress, thus establishing the existence of a physiologically important link between the UPR and autophagy.
Because execution of the UPR transcriptional program leads to ER expansion, it is plausible to assume that ER-phagy serves to provide the opposite effect of reducing the volume of the ER and with it, unfolded ER proteins that have accumulated there. For example, it has been recently shown that the Z variant of human α-1 proteinase inhibitor (A1PiZ) encounters different degradation pathways depending on its expression and aggregation level [69
]. Normally, A1PiZ is a substrate of ERAD. However, when A1PiZ is overexpressed, it is sent to the vacuole via the secretory pathway, and any excess of A1PiZ that aggregates inside the ER is targeted to the vacuole via an autophagy pathway, suggesting that ER-phagy may be induced under these conditions. In liver cells, reduction by autophagy of barbiturate-induced expansion of smooth ER was previously observed when the drug was removed [70
]; similarly, in UT-1 cells, the expanded ER induced by HMG-CoA reductase (an ER membrane protein) overexpression is reduced by autophagy when the expression of the enzyme is tuned down [71
]. Thus the UPR may function in conjunction with ER-phagy to balance ER synthesis with ER degradation as part of the homeostatic control network that adjusts ER abundance up and down. Similarly, pexophagy degrades excess peroxisomes when cells switch carbon sources from using fatty acids to other food stuffs [39
], and mitophagy reduces mitochondrial abundance, e.g., under starvation conditions or under respiring conditions when mitochondria become easily damaged by oxygen radicals [40
]. For pexophagy, Pex14 has been proposed to have a role in the selective targeting of peroxisomes for degradation [75
], but how autophagy targets other organelles for selective sequestration remains an open question.
The ERAD pathway is thought to continually remove unfolded proteins from the ER and channel them to degradation by the proteasome. We have previously shown that ERAD is intimately linked to the UPR; either pathway is necessary for cell survival if the other one is impaired [18
]. Many ERAD genes are UPR targets, and it was their up-regulation during UPR-inducing conditions that let to the discovery of this connection. By contrast to ERAD genes, autophagy genes were not defined as UPR targets in this study, and the connection between the UPR and autophagy escaped attention. Autophagy genes were excluded from the set of UPR target genes because they are subject to dual control: in response to protein misfolding in the ER, they are induced by Hac1i
in the Ire1-dependent UPR pathway, but also by a parallel pathway that can operate in the absence of Ire1 and Hac1. It is likely that this parallel signaling pathway originating from the ER lumen corresponds to the S-UPR previously described to control the expression level of HAC1
]. Studying the regulation of autophagy genes therefore provides a powerful new experimental angle on deciphering the molecular mechanism of Ire1-independent ER-to-nucleus signaling in yeast. Because Hac1i
expression from the glucocorticoid receptor-activated promoter is not sufficient to induce ERA formation, another signal from the ER lumen beyond activating Ire1 must be required. This signal could (directly or indirectly) establish a marker on the ER surface, labeling the organelle as “damaged” for sequestration into ERAs, and it may utilize the same pathway that confers Ire1-independent regulation of ATG8
transcription and, possibly, of other genes encoding components of the autophagy machinery.
The ERAs observed in this study show several remarkable features. First, they have a strikingly homogenous appearance and are largely filled with tightly stacked membrane cisternae. Second, the Sec61-cherry staining and the Sec63-myc immunogold staining show that the cisternae are derived from the ER. This notion is supported by the observation that cells containing ERAs lack expanded ER, which appears to be consumed during ERA formation. Third, the outer membrane of the delimiting double membrane of ERAs is densely studded with ribosomes and thus also derives—at least in part—from the ER. It has been a longstanding and still unresolved question where the delimiting membrane of conventional starvation-induced autophagosomes comes from [77
]. Our finding thus represents a first identification of the origin of the delimiting membrane of an autophagosomal structure by showing that the ER can serve as the membrane source to generate autophagosomal double membranes. Finally, the inner envelope membrane and the membrane of the stacked cisternae for the most part lack bound ribosomes (E). The tight packing of the cisternae is consistent with the absence of ribosomes, which could not be accommodated in the approximately 16-nm space between them (a ribosome is approximately 30 nm in diameter). Taken together, these observations suggest that a sophisticated mechanism must exist that peels ER from the cell cortex, strips off most bound ribosomes, compacts the membrane into tight stacks, and packages the stacks selectively and with exclusion of most of the surrounding cytosol into ERAs by enclosing them in an envelope that is also derived—at least in part—from ER membranes. Hence, ERA formation involves a controlled “self-eating” of the ER.
No ERAs are formed in cells lacking Atg8, which is required for early steps in the biogenesis of autophagosomes. We found that during the UPR, Atg8 is first diffusely distributed throughout the cytosol. At later time points, Atg8 coalesces into discrete foci (PASs). This phenomenon occurred in the vast majority of cells (6 ± 2 PASs per cell at 3 h after UPR induction). At the same time point, ERAs formed in 20% of the cells in apparent juxtaposition to PASs. Notably, there is no overlap in staining. Moreover, and in contrast to nitrogen starvation–induced macroautophagy, no Atg8 is delivered to the vacuole (as indicated by the lack of proteolytic cleavage of GFP-Atg8). In principle, two distinct but not mutually exclusive explanations could account for this observation. First, ERA biogenesis selectively excludes co-packaging of Atg8. Although Atg8-containing PASs may nucleate ERA formation, the fluorescence microscopy images show that their localization remains distinct. If a similar process occurred during formation of classical autophagosomes induced by nitrogen starvation, the less-selective sequestrations of surrounding cytosol might non-selectively co-package Atg8 in proximity. Second, ERAs do not fuse with vacuoles when UPR-inducing conditions are maintained. The role of ERAs in the face of ongoing folding stress would therefore primarily be one of sequestration rather than degradation. Consistent with this idea, vps4Δ pep4Δ
cells lacking vacuolar proteases can live in UPR-inducing conditions despite the fact that they are already sick under normal growth conditions. Cells that are unable to form autophagosomes, however, die upon exposure to folding stress. This is in contrast to macroautophagy during nitrogen starvation, which has the primary purpose to cannibalize portions of the cytoplasm to provide recycled metabolites to the starving cells. vps4Δ pep4Δ
cells cannot degrade autophagocytosed material and therefore die under these conditions [78
]. Either of these two possibilities further supports the notion that ERAs have distinct properties and/or have a distinct fate from classical starvation-induced autophagosomes.
If the main function of ER-phagy is to counteract UPR-induced ER expansion, why do some cells already form ERAs despite ongoing folding stress? We can speculate that an expanded ER could allow cells to isolate potentially toxic unfolded proteins or aggregates into distinct regions of the ER; their preferential packaging into ERAs might serve to make this segregation complete, allow their eventual degradation in bulk, or prevent passing them on to daughter cells. ER-phagy may therefore not only be a homeostatic mechanism to control ER size, but could also serve a detoxification function under certain conditions. The existence of such an additional role of ERAs is supported by the observation that ERAs are not generated in cells expressing Hac1i, arguing that ERA formation under UPR-inducing conditions is not triggered by an expanded ER, but requires the actual presence of unfolded proteins. This idea may also explain why ERAs are found only in a fraction of the cells exposed to folding stress. ERA formation under UPR-inducing conditions might only set in when a large load of unfolded proteins has accumulated, and this may be the case only in some cells. UPR activation may induce almost all cells to eventually downsize their ER through ER-phagy, as judged by the widespread generation of extra PASs. However, only some cells may be challenged by unfolded proteins to such an extent that they trigger ER-phagy despite continuing ER stress. The activation of the Ire1-independent arm of the UPR, leading to S-UPR induction, might increase the fraction of cells that form ERAs during folding stress. It will be interesting to determine whether the fraction of cells containing ERAs increases once the folding stress ceases, as the homeostatic function of ER-phagy may then dominate over its detoxification function. In support of such a switch, we have seen in preliminary experiments that ERAs can fuse with vacuoles after UPR-inducing agents have been washed out and the cells recover from stress (S. Bernales and P. Walter, unpublished data). Thus the delivery of ERAs to the vacuole may be a controlled process that can be turned on and off. In summary, many questions about the molecular mechanisms and the cellular functions of ERAs formation remain, but it seems clear that ER-phagy serves as a countermeasure to ER expansion and helps to bring organelle abundance back into balance.
While this work was under review, Yorimitsu et al. [79
] independently reported that ER stress triggers autophagy. Their results confirm the transcriptional up-regulation of ATG8
and GFP-ATG8 foci formation reported here. Moreover, the authors show that ER stress–induced Atg8 is activated by lipid modification, and that the formation of GFP-ATG8 foci depends on ATG12,
indicating that these structures correspond to PASs seen during starvation-induced macroautophagy. One significant difference is that Yorimitsu et al. report that GFP-Atg8 is degraded, whereas we do not see degradation (). This difference is likely due to growth conditions, as they allow cells to go into stationary phase in which starvation-induced macroautophagy is turned on.
After our work was accepted for publication, Ogata et al. [80
] reported that autophagy is activated and promotes cell survival upon ER stress in mammalian cells.