The UFM1 system regulates ER-phagy through the ufmylation of CYB5R3

Protein modification by ubiquitin-like proteins (UBLs) amplifies limited genome information and regulates diverse cellular processes, including translation, autophagy and antiviral pathways. Ubiquitin-fold modifier 1 (UFM1) is a UBL covalently conjugated with intracellular proteins through ufmylation, a reaction analogous to ubiquitylation. Ufmylation is involved in processes such as endoplasmic reticulum (ER)-associated protein degradation, ribosome-associated protein quality control at the ER and ER-phagy. However, it remains unclear how ufmylation regulates such distinct ER-related functions. Here we identify a UFM1 substrate, NADH-cytochrome b5 reductase 3 (CYB5R3), that localizes on the ER membrane. Ufmylation of CYB5R3 depends on the E3 components UFL1 and UFBP1 on the ER, and converts CYB5R3 into its inactive form. Ufmylated CYB5R3 is recognized by UFBP1 through the UFM1-interacting motif, which plays an important role in the further uyfmylation of CYB5R3. Ufmylated CYB5R3 is degraded in lysosomes, which depends on the autophagy-related protein Atg7- and the autophagy-adaptor protein CDK5RAP3. Mutations of CYB5R3 and genes involved in the UFM1 system cause hereditary developmental disorders, and ufmylation-defective Cyb5r3 knock-in mice exhibit microcephaly. Our results indicate that CYB5R3 ufmylation induces ER-phagy, which is indispensable for brain development.

1. The revised manuscript now includes one experiment on endogenous (ie not overexpressed) CYB5R3 UFMylation (Fig 2a) (where the UFMylation components are not overexpressed) showing a very faint band putatively identified on a CYB5R3 western blot in lysates and enriched in CYB5R3 pulldowns only upon overexpression of the E3 components UFL1 and UFBP1. The authors' prime argument supporting the conclusion that CYB5R3 is a bona fide substrate rests on the data in Fig 2a from lane 5 where they IP CYB5R3 and claim to see the UFM1-CYB5R3 adduct in cells not overexpressing UFL1/UFBP1. However, the putative adduct actually does not appear to have the same mobility as the adducts in the E3 overexpressing lanes (compare Fig 2a lane 5 with lanes 3 and 7 in the anti CYB5R3 panel). In Fig 2b they show a small (<50%) reduction in the intensity of the putative endogenous CYB5R3 adduct following siRNA to UFBP1 when blotted for CYB5R3 but complete loss of what they argue to be the same band in the UFM1 blot. Since the same exact samples are being analyzed in the two blots, it is hard to understand the discrepancy. Perhaps they're not the same bands? 2. Remarkably, the same lysine specificity on CYB5R3 can be recapitulated in a cell free reaction with only the E1 and E2. In other words when there is enough CYB5R3 and charged E2 present selective UFMylation of a single lysine does not depend on a the E3-the specificity determining component in all UBL systems! This suggests that K214 is selectively modified by UFM1 because it is a better nucleophile than the other lysines and argues against UFMylation of CYB5R3 being a physiologically meaningful process.
3. The fact that UFSP2 KO increases the abundance of this species does not substantiate the claim of specificity because UFSP2, recognizes the UFM1 moiety and will cleave all UFM1 adducts.
4. The observation that UFM1-modified CYB5R3 is enriched in membrane fractions does not contribute to substantiating the authenticity of CYB5R3 being a substrate as both it and the UFM1 conjugation machinery are present at the same ER membrane Independently of one another. Moreover , the fractionation is problematic. Cytosol fractions contain ~50% contamination with ER based on BiP distribution in some figures (eg Fig 2) 5. Throughout the manuscript the authors quantify the ratio of UFMylated to unmodified CYB5R3 in bar graphs and note extremely low P-values supporting their conclusions, but it is unclear how these ratios are calculated. In most cases, the fraction of CYB5R3 that's modified is well below 10% (difficult to accurately assess because the unmodified form is so overexposed) yet their quantification shows ratios that must be normalized in some way that is not explained and not consistent. For example compare quantification of UFM1 modified CYB5R3 in Fig 2 with quantification of GFP-UFM1 modified CYB5R3 in Fig 5. Every figure in the paper has at least one panel using a similar quantification approach where the quantification does not seem to correspond to the actual data in the western blots.
6. In Fig 4 the authors use a pH-quenched fluorescent reporter attempt to demonstrate a role for UFMylation in promoting lysosomal degradation of CYB5R3. The visual data presented in panels b and c are impossible to interpret by eye so the authors quantify red puncta and use a statistical approach to claim a significant increase upon overexpression of UFL1/UFBP1 (and of course, the substrate cherry/GFP-CYB5R3) in starved cells. It is unclear why starvation was needed. Visual inspection of the graphical data show large error bars and many outlier points that are difficult to reconcile with the reported P-values, but I am not a statistician.
7. The authors have -somewhat arbitrarily decided to consider the third component of the tripartite UFM1 E3, CDK5RPAP3, as a "lysosome autophagy adaptor" based on data in a single report, while ignoring a substantial body of data supporting the view that this protein is a core constituent of the E3. Indeed, functional insight into the role of this E3 component was recently published on BioRxiv. While it can't be ruled-out that CDK5RAP3 could have two functions, it is concerning that its function in UFM1 ligase activity was not considred in the experimental design or in the data interpretation. It is difficult to understand their rationale for performing all the reconstitution and overexpression experiment with only two of the three components of this E3 (and indeed they show, disconcertingly, that the E3 is entirely dispensable for CYB5R3 UFMylation in vitro). In Fig 5a they finally test the role for CDK5RAP3 and find that its inclusion in the overexpression paradigm actually reduces the amount of UFMylated CYB5R3 (and UFBP1). While one explanation for this result is that CDK5RAP3 promotes the specificity of the E3 for its authentic substrate (in part by suppressing modification of non-authentic substrates) they choose to interpret the CDK5RAP3 effect as supporting a role in lysosomal degradation . But the experiment they present in support of this hypothesis, in Fig 5b (anti CYB5R3 blot) shows a remarkably weak effect of the lysosomal inhibitor BafA (compare lanes 5 and 6). Although at best there is a 20% increase in the GFP-UFM1-CYB5R3 band in the blot, strangely the quantification reports a highly significant (P = 0.003) effect. 10. The claim that CYB5R3 reductase activity is lost from Fig 2i is  All comments and concerns raised by the reviewer have been adequately addressed by the authors. I recommend publication of the revised manuscript.
Reviewer #4 (Remarks to the Author): There is no doubt that the manuscript is greatly improved and that the majority of concerns of the original reviewers have been addressed but a few points should be considered when making the final decision: Reviewer 1's comments are largely addressed adequately bar the following minor point: -I think confirmation that the new endogenous CYB5R3 from microsomal fractions and probing for UFL1 is done under denaturing conditions would help. The methods suggest this is the case but I would make it more explicit in the text and figure legend. If it is not, it should be.
Reviewer 2's comments are largely addressed adequately bar the following minor points: -there are still limitations on data showing the endogenous nature of some of the molecular details of the mechanism. Immunofluorescence of UFL1 localisation remains absent, but on balance it may be impossible to achieve this with current reagents. Short of knock-in of a tag to the endogenous locus using CRISPR/Cas9, which would be extremely time consuming and has no guarantee of methodological success, I think there is little else on balance that can be done here. Overall, although this point has not been addressed completely, I don't think it should preclude publication at this stage and in this journal.
-similarly, there is still a reliance on ectopic expression when using the Fluoppi system to show ATG8 protein interaction with CDK5RAP3. Here I think the authors should at least provide additional data to show that the expression level of these proteins is comparable with endogenous and that mutation of known binding interfaces on ATG8 proteins, e.g. the LDS (if these are indeed involved in CDK5RAP3 interaction) can reduce the Fluoppi signal, thus validating the specificity of this system for detecting ATG8-adaptor/receptor interactions.
All reviewer 3's comments appear to be addressed adequately.
My additional comments -while I do not wish to add undue burden to the revision of this manuscript there is a fundamental point that is not specifically identified by any of the three prior reviewers but that I think is critical to draw the major conclusions of the manuscript. That is, proof of CYB5R3 turnover by autophagy is not per se proof of ER-phagy, despite the outer ER localisation of at least a substantial population of this protein. The authors should really use an abundant general ER marker and show degradation by ER-phagy occurs in an at least partially CYB5R3-dependent manner. For instance, ss-mRFP-GFP-KDEL, or tandem fluorescent RAMP4, etc., could be employed in a tandem fluorescent flux assay, analogously to how CYB5R3 is currently used. Similarly, the CYB5R3 positive autophagic foci identified by colocalisation with autophagy proteins should be demonstrated to contain other ER material via microscopy. N.B. I apologise if some of this was present in the previous version of the manuscript and has been removed, as I have not seen this draft. The response to Reviewer 1 might hint at this with regard to the mRFP-GFP-KDEL. I wasn't completely clear on this.
Reviewer #5 (Remarks to the Author): The revised manuscript has been improved significantly. However, as the reviewer #1 stated: the principal conclusion of the work -that CYB5R3 is a bona fide endogenous substrate of UFMylation is still unconvincing. He also raised many other issues, which are reasonable.
Although almost all the experiments are done in the overexpressed system and the finial in vivo experiments could not guarantee that the effect is caused by the ufmylation BUT NOT by other lysine modifications at the position 214, currently there are no better experiment techniques which could solve this problem.
There are still some minor issues in the manuscript, which need to be fixed before the manuscript is publishable. Because some issues require further experiments, therefore, I suggested a major revision of the manuscript. Please see below for other issues.
1. Figure 1a: The BiP distribution is different in the USFP2 KO cells? What is the reason for this? It should be discussed in the manuscript.  Figure 1d showed that UFL1 is important for the ufmylation of CYB5R3. However, in Figure 1f, ufmylation of CYB5R3 occurs in the absence of UFL1. How could this happen? What caused this discrepancy?
3. Figure 2a: Why is there no heavy chain but with intense light chain? Figure 2a and 2b: Add "Relative" in the y axis of the statistical data. Figure 2b: IP and WB are required to confirm the ufymylation of endogenous CYB5R3. It is hard to tell whether the bands are specific or not if no IP is performed to eliminate the effect from other interacting proteins. The current images showed that there is a slightly less intense band when UFL1 and UFBP1 is knocked down in the anti-CYB5R3 blot. However, in the anti-UFM1 WB, this band is completely disappeared. This inconsistent might indicate this band is not ufmylated CYB5R3. A stricter experiment should be provided to confirm this result.  Figure S6: The fractionation is not perfect. GAPDH is also present in the microsomal fraction and is not consistent under different conditions. (It is not a big problem). 11. In the Discussion, the authors conclude that 2) ufmylated CYB5R3 interacts with the E3 ligase component UFBP1 to facilitate further ufmylation of CYB5R3. However, there is no sufficient evidence to conclude this statement.

Supplementary
12. There are some typos for example, GFP-tagged UFL1 (GFP-UFL1) were...; bonafide Reviewer #6 (Remarks to the Author): My review of the manuscript is only based on the high-speed AFM part, given that this is my area of expertise. I will refrain from commenting on the much broader other aspect of the paper.
Appreciate the efforts that he authors have put in to investigating CYB5R3 with high-speed AFM to observe structural changes. Unfortunately very little information is given in the manuscript about the HS-AFM data, which makes it very difficult to interpret. Based on what criteria did the authors attribute the closed and open states of the molecule? What kind of differences do the authors expect based on the crystallography data? It seems to me that the fluctuations comparing images denoted as "O" with each-other are similar to the fluctuations comparing "O" with "C", or "C" with "C". HS-AFM experiments always have significant differences from experiment to experiment. How many different experiments have the authors performed? How many molecules have they observed per experiment? In how many cases did they see a convincing difference between "O" and "C"? Thank you for your valuable comments, which helped us revise and strengthen our original manuscript. According to the comments raised by Reviewers 1 and 5, we made the following changes.

1.
To prove that CYB5R3 is a bona fide substrate for UFM1, we carefully re-performed several experiments to detect endogenous ufmylated CYB5R3 and the knockdown-induced 15 suppression of UFL1 or UFBP1 ( Fig. 2a and b).

2.
To demonstrate the requirement for E3 in CYB5R3 ufmylation, we performed an in vitro ufmylation assay (Fig. 1g).

3.
We delineated the quantification and statistical methods in this study.

4.
In the revised manuscript, we discussed the role of CDK5RAP3 as an E3 regulatory 20 subunit with reference to related studies.
For specific details, please see our reply to each comment below.

Comment 1
The revised manuscript now includes one experiment on endogenous (ie not overexpressed) 25 CYB5R3 UFMylation (Fig 2a)  argue to be the same band in the UFM1 blot. Since the same exact samples are being analyzed in the two blots, it is hard to understand the discrepancy. Perhaps they're not the same bands?

Reply 1
Thank you for pointing this out. We regret that the results presented in the original manuscript 40 were unclear. We agree with the reviewer that this experiment is key to assuring that CYB5R3 is an authentic substrate of UFM1. Therefore, we thoroughly optimized the experimental settings and performed the experiment multiple times. As shown in Fig. 2a of the revised manuscript, immunoprecipitation of microsomal fractions with CYB5R3 antibody revealed endogenous ufmylated CYB5R3, and the same migrated band was significantly increased by 45 overexpression of E3 (UFL1 and UFBP1). Furthermore, as shown in Fig. 2b of the revised manuscript, we showed that knockdown of UFL1 or UFBP1 significantly decreased the level of endogenous ufmylated CYB5R3.

Comment 2 50
Remarkably, the same lysine specificity on CYB5R3 can be recapitulated in a cell free reaction with only the E1 and E2. In other words when there is enough CYB5R3 and charged E2 present selective UFMylation of a single lysine does not depend on a the E3-the specificity determining component in all UBL systems! This suggests that K214 is selectively modified by UFM1 because it is a better nucleophile than the other lysines and argues against UFMylation of 55 CYB5R3 being a physiologically meaningful process.

Reply 2
This in vitro reaction is a very specific experimental system, i.e., an artificial system for producing large amounts of ufmylated CYB5R3. First, it used an excess quantity of recombinant UFM1 (9 60 μM), UBA5 (2 μM), and UFC1 (5 μM). Second, instead of purified CYB5R3ΔN26, a lysate of E.
coli expressing GST-CYB5R3ΔN26, was mixed with these recombinant enzymes. Under these conditions, CYB5R3 was covalently bound to UFM1 without E3. We cannot explain the exact mechanism involved, so we omitted this data from Figure 1. However, CYB5R3 covalently bound to UFM1 is a valuable tool, so we used it for the pull-down experiments ( Fig. 3b and g). As pointed out by Reviewers 1 and 5, we agree that in the original manuscript there were problems in demonstrating the specificity of E3. To solve this issue, we expressed FLAG-UFL1 and UFBP1-MYC in UFC1-deficient HEK293T cells and purified the FLAG-UFL1 and UFBP1-MYC complex by immunoprecipitation with anti-FLAG antibody (anti-DDDDK-tagged pAb-agarose). In vitro ufmylation assays were performed using recombinant UFM1, UBA5, 70 UFC1, the purified E3 complex, and a microsomal fraction prepared from UFBP1-deficient HEK293T cells or recombinant CYB5R3ΔN26 from E. coli. As shown in Fig. 1g of the revised manuscript, ufmylated CYB5R3 was not formed by the addition of UFM1, UBA5, and UFC1 to microsomal fractions, but it was formed by adding them together with purified E3 complex. By contrast, no ufmylated CYB5R3 was formed in the assay with recombinant CYB5R3ΔN26 75 ( Supplementary Fig. S3c). This implies that E3 and one or more factors in the microsomal fractions are required for ufmylation of CYB5R3. We presented the aforementioned points in the revised manuscript.

Comment 3 80
The fact that UFSP2 KO increases the abundance of this species does not substantiate the claim of specificity because UFSP2, recognizes the UFM1 moiety and will cleave all UFM1 adducts.

Reply 3
The comment may be correct. However, it has been shown that RPL26, which is considered to 85 be an authentic UFM1 substrate, is also increased by ablation or knockdown of UFSP2 Therefore, we believe that the data showing that UFSP2 knockout increases ufmylated CYB5R3, together with Figs. 2a, 2b (detection of endogenous ufmylated CYB5R3 and inhibition of its formation by knockdown of E3 components), and 1f (requirement for E3 in the 90 in vitro ufmylation assay) in the revised manuscript, indicate that CYB5R3 is a substrate of UFM1.

Comment 4
The observation that UFM1-modified CYB5R3 is enriched in membrane fractions does not 95 contribute to substantiating the authenticity of CYB5R3 being a substrate as both it and the UFM1 conjugation machinery are present at the same ER membrane Independently of one another. Moreover, the fractionation is problematic. Cytosol fractions contain ~50% contamination with ER based on distribution in some figures (eg Fig 2). 100 Reply 4 As the reviewer pointed out, this data is only supportive evidence. However, now that the key issues of Fig. 2a and 2b (detection of endogenous ufmylated CYB5R3 and inhibition of its formation by knockdown of E3 components) and Fig. 1f (requirement for E3 in vitro ufmylation assay) are resolved, we would like to keep this data in the revised manuscript as well. In 105 confirming our fractionation experiments, we believe that the Bip antibody we used in the blots was problematic. In fact, blots with anti-calnexin antibody (Santa Cruz Biotechnology, sc-46669), which has been used in many papers (e.g., Rajagopalan et al., Science, 263:387-90, 1994), showed clearer fractionation in comparison with the Bip antibody. Throughout the manuscript the authors quantify the ratio of UFMylated to unmodified CYB5R3 in bar graphs and note extremely low P-values supporting their conclusions, but it is unclear how these ratios are calculated. In most cases, the fraction of CYB5R3 that's modified is well below 10% (difficult to accurately assess because the unmodified form is so overexposed) yet 115 their quantification shows ratios that must be normalized in some way that is not explained and not consistent. For example compare quantification of UFM1 modified CYB5R3 in Fig 2 with quantification of GFP-UFM1 modified CYB5R3 in Fig 5. Every figure in the paper has at least one panel using a similar quantification approach where the quantification does not seem to correspond to the actual data in the western blots. 120

Reply 5
We apologize for not specifying how we calculated the ratio of ufmylated CYB5R3 to CYB5R3. The signals of ufmylated CYB5R3 and CYB5R3 bands were quantified using Multi Gauge V3.2 (FUJIFILM Corporation), and the ratio of ufmylated CYB5R3 to CYB5R3 was 125 calculated. The ratio was 1 when neither overexpression nor knockdown of any E3 components was performed ( Fig. 2a and 2b). P values were determined by t-test (Fig. 2a) or Šidák's test after a two-way factorial analysis of variance (ANOVA) (Fig. 2b). The details are described in the corresponding Figure legends of the revised manuscript. In addition, according to a comment (comment 3) by Reviewer#5, in the y-axis of the graph, "UFM1~CYB5R3/CYB5R3" was 130 substituted with "Relative value of the UFM1~CYB5R3/CYB5R3 ratio." As the reviewer indicated, the amount of ufmylated CYB5R3 was low compared to that of free CYB5R3 under normal culture conditions. We need to find conditions for inducing CYB5R3 ufmylation such that anisomycin treatment (i.e., protein synthesis inhibition) results in ufmylation of RPL26 (Walczak et al., PNAS, 116:1299-1308, 2019Wang et al., Cell Res, 135 30:5-20, 2019). However, we believe that this is beyond the scope of this study.
Currently we are not sure why CYB5R3 ufmylation is induced more effectively when using GFP-tagged UFM1 than FLAG-tagged or endogenous UFM1. However, our data show that the effect of the E3 component on CYB5R3 ufmylation is consistent regardless of Tag type.
To quantify the number of RFP-positive punctae (the number of lysosomes encapsulating CYB5R3-positive ER) per cell, we used a Benchtop High-Content Analysis System (CQ1; Yokogawa Electric, Tokyo, Japan) and CellPathfinder software (Yokogawa Electric). This system allowed us to quantify fluorescently labeled lysosomes in multiple 160 samples at the same time and without bias. The statistical analysis was conducted by Šidák's test after two-way ANOVA. inhibitor BafA (compare lanes 5 and 6). Although at best there is a 20% increase in the GFP-UFM1-CYB5R3 band in the blot, strangely the quantification reports a highly significant (P = 0.003) effect.
As mentioned above, we were able to show both in vivo and in vitro that at least UFL1 190 and UFBP1 are required for ufmylation of CYB5R3 (please see Replies 1 and 2).
As the reviewer pointed out, CDK5RAP3 may act positively on authentic substrates such as RPL26, and negatively on other substrates (Peter et al., EMBO J e111015). This is supported by the fact that different proteins show either decreased or increased ufmylation in CDK5RAP3 knockout mouse tissues (Yang et al., Development, 146, dev169235, 2019). We 195 have described this possibility in the revised manuscript. Actually, we noticed that while CDK5RAP3 had the ability to promote ufmylation of RPL26, it had an inhibitory effect on the ufmylation of UFBP1 and CYB5R3 (data not shown; if the reviewer would like us to present the data, we will do so as "Figure for Reviewer 1").
Statistical analysis of western blots was performed as described in Reply 5. The 200 signals of ufmylated CYB5R3 and CYB5R3 bands were quantified using Multi Gauge V3.2 (FUJIFILM Corporation), and the ratio of ufmylated CYB5R3 to CYB5R3 was calculated. P values were determined by Šidák's test after two-way ANOVA.

Comment 8 205
The structure/model in Fig 2g claims to  The claim that CYB5R3 reductase activity is lost from Fig 2i is  Reply 9 In Figure 2i, the same amount of CYB5R3 and ufmylated CYB5R3 proteins were used. Figure   2j shows that wild-type CYB5R3 (not ufmylated) and mutant CYB5R3 (K214R) have similar reductase activity, implying that the K214 mutant has the same ability as wild type to possess 225 FAD.

General comment 230
All comments and concerns raised by the reviewer have been adequately addressed by the authors. I recommend publication of the revised manuscript.

Reply
Thank you very much for your positive evaluation. 235 Reviewer #4:

General comment
There is no doubt that the manuscript is greatly improved and that the majority of concerns of 240 the original reviewers have been addressed but a few points should be considered when making the final decision:

Reply
Thank you very much for your positive evaluation and valuable suggestions that helped us 245 improve the revised manuscript.

Reply 1
We performed immunoprecipitation with the microsomal fraction denatured with 1% sodium 255 dodecyl sulfate (SDS). Briefly, after cytoplasmic and microsomal fractionation, 10 μl of 10% SDS was added to 100 μl of both fractions. In the next step, 900 μl of IP buffer (20 mM Tris-HCl [pH 7.5], 150 mM NaCl, 1 mM EDTA, 1% NP40 and 1% TX-100), 2 μl of CYB5R3 antibody (GTX84646; GeneTex, Irvine, CA, USA), and 10 μl of Protein G Sepharose 4 Fast Flow (Cytiva, Marlborough, MA, USA) were added to the 100 μl of cytoplasmic and 260 microsomal fractions, and the mixture was mixed under constant rotation at 4C for 3 h. The immunoprecipitates were washed five times with ice-cold IP buffer. The complex was boiled for 5 min in SDS sample buffer in the presence of β-mercaptoethanol to elute proteins. We describe the details in the Material and Methods section of the revised manuscript. We also indicated in the corresponding figure legends that these experiments were performed under denatured 265 conditions.

Comment 2
Reviewer 2's comments are largely addressed adequately bar the following minor points: -there are still limitations on data showing the endogenous nature of some of the molecular 270 details of the mechanism. Immunofluorescence of UFL1 localisation remains absent, but on balance it may be impossible to achieve this with current reagents. Short of knock-in of a tag to the endogenous locus using CRISPR/Cas9, which would be extremely time consuming and has no guarantee of methodological success, I think there is little else on balance that can be done here. Overall, although this point has not been addressed completely, I don't think it should 275 preclude publication at this stage and in this journal.

Reply 2
Thank you for this comment. Unfortunately, our UFL1 antibody was unable to demonstrate subcellular localization of endogenous UFL1. We agree that Tag knock-in to the UFL1 locus 280 would be a powerful tool, but we have not routinely performed such experiments in our lab and they would be time consuming. Therefore, we decided to use the data from GFP-tagged UFL1.

ATG8 protein interaction with CDK5RAP3. Here I think the authors should at least provide additional data to show that the expression level of these proteins is comparable with endogenous and that mutation of known binding interfaces on ATG8 proteins, e.g. the LDS (if these are indeed involved in CDK5RAP3 interaction) can reduce the Fluoppi signal, thus
validating the specificity of this system for detecting  Reply 3 Expression levels of Azami Green-tagged CDK5RAP3 and its mutants (W269A, W294A, W312A) were comparable to the endogenous one ( Supplementary Fig. S8b in the revised manuscript). Likewise, expression levels of Ash-tagged LC3B and GABARAP were also 295 similar to the endogenous ones ( Supplementary Fig. S8b in the revised manuscript). In addition, mutants of three LC3-interacting regions of CDK5RAP3 (W269A, W294A, W312A), which have been reported to be required for interaction with LC3B or GABARAP (Stephani et al., eLife, 9:e58396, 2020), lost this interaction in the Fluoppi assay ( Supplementary Fig. S8c in the revised manuscript). 300

Reply 4
In the original manuscript, we utilized a doxycycline-inducible ER-phagy reporter containing an N-terminal ER signal sequence followed by tandem monomeric RFP and GFP sequences and the ER retention sequence KDEL (Chino et al., Mol Cell, 74:909-921, 2019). We found that 320 UFL1 and UFBP1 expression significantly increased the number of RFP-positive punctae in both nutrient-rich and nutrient-deprived conditions. However, in the first round review of this manuscript, Reviewer#1 pointed out that the experiment with RFP-GFP-KDEL does not provide direct information about the role of CYB5R3, so we have used RFP-GFP-CYB5R3 and its mutants instead. We hope the reviewer understands this process. 325 Reviewer #5:

General comment
The revised manuscript has been improved significantly. However, as the reviewer #1 stated: 330 the principal conclusion of the work -that CYB5R3 is a bona fide endogenous substrate of UFMylation is still unconvincing. He also raised many other issues, which are reasonable. There are still some minor issues in the manuscript, which need to be fixed before the manuscript is publishable. Because some issues require further experiments, therefore, I suggested a major revision of the manuscript. Please see below for other issues. 340 Reply Thank you for your valuable comments that have helped us strengthen our revised manuscript.
According to comments raised by Reviewers 1 and 5, we made the following changes.

2.
To prove that CYB5R3 is a bona fide substrate for UFM1, we performed several experiments to detect endogenous ufmylated CYB5R3 and knockdown-induced suppression of UFL1 or UFBP1 ( Fig. 2a and b).

3.
We delineated the quantification and statistical methods in this study.
For specific details, please see our reply to each comment below. Reply 1  In the fractionation confirmation in Fig. 1a, we believe that the Bip antibody we used in the 360 blots was problematic. In fact, blots with anti-calnexin antibody (Santa Cruz Biotechnology, sc-46669), which has been used in many papers (e.g., Rajagopalan et al., Science, 263:387-90, 1994), showed clearer fractionation regardless of the presence or absence of UFSP2.
 The double bands detected below ufmylated CYB5R3 in the blot using the UFM1 antibody 365 in Fig. 1e are likely endogenous proteins (probably RPL26) to which UFM1 is covalently bound. In Fig. 1e of the revised manuscript, they are referred to as UFM1 conjugates. This in vitro reaction is a very specific experimental system, i.e., an artificial system for producing large amounts of ufmylated CYB5R3. First, it used an excess quantity of recombinant UFM1 (9 380 μM), UBA5 (2 μM), and UFC1 (5 μM). Second, instead of purified CYB5R3ΔN26, a lysate of E.
coli expressing GST-CYB5R3ΔN26, was mixed with these recombinant enzymes. Under these conditions, CYB5R3 was covalently bound to UFM1 without E3. We cannot explain the exact mechanism involved, so we omitted this data from Figure 1. However, CYB5R3 covalently bound by UFM1 is a valuable tool, so we used it for the pull-down experiments ( Fig. 3b and g). 385 In the original manuscript, as pointed out by Reviewers 1 and 5, we agree that there were problems in demonstrating the specificity of E3. To solve this issue, we expressed FLAG-UFL1 and UFBP1-MYC in UFC1-deficient HEK293T cells and purified the FLAG-UFL1 and UFBP1-MYC complex by immunoprecipitation with anti-FLAG antibody (anti-DDDDK-tagged pAb-agarose). In vitro ufmylation assays were performed using recombinant UFM1, UBA5, 390 UFC1, the purified E3 complex, and a microsomal fraction prepared from UFBP1-deficient HEK293T cells or recombinant CYB5R3ΔN26 from E. coli. As shown in Fig. 1g of the revised manuscript, ufmylated CYB5R3 was not formed by the addition of UFM1, UBA5, and UFC1 to microsomal fractions, but it was formed by adding them together with purified E3 complex. By contrast, no ufmylated CYB5R3 was formed in the assay with recombinant CYB5R3ΔN26 395 ( Supplementary Fig. S3c). This implies that E3 and one or more factors in the microsomal fractions are required for ufmylation of CYB5R3. We presented the aforementioned points in the revised manuscript.

Reply 3 415
Thank you for the valuable comments that helped us improve and strengthen our manuscript. As pointed out by Reviewers 1 and 5, the experiments presented in Fig. 2 include those that form the foundation of this study, and we substantially optimized these experiments with the latest attention.
 Mouse monoclonal antibody against CYB5R3 (GTX84646; GeneTex) was used for 420 immunoprecipitation of endogenous CYB5R3 in microsomal fractions of HEK293T cells.
The immunoprecipitants were subjected to immunoblot analysis with a rabbit monoclonal antibody against UFM1 (ab109305; Abcam). The secondary antibody used for its detection was Peroxidase AffiniPure Goat Anti-Rabbit IgG (H+L) (Jackson ImmunoResearch Laboratories, Code: 111-035-144). Therefore, we did not expect to detect either heavy or 425 light chains. However, the product information for the secondary antibody states that it may cross-react with immunoglobulins of other animal species. Perhaps there is a crossover to light chains in mice.
 According to the reviewer's suggestion, we added "Relative" to the y-axis of the statistical data in Fig. 2a and 2b. 430  As shown in Fig. 2a of the revised manuscript, immunoprecipitation of microsomal fractions with anti-CYB5R3 antibody revealed endogenous ufmylated CYB5R3, and the same migrated band was significantly increased by overexpression of E3 (UFL1 and UFBP1). Furthermore, as shown in Fig. 2b of the revised manuscript, we showed that knockdown of UFL1 or UFBP1 significantly decreased the level of endogenous ufmylated CYB5R3. These results were confirmed by immunoblotting with both anti-CYB5R3 and anti-UFM1 antibodies.
 CYB5R3 is anchored to the membrane by myristoylation of the second glycine at the Nterminus, and the 24 amino acids up to the N-terminal form the membrane-binding domain (Murakami et al., J Biochem, 105:312-7, 1989). Thus, CYB5R3 that was detected in the 440 cytoplasmic fraction upon overexpression is considered to be the N-terminal truncated form. In fact, the CYB5R3 detected in this fraction is almost as mobile as ΔN26, in which the N-terminal 26 amino acids are deleted. The aforementioned points are explained in the revised manuscript.
 Figure 2j shows that wild-type CYB5R3 (not ufmylated) and its mutant (K214R) have 445 similar reductase activity, while ufmylated CYB5R3 has lost this activity. Reply 4  As shown in Fig. 1i (Supplementary Fig. S6 in the revised manuscript), ufmylated CYB5R3ΔN26 is detected as a single band immediately after purification. The ufmylated CYB5R3ΔN26 shown in Fig. 1i (Supplementary Fig. S6 in the revised manuscript) was 460 stored at −80°C for 5 months and utilized for the experiment in Fig. 3b. This means that the faster migrating ufmylated CYB5R3ΔN26 proteins detected in Fig. 3b are most likely degradation products that resulted from long-term storage and freeze-thawing. We have indicated that they are degradation products in the corresponding Figure legend   We do not intend to exclude the possibility that UFBP1 binds to free UFM1. In fact, as the reviewer pointed out, UFBP1 bound not only to ufmylated CYB5R3 but also to free UFM1 (Fig. 3g). We noted in the revised manuscript that UFBP1 has the ability to bind to free 475 UFM1.  Reply 5 According to this suggestion, data for three genotype mice are shown in Fig. 5e in the revised manuscript. One Cyb5r3 +/+ , three Cyb5r3 K214R/+ , and four Cyb5r3 K214R/K214R mice were used in this study. Since we used the data (thickness of cortex / distance from aqueduct to lateral edge) of the left and right cortices of the Cyb5r3 K214R/+ and Cyb5r3 K214R/K214R mice and the right 485 cortices of the wild-type and Cyb5r3 K214R/K214R mice, statistical analysis was performed on the wild-type and Cyb5r3 K214R/+ mice together as controls, and these were compared to the Cyb5r3 K214R/K214R mice.

Comment 6 490
Supplementary Figure S5C: What is the lowest band in Western blotting image?

Reply 6
It was thought that it might be a degradation product of purified MBP-tagged UFL1. Therefore, we purified MBP-UFL1 again and performed the same experiment with the MBP-UFL1, and 495 obtained reasonable results (disappearance of fast migrating bands).

Comment 7
Supplementary Figure S6: The fractionation is not perfect. GAPDH is also present in the microsomal fraction and is not consistent under different conditions. (It is not a big problem). 500

Reply 7
We acknowledge that the fractionation in Supplementary Fig. S6 (Supplementary Fig. S9 in the revised manuscript) is not perfect. Except for Supplementary Fig. S6 (Supplementary Fig. S9 in the revised manuscript), the fractionation experiments shown in this study were performed with 505 HEK293T cells. By contrast, in Supplementary Fig. S6 (Supplementary Fig. S9 in the revised manuscript), we used mouse embryonic fibroblasts (MEFs) isolated from wild-type and CYB5R3 K214R/K214R knock-in mice. The detection of ufmylated CYB5R3 in MEFs was more difficult than in HEK293T cells, probably due to its low abundance. Therefore, detection required fractionating a large number of MEFs. As a result, there was some influx of the 510 cytoplasmic fraction. As the reviewer pointed out, since this issue (some influx of the cytoplasmic fraction) is not essential and would require a great deal of effort to prevent, we did not redo this experiment during this revision. We hope that you understand.

Comment 8 515
Supplementary Reply 8 "Peptides (95%)" indicates the number of distinct peptides that were identified with at least 95% 520 confidence by ProteinPilot. We added this explanation as well as those for other columns in the text below Supplementary Table S1 and in Supplementary Dataset1, sheet "Index."

Comment 9
Supplementary Table 2 and Supplementary Table 3: Please pay attention to the significant 525 digits.

Reply 9
Thank you so much for pointing that out. We checked the significant digits of the raw data and corrected the tables accordingly. 530

Reply 10
On page 4, line 120 in the original manuscript we wrote the following: "When the E3 component UFL1 was co-overexpressed, we observed a band representing the MYC-UFM1~CYB5R3-His-FLAG conjugate, which was further enhanced by the expression of UFBP1 (Fig. 1D)." We concluded that UFL1 and UFBP1 are required for the ufmylation of CYB5R3 (Figs. 1g, 2a and 2b, etc., in the revised manuscript). 540

Comment 11
In the Discussion, the authors conclude that 2) ufmylated CYB5R3 interacts with the E3 ligase component UFBP1 to facilitate further ufmylation of CYB5R3. However, there is no sufficient evidence to conclude this statement. 545

Reply 11
We acknowledge that our model is not adequately substantiated by the data presented, as pointed out by the reviewer. In the revised manuscript, we have weakened our argument and removed this model from the main Figure (Fig. 5f) and moved it to Supplementary Fig. S10. 550

Reply 12 555
Thank you for the comment. We corrected the typos throughout the revised manuscript. have they observed per experiment? In how many cases did they see a convincing difference between "O" and "C"?

Reply
We would like to thank this reviewer for this important comment on our HS-AFM experiment. 575

> Based on what criteria did the authors attribute the closed and open states of the molecule?
We divided the images into two groups (the closed and open states) depending on the distances between the two lobes of the molecule (FAD and NADH domains). In the initial version of the manuscript, we used a tip-scan-type HS-AFM apparatus with 1.0-nm/pixel resolution. To obtain 580 clearer images, we re-examined CYB5R3 with a sample-scan-type HS-AFM apparatus. In the revised manuscript, we obtained CYB5R3 images with 0.6-nm/pixel resolution. Because the molecules sometimes change their orientation on the mica surface, it was difficult to divide all the images into two groups (the closed and open states). Therefore, among all the images obtained, we chose and analyzed those in which the two lobes (FAD and NADH domains) were facing up. 585 We measured the distances between the two lobes and obtained histogram data using Kodec software. All of the histograms from six molecules were well reproduced by double-Gaussian distribution whose peaks were at 2.62 ± 0.07 and 4.71 ± 0.11 nm. On the other hand, we simulated AFM images of CYB5R3 (PDB ID: 1umk) using BioAFMviewer software. Then we obtained the cross-sectional profile of the two lobes in the simulated AFM image. The distance between the 590 two lobes was ~2.5 nm, a value close to one of the two peaks of the fitted curve of the real HS-AFM data. In the revised manuscript, representative images of the two conformations are shown in Fig. 2f with simulated AFM data. We manually modeled an open conformation of CYB5R3 by changing the arrangement between the NADH and FAD domains at the hinge region (Leu147) in order to yield a predicted AFM image similar to the experimental one ( Fig. 2f and g). 595 >How many different experiments have the authors performed? How many molecules have they observed per experiment?
In the revised manuscript, we performed three different experiments and collected data from six 600 different molecules in total.
>In how many cases did they see a convincing difference between "O" and "C"?
Based on the histograms of the distances between two lobes, we observed the open state and the 605 closed state at similar frequencies.

REVIEWERS' COMMENTS
Reviewer #4 (Remarks to the Author): I believe that my concerns have been addressed.
I would recommend that in the final revision that the data with the standard ER-phagy reporter (ss-mRFP-GFP-KDEL) that was removed from the first draft of the manuscript is now presented.
Reviewer #5 (Remarks to the Author): The revised manuscript is significantly improved. Below are some minor issues that the authors should take care of.
1. It is hard to tell that the regulation of ER-phagy is indeed caused by the ufmylation of CYB5R3. Similarly, it is not convincing that the microcephaly occurred in the Cyb5r3K214R/K214R knock-in mice is caused by the deficiency of its ufmylation. One cannot rule out the effect of other types of posttranslational modifications on lysine residues. Therefore, it is highly suggested that the authors should also discuss the limitations of their study in the discussion section.
2. Figure 4b and c: Please add magnification or enlarge for the middle panels.
3. Figure 5c: Please add information for the color code at the left side of the images.
Reviewer #6 (Remarks to the Author): The authors have adequately addressed my concerns. The new images are much more convincing.

Reviewer #4:
General comment I believe that my concerns have been addressed.

Reply 5
Thank you for your detailed and thorough peer review.
Comment 1 I would recommend that in the final revision that the data with the standard ER-phagy reporter (ss-mRFP-GFP-KDEL) that was removed from the first draft of the manuscript is now 10 presented.

Reply 1
Thank you for this suggestion. During the revision, we preformed the ER-phagy assay with ss-mRFP-GFP-KDEL using a Benchtop High-Content Analysis System and CellPathfinder 15 software without bias. Although ER-phagy tended to be induced by ufmylation of CYB5R3, we did not recognize significant difference in the assay with mRFP-GFP-KDEL (Supplementary Figure S8 in the revised manuscript). We concluded that since ufmylated CYB5R3 is involved in ER-phagy for restricted ER subpopulation, ssRFP-GFP-KDEL that locates in whole ER is unable to monitor the ufmylated CYB5R3-mediated ER-phagy. We stated this in the result 20 section in the revised manuscript.

General comment
The revised manuscript is significantly improved. 25 Reply Thank you very much for your positive evaluation.

Minor comments 30
Comment 1 It is hard to tell that the regulation of ER-phagy is indeed caused by the ufmylation of CYB5R3.
Similarly, it is not convincing that the microcephaly occurred in the Cyb5r3K214R/K214R knock-in mice is caused by the deficiency of its ufmylation. One cannot rule out the effect of other types of post-translational modifications on lysine residues. Therefore, it is highly 35 suggested that the authors should also discuss the limitations of their study in the discussion section.

Reply 1
As the reviewer pointed out, there is a possibility that at least ufmylation-defective Cyb5r3 40 knock-in mice cause microcephaly due to effects other than ufmylation. Therefore, in the discussion section of the revised manuscript, we explained as follows: Though we do not exclude a possibility that the mutation inhibits other post-translational modification(s) except for the ufmylation, our results suggest that the defect in macro ER-phagy through the ufmylation of CYB5R3 is involved in the pathogenesis of RCM type II. 45 Comment 2 Figure 4b and c: Please add magnification or enlarge for the middle panels.

Reply 2 50
Thank you for the comment. According to this suggestion, we add "Enlarge" in the middle panels (Figure 5b and c in the revised manuscript).

Comment 3
Figure 5c: Please add information for the color code at the left side of the images. 55 Reply 3 We added the information at the left side of the images (Fig. 5b, c e and Fig. 6c in the revised manuscript).