SARS-CoV-2 envelope protein induces LC3 lipidation via the V-ATPase-ATG16L1 axis

  1. Carmen Figueras-Novoa
  2. Lewis Timimi
  3. Elena Marcassa
  4. Raquel Taveira-Marques
  5. Lorin Adams
  6. Ming Jiang
  7. Mary Wu
  8. Beatriz Montaner
  9. Kevin Ng
  10. Giuditta De Lorenzo
  11. Wilhelm Furnon
  12. Vanessa M. Cowton
  13. Nicole Upfold
  14. George Kassiotis
  15. Ruth Harvey
  16. Arvind H Patel
  17. Michael Howell
  18. Rachel Ulferts
  19. Rupert Beale

Reviewed by

Read the full article See related articles

Discuss this preprint

Start a discussion What are Sciety discussions?

Listed in

Log in to save this article

Abstract

Coronaviruses encode envelope (E), a structural component of the virion that is important for assembly and egress. E has proton channel activity that prevents premature rearrangement of the spike glycoprotein as virions encounter acidic compartments as they exit the cell. How infected cells respond to this pH disruption during coronavirus infection is unknown. Here we show that SARS-CoV-2 E ion channel activity triggers the proton pump V-ATPases to recruit the ATG16L1 complex during infection. This results in ATG8 molecules such as LC3B decorating perturbed compartments. This recruitment of autophagy machinery does not inhibit viral replication, rather SARS-CoV-2 exploits this response. Inhibition of the V-ATPase/ATG16L1 interaction using the Salmonella effector SopF inhibits SARS-CoV-2 replication. Careful distinction between use of the autophagic machinery from canonical macroautophagy is required in order to better understand coronavirus replication and for rational targeting of any potential host-directed therapies.

Article activity feed

  1. Review Commons

    Note: This response was posted by the corresponding author to Review Commons. The content has not been altered except for formatting.

    Learn more at Review Commons

    Reply to the reviewers

    We thank the reviewers for their positive comments. Our manuscript is to our knowledge the first to investigate the role of VAIL (V-ATPase—ATG16L1 induced LC3 lipidation), a form of CASM (Conjugation of ATG8s to single membranes) in SARS-CoV-2 replication. We demonstrate that SARS-CoV-2 Envelope (E) induces VAIL and this contributes to viral replication, including by using a reverse genetics system to make an E mutant virus. There have been many high quality studies examining the role of canonical autophagy in SARS-CoV-2 replication and our manuscript does not argue that all or even most LC3 lipidation during infection is via VAIL. We will try to make this point more clearly in the text. We do not think this detracts from the novelty and importance of our manuscript.

    *Reviewer #1 (Evidence, reproducibility and clarity (Required)): *

    • Figueras-Novoa et al present a short report demonstrating the induction of LC3 lipidation on single membranes by SARS-CoV-2 through a noncanonical autophagy pathway referred to as VAIL. The authors utilize elegant genetic tools to show that the induction of LC3 lipidation upon viral infection is mainly due to VAIL rather than canonical autophagy. They demonstrate that the activity of the viral E protein that can cause neutralization of acidic vesicles leads to the activation of non-canonical LC3 lipidation on single membranes. Interestingly, the authors also conclude that the impairment of VAIL leads to a reduction of viral load as a result of a defect in later stages of viral infection, although the underlying mechanism was not further explored. *

    • Overall, this is an elegant and well controlled study that provides a clear conclusion. I only have some minor comments.*

    We thank the reviewer for their assessment of our manuscript.

    In some experiments, LC3 lipidation does not appear to be fully disrupted upon VAIL inhibition (e.g. Fig.'s 1H, 3D, 4A). As other labs have shown that SARS-CoV2 blocks autophagic flux, this could be further clarified in this manuscript as both VAIL and autophagy may be co-induced upon viral infection.

    We agree with the reviewer that there is a contribution of canonical macroautophagy to the LC3B lipidation observed in SARS-CoV-2. We will extend the discussion in the manuscript to clarify this point for the readers.

    Can the authors test the induction of LC3 lipidation in cells expressing K490 mutant of ATG16L1 in ATG16L1 KO cells to compare them with ATG16L1-ATG13 double knockouts?

    The western blot in figure 3F (quantified in Figure 3G) shows LC3B lipidation in response to E expression in ATG16L1-ATG13 double knock out cells reconstituted with wild type ATG16L1 but not in cells complimented with ATG16L1 K490A mutant. We agree that the referee’s suggestion to perform these experiments in the context of infection would be informative. However in spite of numerous attempts, we have so far been unable to generate a cell clone fully devoid of ATG16L1 in a cell line that can be productively infected with SARS-CoV-2. For reasons unclear to us there appears to be a very low level of residual ATG16L1 activity despite multiple different CRISPR/Cas9 targeting attempts. The suggested complementation experiments might still be informative in the context of low level ATG16L1 expression so we will pursue this. Alternatively, as a contingency we can try to produce SARS-CoV-2 infectable cells with mutations in ATG16L1’s binding partner V1H, this interaction is required for VAIL. A further contingency could be to assess LC3B lipidation during infection and treatment with a Vps34 inhibitor, which inhibits canonical autophagy.

    *Minor points: *

    • The difference between Fig. 1F&G is unclear and why the authors are including both analyses. Similarly figures 4G&H.*

    We included both metrics to show that the decrease in LC3B lipidation in cells expressing SopF during infection is robust and observed in two separate readouts. While spot area measures the area of infected cells covered by GFP-LC3B fluorescence, spot intensity is a reading of the intensity of the area defined in an infected cell as being LC3 positive. Theoretically, these measurements could change in different ways. For example, if the same amount of lipidated LC3 were to distribute over a larger area of the cell. We prefer to keep both measurements in the manuscript.

    The authors should show boxed colocalisation of all images, including negative controls. For examples, the authors have shown boxed magnifications in only the lowest panel in Figure 2A but not the upper two panels. Figures 4E&F should include boxed examples. This serves to clarify both positive and negative colocalisation events.

    Boxed magnifications will be added to all images.

    • Reviewer #1 (Significance (Required)): *

    • Overall an elegant and well controlled study demonstrating the induction of non-canonical LC3 conjugation on single membranes (VAIL) during SARS-CoV2 infection. A further exploration of canonical autophagy (as previously published by others) in addition to VAIL would enhance this study.*

    As the reviewer noted, several excellent studies have explored canonical autophagy during SARS-CoV-2 infection, many of which we cite in our manuscript. Our focus, however, is to demonstrate that SARS-CoV-2 E induces LC3 lipidation via VAIL. We believe that exploring the diverse roles of canonical autophagy mechanisms in SARS-CoV-2 infection is beyond the scope of this study.

    *This study is of interest to researchers studying autophagy, viruses, immunology, single membrane LC3 lipidation, and lysosomes as well as potentially clinicians treating SARS-CoV2 infecteted individuals. *

    • This reviewer is experienced in autophagy research.*

    We thank the reviewer for this assessment of our manuscript.

    *Reviewer #2 (Evidence, reproducibility and clarity (Required)): *

    • Major Comments *

    • Figure 1D does not very clearly show an overlap between V1D and LC3B. Both proteins seem broadly present across the cell and there is no easily identifiable change in V1D distribution upon infection. As such the overlay may be purely stochastic. The authors should quantify the observed co-localization events across multiple cells and biological replicates and compare them to other protein(s) with a similar cellular distribution pattern.*

    We agree there is no obvious change in V1D staining on infection. The images in Figure 1D are purely intended to illustrate that LC3 and the V-ATPase can colocalise, not to demonstrate a change in V-ATPase distribution or to suggest a direct interaction. We will make this point more clearly in the text. We will also carry out analyses of the kind (see also response to the first two Minor Comments). We would be happy to provide an alternative method of visualising the V-ATPase (we could use any suitable antibody to the V-ATPase, or the bacterial effector SidK) if required. In response to reviewer 3’s comments, we will carry out a pull-down experiment to test the association of the V-ATPase and ATG16L1 during E expression, as this is a key interaction during VAIL activation.

    Based on Figure 2F the authors suggest that virus entry is unaffected by the inhibition of VAIL in early timepoints. However, according to the figure legend, the timepoint used is 7hpi, while 2D uses 24hpi. Some SARS-CoV-2 papers suggest 7-10 hours is sufficient time to release new virions (Ban-On et al., 2020). As such 7hpi can not necessarily be seen as an early time point. Did the authors test earlier ones? Also, based on this, would it be possible that the effects observed at 24hpi are actually secondary infections, meaning that the virus utilizes pathway components for virion production and a lack thereof reduces infectivity of newly formed virions? In this case it would be interesting to set up an assay that can distinguish between primary and secondary infection to study both individually more closely.

    Whereas 7 hours may be sufficient to release new virions, it is not sufficient to establish infections in other cells – this is why we chose that time point. The observation that there is no difference in the percentage of infected cells at 7 h p.i. (figure 2F) led us to suggest that viral entry is unaffected . We then confirmed this through the pseudovirus assay in Figure 2G, where no difference is found between SopF and mCherry expressing cells. For this assay, GFP-expressing, replication incompetent, lentiviral particles pseudotyped with Spike from different SARS-CoV-2 lineages were used to transduce mCherry and SopF expressing cells. A change in the percentage of GFP-positive cells would indicate an effect on viral entry, but no such change was observed in SopF-expressing cells.

    We agree with the reviewer that the effects observed at 24 hpi are likely due to a defect in subsequent rounds of infection, since no difference was observed at 7 hpi or with our pseudovirus assay. We will attempt to make this point in the text as clearly as possible.

    The authors nicely show in their study an involvement of VAIL in SARS-CoV-2 mediated LC3 lipidation. However, the observed effects are relatively moderate in several experiments, indicating that there may be another contributor to the observed phenotype. It would be nice to highlight this in the discussion and debate potential mechanisms that are causing the observed effects during infection.

    We agree with the reviewer’s analysis. We have discussed the contribution of canonical autophagy in the second paragraph of the discussion, but we will expand on this in a revised manuscript. E expression levels are moderate during infection, other structural proteins such as N and M are present in much higher amounts. Since E is the key protein in VAIL initiation, a moderate effect of VAIL inhibition in perhaps expected. Nonetheless this still plays a crucial role in the viral life cycle.

    *Minor Comments *

    • The re-localization events shown in Fig 3A should be quantified.*

    This quantification of GFP-LC3 relocalisation will be carried out and included.

    • The co-localization events displayed in Fig 4A should be quantified.*

    The quantification of V1D, E and GFP-LC3 will be carried out and included.

    For Figure 2H-K the authors perform KDs of ATG16L1 and ATG13. While the results for the two specific proteins are certainly convincing, the authors would strengthen their argument by testing additional proteins in the autophagy pathway to support their claim that VAIL but not autophagy affects protein abundance of N (OPTIONAL).

    As discussed in response to reviewer 1, we will attempt to infect ATG16L1 KO cells reconstituted with a K490A ATG16L1 mutant, which is an established tool and has been validated to be deficient in VAIL but not canonical autophagy.

    ***Referee cross-commenting** *

    • Overall I agree with the comments of my co-reviewers and I think the suggested experiments/comments are sensible. *
    • I in part already eluted to it my analysis, but I tend to agree with reviewer 3 on the limited effect VAIL seems to have on LC3b lipidation.*

    As outlined above in response to reviewer 1 and below to reviewer 3, we agree that there is a modest contribution of VAIL to overall LC3 lipidation, which correlates with a modest amount of E expression in SARS-CoV-2 infection. VAIL is clearly important for the viral life cycle, thus whatever the proportion of LC3 lipidation attributable to this pathway it must be biologically significant.

    *Reviewer #2 (Significance (Required)): *

    • While previous publications have shown interaction between SARS-CoV2 and autophagy, the authors of this manuscript demonstrate that V-ATPase-ATG16L1 induced LC3 lipidation (VAIL) is activated during infection and affects viral replication. *

    • This study provides an interesting new aspect to host-SARS_CoV-2 interactions. *

    • The manuscript is of interest for people studying virus-host cell interaction, as well as for researchers in the fields of infectious diseases, specifically SARS-CoV2, and autophagy/VAIL*.

    We thank the reviewer for their assessment of our manuscript.

    R*eviewer #3 (Evidence, reproducibility and clarity (Required)): *

    • The interaction of SARS-CoV-2 with canonical autophagy has been well documented. However, whether SARS-CoV-2 infection induces and benefits from non-canonical autophagy is unclear. In this manuscript, the authors demonstrated that SARS-CoV-2 infection induces V-ATPase-ATG16L1-induced LC3 lipidation (VAIL), a form of non-canonical autophagy in which LC3 is conjugated to single membranes. The SARS-CoV-2 envelope protein, through its ion channel activity, triggers the V-ATPase proton pump and induces VAIL during SARS-CoV-2 infection. Inhibiting VAIL during SARS-CoV-2 infection with SopF, a Salmonella effector, attenuates SARS-CoV-2 egress. *

    • While these findings are interesting and demonstrate that SARS-CoV-2 infection triggers VAIL for its own benefit, the mechanism by which VAIL promotes SARS-CoV-2 replication remains unclear. Moreover, the contribution of VAIL to LC3 lipidation during SARS-CoV-2 infection appears to be minimal, as blocking VAIL through SoPF expression only marginally reduced LC3B lipidation (Fig. 1H). Therefore, the contribution of VAIL to LC3 lipidation during SARS-CoV-2 infection is minimal.*

    We thank the reviewer for their assessment of our manuscript. As we have already alluded to in our response, we agree that only part of the LC3 lipidation observed during infection can be attributed to VAIL. There is a reproducible effect on viral replication which we have demonstrated in multiple ways, therefore the contribution of VAIL is of biological importance.

    *Comments: *

    • The authors show that the ion channel activity of E is essential for VAIL induction during SARS-CoV-2 infection. Since V-ATPase recruits the ATG16L complex to induce VAIL, and to clarify how SARS-CoV-2 infection triggers VAIL, the authors should examine whether SARS-CoV-2 infection or the expression of E induces V-ATPase-ATG16L interaction and whether this interaction is disrupted when SopF is expressed.*

    We agree with the reviewer that this would be an informative experiment. We can carry out this experiment in an E expression system, rather than infection. This is due to the difficulty of getting enough material to carry out this kind of pull-down experiment in infected cells (at the time of writing these experiments still have to be carried out in CL3).

    • Since the authors suggest that expression of SopF attenuates viral exit, one would expect that the number of N-positive cells will increase in SopF-expressing cells compared to the mCherry control cells. However, as shown in Figure 2D, this is not the case. Could the authors discuss why N-positive cells will be reduced in SopF-expressing cells when viral egress is impeded in these cells*?

    This is a reflection of multi-cycle kinetics. N is still very strongly expressed in infected cells, even after virions have egressed. SARS-CoV-2 can infect VAIL-deficient cells and expresses the same levels of N prior to subsequent rounds of infection (at 7 hours after infection for example). Egress in VAIL-deficient, SopF-expressing cells is defective. Therefore, fewer cells will be infected in subsequent rounds of infection in SopF expressing cells, resulting in fewer N-positive cells in the SopF expressing cell population (most obvious after 24 hours).

    Figure 2H. The authors show that knockdown of ATG16L1 reduces the expression of N during SARS-CoV-2 infection compared to the controls. To confirm that knockdown of ATG16L1, which is required for both canonical autophagy and VAIL, reduces N staining via VAIL, the authors should examine the impact of SopF expression on N levels in ATG16L KD cells. This experiment will confirm if the reduction in N staining in ATG16L1 KD cells is due to VAIL.

    As stated in the response to reviewer 1, we can attempt this experiment in an ATG16L1 KO system complemented with K490A ATG16L1, which is deficient in VAIL and not canonical autophagy.

    • Figure 2J. The quality of the Western blot data is poor.*

    In this western the exposure is deliberately turned up to show that minimal ATG13 was left after knock down. We will also show the full blot with less exposure – this will demonstrate high quality.

    Also, N appears as a single band in Figure 2J, but appears as double bands in Figures 2A and H. Could the authors explain this?

    An extra band can be seen in 2J for N. However, as the reviewer points out, the intensity of the lower band is fainter than in 2A or 2H. The biology of SARS-CoV-2 N is interesting and complicated, with different truncated isoforms and phosphorylation patterns observed (see for example Mears et al., 2025 PMID:39836705). We observed changes in abundance of the second band between experiments, but this did not obviously depend on VAIL. We therefore consider this to be beyond the scope of this investigation.

    *Reviewer #3 (Significance (Required)): *

    • This manuscript proposes a role for VAIL in LC3 lipidation during SARS-CoV-2 infection. While the findings are interesting, VAIL only marginally contributes to LC3 lipidation during SARS-CoV-2 infection. Therefore, the significance of VAIL to LC3B lipidation during SARS-CoV-2 infection is unclear.*

    Our experiments show unambiguously that VAIL contributes to viral replication. Therefore even if As alluded to above, we do not think a further investigation of canonical macroautophagy and SARS-CoV-2 would enhance the quality of our manuscript. We will try to make our description of the contribution of macroautophagy clearer in the revised manuscript (without providing a full literature review). We also do not think that exploring the nature of the multiple N bands on western blot is within the scope of this paper.

  2. Review Commons

    Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

    Learn more at Review Commons

    Referee #3

    Evidence, reproducibility and clarity

    The interaction of SARS-CoV-2 with canonical autophagy has been well documented. However, whether SARS-CoV-2 infection induces and benefits from non-canonical autophagy is unclear. In this manuscript, the authors demonstrated that SARS-CoV-2 infection induces V-ATPase-ATG16L1-induced LC3 lipidation (VAIL), a form of non-canonical autophagy in which LC3 is conjugated to single membranes. The SARS-CoV-2 envelope protein, through its ion channel activity, triggers the V-ATPase proton pump and induces VAIL during SARS-CoV-2 infection. Inhibiting VAIL during SARS-CoV-2 infection with SopF, a Salmonella effector, attenuates SARS-CoV-2 egress.

    While these findings are interesting and demonstrate that SARS-CoV-2 infection triggers VAIL for its own benefit, the mechanism by which VAIL promotes SARS-CoV-2 replication remains unclear. Moreover, the contribution of VAIL to LC3 lipidation during SARS-CoV-2 infection appears to be minimal, as blocking VAIL through SoPF expression only marginally reduced LC3B lipidation (Fig. 1H). Therefore, the contribution of VAIL to LC3 lipidation during SARS-CoV-2 infection is minimal.

    Comments:

    The authors show that the ion channel activity of E is essential for VAIL induction during SARS-CoV-2 infection. Since V-ATPase recruits the ATG16L complex to induce VAIL, and to clarify how SARS-CoV-2 infection triggers VAIL, the authors should examine whether SARS-CoV-2 infection or the expression of E induces V-ATPase-ATG16L interaction and whether this interaction is disrupted when SopF is expressed.

    Since the authors suggest that expression of SopF attenuates viral exit, one would expect that the number of N-positive cells will increase in SopF-expressing cells compared to the mCherry control cells. However, as shown in Figure 2D, this is not the case. Could the authors discuss why N-positive cells will be reduced in SopF-expressing cells when viral egress is impeded in these cells?

    Figure 2H. The authors show that knockdown of ATG16L1 reduces the expression of N during SARS-CoV-2 infection compared to the controls. To confirm that knockdown of ATG16L1, which is required for both canonical autophagy and VAIL, reduces N staining via VAIL, the authors should examine the impact of SopF expression on N levels in ATG16L KD cells. This experiment will confirm if the reduction in N staining in ATG16L1 KD cells is due to VAIL.

    Figure 2J. The quality of the Western blot data is poor. Also, N appears as a single band in Figure 2J, but appears as double bands in Figures 2A and H. Could the authors explain this?

    Significance

    This manuscript proposes a role for VAIL in LC3 lipidation during SARS-CoV-2 infection. While the findings are interesting, VAIL only marginally contributes to LC3 lipidation during SARS-CoV-2 infection. Therefore, the significance of VAIL to LC3B lipidation during SARS-CoV-2 infection is unclear.

  3. Review Commons

    Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

    Learn more at Review Commons

    Referee #2

    Evidence, reproducibility and clarity

    Major Comments

    Figure 1D does not very clearly show an overlap between V1D and LC3B. Both proteins seem broadly present across the cell and there is no easily identifiable change in V1D distribution upon infection. As such the overlay may be purely stochastic. The authors should quantify the observed co-localization events across multiple cells and biological replicates and compare them to other protein(s) with a similar cellular distribution pattern.

    Based on Figure 2F the authors suggest that virus entry is unaffected by the inhibition of VAIL in early timepoints. However, according to the figure legend, the timepoint used is 7hpi, while 2D uses 24hpi. Some SARS-CoV-2 papers suggest 7-10 hours is sufficient time to release new virions (Ban-On et al., 2020). As such 7hpi can not necessarily be seen as an early time point. Did the authors test earlier ones? Also, based on this, would it be possible that the effects observed at 24hpi are actually secondary infections, meaning that the virus utilizes pathway components for virion production and a lack thereof reduces infectivity of newly formed virions? In this case it would be interesting to set up an assay that can distinguish between primary and secondary infection to study both individually more closely.

    The authors nicely show in their study an involvement of VAIL in SARS-CoV-2 mediated LC3 lipidation. However, the observed effects are relatively moderate in several experiments, indicating that there may be another contributor to the observed phenotype. It would be nice to highlight this in the discussion and debate potential mechanisms that are causing the observed effects during infection.

    Minor Comments

    The re-localization events shown in Fig 3A should be quantified.

    The co-localization events displayed in Fig 4A should be quantified.

    For Figure 2H-K the authors perform KDs of ATG16L1 and ATG13. While the results for the two specific proteins are certainly convincing, the authors would strengthen their argument by testing additional proteins in the autophagy pathway to support their claim that VAIL but not autophagy affects protein abundance of N (OPTIONAL).

    Referee cross-commenting

    Overall I agree with the comments of my co-reviewers and I think the suggested experiments/comments are sensible. I in part already eluted to it my analysis, but I tend to agree with reviewer 3 on the limited effect VAIL seems to have on LC3b lipidation.

    Significance

    While previous publications have shown interaction between SARS-CoV2 and autophagy, the authors of this manuscript demonstrate that V-ATPase-ATG16L1 induced LC3 lipidation (VAIL) is activated during infection and affects viral replication.

    This study provides an interesting new aspect to host-SARS_CoV-2 interactions.

    The manuscript is of interest for people studying virus-host cell interaction, as well as for researchers in the fields of infectious diseases, specifically SARS-CoV2, and autophagy/VAIL.

  4. Review Commons

    Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

    Learn more at Review Commons

    Referee #1

    Evidence, reproducibility and clarity

    Figueras-Novoa et al present a short report demonstrating the induction of LC3 lipidation on single membranes by SARS-CoV-2 through a noncanonical autophagy pathway referred to as VAIL. The authors utilize elegant genetic tools to show that the induction of LC3 lipidation upon viral infection is mainly due to VAIL rather than canonical autophagy. They demonstrate that the activity of the viral E protein that can cause neutralization of acidic vesicles leads to the activation of non-canonical LC3 lipidation on single membranes. Interestingly, the authors also conclude that the impairment of VAIL leads to a reduction of viral load as a result of a defect in later stages of viral infection, although the underlying mechanism was not further explored.

    Overall, this is an elegant and well controlled study that provides a clear conclusion. I only have some minor comments.

    In some experiments, LC3 lipidation does not appear to be fully disrupted upon VAIL inhibition (e.g. Fig.'s 1H, 3D, 4A). As other labs have shown that SARS-CoV2 blocks autophagic flux, this could be further clarified in this manuscript as both VAIL and autophagy may be co-induced upon viral infection. Can the authors test the induction of LC3 lipidation in cells expressing K490 mutant of ATG16L1 in ATG16L1 KO cells to compare them with ATG16L1-ATG13 double knockouts?

    Minor points:

    The difference between Fig. 1F&G is unclear and why the authors are including both analyses. Similarly figures 4G&H.

    The authors should show boxed colocalisation of all images, including negative controls. For examples, the authors have shown boxed magnifications in only the lowest panel in Figure 2A but not the upper two panels. Figures 4E&F should include boxed examples. This serves to clarify both positive and negative colocalisation events.

    Significance

    Overall an elegant and well controlled study demonstrating the induction of non-canonical LC3 conjugation on single membranes (VAIL) during SARS-CoV2 infection. A further exploration of canonical autophagy (as previously published by others) in addition to VAIL would enhance this study.

    This study is of interest to researchers studying autophagy, viruses, immunology, single membrane LC3 lipidation, and lysosomes as well as potentially clinicians treating SARS-CoV2 infecteted individuals.

    This reviewer is experienced in autophagy research.

  5. Version published to 10.1101/2025.06.12.659257 on bioRxiv