In-silico docking platform with serine protease inhibitor (SERPIN) structures identifies host cysteine protease targets with significance for SARS-CoV-2

  1. Joaquín J Rodriguez Galvan
  2. Maren de Vries
  3. Shiraz Belblidia
  4. Ashley Fisher
  5. Rachel A Prescott
  6. Keaton M Crosse
  7. Walter F. Mangel
  8. Ralf Duerr
  9. Meike Dittmann

Reviewed by

Read the full article

Listed in

Log in to save this article

Abstract

Serine Protease Inhibitors (SERPINs) regulate protease activity in various physiological processes such as inflammation, cancer metastasis, angiogenesis, and neurodegenerative diseases. However, their potential in combating viral infections, where proteases are also crucial, remains underexplored. This is due to our limited understanding of SERPIN expression during viral-induced inflammation and of the SERPINs’ full spectrum of target proteases. Here, we demonstrate widespread expression of human SERPINs in response to respiratory virus infections, both in vitro and in vivo , alongside classical antiviral effectors. Through comprehensive in-silico docking with full-length SERPIN and protease 3D structures, we confirm known inhibitors of specific proteases; more importantly, the results predict novel SERPIN-protease interactions. Experimentally, we validate the direct inhibition of key proteases essential for viral life cycles, including the SERPIN PAI-1’s capability to inhibit select cysteine proteases such as cathepsin L, and the serine protease TMPRSS2. Consequently, PAI-1 suppresses spike maturation and multi-cycle SARS-CoV-2 replication. Our findings challenge conventional notions of SERPIN selectivity, underscore the power of in-silico docking for SERPIN target discovery, and offer potential therapeutic interventions targeting host proteolytic pathways to combat viruses with urgent unmet therapeutic needs.

SIGNIFICANCE

Serine protease inhibitors (SERPINs) play crucial roles in various physiological processes, including viral infections. However, our comprehension of the full array of proteases targeted by the SERPIN family has traditionally been limited, hindering a comprehensive understanding of their regulatory potential. We developed an in-silico docking platform to identify new SERPIN target proteases expressed in the respiratory tract, a critical viral entry portal. The platform confirmed known and predicted new targets for every SERPIN examined, shedding light on previously unrecognized patterns in SERPIN selectivity. Notably, both key proteases for SARS-CoV-2 maturation were among the newly predicted targets, which we validated experimentally. This underscores the platform’s potential in uncovering targets with significance in viral infections, paving the way to define the full potential of the SERPIN family in infectious disease and beyond.

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

    1. General Statements

    *We thank the reviewers for the overwhelmingly positive feedback on our initial submission. *

    Reviewer 1: “Overall this is a very simple, although extensive and excellent, study analyzing a wide range of data form (sic) patients with bronchoalveolar lavage and epithelial cell samples, human epithelial cell cultures after infection with a range of respiratory viruses as well as the development of a 3D in silico analysis of potential protease and serpin interactions.”

    Reviewer 2: “Overall, this paper will be interesting to a specialized audience that is interested in SERPIN function. The SERPIN expression data during viral infection, discovery of CTSL as a target of PAI-1, and evidence that PAI-1 can inhibit SARS-CoV-2 replication, will move that field forward.”

    No new experiments were requested, but some were either suggested or explicitly marked optional. We thus focused the initial 4-week-revision on performing new experiments aimed to enhance our study’s significance and impact by validating the heart of our study: the data from the in-silico docking screen.

    Reviewer 1: “Further analysis of the detected proteases that are reported here to bind to PAI-1 would be of great interest.”

    Reviewer 2: “It is exciting that they predicted CTSL as a target of PAI-1, but it is not obvious that this is a generalizable approach without further hypothesis testing.”

    Thus, we performed additional protease activity assays to validate SERPIN-protease pairs from the in-silico-screen. The results elevate our study above the proof-of-principle state. Beyond their described roles in infectious disease, the two SERPINs that are now tested in more detail (SERPINB2, plasminogen activator inhibitor 2 and SERPINE1, plasminogen activator inhibitor 1) also play critical roles in cancer, neurodegeneration, aging, and cardiovascular disease. (Bouton et al., EMBO Mol Med 2023 Vol. 15 No. 6; Zhang et al., EMBO Mol Med 2023 Vol. 15 No. 9; Bode et al., EMBO Journal 1986 Vol. 5 No. 10; Uhl et al., EMBO Mol Med 2021 Vol. 13 No. 6). Given these multifaceted roles, we anticipate that our discovery of new SERPIN-protease binders and non-binders will advance various areas of human disease driven by SERPIN biology.

    2. Description of the planned revisions

    *We believe that the planned revisions outlined below can be finalized within 1-2 months. *

    Reviewer 1: “Further analysis of the detected proteases that are reported here to bind to PAI-1 would be of great interest.”

    *At the time of the 30-day revision, recombinant SERPINB1 (LEI) and SERPING1 (C1-INH) were still backordered with an estimated shipping date the week of resubmission. Once delivered, we will perform protease activity assays with LEI or C1-INH and uPA, TMPRSS2, Cathepsin L, and Cathepsin B to bring up the number of validated SERPIN:protease interactions from 8 to 16. *

    Reviewer 2, major points:

    1. Further, to strengthen the conclusions of this data the authors should include additional controls. One would be to use trixplanin as they did in previous panels to show that PAI-1 is necessary. Further, if the authors generate mutant PAI-1 that is unable to inhibit TMPRSS2 (see comment 11 below), they could also use this as a control to show the necessity of functional PAI-1.

    *We agree that these optional experiments would increase rigor. We generated plasmids containing mutated PAI-1 that we can use in spike cleavage assays as suggested and can perform this experiment. *

    *We can unfortunately not use triplaxinin on cells, as our preliminary data show that it is quite cytotoxic at the concentrations required to inhibit PAI-1. *

    1. For Figures 4I-J, is it possible to also blot for S1 cleavage? If possible, this optional data would be helpful to understand whether the entire cleavage process is disrupted or only S2 to S2' especially given that visually it appears as if the full length is more depleted in the condition with PAI-1 suggesting that it is cleaving spike better into S1 and S2. Could also suggest that the dynamics of cleavage are shifted rather than impaired?

    *S1 cleavage is shown indirectly in (now) Figure 5f,g – the main product of S1 cleavage is the fragment annotated as S2. Due to high levels of endogenous furin in BHK cells, this cleavage always occurs in this experimental setting. It is true that we have not shown the effects of PAI-1 inhibits on S1 cleavage– we can include that control in the above optional experiment (point 9). We do not expect PAI-1 to have an effect on S1 cleavage, as it is well-established that it does not inhibit furin. *

    Reviewer 2, minor points

    1. As a supplemental figure, can the authors show a complex blot (similar to Figure 4F) for CTSB to show that is does not complex with PAI-1.

    *Purified active CTSB is not commercially available, but we can attempt to perform gel shift analysis on the samples from the in vitro protease assay. Due to the presence of proteinaceous substrate in these samples, we have previously observed lot of background on the gel, but we can attempt it and include it in a revised manuscript if reviewer/editor find it useful. *

    3. Description of the revisions that have already been incorporated in the transferred manuscript

    Reviewer 1:

    Summary:

    The authors do not reference the prior work which has examined cross class serpins, viral and mammalian, - this should be noted as alternative protease targets are known.

    *Thank you – please see our response in point 1 below. *

    The bronchiolar lavage analysis is excellent but cannot differentiate epithelial cell and associated immune cells and their roles in the response.

    We apologize for not making this clear – scRNAseq can indeed differentiate between different cell types using cell-type specific expression markers for each individual cell. This is how we were able to retrieve expression data specific for individual cell types. The reviewer is correct in that an expression analysis cannot show the role of individual cell types in the antiviral response. However, as epithelial cells are the primary cell type infected by SARS-CoV-2 gene expression patterns in these epithelial cells may show us cell-intrinsic effectors that are upregulated in response to viral infection. We now revised language in this paragraph to make this clearer (lines 156-162).

    PAI-1 does not seem to be present in the bronchoalveolar lavage samples.

    We do not know if PAI-1 is present, as we did not analyze protein levels in these samples. The gene expression data suggests that SERPINE1, the gene encoding PAI-1, is expressed at low levels in the epithelial cell subset at baseline, and expressed at slightly higher levels in individuals with severe COVID-19 (Figure 1c). This is consistent with previously published data on SERPINE1 gene expression upon viral infection (Dittmann et al., Cell, 2015).

    Further discussion of prior work with cross class serpins and also the limitations of the in-silico analyses and the lavage specimens should be provided.

    Prior work evidencing other cross class serpin protease targets as well as limitations related to the analyses as discussed in the critiques above should be noted and the abstract and title could better describe and define the studies as performed.

    *Thank you for raising these important points. For cross-class SERPINs, please see our response to point 1. The limitations of in silico analyses are discussed in-depth in a paragraph of the discussion (lines 608-631). We also discuss discrepancies observed between SERPIN expression in lavage specimens and in HAEC – please advise whether this is sufficient or needs bolstering (lines 546-564). We revised both title and abstract to better describe and define the studies as performed. *

    These correlations between changes in serpin and protease expression with viral infections and potential new interactions for serpins with previously non identified proteases is of clear interest. This shows an excellent correlation but as with big data sets this does not provide a true cause and effect - rather providing new potential directions for analysis of these interactions in viral infections in lung epithelium and this is valuable as a basis for ongoing studies.

    *We are in agreement with the lack of cause and effect –to our knowledge, we make no such claim from the gene expression data. We state that we used the expression data to guide the selection of SERPINs for our in-silico screen (lines 317-319). We then validated select data from our in-silico screen in vitro, which provides true cause and effect (Figures 4 and 5). *

    Major points:

    • Cross class serpin interactions are known and have been reported for at least two viral serpins Serp-1 and CrmA - both of which bind cysteine proteases as well as serine proteases as well as the mammalian SCCA serpins. *Thank you for bringing these two examples to our attention – we added them to the discussion (lines 648-652). We now also emphasized throughout the manuscript that the novelty of our findings is in PAI-1 cross-class inhibition, specifically, which has not been previously reported despite PAI-1 being an extremely well-studied SERPIN. *

    *We also would like to mention that in our opinion the scientific advance provided by our in-silico screen is not limited to the identification of new PAI-1 targets, but also provides a birds-eye view on SERPIN selectivity in a specific proteolytic landscape. For example, to our knowledge, it was unknown that SERPINB1 is promiscuous and that SERPINC1 is more selective, which our docking predicted. It was unknown that most TMPRSSs are unlikely SERPIN targets and that those that are SERPIN targets need to be in their active state to bind. The unsupervised clustering in Figure 4b (both on the SERPIN and on the protease side) predicts such unrecognized patterns in SERPIN selectivity. *

    • The protease targets are reported to vary when interacting with glycosaminoglycans such as heparan sulfate - PAI-1 inhibits thrombin in the presence of heparin - thus while a canonical serpin suicide inhibition is considered specific - it can vary. This is noted in the discussion *Yes, we agree *(lines 608-610).

    • What is the potential impact of the noted interactions of PAI-1 with other proteases such as cathepsin - PAI-1 is considered to have predominately extracellular functions, but prior work indicates internalization of PAI-1 when bound to the uPA/uPAR complex with alterations in intra cellular activation *This is correct and PAI-1 internalization is cited and mentioned in discussion *(lines 620-624). We now also added data on SARS-CoV-2 variant Omicron BA.1, which predominantly uses CTSL for maturation, and we show is also inhibited by PAI-1 (new Figure 5).

    • This is supported by basic in vivo and in vitro serpin and protease interactions that are demonstrated confirming in silico analyses, eg. gel shift analyses or even Mass spectrometry analysis particularly for PAI-1 Yes, this is the data shown in Figure 4. We now also added protease activity assays for other SERPIN-protease pairs, thereby elevating our study above the proof-of-principle state. *This was also a suggestion raised by reviewer 2. *
    • Per the authors "To date, three SERPINs have been studied in the context of innate antiviral defense: PAI- 1 (encoded by SERPINE1) against influenza viruses encoding hemagglutinin H1 and SARS-CoV-2, by impeding the proteolytic maturation of H1 or spike, respectively19,20; alpha-1-antitrypsin (encoded by SERPINA1) and antithrombin (encoded by SERPINC1) against SARS-CoV-2, likely through the inhibition of TMPRSS2, by reducing maturation of spike, although direct inhibition of TMPRSS2 by either SERPIN was not shown". This is partially complete however other serpins such as C1Inh and one virus derived serpin that have been analyzed for efficacy in treating SARS *Thank you for mentioning this, we added the information to the introduction **(lines 106-111). *

    • While TMPRSS2 is indeed a serine protease - Beneficial effects of some serpins may be due to modulation of the immune response as opposed to selective anti-viral responses. The immune / cytokine storm and coagulopathies (with clotting and even hemorrhage) seen in the excess inflammatory response that causes respiratory vascular leak and severe viral sepsis. PAI-1 targets tPA and uPA - uPA has marked proinflammatory actions when bound to the uPA receptor (uPAR) and can activate growth factors and MMPs which can enhance immune cell invasion - PAI-1 binds to the uPA / uPAR complex which can thus also alter inflammatory cell responses and cell activation when internalized. *Thank you for bringing up this point. The role of SERPINs in inflammation and anti-viral immune responses is indeed well-established. While our study focuses on cell-intrinsic antiviral roles of SERPINs by shutting down pro-viral proteases, which is much less established, we now added this to the results section for clarification *(line 153-156).

    • The RCL does in general incorporate P4 to P4' but can vary from this specific P4 to P4' sequence *Yes, we agree. *

    • How accurately does in silico protease serpin analysis predict real interactions? - this should be discussed as HADDOCK may have some limitations - This is outside my field of expertise *We added an in-depth paragraph on how HADDOCK operates to the results section to help readers not familiar with the technique (lines 248-290). We discuss the limitations of HADDOCK in depth in the discussion section **(lines 608-631)– please advise whether this needs additional information. *

    *We argue that, with the limitations stated in the discussion, our in-silico method predicts interactions well, as shown by the correct prediction of known binders and non-binders, as well as of new binders (PAI-1 to *active* TMPRSS2 and CTSL) and a new non-binder (CTSB). *

    *As with any screening method, results require validation via another method, which we performed for select SERPINs and proteases. In fact, the revised manuscript now features in vitro validation of 8 SERPIN-protease pairs (Figure 4a, b), with 8 additional planned (see “planned revisions” section). *

    • The data from a published study examining bronchoalveolar lavage fluid single cell transcriptional analysis from patients with and without COVID - mild and severe - and with comparison to patients without COVID does demonstrate altered protease and serpin activity - but does not indicate specific interactions *We agree with this statement partially. We disagree in that the data does not demonstrate altered protease and SERPIN activity; it instead demonstrates changes in gene expression levels. We agree in that this does indeed not indicate specific interactions. *

    • What is the significance for changes in gene expression in epithelial cells versus macrophage T and B cells looks - This looks like a small change like a small change in the mean values Figure 1b *We performed additional statistical analyses on the Figure 1 data – please refer to Reviewer 2 point 1. *

    • Of interest - is the brocholaveolar lavage fluid likely to contain both epithelial cells as well as immune response macrophage, T cells and NK cells etc - one assumes single cells were identified and isolated- Is this defined? *Apologies if this was unclear. Yes, the BALF contains all of these cell types. We now added some sentences to the results section explaining scRNAseq and analyses in more detail **(lines 147-162). *
    • The known previously reported target proteases for PAI-1 should be noted *Agreed; it is noted in the results section where we first speak about PAI-1 target specificity *(line 379-382).

    SERPINE1 is not noted in figure 1 - this is PAI-1 - but is seen in the HAEC infection model data

    SERPINE1 is indeed not significantly upregulated in Figure 1, but is significantly upregulated in HAEC upon infection with Reovirus and parainfluenzavirus 3, and upon IFN stimulation (new Supplemental Tables S1 and S2). The possible reasons for discrepancies between the BALF and HAEC data are discussed in lines 546-564.

    • “To overcome this limitation, we developed a computational method to predict 3D interactions between SERPINs and proteases, simulating the binding process depicted in Supplemental Figure 1a. Specifically, we employed High Ambiguity Driven protein- protein Docking (HADDOCK), a tool that predicts complex structures, integrating experimental and computational data35,36." This analysis looks to be extensive however this is a correlation - not a true analysis of cause and effect. We agree on the first point – to our knowledge, our study provides the most extensive SERPIN target discovery process (testing 480 SERPIN-protease interactions). We disagree on the point that our results provide a mere correlation. If you will, we performed a computer-modeled interaction experiment that yields predicted binding energies between each SERPIN with each tested protease. We added a paragraph on how HADDOCK operates to the results section to help readers unfamiliar with the technique. As with any screening method, results need to be validated with another method, which we did for select SERPINs and proteases (Figure 4a, b). This does however have the potential to identify significant interactions *We certainly agree on this point. * In future it might be of interest to assess PAI-1 given to infected cultures to assess viral replication and titers or perhaps examine a knock out cell model? *We did exactly the former in Figure 4 (now 5). *
    • As PAI-1 was identified as having new cathepsin protease binding in addition to TMPRSS2 - the authors did demonstrate inhibition of the new targets on fluorometric analysis and also demonstrated interaction by gel shift - This is excellent *Thank you. *
    • The title and the abstract could be better written and more clearly indicate the extent of the analyses performed and the discovery of alternate protease targets for PAI-1 We modified both title and abstract.
    • Was the SARS CoV2 lung epithelial cell culture analysis performed in BSL3? *Yes. All SARS-CoV-2 infection experiments were performed in a BSL3 environment. We added this information throughout the Methods section, and also generated a new Methods section on Biohazards *(lines 779-797).

    __Minor critiques __

    1. Results section heading "SERPINs are differentially expressed individuals with COVID-19 and in response to respiratory virus infection in a model of the human airway epithelium." The word in needs to be inserted between expressed and individuals *Thank you for catching this – we fixed the sentence (lines 128-129). *

    Reviewer 2:

    Major points:

    1. The rigor of the results presented in Figure 1 are unclear. For the COVID-19 analyses (Figure 1), only one dataset is used, and no statistical analyses are performed to determine to what degree any of the changes they observe are significant relative to variation in the dataset. This makes it difficult to determine how much can be extrapolated from these data.

    We agree that performing statistics on the BALF dataset would be ideal. However, the BALF contains only two non-infected individuals (intubated gun-shot victims), limiting our possibilities for statistical analysis.

    *For Figure 1b, we overcame this limitation by adding statistical analysis of upregulated expression values between cell types (i.e. by analyzing differences of upregulation of given SERPIN in epithelial cells compared to macrophages; Supplemental Table S1). We also performed statistical analysis on upregulation for individual SERPINs compared to housekeeping gene B2M (Supplemental Table S1). This revealed that SERPINs statistically significantly upregulated in severe COVID-19 in most cell types, including epithelial cells, in which SERPIN function has not been broadly studied. Upregulation was not statistically significant in mild COVID-19 samples, likely due to the n=3 (as compared to n=6 in the severe COVID-19 group). *

    *As for analysis of Figure 1c, we could theoretically perform analysis of differential levels between mild and severe COVID-19, but this is not the question we are trying to answer. The question is whether epithelial cells express SERPINs and proteases, and whether there is an upregulation of either in infected individuals. We now state the limitation of lacking statistical power in the figure legend and the text (lines 176-177). *

    1. Similarly, the qPCR data presented in Figure 2 are presented with no statistical analyses. Results should not only be presented with fold change but also p-values that are adjusted for multiple testing.

    *We now present p-values in Supplemental Table S2. Of note, data obtained with the experimental system of polarized airway epithelial cultures, differentiated over several weeks, tends to be noisier than that obtained with cell lines. Despite this, a number of SERPINs reach statistical significance. *

    1. How is the dotted line drawn in Figure 3C and D? It would appear there is very little in terms of HADDOCK score to distinguish a predicted "binder" from "non-binder". Also, they later show that CTSB is non inhibited, and yet in Figure 3C it is below the dotted line. Can the authors more clearly delineate how one might use their dataset shown in Figure 3B to accurately predict targets of SERPINs?

    *This is a valid point. We added a more in-depth description to the results section on how we define “binders” and “non-binders” **(lines 324-331 and Figure 3 legend). We added raw data graph with the thresholds in Supplemental Figure 3d. We further added and defined a threshold line to the PAI-1:CTSs graph (Figure 3c). It is now evident that CTSL, A, F, K score as high-confidence “binders”, while CTSB and others do not. We also added the normalization process and the visual assessment of top-scoring complexes to the in silico docking screen schematic in Figure 3a and the respective figure legend to guide readers. *

    1. Based on this, it would be preferable for the authors to tone down their claims about the broad applicability of this approach to predict SERPIN-protease interactions. It is true that they have used it to accurately predict PAI-1-CTSL interactions, but to make such a broad claim about the generalizable nature of this approach would require testing several more SERPIN-protease pairs (both binders and non-binders) to clearly define the scores and parameters that can used to robustly predict interactions.

    We thank the reviewer for this criticism. We now address this in the text as outlined in our response to point 3 above. As with any screening method, the results require to be validated via an alternative approach, which we did in the initial submission for TMPRSS2 and CTSL as binders and CTSB as a non-binder. The revised manuscript now features additional in vitro validation of binders and non-binders for a total of 8 SERPIN-protease combinations (Figure 4a, b), which were all correctly predicted by our in-silico method. Two more SERPINs will be added in the final revision (see “planned revisions” section). Our study provides ample data for future studies validating additional predicted pairs and characterizing their biological function, in infectious disease and beyond.

    1. In Figure 3D, the authors mutate all eight modeled RCL residues to alanine to create a LOF mutant that has a higher HADDOCK score. Single residue mutations would be more convincing for their model, and would be more informative in terms of their predicted models of interactions.

    *We now performed the docking with the single mutant, please see new Figure 3c. *

    1. Further, in Figure 4G lanes 2-4, the PAI-1 band at ~38kDa is not present. Can the authors explain this?

    *This is likely because CTSL digests PAI-1 working at its optimum pH (aka “the protease wins”). We removed the panel from the manuscript. *

    1. In Figure 4I, the authors claim that the addition of PAI-1 is inhibiting cleavage of the SARS-CoV-2 spike protein (S2) based on densitometry quantifications. However, it is unclear how the authors are normalizing their data, nor whether the experiments (and therefore quantification) are from a single experiment or multiple replicates. Could the authors explain the quantification further and provide replicate information (including statistical support) if those experiments were performed?

    Thank you for pointing this out. An explanation has now been added to the Figure 5 legend.

    __Minor comments: __

    1. The authors speculate about SERPINA1 regulation during viral infection, suggesting an active process of "viral evasion". However, it would appear that even upon interferon treatment in Figure 2C, SERPINA1 expression is decreased. Based on that, the authors should soften their claims about the cause of downregulation of SERPINA1.

    *Thank you for pointing this out – we softened the language on this point (lines 225-228). *

    1. In Figure 2C, do the authors have an explanation or hypothesis for why SERPINE1 is less upregulated at 72hrs when compared to 24hr infection of SARS-CoV-2?

    *We can only speculate on this point. It is possible that one or several of the SARS-CoV-2 accessory proteins modulate SERPINE1 expression in a time-dependent manner. *

    1. Can the authors demonstrate how the docking structure of the TMPRSS2 zymogen differs from the active version (especially zooming in on the interface of PAI-1 and the protease)? This could be supplemental data but can the authors show a panel like that in Figure 3F to show how the interface between PAI-1 and TMPRSS2 zymogen looks. Does the inactive TMPRSS2 not interface well with the RCL? Or what is leading to the decreased HADDOCK score?

    *We added an extensive paragraph on how HADDOCK operates to the results section to introduce how the HADDOCK score is calculated **(lines 248-290). We also added a visual of the top-scoring docking complex of PAI-1 and the TMPRSS2 zymogen (Figure 3d) to illustrate the differences in binding. *

    1. In methods, uPA fluorometric protease assay information is missing. Please add this information.

    *Thank you for catching this – we added the information *(line 890).

    1. It is a bit confusing that Figure 4K is the quantification of assays shown in Figure 4A-C, rather than quantification of any of the intervening figure panels. It might be clearer to move this quantification next to 4A-C so that it is clearer.

    *Thank you for the suggestion – Figure 4 has been restructured. *

    1. In Figure 4H, the authors show that addition of recombinant PAI-1 decreases the number of SARS-CoV-2 nucleoprotein positive cells. Have the authors examined whether this decreases the viral titers as well?

    *Yes, this is now part of the (new) Figure 5. *

    1. In the text, the authors suggest that PAI-1 inhibition of CTSL is surprising/novel. The authors should reconsider phrasing this since there are several other SERPINs that have been shown to inhibit other cathepsins, making this appear less surprising than the authors are suggesting.

    *Thank you for pointing this out. We have now clarified throughout the manuscript that while other SERPINs indeed are known to inhibit cathepsins, this had not been previously shown for the extremely well-studied SERPIN PAI-1 with over 15,000 pubmed entries. We also added the implications of this PAI-1-specific finding to the discussion section. *

    __Significance: __

    The claim of novelty about TMPRSS2 is confusing. In their previous paper (reference 19) they show that PAI-1 inhibits TMPRSS2 activity. These data are clearly shown in Figure 4C & 4D of that paper and are summarized in their sentence in the discussion: "Here, we find three new PAI-1 protease targets: human tryptase (tryptase Clara; club cell secretory protein), HAT, and TMPRSS2 ...". In this current paper, although they characterize the PAI-1-TMPRSS2 interaction in more detail than in their previous paper, they have truly only discovered one new target for PAI-1, which is CTSL.

    *Thank you for pointing this out – we softened language on the novelty of TMPRSS2 as a PAI-1 target **throughout the manuscript. We further clarify that the novelty is that TMPRSS2 has to be in its active form to be inhibited by PAI-1, which was previously unknown (lines 392, 432). The revised manuscript now also provides validation of total 8 predicted binders and non-binders for 2 (Figure 4 b,c), with 8 more pending (see “planned revisions” section). As those two (future four) SERPINs have various roles in cancer, cardiovascular disease, neurodegeneration, and immunity, our findings have impact beyond their antiviral potential, thereby increasing the overall significance of the manuscript. *

    4. Description of analyses that authors prefer not to carry out

    Reviewer 1 major point:

    The more common names for the SERPINS as detected in COVID alveolar lavage samples would be helpful in figure 1 - and specifically labelling PAI-1 as this is a focus for this study - together with the known SERPIN nomenclature or under abbreviations - For example SERPINB2 is PAI-2 and SERPING1 is C1INH and SERPINA1 is alpha 1 antitrypsin *Thank you for this suggestion. We tried to keep the SERPIN nomenclature consistent throughout the manuscript, in that the SERPIN genes are referred to by their gene name (i.e., SERPINE1), while the proteins are referred to by their protein name (i.e., PAI-1). Editor and/or Reviewer 1, please advise whether this is acceptable or should be changed. We also added the protein corresponding names in the figure legend. *

    Why does supplemental figure 2 show SERPINB1 and not PAI-1. *We performed this computer-modeled experiment (docking SERPINs to known binders and known non-binders) for each SERPIN tested in the study. This was needed to obtain thresholds to define likely binders and likely non-binders. We chose to show SERPINB1 in this supplemental figure because it is well-described with regards to binders and non-binders (the latter, as “negative result”, is not always published for a given SERPIN). We also did not want to narrow the study immediately to PAI-1, as we believe our screen is a generalizable method and our findings are valid beyond PAI-1. We can easily show any other SERPIN here - editor and/or Reviewer 1, please advise. *

    Reviewer 2 major point:

    1. Figures 4F and 4G are rather confusing. First, in Figure 4F, amount of PAI-1 in lane 1 is not the same as in the lanes with CTSL. The biggest concern with this is that there is a second, higher MW band that is present in lane 1 (also in Figure 4G lane 1) that runs near the band in lanes 2&3 that is marked as the PAI-1-CTSL complex. Although it does appear that the band in lane 1 and lanes 2&3 are slightly different sizes, it is hard to say that conclusively when the amounts of PAI-1 are different. Can the authors repeat this assay to load consistent amounts PAI-1 across all conditions and even potentially separate the top bands to more convincingly show that the band in lanes 2&3 is not in the PAI-1 alone control?

    *The upper band is an impurity that disappears upon addition of a protease to the reaction. We confirmed that this band is neither PAI-1 nor CTSL via western blot with PAI-1- or CTSL-specific antibodies. Should reviewer 2 and/or the editor feel that we should repeat the experiment with more loading in the first lane, we can certainly do so. Please advise. *

    1. The authors show that exogenous PAI-1 can inhibit SARS-CoV-2 in a multicycle infection in Figure 4H. However, this could be acting at multiple points during the viral infection cycle. A clearer virology experiment to support their model would be to perform single-cycle infections to show that the virus fails to productively infect the cell. For instance, have the authors attempted a high MOI, single-cycle infection to see whether they can detect uncleaved spike protein to show inhibition of cleavage? Or show that no early products of viral infection are produced? While this type of experiment is optional in that it is not required to support the claim that PAI-1 inhibits multicycle SARS-CoV-2 infection, it would support the conclusion that PAI-1 is inhibiting viral entry.

    *We agree with the reviewer. We did expand on the virology by using now two strains of SARS-CoV-2 with different proteolytic needs, ancestral WA-1 and Omicron BA.1. We also performed titer analysis (all in Figure 5). *

    *However, the other suggested experiments would represent a substantial amount of work in a BSL3 environment. We thus would prefer not do these experiments (as the reviewer states, it is optional), and instead tone down the manuscript to make clear we make no claims on viral entry. *

    Reviewer 2 minor point:

    1. One (optional) way to extend these data and support their molecular model would be to mutate residues in PAI-1 that they predict are important for protease inhibition. As their source of PAI-1 currently is commercial, this would require purification of WT and variant PAI-1, which is clearly an undertaking. However, these data would strongly support their modeling and the importance of these residues in engaging with the proteases and springing the mousetrap for their in-vitro/in-vivo experiments (as suggested by data shown in Figure 3F and explained in text). Further, the authors can use these mutants to do some of the functional experiments in Figure 4 as a negative control, and potentially even separate the role of PAI-1 in inhibition of CTSL and TMPRSS2 in terms of SARS-CoV-2 inhibition.

    *We agree that these (optional) experiments would be beautiful and are indeed part of future studies on the subject. We feel that they exceed the scope of this current manuscript. *

  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 #2

    Evidence, reproducibility and clarity

    Summary:

    Rodriguez Galvan et al. use a combined computational and functional approach to identify a novel target for the protease inhibitor, PAI-1 (SERPINE1), and show that exogenous PAI-1 can inhibit SARS-CoV-2 replication. They first use a COVID-19 dataset to identify SERPINs that are differentially expression in individuals with mild and severe COVID-19. They further use experimental infections of a model of human airway epithelium to identify SERPINs that are upregulated in response to several viruses, as well as treatment with interferon. Using this panel of SERPINs and a panel of host proteases, they use computational docking to predict SERPINs that may inhibit human proteases that may be relevant for viral infection. Using these predictions, they show that PAI-1 inhibits TMPRSS2 (previously shown) and CTSL (newly shown in this study), two proteases with relevance for SARS-CoV-2 infection. They finally show that extracellular addition of PAI-1 inhibits multicycle replication of SARS-CoV-2.

    Major comments:

    1. The rigor of the results presented in Figure 1 are unclear. For the COVID-19 analyses (Figure 1), only one dataset is used, and no statistical analyses are performed to determine to what degree any of the changes they observe are significant relative to variation in the dataset. This makes it difficult to determine how much can be extrapolated from these data.
    2. Similarly, the qPCR data presented in Figure 2 are presented with no statistical analyses. Results should not only be presented with fold change but also p-values that are adjusted for multiple testing.
    3. How is the dotted line drawn in Figure 3C and D? It would appear there is very little in terms of HADDOCK score to distinguish a predicted "binder" from "non-binder". Also, they later show that CTSB is non inhibited, and yet in Figure 3C it is below the dotted line. Can the authors more clearly delineate how one might use their dataset shown in Figure 3B to accurately predict targets of SERPINs?
    4. Based on this, it would be preferable for the authors to tone down their claims about the broad applicability of this approach to predict SERPIN-protease interactions. It is true that they have used it to accurately predict PAI-1-CTSL interactions, but to make such a broad claim about the generalizable nature of this approach would require testing several more SERPIN-protease pairs (both binders and non-binders) to clearly define the scores and parameters that can used to robustly predict interactions.
    5. In Figure 3D, the authors mutate all eight modeled RCL residues to alanine to create a LOF mutant that has a higher HADDOCK score. Single residue mutations would be more convincing for their model, and would be more informative in terms of their predicted models of interactions.
    6. Figures 4F and 4G are rather confusing. First, in Figure 4F, amount of PAI-1 in lane 1 is not the same as in the lanes with CTSL. The biggest concern with this is that there is a second, higher MW band that is present in lane 1 (also in Figure 4G lane 1) that runs near the band in lanes 2&3 that is marked as the PAI-1-CTSL complex. Although it does appear that the band in lane 1 and lanes 2&3 are slightly different sizes, it is hard to say that conclusively when the amounts of PAI-1 are different. Can the authors repeat this assay to load consistent amounts PAI-1 across all conditions and even potentially separate the top bands to more convincingly show that the band in lanes 2&3 is not in the PAI-1 alone control?
    7. Further, in Figure 4G lanes 2-4, the PAI-1 band at ~38kDa is not present. Can the authors explain this?
    8. The authors show that exogenous PAI-1 can inhibit SARS-CoV-2 in a multicycle infection in Figure 4H. However, this could be acting at multiple points during the viral infection cycle. A clearer virology experiment to support their model would be to perform single-cycle infections to show that the virus fails to productively infect the cell. For instance, have the authors attempted a high MOI, single-cycle infection to see whether they can detect uncleaved spike protein to show inhibition of cleavage? Or show that no early products of viral infection are produced? While this type of experiment is optional in that it is not required to support the claim that PAI-1 inhibits multicycle SARS-CoV-2 infection, it would support the conclusion that PAI-1 is inhibiting viral entry.
    9. In Figure 4I, the authors claim that the addition of PAI-1 is inhibiting cleavage of the SARS-CoV-2 spike protein (S2) based on densitometry quantifications. However, it is unclear how the authors are normalizing their data, nor whether the experiments (and therefore quantification) are from a single experiment or multiple replicates. Could the authors explain the quantification further and provide replicate information (including statistical support) if those experiments were performed? Further, to strengthen the conclusions of this data the authors should include additional controls. One would be to use trixplanin as they did in previous panels to show that PAI-1 is necessary. Further, if the authors generate mutant PAI-1 that is unable to inhibit TMPRSS2 (see comment 11 below), they could also use this as a control to show the necessity of functional PAI-1.
    10. For Figures 4I-J, is it possible to also blot for S1 cleavage? If possible, this optional data would be helpful to understand whether the entire cleavage process is disrupted or only S2 to S2' especially given that visually it appears as if the full length is more depleted in the condition with PAI-1 suggesting that it is cleaving spike better into S1 and S2. Could also suggest that the dynamics of cleavage are shifted rather than impaired?
    11. One (optional) way to extend these data and support their molecular model would be to mutate residues in PAI-1 that they predict are important for protease inhibition. As their source of PAI-1 currently is commercial, this would require purification of WT and variant PAI-1, which is clearly an undertaking. However, these data would strongly support their modeling and the importance of these residues in engaging with the proteases and springing the mousetrap for their in-vitro/in-vivo experiments (as suggested by data shown in Figure 3F and explained in text). Further, the authors can use these mutants to do some of the functional experiments in Figure 4 as a negative control, and potentially even separate the role of PAI-1 in inhibition of CTSL and TMPRSS2 in terms of SARS-CoV-2 inhibition.

    Minor comments:

    1. The authors speculate about SERPINA1 regulation during viral infection, suggesting an active process of "viral evasion". However, it would appear that even upon interferon treatment in Figure 2C, SERPINA1 expression is decreased. Based on that, the authors should soften their claims about the cause of downregulation of SERPINA1.
    2. In Figure 2C, do the authors have an explanation or hypothesis for why SERPINE1 is less upregulated at 72hrs when compared to 24hr infection of SARS-CoV-2?
    3. Can the authors demonstrate how the docking structure of the TMPRSS2 zymogen differs from the active version (especially zooming in on the interface of PAI-1 and the protease)? This could be supplemental data but can the authors show a panel like that in Figure 3F to show how the interface between PAI-1 and TMPRSS2 zymogen looks. Does the inactive TMPRSS2 not interface well with the RCL? Or what is leading to the decreased HADDOCK score?
    4. In methods, uPA fluorometric protease assay information is missing. Please add this information.
    5. It is a bit confusing that Figure 4K is the quantification of assays shown in Figure 4A-C, rather than quantification of any of the intervening figure panels. It might be clearer to move this quantification next to 4A-C so that it is clearer.
    6. In Figure 4H, the authors show that addition of recombinant PAI-1 decreases the number of SARS-CoV-2 nucleoprotein positive cells. Have the authors examined whether this decreases the viral titers as well?
    7. As a supplemental figure, can the authors show a complex blot (similar to Figure 4F) for CTSB to show that is does not complex with PAI-1.
    8. In the text, the authors suggest that PAI-1 inhibition of CTSL is surprising/novel. The authors should reconsider phrasing this since there are several other SERPINs that have been shown to inhibit other cathepsins, making this appear less surprising than the authors are suggesting.

    Significance

    Assessment and impact:

    This paper brings attention to the potential role of SERPINs in viral pathogenesis. The datasets shown in Figure 1 and 2, with the statistical caveats described above, are interesting demonstrations of the regulation of SERPINS during viral infection. In particular, the comparison of different viruses, and viruses compared to interferon alone, in Figure 2B is intriguing. These data are the strongest points of the paper.

    The impact of the computational modeling is difficult to assess. While they have used this dataset to predict one novel interaction (CTSL) with PAI-1, the generalizable nature of this approach to broadly predict SERPIN-protease interactions is unclear since they have not tested or validated any other SERPIN-protease pairs. One major concern is the one raised in Major comments 3&4 above, which is that the score difference between a "non-binder" (CTSB) and a "binder" (uPa) is very small. It is exciting that they predicted CTSL as a target of PAI-1, but it is not obvious that this is a generalizable approach without further hypothesis testing.

    The claim of novelty about TMPRSS2 is confusing. In their previous paper (reference 19) they show that PAI-1 inhibits TMPRSS2 activity. These data are clearly shown in Figure 4C & 4D of that paper and are summarized in their sentence in the discussion: "Here, we find three new PAI-1 protease targets: human tryptase (tryptase Clara; club cell secretory protein), HAT, and TMPRSS2 ...". In this current paper, although they characterize the PAI-1-TMPRSS2 interaction in more detail than in their previous paper, they have truly only discovered one new target for PAI-1, which is CTSL.

    Finally, the data on SARS-CoV-2 are intriguing and contribute to an emerging field on antiviral SERPINs. This reveals an additional virus that is inhibited by PAI-1, to add to their previous discoveries (reference 19) of influenza virus and Sendai virus inhibition by PAI-1. Future virology experiments, and experiments with mutants that ideally separate the ability of PAI-1 to inhibit TMPRSS2 versus CTSL, will further reveal the step of viral replication that is inhibited, and reveal the contribution of inhibition of TMPRSS2, CTSL, or any other PAI-1 targets, on SARS-CoV-2 replication.

    Audience: Overall, this paper will be interesting to a specialized audience that is interested in SERPIN function. The SERPIN expression data during viral infection, discovery of CTSL as a target of PAI-1, and evidence that PAI-1 can inhibit SARS-CoV-2 replication, will move that field forward.

    Field of expertise: Biochemistry, host-virus interactions

  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 #1

    Evidence, reproducibility and clarity

    Title - In-silico docking platform with serine protease inhibitor (SERPIN) structures identifies host cysteine protease targets with significance for SARS-CoV-2

    Authors - Joaquín J Rodriguez Galvan, Maren de Vries, Shiraz Belblidia, Ashley Fisher, Rachel A Prescott, Keaton M Crosse, Walter F. Mangel, Ralf Duerr, Meike Dittmann

    Summary

    The finding that PAI-1 has cross class serpin functions is of definite interest given the roles of PAI-1 in regulation of physiological processes as well as in driving pathology. PAI-1 is generally considered to be a key regulator of thrombolysis and thus an effect on other pathways and even intracellular pathways is of interest. Examining airway epithelial proteases and serpins is of definite interest in respiratory viral infections. Broadening the targets for serpins is also of very definite interest. This study ranges from an overview of prior published work on bronchoalveolar lavage samples and serpin expression, a tissue culture analysis of lung epithelial cells and expression of proteases and serpins is assessed. In addition selective changes in serpin expression and protease targets are assessed by in silico analysis as well as proof of concept via Western blot and fluorometric analysis. This is an extensive study and of definite interest.

    There are some limitations as with any study, albeit the study overall is excellent. The authors do not reference the prior work which has examined cross class serpins, viral and mammalian, - this should be noted as alternative protease targets are known. They do mention the modulation of protease targets by glycosaminoglycans in the discussion. Further, serpins are inhibitors, thus while the RCL provides a target for a protease, but the response may not be fully selective in vivo as the protease has to be present and active to complete the serpin protease interaction. The bronchiolar lavage analysis is excellent but cannot differentiate epithelial cell and associated immune cells and their roles in the response. PAI-1 does not seem to be present in the bronchoalveolar lavage samoles. Further discussion of prior work with cross class serpins and also the limitations of the in silico analyses and the lavage specimens should be provided. Further analysis of the detected proteases that are reported here to bind to PAI-1 would be of great interest. The data from the bronchoalveolar lavage is published.

    Overall this is a very simple, although extensive and excellent, study analyzing a wide range of data form patients with bronchoalveolar lavage and epithelial cell samples, human epithelial cell cultures after infection with a range of respiratory viruses as well as the development of a 3D in silico analysis of potential protease and serpin interactions. These correlations between changes in serpin and protease expression with viral infections and potential new interactions for serpins with previously non identified proteases is of clear interest. This shows an excellent correlation but as with big data sets this does not provide a true cause and effect - rather providing new potential directions for analysis of these interactions in viral infections in lung epithelium and this is valuable as a basis for ongoing studies. Prior work evidencing other cross class serpin protease targets as well as limitations related to the analyses as discussed in the critiques above should be noted and the abstract and title could better describe and define the studies as performed.

    Critique

    Major

    1. Cross class serpin interactions are known and have been reported for at least two viral serpins Serp-1 and CrmA - both of which bind cysteine proteases as well as serine proteases as well as the mammalian SCCA serpins
    2. The protease targets are reported to vary when interacting with glycosaminoglycans such as heparan sulfate - PAI-1 inhibits thrombin in the presence of heparin - thus while a canonical serpin suicide inhibition is considered specific - it can vary. This is noted in the discussion
    3. What is the potential impact of the noted interactions of PAI-1 with other proteases such as cathepsin - PAI-1 is considered to have predominately extracellular functions, but prior work indicates internalization of PAI-1 when bound to the uPA/uPAR complex with alterations in intra cellular activation
    4. This is supported by basic in vivo and in vitro serpin and protease interactions that are demonstrated confirming in silico analyses, eg. gel shift analyses or even Mass spectrometry analysis particularly for PAI-1
    5. Per the authors "To date, three SERPINs have been studied in the context of innate antiviral defense: PAI- 1 (encoded by SERPINE1) against influenza viruses encoding hemagglutinin H1 and SARS-CoV-2, by impeding the proteolytic maturation of H1 or spike, respectively19,20; alpha-1-antitrypsin (encoded by SERPINA1) and antithrombin (encoded by SERPINC1) against SARS-CoV-2, likely through the inhibition of TMPRSS2, by reducing maturation of spike, although direct inhibition of TMPRSS2 by either SERPIN was not shown". This is partially complete however other serpins such as C1Inh and one virus derived serpin that have been analyzed for efficacy in treating SARS
    6. While TMPRSS2 is indeed a serine protease - Beneficial effects of some serpins may be due to modulation of the immune response as opposed to selective anti-viral responses. The immune / cytokine storm and coagulopathies (with clotting and even hemorrhage) seen in the excess inflammatory response that causes respiratory vascular leak and severe viral sepsis. PAI-1 targets tPA and uPA - uPA has marked proinflammatory actions when bound to the uPA receptor (uPAR) and can activate growth factors and MMPs which can enhance immune cell invasion - PAI-1 binds to the uPA / uPAR complex which can thus also alter inflammatory cell responses and cell activation when internalized.
    7. The RCL does in general incorporate P4 to P4' but can vary from this specific P4 to P4' sequence
    8. How accurately does in silico protease serpin analysis predict real interactions? - this should be discussed as HADDOCK may have some limitations - This is outside my field of expertise
    9. The data from a published study examining bronchoalveolar lavage fluid single cell transcriptional analysis from patients with and without COVID - mild and severe - and with comparison to patients without COVID does demonstrate altered protease and serpin activity - but does not indicate specific interactions
    10. What is the significance for changes in gene expression in epithelial cells versus macrophage T and B cells looks - This looks like a small change like a small change in the mean values Figure 1b
    11. The more common names for the SERPINS as detected in COVID alveolar lavage samples would be helpful in figure 1 - and specifically labelling PAI-1 as this is a focus for this study - together with the known SERPIN nomenclature or under abbreviations - For example SERPINB2 is PAI-2 and SERPING1 is C1INH and SERPINA1 is alpha 1 antitrypsin
    12. Of interest - is the brocholaveolar lavage fluid likely to contain both epithelial cells as well as immune response macrophage, T cells and NK cells etc - one assumes single cells were identified and isolated- Is this defined?
    13. The known previously reported target proteases for PAI-1 should be noted
    14. SERPINE1 is not noted in figure 1 - this is PAI-1 - but is seen in the HAEC infection model data
    15. "To overcome this limitation, we developed a computational method to predict 3D interactions between SERPINs and proteases, simulating the binding process depicted in Supplemental Figure 1a. Specifically, we employed High Ambiguity Driven protein- protein Docking (HADDOCK), a tool that predicts complex structures, integrating experimental and computational data35,36." This analysis looks to be extensive however this is a correlation - not a true analysis of cause and effect This does however have the potential to identify significant interactions - In future it might be of interest to assess PAI-1 given to infected cultures to assess viral replication and titers or perhaps examine a knock out cell model?
    16. Why does supplemental figure 2 show SERPINB1 and not PAI-1
    17. As PAI-1 was identified as having new cathepsin protease binding in addition to TMPRSS2 - the authors did demonstrate inhibition of the new targets on fluorometric analysis and also demonstrated interaction by gel shift - This is excellent
    18. The title and the abstract could be better written and more clearly indicate the extent of the analyses performed and the discovery of alternate protease targets for PAI-1
    19. Was the SARS CoV2 lung epithelial cell culture analysis performed in BSL3?

    Minor critiques

    1. Results section heading "SERPINs are differentially expressed individuals with COVID-19 and in response to respiratory virus infection in a model of the human airway epithelium." The word in needs to be inserted between expressed and individuals

    Significance

    Overall this is a very simple, although extensive and excellent, study analyzing a wide range of data form patients with bronchoalveolar lavage and epithelial cell samples, human epithelial cell cultures after infection with a range of respiratory viruses as well as the development of a 3D in silico analysis of potential protease and serpin interactions. These correlations between changes in serpin and protease expression with viral infections and potential new interactions for serpins with previously non identified proteases is of clear interest. This shows an excellent correlation but as with big data sets this does not provide a true cause and effect - rather providing new potential directions for analysis of these interactions in viral infections in lung epithelium and this is valuable as a basis for ongoing studies. Prior work evidencing other cross class serpin protease targets as well as limitations related to the analyses as discussed in the critiques above should be noted and the abstract and title could better describe and define the studies as performed.

  4. Version published to 10.1101/2022.11.18.517133v2 on bioRxiv
  5. Version published to 10.1101/2022.11.18.517133v1 on bioRxiv