Genetically encoded discovery of perfluoroaryl macrocycles that bind to albumin and exhibit extended circulation in vivo

Peptide-based therapeutics have gained attention as promising therapeutic modalities, however, their prevalent drawback is poor circulation half-life in vivo. In this paper, we report the selection of albumin-binding macrocyclic peptides from genetically encoded libraries of peptides modified by perfluoroaryl-cysteine SNAr chemistry, with decafluoro-diphenylsulfone (DFS). Testing of the binding of the selected peptides to albumin identified SICRFFC as the lead sequence. We replaced DFS with isosteric pentafluorophenyl sulfide (PFS) and the PFS-SICRFFCGG exhibited KD = 4–6 µM towards human serum albumin. When injected in mice, the concentration of the PFS-SICRFFCGG in plasma was indistinguishable from the reference peptide, SA-21. More importantly, a conjugate of PFS-SICRFFCGG and peptide apelin-17 analogue (N3-PEG6-NMe17A2) showed retention in circulation similar to SA-21; in contrast, apelin-17 analogue was cleared from the circulation after 2 min. The PFS-SICRFFC is the smallest known peptide macrocycle with a significant affinity for human albumin and substantial in vivo circulation half-life. It is a productive starting point for future development of compact macrocycles with extended half-life in vivo.


"
The switching from DFS to PFS also exhibited a minimal effect on the binding of peptides 5b and 5c (Figure S20)."This statement should be expanded -in the figure it looks like there is higher signal decrease by PFS compared to DFS. 8. Can the authors highlight the importance of binding to a different site than other known albumin binders?Also, is the binding site for SA-21 known and does it bind to a different site than 5c? 9.In Figure 8, what is the unit for the y axis on the peptide retention graph?Is the n = 3 referring to biological or technical replicates?It may be beneficial to report the calculated serum half-lives of the peptides.
Reviewer #2 (Remarks to the Author): The present work describes the selection of phage-encoded macrocyclic peptides, modified with cysteine-reactive perfluoroaromatic linkers, that are capable of binding human serum albumin (HSA).The best isolated perfluoroaryl-macrocycle, named 5c, binds HSA with Kd values of 4 -6 μM.The macrocycle 5c cross-reacts with murine albumin and does not bind two unrelated control proteins.Docking studies suggested that the primary HSA binding site for perfluoroaryl-macrocycle 5c is the fatty acid binding site 1 (FA1).Next, the half-life of compound 5c was assessed in vivo.Overall, the manuscript yields "mixed feelings".I like the approach; the topic covered in this article is likely to be of interest to a reasonable number of scientists working in this field and certainly worthy of investigation.The manuscript is concise and easy to read.I am a little less enthusiastic about the characterization of the identified binder.To be honest, I am struggling with this flaw, I would have had less hesitation if the authors had assessed and validated their technology against multiple targets or at least provided a more thoughtful characterization of the isolated HSA binders.I don't know if this is a Nature Communicationlevel work, but this should be an editor level decision.
The novelty is the "Achille's heel" of this manuscript as it looks like more an extension of previous works by the same group, which, one may say, "scoops the originality of the approach".Indeed, the design, chemical modification, and next-generation sequencing of such libraries have been previously detailed in "Compositional Bias in Naive and Chemically-modified Phage-Displayed Libraries uncovered by Paired-end Deep Sequencing, Sci.Rep. 2018, 8 (1), 1214" (reference 44) and in "Rapid biocompatible macrocyclization of peptides with decafluoro-diphenylsulfone, Chem.Sci.2016, 7 (6), 3785-3790" (reference 42) by the same group.The current manuscript adds an extra step and describes the application of such libraries for the selection of PFS-modified synthetic macrocycles against HSA.
The methodology is detailed and clearly described.The technology, as described, appears to work well for HSA.
The conclusions drawn by the authors are not always convincing or supported by clear experimental data.I recommend the authors to provide a more balanced evaluation of their findings as well as of the challenges that still remain.The technology as described is proved to work for one only target so far.
Below are listed a few points where the clarity of the manuscript can be improved.They are meant to make this work stronger and more impactful.
Major revisions: 1. Page 5, line 86: "The main mechanism leading to the long half-life of albumin and antibody are similar: both proteins interact with FcRn on the surface of immune cells".This sentence is partially incorrect.The authors might have mistaken FcRn with FcyRs.FcRn is primarily expressed on the surface of endothelial cells, the major site of IgG catabolism and recycling, and on the surface of interstitial macrophages.I suggest the authors to edit the sentence and eventually cite these two works: (https://doi.org/10.1016/j.jconrel.2021.07.007) 2. Page 7 (line 127).The authors state "There is a need for development of other albumin binding peptide modalities that have lower molecular weight".Why do we need lower molecular weight molecules?What would be the advantages of smaller peptides with low affinity versus slightly bigger peptides but with higher affinity?Why not use then available small molecules capable of binding albumin?The literature is plenty of small molecules (e.g., evans blue, drugs, fatty acids, etc.) that have been successfully used to prolong the half-life of peptides and small proteins such as insulin.I suggest that the authors carefully clarify this point that, at the end, is the aim behind all their work on finding smaller peptide macrocycles against HSA.
3. Page 7 (line 133).The authors state "We hypothesized that a perfluoroaromatic linchpin would serve as useful pharmacophore recognized by one of the binding sites of HSA similarly to the binding of fatty acid in lipidated peptides".Do the authors have experimental evidence supporting their hypothesis that the two perfluoroaromatic compounds (OFS and PFS) applied in this study are capable of binding HSA?If yes, please provide the experimental data.If no, please measure the binding affinity of both OFS and PFS to HSA.
4. Page 9.The difference in enrichment observed between the first (2-fold increase) and the second (200-fold increase) discovery campaigns against HSA is impressive.Please provide a critical and thoughtful explanation for this. 5. Page 9 and 10.Different selection strategies employed against the same target yielded different peptide consensus sequences.Please provide a critical and thoughtful explanation for this.
6. Page 14.The fact that two macrocyclic peptides, 5c and 6c, that differ by only three amino acids, display a 100-fold difference in Kd values is remarkable and must be stressed.This data would support the fact that the binding is not driven by the perfluoroaromatic linker only, but the peptide also appears to play a role.Please provide additional insights on this finding.
7. I must admit that I am very surprised by the ITC data shown in Figure S13, S14 and S15.The KD values reported for SA-21 peptide are very far from those described in literature (~15-fold higher in figure S13, ~100-fold higher in figure S14 and ~25-fold higher in figure S15).This makes me think that something in the design of the experiment, preparation of the protein and/or in the setting of the instrument went wrong.I suggest that the authors characterize macrocycle binding using a complementary technique such as SPR, which is the same technique used by Dennis and colleagues to determine the affinity of SA-21 (Kd = 467 nM).I must also add that the multi-site binding behaviour observed by the authors for all peptides when performing ITC measurements is not surprising as HSA is known to be highly flexible and undergoes conformational changes upon binding.8. Page 16.Docking studies indicate that perfluoroaryl-macrocycle 5c binds fatty acid binding site 1 (FA1).However, no experimental data have been shown in support of this.The authors should experimentally confirm their in silico findings.There are several commercially available drugs capable of binding FA1 site (please see two manuscripts listed below) that can be used as probe in competitive NMR, FP, ITC and SPR studies.Why not try to co-crystallise 5c in complex with HSA?  6.The authors showed that the binding of macrocycle 5c did not decrease in the presence of any of the three commercially available drugs tested.However, if we carefully analyse the structure of HSA in complex with diclofenac described by Zhang Y. et al. (Chem Biol Drug Des 86: 1178-1184; PDB: 4Z69), we realize that one molecule of diclofenac is bound to FA1 site in the IB subdomain.If the docking experiments predict that 5c binds to FA1, why did the authors not observe competition when performing the 19F NMR experiments in the presence of diclofenac?Please provide a critical and thoughtful explanation for this.
10. Page 19.The authors should determine and provide all the major pharmacokinetic parameters such as elimination half-life (t1/2), maximal concentration (Cmax), volume of distribution (Vd), clearance (CL), and so forth.11.Ideally, I would have liked to see a better characterization of the binding mode of macrocycle 5c in complex with HSA.For instance, it would be nice to see if 5c competes with other HSA-binding peptides (e.g., DX-263, SA-21 and FITC labeled linear peptide).Similarly, it would be interesting to investigate the metabolic stability of 5c in comparison to previously reported HSA-binding peptides.I believe all these additional data would add significance and robustness to the whole story.
Minor revisions: 1 Figure 1ABC.Chemical structures of previously reported albumin-binding peptides DX-263 (A), SA-21 (B) and FITC labeled linear peptide (C) don't add something new to the work and should, in my opinion, be moved to the supplementary information section.
2 Figure 1D.This scheme is somehow repeated in Figure 2 (top).One of the two draws is sufficient to understand the technology.
4 Figure 5B.Please add here also the FP assay for BODIPY labelled 5c titrated against mouse serum albumin.
5 Figure 6A.Please label the different fatty acid binding sites (FAs) on the two HSA structures reported.This will be helping the reader to quickly locate the different FA sites discussed in the text.
6 Figure 7A.The resolution of this figure is very low and must be improved.I also advise the author to label the different FAs on the HSA structure.
7 Figure 8.The author should plot the pharmacokinetic data as a canonical standard curve rather than bar diagrams.
8 There are several typos in the text which need to be addressed.I suggest that the authors read again the paper carefully.
Reviewer #3 (Remarks to the Author): The authors present a new linkage using decafluoro-diphenylsulfone which they implement in a genetically encoded library to discover potent albumin binder.Albumin is an interesting target to prolong circulation half live of small molecules and peptides.
The presented strategy follows the idea of bridging peptides via cysteine functionalization, a well reported strategy of the generation of (bi-)cyclic peptides.<sup>1</sup>The octafluoro-diphenylsulfone crosslinked macrocyclic peptides represent a novel class of bridged macrocycle.The linker is unstable towards nucleophilic cysteines and was replaced by a nearly isosteric perfluorophenyl-sulfide for investigations on identified binders after the phage display.The authors showed that the replacement didn't alter binding.Still this can be considered as obstacle, that other bridging strategies do not face.The authors state, that this attenuated reactivity towards thiols can be used advantageous, to covalently bound to cysteines in proteins.No further experiments were performed into this direction.
The authors present an optimized phage display screening against albumin and identify enriched sequences.This process is underlined with a lot of thorough data.They further validate those candidates by 19F-NMR using the fluorinated handle.A main advantage of the strategy, according to them, is the implementation of a fluorinated linker that allows the use of 19F-NMR.Despite showing the occurrence of binding events, the signals obtained from the NMR analyses are unsuitable for drawing conclusions on binding strength of the selected samples.Furthermore, a fluorescence polarization binding assay was used to determine KDs of selected hits.The best binder achieves a KD-value of 4-6 uM.It competes with fatty acids (binding affinities between 0.5 and 60 μM) but lags behind the top peptidic binders (15 nM<sup>2</sup> or 39 nM [H. A. P. Revets and C. Boutton, Peptides capable of binding to serum albumin and compounds, constructs and polypeptides comprising the same, WO2011095545, 2011]).<sup>3</sup>Regarding the small genetic library and approach of proving a concept, this value can be considered good.
In the next step they use computational methods to compare binding pockets of the most potent binder (5c) with known albumin binders.In my opinion, this experiment does not follow the thread of the report.The binding site of the peptide is not determined, the representation of the data is inadequate and I am missing the outcome of this paragraph.(Probably the authors are self-aware of that, because this part is also missing in the conclusion.)Finally, the authors test their compound in vivo by evaluating serum stability of their top binder (5c) in mouse.The data shows increased half live circulation compared to the negative control.This experiment is a nice complementation of their in vitro data and totally reasonable for the stage of this project but is indeed, as stated, a minimalistic starting point.
The authors use a sound methodology to proof their new concept.The right conclusions are drawn from the presented data.Nevertheless, I am not convinced that this study is suitable for publication in nature communications yet.I can´t see any superiority of the novel linker compared to established strategies.The need for replacement within the identification of a binder in combination with missing significance of the presented 19F-NMR data aren't convincing.In my opinion the computational data is meaningless in the presented context.
To make this an impactful report I would expect the authors to pick up one of the loose ends within the story: 1) The attenuated reactivity of the octafluoro-diphenylsulfone macrocycles for covalent linkage as advantageous feature 2) Getting profound knowledge about the binding of the identified motive to albumin by profound computational studies or crystal structures 1

Reviewer #1 (Remarks to the Author):
This is a relatively thorough and well-written article that describes the development of a macrocyclic peptide that binds to albumin.The research has the potential to be helpful in extending the serum stability of peptides, as the identified peptide can be incorporated into other peptide libraries to identify compounds that bind to other targets.
We thank the reviewer for this positive evaluation of the manuscript.

R1.1. However, the manuscript stopped short without showing any major application. A critical application will be necessary to show the real potential impact of the identified macrocyclic peptide.
New results demonstrate that RFF peptide (14c) can extend the circulation half-life of attached payloads (e.g.therapeutically relevant peptide apelin-17 analogue).We have incorporated the following data and figure (Fig. 6) into the main text.
"…we observed that conjugation of PFS-SICRFFCGGG to apelin-17 analogue (N3-PEG6-NMe17A) increased its plasma retention significantly.We could not detect apelin-17 analogue in plasma even 2 minutes after injection and such rapid clearance of apelin analogues is consistent with previous observations. 1In contrast, plasma concentration of compound 20n, produced by ligation of N3-PEG6-NMe17A 2 to PFS-SICRFFCGGGZ (20c) by Cu-catalyzed 1,3-dipolar cycloaddition was indistinguishable from the plasma concentration of 20c after 3 hours.Such enhanced circulation of apelin-17 analogue promoted by a relatively small albumin-binding peptide is a promising starting point for more in-depth investigation and benchmarking to previously reported conjugates of apelin and albumin binding antibody 3 or apelin and 40 kDa PEG group. 4"

R1.2. A direct comparison with the conjugation with a long fatty acid for prolonging serum half lives of a peptide drug will be necessary as well. Therefore, a major revision with a demonstrated application is required.
Careful comparison of the attachment of all major albumin binders (including SA-21, 5 4-(piodophenyl)butyric acid, 6 AlbudAb TM7 , Albubinder1 8 , and lipids) would require significant additional experimentation, including prohibitively expensive in vivo pharmacokinetic studies.It is beyond the scope of our current manuscript.6][7][8] Such studies usually appear in follow-up publications.Requesting it here introduces an unfair double standard.
Additional minor concerns are provided below.

R1.3. Can the authors comment on why an N-terminal serine library was used? This N-terminal serine shows no purpose.
In this publication, we used libraries previously developed and characterized in our lab.Our lab is generally interested in attachment of payloads via N-terminal Ser; hence all phage libraries produced in our lab contain a constant N-terminal Serine.The nature of the N-terminal residue is not relevant to the selection, and constant N-terminal residues are almost always present in phage display vectors to avoid biases associated with processing of randomized N-terminal residue.The 19 F NMR binding assay showed that the KD value increased from 3.6 µM to 12.1 µM for the Ser→ Ala mutant.
We added the following sentences to the manuscript to clarify these points: "The SXCX3C library has been used in our previous publications successfully using only one round of sequencing [9][10][11] .It was sufficiently small (160,000 variants) to be fully covered by NGS and copy number of each member of the naïve library was sufficiently large to afford quantification of the selection by NGS and DE analysis." We observed that R4A or F5A changes in 14c ablated the binding between the perfluoromacrocycles and HSA.In contrast, the S1A and I2A changes resulted in smaller changes in KD of 3-and 10-fold respectively.The alanine scans indicate that variation to any of the four discovered amino acids in SXCXXXC sequence negatively affects the binding.Even serine, a constant residue in selection, becomes modestly important in albumin recognition as well."R1.4.This may just be an issue with formatting, but Figure 2 A and B are confusing and it is hard to distinguish the parts of the figure.
We amended the figure 2A and B and added the following clarification to the legend: "Panning with chemically modified phage libraries (A) Modification of phage-displayed SXCX3C disulfide library by DFS to yield OFS-SXCX3C library (B) The OFS-SXCX3C library panned against a mixture of biotinylated HSA and His-tag expressed T4-GP in solution containing unlabelled milk proteins.Targets were captured separately with avidin beads and Ni-NTA beads affinity beads.In the negative control, OFS-SXCX3C library was panned against biotinylated ConA and captured with avidin beads."

R1.5. Can the authors comment on the % cyclization of the 3mer phages being cyclized with OFS? Are there any experimental details confirming that such a small library will cyclize?
The OFS linchpin utilized in this study was adapted from a previous study, where we confirmed that OFS could modify 85% of the displayed peptides. 12We further validated the conversion of peptides of SXCX3C library in this manuscript.(New Supplementary Fig. S3).Supplementary Fig. S3.Quantification of modification of SXC3C libraries by DFS.(A) M13-phage displayed disulfide library was reduced with TCEP and the exposed cysteine thiols were modified with biotin-peg iodoacetamide (BIA).The streptavidin capture (capture 1) of the biotinylated library reveals the percentage of available cysteines (B) M13-phage displayed thiol library was modified with DFS (C) Left over cysteines were monitored by a second biotinylation with BIA followed by a streptavidin capture (capture 2) (D) The difference between capture 1 and 2 reveals the percentage of library modified by DFS.

R1.6. Could the authors expand on their reasoning to only perform one round of selection for the third campaign? Is this related to the total phage input compared to the library size?
We added the following clarification to the results: " The SXCX3C library has been used in our previous publications successfully using only one round of sequencing [9][10][11] .It was sufficiently small (160,000 variants) to be fully covered by NGS and copy number of each member of the naïve library was sufficiently large to afford quantification of the selection by NGS and DE analysis."

R1.7. Can the authors expand on their analysis of the ITC experiments and give a possible explanation for why they did not show 1:1 binding?
We added the following clarification to the results: "… multi-step binding to albumin in ITC experiments has been observed in previous reports. 13" R1.8.To highlight the necessity of the PFS linkage, it would be beneficial to compare the affinity to the linear peptide and/or disulfide cyclized peptide to show that they either do not bind or have considerably lower affinity "The concentration of SA-21 in plasma at 2 and 60 min decreased subtly from 410 4 to 310 4 ng/mL whereas concentration of macrocycle 14c in plasma was 3 and 1.510 4 ng/mL at the same time points (Fig. 6B).In contrast, four derivatives of 14c were excreted at 2 minutes or were not detectable at 60 minutes.The fast-clearing analogues were linear SICRFFCGGG with the cysteines alkylated by iodoacetamide (14l), the alanine mutants PFS-SICAFFCGGG (23c) and PFS-SACRFFCGGG (21c), and the control PFS-STCQGECGGG sequence (17c);" and "Replacing PFS by hexafluorobenzene (HFB, 14j) and decafluorobiphenyl (DFB, 14k) did not ablate retention in plasma completely (Fig. 6A); however, concentration of HFB-and DFBmodified macrocycles after 60 minutes were factor of 10 lower when compared to PFS-modified 14c parent.These observations reinforced the important structural features of 14c: both the amino acids sequence of 14c and specific conformational constraint of this sequence imposed by PFS linker are critical for even a short-term retention in circulation." and "Encouraged by the initial evaluation of serum retention, we extended the study to three hours.Gratifyingly we observed that the concentration of 14c and SA-21 in plasma after 3 hours were indistinguishable (Fig. 6C).Analogs of 14c in which PFS was replaced by HFB (14j) or DFB (14k) perfluoroarenes consistently showed over 10x decrease in plasma concentration; a decrease of nearly 100x was observed for 14m analogue in which perfluoroarene was substituted by metabromoxylene linker (MBX) (Fig. 6C)."R1.9."The switching from DFS to PFS also exhibited a minimal effect on the binding of peptides 5b and 5c (Figure S20)."This statement should be expanded -in the figure it looks like there is higher signal decrease by PFS compared to DFS.
We added the following clarification to the result section: "The switching from DFS (18f) to PFS (18d) exhibited no difference in FP experiment (Supplementary Fig. S18A), but minor differences were observed by NMR (Supplementary Fig. S18B).".

R1.10. Can the authors highlight the importance of binding to a different site than other known albumin binders? Also, is the binding site for SA-21 known and does it bind to a different site than 5c?
Although SA-21 peptide was discovered more than 20 years ago and used in multiple publications, the binding site for SA-21 was never determined.As of today, the team of Dennis and co-workers have not been able to map the site either by inhibition studies or any other methods (personal email communication with Mark Dennis).8, what is the unit for the y axis on the peptide retention graph?Is the n = 3 referring to biological or technical replicates?It may be beneficial to report the calculated serum half-lives of the peptides.

R1.11. In Figure
We have amended this figure.Note that it is now moved to the supplementary information section (Supplementary Fig. S32).

Reviewer #2 (Remarks to the Author):
The present work describes the selection of phage-encoded macrocyclic peptides, modified with cysteine-reactive perfluoroaromatic linkers, that are capable of binding human serum albumin (HSA).The best isolated perfluoroaryl-macrocycle, named 5c, binds HSA with Kd values of 4 -6 μM.The macrocycle 5c cross-reacts with murine albumin and does not bind two unrelated control proteins.Docking studies suggested that the primary HSA binding site for perfluoroaryl-macrocycle 5c is the fatty acid binding site 1 (FA1).Next, the half-life of compound 5c was assessed in vivo.Overall, the manuscript yields "mixed feelings".I like the approach; the topic covered in this article is likely to be of interest to a reasonable number of scientists working in this field and certainly worthy of investigation.The manuscript is concise and easy to read.
We thank the reviewer for their positive feedback.

R2.1. I am a little less enthusiastic about the characterization of the identified binder. To be honest, I am struggling with this flaw, I would have had less hesitation if the authors had assessed and validated their technology against multiple targets or at least provided a more thoughtful characterization of the isolated HSA binders. I don't know if this is a Nature Communication-level work, but this should be an editor level decision.
Replies to comments R1.1, R1.2, R2.11, R2.12, R2.14, new figures 3, 5, 6, and Supplementary Figures S21-S28, S29-S32 provide a more thorough characterization of HSA binders R2.2.The novelty is the "Achille's heel" of this manuscript as it looks like more an extension of previous works by the same group, which, one may say, "scoops the originality of the approach".Indeed, the design, chemical modification, and next-generation sequencing of such libraries have been previously detailed in "Compositional Bias in Naive and Chemically-modified Phage-Displayed Libraries uncovered by Paired-end Deep Sequencing, Sci.Rep. 2018, 8 (1), 1214" (reference 44) and in "Rapid biocompatible macrocyclization of peptides with decafluoro-diphenylsulfone, Chem.Sci.2016, 7 (6), 3785-3790" (reference 42) by the same group.The current manuscript adds an extra step and describes the application of such libraries for the selection of PFS-modified synthetic macrocycles against HSA.The methodology is detailed and clearly described.The technology, as described, appears to work well for HSA.
Our manuscript describes a previously unreported selection of perfluoro-aryl libraries (specifically three separate selection campaigns of various perfluoro-aryl cyclic architectures).The initial synthesis of such libraries is cited correctly, and they have been appropriately cited in our manuscript.Characterization of the disulfide precursor libraries by next-generation sequencing is also cited appropriately, although the purpose of the cited manuscript is very divergent from the current manuscript, and it aimed to characterize sequencing and expression biases rather than any selection.We believe disclosing prior research directions should not be a barrier to this publication.Prior publications cover very different areas that do not overlap with most topics in this publication.

R.2.3. The conclusions drawn by the authors are not always convincing or supported by clear experimental data. I recommend the authors to provide a more balanced evaluation of their findings as well as of the challenges that still remain. The technology as described is proved to work for one only target so far.
We provide additional experimental data per specific request of all three reviewers with the exception of very onerous requests (e.g.comparison of all known albumin binders in-vivo.)Below are listed a few points where the clarity of the manuscript can be improved.They are meant to make this work stronger and more impactful.
Major revisions: R2.4.Page 5, line 86: "The main mechanism leading to the long half-life of albumin and antibody are similar: both proteins interact with FcRn on the surface of immune cells".This sentence is partially incorrect.The authors might have mistaken FcRn with FcyRs.FcRn is primarily expressed on the surface of endothelial cells, the major site of IgG catabolism and recycling, and on the surface of interstitial macrophages.5] " We have cited the mentioned work.

R2.5. Page 7 (line 127). The authors state "There is a need for development of other albumin binding peptide modalities that have lower molecular weight". Why do we need lower molecular weight molecules? What would be the advantages of smaller peptides with low affinity versus slightly bigger peptides but with higher affinity? Why not use then available small molecules capable of binding albumin?
The literature is plenty of small molecules (e.g., evans blue, drugs, fatty acids, etc.) that have been successfully used to prolong the half-life of peptides and small proteins such as insulin.I suggest that the authors carefully clarify this point that, at the end, is the aim behind all their work on finding smaller peptide macrocycles against HSA.
Our manuscript already contained a proposed application of the short peptide structures: "Similarly short peptides that bind to HSA could be used in tandem with therapeutic peptide or protein sequences to dial in predictable half-life for such therapeutics.Such short albuminbinding peptides could empower the development of many future therapeutic peptides because they could be built into any genetically encoded peptide library (e.g., displayed on phage, RNA, and other platforms) to give rise to billion-scale libraries with predictable in vivo half-life.However, short HSA-binding peptides are scarce." Few other discussion points have been added to the main text.
"Small-molecules have been developed with astonishing affinity for human albumin, but such molecules exhibit no binding to mouse albumin and no retention in mice, unless such mouse is engineered to express human albumin. 8Mid-size peptide macrocycle described in this report might provide an interesting opportunity for sufficient affinity but also much desired crossreactivity between the species."and "… the SICRFFC-motif can be easily re-introduced into phage-displayed libraries to serve as a constant N-terminal albumin binding motif and giving rise to libraries with predictable circulation half-life.The compact nature of the macrocycle allows introducing this albuminbinding moiety to diverse display platforms within the initial stages of screening of geneticallyencoded libraries, making it possible to perform de-novo discovery of peptide and macrocyclic modalities with predictable circulation life-time in animal models."R2.6 Page 7 (line 133).The authors state "We hypothesized that a perfluoroaromatic linchpin would serve as useful pharmacophore recognized by one of the binding sites of HSA similarly to the binding of fatty acid in lipidated peptides".Do the authors have experimental evidence supporting their hypothesis that the two perfluoroaromatic compounds (OFS and PFS) applied in this study are capable of binding HSA?If yes, please provide the experimental data.If no, please measure the binding affinity of both OFS and PFS to HSA.
Based on all data collected to date, it is appropriate to amend the statement as: "We hypothesized that a perfluoroaromatic linchpin might serve as a useful pharmacophore and it might be recognized by one of the binding sites of HSA similar to the binding of fatty acid; however, as we observed in the NMR studies and late-stage pharmacokinetic evaluation of albumin-binding macrocycles, the primary function of perfluoro-linchpin is to constrain the discovered peptide macrocycle in a productive albumin-binding conformation.While perfluoroaromatic linchpin alone does not equip a random peptide with albumin-binding properties, the change in the shape of perfluoroaromatic linchpin is detrimental to the albuminbinding properties and in vivo circulation."

R2.7. Page 9. The difference in enrichment observed between the first (2-fold increase) and the second (200-fold increase) discovery campaigns against HSA is impressive. Please provide a critical and thoughtful explanation for this.
We added the following clarification: "…the increase in round-to-round phage recovery was only modest.Despite depletion on protein A, the enriched population bound equally to HSA and protein A (Supplementary Fig. S1D).To mitigate these problems, the second discovery campaign increased the stringency and altered the immobilization of HSA between 96-well plate in rounds 1 and 3 and biotinylated HSA immobilized onto streptavidin beads in round 2 (Supplementary Fig. S2A).Such changes to selection stringency led to 200-fold increase in recovery in round 3 when compared to rounds 1 and 2."

R2.8. Page 9 and 10. Different selection strategies employed against the same target yielded different peptide consensus sequences. Please provide a critical and thoughtful explanation for this.
We added the following clarification: "In summary, three selection strategies against HSA yielded divergent binding motifs.Such divergence of selection campaigns is not surprising because protein immobilization, depletion and amplification strategies were different among these strategies selection."R2.9.Page 14.The fact that two macrocyclic peptides, 5c and 6c, that differ by only three amino acids, display a 100-fold difference in Kd values is remarkable and must be stressed.This data would support the fact that the binding is not driven by the perfluoroaromatic linker only, but the peptide also appears to play a role.Please provide additional insights on this finding.
We added the following new text and data: " A substantial change in measured KD values resulting from subtle amino-acid changes in PFS-macrocycles highlighted that the binding is not driven by the perfluoroaromatic linker alone.The peptide sequence plays a major role.Alanine scan of 14c further reinforced this observation (Fig. 3B-3C).We observed that R4A or F5A changes in 14c ablated the binding between the perfluoro-macrocycles and HSA.In contrast, the S1A and I2A changes resulted in smaller changes in KD of 3-and 10-fold respectively.The alanine scans indicate that variation to any of the four discovered amino acids in SXCXXXC sequence negatively affects the binding.Even serine, a constant residue in selection, becomes modestly important in albumin recognition as well."S13, S14 and S15.The KD values reported for SA-21 peptide are very far from those described in literature (~15-fold higher in figure S13, ~100-fold higher in figure S14 and ~25-fold higher in figure S15).This makes me think that something in the design of the experiment, preparation of the protein and/or in the setting of the instrument went wrong.I suggest that the authors characterize macrocycle binding using a complementary technique such as SPR, which is the same technique used by Dennis and colleagues to determine the affinity of SA-21 (Kd = 467 nM).I must also add that the multi-site binding behaviour observed by the authors for all peptides when performing ITC measurements is not surprising as HSA is known to be highly flexible and undergoes conformational changes upon binding.

R2.10. I must admit that I am very surprised by the ITC data shown in Figure
We have struggled with ITC characterization of albumin-peptide interactions, and, as reviewers point out, we saw very unsatisfactory, variable outcomes.We agree that such a problem might be inherent to albumin peptide interactions, variability of albumin of preparation of albumin solutions, or other factors such as the inherent binding of albumin to nearly every physical surface.We note similar multi-site binding in other publications and a complete lack of ITC data, including the three cornerstone peptide binding papers from Genentech 5 , Dyax corp 16 , and the Heinis 17 group.Our group attempted the SPR several times, but we observed a very unusual negative index of refraction change, in the interest of clarity, it might be best not to show these SPR data in this manuscript.Re: Why not co-crystallize?This is an unusually onerous request for a review process.We would like to disclose that we tried crystallization for many years in collaboration with Scott Lovell, who is the director of the Protein Structure and X-Ray Crystallography Laboratory at the University of Kansas.Our attempts to co-crystallize 14c with HSA failed.Given the uncertainty of crystallization and prohibitive expenses, we discontinued our attempts.

R2.11. Page 16. Docking studies indicate that perfluoroaryl-macrocycle 5c binds fatty acid binding site 1 (FA1). However, no experimental data have been shown in support of this. The authors should experimentally confirm their in silico findings. There are several commercially available drugs capable of binding FA1 site (please see two manuscripts listed below) that can be used as probe in competitive NMR, FP, ITC and SPR studies. Why not try to co-crystallise 5c in complex with
The following text was added to main text "To identify the binding location of PFS-SICRFFCGGG (14c), we attempted to co-crystalize 14c with HSA but were not successful." The following text was added to the main text to provide our best attempts to link computational and experimental predictions: "Nine distinct sites on HSA bind to fatty acids 18 , some of which also accommodate ibuprofen and diclofenac [19][20] (Supplementary Fig. S21, Fig. 5A).Docking of 14c to these nine sites identified binding site 1, which is close to Hemin binding site, as the most favorable (Fig. 5A).The docking score for site 1 averaged over all the docking calculations across five distinct HSA structures (PDB 1e7e, 1e7f, 1e7g, 1e7h, 1e7i) was -8.95 ± 1.0 kcal/mol whereas the scores for the next sites were -6.6 ± 1.1, -6.2 ± 0.7, and -6.0 ± 1.2 kcal/mol (Supplementary Fig. S22).Observed docking preference away from known ibuprofen and diclofenac sites corroborated the experimental observation (Supplementary Fig. S22 and S23).As site 1 is hemin-binding site, a desired experimental validation of this prediction would be inhibition of HSA:14a interaction by hemin.We attempted to measure 14c:HSA interactions in the presence of hemin, but the results were inconclusive due to strong association of hemin and 14c in 19 F NMR experiments."and "The PMF corroborated that pocket 1 has significantly stronger binding energy than the others (Fig. 5A, Supplementary Fig. S24).Furthermore, the calculated G of free energy perturbation calculations (FEP) for 14c and five Ala mutants of 14c were in alignment with KD measured for the same alanine mutants in the 19 F NMR assay (Fig. 3B-C).The biggest loss of function for R4A mutant, highlighted the importance of Arginine interaction for albumin binding (Fig. 5E-G); in contrast, both NMR and FEP agreed on a relatively minor role of N-terminal Ser and minor loss for S1A mutant."R2.12. Figure 6.The authors showed that the binding of macrocycle 5c did not decrease in the presence of any of the three commercially available drugs tested.However, if we carefully analyse the structure of HSA in complex with diclofenac described by Zhang Y. et al. (Chem Biol Drug Des 86:  1178-1184; PDB: 4Z69), we realize that one molecule of diclofenac is bound to FA1 site in the IB subdomain.If the docking experiments predict that 5c binds to FA1, why did the authors not observe competition when performing the 19F NMR experiments in the presence of diclofenac?Please provide a critical and thoughtful explanation for this.
We added the following explanation to the supplementary information section.
"When examining the superposition of diclofenac with 14c (docking) in IB site of HAS (Supplementary Fig. S21), the overlap of 14c and diclofenac occurs only on one side of 14c.14c binds to 19 amino acids of HSA, and diclofenac binds to 9 amino acids of HSA, and 7 out of 9 amino acids which bind to diclofenac also bind to 14c (L154, G189, Y161, K190, R186, H146, I142).Furthermore, the IB site of HSA in the crystal structure 4z69 binds simultaneously to both diclofenac and a fatty acid so that diclofenac binding could be modulated by the fatty acid binding.There is no fatty acid in the 14c-HSA complex obtained by docking.Therefore, 14c has a larger binding site to FA1 than diclofenac, and a more favorable docking score of -10 kcal/mol, compared to diclofenac's binding score of -6.6 kcal/mol (Supplementary Fig. S21D)."R2.13.Page 19.The authors should determine and provide all the major pharmacokinetic parameters such as elimination half-life (t1/2), maximal concentration (Cmax), volume of distribution (Vd), R2.21. Figure 8.The author should plot the pharmacokinetic data as a canonical standard curve rather than bar diagrams.
We have extended the data and discussion on pharmacokinetic properties of 14c and its analogues.New Figure 6 is presented as curves for ng/mL of compounds.

R2.22.
There are several typos in the text which need to be addressed.I suggest that the authors read again the paper carefully.
We thank the reviewer for their feedback, and we have corrected the typos to the best of our ability.

R3.1. The authors present a new linkage using decafluoro-diphenylsulfone which they implement in a
genetically encoded library to discover potent albumin binder.Albumin is an interesting target to prolong circulation half live of small molecules and peptides.The presented strategy follows the idea of bridging peptides via cysteine functionalization, a well reported strategy of the generation of (bi-)cyclic peptides. 1  (7), 502-7.
We thank the authors for the evaluation and historical context.

R3.2. The octafluoro-diphenylsulfone crosslinked macrocyclic peptides represent a novel class of bridged macrocycle. The linker is unstable towards nucleophilic cysteines and was replaced by a nearly isosteric perfluorophenyl-sulfide for investigations on identified binders after the phage display.
The authors showed that the replacement didn't alter binding.Still this can be considered as obstacle, that other bridging strategies do not face.
We thank the reviewer for their evaluation.While the unwanted reactivity of OFS was an obstacle during the characterization stage, we overcame it by replacement of DFS with PFS.

R3.3. The authors state, that this attenuated reactivity towards thiols can be used advantageous, to covalently bound to cysteines in proteins. No further experiments were performed into this direction.
We believe this type of experiment is beyond the scope of this manuscript and hope to explore this avenue in future projects.It would not be fair to diminish the value of the current manuscript based on the forward-looking proposals in the conclusion section.

R3.4.
The authors present an optimized phage display screening against albumin and identify enriched sequences.This process is underlined with a lot of thorough data.They further validate those candidates by 19F-NMR using the fluorinated handle.A main advantage of the strategy, according to them, is the implementation of a fluorinated linker that allows the use of 19F-NMR.Despite showing the occurrence of binding events, the signals obtained from the NMR analyses are unsuitable for drawing conclusions on binding strength of the selected samples.
We appreciate the reviewer's comments on our use of 19 F NMR for characterizing the binding of our lead candidate.We believe that the reviewer underestimates the power of 19 F NMR, as it has been shown to be a valuable tool for studying ligand-protein interactions in many previous studies.We would like to clarify that our use of 19 F NMR allowed us to determine the KD value for binding and observe binding events, which are crucial pieces of information for understanding the mechanism of action of our lead candidate.Furthermore, we obtained quantitative KD and structure-activity relationship data for alanine mutants of the lead candidate using 19 F NMR.
To illustrate the sheer beauty and power of 19 F NMR incorporate the 19 F NMR measurements in main text Fig. 3, and include this figure here for the reviewer's convenience.R3.5.Furthermore, a fluorescence polarization binding assay was used to determine KDs of selected hits.The best binder achieves a KD-value of 4-6 uM.It competes with fatty acids (binding affinities between 0.5 and 60 μM) but lags behind the top peptidic binders (15 nM 2 or 39 nM [H. A. P. Revets and C. Boutton, Peptides capable of binding to serum albumin and compounds, constructs and polypeptides comprising the same, WO2011095545, 2011]). 3Regarding the small genetic library and approach of proving a concept, this value can be considered good.REF 2. Zorzi, A.; Middendorp, S. J.; Wilbs, J.; Deyle, K.; Heinis, C., Acylated heptapeptide binds albumin with high affinity and application as tag furnishes long-acting peptides.Nat Commun 2017, 8, 16092.
We thank the reviewer for this analysis and designation of the identified ligands as "good" R3.6.In the next step they use computational methods to compare binding pockets of the most potent binder (5c) with known albumin binders.In my opinion, this experiment does not follow the thread of the report.The binding site of the peptide is not determined, the representation of the data is inadequate and I am missing the outcome of this paragraph.(Probably the authors are self-aware of that, because this part is also missing in the conclusion.) We added the following summary: "In summary, while we failed to co-crystalize 14c with HSA, the combined findings from FP and NMR, drug inhibition, docking/MD, and aligned performance of Ala-mutants in FEP and NMR studies offer a useful guide for optimization of 14c and its use in delivery.For example, both the binding pose in site1, and NMR/FP studies suggest that both C-and N-termini of the 14c might be accessible as a plausible location for the attachments of payloads to 14c.We followed up on these suggestions in pharmacokinetic studies." We also expanded the computational studies and their comparison to experimental as summarized below: "To further evaluate prioritization of albumin site 1 as a plausible binding site for 14c, we performed NVT (constant temperature, constant volume) and NPT (constant temperature, constant pressure) simulations of 14c in site 1 of HSA solvated with explicit TIP3P water molecules and calculated binding free energy using steered molecule dynamics (SMD) and umbrella sampling techniques.We then calculated the potential of mean force (PMF) of the unbinding process in biased MD simulations with a harmonic potential whose interaction center is located at a specific distance between the binding pocket and the center of mass of 14c.The G was about -7.0 kcal/mol for pocket 1, while the G for the other pockets was less than -4.2 kcal/mol.The PMF corroborated that pocket 1 has significantly stronger binding energy than the others (Fig. 5A, Supplementary Fig. S26).Furthermore, the calculated G of free energy perturbation calculations (FEP) for 14c and five Ala mutants of 14c were in alignment with KD measured for the same alanine mutants in the 19 F NMR assay (Fig. 3B-C).The biggest loss of function for R4A mutant highlighted the importance of Arginine interaction for albumin binding (Fig. 5E-G); in contrast, both NMR and FEP agreed on a relatively minor role of N-terminal Ser and minor loss for S1A mutant."R3.7.Finally, the authors test their compound in vivo by evaluating serum stability of their top binder (5c) in mouse.The data shows increased half live circulation compared to the negative control.This experiment is a nice complementation of their in vitro data and totally reasonable for the stage of this project but is indeed, as stated, a minimalistic starting point.
We invite the reviewer to see new in vivo experiments summarized in Fig. 6 and answers to R1.1 and R1.2 R3.8.The authors use a sound methodology to proof their new concept.The right conclusions are drawn from the presented data.Nevertheless, I am not convinced that this study is suitable for publication in nature communications yet.I can´t see any superiority of the novel linker compared to established strategies.The need for replacement within the identification of a binder in combination with missing significance of the presented 19F-NMR data aren't convincing.In my opinion the computational data is meaningless in the presented context.
We are confident that new data added to this manuscript (Fig. 3, 5, and 6) address these concerns.R3.9.To make this an impactful report I would expect the authors to pick up one of the loose ends within the story: 1) The attenuated reactivity of the octafluoro-diphenylsulfone macrocycles for covalent linkage as advantageous feature 2) Getting profound knowledge about the binding of the identified motive to albumin by profound computational studies or crystal structures 3) Establishment of a 19F-NMR methodology with conclusive output Answers to R3. 6 and R2.11 expand on our docking MD simulations and add desired (profound) knowledge.For example, alanine mutants and KD results obtained by 19F NMR are wellcorrelated.These observations are further corroborated by in vivo pharmacokinetic studies (Fig. 6).