Frances R.
Smith
a,
Declan
Meehan
a,
Rhys C.
Griffiths
a,
Harriet J.
Knowles
a,
Peiyu
Zhang
b,
Huw E. L.
Williams
c,
Andrew J.
Wilson
bd and
Nicholas J.
Mitchell
*a
aSchool of Chemistry, University of Nottingham, University Park, Nottingham, NG7 2RD, UK. E-mail: nicholas.mitchell@nottingham.ac.uk
bSchool of Chemistry, University of Leeds, Woodhouse Lane, Leeds, LS2 9JT, UK
cBiodiscovery Institute, University of Nottingham, University Park, Nottingham, NG7 2RD, UK
dSchool of Chemistry, University of Birmingham, Edgbaston, Birmingham, B15 2TT, UK
First published on 14th May 2024
Synthetic methods that enable the macrocyclisation of peptides facilitate the development of effective therapeutic and diagnostic tools. Herein we report a peptide cyclisation strategy based on intramolecular interception of visible-light-mediated cysteine desulfurisation. This method allows cyclisation of unprotected peptides in an aqueous solution via the installation of a hydrocarbon linkage. We explore the limits of this chemistry using a range of model peptides of increasing length and complexity, including peptides of biological/therapeutic relevance. The method is applied to replace the native disulfide of the peptide hormone, oxytocin, with a proteolytically/redox-stable hydrocarbon, and internal macrocyclisation of an MCL-1-binding peptide.
Many of the synthetic methods developed to cyclise peptides5–8 take inspiration from nature and utilise the canonical residues, cysteine (Cys) and lysine (Lys), or the native N- and C-terminal functionality to form disulfide,5 thioether,9,10 and amide bonds (Fig. 1). Cyclisation via amide bond formation has been applied using both synthetic methods (peptide ligation11–13) and biological approaches exploiting intein chemistry (e.g., SICLOPPS14,15), mRNA display,10 and ligase enzymes.16 Due to the nucleophilicity of the thiol sidechain, numerous Cys-selective reactions (beyond disulfide and thioether formation) have been developed/repurposed for peptide cyclisation. These include the formation of bicycles17via alkylating scaffolds,18–21 bridging Cys residues using perfluoroaryl braces/fluorine displacement22,23 (and alternative bridging groups24,25), thiol coordination to bismuth,26 thiol-addition chemistry,27,28 and desulfurative replacement of a disulfide bridge.29 Incorporation of non-standard amino acids enables the exploitation of bioorthogonal chemistry such as azide–alkyne cycloaddition30,31 (i.e., ‘Click Chemistry’), Staudinger ligation32 and azide-phosphonite chemistry.33 Simple imine34 and oxime35 bond formation has been utilized, as well as more complex transition metal (TM)-catalysed36–41 and multi-component chemistry.42 Radical reactions such as thiol–ene43–45 and, more recently, decarboxylative photoredox catalysis46 and C–H alkylation47 have also been successfully applied. However, among the broad range of available synthetic techniques, Grubbs' ruthenium (Ru)-catalysed ring-closing olefin metathesis (RCM)48 reaction has found global application as a method-of-choice for peptide cyclisation due to effective formation of a proteolytic, hydrolytic, and redox-stable hydrocarbon linkage.49–53 Whilst an undeniably powerful method, RCM is usually conducted on protected peptides in organic solvent, and access to the saturated hydrocarbon necessitates reduction under harsh conditions. New, operationally simple techniques that retain the benefits of RCM, but that work effectively in more sustainable solvents, would offer an impactful alternative to this universally popular method.
Fig. 1 Commonly applied peptide macrocyclisation strategies; interception of desulfurisation as an approach to macrocyclisation of unprotected peptides. |
We recently reported a novel strategy for the site-selective modification of peptides and proteins via visible-light-mediated desulfurative C(sp3)–C(sp3) bond formation.54,55 Desulfurisation of Cys (and alternative non-proteinogenic thiol-containing amino acids56–58) can be applied post-peptide ligation as an elegant method to access a broad range of ligation junctions and facilitate chemical protein synthesis.56,59–67 A widely used free-radical-mediated Cys desulfurative protocol,59 developed by Danishefsky and co-workers, proceeds via a thiophosphoranyl radical species generated using a radical initiator (VA-044) to form a thiyl radical from the thiol sidechain of Cys in the presence of the water-soluble phosphine, tris(2-carboxyethyl)phosphine hydrochloride (TCEP). β-Scission of the thiophosphoranyl radical62 produces a peptide ‘alanyl’ radical which, in the presence of a suitable thiol additive (e.g. glutathione), will abstract an H-atom to yield the residue, alanine (Ala), at the ligation junction.59 In previous work, we demonstrated desulfurisation using an iridium(III) photocatalyst (PC) and employed alkenes to intercept the alanyl radical species, enabling installation of Lys sidechains carrying natural modifications as well as effective mimics of this modified sidechain.54,55 This reaction is initiated via excitation of the Ir(III) PC by a photon of visible light. The activated catalyst is then reduced by the thiol group producing a thiol radical cation which forms a thiyl radical on deprotonation. In the presence of TCEP, the thiophosphoranyl radical is formed; β-scission of this species produces the ‘alanyl’ radical60 which is trapped by the alkene. Due to the requirement to out-compete H-atom abstraction during this process, a large excess of the alkene is required (a minimum of 200 equivalents). By installing an appropriate alkene into a peptide containing a Cys residue, we postulated that intramolecular trapping of the radical produced upon desulfurisation may proceed preferentially to H-atom abstraction, essentially allowing us to use an equimolar equivalent of the alkene (Fig. S1†). The cyclic peptide radical produced during this reaction will be quenched to form the macrocyclic product; H-atom transfer (HAT) from the thiol group of remaining starting peptide is a likely pathway. The resulting thiyl radical can then continue the cycle or be reduced by the catalyst and protonated.68 Oxygen in the buffer has also been identified as an oxidant for the catalyst during Ru-mediated desulfurisation.60
The chemoselectivity of this chemistry should ensure that the reaction enables efficient cyclisation of unprotected peptides in aqueous solution. If realised, this technique would be a valuable addition to the toolkit available to researchers for the production of cyclic peptides.
Entry | Peptide | [Peptide]a/mM | [TCEP]/mM | Ir(III) PC (mol%) | Reaction durationb (min) | Isolated yieldc [% conversiond] |
---|---|---|---|---|---|---|
a Reaction conducted in 10% acetonitrile (MeCN)/6 M Gdn·HCl, 0.1 M Na2HPO4, pH 7.5–8.0. b LED strips, photochem set up 1 (ESI†). c Peptide products isolated by preparative HPLC. d % Conversion calculated via analytical HPLC. e Combined yield from both diastereomers. f Diastereomeric ratio calculated via analytical HPLC. g No reaction observed in the absence of blue light. h Reaction completed in 15 min using a PhotoRedOx box (photochem set up 2 – ESI†). | ||||||
1 | 3 | 0.5 | 5 | 5 | 10 | 77%e (dr 69:31)f |
2 | 4 | 0.5 | 5 | 5 | 60 | [72%] |
3 | 4 | 0.5 | 2.5 | 5 | 60 | [38%] |
4 | 4 | 0.5 | 5 | 1 | 60 | [36%] |
5 | 4 | 0.5 | 25 | 5 | 60 | 78% |
6g | 7a | 0.5 | 25 | 1 | 45 [15]h | 79% |
7 | 10 | 0.5 | 25 | 1 | 60 | 32% |
Quantitative conversion of the starting sequence to cyclised product, with negligible formation of the linear desulfurised by-product (formed via the alanyl radical abstracting an H-atom), is a gratifying initial result that allows high-yielding and rapid access to macrocyclic peptides. However, the formation of diastereomers is not ideal. Peptide 4 carries an allyl group as the radical trap; while this moiety would generate a less stable secondary radical upon addition of the alanyl radical, our previous results gave us confidence that intramolecular trapping should still out-compete H-atom abstraction. Under the conditions described, the starting peptide (4) was consumed within 60 min leading to the production of a major product (72% conversion to product by analytical HPLC, entry 2). It was noted that the reaction was equally effective without using degassed buffer; therefore, this step was omitted from the protocol. Prior to scaling the reaction up for isolation, a brief optimisation study was undertaken. It was observed that a reduction in the equivalents of TCEP led to a dramatic decrease in conversion to the product (Entry 3; 38% conversion to product 6), as did decreasing the mol% loading of the PC from 5 to 1 mol% (entry 4). When scaling up to an isolable yield it was observed that the conditions detailed in entry 2 were not optimal; an increase in the equivalents of TCEP to 50 (25 mM) was necessary to maintain high conversion to product. Using these adjusted conditions (entry 5) the desired product (6) was isolated in 78% yield and characterised via MS and NMR spectroscopy. The remaining mass balance for these reactions was the linear desulfurised by-product. No peptide degradation or epimerisation was observed over the course of the reaction.
Due to the need to synthesise the allyl-protected serine (Ser) building block, a more readily accessible option was sought. The commercially available amino acid, Fmoc-allyl-Gly-OH (alGly, alG), would afford a cyclic peptide with a butyl hydrocarbon linker. Therefore, Fmoc-protected alGly was incorporated into a model peptide (H-(alG)AFAC-NH2; 7a) and cyclised using the optimised conditions based on entry 5, Table 1 with a decrease in the PC loading to 1 mol% (entry 6), a change which which did not hinder conversion to the product. The reaction proceeded to completion as expected over 45 min and the desired product (8a) was isolated in an excellent yield of 79% by preparative HPLC (Fig. 2). The following conditions were therefore identified as optimal: 0.5 mM peptide, 1 mol% PC, 25 mM TCEP, in 10% MeCN/6 M Gdn·HCl, 0.1 M Na2HPO4, pH 7.5–8.0 (degassing step omitted). The reaction was then repeated using a PhotoRedOx Box equipped with a 34 mW cm−2, 450 nm LED (HeptatoChem). The ratio of desired product to by-product remained consistent with the reaction performed using blue LED light strips, however the rate of the reaction was enhanced, reaching completion in just 15 min (Fig. S49–S54†).
While 1H NMR analysis of 8a confirmed that the allylic protons were not present (indicating successful cyclisation) (Fig. 2), further analysis was sought to fully characterise the macrocycle for peptide 8a. A TOCSY NMR experiment was carried out on this model and a complete assignment of the macrocycle was achieved (Fig. 3). Proton environments in the hydrocarbon linker were found to couple to Hα signals on residues at each end of the macrocycle indicating successful cyclisation. Moreover, the Hα signal from what was initially the Cys residue prior to cyclisation coupled to the Hα of the first Ala residue, which could only occur as a result of macrocycle formation. A significant chemical shift dispersion suggests an ordered structure. Furthermore, the phenylalanine (Phe) Hα resonances show an NOE interaction to the Hα of the allyl glycine position, suggesting that the macrocycle is strained.
Fig. 3 Hα region for the TOCSY NMR spectrum of 8a confirming successful formation of the macrocycle. |
To further explore the optimised reaction, we considered buffer composition, the phosphine additive, and alternative desulfurisation conditions. Cyclisation of model 7a in PBS did not proceed cleanly (Fig. S55†), while employing HEPES buffer did lead to clean conversion of the starting peptide, however, the undesired desulfurised linear by-product was the dominant product (Fig. S56†). Thus, we conclude that cyclisation is most effective in a buffer containing a high concentration of chaotropic salt. Replacing TCEP with the water soluble phosphine, 3,3′,3′′-phosphanetriyltris(benzenesulfonic acid) trisodium salt (TPPTS) did not lead to effective conversion of the starting peptide 7a (Fig. S57†). Employing 1,3,5-triaza-7-phosphaadamantane (PTA) initially appeared to eliminate the production of the undesired linear desulfurised by-product (Fig. S58†). However, repetition of the reaction on an isolable scale and purification by preparative HPLC revealed the by-product and other impurities eluting under the broader product peak (Fig. S59–S62†) which ultimately limited the isolated yield. Finally, the peptide H-(alG)AFAC-OH (7b) was synthesised on 2-CTC resin to probe the cyclisation on peptides bearing a C-terminal carboxylic acid. While conversion to the desired product was observed (Fig. S68†), the ratio of by-product to macrocycle was less favourable when compared to C-terminal amide peptides. In addition, for this simple model, the retention times for the desired product and by-product were very similar making separation a significant challenge.
Several methods for the desulfurisation of Cys residues have been reported in the literature, these include; photo-induced desulfurisation69 using a ruthenium PC,60 photo-desulfurisation in flow,61 accelerated desulfurisation using tetraethylborate (NaBEt4),63 desulfurative borylation,64 and the exploitation of phosphite66 and phosphine-dependent pathways.65 We attempted to adapt and explore two appropriate examples.60,63 However, both the use of NaBEt463 and a ruthenium photocatalyst60 failed to improve on our reported conditions (Fig. S63 and S64–S67†).
To enable the synthesis of peptide macrocycles with longer hydrocarbon linkages, the amino acid pentenyl glycine (pGly, pG) was incorporated into the N-terminus of a simple model peptide (H-(pG)AFAC-NH2; 10) and this peptide subjected to the optimised cyclisation conditions. The desired product (11) was successfully formed but in low yield compared to the alGly example (Table 1 and Fig. 4). Furthermore, in addition to the use of non-proteinogenic amino acids carrying alkenes, we also explored the on-resin installation of similar radical traps. The N-terminus of H-AFAC-NH2 was functionalised on-resin using a solution of pentafluorophenyl acrylate46 (12) to afford the linear acrylamide peptide, 13. When subjected to the optimised cyclisation conditions the starting peptide was fully consumed within 60 min; LC-MS analysis of the crude reaction material suggested the formation of an interesting phosphonium salt by-product (Fig. S74†) but no cyclised material was observed.
Fig. 4 Linear starting peptides (10, 14–20) and the macrocyclic peptide products formed via desulfurative C–C bond formation (11, 21–27); macrocycle size indicated. |
A final 14-residue model peptide (49) was prepared to explore the preparation of larger (44-membered) macrocycles. Under the optimised cyclisation conditions the starting peptide was fully consumed to afford an inseparable mixture of the desired cyclised product and the desulfurised linear by-product. Increasing the loading of the PC to 5 mol% resulted in formation of the postulated phosphonium by-product observed for acrylamide model 13 (Fig. S144–S146†). This by-product (50) was isolated in 58% yield; 31P NMR analysis gave a single phosphorus resonance at 36.6 ppm. Our strategy is, therefore, highly effective for the formation of peptide macrocycles up to a 26-membered ring, and tolerates the full range of proteinogenic chemical functionality found across the proteome. Access to larger macrocycles may be possible, but will be dependent on successful separation of the desired product from the linear desulfurised by-product.
Finally, to explore internal peptide ‘stapling’ using this strategy, we selected a region of a BH3 protein – BID80–102, that binds MCL-1 to regulate apoptosis.53,72 This PPI is known to play a significant role in cancer development and progression.53,72 Three sequences (54–56) were synthesised with alGly and Cys positioned to afford either an internal i, i + 4 staple (57, 59; Fig. 5 and ESI†) or i, i + 7 staple (58; Fig. 5 and ESI†), fixing the length of either one or two full helical turns of the peptide, respectively. In addition, peptide macrocycle 26 is a head-to-tail macrocycle representing the i, i + 4 binding region of these longer sequences. The staple for all three peptides was successfully formed under the standard conditions, albeit in lower yield than the N- to C-terminal cyclisations previously explored (57–59). The alpha-helical structure of peptide 59 was assessed via CD spectroscopy. While macrocyclisation did increase the helicity by a few percent compared to the native BH3 sequence (22% compared to 19%), the starting peptide carrying the allyl glycine residue had a relatively high helical content (38%, Fig. S159†). This can be rationalised by considering the strain on the macrocycle imposed by the linker, and the fact that the native residues glutamine (Gln) and Ser were switched for alGly and Cys, respectively. These original residues have higher helical propensities relative to their replacements. Competitive inhibition studies (measured via fluorescence anisotropy) using MCL-1 and the fluorophore-labelled WT BH3 sequence demonstrated slightly lower inhibitory potency (26 ± 4 mM) for 59 compared to BID-wt (7.4 ± 0.9 mM) (Fig. S160†).
Footnote |
† Electronic supplementary information (ESI) available. See DOI: https://doi.org/10.1039/d3sc05865d |
This journal is © The Royal Society of Chemistry 2024 |