Post-translational insertion of boron in proteins to probe and modulate function

Boron is absent in proteins, yet is a micronutrient. It possesses unique bonding that could expand biological function including modes of Lewis acidity not available to typical elements of life. Here we show that post-translational Cβ–Bγ bond formation provides mild, direct, site-selective access to the minimally sized residue boronoalanine (Bal) in proteins. Precise anchoring of boron within complex biomolecular systems allows dative bond-mediated, site-dependent protein Lewis acid–base-pairing (LABP) by Bal. Dynamic protein-LABP creates tunable inter- and intramolecular ligand–host interactions, while reactive protein-LABP reveals reactively accessible sites through migratory boron-to-oxygen Cβ–Oγ covalent bond formation. These modes of dative bonding can also generate de novo function, such as control of thermo- and proteolytic stability in a target protein, or observation of transient structural features via chemical exchange. These results indicate that controlled insertion of boron facilitates stability modulation, structure determination, de novo binding activities and redox-responsive ‘mutation’.

B oron's electronic proximity to carbon has long drawn comparisons and highlighted their different roles in nature 1 . While both show more extensive speciation 2 compared to most other elements, boron's (semi)dynamic interactions find it differentially sequestered across the domains of life. Insertion of B(OH) n structures into nature-that is borylation ( Fig. 1 and Extended Data Fig. 1)-appears to generate endogenous functional effects not available through the traditional modes of biological bonding based on the organic chemistry of carbon, hydrogen, nitrogen and oxygen alone. Yet, despite this, it plays only a minor role in current 'native' biology.

boronates).
Notably, ligand binding can be accomplished even in competing Lewis-basic solvents such as water.
Despite its implicated use, the nonanchored (Extended Data Fig. 1a) character of natural borylation prevents explanation of its precise functional roles. In plants, for example, the essential nutritional role of boron drives postbiosynthetic borylation; in muro cis-diol engagement of apiosyl sugar residues by borylation (+'B(OH) 2 ') critically regulates cell-wall strength 4 . However, this nonanchored borylation cannot be predetermined (Extended Data Fig. 1a), resulting in an inability to control or exploit its effects.
Methods for introduction of boron-containing moieties into larger biomolecules remain limited. This is made challenging by the many Lewis bases in biomolecules that could sequester them (and hence inhibit installation). Two general methodologies for site-selective introduction of boronic acids into proteins have been reported (Extended Data Fig. 2a): biosynthetic incorporation via tyrosine mimicry 5 , or attachment of a prosthetic group 6 . Biosynthetic incorporation can be limited by reduced expression yields; boronic acids act as inhibitors of protein translation 7 . The alternative use of prosthetic groups necessitates larger, linker-based constructs 6,8 .
We reasoned that boron-carbon bond formation might allow the insertion of a minimal boryl moiety ('B(OH) 2 ' , Extended Data Fig. 1b) with precise control of its site ('anchored') and unique functions via direct programmable 'editing' into biomolecules. The alkylborono-aminoacid boronoalanine (Bal, Fig. 1a) represents a minimal borylated residue. It is a challenging amino acid to isolate 9 ; the boronyl sidechain readily engages its own alpha amino or carboxy moieties 9,10 . Such expected Lewis-acidic coordination hampers typical peptide assembly.
Direct late-stage Cβ-Bγ bond formation, ideally from an unprotected borylation source, might allow ready access to Bal in proteins via a tag-and-modify approach 11 from the flexible intermediate residue dehydroalanine (Dha) (Fig. 1b). Dehydroalanine residues (Dha) can now be readily introduced into proteins via a variety of chemical and biological methods and therefore provide a versatile 'tag' for site-selective protein modification 12 . Given the precedent for useful chemo-and regioselective hydroborylation reactions of terminal alkenes, including in Dha dipeptide models 13 , as well as procedures using aqueous solvent under Cu(II) catalysis with appropriate N-ligands on nonpeptidic systems 14,15 , we considered that the terminal olefin of Dha would allow distinct reactivity over other proteinogenic residues, thereby allowing selective carbon-boron bond formation in proteins under potentially benign conditions. Notably, during the final stage of this work elegant, proof-of-principle borylation of longer peptides, ribosomally

Bal allows intramolecular dynamic Lewis acid-base pairing.
Alkyl boronates show Lewis acidity toward hard Lewis bases in pairing (LABP). The extent is not only kinetically dependent 6 but can be externally modulated ('Wulff-type' 3 , see also ref. 19 ). Simple, unmodulated alkylboronic acids usually exhibit pK a higher than arylboronic acids (for example, pK a methylboronic acid 10.7 and pK a phenylboronic acid 8.8) rendering them typically 20 less suitable for binding applications in aqueous media 21 . Titration revealed that Bal displays higher acidity (pK a = 8.31, Supplementary Fig. 3) while retaining minimal size.
We probed the effect of installation of Bal at different protein sites, folds and environments (exposed versus enclosed) ( Fig. 1c and Supplementary Fig. 4). Intact protein mass spectra showed distinct peaks for three different borylated species (Fig. 2a and Supplementary Table 6). These corresponded to free Bal and also different ligated forms: mono-and di-substituted boronyl (Fig. 2a,b). These signatures of Bal's coordination-state (free, mono-, di-: Fig. 2a,b) therefore probed protein environment through protein-LABP (PLABP).
Analysis of putative ligands suggested ( Fig. 2b and Supplementary Table 6) protein-specific boronyl substitution that could be linked with the respective Lewis base environment. Substitution ranged from negligible (for example, in mCherry-Bal131, consistent with the solvent-exposed location of outer barrel site 131), to high amounts of mono-substituted (for example, in AcrA-Bal123, PanC-Bal44, in more structured enclosed regions, presenting internal ligands), to mainly di-substituted boronates (for example, in Histone H3-Bal10, consistent with a flexible N-terminal tail that can provide multiple internal ligands).

Bal generates de novo intermolecular binding function.
This engagement by Bal with internal Lewis bases also suggested potential in binding Lewis-basic moieties in intermolecular partners. Moreover, our observations suggested potential for Wulff-type 3 and/ or competing modulation of substitution. In this way, possible competition between intra-versus intermolecular engagement might provide a mode for higher selectivity than small-molecule boronic acid 'sensors' and reduced oligomerization 22 inside proteins.
We had already established compatibility with existing, endogenous intermolecular binding functions; borylation of PstS and Annexin V did not remove inherent phosphate binding (phosphate and phosphatidylserine, respectively). As an initial test of de novo ('host-guest') binding function, H3-Bal10 was surveyed with a range of possible guest ligands (Extended Data Fig. 7 and Supplementary Fig. 7). Wild-type (WT) Histone H3 has no integral function as a host receptor. We tested for intermolecular engagement with biologically derived poly-ols (terpenes, glycans).
Using suitably labeled ligands we observed (by 19 F-NMR) 'capture' by intermolecular PLABP (Fig. 3a,b) in H3-Bal10. This not only displayed single-site, saturation behavior as a host but also reversible and selective competition by alternative ligands (Fig. 3a).
We not only probed (Fig. 3b) 'guest'-ligand type (terpene-poly-ol versus sugar-poly-ol) but also site-dependency within the 'host' protein. Site variant H3-Bal9 showed consistently enhanced de novo host affinity over H3-Bal10 in its engagement with guest diols (down to low mM K D ). A clear selectivity for diol-type (>11-fold terpene 4 > sugar 5) was also observed (Fig. 3b). In all cases, corresponding nonborylated histone H3 variants showed no measurable affinity for intermolecular guest, confirming the critical role of Bal ( Supplementary Fig. 6). Moreover, attempted use of isolated small-molecule model Ac-Bal-NHBn bearing Bal as a host led only to ready oligomerization, precipitation and/or formation of complex mixtures not reflective of direct host-guest binding (Extended Data Fig. 7), confirming the critical role of placing Bal inside a suitable protein environment to exploit this de novo binding function.
This successful creation of de novo diol binding also suggested the potential for diol detection in more complex (for example, cellular) environments where varied diols (for example, cell-surface carbohydrates) are abundant. Representative glycosylated mammalian (CHO) cells were incubated with variants of the red-fluorescent protein mCherry (Fig. 3c). Flow cytometry revealed that while both nonborylated negative controls displayed no substantial cellular binding, site-selectively borylated mCherry-Bal131 showed clear cellular binding (>80% positive for mCherry-Bal131 versus <5% for nonborylated, Fig. 3d). In this way, mCherry, which has no inherent glycan-or cell-binding capacity, was bestowed with de novo cell-surface recognition. While detection systems based on fluorescent proteins as components in FRET-type sensors of small-molecule sugars have been developed 23 , this represents a de novo sugar-detecting fluorescent protein that uses direct binding.

Reactive PLABP enables footprinting to Ser/d 1 -Ser residues.
These successful uses of Bal in dynamic, reversible PLABP to convey binding function caused us to consider whether Bal could be exploited as both a dynamic and subsequently reactive Lewis acid motif (Figs. 1a and 4). Engagement with suitably reactive Lewis bases could allow reactive Lewis acid-base probe pairing.   reveals that, once Lewis acid Bal is inserted into different environments, it engages local Lewis bases to differing extents according to substitution state: un-, mono-or di-substituted, as judged by m/z difference of 0, −18 or −36 Da, respectively (right). In each case, possible Lewis-basic sidechains can be identified, for example, at Bal10 in H3-Bal10, Thr13 or Lys9 (left), but it should be emphasized that these data do not allow direct identification of such ligands. b, The site-dependent coordination states detected by protein LC-MS. n = 3, mean ± s.d. given; *n = 1. c, NMR investigations on small-molecule model Ac-Bal-NHBn show pH-dependent boronate formation (turquoise) as well as oxaborolane formation (rose) with the C-terminal amide C = O as one specific mode of PLABP. Fast exchange can be observed for boronate formation, while slow exchange is observed for oxaborolane formation (NaP i -buffered D 2 O).
The permanence of this reactive transformation would then create a 'footprint' of LABP.
Several kinds of reactive oxygen species (ROS) are reactive alpha-nucleophiles with the potential to allow the conversion of Bal to Ser via PLABP followed by migratory C-O bond formation (Fig. 4a,b). Not only would this 'record' interaction with ROS, it would allow synthetic access to Ser from Dha via Bal, thereby extending post-translational mutagenesis 24 by providing a method for β,γ-C-O formation (Fig. 4b).
Finally, given this demonstrated potential for intended oxidation of the Bal motif, we also probed the longer-term stability of Bal under background, potentially oxidative, conditions (Extended Data Fig. 4b,c). The isolated Bal residue proved stable in buffered solution under ambient conditions up to 100 h (Extended Data Fig. 4b); in a protein context, Bal oxidation in histone H3-Bal9 under ambient conditions in solution after 1 week was essentially comparable to that of background nonspecific oxidation (roughly 30%, Extended Data Fig. 4c).

PLABP modulates both thermo-and proteolytic stability.
Having demonstrated dynamic and reactive PLABP in single proteins, we explored application in more complex systems. Nonanchored borylation (Extended Data Fig. 1a) can play a stabilizing role in nature. This suggested the potential for design of protein stability, perhaps even in a chemoselective manner (for example, OH over SH engagement). We first probed the use of PLABP for altering thermostability (Fig. 5a). We chose three proteins with Bal placed at sites within different secondary-structure motifs with varying levels of intramolecular engagement with potential for PLABP-mediated stabilization: partially engaged in PstS-Bal197 (≥80% mono-substituted, start of alpha-helix-197-201, directed toward nearby alpha helices) 27 ; strongly engaged in Annexin V-Bal316 (≥40% di-substituted alpha-helix site, end of C-terminal-alpha-helix-20 that lies parallel and interacts (C 316 = O•••Arg285) with helices 18/19) 28 and multiply engaged in Npβ-Bal61 (≥20% of all three forms non-, mono-, di-substituted in pentapeptide-beta-strand motif with three strand-to-strand (C = O•••HN)-bonds) 17 .
Next, we tested PLABP for stability control in multi-protein cascades. Bal's modulation of thermolytic stability prompted us to explore proteolytic stability. We reasoned that borylation might selectively and locally inhibit (and hence confer resistance to) degradation by serine proteases while maintaining maturation susceptibility by other (for example, activating cysteine proteases)    Table 9). This, in turn, enables not only record of the interaction (footprinting) but also post-translational mutagenesis to Ser and its isotopolog d1-Ser through migratory Cβ-Bγ to Cβ-Oγ bond formation. enzyme types. SUMO1, a critical regulator of various cellular processes 30 , is converted from zymogen precursor pre-SUMO1 ( Fig. 5c) to matured SUMO1 before attachment to protein substrates. This SUMOylation cascade is mediated by protease SENP1 before SUMO-activating enzyme SAE1 and SUMO-conjugating enzyme Ubc9 transfer this matured SUMO to protein; all three use active-site Cys. We tested Bal in this full SUMOylation cascade (Fig. 5c,d). First, pre-SUMO1 was site-selectively borylated to create pre-SUMO1-Bal51. Next, in the presence of degradative serine protease (chymotrypsin), comprehensive digestion was observed for WT pre-SUMO1, showing poor proteolytic stability (Fig. 5d, right). However, insertion of Bal51 into the most susceptible central region (pre-SUMO1-Bal51) conferred greatly increased stability ( Fig. 5d, left). At the same time, full compatibility of Bal with cysteine protease and ligases required for the SUMOylation cascade not only allowed unperturbed maturation of pre-SUMO1-Bal51 (by Cys-protease SENP1) but also use of resulting matured SUMO1-Bal51 in subsequent successful SUMOylation (by SAE1/ Ubc9) of protein fragment RanGAP1 (Fig. 5e). In this way, Bal insertion into pre-SUMO1 allowed use of PLABP to convey selective proteolytic protection toward serine protease while allowing full zymogenic processing by cysteine-dependent enzymes. These observations are mechanistically consistent with small-molecule boronyl inhibitors of serine proteases 7 but represent examples now of de novo engineering of such benefits of chemoselective function into intact proteins to control stability in complex cascades.   Comparison of representative thermal denaturation melting curves for PstS-Cys197 (red) and PstS-Bal197 (green) shows range widening on borylation. c, De novo chemoselective function that exploits recognition of 'hard' Lewis bases (for example, active-site Ser) over 'soft' ones (for example, active-site Cys) can be introduced into proteins to convey stability toward Ser-mediated proteolysis while retaining Cys-mediated proteolytic maturation. d, Borylated pre-SUMO1-Bal51 is locally protected from degradative serine protease (Ser-based catalytic triad) in its most susceptible central core bearing site 51. e, electrophoretic analysis (performed in a qualitative single experiment) showing that pre-SUMO1-Bal51 can be successfully processed in a full SUMOylation cascade mediated by cysteine proteases and ligases, despite its engineered unnatural form.

Bal allows dative-contact induced chemical exchange (DICE).
Initial investigations had revealed that Bal can flexibly engage peptidic backbone via PLABP (above and Extended Data Fig. 2c). This suggested that Bal might allow observation of contacts in more complex systems via protein NMR through modulation of signal intensity by PLABP-mediated chemical exchange, specifically DICE. Paramagnetic relaxation enhancement (PRE) measures long-range contacts (for example, <15-24 Å using nitroxide labels) in disordered proteins 31 . There are, however, two drawbacks: lack of quantitative short-range information and incorporation of a spin label-prosthetic that can itself affect residual structure. Generalized, benign, 'short-and-long' contact measurement (for example, via DICE) would therefore prove valuable. Notably, a small-molecule model (Fig. 2c) revealed propensity of the Bal-B γ to associate with carbonyls transiently and reversibly, forming dative contacts needed for DICE under physiological conditions. In proteins, increased concentration of backbone-C = O could provide many such contacts. Moreover, since it is dynamic under these conditions, we reasoned that such C = O•••B γ engagement would not alter underlying structural preferences, thereby acting as a benign 'observer' (Fig. 6a).
To test this, we first characterized Bal as an NMR probe in proteins that form higher-order complexes. First, comparative spectra of denatured H3-Bal9 and H3-Ser9 revealed similar signal intensities for most residues. Next, using 13 CO-15 N correlation spectra, six resonances in H3-Bal9 were found to be entirely absent, with no discernable change in the chemical shift of the remaining residues. These observations indicated that transient C = O•••B γ contacts occur in a manner that induces intermediate exchange, thus generating loss of signal intensity. This was further confirmed by similar observations in N-H correlation spectra (Extended Data Fig. 9a); most notably, we observed that the signal loss increased markedly with temperature. Together these observations not only confirmed the benign nature of Bal as an 'observer' but also suggested that PLABP could be used to measure residual structure in disordered proteins by simultaneously examining local and remote contacts by using signal loss that correlates with the probability of contact between a backbone site and Bal.
Having established initial viability of Bal as a structural probe, we next tested it in the sampling of transient interactions. These are of special interest in the characterization of intrinsically disordered regions (IDRs) in proteins, which can naturally adopt an ensemble of different states 32 . Typically, PRE is used to detect only long-range contacts in such systems with the drawbacks noted above of label-induced artifacts and local short-range signal broadening ('bleaching') 32 . However, by contrast, use of DICE revealed region-selective loss of signal intensity in the manner anticipated (Extended Data Fig. 9) that probed not only transient long-distance contacts (permitting conclusions on residual structure), but also, and in contrast to PRE, DICE allowed simultaneous detection of short-range contacts.
We chose next the nucleosome as a more complex model system to test DICE using Bal as an NMR 'observer' of IDRs. Each nucleosome consists of a histone octamer (two each of H2A, H2B, H3, H4) wrapped by DNA 33 . Histones display disordered tails at their termini, which are subjected to a wide range of post-translational modifications (PTMs) suggested to regulate a multitude of cellular functions 33 . Of these, histone H3 phosphorylation at Ser10 by Aurora B kinase has been shown to cause two seemingly contradictory effects: chromatin condensation in mitosis and chromatin relaxation (and so gene expression) in interphase [34][35][36] . Moreover, such H3-Ser10 phosphorylation also seems to play a crucial role in switching mediated by the key heterochromatin protein 1 (HP1): HP1 is recruited by trimethylation of Lys9 in H3 whereas phosphorylation of Ser10 in H3 leads to its ejection 37 . While the central, structured part of the nucleosome has been described with high resolution 38 , detailed structural descriptions of the intrinsically disordered histone tails and so knowledge of the influence of PTMs (such as phosphorylation of Ser10 of H3) on structure is essentially lacking 33,39,40 .
We constructed a nucleosome containing 15 N-labeled, borylated, histone H3 (Fig. 6a). Bal was placed strategically next to phosphorylation site Ser10, replacing Lys9 (to create 15 N-H3-Bal9). Since Lys9 is implicated in Ser10-phosphorylation-dependent HP1 recruitment, we reasoned that this site would allow suitable direct observation of phosphorylation-induced structural change. In agreement with the established intermediate exchange regime, intensity comparison between the 1 H-15 N-HSQC spectra obtained for the fully assembled Bal-containing nucleosome (in comparison with its WT variant) immediately revealed a unique, residue-specific intensity modulation 'double-dip' pattern ( Fig. 6b) corresponding to the presence of a residual structure consisting of two alpha helices (at H3 residues 3-10 and 17-28) connected by a beta-turn. Consistent with induction of tail structure within the intact nucleosome, this also represented a marked increase when compared to isolated, denatured H3 histone (Extended Data Fig. 9c). These observations were further supported by secondary chemical shift (SCS) analyses 41 (Supplementary Fig. 9) where, in WT nucleosome, we saw strong helical preferences in the same regions identified by Bal as an observer probe.
Next, having used DICE to observe this transient structural motif, we tested the effect of phosphorylation. Notably, on treatment with Aurora B kinase, transient structure was released in favor of increased tail flexibility, leading to a reduction in PLABP (as observed by reduced Bal-induced intensity modulation via DICE, Fig. 6b); again, this was further confirmed by SCS analysis. The ratio of signal intensities ± phosphorylation did not alone reveal the patterns in structural rearrangements from the changes in signal intensities. Moreover, the pattern of signal intensity change was essentially the same for corresponding samples bearing ±Bal as a probe. Together, these combined ±Bal and ±phosphorylation comparisons confirmed that, under these conditions, the presence of Bal caused no perturbation to the relevant underlying structural equilibria. Simple comparison of intensity patterns of spectra obtained ±Bal allowed direct read out of the regions that are making contact via loss in signal intensity using the DICE method. In this way, Bal9 at the end of the first observed helix not only sampled long-range contacts, but also concurrently provided structural information in its direct vicinity (Extended Data Fig. 9e) leading to a mechanistic model where phosphorylation of H3 in the nucleosome opens up transient structure in the tail of H3 with seemingly diverse functional consequences (Fig. 6).
In all cases, control experiments ( Fig. 6a and Supplementary  Fig. 10) revealed no differences in either induced structures or phosphorylation activity by kinase between an intact WT nucleosome assembly and the H3-Bal9 variant (Fig. 6b), further confirming the benign nature of Bal as a structural probe. Together this highlighted the unique use of Bal as a protein NMR probe of residual structure using the DICE method that exploits PLABP-mediated dynamic, dative sampling of its spatial environment.

Discussion
Despite being an element essential to many organisms, only a few boron-containing natural products have been described. In all of these, function is endowed by boron as its borate (-O-B-X) rather than boronate (-C-B-X) form. Examples include macrolides boromycin and borophycin 42 , bacterial signaling molecule AI-2 (ref. 43 ) and rhamnogalacturonan II (RG-II) 44 . Our results now indicate that endowing biomolecules with the potential for exploiting the C-B bond ('anchoring' of boron) can lead to wider application of the properties of boron as an element in biology.
Site-selective C β -B γ bond formation in proteins now generates a minimal borylated residue, Bal, which displays diverse functional traits based on dative bonding. It should be noted that our analyses of diastereoselectivity indicate that in this method Bal (and/or product Ser) are formed as a mixture of l/d-epimers in diastereometric ratios of <2/1 to 1/1 (Extended Data Fig. 10 and Methods). This highlights that not only do the functions we observe arise from site epimers (either together or alone), but also, given the low observed substrate control in diastereometric ratio, highlights the future potential for ligand/catalyst control in stereoselectivity.
Bal now confers on proteins a de novo Lewis-acidic binding function that allows complementary dynamic and even reactive Lewis acid-base pairing with striking selectivity and high biocompatibility toward both intra-and intermolecular partners. Use of appropriately reactive Lewis bases allows migratory conversion of the C β -B γ bond to C β -O γ bond formation, in turn giving site-selective chemical, post-translational access to Ser and isotopolog d1-Ser. Such reactivity complements approaches in other systems where, for example, peroxynitrite-reactive chromophores containing p-boronophenylalanine in fluorescent proteins have been used to generate elegant sensor systems 45 .
Boronic acids are well-known, aqueous-compatible binding motifs that can exploit chelation-enhanced selectivity, such as for poly-ols over mono-ols, allowing recognition even in water 46 Fig. 6 | exploitation of Bal in PLABP using the DICe method as a direct probe of IDrs found during epigenetic modification in nucleosomes. a, Assembly of nucleosomes containing histone H3-Bal9 allowed for exploitation of dynamic PLABP for the probing of structure via NMR using DICe. b, Nucleosome phosphorylation leads to a structural change in the tail of histone H3 that can be observed as a change in Bal-mediated backbone binding. Bal acts as a direct NMR 'observer' sensitive to structural change. Comparison of the NMR intensities of borylated and WT nucleosomes uniquely reveals both short-and long-range structural information. Comparison of intensity modulation after phosphorylation of both Bal-containing nucleosomes and WT nucleosomes reveals no notable differences, confirming Bal's benign nature as an 'observer'. Values represent the mean ± s.d. from n = 3 measurements of the same sample.
Equivalent simple, small-molecule boronate hosts can occupy complex equilibria that prevent precise 'programmed' use. Yet the use of Bal in proteins seemingly allows more discrete control, free of oligomerization, precipitation and other confounding effects. Monodentate 48 as well as more complex bi-and tridentate interactions (O,N and O,O,N) of small-molecule boronic acid-based enzyme inhibitors with Lewis-basic sidechains have been previously described 49 , but are solely based on boronates as external unanchored ligands making use of a predefined protein environment. Future applications of multi-faceted dual-and multi-receptor systems can now be envisaged. For example, both Annexin V-Bal316 and PstS-Bal197 remain active host receptors of their respective natural ligands (phosphatidyl-Ser and phosphate, respectively) while at the same time carrying a Bal residue for putative additional binding. While cooperativity between natural and de novo (Bal) binding was neither designed nor tested here, our observations of independent function suggest promise for de novo 'dual-mode binding' . Moreover, given the precedent in small-molecule boronates 50 , designed cooperativity between functional groups in multiple unnatural amino acids, for example Bal and formyl-glycine, can also now be envisaged.
Given the advent of boron-based small-molecule drugs (including boron-containing peptides), such site-selective engineering of Bal into proteins could create unique proteins carrying engineered binding sites, tailored responses and selective stability for altered biological function as biotherapeutics 51,52 . Indeed, our creation here of proteins with tailored, selective responses toward thermolysis and proteolysis that still retain their biological functions and compatibility suggests many avenues.
The successful application of Bal in multi-protein systems (SUMOylation cascades or nucleosome phosphorylation) with full retention of natural biological use highlights the residue's compatibility. For example, installation of Bal9 in place of Lys9 in histone H3 of a nucleosome did not impede native enzymatic phosphorylation ( Supplementary Fig. 8) and also permitted detection of H3-pSer10 product using highly specific antibodies ( Supplementary Fig. 11).
DICE-NMR via PLABP allowed detection of a transient structural consequences of Aurora B kinase-mediated phosphorylation on the histone H3 tail. This in turn suggests a more complex structural mechanism driving observed HP1 ejection than had previously been supposed 37 (that is, more than simple occlusion of HP1 from methylated Lys9 by proximal pSer10).
Although the detailed mechanism of DICE-NMR is beyond the scope of the present study, drawing on our current data, Bal-B γ is selectively associating and disassociating with carbonyl oxygen atoms; in the context of many protein C=O bonds, this is N-site chemical exchange. During association, chemical shifts of adjacent atoms will be altered (δω), which modulates signal intensities in a manner that depends on both δω and overall exchange rate (k ex ). In protein-DICE-NMR the effect should be most clear through experiments involving an amide nitrogen or carbonyl oxygen; electron density around the carbonyl carbon is largely unaffected. The effect will be discerned as loss of signal (Extended Data Fig. 6b) and then on backbone HN, where signal intensities are modulated (Fig. 6b). A small population of the boron-associated state (<1%) can, in principle, lead to the substantial intensity changes observed here. Temperature dependence suggests fast-intermediate exchange with k ex /δω > 1. Overall, every atom transiently modulated by Bal will lead to reduction in signal intensity that reflects chemical accessibility of the site and so, in turn, reveals residual structure in disordered chains. In principle, such Bal associations might alter structural preferences. However, we do not see any substantial change in the chemical shift of observed residues and see similar structural arrangements using both DICE and PREs; there is therefore no indication, in this case, that underlying structural preferences are altered.
Finally, it should also be noted that the chemical roles and reactivities of organoboranes are wide in scope 46 and so Bal in proteins may now enable further, more diverse applications. For example, while we have demonstrated de novo binding function (dynamic PLABP) toward naturally derived ligands and stoichiometric reactivity (for example, Bal → Ser through reactive PLABP) we can envisage future use of host-guest interactions with nonnatural ligands and even as activatable or de novo catalytic residues 53 .

Influence of concentration.
A stock solution of Ac-Bal-NHBn (30 mg ml −1 ) in NaP i buffer (20 mM, pH 8.5, D 2 O) was prepared and the pH was adjusted to pH 8.5 using 1 M NaOH in D 2 O. A two-fold dilution series was prepared by dilution of the stock solution into the same volume of NaP i buffer (20 mM, pH 8.5, D 2 O). The pH of the resulting solutions was checked before analysis via NMR spectroscopy (Supplementary Table 12). General protocol for protein borylation. Stock solutions of copper(II) sulfate, 4-picoline and tetrahydroxydiboron were prepared in milliQ water directly before the reaction was conducted.
To a solution of dehydroalanine-mutant protein (roughly 1 mg ml −1 ) in borylation buffer (100 mM NaP i , 3 M Gdn•HCl, pH 7.0) were added aliquots of the previously prepared stock solutions of (copper(II) sulfate, 5-60 equiv.; 4-picoline, 12.5-150 equiv.; tetrahydroxydiboron, 50-600 equiv.). The mixture was briefly vortexed and incubated at room temperature for 10 min. Protein samples can be purified by gel filtration or dialysis. Conversions were determined by liquid chromatography-MS (LC-MS) and protein recovery was quantified by absorbance at 280 nm (A 280 ) spectrophotometry (see Supplementary Note 3 for details).

Sample analysis. Samples were analyzed on a Perkin Elmer NexION
2000B ICP-MS. Calibration was conducted using dilutions from certified reference material (River Water, SLRS-6). All samples were spiked with 1 ppb Rh and the internal standard was used to adjust for instrument drift. All samples were measured in duplicate.
Intact protein mass spectrometry. Intact protein LC-MS was performed on a Waters Xevo G2-S quadrupole time of flight (QTOF) spectrometer equipped with a Waters Acquity ultrahigh-performance liquid chromatography (UPLC), on a Waters Xevo G2-XS QTOF spectrometer equipped with a Waters Acquity UPLC or an AB Sciex TripleTOF 6600 spectrometer equipped with a Shimadzu high-performance liquid chromatography (HPLC) system. Separation was achieved using a Thermo Scientific ProSwift RP-2H monolithic column (4.6 × 50 mm) using water + 0.1% formic acid (Solvent A) and acetonitrile + 0.1% formic acid (Solvent B) as mobile phase at a flow rate of 0.3 ml min −1 and running a 10 min linear gradient as follows: 5% Solvent B for 1 min, 5 to 95% Solvent B over 6 min, 95 to 5% Solvent B over 1 min, 5% Solvent B for 2 min. Spectra were deconvoluted either using MassLynx (Waters) and the 'MaxEnt1' deconvolution algorithm or Analyst (AB Sciex) using the 'Reconstruct Protein' algorithm. Conversions were calculated from peak intensities (Waters) or peak areas (AB Sciex) dividing the value for the product by the sum of the values for products and (residual) starting material. Impurities present before the reaction were not considered. For details, see Extended Data Fig. 5 and Supplementary Note 3.

Protein MS/MS. An aliquot of the protein sample was denatured by addition of 8 M urea solution to a final concentration of 4 M.
TCEP reducing agent was added to a final concentration of 10 mm. The sample was diluted fourfold with 50 mm TEAB buffer solution. Digestion buffer (50 mm TEAB, 50 mm TCEP, 2 mm EDTA) was added (one tenth of sample volume). ArgC was added in a ratio 1:20 and the sample was incubated at 37 °C overnight. The sample was analyzed using a standard MS/MS procedure.
LC-MS/MS data were analyzed in PEAKS Studio v.8.5 or PEAKS Studio X (Bioinformatics Solutions Inc.). The data were searched against the known protein sequence. The following settings were used: precursor mass tolerance at 15 ppm; fragment mass tolerance at 0.5 Da; maximum number of missed cleavages, 3 and nonspecific cleavage at one end of the peptide. The following variable PTM were selected in addition to the modification of interest: oxidation (+15.99 Da, Met), deamidation (+0.98 Da, Asn, Gln) and carbamylation (+43.01 Da, Lys). An FDR or 1% on peptide level was applied.
Histone H3-Bal9 stability. Lyophilized Histone H3-Bal9 was dissolved in denaturing buffer (NaPi 100 mM, 3 M Gdn•HCl, pH 7.0) at 1 mg ml −1 . The solution was incubated under ambient atmosphere at room temperature in a microcentrifuge tube for 1 week. LC-MS analysis revealed the presence of Histone H3-Ser9 (33%) as a result of Bal oxidation. In agreement with ambient oxidation, further oxidized species were observed (M + 16 Da) (Extended Data Fig. 4).

Determination of the enantiomeric ratio of Ac-Bal-NHBn.
Ac-Bal-NHBn was dissolved in binding buffer (40 mM NaP i , 5 M urea, pH 7.0, 10% D 2 O) at a final concentration of 500 µM. Then 10 equiv. of chiral shift reagent were added and the sample was vortexed. The sample was transferred to a NMR tube and analyzed on a Bruker AVIII HD 500 MHz spectrometer equipped with BBFO SMART probe (2,060 scans, d1 = 2 s). A ratio of approximately one-to-one was detected after integration of the peaks (Supplementary Fig. 26).

Determination of the diastereometric ratio on Histone H3-Bal9.
Histone H3-Bal9 was dissolved in binding buffer (40 mM NaP i , 5 M urea, pH 7.0, 10% D 2 O) at a final concentration of 144 µM. Then 10 equiv. of chiral shift reagent were added and the sample was vortexed. The sample was transferred to a NMR tube and analyzed on a Bruker AVIII 600 MHz spectrometer equipped with a Prodigy N2 broadband cryoprobe (3,500 scans, d1 = 2 s). A ratio of approximately one-to-one was detected after integration of the peaks (Extended Data Fig. 10).

Determination of stereoselectivity using proteolytic LC-MS.
Cloning, expression and purification of Histone H3 TEV -Cys2 was done as follows. The plasmid encoding the sequence for expressing the WT protein Xenopus laevis Histone H3.3, lacking cysteines was used as a PCR template. A tobacco etch virus (TEV) protease consensus sequence (ENLYFQG) was added between the second and third residues of the Histone sequence (AR(TEV)TKQ…), and the second residue was mutated to Cys (R to C) by including the desired mutations in the forward primer, as well as necessary restriction enzyme sites. The primers used for PCR were: After amplification, the PCR product was digested with BamHI and NcoI along with a pET3d vector, ligated into the vector and transformed into XL-10-Gold cells (Agilent). The plasmid and resulting protein sequence were confirmed via Sanger Sequencing, and the H3-Cys2, N-TEV Histone was expressed and purified as described for the other histones.
Dha formation, borylation and oxidation. Dha formation, borylation and oxidation were conducted as described for Histone H3 Cys9 and Histone H3 Cys10. LC-MS analysis. Samples (TEV digest or reference peptides) were analyzed on a Waters Xevo G2-XS QTOF mass spectrometer equipped with a Water Acquity UPLC. Separation was achieved on a ACQUITY UPLC BEH C18 column (Waters) using water + 0.1% formic acid (Solvent A) and MeCN + 0.1% formic acid (Solvent B) as mobile phase at a flow rate of 0.2 ml min −1 and using a 10 min linear gradient as follows: 5% Solvent B for 1 min, 5 to 25% over 9 min. The recorded data were processed in MassLynx 4.1 (Waters) by generating extracted-ion chromatograms (971 ± 1 Da) followed by integration of the corresponding peaks. Comparison with authentic peptide references allowed for assignment of the l-Ser and d-Ser epimers.

Reactive PLABP for the synthesis of Ser and d1-Ser mutants.
Nondeuterated. To a solution of purified borylated protein in borylation buffer (NaP i 100 mM, pH 7.0, 3 M Gdn•HCl) was added H 2 O 2 to a final concentration of 5 to 20 mM and the mixture was incubated at room temperature for 10 min. The concentration of H 2 O 2 required depends on the specific protein.
Deuterated. Dehydroalanine-bearing protein in borylation buffer (100 mM NaP i , 3 M Gdn•HCl, pH 7.0) was lyophilized and redissolved in D 2 O. This procedure was repeated twice. The lyophilized protein was then reconstituted with D 2 O and borylated according to the general protocol to yield a d1-Bal mutant. Oxidation to the deuterated serine residue was performed as described above.
Calculation of residue accessibility. The solvent-accessible surface area was calculated using FreeSASA 54 . The Lee and Richards algorithm 55 was used on Protein Data Bank (PDB) structures having the site of interest mutated to Cys. The number of slices was set to 100, atom radii from NACCESS were used and different probe sizes (1.00, 1.40 and 2.80 Å) were used. The absolute values obtained for the solvent-accessible surface area of the respective Cys residues were set in relation to a tripeptide Gly-Cys-Gly using the value published by Tien et al. 56 (Supplementary Fig. 4). 1 H-NMR. Small-molecule model compound Ac-Bal-NHBn (2) (see Supplementary Note for synthesis and purification) was dissolved in buffer (20 mM NaP i in D 2 O) at a concentration of 20 mg ml −1 . The pH was adjusted with concentrated DCl or 1 M NaOH in D 2 O to obtain seven samples having a pH ranging from 5.4 to 12.0. Standard 1 H-NMR experiments were conducted on each sample and species were quantified by integration of the C β -H signals, which are distinctly different for the boronic acid and the oxaborolane. The data were processed with GraphPad Prism v.8.0.0 (GraphPad Software Inc.) and fitting was performed using a sigmoidal standard curve.

Determination of poly-ol binding to Bal proteins by NMR.
Borylated protein was dissolved in NMR binding buffer (NaP i 40 mM, pH 7.0, 5 M urea, 10% D 2 O). Fluorinated poly-ol (see Extended Data Fig. 7a for poly-ols used and Supplementary Note 1 for their synthesis) was added from a stock solution in dimethylsulfoxide (DMSO) (typically 10 equiv.), the mixture was transferred to an NMR tube and analyzed via 19 F-NMR on a Bruker Avance III HD 600 MHz NMR spectrometer equipped with a Prodigy N 2 broadband cryoprobe. Peaks for bound poly-ols appeared downfield of the nonbound species. Binding was quantified by peak integration for a known protein, and fluorinated poly-ol concentrations and experiments were conducted in triplicate ( Supplementary Fig. 7).

Determination of protein melting curves.
Melting curves were recorded on a Prometheus NT.48 differential scanning fluorimeter (NanoTemper Technologies). Samples were loaded into high sensitivity quartz capillaries. Unfolding was detected during heating using a linear temperature gradient (20 to 95 °C, 1 °C min −1 ) and an excitation power of 50% (100% for Npβ). Melting curves were recorded at three different concentrations in phosphate buffer (20 mM NaP i , 50 mM NaF, pH 7.4). The data were analyzed using PR.Stability Analysis v.1.0.2 software (NanoTemper Technologies) and GraphPad Prism v.8.0.0 (GraphPad Software Inc.).

Stability in SUMOylation cascade.
Pre-SUMO1 maturation. Here, 200 µl (0.67 mg ml −1 ) of a solution in TRIS buffer (100 mM TRIS base, pH 7.0) of pre-SUMO1-Bal51 were incubated with 10 µl of SENP1 (0.14 mg ml −1 ) at 37 °C for 4 h. Then 20 µl of fresh Ni-NTA resin was washed with water (2 × 500 µl) and added to the reaction mixture. The mixture was incubated for 15 min and the resin was removed by filtration. The filtrate was concentrated to 50 µl using a VivaSpin concentrator (molecular weight cutoff (MWCO) 5 kDa) and used directly for in vitro SUMOylation.

SUMOylation of a model protein.
In vitro SUMOylation of RanGAP1 fragment was conducted using an in vitro SUMOylation assay kit (Abcam) on a 20 µl scale using matured SUMO1-Bal51 and following the manufacturer's general protocol. A control experiment was conducted in the absence of matured SUMO1-Bal51. RanGAP1 SUMOylated with SUMO1-Bal51 was detected via western blot using the supplied primary antibody and a polyclonal goat antirabbit IgG alkaline phosphatase secondary antibody. Controls and reaction were analyzed via SDS-PAGE (10% bis-TRIS gel, 150 V, 45 min, MES buffer). The proteins were transferred to a nitrocellulose membrane. The membrane was blocked for 1 h in TBS-T (pH 7.5) supplemented with 5% BSA. The membrane was incubated with the primary rabbit anti-SUMO1 polyclonal antibody (dilution 1:1,000) for 1 h at room temperature and washed three times for 5 min before the secondary goat antirabbit polyclonal antibody alkaline phosphatase conjugate was added (dilution 1:1,000). The membrane was incubated for another hour, washed three times and SUMO1 was visualized using NBT/BCIP substrate solution (ThermoFisher) (Supplementary Figs. 29 and 30).

Annexin V binding. FITC-labeling of Annexin V-Bal316.
A solution of Annexin V-Bal316 in phosphate buffer (500 μl, 0.15 mg ml −1 , 2.09 nmol, 1.00 equiv.) was exchanged into sodium bicarbonate buffer (100 mM, pH 9.4) and concentrated to 200 μl using VivaSpin 500 concentrators (5 kDa MWCO, Satorius). Then 20 μl (81.6 μg, 209 nmol, 100 equiv.) of a freshly prepared solution of fluorescein 5(6)-isothiocyanate (FITC) (4.1 mg) in DMSO (1,000 μl) was added to the protein solution and the mixture was vortexed briefly. The mixture was shaken (400 r.p.m.) in the dark at 37 °C for 2 h and purified using a PD Minitrap G-25 size exclusion column (GE Healthcare) previously equilibrated with HEPES buffer (10 mM HEPES, 140 mM NaCl, pH 7.4) according to the manufacturer's instructions. The purified solution of FITC-labeled Annexin V-Bal316 was concentrated using VivaSpin 500 concentrators (5 kDa MWCO, Satorius) and stored on ice until used. The dye-to-protein ratio was determined spectrophotometrically on a NanoDrop 8000 spectrophotometer (Thermo Scientific) measuring at λ = 280 nm and λ = 494 nm. A dye-to-protein ratio of 0.95 was detected. Flow cytometry was performed on a BD FACSCalibur cell analyzer using BD FACSDiva software. A minimum of 10,000 cells per sample was analyzed using the 488 nm laser and a 530/30 nm bandpass filter.

mCherry binding. Determination of absorption maxima.
Absorption spectra of different mCherry mutants were measured on a BMG Labtech SPECTROstar Nano (full spectrum, 350-900 nm, 1 nm interval) using 200 µl of protein solution in a clear 96-well plate at a concentration of approximately 10 µM. Measured optical density (OD) values were used to calculate the relative absorbance of the respective sample. The absorbance maximum was determined to be at 586 nm for all three mutants-mCherry-Cys131, mCherry-Sulfonium131 and mCherry-Bal131.
Flow cytometry. To 100 µl of CHO-WT cells (roughly 10 6 cells) in fluorescence activated cell sorting (FACS) buffer (Dulbecco's phosphate-buffered saline, pH 8.0, 2% FBS) were added 300 µl of mCherry mutant (0.28 mg ml −1 , 10 µM) or 300 µl of FACS buffer (control). The cells were shaken on ice for 20 min at 300 r.p.m. The samples were centrifuged at 400g for 3 min at 4 °C. The liquid was removed and the cell pellet was resuspended in 1,000 µl of FACS buffer, centrifuged at 400g for 3 min at 4 °C and the buffer was removed. The washing step was repeated once before the cell pellet was suspended in 400 µl of FACS buffer. Flow cytometry was performed on a BD LSRFortessa X-20 cell analyzer using BD FACSDiva 8.0 software. A minimum of 10,000 cells per sample was analyzed using the 561 nm laser and a 610/20 nm bandpass filter. The 640 nm laser in combination with a 780/60 nm bandpass filter was used as a reference channel. The data were analyzed using FlowJo v.10 software.

PstS competition assay. A stock solution of Phosphate Sensor
Protein CON-NMR. Lyophilized samples were dissolved in 330 µl of NMR buffer (50 mM NaP i , 3 M Gdn•HCl, 5% D 2 O, pH 7.0). NMR experiments on the triple labeled denatured histone H3 samples were performed on a 14.1T Varian Inova spectrometer equipped with a 5 mm z axis gradient triple resonance room temperature probe. Three-dimensional HNCO experiments were recorded using a random sparse sampling schedule that contained 1 H/ 13 C/ 15 N 1,600/50/30 complex points from a maximum of 1,600/150/50 points. The spectral widths were 7,993/849/1,600 Hz, with eight scans recorded per free induction decay (FID) and an interscan delay of 0.6 s for a total acquisition time of 12 h and 24 min. The spectra were processed in NMRpipe 7 with the SMILE reconstruction algorithm using linear prediction, a sine-bell window function, zero filling and phase correction (Extended Data Fig. 9).
Nucleosome NMR study. Nucleosome assembly. Octamer reconstitution, nucleosome assembly and 145 bp DNA products were all performed essentially as previously described 57 using borylated Histone H3 where applicable.
Cloning and mutagenesis. The H3-SUMO fusion was created by overlap PCR using primers given in the table below. The fusion protein consisted of an N-terminal His-tagged SUMO for solubility followed by a 'Tobacco Etch Virus nuclear inclusion A endopeptidase' (TEV) recognition site and a C-terminal H3 tail (residues 1-44). TEV protease can cleave a variant of its optimal recognition sequence (ENLYFQ\S) in which the C-terminal serine is replaced with alanine (ENLYFQ\A). Digestion of the construct resulted in a WT tail with an N-terminal alanine without any additional amino acids. The WT Xenopus laevis histone plasmids were a kind gift from R. Klose (Oxford, Department of Biochemistry). The '601' 145 bp was a kind gift from J. Min.
Primers used for mutagenesis and cloning:  H3-tail peptide. Preparation. The production of the labeled H3-tail peptide was performed as described by Lundstrom et al. 58 with minor modifications. The plasmid encoding the H3(1-45)-SUMO construct were transformed into BL21(DE3)pLysS chemically competent Escherichia coli and grown on LB-agar plates containing ampicillin (100 mg l −1 ) and chloramphenicol (34 mg l −1 ). The next evening, 2-15 ml of LB medium containing the same antibiotics was inoculated with one colony from the plate. These were grown overnight and then centrifuged, and the cells were resuspended in 100 ml of M9 medium (containing 15 61 . With 160 scans in the direct dimension the total acquisition time was 18 h. 15 N R 2 measurements. 15 N R 2 rate measurements were performed using the HSQCT2ETF3GPSI standard Bruker pulse sequence. An interscan delay of 5 s was used. With 64 scans in the direct dimension the total acquisition time for four planes was 24 h. The set delay period of 16.96 ms was repeated 1, 5, 10 and 15 times. The separate two-dimensional (2D) planes with different delays were combined and processed using NMRPipe 59 and peak fitting was performed with FuDa (https://www.ucl.ac.uk/hansen-lab/fuda/).

PRE.
For the attachment of a spin label, the histone H3 cysteine mutant (H3-Cys36) containing nucleosome was first dialyzed into no-salt buffer containing no DTT for 4 h. Residual DTT was removed by VivaSpin. Then 10 equiv. (1-oxyl-2,2,5,5-tetramethyl-Δ3-pyrroline-3-methyl) methanethiosulfonate (Toronto Research Chemicals O875000; dissolved in MeCN) was added and left to incubate for 1 h at room temperature and then overnight at 4 °C. Subsequently, excess methanethiosulfonate was removed by Vivaspin. After R 2 rates were recorded with the spin label, the spin label was reduced by the addition of 10 mM sodium ascorbate and incubation at 4 °C overnight. PRE rates were calculated from 1 H R 2 rates using: To calculate errors, these spectra were recorded three times for the following samples: WT nucleosome, phosphorylated WT nucleosome, Histone H3-Bal9-containing nucleosome, denatured Histone H3-Ser10 (WT) and denatured Histone H3-Bal9. One measurement was made of the phosphorylated Histone H3-Bal9-containing sample, and errors for this sample were calculated by using the average peak noise found in the other spectra. These spectra were processed using a sine-bell window function, zero filling and phase correction in both dimensions. The peaks were then manually identified in SPARKY 60 by mapping the assignments from Zhou et al. 62 . With the exception of phosphorylated Histone H3-Bal9-containing nucleosome, the intensities were averaged across the three spectra and the standard deviations calculated. All nucleosome samples were measured in NMR buffer (1 mM EDTA, 10 mM NaP i pH 6.5, 5% v/v D 2 O). Denatured histones were measured in denaturing NMR buffer (50 mM NaP i , 3 M Gdn•HCl, 5% v/v D 2 O, pH 7.0).

Concentration determination.
To quantify the concentration of nucleosomes, 1 H spectra were recorded with 8,192 complex points and a sweep width of 10,417 Hz for an acquisition time of 786 ms. Then 32 scans per FID were recorded with an interscan delay of 1 s. The spectra were processed in NMRPipe 59 using a sine-bell window function, zero filling, phase correction and linear baselining and then the methyl regions (0.62 to 0.98 p.p.m.) were extracted for comparison. To quantify the concentration of denatured H3 histones, 1 H spectra were recorded with 31,248 complex points and a sweep width of 10,417 Hz for an acquisition time of 3 s. Next, 32 scans per FID were recorded with an interscan delay of 1 s. The spectra were processed in NMRPipe 59 using a sine-bell window function, zero filling, phase correction and linear baselining and then extracted the methyl regions (0.7 to 1.14 p.p.m.) for comparison ( Supplementary Fig. 14).
Temperature variation of denatured histones. The same experimental protocol was then applied to denatured Histone H3-Ser9 and denatured Histone H3-Bal9 samples. 2D 1 H-15 N BEST-transverse relaxation optimized spectra were recorded three times at three temperatures: 288, 298 and 303 K. Processing again involved a sine-bell window function, zero filling and phase correction in both dimensions. In the absence of assignments, isolated peaks in the glycine-serine-threonine region were peak picked in SPARKY 60 . Peak intensities were averaged and standard deviations calculated for errors. Denatured histones were measured in denaturing NMR buffer (50 mM NaP i , 3 M Gdn•HCl, 5% v/v D 2 O, pH 7.0) (Extended Data Fig. 9).
Reporting Summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/ s41589-021-00883-7.

M et ho ds
Small-molecule chemical synthesis. Detailed synthetic procedures are available in Supplementary Note 1. pH dependency. The pH dependency of oxaborolane formation on model compound Ac-Bal-NHBn was observed by 1 H-NMR using solutions of Ac-Bal-NHBn (20 mg ml −1 ) in NaP i buffer (20 mM in D 2 O). The pH was adjusted with concentrated DCl and 1 M NaOH in D 2 O. Quantification was conducted by integration of the C β -H signals, which are distinctly different for the boronic acid and the oxaborolane. The data were processed with GraphPad Prism v.8.0.0 (GraphPad Software Inc.) and fitting was performed using a sigmoidal standard curve (Supplementary Table 11) (where IC 50 is half-maximum inhibitory concentration): ) .