Targeted protein degradation in mycobacteria uncovers antibacterial effects and potentiates antibiotic efficacy

Won, Harim I.; Zinga, Samuel; Kandror, Olga; Akopian, Tatos; Wolf, Ian D.; Schweber, Jessica T. P.; Schmid, Ernst W.; Chao, Michael C.; Waldor, Maya; Rubin, Eric J.; Zhu, Junhao

doi:10.1038/s41467-024-48506-8

Download PDF

Article
Open access
Published: 14 May 2024

Targeted protein degradation in mycobacteria uncovers antibacterial effects and potentiates antibiotic efficacy

Nature Communications volume 15, Article number: 4065 (2024) Cite this article

2702 Accesses
16 Altmetric
Metrics details

Subjects

Abstract

Proteolysis-targeting chimeras (PROTACs) represent a new therapeutic modality involving selectively directing disease-causing proteins for degradation through proteolytic systems. Our ability to exploit targeted protein degradation (TPD) for antibiotic development remains nascent due to our limited understanding of which bacterial proteins are amenable to a TPD strategy. Here, we use a genetic system to model chemically-induced proximity and degradation to screen essential proteins in Mycobacterium smegmatis (Msm), a model for the human pathogen M. tuberculosis (Mtb). By integrating experimental screening of 72 protein candidates and machine learning, we find that drug-induced proximity to the bacterial ClpC1P1P2 proteolytic complex leads to the degradation of many endogenous proteins, especially those with disordered termini. Additionally, TPD of essential Msm proteins inhibits bacterial growth and potentiates the effects of existing antimicrobial compounds. Together, our results provide biological principles to select and evaluate attractive targets for future Mtb PROTAC development, as both standalone antibiotics and potentiators of existing antibiotic efficacy.

Legionella effector LnaB is a phosphoryl-AMPylase that impairs phosphosignalling

Article 22 May 2024

The intrinsic substrate specificity of the human tyrosine kinome

Article Open access 08 May 2024

Targeted protein degradation: from mechanisms to clinic

Article 29 April 2024

Introduction

Antibiotics, like many small molecule therapeutics, traditionally exert their mechanism of action through modulating a specific molecular target, whether by directly binding to an enzymatic active site or inducing allosteric conformational changes. Efforts to develop new antibiotics largely apply this same paradigm, but development remains challenging due to the high failure rate¹ and high cost². As a result, the discovery of new antibiotics has stalled in recent decades, including of those to treat bacterial infections like tuberculosis (TB), which remains one of the world’s leading infectious killers. The continued use of old antibiotics has compounded this problem, by giving rise to multi-drug resistant TB strains which are incredibly difficult to treat³. Because the traditional approach of single target modulation has slowed in its ability to discover powerful new anti-tubercular agents, we sought to apply a novel modality to TB antibiotic development.

Targeted protein degradation (TPD) is an emerging therapeutic modality that has progressed from concept⁴ to Phase 3 clinical trials⁵ within the past two decades. The major class of molecules that enable protein level regulation through TPD are the proteolysis-targeting chimeras (PROTACs). These molecules are heterobifunctional, comprising two small molecule ligands joined by a chemical linker. One of the ligands binds an E3 ubiquitin ligase while the other binds a given target protein. The PROTAC induces the proximity of both the target protein and the E3 ligase, resulting in polyubiquitination of the target and its subsequent degradation by the eukaryotic proteasome. Other TPD approaches employ the lysosomal⁶ and autophagic^7,8 degradation machinery. Despite its great potential, the development of TPD-based antimicrobials remains largely conceptual. Rather than using host degradation machinery in a TPD strategy for bacteria, induced autoproteolysis would instead deliver proteins essential for bacterial viability to bacterial degradation machinery, resulting in their degradation and subsequent cell death or growth inhibition. In mycobacteria, options for degradation machinery include the mycobacterial proteasome, which uses the PafA ligase (a system analogous to the ubiquitination system employed by PROTACs⁹), or the Clp proteolytic complexes (i.e., ClpC1P1P2 and ClpXP1P2, which are essential in mycobacteria¹⁰).

Recently, the first TPD proof-of-concept in bacteria demonstrated that heterobifunctional bacterial PROTACs (BacPROTACs) could mediate the inducible degradation of target proteins in Mycobacterium smegmatis (Msm) cells^11,12. This approach relied on direct delivery of target proteins to the multicomponent ClpC1P1P2 proteolytic complex (Fig. 1a), resulting in their degradation. In this work, BacPROTAC-mediated degradation of different targets was shown to induce sensitivity to D-cycloserine or auxotrophy for L-threonine, highlighting the potential of TPD as the foundation of a new class of antibiotic degraders¹¹. The same group later reported that dimeric-Cyclomarin A analog could kill M. tuberculosis (Mtb) by inducing auto-degradation of ClpC1 and other ClpC homologs in mycobacteria¹². However, it remains unclear whether all proteins are equally viable targets for this approach and for future drug development.

**Fig. 1: The FRB-FKBP dimerizable domains can be used to induce proximity in mycobacteria.**

In this work, we develop a rational platform for TPD target selection in mycobacteria and survey a set of diverse, essential mycobacterial proteins to examine their suitability for a TPD approach. Our genetic platform enables the rapid assessment of different target proteins for TPD, the output of which can subsequently train machine learning models to uncover critical protein features which underlie TPD potential. We also find that targeted degradation of particular substrates is sufficient to cause cellular growth inhibition and to potentiate killing by clinical and pre-clinical TB antibiotics even when targeted degradation alone functions at sub-inhibitory levels.

Results

A chemical genetic platform for evaluating targeted protein degradation in mycobacteria

The development of a targeted protein degradation (TPD) strategy for Mtb infection necessitates the identification of protein targets that are required for bacterial growth and are amenable to proximity mediated degradation by bacterial proteolytic systems. However, the systematic evaluation of vulnerable protein targets for TPD drug development is infeasible using a chemical-forward approach (i.e., identifying chemical ligands to over 400 genes that are essential for Mtb growth¹⁰). Therefore, we sought to develop a platform for rational target selection, which would enable the prioritization of targets before chemical screening for binding ligands. To accomplish this, we needed a system that would enable regulated, inducible proximity of native proteins and a mycobacterial proteolytic complex like ClpC1P1P2, which has recently been used for TPD in bacteria^11,12. For this purpose, we chose to create genetic fusions with FRB, the rapamycin-binding domain of mTOR, and FKBP12, which binds both rapamycin and FK506¹³; FRB and FKBP are small (~12 kDa) proteins which dimerize with nanomolar affinity only in the presence of rapamycin, forming a ternary complex^14,15. We fused ClpC1 with the FRB domain and candidate proteins with FKBP, creating strains where the presence of rapamycin induces the proximity of target proteins to ClpC1 (Fig. 1b). To enable rapid screening of targets which were most effectively degraded, we built this system in the non-pathogenic mycobacterial model species Msm.

Though FRB and FKBP have been widely used to conditionally dimerize proteins in mammalian cells¹⁵ and, more recently, in Escherichia coli¹⁶, they had not yet been evaluated in live mycobacterial cells. Therefore, we first assessed whether rapamycin-dependent dimerization occurred in Msm cells by fusing each domain to the NanoLuc (Nluc) 11 S (large) fragment and 114 (small) peptide split reporter system¹⁷. Compared to other split reporter systems (e.g., fluorescent proteins, β-galactosidase), the split Nluc (NanoBiT) system features the rapid signal detection, sensitivity, and dynamic range of luciferase assays while also being optimized to have minimal affinity between the split components. We expressed the FRB and FKBP Nluc fusion proteins in Msm and observed rapamycin-dependent luciferase activity only when both rapamycin-binding domains were fused to the split Nluc fragments (i.e., FRB-11s and FKBP-114), and this activity appeared to saturate at a rapamycin concentration of 1 μg·mL⁻¹ (Fig. 1c). We also observed the characteristic hook effect resulting from the formation of unproductive binary complexes at high concentrations rather than necessary ternary complexes required for induced proximity^18,19. Importantly, all concentrations of rapamycin we used to induce proximity in this work were inert with respect to bacterial growth and were orders of magnitude below the highest tested concentration which itself also did not inhibit bacterial growth in wild-type mc²155 strain Msm cells (Supplementary Fig. 1).

To generate the ClpC1 and FRB fusion protein, the FRB domain could be fused to either the N-terminal or C-terminal end. Based on the crystal structures of the Staphylococcus aureus ClpC and Mtb ClpC1 hexamer^20,21, we reasoned while N-terminal fusion might facilitate the direct delivery of target proteins to the ClpC1 hexamer pore, the genetic engineering of the ClpC1 C-terminus would be better tolerated by cells given the degree to which the N-terminal ends were nestled among the ClpC1 N-terminal domains (NTDs) compared to the relative accessibility of the C-terminal end (Fig. 1d). To compare these two scenarios, we purified both the N-terminal and C-terminal MtbClpC1 fusions (FRB-ClpC1 and ClpC1-FRB, respectively; Supplementary Fig. 2a) to measure each fusion protein’s ability to degrade enhanced green fluorescent protein tagged with ssrA (eGFP-ssrA). This target was selected as WT MtbClpC1 in association with ClpP1P2 had been previously shown to be able to degrade eGFP-ssrA in vitro^22,23. Indeed, we found FRB-ClpC1 had no measurable protease activity, while ClpC1-FRB was able to degrade eGFP-ssrA, albeit to a lesser degree than WT ClpC1 (Fig. 1e). We also observed this trend among the different ClpC1 fusion variants without ClpP1P2 when measuring ATPase activity, though here, FRB-ClpC1 showed some minimal activity (Supplementary Fig. 2b, c).

To generate a strain carrying a single copy of clpC1-frb under the native clpC1 promoter, we replaced wild-type clpC1 with a clpC1-frb allele via an L5 integrase allele swap²⁴ (Fig. 1f, top). We generated viable Msm strains with clean swaps for clpC1-frb alleles at a rate of 45.8%, successfully verifying that ClpC1-FRB remains functional in vivo and can complement wild-type ClpC1’s essential role in normal mycobacterial growth (Fig. 1f, bottom). We further confirmed that the addition of the FRB domain did not disrupt cellular physiology by measuring growth and found that the strain expressing ClpC1-FRB shared growth characteristics with parental Msm (Supplementary Fig. 3). This strain served as the basis for our platform to assess TPD in mycobacteria and examine whether we could mediate the degradation of specific target proteins by induced proximity to ClpC1.

Rapamycin-induced proximity to ClpC1 is sufficient to degrade a native mycobacterial protein

To test whether rapamycin-induced proximity to ClpC1 was sufficient for mycobacterial protein degradation, we constructed target protein fusions with FKBP and eGFP, introduced them as merodiploid copies into the Tweety phage attB site²⁵ in both a WT clpC1 and ΔclpC1 L5::clpC1-frb background (Fig. 2a, b), and measured the effect of rapamycin on fluorescence in Msm. Before testing different native targets, we assessed whether the minimal backbone, FKBP-eGFP alone, could be directed for degradation by ClpC1 or if degradation was more substrate selective. We found that heterologously expressed FKBP-eGFP was not degraded regardless of clpC1 background (Fig. 2c), suggesting either that it is a poor ClpC1 substrate (perhaps due to eGFP’s characteristic, protease-resistant tight folding²⁶) or that the geometry of delivery we achieved is incompatible with efficient degradation. To test whether the absence of rapamycin-induced degradation was due to a possible cryptic obstruction of the rapamycin-mediated protein-protein interaction, we examined the subcellular distribution of the FKBP-eGFP signal with or without rapamycin treatment. Previous work indicated that full-length ClpC1-eGFP locates near the plasma membrane and enriches at the subpolar region in a non-homogeneous manner²⁷ which we also observed (Supplementary Fig. 4a, b). While FKBP-eGFP in the clpC1-frb background diffuses in the cytosol, the addition of rapamycin results in re-localization of FKBP-eGFP to the subpolar membrane compartment (Supplementary Fig. 4c), indicating that rapamycin successfully mediates the physical interaction between FKBP-eGFP and ClpC1-FRB and further suggesting that FKBP-GFP alone is a suboptimal substrate for ClpC1.

**Fig. 2: Rapamycin redirects a native mycobacterial protein for degradation by ClpC1-FRB.**

We next selected RpoA, a small (38 kDa) and essential component of the RNA polymerase (RNAP) holoenzyme, to test whether the addition of an endogenous protein could potentiate proximity-induced degradation by ClpC1. Indeed, we found that rapamycin is sufficient to direct RpoA-FKBP-eGFP for degradation, with substantial loss of eGFP signal only in the clpC1-frb background (Fig. 2d). To validate that the observed loss of eGFP signal was the result of bona fide protein degradation, we showed a rapamycin-dependent reduction in protein levels of the full length RpoA-FKBP-eGFP fusion protein only in the clpC1-frb background (Fig. 2e). Furthermore, to assess the dynamics of this targeted protein degradation of RpoA, we examined individual cells by time-lapse fluorescence microscopy and observed a rapamycin-dependent reduction of eGFP signal only in the clpC1-frb background within 8 h of induction (Fig. 2f, g). Our findings demonstrate that a genetically encoded system for chemically-induced proximity to ClpC1 which is orthogonal to the chemical approach developed by ref. ¹¹., is sufficient to degrade a native mycobacterial protein.

Mycobacterial proteins are differentially susceptible to ClpC1-mediated TPD

To identify suitable endogenous targets for TPD more systematically, we sought to fuse a set of native mycobacterial proteins to FKBP-eGFP and to assess their degradation potential (Fig. 3a and Supplementary Fig. 6a). The fluorescent degradation kinetics of RpoA as measured by flow cytometry (Fig. 3b) correlated well with the signal observed by time-lapse microscopy (Supplementary Fig. 6b), motivating our use of flow cytometry as the basis for our target screening assay. To increase the generalizability of our screen in Msm to Mtb, we focused on protein orthologs which are highly conserved between the two species and vulnerable to depletion in both^10,28,29. In total, we identified 348 conserved Msm proteins that met these criteria (Methods, Supplementary Data 1), from which we chose 54 proteins with diverse functional and biochemical properties (Supplementary Fig. 7a, b). Using high-throughput flow cytometry, we quantified the effect of rapamycin on the fluorescence of these targets over time and found that most of the tested native proteins were able to be targeted for degradation by rapamycin (Fig. 3c and Supplementary Fig. 7c) and that the observed fluorescent signal decay could be modeled by an exponential decay function (Supplementary Fig. 7d), allowing us to calculate a rate constant for each target (Fig. 3d and Supplementary Data 2). While some targets displayed slower degradation kinetics – such as PyrG, which did not reach maximal, steady state degradation until 12 h (dG/dt = 0.026) – we found that others, such as the transcription factor RbpA, were rapidly degraded to a steady state baseline by 3 h (dG/dt = 1.414; Fig. 3d and Supplementary Fig. 7d).

**Fig. 3: Mycobacterial proteins are differentially susceptible to degradation by ClpC1-FRB.**

Interestingly, although all targets were expressed from the same constitutive promoter, we observed variance in the steady state fluorescence among the strains in the absence of rapamycin, indicating that there may be some basal post-translational regulation of some targets, possibly through proteolysis. We wondered whether these actively degraded proteins could also be more susceptible to TPD, if they are naturally recognized by endogenous proteases. Indeed, we found an inverse correlation with the log₂ degradation rate of each target and its steady state fluorescence (Fig. 3e; Pearson’s r = −0.69, P = 6.59 × 10⁻⁹). We also leveraged previously published transcriptomic³⁰ and quantitative proteomic³¹ datasets of the Mtb H37Rv lab strain to estimate the steady-state protein/RNA ratios of Mtb genes. We found that this orthogonal metric, which roughly reflected the composite effect of translation efficiency and protein turnover, also exhibited a mild, though statistically insignificant, negative correlation with the log₂-transformed degradation rate of the corresponding Msm homologs (Supplementary Fig. 8a). Together, these results imply that this intriguing relationship between steady-state protein abundance and protein degradability is probably not confounded by unknown technical caveats but is rather a reflection of intrinsic biological differences among these proteins.

To assess the effect of rapamycin-induced proximity to ClpC1 on protein localization and protein levels, we imaged each strain using fluorescence microscopy and again observed high concordance between the flow cytometry and microscopy datasets (Fig. 3f and Supplementary Fig. 8b, c). We also confirmed the gradient of degradation potential of the tested targets (Supplementary Fig. 8c). Notably, for several substrates, including NrdE, SecA1, and InhA, the addition of rapamycin resulted in profound re-localization of fluorescent signal (Fig. 3f, Supplementary Fig. 8c). Though these observations could reflect biology, these punctate foci could alternatively form due to oligomerization of eGFP due to increases in local concentrations near ClpC1-FRB. Finally, we examined whether rapamycin is stable in Msm culture conditions and found that eGFP levels remain low for at least 48 h post-inoculation, indicating that rapamycin is not rapidly metabolized below effective levels in that time period (Supplementary Fig. 9).

Protein sequence and structural features underlie TPD potential

The 54 substrates depicted in Fig. 3e vary not only in their measured degradation potential, but also by other primary sequence features, such as isoelectric point and aliphatic index (Supplementary Fig. 7b). Therefore, we sought to leverage the diversity of this dataset to query the sequence and structural determinants underlying TPD potential. We assembled a diverse set of numeric protein descriptors, encompassing quantitative structure-activity relationship (QSAR) metrics^32,33, AlphaFold2 prediction confidence³⁴, protein disorder propensity³⁵, and additional metrics, and applied them to both the full-length proteins and their terminal segments (Methods, Supplementary Data 2). After feature selection, 40 of the 485 protein features describing the 54 substrates and their measured degradation constants (the training set) were used to train a Lasso regression model (Fig. 4a, Supplementary Data 2). To evaluate the trained linear model, we constructed additional reporter strains for 18 protein substrates and measured their degradation rate upon rapamycin treatment. We found high concordance between the predicted and the measured degradation rate of the additional 18 substrates (the validation set; Fig. 4a, Supplementary Fig. 10, Supplementary Data 2). The performance of the relatively simple linear model implied linear or near-linear relationships between a subset of sequence or structural features and protein degradation potential. Indeed, we found that many of the selected features exhibited a statistically significant correlation with protein degradation potential (Supplementary Fig. 11a). A key example is the mean of disorder propensity of the N-terminal 30 residues (Fig. 4b), which suggested that the more disordered the N- terminal sequence is, the more efficiently TPD could act on the protein. The predictive power of this single numeric feature was further revealed by the terminal Disorder Propensity (flDPnn) profiles of the 72 tested substrates which implied that the well-degraded proteins tended to have disordered N- termini, but not necessarily disordered C-termini (Fig. 4b, c). Notably, when we combined the two batches of protein substrates which were used for model training and validation and compared their measured degradation potential to the steady-state protein-to-RNA ratio of their Mtb homologs, we indeed observed a statistically significant inverse correlation (Supplementary Fig. 11b), which further supported our hypothesis that efficient TPD is reliant on the substrate’s natural susceptibility to endogenous proteases. Having validated our linear prediction model, we applied it to predict the degradability of 348 essential proteins which are highly conserved between Mtb and Msm, and identified TPD candidates which are also vulnerable to transcriptional perturbation via CRISPRi²⁸ (Fig. 4e, Supplementary Data 3).

**Fig. 4: Machine learning uncovers features underlying protein degradation potential.**

TPD of native mycobacterial proteins inhibits bacterial growth and potentiates existing antimycobacterial compounds

In our previous experiments revealing that induced proximity drives the degradation of several mycobacterial proteins, we did not expect to see any impacts on growth because the strains we generated for these experiments were merodiploid for the protein being degraded (i.e., the genes were present in two copies), with only one of the copies tagged with fkbp-egfp. We therefore sought to test whether targeted degradation of native mycobacterial proteins predicted by our linear model to be highly susceptible to TPD (Fig. 4e) impacts bacterial growth. We accomplished this by endogenously tagging a subset of targets at their chromosomal loci with fkbp-egfp or fkbp-flag by recombineering in a ΔclpC1 L5::clpC1-frb background (Fig. 5a) and measuring the effect of rapamycin on bacterial growth. We selected targets spanning a range of predicted degradability and included known (e.g., RpoB), and potential (e.g., MmpL3, NusG³⁶ and RbpA³⁷) drug targets which our model predicted to be efficiently degraded in our system (Fig. 4e). We observed that rapamycin-mediated degradation of NusG, RbpA, Ffh, MmpL3, and AtpA resulted in substantial growth delay in cells grown in liquid media, whereas the targeted degradation of RpoB, RpsD, MutT4, SecA1, PyrG, and LeuS had little to no effect on bacterial growth (Fig. 5b).

**Fig. 5: Targeted degradation of native mycobacterial proteins inhibits bacterial growth and potentiates existing antibiotics.**

We then tested the extent to which endogenously tagged substrates could be degraded in a rapamycin-dependent fashion using fluorescence microscopy (Fig. 5c). It is worth noting that while most tested substrates were tagged with FKBP-eGFP, thereby permitting subsequent microscopy imaging, AtpA was excluded from this experiment as we found a cryptic duplication of the AtpA locus in the FKBP-eGFP tagged strain, but not in the FKBP-FLAG tagged counterpart (Supplementary Fig. 12a, b). Even though we expected the fluorescence of endogenous substrates to be confounded by variations in gene expression (Supplementary Fig. 12c), there still exists a robust positive correlation between the normalized fluorescence decrease of these FKBP-eGFP tagged endogenous substrates and their predicted degradation potential (Supplementary Fig. 12c–e)

In addition to changes in growth dynamics, fluorescence intensity, and protein localization (Fig. 5b, c), we also observed that rapamycin-induced degradation of several substrates, such as MmpL3, resulted in aberrant cell morphology that was reminiscent of reported morphological changes due to transcriptional perturbation³⁸ or chemical inhibition³⁹ (Fig. 5c, Supplementary Fig. 12f, g). Together, these results imply that our genetic platform effectively models chemically induced TPD of endogenous substrates, and that ClpC1-mediated degradation of certain substrates strongly inhibits mycobacterial growth.

The depletion of antibiotic targets, either at the protein or RNA level, has been shown to alter bacterial responses to compounds which inhibit those targets^40,41. We therefore sought to test whether targeted degradation of protein complex components could sensitize bacteria to clinically-relevant antibiotics that target the components themselves or other members of the complexes they form (Fig. 5d). There was no sensitization to any antibiotic tested when FKBP-eGFP was used as the substrate (Supplementary Fig. 13a). By contrast, we observed over 10- and 20-fold decreases in the half-maximal minimum inhibitory concentration (MIC₅₀) of the RNA polymerase inhibitor rifampicin, when the polymerase β subunit RpoB and the transcription facilitator RbpA were targeted for degradation^37,40, respectively (Fig. 5e). Fidaxomicin is another FDA-approved RNAP inhibitor that has regained attention for its excellent in vitro bactericidal activity against Mtb⁴². A recent study demonstrated that fidaxomicin jams mycobacterial RNAPs via direct interactions with the core complex and RbpA⁴³ and reported that removing the N-terminal tail of Msm RbpA weakened fidaxomicin binding as well as its growth inhibition activity. In support of the hypothesis that full-length RbpA is critical for maintaining the structural and functional integrity of mycobacterial RNAPs in the presence of both inhibitors, we found that the rapamycin induced degradation of RbpA indeed lowered the fidaxomicin MIC₅₀ by over four-fold (Fig. 5f). We also tested the impact of targeted degradation of RbpA on rifampicin-resistant (rif^R) isolates and found that RbpA depletion could re-sensitize rif^R Msm isolates to rifampicin, albeit to variable degrees depending on the specific mutation that conferred resistance (Supplementary Fig. 13b).

Compared to RpoB and RbpA, targeted degradation of MmpL3 and AtpA had a milder but still pronounced impact on Msm’s susceptibility to the MmpL3 inhibitor SQ109 and the ATP synthase inhibitor bedaquiline, respectively (Fig. 5g, h). Degrading RpoB, RbpA, and MmpL3 did not sensitize Msm to the antibiotics streptomycin and linezolid which inhibit translation by targeting the 30 S and 50 S ribosomal subunits, respectively, but degrading AtpA resulted in mild sensitization to both antibiotics which could suggest that these cells were in a compromised state that is more susceptible to antibiotic stress more generally (Supplementary Fig. 13c, d) Because bedaquiline has been shown to elicit a much delayed bactericidal activity against Mtb^44,45, we additionally measured the survival fraction of Msm exposed to 10× the observed MIC₅₀ of bedaquiline in the presence or the absence of rapamycin-induced degradation of AtpA. We found that targeted degradation of AtpA substantially accelerated cell killing by bedaquiline compared to cells with no degradation, which did not demonstrate any killing at all (Fig. 5i, P < 0.05 for 27 and 48 h). In the absence of bedaquiline, rapamycin had no discernible effect on bacterial growth (Supplementary Fig. 13e). Taken together, our findings highlight current antibiotic targets as pathways which can also be attacked by ClpC1-mediated targeted protein degradation, yielding direct effects on inhibiting bacterial growth, but importantly also sensitizing and resensitizing bacteria to currently approved antibiotics and accelerating their killing kinetics.

Discussion

In this work, we developed and validated a platform for the systematic identification of promising targets in the development of BacPROTAC targeted degrader antibiotics against mycobacteria. Genetic fusions to the FRB and FKBP domains enable the induced proximity of targets and the ClpC1 ATPase using the small molecule rapamycin, facilitating the subsequent proteolytic degradation of endogenous proteins. We show that many, but not all proteins we screened are effectively degraded using this approach, and that certain sequence properties, such as the disorder propensity of a protein’s N-terminus, are predictors of degradation potential of mycobacterial proteins. Moreover, we demonstrate that TPD-mediated depletion of certain proteins represses bacterial growth and sensitizes Msm to antibiotics which target the same pathway. Our results highlight the potential of TPD as a new modality for antimicrobial therapeutics.

When selecting targets for a TPD strategy, access to the target surface by the degradation mediating machinery (e.g., E3 ligase for PROTACs) is essential. The geometry of the nascent ternary complex formed when targets are brought in proximity to degradation machinery is therefore a critical factor in determining whether proteins can be degraded or not⁴⁶. In addition to the limitations on our ability to vary the geometry of potential targets, an additional layer of complexity to our ClpC1-mediated approach is that ClpC1 degradation has its own structural determinants. The rules governing whether ClpC1 will degrade a given protein are not well known. Previously, Lunge and others conducted quantitative proteomics to query the substrate pool of ClpC1 and reported that mycobacterial ClpC1 may preferentially degrade proteins with disordered termini⁴⁷. Our study, which approached this question using an orthogonal method, provided additional evidence to their hypothesis. The lack of association between the disorder propensity of protein C-terminus and degradation potential is probably because in our system, the C-termini of protein substrates are fused to FKBP-eGFP and pulled away from the entry pocket of the ClpC1 unfoldase. While further structural evidence to fully elucidate ClpC1’s substrate selectivity is still needed, our study along with previous ones may shed light on which native protein substrates should be prioritized for BacPROTAC drug design and screening.

Although we ultimately selected the ClpC1P1P2 proteolytic complex as the mediator of TPD in mycobacteria in our system, there are other endogenous protein degradation systems that may also be suitable⁴⁸. The Pup proteasome system (PPS) is more analogous to PROTACs⁴⁹; the mycobacterial PPS uses the pupylating ligase PafA, which is structurally distinct but functionally akin to human ubiquitin ligases, to mark endogenous proteins for degradation by the mycobacterial proteasome complex. In our proposed scheme of BacPROTAC, whereby ClpC1 is solely responsible for target binding and unfolding, the ClpC1 unfoldase should be physically occupied until the substrate is fully degraded. By contrast, a PafA-targeting BacPROTAC would deliver cognate substrates to PafA to be pupylated, after which PafA could, in principle, be recycled more quickly for the next round of substrate labeling. However, the PafA ligase and the bacterial proteasome likely have their own specificities and might only be appropriate for a restricted class of proteins⁵⁰. Given such fundamental differences in these degradation complexes, the designing principles of ClpC1- or PafA-targeting BacPROTAC compounds are foreseeably very different; therefore, a proof-of-principle for the latter should be established to further investigate the PPS’s suitability for a TPD approach in mycobacteria.

Using microscopy, we noted that for some, but not all targets, rapamycin addition resulted in the formation of fluorescent puncta. This raises the possibility that delivery of target proteins to ClpC1 could have a dual effect: first, proximity-mediated degradation of that protein, and second, forced removal of that protein from its native localization (Supplementary Fig. 8c), potentially perturbing optimal function or complex formation. This forced removal may be sufficient to cause unfavorable outcomes in bacteria even without substantial degradation; for example, Ffh was only degraded by about 30% of its steady state levels (Supplementary Fig. 7d), yet we observed substantial growth delays in liquid media when degraded (Fig. 5b). There is some evidence for this concept in Mtb, where it has been shown that small molecules that disrupt a native protein-protein interaction show strong bactericidal activity⁵¹. We posit that in addition to TPD, chemically-induced protein re-localization and aggregation may also serve as promising revenues for novel antibiotic discovery; however, further studies are needed to evaluate the feasibility and limitations of these methods, which are all based on the chemical perturbation of endogenous protein-protein interactions.

BacPROTACs exhibit great potential for the modular design of next-generation antimicrobials, as was demonstrated by our study and others previously^11,12. Nevertheless, technical hurdles in chemical binder screening, chemical synthesis of large modularized compounds, pharmacological optimization for improved membrane permeability and oral availability, host toxicity, and other properties pose major challenges in developing a drug-like BacPROTAC compound⁴⁸. Moreover, there exists only limited information regarding the frequency and mechanisms of BacPROTAC resistance^11,12. While we envision that the degradation potential of novel essential substrates, as predicted by our machine learning model, may facilitate target prioritization (Fig. 4e, Supplementary Data 3), the druggable potential of these targets would still require orthogonal validation, which could still be resource-demanding. In this case, prioritizing degradation-prone proteins which are also critical components of established drug targets could be advantageous, as BacPROTAC compounds that induce TPD of these substrates might further sensitize the bacteria to existing antibiotics and potentially accelerate the clearance of infection.

Methods

Bacterial strains and culture conditions

Mycobacterium smegmatis mc²155 and derivative strains were grown in Middlebrook 7H9 liquid media (Millipore) supplemented with ADC (5 g·L⁻¹ bovine albumin fraction V, 2 g·L⁻¹ dextrose, 0.003 g·L⁻¹ catalase, 0.85 g·L⁻¹ NaCl), 0.2% (v/v) glycerol, and 0.05% (v/v) Tween-80 or plated on LB agar or 7H10 agar. Antibiotic selection concentrations for Msm were: 25 μg·mL⁻¹ kanamycin, 20 μg·mL⁻¹ nourseothricin, 50 μg·mL⁻¹ hygromycin, and 20 μg·mL⁻¹ zeocin. Antibiotic selection concentrations for Escherichia coli were: 50 μg·mL⁻¹ kanamycin, 40 μg·mL⁻¹ nourseothricin, 100 μg·mL⁻¹ hygromycin, and 50 μg·mL⁻¹ zeocin. All strains were cultured at 37 °C. Rapamycin (Sigma) was dissolved in DMSO to a stock concentration of 1 mg·mL ⁻¹ or 10 mg·mL⁻¹. Plasmids and strains generated in this study are summarized in Supplementary Data 4 and 5. Primers used in this study are summarized in Supplementary Data 6.

Strain construction

mc²155 ΔclpC1 L5::clpC1-frb strain

The clpC1-frb strain was made by L5 allele swap. First, a clpC1 (msmeg_6091) merodiploid strain was generated by amplifying clpC1 with 500 bp of the sequence upstream of the gene to capture the native promoter, assembling into a kan^r plasmid with the L5 integrase, and integrating as a single copy at the L5 phage chromosomal attachment site (attB)⁵². Native clpC1 was knocked out by mycobacterial recombineering (phage-mediated homologous recombination)⁵³, replacing the native clpC1 allele with a zeocin selection marker, flanked by loxP sites. The knockout construct comprised ~500 bp homology arms upstream and downstream of clpC1 surrounding the floxed zeocin selection marker and was assembled into a pL5 vector⁵⁴. Next, clpC1 (with native promoter) and frb codon-optimized for mycobacteria were amplified, assembled into a ntc^r plasmid with the L5 integrase, and transformed into the ΔclpC1 L5::clpC1 strain; successful swaps acquired nourseothricin resistance at the expense of kanamycin resistance, whereas double integrants contained both markers. A 1:1 ratio of single to double integrants indicates that there is no fitness advantage to either allele.

Fluorescent target-fkbp-egfp fusions

The genes listed in Fig. 3e, Supplementary Figs. 7, 10, and Supplementary Data 2, as well as fkbp and egfp codon optimized for mycobacteria were amplified, assembled under control of the high-strength P_UV15TetO into a kan^r plasmid with the Tweety integrase, and transformed in the WT clpC1 or ΔclpC1 L5::clpC1-frb background. All gene IDs for the genes used in this study are listed in Supplementary Data 2.

Chromosomal locus target-fkbp fusions

The genes listed in Fig. 4e (red filled) and Fig. 5b were tagged with codon-optimized fkbp-flag (for AtpA only) or fkbp-egfp (the remaining chromosomally tagged strains) at their native locus by recombineering, by adding the coding sequence in frame immediately before each gene’s stop codon. First, the zeocin marker at the clpC1 chromosomal locus was removed by transformation with a Cre plasmid that is sucrose curable by SacB⁵⁵. The knock in construct comprised 500 bp homology arms upstream and downstream of each gene’s stop codon surrounding the floxed zeocin marker and was assembled into a pET-26b+ vector. All plasmids used in this work were cloned using isothermal assembly⁵⁶ and plasmid maps are available upon request.

Transformation

E. coli DH5α or XL1b competent cells were prepared using rubidium chloride and transformed by heat shock. Electrocompetent Msm cells were prepared by growing cells to mid-log phase (OD₆₀₀ ≈ 0.5), washing 3× with 10% (v/v) glycerol pre-chilled to 4 °C, then resuspending in 1/100^th of the initial culture volume with 10% glycerol. DNA (200 ng for integrating plasmids, 1 μg for recombineering) was added to 100 μL competent cells and incubated on ice for 5 min, followed by electroporation in a 2 mm cuvette at 2500 V, 125 Ω (plasmids; 1000 Ω recombineering), 25 μF using an ECM 630 electroporator (BTX). 7H9 media (1 mL) was added to the cells, which recovered for 3 h with shaking at 37 °C, before spreading on LB agar and incubating at 37 °C for 2–3 days.

Rifampicin-resistant strains

The rifampicin resistant strains (Supplementary Fig. 13b) were generated from the parent clpC1-frb, rbpA::fkbp-egfp and clpC1-frb, rpoB::fkbp-egfp strains. For each strain, 1 mL of Msm cells at OD₆₀₀ = 1.0 was plated on each of several 7H10 agar plates containing 100 μg·mL⁻¹ rifampicin and incubated for four days. A 462 bp region of rpoB containing the rifampicin resistance determining region was then amplified from individual colonies and sequenced to confirm the presence of rifampicin-resistance mutations.

NanoLuc assay

To validate rapamycin-mediated FRB-FKBP dimerization in mycobacteria, Msm cells were grown to mid-log phase (OD₆₀₀ ≈ 0.5) and plated at a final OD₆₀₀ = 0.1 with 5-fold serial dilutions of rapamycin. The Nano-Glo® 5× reagent (furimazine substrate + buffer; Promega) was prepared and added to each sample. Luminescence was measured in opaque white plates after incubation with linear shaking at 37 °C for 5 min using a Spark 10 M plate reader (Tecan).

Protein structure modeling

The crystal structures of the stabilized mutant MtbClpC1 hexamer (PDB: 8A8U)²¹, the Msm ATP synthase complex (PDB: 7NJK)⁵⁷, the Mtb RNA polymerase holoenzyme in complex with RbpA (PDB: 6C05)⁴², and MsmMmpL3 (PDB: 6AJG)⁵⁸ were modeled using PyMOL (Schrodinger). The MtbClpC1 NTD trimer was modeled by aligning three individual structures of the MtbClpC1 NTD (PDB: 6PBS)⁵⁹ with a E. coli ClpB NTD trimer (PDB: 6OG3)⁶⁰ as previously in ref. ²¹.

Protein expression and purification

The Mtb clpC1 (rv3596c) allele and frb codon optimized for mycobacteria were amplified and assembled to encode both N- and C-terminally fused gene products under control of the T7 promoter into a pET-26b+ plasmid, which encodes a 6xHis tag at the C-terminal end of the gene product. Constructs were transformed into BL21-CodonPlus(DE3)-RP cells (Agilent), which enables efficient expression of GC-rich genes in E. coli.

BL21 strains were grown in LB broth (Lennox) with shaking at 37 °C, until they reached OD₆₀₀ ≈ 0.8; protein expression was induced overnight with 100 μM IPTG and incubating with shaking at 18 °C overnight. Cells were harvested by centrifugation, subjected to a −80 °C freeze-thaw cycle, and resuspended in protein binding buffer (25 mM Gomori buffer pH 7.6, 100 mM KCl, 5% (v/v) glycerol). Next, cells were treated with 12.5 μg·mL⁻¹ lysozyme for 5 min and then lysed by sonication (30% power pulses for 30 s, alternating with ice incubation for 1 min; repeat 4×). Cell debris was removed by centrifugation and His-tagged ClpC1 proteins were purified from supernatants with Ni-NTA agarose beads (Qiagen). After washing in binding buffer supplemented with 20 mM imidazole, proteins were eluted with binding buffer supplemented with 100 mM and 200 mM imidazole. Input, flow through, and both elutions were examined by SDS-PAGE by loading samples in a NuPAGE 4–12% Bis-Tris precast gradient gel (Invitrogen) using MES SDS buffer.

Eluted proteins were concentrated and desalted with an Amicon® Ultra-4 Centrifugal Filter Device (30 kDa MWCO, Millipore Sigma), with equilibration in protein binding buffer lacking imidazole; protein concentrations were ascertained with the Pierce™ Coomassie Plus (Bradford) Assay Reagent (ThermoScientific) and comparison to a BSA standard curve. Purified fractions were pooled, aliquoted, and stored at −80 °C. FRB-tagged MtbClpC1 purification was conducted as one independent experiment and WT MtbClpC1, MtbClpP1, and MtbClpP2 were purified previously^22,61.

In vitro assays

ClpC1P1P2 protease activity

Degradation of eGFP-ssrA by purified MtbClpC1P1P2 was measured by examining loss in fluorescence signal over time. MtbClpC1 (WT and FRB-tagged proteins, 0.26 μM hexamer), MtbClpP1P2 (0.26 μM tetradecamer), and eGFP-ssrA (1.25 μM) were mixed in buffer containing: 25 mM Gomori buffer pH 7.6 supplemented with 100 mM KCl (Sigma), 5% glycerol (Sigma), 8 mM Mg-ATP (Sigma), and 2.5 mM Bz-Leu-Leu⁶² (Laboratory of Dr. William Bachovchin, Tufts University). Fluorescence signal decay was measured (Ex/Em: 485 nm / 520 nm) continuously in a SpectraMax M5 microplate reader (Molecular Devices) at 37 °C for 35 min.

ClpC1 ATPase activity

ATP hydrolysis by purified MtbClpC1 was measured using a pyruvate kinase and lactate dehydrogenase (PK/LDH) coupled enzymatic assay in which absorbance at 340 nm is an inverse proxy for the rate of ATP hydrolysis. MtbClpC1 (WT and FRB-tagged proteins, 0−0.7 μM hexamer), 1 mM phosphoenolpyruvate (Sigma), and 20 U·mL⁻¹ PK/LDH (Sigma) were mixed in buffer containing: 50 mM Tris pH 7.6 (Sigma), 100 mM KCl (Sigma), 1 mM DTT (Sigma), 1 mM NADH (Sigma), and 4 mM Mg-ATP (Sigma). Absorbance₃₄₀ was measured continuously in a SpectraMax M5 microplate reader (Molecular Devices) at 37 °C for 32 min.

Flow cytometry

To measure the fluorescence of the target reporter fusions in the presence of rapamycin, Msm cells were grown to mid-log phase (OD₆₀₀ ≈ 0.5) and plated at a final OD₆₀₀ = 0.2 (for the time course flow cytometry, OD₆₀₀ = 0.1) with the denoted concentration of rapamycin or DMSO (vehicle control). The flow cytometry screening was carried out in technical duplicate where the same bacterial culture was inoculated into replicate wells containing either DMSO or rapamycin. Cells were incubated with shaking at 700 rpm at 37 °C for the specified times (0 h, 3 h, 6 h, 9 h, 12 h), diluted in 7H9 media (to remain under 10,000 events·sec⁻¹), and fluorescence was quantified on a MACSQuant® VYB flow cytometer using the green fluorescence laser and filter set (channel B1, Ex/Em: 488 nm / 525/50 nm) with at least 10,000 events captured for each sample. To suppress spurious events due to noise or cell debris, event triggers were applied on forward scatter peak height (FSC-H > 0.8) and side scatter area (SSC-A > 1.2). To exclude cellular aggregates and morphological outliers, gates were applied with FlowJo (v. 10.8), and a representative gating strategy can be found in Supplementary Fig. 6a. The remaining events were normalized to mode and presented as histograms displaying log₁₀-transformed green fluorescence intensity area (AlexaFluor488-A), using GFP signal as a proxy for target protein levels in vivo. Full flow cytometry plots for all the training and validation reporter strains are included in Supplementary Figs. 7c and 10c, respectively.

Western blots

To directly examine protein levels when redirected to ClpC1-FRB, Msm cells were grown to mid-log phase (OD₆₀₀ ≈ 0.5), pelleted by centrifugation, and resuspended in 1× TBS supplemented with the cOmplete™ Protease Inhibitor Cocktail (Roche). Resuspended cells were aliquoted into FastPrep® 2 mL Lysing Matrix tubes (MPBio) and lysed by beat beating with a Precellys 24 tissue homogenizer (Bertin Technologies; 6,500 rpm for 45 s, alternating with ice incubation for 2 min; repeat 3×). Cell debris was removed by centrifugation, supernatant protein content was evaluated by NanoDrop 1000 (ThermoScientific), and all samples were normalized to the same protein concentration in 1× TBS with protease inhibitor. Normalized protein aliquots were treated with TURBO™ DNase (Invitrogen) for 15 min at 37 °C to digest mycobacterial genomic DNA and then denatured with boiling in Laemmli buffer. For all samples, 140 μg of protein and PageRuler™ Prestained Protein Ladder (ThermoScientific) was loaded onto NuPAGE 4–12% Bis-Tris precast gradient gels (Invitrogen) using MES SDS buffer, transferred to PVDF membranes using a Trans-Blot Turbo Transfer System (Bio-Rad), and blocked with 1:1 SEA-BLOCK blocking buffer (ThermoScientific) and 1× PBS with shaking for 1 h at 25 °C. Membranes were probed with the specified antibodies in 1:1 SEA-BLOCK and 1× PBS-Tween (phosphate-buffered saline with 0.1% (v/v) Tween-20) with shaking for 1 h each primary and secondary at 25 °C. Western blots were imaged using the Odyssey® CLx Infrared Imaging System (LI-COR) and processed using LI-COR ImageStudio. The full uncropped scans of the blot shown in Fig. 2e can be found in Supplementary Fig. 5. The mouse IgG2b monoclonal RpoB antibody (Invitrogen, clone #8RB13) was diluted 1:1,000⁶³. The rabbit IgG monoclonal GFP antibody (Abcam, clone #EPR14104) was diluted 1:1,000⁶⁴. The IRDye® 680RD goat IgG (H + L) anti-mouse and IRDye® 800CW goat IgG (H + L) anti-rabbit (LI-COR) fluorescent secondary antibodies were diluted 1:15,000.

Time-lapse fluorescence microscopy and data analysis

To generate the time-lapse data for targeted degradation of RpoA-FKBP-eGFP in Fig. 2f, cells were grown until late log phase (OD₆₀₀ ≈ 1.0)) and seeded on a 2.0% agarose pad containing 1× concentration of 7H9, 0.1% w/v casamino acid (BD), 0.2% w/v glucose, 0.2% v/v glycerol and 0.1 μg·mL⁻¹ rapamycin. The agarose pad was cast in a 12 × 12 mm² customized plastic frame and placed in a low-evaporation imaging disk (MatTek Corp.) to enable long-term time-lapse experiments. Fluorescence and phase-contrast images were acquired every 10 min over a 9 h period with a 100× Plan Apo oil objective (NA = 1.45) using a Nikon Ti-E inverted, widefield microscope equipped with a Nikon Perfect Focus system with a Piezo Z drive motor, Andor Zyla sCMOS camera, and an Agilent MLC400 Monolithic laser combiner. Fluorescence signals were acquired using a 6-channel Spectra X LED light source and the Sedat Quad filter set. The excitation and emission filters used for the green fluorophores (eGFP) were Ex. 470/24 nm and Em. 515/25 nm. Time-lapse snapshots were rendered using customized Python scripts based on the Python package microfilm (https://github.com/guiwitz/microfilm). To analyze time-lapse data, our previously established microscopy image analysis pipeline MOMIA⁶⁵ was modified to track the fluorescence of individual cells and emergence of new cell generations. Briefly, full-sized (2048 × 2048) microscopy image stacks were manually cropped to capture the growth of individual microcolonies. A Residual-Attention-Unet model⁶⁶ was trained to perform cell segmentation from phase contrast images. Segmented objects of consecutive frames were linked based on their overlaps in adjacent frames and their resemblance in cell morphology. The initial linkage was refined iteratively to minimize linkage drift. Cellular fluorescence per frame was estimated by the mean intensity of cellular ROI of the fluorescence image, which had been cropped and removed of background fluorescence using the rolling-ball method.

Target protein microscopy image acquisition

To capture the microscopy data for targeted degradation of all 72 target proteins examined in Fig. 3f and Supplementary Figs. 8c and 10e, cells were sampled from the flow cytometry experiment at 9 h after inoculation and seeded on 2.0% agarose pads. Fluorescence and phase-contrast images were acquired using the same microscope configurations as used for the time-lapse experiments. Image segmentation and fluorescence intensity quantification were performed using a Python analysis pipeline similar to the one used for time-lapse image analysis. Image crops were rendered using customized Python scripts powered by microfilm (guiwitz.github.io/microfilm).

Protein feature extraction

The full list of conserved orthologs shared by Msm and Mtb was generated by combining ortholog annotations from the Mycobacteria Systems Resources⁵⁷ and Mycobrowser²⁹, which yielded a total number of 2450 protein candidates. Gene essentiality and vulnerability annotations were referenced from the seminal work from Bosch and others^10,28. After removing non-essential candidates, we obtained 354 essential orthologs. We also included 3 genes (msmeg_3858/rbpA, msmeg_1435/rpsJ, msmeg_2299/nrdE) for which their Mtb counterparts were annotated as essential whereas the annotation for these Msm copies were ambiguous. Proteins that are shorter than 100 amino acids or longer than 1500 amino acids were manually excluded from the candidate list. For the remaining 348 proteins, AlphaFold prediction confidence scores (pLDDt) were retrieved from the AlphaFold Protein Structure Database (https://alphafold.ebi.ac.uk/)^34,58, and Disorder Potential predictions were downloaded from the flDPnn web server (http://biomine.cs.vcu.edu/servers/flDPnn/)³⁵. For each protein, we generated five test subjects: the N- terminal 15 or 30 amino acids, the C- terminal 15 or 30 amino acids, and the full-length protein. For each subject, we generated a comprehensive set of quantitative structure-activity relationship (QSAR) metrics using the Python package peptides (https://github.com/althonos/peptides.py) and computed the average and standard deviation of the sectional QSAR, pLDDt, and flDPnn scores. To estimate the length of the unstructured stretch of the termini, we defined the lower or higher bound for flDPnn or pLDDt scores as 0.2 and 50, respectively, and computed the fraction of disordered or poorly predicted terminal residues. The 348 conserved proteins and their corresponding protein features are listed in Supplementary Data 1.

Machine learning and prediction

Z-score data normalization was performed on each of the 485 numeric features across the 348 proteins. Feature selection was carried out using the SelectKBest function scored of the Python package Scikit-learn⁶⁷. Here we used data of the substrates of the first batch (54 proteins in total) as the full training set, which was further split into two subsets at a ratio of 0.67 and 0.33, respectively. The random train-test splitting and SelectKBest scoring were repeated 10 times to find features which better correlate with the log₂-transformed degradation constants. The scores were averaged and ranked, and the top 40 features were selected for model training. Multiple Lasso regression models were trained with varying L1 regularization terms on the full training set and evaluated on the validation dataset (comprised of the 18 proteins of the second experimental batch) by Pearson’s correlation. The best model achieved a Pearson’s correlation coefficient over 0.6 and a statistically significant P value (0.0057), The selected model was used to predict the log₂-transformed degradation constant as depicted in Fig. 4e.

atpA locus PCR validation

To test for the presence of an atpA locus duplication (Supplementary Figs. 12a, b), primers were designed to amplify regions corresponding to the wild-type or tagged locus. A primer pair to amplify a region of rpoB was included as a control. For each reaction, 16 ng of genomic DNA from the indicated strains was used as a template and amplified using KOD Xtreme™ Hot Start DNA Polymerase (Sigma-Aldrich).

Minimum inhibitory concentration (MIC) assay

To calculate their dose response to antibiotics and targeted degradation, Msm cells were grown to mid-log phase (OD₆₀₀ ≈ 0.5) and plated at a final OD₆₀₀ = 0.0015 in each well of a 96-well plate containing serial dilutions of the specified antibiotics. For assays with concurrent targeted degradation, all wells were supplemented with the specified concentration of rapamycin or DMSO. Plates were incubated with drug with shaking at 150 rpm at 37 °C overnight. Resazurin was added to a final concentration of 0.002% (w/v), plates were incubated up to 18 h, and fluorescence conversion was measured (Ex/Em: 560 nm/590 nm) in a SpectraMax M2 microplate reader (Molecular Devices) or a Spark 10 M plate reader (Tecan). Background fluorescence was subtracted from wells without cells and fluorescence signal was normalized to wells without any compound. A nonlinear regression was used to fit a sigmoid curve to the dose response data and calculate the half-maximal minimum inhibitory concentration (MIC₅₀) using GraphPad Prism.

Antibiotic kill kinetics curves

To examine the effect of targeted degradation on bedaquiline killing, Msm cells were grown to mid-log phase (OD₆₀₀ ≈ 0.5) and plated at a final OD₆₀₀ = 0.08 in 24 well plates; wells were supplemented with 0× or 10× their measured MIC of bedaquiline and with 0.5 μg·mL⁻¹ rapamycin or DMSO. Plates were incubated for 2 days with shaking at 155 rpm at 37 °C and cells were sampled at 0, 4, 8, 20, 27, and 48 h after plating. Sampled cells were serially diluted and plated on plain LB agar; plates were incubated for 2–3 days at 37 °C and individual colony forming units (CFUs) were counted.

Bacterial growth kinetics curves

To examine the effect of targeted degradation on bacterial growth in liquid media, Msm cells were grown to mid-log phase (OD₆₀₀ ≈ 0.5) and plated at a final OD₆₀₀ = 0.0015. Wells were supplemented with the specified concentration of rapamycin or DMSO. Absorbance (OD₆₀₀) and/or eGFP fluorescence (Ex/Em: 488 nm/510 nm) was measured in clear 96 well plates sealed with a Breathe-Easy® membrane (Sigma) with linear shaking at 37 °C for 48 h using a Spark 10 M plate reader (Tecan).

Statistical analysis

The statistical analyses and correlation plots described in this study were performed using the python package SciPy⁶⁸ or GraphPad Prism. All experiments were performed at least twice (biological replicates), unless otherwise noted. Technical replicates were measured within the same experiment from distinct samples, unless otherwise noted. Means were compared using a two-tailed Student’s t-test with a Holm-Šídák correction. Adjusted P < 0.05 was considered significant. For the microscopy experiments at least 70 cells were analyzed for each shown image and the image data shown are representative of multiple fields. No statistical methods were used to determine sample sizes used in this work and the investigators were not blinded to sample identity.

Software

Fiji (1.52p), Scikit-learn (1.0.1), Scikit-image (0.19.3), OpenCV2 (4.5), NumPy (1.22.4), Tensorflow (2.6), Keras (2.6), Pandas (1.3.4), and Python (3.8) to analyze plate and microscopy images; SciPy (1.10.0), Matplotlib (3.6.2) and GraphPad Prism (10.0.3) were used to perform statistical analysis and to create data graphs. Flowjo (10.8.2), NumPy, and SciPy were used to analyze flow cytometry data. Peptides.py (v0.3.2) was used to extract sequence features from proteins and oligopeptides.

Data availability

Example unprocessed microscopy images, clips of the timelapse microscopy data, and flow cytometry data generated in this study have been deposited in the Zenodo database under the following https://doi.org/10.5281/zenodo.11004395. Due to size limitations, the rest of the microscopy data are available upon request. AlphaFold protein structure files were downloaded from the AlphaFold Protein Structure Database and are available at https://alphafold.ebi.ac.uk. Source data are provided with this paper.

Code availability

All code used to generate primers and analyze data has been deposited on Zenodo and is available under the https://doi.org/10.5281/zenodo.11004395.

References

Payne, D. J., Gwynn, M. N., Holmes, D. J. & Pompliano, D. L. Drugs for bad bugs: confronting the challenges of antibacterial discovery. Nat. Rev. Drug Discov. 6, 29–40 (2007).
Article CAS PubMed Google Scholar
Årdal, C. et al. Insights into early stage of antibiotic development in small- and medium-sized enterprises: a survey of targets, costs, and durations. J. Pharm. Policy Pract. 11, 8 (2018).
Article PubMed PubMed Central Google Scholar
Global Tuberculosis Programme. Global Tuberculosis Report 2023. 43–44 (Geneva: World Health Organization).
Sakamoto, K. M. et al. Protacs: Chimeric molecules that target proteins to the Skp1–Cullin–F box complex for ubiquitination and degradation. Proc. Natl Acad. Sci. Usa. 98, 8554–8559 (2001).
Article ADS CAS PubMed PubMed Central Google Scholar
Campone, M. et al. VERITAC-2: A global, randomized phase 3 study of ARV-471, a proteolysis targeting chimera (PROTAC) estrogen receptor (ER) degrader, vs fulvestrant in ER+/human epidermal growth factor receptor 2 (HER2)- advanced breast cancer. J. Clin. Oncol. 41, TPS1122–TPS1122 (2023).
Article Google Scholar
Banik, S. M. et al. Lysosome-targeting chimaeras for degradation of extracellular proteins. Nature 584, 291–297 (2020).
Article ADS CAS PubMed PubMed Central Google Scholar
Takahashi, D. et al. AUTACs: Cargo-Specific Degraders Using Selective Autophagy. Mol. Cell 76, 797–810.e10 (2019).
Article CAS PubMed Google Scholar
Li, Z. et al. Allele-selective lowering of mutant HTT protein by HTT–LC3 linker compounds. Nature 575, 203–209 (2019).
Article ADS CAS PubMed Google Scholar
Pearce, M. J., Mintseris, J., Ferreyra, J., Gygi, S. P. & Darwin, K. H. Ubiquitin-like protein involved in the proteasome pathway of Mycobacterium tuberculosis. Science 322, 1104–1107 (2008).
Article ADS CAS PubMed PubMed Central Google Scholar
DeJesus, M. A. et al. Comprehensive Essentiality Analysis of the Mycobacterium tuberculosis Genome via Saturating Transposon Mutagenesis. mBio 8, e02133-16 (2017).
Article PubMed PubMed Central Google Scholar
Morreale, F. E. et al. BacPROTACs mediate targeted protein degradation in bacteria. Cell 185, 2338–2353.e18 (2022).
Article CAS PubMed PubMed Central Google Scholar
Hoi, D. M. et al. Clp-targeting BacPROTACs impair mycobacterial proteostasis and survival. Cell 186, 2176–2192.e22 (2023).
Article CAS PubMed Google Scholar
Chen, J., Zheng, X. F., Brown, E. J. & Schreiber, S. L. Identification of an 11-kDa FKBP12-rapamycin-binding domain within the 289-kDa FKBP12-rapamycin-associated protein and characterization of a critical serine residue. Proc. Natl. Acad. Sci. 92, 4947–4951 (1995).
Article ADS CAS PubMed PubMed Central Google Scholar
Choi, J., Chen, J., Schreiber, S. L. & Clardy, J. Structure of the FKBP12-Rapamycin Complex Interacting with Binding Domain of Human FRAP. Science 273, 239–242 (1996).
Article ADS CAS PubMed Google Scholar
Banaszynski, L. A., Liu, C. W. & Wandless, T. J. Characterization of the FKBP·Rapamycin·FRB Ternary Complex. J. Am. Chem. Soc. 127, 4715–4721 (2005).
Article CAS PubMed Google Scholar
Cui, B. et al. Bioluminescence Resonance Energy Transfer System for Measuring Dynamic Protein-Protein Interactions in Bacteria. mBio 5, e01050-14 (2014).
Article PubMed PubMed Central Google Scholar
Dixon, A. S. et al. NanoLuc Complementation Reporter Optimized for Accurate Measurement of Protein Interactions in Cells. ACS Chem. Biol. 11, 400–408 (2016).
Article ADS CAS PubMed Google Scholar
Tate, J. & Ward, G. Interferences in Immunoassay. Clin. Biochem. Rev. 25, 105–120 (2004).
PubMed PubMed Central Google Scholar
Douglass, E. F. Jr., Miller, C. J., Sparer, G., Shapiro, H. & Spiegel, D. A. A Comprehensive Mathematical Model for Three-Body Binding Equilibria. J. Am. Chem. Soc. 135, 6092–6099 (2013).
Article CAS PubMed PubMed Central Google Scholar
Carroni, M. et al. Regulatory coiled-coil domains promote head-to-head assemblies of AAA+ chaperones essential for tunable activity control. eLife 6, e30120 (2017).
Article PubMed PubMed Central Google Scholar
Weinhäupl, K. et al. Structure of the drug target ClpC1 unfoldase in action provides insights on antibiotic mechanism of action. J. Biol. Chem. 298, 102553 (2022).
Article PubMed PubMed Central Google Scholar
Fraga, H. et al. Development of high throughput screening methods for inhibitors of ClpC1P1P2 from Mycobacteria tuberculosis. Anal. Biochem. 567, 30–37 (2019).
Article CAS PubMed Google Scholar
Schmitz, K. R. & Sauer, R. T. Substrate delivery by the AAA+ ClpX and ClpC1 unfoldases activates the mycobacterial ClpP1P2 peptidase. Mol. Microbiol. 93, 617–628 (2014).
Article CAS PubMed PubMed Central Google Scholar
Pashley, C. A. & Parish, T. Efficient switching of mycobacteriophage L5-based integrating plasmids in Mycobacterium tuberculosis. FEMS Microbiol. Lett. 229, 211–215 (2003).
Article CAS PubMed Google Scholar
Pham, T. T., Jacobs-Sera, D., Pedulla, M. L., Hendrix, R. W. & Hatfull, G. F. Comparative genomic analysis of mycobacteriophage Tweety: evolutionary insights and construction of compatible site-specific integration vectors for mycobacteria. Microbiol. Read. Engl. 153, 2711–2723 (2007).
Article CAS Google Scholar
Bokman, S. H. & Ward, W. W. Renaturation of Aequorea green-fluorescent protein. Biochem. Biophys. Res. Commun. 101, 1372–1380 (1981).
Article CAS PubMed Google Scholar
Murphy, K. C. et al. ORBIT: a New Paradigm for Genetic Engineering of Mycobacterial Chromosomes. mBio 9, e01467–18 (2018).
Bosch, B. et al. Genome-wide gene expression tuning reveals diverse vulnerabilities of M. tuberculosis. Cell 184, 4579–4592.e24 (2021).
Article CAS PubMed PubMed Central Google Scholar
Kapopoulou, A., Lew, J. M. & Cole, S. T. The MycoBrowser portal: a comprehensive and manually annotated resource for mycobacterial genomes. Tuberc. Edinb. Scotl. 91, 8–13 (2011).
CAS Google Scholar
Płociński, P. et al. Proteomic and transcriptomic experiments reveal an essential role of RNA degradosome complexes in shaping the transcriptome of Mycobacterium tuberculosis. Nucleic Acids Res. 47, 5892–5905 (2019).
Article PubMed PubMed Central Google Scholar
Schubert, O. T. et al. Absolute Proteome Composition and Dynamics during Dormancy and Resuscitation of Mycobacterium tuberculosis. Cell Host Microbe 18, 96–108 (2015).
Article CAS PubMed Google Scholar
Osorio, D., Rondón-Villarreal, P. & Torres, R. Peptides: A Package for Data Mining of Antimicrobial Peptides. R. J. 7, 4 (2015).
Article Google Scholar
Hellberg, S., Sjöström, M., Skagerberg, B. & Wold, S. Peptide quantitative structure-activity relationships, a multivariate approach. J. Med. Chem. 30, 1126–1135 (1987).
Article CAS PubMed Google Scholar
Jumper, J. et al. Highly accurate protein structure prediction with AlphaFold. Nature 596, 583–589 (2021).
Article ADS CAS PubMed PubMed Central Google Scholar
Hu, G. et al. flDPnn: Accurate intrinsic disorder prediction with putative propensities of disorder functions. Nat. Commun. 12, 4438 (2021).
Article ADS CAS PubMed PubMed Central Google Scholar
Delbeau, M. et al. Structural and functional basis of the universal transcription factor NusG pro-pausing activity in Mycobacterium tuberculosis. Mol. Cell 83, 1474–1488.e8 (2023).
Article CAS PubMed PubMed Central Google Scholar
Bortoluzzi, A. et al. Mycobacterium tuberculosis RNA Polymerase-binding Protein A (RbpA) and Its Interactions with Sigma Factors. J. Biol. Chem. 288, 14438–14450 (2013).
Article CAS PubMed PubMed Central Google Scholar
De Wet, T. J., Winkler, K. R., Mhlanga, M., Mizrahi, V. & Warner, D. F. Arrayed CRISPRi and quantitative imaging describe the morphotypic landscape of essential mycobacterial genes. eLife 9, e60083 (2020).
Article PubMed PubMed Central Google Scholar
Tahlan, K. et al. SQ109 targets MmpL3, a membrane transporter of trehalose monomycolate involved in mycolic acid donation to the cell wall core of Mycobacterium tuberculosis. Antimicrob. Agents Chemother. 56, 1797–1809 (2012).
Article CAS PubMed PubMed Central Google Scholar
Wei, J.-R. et al. Depletion of antibiotic targets has widely varying effects on growth. Proc. Natl Acad. Sci. 108, 4176–4181 (2011).
Article ADS CAS PubMed PubMed Central Google Scholar
Li, S. et al. CRISPRi chemical genetics and comparative genomics identify genes mediating drug potency in Mycobacterium tuberculosis. Nat. Microbiol. 7, 766–779 (2022).
Article PubMed PubMed Central Google Scholar
Quan, D., Nagalingam, G., Payne, R. & Triccas, J. A. New tuberculosis drug leads from naturally occurring compounds. Int. J. Infect. Dis. 56, 212–220 (2017).
Article CAS PubMed Google Scholar
Boyaci, H. et al. Fidaxomicin jams Mycobacterium tuberculosis RNA polymerase motions needed for initiation via RbpA contacts. eLife 7, e34823 (2018).
Article PubMed PubMed Central Google Scholar
Diacon, A. H. et al. 14-day bactericidal activity of PA-824, bedaquiline, pyrazinamide, and moxifloxacin combinations: a randomised trial. Lancet Lond. Engl. 380, 986–993 (2012).
Article CAS Google Scholar
Diacon, A. H. et al. Randomized dose-ranging study of the 14-day early bactericidal activity of bedaquiline (TMC207) in patients with sputum microscopy smear-positive pulmonary tuberculosis. Antimicrob. Agents Chemother. 57, 2199–2203 (2013).
Article CAS PubMed PubMed Central Google Scholar
Smith, B. E. et al. Differential PROTAC substrate specificity dictated by orientation of recruited E3 ligase. Nat. Commun. 10, 131 (2019).
Article ADS PubMed PubMed Central Google Scholar
Lunge, A., Gupta, R., Choudhary, E. & Agarwal, N. The unfoldase ClpC1 of Mycobacterium tuberculosis regulates the expression of a distinct subset of proteins having intrinsically disordered termini. J. Biol. Chem. 295, 9455–9473 (2020).
Article CAS PubMed PubMed Central Google Scholar
Sarathy, J. P., Aldrich, C. C., Go, M.-L. & Dick, T. PROTAC antibiotics: the time is now. Expert Opin. Drug Discov. 18, 363–370 (2023).
Article CAS PubMed PubMed Central Google Scholar
Jastrab, J. B. & Darwin, K. H. Bacterial Proteasomes. Annu. Rev. Microbiol. 69, 109–127 (2015).
Article CAS PubMed PubMed Central Google Scholar
Festa, R. A. et al. Prokayrotic Ubiquitin-Like Protein (Pup) Proteome of Mycobacterium tuberculosis. PLoS ONE 5, e8589 (2010).
Article ADS PubMed PubMed Central Google Scholar
Lin, Y. et al. Identification of antituberculosis agents that target ribosomal protein interactions using a yeast two-hybrid system. Proc. Natl Acad. Sci. 109, 17412–17417 (2012).
Article ADS CAS PubMed PubMed Central Google Scholar
Lee, M. H., Pascopella, L., Jacobs, W. R. & Hatfull, G. F. Site-specific integration of mycobacteriophage L5: integration-proficient vectors for Mycobacterium smegmatis, Mycobacterium tuberculosis, and bacille Calmette-Guérin. Proc. Natl Acad. Sci. Usa. 88, 3111–3115 (1991).
Article ADS CAS PubMed PubMed Central Google Scholar
van Kessel, J. C. & Hatfull, G. F. Recombineering in Mycobacterium tuberculosis. Nat. Methods 4, 147–152 (2007).
Article PubMed Google Scholar
Kieser, K. J. et al. Phosphorylation of the Peptidoglycan Synthase PonA1 Governs the Rate of Polar Elongation in Mycobacteria. PLoS Pathog. 11, e1005010 (2015).
Article PubMed PubMed Central Google Scholar
Pelicic, V., Reyrat, J. M. & Gicquel, B. Expression of the Bacillus subtilis sacB gene confers sucrose sensitivity on mycobacteria. J. Bacteriol. 178, 1197–1199 (1996).
Article CAS PubMed PubMed Central Google Scholar
Gibson, D. G. et al. Enzymatic assembly of DNA molecules up to several hundred kilobases. Nat. Methods 6, 343–345 (2009).
Article CAS PubMed Google Scholar
Judd, J. A. et al. A Mycobacterial Systems Resource for the Research Community. mBio 12, e02401–e02420 (2021).
Article CAS PubMed PubMed Central Google Scholar
Varadi, M. et al. AlphaFold Protein Structure Database: massively expanding the structural coverage of protein-sequence space with high-accuracy models. Nucleic Acids Res 50, D439–D444 (2022).
Article CAS PubMed Google Scholar
Wolf, N. M. et al. Structure of the N-terminal domain of ClpC1 in complex with the antituberculosis natural product ecumicin reveals unique binding interactions. Acta Crystallogr. Sect. Struct. Biol. 76, 458–471 (2020).
Article ADS CAS Google Scholar
Rizo, A. N. et al. Structural basis for substrate gripping and translocation by the ClpB AAA+ disaggregase. Nat. Commun. 10, 2393 (2019).
Article ADS PubMed PubMed Central Google Scholar
Gao, W. et al. The cyclic peptide ecumicin targeting ClpC1 is active against Mycobacterium tuberculosis in vivo. Antimicrob. Agents Chemother. 59, 880–889 (2015).
Article PubMed PubMed Central Google Scholar
Li, M. et al. Structure and Functional Properties of the Active Form of the Proteolytic Complex, ClpP1P2, from Mycobacterium tuberculosis. J. Biol. Chem. 291, 7465–7476 (2016).
Article CAS PubMed PubMed Central Google Scholar
Zhu, J.-H. et al. Rifampicin can induce antibiotic tolerance in mycobacteria via paradoxical changes in rpoB transcription. Nat. Commun. 9, 4218 (2018).
Article ADS PubMed PubMed Central Google Scholar
Campbell, B. C., Paez-Segala, M. G., Looger, L. L., Petsko, G. A. & Liu, C. F. Chemically stable fluorescent proteins for advanced microscopy. Nat. Methods 19, 1612–1621 (2022).
Article CAS PubMed PubMed Central Google Scholar
Zhu, J. et al. Spatiotemporal localization of proteins in mycobacteria. Cell Rep. 37, 110154 (2021).
Article CAS PubMed PubMed Central Google Scholar
Ni, Z-L. et al. RAUNet: Residual Attention U-Net for Semantic Segmentation of Cataract Surgical Instruments. Lect. Notes Comput. Sci. 11954, 139–149 (2019).
Pedregosa, F. et al. Scikit-learn: Machine Learning in Python. J. Mach. Learn Res 12, 2825–2830 (2011).
MathSciNet Google Scholar
Virtanen, P. et al. SciPy 1.0: fundamental algorithms for scientific computing in Python. Nat. Methods 17, 261–272 (2020).
Article CAS PubMed PubMed Central Google Scholar

Download references

Acknowledgements

We thank members of the Eric J. Rubin and Sarah Fortune labs for helpful discussions during this work and Douaa Mugahid for thoughtful feedback on the manuscript. We thank Kenan Murphy for sharing the ClpC1-eGFP Msm strain. Funding: This material is based upon work supported by the NSF Graduate Research Fellowship under Grant No. DGE1745303 (H.W.), the Harvard GSAS Herchel Smith Graduate Fellowship (H.W.), the Marcus Urann Graduate Fellowship (H.W.), the National Institute of General Medical Sciences award T32GM007753 and T32GM144273 (S.Z.), the NIH T32 Molecular, Cellular, and Developmental Mechanisms grant 5T32GM145407 (S.Z.), the NSF Graduate Research Fellowship under Grant No. DGE2140743 (E.S.), the NIH/NIAID award U19 AI142735 (E.J.R.), and funding provided by Grace Wang and Josef Tatelbaum. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institute of General Medical Sciences or the National Institutes of Health.

Author information

These authors contributed equally: Harim I. Won, Samuel Zinga.

Authors and Affiliations

Department of Immunology and Infectious Diseases, Harvard T.H. Chan School of Public Health, Boston, MA, 02115, USA
Harim I. Won, Samuel Zinga, Olga Kandror, Tatos Akopian, Ian D. Wolf, Jessica T. P. Schweber, Michael C. Chao, Maya Waldor, Eric J. Rubin & Junhao Zhu
Department of Biological Chemistry and Molecular Pharmacology, Harvard Medical School, Blavatnik Institute, Boston, MA, 02115, USA
Ernst W. Schmid
CAS Key Laboratory of Pathogen Microbiology and Immunology, Institute of Microbiology, Chinese Academy of Sciences, Beijing, China
Junhao Zhu

Authors

Harim I. Won
View author publications
You can also search for this author in PubMed Google Scholar
Samuel Zinga
View author publications
You can also search for this author in PubMed Google Scholar
Olga Kandror
View author publications
You can also search for this author in PubMed Google Scholar
Tatos Akopian
View author publications
You can also search for this author in PubMed Google Scholar
Ian D. Wolf
View author publications
You can also search for this author in PubMed Google Scholar
Jessica T. P. Schweber
View author publications
You can also search for this author in PubMed Google Scholar
Ernst W. Schmid
View author publications
You can also search for this author in PubMed Google Scholar
Michael C. Chao
View author publications
You can also search for this author in PubMed Google Scholar
Maya Waldor
View author publications
You can also search for this author in PubMed Google Scholar
Eric J. Rubin
View author publications
You can also search for this author in PubMed Google Scholar
Junhao Zhu
View author publications
You can also search for this author in PubMed Google Scholar

Contributions

H.W., S.Z., J.Z., and E.R. conceived and designed the study; H.W., J.Z., S.Z., O.K., J.S., and MW performed experiments; H.W., J.Z., S.Z., E.S., and I.W. performed data analysis; O.K., T.A., M.C., E.R. gave technical and conceptual advice; H.W. and J.Z. wrote the original draft; H.W., J.Z., S.Z., M.C., I.W., O.K., T.A., J.S., M.W., E.S., and E.R. reviewed and edited the manuscript; J.Z. and E.R. supervised this study.

Corresponding authors

Correspondence to Eric J. Rubin or Junhao Zhu.

Ethics declarations

Competing interests

The authors declare no competing interests.

Peer review

Peer review information

Nature Communications thanks Richard Lee, and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary information

Supplementary Information

Peer Review File

Description of Additional Supplementary Files

Supplementary Dataset 1

Supplementary Dataset 2

Supplementary Dataset 3

Supplementary Dataset 4

Supplementary Dataset 5

Supplementary Dataset 6

Source data

Source Data

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.

Reprints and permissions

About this article

Cite this article

Won, H.I., Zinga, S., Kandror, O. et al. Targeted protein degradation in mycobacteria uncovers antibacterial effects and potentiates antibiotic efficacy. Nat Commun 15, 4065 (2024). https://doi.org/10.1038/s41467-024-48506-8

Download citation

Received: 14 February 2023
Accepted: 03 May 2024
Published: 14 May 2024
DOI: https://doi.org/10.1038/s41467-024-48506-8

Comments

By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.