Zinc starvation uncovers bacterial host-specific proteases that shape NDM adaptability in Acinetobacter baumannii

ABSTRACT

Metallo-β-lactamases, particularly New Delhi metallo-β-lactamase (NDM), threaten antibiotic therapy by disseminating across diverse bacteria. The rise of NDM-producing carbapenem-resistant Acinetobacter baumannii (Ab) highlights the risk of global spread of this pathogen. NDM-mediated resistance depends on periplasmic proteostasis, which regulates the levels of folded, metalated, and active proteins. In Escherichia coli, zinc limitation—common at infection sites—causes loss of the essential metal cofactors in NDM, leading to protein degradation via a specific periplasmic quality control system. Our study reveals that the mechanisms regulating NDM-1 stability under zinc starvation are highly host-dependent. Notably, we identify the proteases DegP and CtpA as responsible for NDM-1 degradation in Ab, differing from the process in E. coli. In-cell stability of NDM in Ab is highly variable depending on minor mutations in CtpA from clinical strains, as well as by mutations in the allelic variants of this β-lactamase. Particularly, NDM-5 displays a higher stability and confers an enhanced resistance phenotype that may help Ab thrive under zinc-limiting conditions. These results reveal selective pressures driving NDM adaptation to each bacterial host. Understanding how pathogens engage their periplasmic metabolism to regulate NDM levels offers insights into the overlooked role of host-specific adaptation of resistance mechanisms. These findings highlight the importance of developing host-directed therapeutic interventions based on the understanding of protein cell homeostasis. In this regard, exploiting host-specific proteases as targeted tools to destabilize resistance enzymes represents a novel therapeutic avenue for curbing the spread of NDM in different pathogens.

IMPORTANCE

The alarming rise of Acinetobacter baumannii producing New Delhi Metallo-β-lactamase (NDM) threatens last-line antibiotic therapies. While β-lactamase dissemination is often accounted for the underlying genetics, the biochemical mechanisms involved in the adaptation of these enzymes within specific bacterial hosts are scarcely known. Here, we show that the stability of NDM differs significantly between A. baumannii and Escherichia coli, due to the varying roles of periplasmic proteases involved in NDM degradation in each host. Variant NDM-5 exhibits enhanced stability and confers increased antibiotic resistance in A. baumannii under zinc-limited conditions (common in infection sites). These findings underscore the role of host-specific proteostasis in shaping the adaptation of resistance determinants and suggest new strategies to combat antibiotic resistance.

INTRODUCTION

Antibiotic resistance is a major global public health threat (1), with the alarming dissemination of carbapenem resistance limiting our current therapeutic options (2). The expression of carbapenemases across different bacterial hosts compromises the efficacy of these potent β-lactam drugs (3–6). The latest report of the World Health Organization has identified carbapenem-resistant Enterobacterales and carbapenem-resistant Acinetobacter baumannii (CRAb) isolates as critical-priority pathogens (7). Metallo-β-lactamases (MBLs) are zinc-dependent enzymes that stand as the largest group of clinically relevant carbapenemases, able to inactivate also penicillins and cephalosporins (8–10). The lack of clinically available MBL inhibitors (11, 12) and the increasing emergence of dual carbapenemase producers (harboring both serine-β-lactamases and MBLs) (13–16) represent a major therapeutic problem, leading to difficult-to-treat or even pan-drug-resistant strains.

CRAb has become a major nosocomial pathogen with high mortality rates (ca. 60% in community-acquired pneumonia and 43.4% in bloodstream infections) (17) exacerbated by its ability to acquire resistance determinants (16, 18–21) and its genetic plasticity, which allows it to withstand variable environmental conditions (22–24). This resistance extends to last-line treatments, including recently approved drugs such as cefiderocol (25–28) and sulbactam/durlobactam (SUL/DUR) (29). While the SUL/DUR combination is effective against Ab strains expressing class D carbapenemases, it lacks activity against MBLs (30). Outbreaks involving MBLs or dual carbapenemase-producing Ab are increasingly reported across Asia, Europe, Africa, and Latin America (17, 31–35). In contrast, MBLs remain relatively uncommon in CRAb isolates from the United States. However, concern is present that this scenario foreshadows a global spread of MBLs in Ab and highlights the urgent need to better understand the adaptation of Ab to host MBLs efficiently.

Clinically relevant MBLs include the soluble periplasmic enzymes VIM, IMP, and SPM, as well as the New Delhi metallo-β-lactamase (NDM) family, which are lipidated enzymes anchored to the inner leaflet of the outer membrane (36). NDM has experienced the widest and fastest worldwide dissemination, with 91 allelic variants identified to date (9, 37–39). Expression of these MBLs is linked to genes in mobile genetic elements that can disseminate across various bacterial species (4, 40, 41). However, MBLs exhibit distinct host specificities that are not fully understood. For instance, NDM enzymes have spread extensively among Enterobacterales and non-fermenters, whereas SPM-1 and VIM-2 are primarily found in Pseudomonas aeruginosa (Pa) (38, 42, 43). In addition, NDM enzymes are unique among MBLs in being able to hydrolyze cefiderocol (44).

Recent work has uncovered that the host specificity of MBLs is related to the proteostasis pathway specific to each enzyme within a given bacterial host (45–50). Despite this progress, a critical mechanistic understanding of this process in different bacterial hosts is missing, hindering our ability to fully leverage the connection between MBL proteostasis and host specificity. The whole proteostasis pathway encompasses (i) protein maturation, including biogenesis, folding, and metal uptake in the periplasm (46, 47), and (ii) degradation, a process that clears misfolded or malfunctioning proteins that impair bacterial fitness (51).

Protein maturation is a key factor influencing the host specificity of MBLs, since the processing efficiency of the signal peptides of these enzymes is highly host dependent (45). NDM-1 is adequately processed across multiple bacterial hosts, including Ab. In contrast, VIM-2 and SPM-1 impose a significant fitness cost when expressed in Ab, since poor signal peptide processing leads to the accumulation of toxic, unprocessed precursors, limiting their dissemination in this host (45, 52). These findings agree with epidemiologic data: a recent analysis of 28,330 Ab isolates available in databases revealed that NDM variants were present in 5.4% of them, compared to only 0.05% harboring the bla_VIM gene, supporting these predictions (17). Additionally, a bioinformatics analysis suggested that bla_NDM is a chimeric gene likely constructed within Ab (53). Recent work also revealed that serine-dependent OXA carbapenemases are membrane-bound in Ab, revealing that this localization favors β-lactamase-mediated resistance in this organism (54). Overall, these observations indicate that Ab is a suitable host for expressing NDM without eliciting significant cellular vulnerabilities.

Knowledge about periplasmic MBL degradation is more scarce. The stability of MBLs in the periplasm depends on the availability of the essential zinc cofactor (36). During an infection, the innate immune response releases neutrophils containing calprotectin, a metal-chelating protein that sequesters zinc ions at the host-pathogen interface (55–57). This reduces the periplasmic zinc levels, leading to the loss of metal ions from MBLs and the resulting accumulation of non-metalated, apo-MBLs (36). These species are unstable in the periplasm, with apo-NDM being degraded in Escherichia coli (Ec) (36, 51) by a specific periplasmic quality control (PQC) system (51, 58). In contrast, the mechanisms governing MBL stability and the corresponding periplasmic PQC systems in Ab remain unexplored, representing a critical knowledge gap to address in combating antibiotic resistance mediated by this pathogen.

Given the rising prevalence of NDM-producing Ab strains, it becomes essential to understand the adaptation mechanisms of this enzyme to this bacterial host by examining NDM proteostasis. Understanding the factors governing the stability of NDM variants in Ab under zinc-limiting conditions is key to accounting for the dissemination of this resistance determinant into CRAb. To address this, we aimed to answer the following questions: (i) How are MBLs, particularly NDM, affected by zinc limitation in Ab? (ii) Are apo-MBLs degraded in Ab during zinc limitation? (iii) Which proteases are involved in this degradation? (iv) How do different NDM variants respond to zinc limitation in Ab?

Inspired by these questions, we found that NDM-1 degradation under zinc limitation is host-dependent. We demonstrate that differences in protein stability between Ab and Ec are driven by distinct host PQC systems mediating its degradation, as well as strain-specific variability within Ab. Clinical NDM variants also displayed a host-dependent behavior, demonstrating that the periplasmic environment imposes unique selective pressures that drive host-specific enzyme adaptation. In particular, NDM-5 is identified as a variant able to overcome the challenge of zinc restriction in Ab. These findings highlight the importance of understanding the bacterial host context when evaluating resistance enzymes. Unveiling these mechanisms is essential for developing more effective, host-adapted strategies to counteract the spread of antibiotic resistance in clinical settings.

RESULTS

Human neutrophils impair NDM-1-mediated resistance in Ab

We first explored the impact of metal restriction on NDM-1-mediated resistance in different bacterial hosts. In previous studies, metal starvation was elicited by using calprotectin or other small molecules as chelators (36). Aiming to mimic physiological conditions, we designed an ex vivo experiment in which we exposed cells of Ab ATCC17978 and Ec ATCC25922 expressing NDM-1 to human neutrophils in the presence of imipenem (IMI) and then evaluated bacterial survival. To assess neutrophil-driven effects unrelated to zinc limitation, we included strains expressing serine-carbapenemases, whose activity does not depend on a bound metal ion. With this aim, we selected KPC-2 for Ec and OXA-24 for Ab (2), which are among the most frequent serine-β-lactamases in clinical strains of these bacteria. Enzymes were expressed using the pMBLeOA vector (designed to express MBLs in different bacterial hosts with their own native signal peptides); isopropyl β-D-thiogalactopyranoside (IPTG)-induced expression was adjusted to produce periplasmic protein levels comparable to those observed in clinical isolates without inducing bacterial fitness, as already reported (36, 45).

Bacteria were exposed to human neutrophils in the presence or absence of sub-inhibitory levels of IMI (Fig. 1A). Bacterial survival was then assessed for each condition by comparing colony-forming units (CFUs) relative to untreated controls (Fig. S1). Based on these data, we estimated the synergistic effect of neutrophils and IMI in decreasing the CFUs. Synergy was not observed in the control strains expressing OXA-24 and KPC-2. Instead, the impact of neutrophils depended on the bacterial host upon expression of NDM-1 (Fig. 1B). Treatment with IMI in the presence of neutrophils showed a strong synergy in Ab expressing NDM-1 that was absent in Ec-NDM-1 (Fig. 1B). This experiment demonstrates that neutrophils have a higher capability of compromising NDM-mediated resistance when expressed in Ab, revealing a bacterial host-specific effect. This can be attributed to the zinc limitation response imposed by neutrophils, as no host-dependent differential response was observed when serine-β-lactamases (whose activities do not depend on zinc ions) were expressed.

Fig 1.

Neutrophil-mediated zinc starvation impairs carbapenem resistance in Ab NDM-1. (A) Bacterial growth was assessed under four conditions in DMEM supplemented with heat-inactivated FBS: (1) untreated (control), (2) exposed to sub-inhibitory concentrations of IMI, (3) incubated with neutrophils alone (Neutrophils), or (4) treated with both neutrophils and IMI (Neutrophils + IMI). (B) CFUs were quantified for each condition to evaluate bacterial viability (Fig. S1). Synergy between neutrophils and IMI against the different bacteria was calculated as Δlog CFUs (Neutrophils + IMI) – [Δlog CFUs (Neutrophils) + Δlog CFUs (IMI)]. The evaluated bacteria were Ab ATCC 17978 expressing either NDM-1 or OXA-24 and Ec ATCC 25922 expressing NDM-1 or KPC-2. β-lactamase expression was induced with IPTG: 20 µM for NDM and 100 µM for serine β-lactamases (OXA-24 and KPC-2), representing the lowest IPTG concentrations that maximize MIC values without inducing toxicity. Experiments were conducted with neutrophils isolated from at least four independent donors. Gray circles represent individual blood donors. Error bars indicate SD. Asterisks indicate statistically significant synergy values different from zero **P < 0.01; ns: not significant differences (P ≥ 0.05).

The impact of zinc limitation on MBL-mediated resistance is host-dependent

Based on these initial results, we next explored the effect of zinc limitation in different bacterial hosts expressing several MBLs. To this end, we selected four clinically relevant MBLs—NDM-1, VIM-2, IMP-1, and SPM-1—and three model strains of priority pathogens (7): Ab ATCC 17978, Ec ATCC 25922, and Pa PAO1.

Zinc limitation was induced in these experiments using an approach that enables multiple and more challenging experiments, that is, adding dipicolinic acid (DPA). DPA is a small metal-chelating molecule that mimics the effect of the metal scavenging protein calprotectin present in neutrophils and does not impair bacterial growth (36). MBLs were expressed using the pMBLeOA plasmid in all strains, as described above.

We assessed the resistance phenotype of the three bacterial strains expressing each of the four selected MBLs by measuring ceftazidime’s minimum inhibitory concentration (MIC) under increasing concentrations of DPA. This cephalosporin usually displays larger MIC values than carbapenems, enabling us to sample a larger dynamic range of antibiotic concentrations. In line with previous observations in Ec DH5α (36), MIC values of ceftazidime against MBL-expressing bacteria decreased as zinc availability was restricted in all cases (Fig. S2). However, the impact of zinc limitation on MBL-mediated resistance depended on the bacterial host (Fig. 2A). Among the three pathogens, MBL-producing Ab was the most susceptible host, exhibiting the largest reduction in MIC ratio upon zinc depletion, compared to Pa and Ec, which displayed comparable responses (Fig. 2A; Fig. S2). These data support the results with neutrophils and validate the use of DPA for these experiments.

Fig 2.

MBLs exhibit host-dependent behavior under zinc limitation. (A) Percentage of resistance retained under zinc deprivation across different bacterial hosts, calculated from the decrease in Ceftazidime MIC ratio curves after exposure to varying concentrations (µM) of the zinc chelator DPA (Fig. S2) for Ec ATCC 25922 (Ec, purple), Pa PAO1 (Pa, blue), and Ab ATCC 17978 (Ab, pink), each expressing VIM-2, NDM-1, IMP-1, or SPM-1. Data represent the mean of three independent replicates; gray circles represent each individual replicate; error bars indicate standard deviations (SD). Asterisks denote statistically significant differences from Ab, ***P < 0.001, ****P < 0.0001. (B) Stability of MBLs under zinc limitation in different bacterial hosts. MBL levels (% of remaining enzyme) over time after the addition of 200 µM DPA to Ec (purple), Pa (light blue), and Ab (pink) expressing the MBLs IMP-1, SPM-1, VIM-2, and NDM-1. Protein levels were quantified by WB, normalized to untreated controls, and expressed as the percentage of the initial protein level remaining after DPA treatment at each time point. Data were fitted to an exponential decay model. Values represent the mean of three independent biological replicates (solid circles), with individual replicates shown as empty circles; error bars indicate SD. Asterisks indicate statistically significant differences, ***P < 0.001, ****P < 0.0001 ; ns: not significant differences (P ≥ 0.05).

SPM-1 and IMP-1 were the two MBLs most refractory to the impact of zinc limitation across all three examined hosts. In contrast, VIM-2 and NDM-1 were the most susceptible ones. These findings indicate that the impact of zinc limitation varies significantly depending on both the specific MBL and the bacterial host.

The in-cell stability of apo-NDM is host-dependent

The reduced MIC values in the presence of an external zinc scavenger are due to the dissociation of the metal ions from the active sites of the MBLs, which render them inactive (36, 59). During this process, the stability of MBLs is highly affected in the bacterial periplasm. The metal-depleted form of MBLs (apo-MBLs), despite being stable in vitro, is prone to aggregation or degradation by periplasmic proteases in Ec (36, 51). In this regard, we hypothesized that the stability of each apo-MBL could be different among distinct bacterial hosts.

To address this, we investigated the in-cell stability of the four studied MBLs (NDM-1, VIM-2, IMP-1, and SPM-1) by following the time-dependent levels of these proteins in the periplasm of Ec, Ab, and Pa upon zinc restriction. We resorted to the same approach described, using the expression plasmid pMBLeOA. Expression levels were optimized for each protein and bacterial host by adjusting the concentration of added IPTG to obtain comparable initial protein levels (Fig. S3). All MBLs were linked to a C-terminal Strep-tag, which allows uniform quantification by immunoblotting without affecting the protein localization or the resistance phenotype (36). MBL stability was assessed by quantifying cell protein levels by Western blotting (WB) at various time points after inducing metal deprivation by DPA. For membrane-bound NDM-1, detection was performed in whole cells, while for the soluble enzymes, protein quantitation was performed in the corresponding periplasmic fractions.

Both IMP-1 and SPM-1 were stable upon zinc depletion over 2 hours in all three bacteria (Fig. 2B; Fig. S4). In contrast, VIM-2 and NDM-1 showed a decay in protein levels within this time frame, disclosing a reduced stability across all tested bacterial hosts. In addition, the stability of each protein was strongly dependent on the bacterial species (Fig. 2B; Fig. S4). For instance, VIM-2 presents half-lives (t_₁/₂) of 20 min in Ab, 27 min in Ec, and 64 min in its frequent host, Pa (Fig. S4E). Notably, NDM-1 exhibited the greatest sensitivity to zinc limitation in Ab, being eight to nine times less stable in this pathogen (t_₁/₂: 5 min) compared to Ec (t_₁/₂: 38 min) and Pa (t_₁/₂: 45 min) (Fig. S4E). Comparable results were observed using an alternative, non-cell-permeable metal chelator (Fig. S5). To discard the eventual impact of DPA in gene expression and/or protein synthesis, we performed similar experiments inhibiting protein synthesis with chloramphenicol. The finding of similar NDM-1 levels under these conditions (Fig. S6) confirms that the time-dependent decay of protein levels is due to degradation.

Overall, these experiments show that MBL stability under zinc-limiting conditions is host-dependent. Based on the low stability observed for NDM-1 in Ab under zinc depletion, we decided to study the underlying mechanisms.

The host-specific proteases CtpA and DegP are responsible for NDM-1 degradation in Ab

Apo-NDM-1 is degraded in Ec by a specific PQC system that involves the concerted action of two periplasmic proteases: the carboxy-terminal protease Prc and the serine endopeptidase DegP (51). Prc is the main protease in this process, recognizing specific residues and secondary structure elements in apo-NDM-1, while DegP further degrades the peptides released by the action of Prc (58). In Ab ATCC 17978, homologs of Prc (Locus A1S_0493) and DegP (Locus A1S_2525) share 32% and 31% sequence identity, respectively, with the corresponding proteins in Ec (Fig. S7). To investigate their role in NDM-1 degradation, we constructed knockout mutants of these genes in Ab and monitored the time-dependent stability of NDM-1 under zinc-limiting conditions. IPTG concentration was calibrated to ensure comparable initial levels of NDM in each strain (Fig. S8).

Inactivation of Prc in Ab (Ab prc⁻) did not have an impact on NDM-1 stability (Fig. 3; Fig. S9), revealing that, unlike in Ec, Prc is not involved in NDM-1 degradation in Ab. In contrast, inactivation of DegP (Ab degP⁻) led to a substantial increase in NDM-1 stability, with a fivefold longer half-life (t_₁/₂) in the periplasm compared to the wild-type (WT) strain (Fig. 3; Fig. S9C). The absence of DegP, however, did not totally abolish NDM-1 degradation. Given that DegP is a housekeeping protease that may lack specificity, we looked for the presence of additional proteases involved.

Fig 3.

Stability of NDM-1 in Ab and Ec protease mutants under zinc deprivation. (A) Relative NDM-1 level (% of remaining enzyme) at 15 min after treatment with 200 µM DPA in Ab ATCC 17978 WT, or mutants lacking proteases: prc⁻, degP⁻, ctpA⁻, and ctpA⁻degP⁻. (B) Relative NDM-1 level (% of remaining enzyme) in Ec DH5α WT, or mutants lacking proteases: Δprc, ΔdegP, and ΔdegPΔprc at 10 min after treatment with 1 mM DPA, extracted from Gonzalez et al., (2023) (51). IPTG induction was calibrated to ensure equal initial NDM-1 expression across all Ab strains (Fig. S8). Protein levels were quantified by WB, normalized to untreated controls, and expressed as the percentage of the initial NDM-1 remaining. Data represent the mean of three independent experiments; black circles represent individual replicates; error bars indicate SD. Asterisks denote statistically significant differences from Ab WT/ Ec WT, **P < 0.01, ***P < 0.001, ****P < 0.0001, ns: not significant differences (P ≥ 0.05). (C) Schematic representation of the PQC systems responsible for NDM-1 degradation in Ec and Ab under zinc-limited conditions in the periplasmic space. Under zinc-replete conditions, the enzyme is metalated and stable. Upon zinc limitation, NDM-1 becomes demetallated (apo-NDM), exposing its flexible C-terminal region and triggering degradation. In Ec, apo-NDM is primarily degraded by the periplasmic protease Prc, and the resulting peptide fragments are further processed by DegP. In Ab, both CtpA and DegP contribute to NDM-1 degradation through a yet-uncharacterized mechanism. OM: outer membrane; IM: inner membrane.

Ab encodes an additional periplasmic carboxy-terminal protease, CtpA (locus A1S_0231), absent in Ec. This protease is 288 amino acids smaller than Prc (660 amino acids), with a 16% sequence identity (Fig. S7). Inactivation of CtpA by introduction of a premature stop codon (Ab ctpA⁻) resulted in increased NDM-1 stability, with a fourfold longer half-life compared to the WT strain (Fig. 3; Fig. S9C). To assess the combined effect of the two proteases, we constructed the double mutant Ab ctpA⁻degP⁻. This strain exhibited a remarkable increase in NDM-1 stability relative to the WT strain, with a 16-fold longer half-life, nearly abolishing enzyme degradation (Fig. 3; Fig. S9C). To further support the activity of these proteases toward NDM-1, we performed a heterologous complementation experiment. Expression of CtpA and DegP from Ab in an Ec DH5α ΔprcΔdegP background restores NDM degradation, confirming the role of these proteases in degrading NDM-1 (Fig. S10).

These findings disclose significant differences in the mechanism of NDM-1 degradation between Ec and Ab, such as the replacement of Prc by CtpA, and a higher involvement of housekeeping protease DegP (Fig. 3). We conclude that the host-specific stability in the periplasm can be accounted for by identifying the PQC system in each bacterium.

Adaptation of NDM variants reveals a selective advantage of NDM-5 under zinc starvation in Ab

The molecular features endowing IMP-1 and SPM-1 with a substantially higher stability than NDM-1 in Ab are difficult to pinpoint due to the low sequence identity (8% and 20%, respectively) between these MBLs. However, the study of allelic variants enables the identification of mutational hotspots that can tune the stability towards the periplasmic proteolytic machinery. Clinical NDM variants usually differ by two to three mutations to NDM-1 and are limited to a reduced number of residues outside the active site, thus representing an ideal system to identify sequence determinants of stability (59). Figure 4A depicts the variants and substitutions of interest studied here.

Fig 4.

NDM clinical variants exhibit distinct stability and resistance profiles in Ab under zinc starvation. (A) Crystal structure of NDM-1 (PDB: 5zgx) highlighting the mutations present in the studied variants: V88L (green), M154L (violet), and A233V (yellow), and summary table of the analyzed NDM variants, indicating their specific mutations, color-coded to match the structure. (B) Stability of NDM variants over time following zinc chelation with 500 µM DPA in Ab ATCC 17978. The curves represent the following variants: NDM-1 (pink), NDM-4 (purple), NDM-5 (blue), NDM-6 (yellow), NDM-15 (orange), and NDM-24 (green). NDM expression was induced with 100 µM IPTG; protein levels were quantified by WB, normalized to untreated controls, and expressed as the percentage of the initial protein level remaining after DPA treatment at each time point. Data were fitted to an exponential decay. Values represent the mean of three independent biological replicates (solid circles), with individual replicates shown as empty circles; error bars indicate SD. Asterisks indicate statistically significant differences from NDM-1, ****P < 0.0001; ns: not significant differences (P ≥ 0.05). (C) Percentage of resistance retained under zinc deprivation calculated from the decrease in cefotaxime MIC ratio curves after exposure to varying concentrations (µM) of the zinc chelator DPA (Fig. S11) for Ab expressing different NDM variants: NDM-1 (pink), NDM-4 (purple), NDM-5 (blue), NDM-6 (yellow), NDM-15 (orange), and NDM-24 (green) with 100 uM IPTG. Data represent the mean of three independent biological replicates; black circles represent individual replicates; error bars indicate SD. Asterisks denote statistically significant differences compared to NDM-1, ***P < 0.001; ns: not significant differences (P ≥ 0.05).

Substitution M154L (present in NDM-4) enhances the zinc affinity of NDM, therefore improving the resistance phenotype under metal starvation (59, 60). While this variant did not show an increased stability in Ec, variants containing other substitutions (generally combined with M154L) displayed enhanced stability toward the PQC system of Ec, such as NDM-6 (A233V), NDM-24 (V88L), NDM-5 (V88L, M154L), and NDM-15 (M154L, A233V) (59). We wondered whether differences in the PQC systems of distinct bacterial hosts could influence the adaptation of NDM variants. To address this, we tested the behavior of NDM variants in Ab by assessing their resistance and stability under DPA-imposed zinc-limiting conditions.

The resistance phenotype upon zinc starvation in Ab when expressing these variants revealed that NDM-5 is clearly the most refractory to this perturbation (Fig. 4C; Fig. S11). Expression of some of these variants elicited a fitness cost in Ab under permissive conditions (without antibiotics), as revealed by the growth curves, with NDM-5 being the variant that least affected bacterial growth (Fig. S12). This contrasts with the situation in Ec, where the expression of NDM variants did not affect bacterial fitness (Fig. S12). When evaluating stability, all variants exhibited significantly longer t1/2 values than NDM-1 in Ab, with 7- to 22-fold increases (Fig. 4B; Fig. S13D). Notably, the M154L substitution (NDM-4), which does not have significant effect on enzyme stability in Ec (59), prolonged the protein lifetime by sevenfold in Ab (Fig. 4B; Fig. S13D). This reveals that mutations may be better suited to escape or better resist the action of the periplasmic PQC system in specific bacterial hosts, reflecting host-driven evolutionary pressures. Among the clinical variants studied under zinc limitation, NDM-5 demonstrated both the highest MIC values (Fig. 4C) and greatest stability in Ab (Fig. 4B). Notably, while this variant showed the highest expression levels (Fig. S13A), it maintained the lowest fitness cost among all tested alleles (Fig. S12), demonstrating an efficient adaptation to zinc-deprived conditions in this bacterial host.

NDM stability varies among Ab strains

To assess whether intraspecies variation could affect the stability of NDM in the periplasm, we examined how different Ab backgrounds influence the behavior of NDM-1 under zinc-limiting conditions. Specifically, we used two clinical isolates: AB5075, transformed with the same plasmid used in the reference strain (pMBLeOA_bla_NDM-1), and AMA40, which naturally harbors bla_NDM-1. We observed notable differences in NDM-1 stability among these Ab strains. In AB5075, the enzyme’s half-life increased ninefold compared to ATCC 17978 (Fig. 5; Fig. S14). In contrast, the stability of apo-NDM-1 in AMA40 was similar to that observed in the reference strain (Fig. 5; Fig. S14). These findings reveal that different genetic backgrounds in clinical Ab strains can significantly influence NDM degradation.

Fig 5.

Strain-dependent differences in NDM-1 stability in Ab. Stability analysis of NDM-1 over time following Zn chelation with 500 µM DPA in Ab ATCC 17978 (pink), Ab AB5075 (brown), and Ab AMA40 (gray). Protein levels were quantified by WB, normalized to untreated controls, and expressed as the percentage of the initial protein level remaining after DPA treatment at each time point, fitted to an exponential decay. Values represent the mean of three independent biological replicates (solid circles), with individual replicates shown as empty circles; error bars indicate SD. Asterisks indicate statistically significant differences from Ab ATCC 17978 ***P < 0.001; ns: not significant differences (P ≥ 0.05).

Sequence analysis of the proteases among these Ab strains revealed that DegP is fully conserved, whereas CtpA displays amino acid substitutions in the signal peptide region in both AB5075 and AMA40, as well as additional changes in the N-terminal low-complexity region (LCR) of the mature protein in AMA40 (Fig. S15). These subtle differences in the PQC components could contribute to the observed variation in NDM stability across Ab strains.

DISCUSSION

The global dissemination of MBLs continues to compromise antimicrobial therapies, with NDM emerging as the most widespread and clinically relevant resistance determinant (38, 39). Unlike other MBLs such as SPM-1 or VIM, which remain restricted mainly to Pa, NDM has successfully spread across a broad range of bacterial hosts, including Klebsiella pneumoniae, Ec, Pa, and Ab (38, 61). This adaptability of NDM has been attributed to the optimal expression and efficient maturation process of this enzyme in different hosts without imposing a significant fitness cost (45). However, environmental conditions can challenge the resistance phenotype of properly folded and metalated MBLs. This is the case of zinc limitation elicited by the immune system response, which gives rise to the inactive apo-MBL forms, unable to confer resistance (36, 55, 62). Zinc starvation not only leads to enzyme inactivation, since the bacterial metabolism can be negatively affected by the accumulation of apo-MBL species prone to aggregation, which may elicit cellular vulnerabilities. To counteract this stress, bacteria rely on PQC systems to eliminate these unstable apo-forms (36, 49, 51, 52), at the same time affecting antibiotic resistance by impeding Zn(II) rebinding and restoration of enzyme activity. Therefore, understanding the whole process of MBL proteostasis in the periplasm is relevant to designing host-specific therapeutic approaches.

Herein, we show that the resistance phenotype of MBL producers under zinc limitation depends on the bacterial host. Specifically, Ab is more sensitive to the metal scavenging action of neutrophils than Ec when expressing NDM-1. Upon adding a chelating agent, MBLs are demetallated and therefore inactivated, as revealed by the decrease in MICs of three different bacteria expressing four MBLs. However, the time evolution of apo-protein levels under these conditions revealed different in-cell protein stabilities, depending on both the protein and the bacterial host. Specifically, NDM-1 exhibited a significantly reduced stability in its apo form within the Ab periplasm compared to its behavior in Ec and Pa. In contrast, other MBLs, such as IMP and SPM, maintained high stability across all three pathogens.

Apo-NDM-1 is degraded in the periplasm of Ec by a PQC system that involves the concerted action of two C-terminal proteases, Prc and DegP (51). Prc recognizes the disordered C-terminus of apo-NDM-1 and releases peptide fragments that are cleaved by the housekeeping protease DegP, giving rise to shorter fragments (58). In contrast, Prc does not play a significant role in the degradation of apo-NDM-1 in Ab. Instead, DegP is more relevant in this bacterium, acting in concert with CtpA, a host-specific protease not present in Ec. This reveals that PQC systems vary in composition and relative contribution of each component, accounting for the different stability of NDM-1 in the two hosts. While our in vivo and heterologous complementation data support the role of CtpA and DegP in NDM-1 proteolysis, evidence of direct cleavage would require future in vitro assays with purified proteins.

The stability of apo-NDM-1 also varies significantly among Ab strains that share the same PQC system with minor sequence variations. CtpA carries mutations in the signal peptide in strains AB5075 and AMA40 (Fig. S15), which could affect the efficiency of protein secretion, and therefore, the periplasmic protein levels compared to the reference strain. Additionally, AMA40 presents substitutions in the N-terminal stretch of the mature protein, known as the LCR. At the moment, CtpA from Ab has not been characterized biochemically, precluding a proper molecular understanding of the role of this region. However, inspection of the crystal structure of the homologous CtpA from Pa suggests that this domain is potentially involved in the interaction with membrane partners (63). Therefore, mutations in this hitherto unexplored region could affect substrate recognition by CtpA. In any case, these findings suggest that minor substitutions in the PQC components can substantially impact their efficiency. As a result, despite apo-NDM-1 being more susceptible to degradation in the reference strain, it can be more tolerant to zinc starvation in some clinical isolates.

The susceptibility of apo-NDM to degradation in the periplasm is due to the flexibility of the C-terminal stretch of this protein upon zinc removal, which is recognized and cleaved by Prc in Ec (51). Substitution A233V (present in NDM-6) quenches this flexibility, therefore impairing degradation in Ec (51, 59). Since A233V is also stabilizing in Ab, we conclude that the susceptibility of apo-NDM variants to the PQC system in Ab depends also on the flexibility of the C-terminal region, consistent with the C-terminal recognition previously shown for the homologous Pa CtpA (64).

NDM variants are currently evolving to overcome the impact of zinc starvation, either by improving the zinc binding affinity or by resisting proteolysis (51, 59). Here, we found intriguing differences in the stability of specific variants between the two hosts. The substitution M154L is only responsible for an increased zinc affinity in Ec (59) but stabilizes the apo-protein in Ab against proteolysis, as evidenced by the higher stability of NDM-4 and NDM-15, revealing a synergistic action of M154L and A233V (Fig. 4). The host-dependent pleiotropic effect of M154L could account for its high prevalence among NDM alleles (48%). This is particularly relevant because enhanced stability has been shown to allow NDM variants to better sustain resistance under metal limitation, likely by improving their ability to compete for zinc (59). Finally, NDM-5 (V88L M154L) is the most stable variant in Ab, likely due to the additive effect of two stabilizing mutations. The recent widespread dissemination of NDM-5 among Enterobacterales (65–67), which in some regions prevails over NDM-1 (68), suggests that this variant could easily adapt to Ab based on its high stability and high resistance levels upon zinc limitation, together with a lower fitness cost. The high number of NDM allelic variants (91) that harbor M154L, V88L, and A233V combined with other substitutions makes it compelling to study the impact of how these combinations impact the tolerance to zinc starvation.

Our findings reveal that resistance cannot be fully understood outside the context of the bacterial host and its stress physiology. PQC systems vary across bacterial species and strains, creating host-specific selective pressures that influence NDM stability and potentially drive its adaptation. The elucidation of how distinct pathogens engage their PQC machinery to modulate NDM levels under zinc-limited conditions provides a conceptual framework to understand the enzyme’s adaptation to diverse bacteria. These insights deepen our understanding of the clinical evolution of MBLs and open the door to exploiting host-specific proteases as targeted tools to destabilize resistance enzymes.

We conclude that β-lactamases from the NDM family display a remarkable plasticity to tolerate zinc starvation across distinct periplasmic environments, facilitating their adaptation to diverse bacterial hosts. Given the clinical concern posed by NDM expression in Ab and the marked strain dependence of their behavior under zinc-limiting conditions typical of an infection site, our findings underscore the importance of tailoring therapeutic strategies based on the specific bacterial host. In the case of Ab, a highly adaptable and genomically plastic pathogen, such precision is essential to prevent further dissemination of bla_NDM genes in clinical settings. The increasing prevalence of NDM in addition to other OXA carbapenemases raises the concern that our future therapeutic pipeline may not be adequate to treat MDR Ab infections, particularly in settings with limited infection control measures. In addition, the insights obtained examining NDM proteostasis under zinc starvation across bacterial hosts reveal new opportunities to develop targeted therapies that exploit host-specific MBL vulnerabilities.

MATERIALS AND METHODS

Bacterial strains and growth conditions

The bacterial strains used in this study included Ab ATCC 17978, Pa PAO1 (non-fermenting organisms), and Ec ATCC 25922 (Enterobacteriaceae). For each organism, a set of isogenic clones was constructed by transformation with the plasmid pMBLeOA, either empty or carrying different WT MBL genes (bla_NDM-1, bla_VIM-2, bla_SPM-1, and bla_IMP-1). Additionally, Ab ATCC 17978 was transformed with pMBLeOA-bla_OXA-24 and pMBLeOA carrying bla_NDM variants (NDM-4, NDM-5, NDM-6, NDM-15, and NDM-24), which were generated as described below. Ec DH5α harboring pMBLe plasmids expressing the aforementioned NDM variants were used for growth curve comparisons with Ab. Also, Ab AB5075 ΔOXA-23 was used to express pMBLeOA-bla_NDM-1, while Ec ATCC 25922 was also transformed with pMBLeOA-bla_KPC-2. Cells were routinely grown aerobically at 37°C in Luria-Bertani (LB) broth (Difco) or on LB agar plates, unless otherwise specified. When necessary, gentamicin (Sigma-Aldrich) was added at a final concentration of 20 µg/mL to maintain plasmid selection.

Strains and plasmid construction

Plasmid isolation, DNA purification, restriction enzyme digestion, and ligation were performed following standard protocols (69). Reagents for purification were from Promega, while other reagents and chemicals were from Sigma-Aldrich. Oligonucleotides and enzymes were sourced from Macrogen and Thermo Fisher Scientific, respectively. The pMBLeOA plasmid, which contains an origin of replication compatible with Ab (45), was used to express the different β-lactamases. This vector allows the expression of full-length β-lactamase genes, including their native signal peptides, under the control of an IPTG-inducible pTac promoter (36). Additionally, it enables the expression of β-lactamases fused to a C-terminal Strep-tag II, which facilitates standardized protein quantification by Western blot. MBL expression was induced with low IPTG concentrations (5–80 µM) to avoid overexpression artifacts and better mimic β-lactamase levels observed in clinical strains (36). The coding sequences for these enzymes had been previously cloned into pMBLe. Therefore, each bla gene was subcloned into pMBLeOA using BamHI and HindIII restriction sites. Constructs were verified by PCR using plasmid-specific primers that anneal outside the multiple cloning site (MCS) (pMBLe_Fw and pMBLe_Rv [Table S1]). Bacterial strains were transformed by electroporation with either the empty vector or pMBLeOA carrying the different MBLs, as previously described (70).

Generation of Ab mutants

To obtain inactive protease mutants in Ab, two approaches were used: the suicide vector technique with the plasmid pET-28a + for prc and degP and CRISPR-Cas with the modified plasmid pMBEC6 for ctpA due to the failure of the former approach. For the suicide vector technique, the protocol by Aranda et al. (71) was followed, using the plasmid pET-28a⁺, non-replicative in Ab. Internal fragments of prc and degP (amplified from Ab by PCR using PrcInt_Fw_HinIII/PrcInt_Rv_BamHI and DegPInt_Fw_HinIII/DegPInt_Rv_BamHI [Table S1]) were cloned into this vector via BamHI and HindIII. The constructs pET-28a⁺-prc and pET-28a⁺-degP were verified by PCR using T7_termiantor_Fw/T7_Promotor_Rv (Table S1), which hybridize outside the MCS, followed by sequencing. These plasmids were used to transform Ab. Since pET-28a⁺ is non-replicative in this strain, homologous recombination led to the disruption of prc and degP. Mutants were selected on LB agar supplemented with the plasmid selection marker kanamycin (20 µg/mL, Sigma-Aldrich) and confirmed by PCR using T7_Promotor_Rv + PrcExt_Fw/DegPExt_Fw (upstream of the genes) and T7_Terminator_Fw + PrcExt_Rv/DegPExt_Rv (downstream of the genes) (Table S1). Finally, following the previously described protocol, Ab prc⁻ and degP⁻ mutants were transformed with pMBLeOA and pMBLeOA_bla_NDM-1.

A CRISPR-Cas9-based base editing approach was applied to introduce a premature stop codon into ctpA in Ab, using the pMBEC6 vector as described by Volke et al. (72). Since this plasmid is non-replicative in Ab, a replication origin compatible with this species (OA) was PCR-amplified from pMBLeOA using primers “Fw_OA_CP” and “Rv_OA_CP” (Table S1) and cloned into pMBEC6 via SalI and XmaI restriction sites to generate pMBEC6_OA. To target ctpA, a spacer sequence was designed using the CRISPy web service (73), based on the Ab ATCC 17978 reference genome (GenBank CP000521) (74). The selected spacer (5′-ACTCATTCAATATACTGAAG-3′) directed the introduction of a stop codon at position 106, replacing the native glutamine codon. The synthetic spacer, included in the “CtpA_CRISPR_Fw” primer (Table S1), was cloned into pMBEC6_OA via Golden Gate assembly (75), following Volke et al.’s protocol (72). The final construct was verified by Sanger sequencing. Ab was electroporated with the resulting plasmid (pMBEC6_OA_ctpA). Cells were incubated in LB for 2 h at 37°C with agitation (200 rpm) to allow recovery, and then 100 µL of culture was inoculated into 10 mL LB supplemented with 30 µg/mL gentamicin and incubated overnight. To cure the editing plasmid, 100 µL of culture was transferred into 10 mL LB containing 10% (wt/vol) sucrose and incubated overnight. Serial dilutions were plated on LB agar to isolate individual clones. The presence of the ctpA mutation was confirmed by sequencing using primers “CtpAExt_Fw” and “CtpAExt_Rv” (Table S1), which bind outside the edited region. One confirmed Ab ctpA⁻ mutant clone was subsequently transformed with pMBLeOA and pMBLeOA_bla_NDM-1, following the previously described transformation protocol. To generate the double mutant Ab ctpA^−*degP⁻, the CRISPR-Cas methodology was applied using the plasmid pMBEC6_OA_ctpA on the Ab degP⁻ strain, following the same procedure described above. The mutation was verified as outlined previously.

MIC determinations

MIC determinations were performed in LB medium using the agar macrodilution method according to the Clinical and Laboratory Standard Institute guidelines (76). To assess the impact of zinc availability on antibiotic resistance, the growth medium was supplemented with varying concentrations of the metal chelator DPA (Merck, >98%). The MIC ratio was calculated as follows: MIC ratio = [(MIC_MBL + DPA – MIC_control + DPA) / (MIC_MBL 0 µM DPA – MIC_control 0 µM DPA)]. The different DPA concentrations used are listed (50, 100, 150, 200, 350, 500, and 750 µM). In this equation, MIC_MBL + DPA and MIC_control + DPA represent the MIC values for bacteria carrying plasmids pMBLeOA-bla or pMBLeOA, respectively, under each chelation condition, while MIC_MBL 0 µM DPA and MIC_control 0 µM DPA refer to measurements without DPA supplementation. For NDM-1, VIM-2, SPM-1, and IMP-1 in Ab, Ec, and Pa, the MICs were tested against ceftazidime (Sigma-Aldrich). For NDM variants in Ab, MICs were tested against cefotaxime (Sigma-Aldrich). The area under the curve (AUC) of the MIC ratio across all DPA concentrations for each enzyme-bacterial host combination was calculated and normalized to the AUC of a theoretical curve representing full resistance (i.e., a constant MIC ratio across all DPA concentrations), and the result was expressed as the percentage of resistance retained.

Impact of human neutrophils on MBL-mediated antibiotic resistance

Peripheral blood was obtained from healthy adult human donors via venipuncture and collected in anticoagulated tubes. Neutrophils were isolated by centrifugation on Ficoll-Paque, followed by dextran sedimentation and hypotonic lysis. Purity of neutrophil preparation was over 98%, and the cells were used immediately after isolation. Ab and Ec strains were diluted 1:100 from overnight LB cultures and grown for 2 hours. β-lactamase induction was then assessed by adding 20 µM IPTG for NDM or 100 µM IPTG for serine-β-lactamases, and the cultures were incubated for an additional time of 2 hours. These cultures were washed with 0.9% (wt/s) NaCl and resuspended in Dulbecco’s modified Eagle medium (DMEM) (Gibco) medium supplemented with 10% fetal bovine serum (FBS) (Gibco), previously inactivated by heating at 65°C for 30 min. Subsequently, 10⁵ CFU were incubated alone or co-incubated with neutrophils at a concentration of 10⁶ cells for 20 min. Imipenem (IMI) was then added at sub-inhibitory concentrations (sub-inhibitory MICs for each strain in DMEM-FBS media: 0.5 µg/mL for Ab (NDM-1 or OXA-24), 3 µg/mL for Ec NDM-1, and 0.125 µg/mL for Ec KPC-2) when appropriate, and the cultures were incubated for an additional 1 hour and 40 min. Finally, CFU/mL were counted in each condition (neutrophils alone, IMI alone, or the combination Neutrophils + IMI) and synergy between neutrophils and IMI was calculated as Δlog CFUs (Neutrophils + IMI) – [Δlog CFUs (Neutrophils) + Δlog CFUs (IMI)].

Periplasm isolation

Extraction of periplasmic proteins was performed as previously described (47). Briefly, 2–3 mL of Ec, Pa, or Ab pMBLeOA-bla_{VIM-2/IMP-1/SPM-1} cultures were pelleted, and cells were washed once with 20 mM Tris, 150 mM NaCl, pH 8.0. The washed cells were resuspended in 20 mM Tris, 0.1 mM EDTA, 20% wt/vol sucrose, 1 mg/mL lysozyme (from chicken egg white, Sigma-Aldrich, protein ≥ 90%), 0.5 mM PMSF, pH 8 (resuspension volume was normalized according to the formula V = 100 µL × OD₆₀₀ × Vc, where Vc is the starting volume of culture sample), incubated with gentle agitation at 4°C for 30 min, and finally pelleted, with the periplasmic extract in the supernatant.

MBL detection and stability assay

The bacterial strains were grown at 37°C until reaching an OD₆₀₀ of 0.4. At this moment, the expression of the different MBLs was induced by adding IPTG (concentration specified for each experiment). After 2 hours of induction at 37°C, the cultures were divided into two equal portions: one was treated with DPA (concentration specified for each experiment), while the other served as an untreated control. Both cultures were incubated at 37°C, and samples were collected at different time points. Cell fractionation was performed to isolate the periplasmic fraction for soluble MBLs for protein analysis, while total cell lysates were used for NDMs. MBL protein levels were assessed by SDS-PAGE followed by WB using Strep-Tag II monoclonal antibodies (Sigma-Aldrich) at a 1:1,000 dilution (from a 200 µg/mL stock). Exceptionally, WB for AB5075 was developed using chemiluminescence, throwing an anti-Strep-Tag II antibody conjugated with horseradish peroxidase. Protein band intensities were quantified from PVDF membranes using Gel Analyzer software. The values from the DPA-treated cultures were normalized to those of the untreated control at each time point, and the resulting ratio was multiplied by 100 to determine the percentage of remaining protein over time. As loading controls for WB, the cytoplasmic protein GroEL was used for total cell lysates, while the periplasmic maltose-binding protein (MBP) was used for periplasmic extracts. In the case of Ab, SDS-PAGE of extracted periplasmic proteins served as a sample concentration control, as this organism lacks an MBP homolog to that of Ec.

Inhibition of protein synthesis under zinc-limiting conditions was achieved by adding chloramphenicol (200 µg/mL) to the cultures immediately after the induction period and before culture division for DPA (200 µM) treatment. Aliquots were processed 30 min after treatment, following the same procedure described above.

Heterologous complementation assay

The Ab ATCC 17978 genes ctpA and degP were obtained as synthetic genes cloned into the pUC vector (GeneUniveral) and subsequently subcloned into the plasmid pBAD under an arabinose-inducible promotor, using the restriction enzymes HindIII and BamHI, yielding the pBAD_degP_Ab and pBAD_ctpA_Ab constructs. These plasmids, including the empty vector pBAD, were then transformed into Ec DH5α ΔprcΔdegP carrying pMBLe_OA_bla_NDM-1.

To analyze NDM-1 stability in this background, cultures were treated as described above, with the following modifications during the induction step: IPTG (20 µM) was added to induce NDM-1 expression from pMBLeOA, and L-arabinose was added to induce protease expression from pBAD at comparable levels (0.005% for CtpA and 0.01% for DegP), as well as in the control with the empty vector. After 2 hours of induction, cultures were divided into equal portions, with one treated with 500 µM DPA and the other left as untreated control. NDM-1 stability was assessed 120 min later by WB, as described above.

Growth curves

The effect of MBL expression on bacterial growth was assessed by measuring cell density at OD_₆₀₀ every 30 min using a BioTek Synergy two multimode microplate reader for 15 hours at 37°C. All bacterial strains were grown in LB broth supplemented with gentamicin at 37°C and 200 rpm. Overnight cultures were diluted 1:100 in LB and inoculated in duplicate into a sterile 96-well microplate (Greiner Bio-One) containing gentamicin (20 µg/mL), with or without IPTG induction, as indicated. Data represent the average of at least three independent experiments.

Multiple sequence alignment

Protein sequences of proteases were aligned using CLUSTAL 2.1 (77) and analyzed with ESPript 3.0 (78).

Statistical analysis

All statistical analyses were performed using GraphPad Prism version 9. Protein stability curves were fitted using a one-phase exponential decay model. Bar graphs showing the percentage of resistance retained, and t_1/2 and synergy assays were analyzed using ordinary one-way analysis of variance (ANOVA). Additionally, selected time points from protein stability curves and a specific DPA concentration from MIC decay experiments were also analyzed using one-way ANOVA. ANOVA was followed by either Dunnett’s or Tukey’s multiple comparisons test, as appropriate. CFU counts between two conditions were compared using an unpaired t-test. All experiments were performed with at least three independent biological replicates. Asterisks indicate statistically significant differences with *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001.

ETHICS APPROVAL

The participation of human blood donors was reviewed and approved by the Central Ethical Committee of the Buenos Aires Province (ACTA-2024-11311995-GDEBA-CECMSALGP). Written informed consent to participate in this study was provided by the participants. Protocols of this study agreed with international ethical standards according to Helsinki Declaration and amendments.

SUPPLEMENTAL MATERIAL

The following material is available online at https://doi.org/10.1128/mbio.03447-25.

Supplemental material. mbio.03447-25-s0001.docx.

Figures S1-S15 and Table S1.

ASM does not own the copyrights to Supplemental Material that may be linked to, or accessed through, an article. The authors have granted ASM a non-exclusive, world-wide license to publish the Supplemental Material files. Please contact the corresponding author directly for reuse.