Adaptor protein mediates dynamic pump assembly for bacterial metal efflux

Two Diffusion States of CusA Shift in Relative Population Under Metal Stress.

To visualize and examine the assembly dynamics of CusCBA in a living E. coli cell, we chose to tag CusA at its C terminus with the photoconvertible fluorescent protein mEos3.2 (26, 27) at its chromosomal locus [i.e., CusA^mE, referred to as the wild-type (WT) cell; SI Appendix, section 1.1]. The C terminus of CusA is exposed to the cytosol (21), whereas for CusB and CusC, being periplasmic and outer membrane proteins, respectively, makes it problematic for the folding and maturation of fluorescent protein tags (28). Cell growth assays and protein gel analyses show that CusA^mE is functional, and the mE tag is not cleaved in the cell (SI Appendix, Figs. S3 and S4A). Photoactivation localization microscopy (PALM) (29⇓–31) and confocal fluorescence microscopy confirm that CusA^mE is membrane localized (SI Appendix, Fig. S5). Oligomerization analysis of the imaging results from PALM also supports that CusA^mE molecules are expectedly stable trimers (21, 24) in the cell (SI Appendix, section 5).

Using time-lapse stroboscopic imaging (18, 19) (4-ms laser exposure; 60-ms lapse), we tracked the motions of single photoconverted CusA^mE proteins in a cell at tens of nanometer precision until their mE tags photobleached (Fig. 1A). This stroboscopic imaging approach allows us to image diffusive proteins with effective diffusion constants up to ∼3.7 μm²⋅s⁻¹, sufficient even for small proteins freely diffusing in the cytosol with effective diffusion constants of typically ∼3–4 μm²⋅s⁻¹ (19). Cells are viable under our imaging conditions (SI Appendix, Fig. S7). From this single-molecule tracking, we obtained the probability distribution function (PDF) of the displacement length r per time lapse for single CusA^mE proteins. This PDF(r), although containing rich information on the diffusive behaviors of CusA^mE, is skewed because of the confinement effect from the 2D projection of diffusion on the curved E. coli membrane surface (32). We thus applied an inverse transformation of confined displacement distribution (ITCDD) (32, 33) to deconvolute out this confinement effect, so as to extract reliably the minimal number of diffusion states, as well as their intrinsic diffusion constants and fractional populations (Fig. 1B and SI Appendix, section 8).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 1.

SMT reveals dynamic population shifts between two different CusA^mE diffusion states. (A) Overlay of many position trajectories of single CusA^mE proteins in a living WT E. coli cell. The blue dashed line is the cell boundary. (B) ITCDD-corrected histogram of displacement length r per time lapse (60 ms) of CusA^mE from a total of 8,354 molecules and ∼350 WT cells grown in a copper-depleted medium. Solid lines represent the overall PDF (black, SI Appendix, Eq. S2) and the resolved fast (yellow) and slow (red) diffusion states, all multiplied by a scaling factor to account for the actual number of measured displacements. (C and D) Fractional populations of the fast (A_f, yellow) and slow (A_s, red) diffusion states (C), and cellular concentrations (D, in terms of CusA trimers), of CusA^mE in various strains, grown in a copper-depleted medium (−) or stressed with [Cu²⁺] of 500 μM for 30 and 90 min. All error bars are SD.

We first examined CusA^mE in WT cells grown in a copper-depleted medium. The ITCDD-corrected PDF(r) is sufficiently fitted by two diffusion states (Fig. 1B and SI Appendix, section 8). The slower diffusion state has a diffusion constant D_s = 0.052 ± 0.011 μm²⋅s⁻¹, which is almost stationary with our localization precision (∼20 nm), and a fractional population A_s = 55 ± 6%; the faster state has D_f = 0.27 ± 0.06 μm²⋅s⁻¹ and A_f = 45 ± 6% (Fig. 1C).

We then examined CusA^mE in WT cells stressed for 30 min under 500 μM Cu²⁺, which is known to cause maximal transcription induction of the cusCFBA operon in the cell (20). The same two diffusion states are resolved. However, their fractional populations have shifted significantly toward the slow state with A_s = 91 ± 6% (Fig. 1C), even though the cellular CusA^mE concentration shows no significant change (0.13 ± 0.11 μM) under copper stress (Fig. 1D).

Metal-Responsive Dynamic CusCBA Pump Assembly.

We assigned the faster diffusion state of CusA^mE to be dominated by free CusA trimers that are not assembled into CusCBA complexes and that might even have a few CusB protein molecules attached (Fig. 2A, Left). This attribution comes from the following reasoning. First, its diffusion constant (D_f ∼ 0.27 μm²⋅s⁻¹) is consistent with those reported for simple inner-membrane proteins (0.18–0.52 μm²⋅s⁻¹) (34⇓–36). Second, the interaction affinity of CusA and CusB in their apo forms is in the micromolar range (24); with cellular CusA protein concentration in the submicromolar range (Fig. 1D), some disassembly of the CusCBA complex is likely, especially when cells are grown in copper-depleted media.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 2.

Adaptor protein-mediated metal-responsive dynamic assembly–disassembly model of CusCBA complex. (A) In the absence of the metal substrate (Cu⁺ or Ag⁺), CusCBA is dominantly disassembled (Left). Upon binding the metal substrate (i.e., M⁺) at its trimethionine site, CusB changes conformation and shifts the equilibrium toward the fully assembled CusC₃B₆A₃ complex for efflux (Right). IM, inner membrane; OM, outer membrane. (B) Schematic potential energy surface of the metal-responsive dynamic assembly shift of CusCBA, where the changes in energy barriers reflect the changes of the assembly and disassembly rates.

We assigned the slower diffusion, almost stationary, state (D_s ∼ 0.052 μm²⋅s⁻¹) of CusA^mE to contain a significant contribution from the fully assembled CusCBA complex (i.e., CusC₃B₆A₃; Fig. 2A, Right). This CusCBA complex is expected to not diffuse much and be almost stationary because it traverses the periplasm (and the peptidoglycan layer), directly linking the inner and outer membranes. This slower diffusion state of CusA^mE might also have contributions from CusA trimers clustered together within the inner membrane, as it is known that membrane protein clustering can slow their motions significantly (37). Moreover, the distribution of pairwise distances between the tracked CusA^mE molecules shows a peak at a shorter distance than that from a simulation of randomly distributed CusA trimers in the membrane, supporting some level of clustering of CusA^mE in the cell (Fig. 3A, blue vs. black lines).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 3.

Pairwise distance distribution and localization distribution for identifying clustering and bipolar accumulation of CusA. (A) Normalized pairwise distance distribution (PWD) of tracked CusA^mE molecules in WT strain (without and with 30-min copper stress) and ΔcusCB strain without copper stress, as well as from a simulation of random localizations. (B) Two-dimensional distribution of all CusA^mE initial localizations in the ΔcusCB strain (8²-nm² bin size) without copper stress. Histogram normalized to the maximum count and overlaid on the average cell boundary (red solid line). (C) Difference of normalized PWD (ΔPWD) of ΔcusCB from that of WT strain without copper stress in A. Gray dotted lines: 99.7% confidence bounds. (D, Top) One-dimensional normalized localization distribution (LD) by projecting B onto the long axis of the cell (red line) and that from the simulation of random localizations (black dashed line). (D, Bottom) Difference between data and simulation (ΔLD, red line) and the 99.7% confidence bounds (gray dotted lines). (E–H) Same as C and D, but for the ΔcusC and ΔcusB strains.

We further attributed the population shift toward the slower diffusion state under copper stress to an increased population of the CusCBA complex (Fig. 2A) (rather than increased CusA clustering, as no shortening of the pairwise distances was observed; Fig. 3A, blue vs. green lines). This shift would thus reflect a metal-responsive dynamic CusCBA assembly that would allow the cell for more effective metal efflux.

To validate our assignments, we examined CusA^mE in a cusC and cusB double-deletion strain (i.e., ΔcusCB) where no CusCBA complexes can exist. Surprisingly, SMT still resolves the two diffusion states in this strain grown in the copper-depleted medium, where the fractional population A_s of the slower state is comparable to that in the WT strain (Fig. 1C). We postulated that the slower diffusion state here must come from significant CusA trimer clustering in the membrane (SI Appendix, Fig. S13A). Consistent with our postulate, pairwise distance distribution of tracked CusA^mE molecules in this ΔcusCB strain shows significant shortening compared with that in the WT strain (Fig. 3 A and C). Past studies have shown that membrane protein clustering is often correlated with their spatial segregation toward the two poles of an E. coli cell, possibly because larger membrane curvature restricts cluster escaping (38). We therefore examined the localization distribution of CusA^mE in this ΔcusCB strain: compared with that of the WT strain and the simulated random distribution, a bipolar accumulation of CusA^mE is clearly observable (Fig. 3 B and D), further supporting our CusA clustering postulate in this ΔcusCB strain.

We further examined this ΔcusCB double-deletion strain after 30-min copper stress. The significant CusA trimer clustering persists, as reflected by the shortened pairwise distance distribution and bipolar accumulation of CusA^mE (SI Appendix, Figs. S10B and S12B). However, no population shift is observable toward the slower diffusion state of CusA^mE (Fig. 1C), indicating that copper stress does not induce significant changes in CusA trimer clustering. More important, the lack of a copper-induced population shift here indicates that the metal-induced CusCBA assembly in the WT strain requires the presence of CusB, CusC, or both, and is not a result of changes in the extent of CusA trimer clustering.

Adaptor Protein CusB Is a Key Responsive Protein.

To identify whether CusC or CusB is a key player for the metal-responsive assembly of CusA into the CusCBA complex, we examined the single-deletion ΔcusC and ΔcusB strains. Relative to the double-deletion ΔcusCB strain, these two strains are equivalent to back-introducing the CusB or CusC protein into the cell individually. For both of these single-deletion strains grown in the copper-depleted medium, SMT of CusA^mE still resolves the two diffusion states with their respective fractional populations comparable to those in the ΔcusCB strain (Fig. 1C). For the ΔcusC strain, the slow diffusion state must still be dominated by CusA trimer clustering (SI Appendix, Fig. S13B) because the shortened pairwise distance distribution and the accompanying bipolar accumulation of CusA^mE are still apparent (Fig. 3 E and F).

The similarities between the ΔcusC and ΔcusCB strains therefore indicate that, once CusC is deleted, CusA will significantly cluster and accumulate toward the poles of the cell, in which CusB does not play significant roles even though CusB may interact with one or both of the diffusion components of CusA. The changes in CusA’s clustering could come from the lack of interaction with CusC increasing CusA trimer’s affinity toward itself, resulting in more CusA trimer clustering and concurrent accumulation toward the poles. Consistently, no copper-induced population shift was observed between the diffusion states in the ΔcusC strain (Fig. 1C), as in the ΔcusCB strain, both of which reflect that CusA trimer clustering has no significant dependence on copper stress.

In contrast, for the ΔcusB strain, the shortened pairwise distance distribution and bipolar accumulation of CusA^mE have either vanished or diminished significantly compared with the ΔcusCB strain (Fig. 3 G and H vs. C and D). Thus, the slower diffusion state here must not be dominated by CusA trimer clustering. Past biochemical studies (7) have shown that, for the RND family drug efflux complex AcrBA–TolC, the inner-membrane pump AcrB can directly interact with the outer-membrane channel protein TolC in the absence of the adaptor protein AcrA (39, 40). A protein docking model (24) also showed that CusA can interact with CusC directly. We therefore attributed the slower state in this ΔcusB strain to have a significant contribution from the assembled bicomponent CusCA complex (SI Appendix, Fig. S13C), whose diffusive behavior should be similar to that of the CusCBA complex, as it also traverses the periplasm and connects the inner and outer membranes (although it is likely nonfunctional for copper or silver efflux). Importantly, copper stress does not cause a population shift from the faster (free CusA) to the slower (CusCA complex) state in this ΔcusB strain (Fig. 1C), in contrast to the WT strain, in which the slower state has a significant contribution from the CusCBA complex. Therefore, CusB is a key protein for the metal-induced assembly of the CusCBA complex, whereas CusC, although being an essential component of CusCBA assembly and able to interact directly with CusA in the absence of CusB, does not contribute to the metal response in shifting CusCBA toward the assembled state.

Molecular and Kinetic Basis of CusB’s Metal Sensing in Shifting CusCBA Assembly.

A likely molecular component of CusB in mediating the metal-responsive assembly of the CusCBA complex is its trimethionine site (i.e., M21, M36, and M38), which can bind Cu⁺ and Ag⁺ under physiological conditions and undergo metal transfer with the metallochaperone CusF, which is also part of the cusCFBA operon (41⇓–43). To test this, we made a cusB(ΔN) strain, in which CusB’s N-terminal residues 1–61 containing this site were truncated (42); the signal peptide sequence was maintained so as to not affect its targeting to the periplasm. In addition to resolving the faster diffusion state, SMT of CusA^mE in this cusB(ΔN) strain still reveals the slower diffusion state, with a fractional population comparable to that in the WT strain (Fig. 1C). More important, this truncation abolishes the copper-induced population shift toward the slower state (Fig. 1C), whereas pairwise distance distribution and localization distribution analyses show no increase in CusA clustering and no bipolar accumulation under copper-depleted or copper-stressed conditions (SI Appendix, Figs. S10E and S12E). These results suggest that CusB’s N-terminal domain, and thus its trimethionine metal-binding site, is essential for its ability to mediate the metal-responsive assembly of the CusCBA complex.

To further probe the role of CusB’s trimethionine metal-binding site, we created a cusB(M36I) strain in which one of the metal-binding methionine (Met36) was mutated to isoleucine to abolish CusB’s specific binding of copper or silver (22, 43). Pairwise distance distribution and localization distribution analyses show a slight increase in CusA clustering and its bipolar accumulation (SI Appendix, Figs. S10F and S12F), which is also reflected by the slight increase of the slow state population compared with the WT strain (Fig. 1C). More important, this mutation abolishes the copper-induced population shift toward the slow state, further supporting that metal binding to CusB’s trimethionine site is essential for the metal-responsive assembly of the CusCBA complex in the cell.

Past studies have shown that CusB can obtain copper and silver from the periplasmic metallochaperone CusF through direct protein–protein interactions (41⇓–43). To probe whether CusF is involved with CusB in mediating the metal-responsive CusCBA assembly, we further studied CusA^mE in a ΔcusF strain. Under copper-depleted conditions, SMT of CusA^mE still resolves the slower and faster diffusion states, with fractional populations comparable to those in the WT strain (Fig. 1C). Pairwise distance distribution and localization distribution analyses show neither an increase in CusA clustering nor a bipolar accumulation in this ΔcusF strain (SI Appendix, Figs. S10G and S12G). Moreover, under copper stress, a clear population shift is observable toward the slower diffusion state, comparable in magnitude to that of the WT strain (Fig. 1C). Therefore, CusF does not play a significant role in CusB’s ability to bind metal and mediate the dynamic assembly of CusCBA, and CusB can sense and bind metal directly in the periplasm.

To probe the kinetic basis of CusB-mediated metal-responsive CusCBA assembly, we further analyzed the displacement-vs.-time trajectories of single CusA^mE in the WT strain (SI Appendix, Fig. S14), obtainable from the tracking trajectories as in Fig. 1A. Thresholding this displacement trajectory with an upper displacement limit selects out those small displacements and gives the estimate of the individual time durations (i.e., the microscopic residence time τ) of a single CusA^mE within an assembled CusCBA complex. Using a two-state kinetic model comprising the disassembled and assembled CusA, as well as correcting for the photobleaching/blinking kinetics of the mE tag, we could analyze the distribution of τ to obtain k_d, the apparent first-order disassembly rate constant of the CusCBA complex (SI Appendix, section 11.1 and Fig. S14 B–D). Assuming a quasiequilibrium between the two states further allowed us to use their fractional populations to derive k_a, the pseudo–first-order assembly rate constant (SI Appendix, section 11.1). Using simulations of two-state membrane diffusions, we further validated the reliability of our analysis in obtaining kinetic trends (SI Appendix, section 11.2).

The disassembly and assembly rate constants are on the order of 10⁰ to 10¹ s⁻¹, corresponding to subsecond timescales (Fig. 4). In the WT strain, the disassembly rate constant decreases while the assembly rate constant increases with 30-min copper stress (Fig. 4A). Similar trends are clear in the ΔcusF strain, but the values of the rate constants are slightly larger (Fig. 4B). Therefore, when CusB binds metal, it not only increases the assembly kinetics but also decreases the disassembly kinetics of the CusCBA complex, possibly via altering the potential energy surfaces of complex assembly and synergistically resulting in a dynamic population shift toward the assembled state (Fig. 2B). CusF here, although not significantly affecting the thermodynamics of the population shift, could facilitate the kinetics underlying the assembly. It is worth noting that we only analyzed the kinetics in the WT cusA^mE and ΔcusF strains, where the slow diffusion state is dominated by the assembled CusCBA complex and CusA trimer clustering is insignificant. Nevertheless, the kinetics of assembly–disassembly here has some contamination from clustering–unclustering kinetics. However, the change in kinetics upon Cu stress here is reliable because CusA clustering is independent of Cu stress.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 4.

Disassembly and assembly kinetics of CusCBA complex. Copper stress dependence of the apparent assembly (k_a) and disassembly rate constants (k_d) for the WT cusA^mE (A) and ΔcusF strains (B).