Hydrophilic nanoparticles that kill bacteria while sparing mammalian cells reveal the antibiotic role of nanostructures

Synthesis and characterization of hydrophilic NPPBs with well-defined sizes

To study the antibiotic role of nanostructures, we designed model NPPBs consisting of inert silica nanospheres of systematically varied diameters grafted with poly(4-vinyl-N-methylpyridine iodide) (P4MVP) brushes of similar degree of polymerization (i.e., DP ~ 32). Both silica and P4MVP have been explored for biomedical applications. For example, P4MVP has been studied for gene delivery^35,36; silica is listed amongst the “generally regarded as safe” substances by the Food and Drug Administration (ID Code: 14808-60-7)³⁷, and the long-term effect of silica nanoparticles on human health is still under scrutinization³⁸. The metabolization and ultimate fate of NPPBs in human body are not known, but the underlying mechanism of nanostructure-induced transformation of antimicrobial activity revealed using this model system should still have broad implications.

To synthesize the model NPPBs, silica nanospheres of well-controlled sizes carrying surface derivatives of the initiator moiety (i.e., α-bromoisobutyryl bromide) for atom transfer radical polymerization (ATRP) were first synthesized. Surface-initiated ATRP (SI-ATRP) was then performed to graft well-defined poly(4-vinylpyridine) (P4VP) brushes on the silica nanospheres (i.e., SiO₂@P4VP), followed by a quaternization reaction with methyl iodide³⁹ to turn P4VP into hydrophilic and cationic P4MVP brushes (Fig. 1a). Experimental details on the synthesis and characterization of model NPPBs are included in Supplementary Information.

We used a series of characterization methods such as transmission electron microscopy (TEM), gel permeation chromatography (GPC), nuclear magnetic resonance (NMR) spectroscopy, Fourier transform infrared (FTIR) spectroscopy, and thermal gravimetric analysis (TGA) together with brush cleavage experiments to determine the silica nanosphere diameter (d_silica), polymer brush size (DP) and polydispersity index (PDI), the degree of quaternization, and graft density (Supplementary Figs. 5–7). We confirmed the successful synthesis of six model NPPBs consisting of P4MVP brushes of a similar size (i.e., DP = 32 ± 3) covalently grafted onto silica nanospheres of systematically varied diameters (i.e., d_silica ~ 7–270 nm). Representative TEM pictures of the silica nanospheres before (Fig. 1b–g) and after SI-ATRP of the P4VP brushes (Fig. 1h, i) are shown. The P4VP brush layer appears as a light-gray shell (pointed by red arrow) overlaid on top of the dark silica nanospheres due to their electron density contrast. The thickness of this shell expands when the P4VP brush size increases³⁹. All NPPBs are positively charged with similar zeta potentials (Supplementary Fig. 8), which also reflect their similarly sized P4MVP brushes. As we reported before³⁹, the zeta potential of NPPBs (i.e., the charge state of P4MVP brushes) is independent on pH.

A summary of the structural characteristics of all model NPPBs is listed in Table 1, in which the hydrophilic nanoparticles were individually named as “Sm–Pn”, where “S” and “P” stand for the silica nanosphere and polymer brush, respectively, with m and n denoting their sizes in nanometer or DP. For example, S25 refers to bare silica nanospheres of d_silica = 25 nm, and S25–P31 refers to the hydrophilic NPPB consisting of S25 covalently grafted with P4MVP brushes of DP = 31. We used the DPs of free P4VP grown simultaneously in the same synthesis batches of SiO₂@P4VP to represent the brush sizes on individual NPPBs, as little difference exists between the surface-bound brushes versus free polymers grown simultaneously during the synthesis of polymer brushes via controlled/“living” polymerization³⁹. We further validated their similarity by cleaving the P4VP brushes and comparing their GPC profiles with those of the free P4VP polymers grown concurrently in the same synthesis batch (Supplementary Fig. 6). Considering that the P4MVP brushes are polyelectrolytes, we estimated their radius of gyration (i.e., for DP = 32) to be ~2.3 nm and critical graft density for mushroom-to-brush transition to be ~0.06 chain/nm², which is far below that of all model NPPBs (Table 1).

Antimicrobial activity and cytotoxicity of model NPPBs

We used standard bacteria killing and inhibition assays^22,40,41 against two representative strains from each bacterial family, Gram− Escherichia coli (i.e., E. coli) and PA14, and Gram+ Staphylococcus aureus (i.e., S. aureus) and MU50, respectively, in which PA14 (the tobramycin and gentamycin-resistant Pseudomonas aeruginosa) and MU50 (the methicillin, oxacillin, and vancomycin-resistant S. aureus) are clinical multidrug resistant bacterial strains, to obtain the minimum bactericidal concentration (MBC) and inhibitory concentration (MIC). We quantified cytotoxicity by standard hemolysis and MTT assays against HRBCs and HEK-293 cells, respectively, to obtain HC₅₀ (i.e., the concentration at which 50% of the HRBCs are lysed)²² and IC₅₀ (i.e., the concentration at which the viability of HEK-293 cells is reduced by 50%)²³. We also used live/dead cell staining assays to directly observe the wellbeing of bacteria and HEK-293 cells^22,23.

As controls for NPPBs, the hydrophilic brush polymer P4MVP₂₈ (i.e., P28) by itself is non-hemolytic and does not show MIC against Gram− E. coli and PA14, nor MBC against E. coli up to 512 μg/mL that we tested²². Although it reaches MIC against Gram+ S. aureus (MIC = 24 μg/mL) and MU50 (MIC = 128 μg/mL), which is consistent with previous reports that cationic compounds are in general strongly bacteriostatic against Gram+ bacteria^42,43, no MBC is observed against S. aureus up to 512 μg/mL that we tested²². The bare silica nanospheres also show no MIC nor MBC against both Gram− E. coli and Gram+ S. aureus up to 4000 μg/mL that we tested (Supplementary Fig. 9). Besides their cordiality with bacteria, both P28 and silica nanospheres by themselves show no cytotoxicity when tested by hemolysis²² and live/dead cell staining assays (Supplementary Fig. 9), respectively.

Formation of NPPBs by grafting the non-toxic and non-bactericidal hydrophilic brush polymers onto silica nanospheres doesn’t alter their amicability with mammalian cells, but it incurs a collective transformation of their antimicrobial potential against both Gram+ and Gram− bacteria, including clinical multidrug-resistant PA14 and MU50 strains (Fig. 2a, b). The hemolysis (Fig. 2c) and MTT assays (Fig. 2f) show that the resultant hydrophilic NPPBs behave similarly as the P28 brush by itself or the silica nanospheres alone, revealing no HC₅₀ against HRBCs nor IC₅₀ against HEK-293 cells. In contrast, a nanostructure-induced transformation of antimicrobial activity shows up in both the MBC (Supplementary Fig. 10) and MIC assays against Gram− (Fig. 2a) and Gram+ bacteria (Fig. 2b), respectively. A list of MICs, MBCs, HC₅₀, and IC₅₀ values of model NPPBs are summarized in Table 2. Interestingly, the acquired antimicrobial potential due to the assembly of polymer brushes into nanostructures depends critically on the nanoparticle size: it intensifies with smaller nanoparticles (d_silica ≤ 50 nm) but subsides quickly with larger ones (d > 50 nm). For example, the MICs for S7–P35 and S50–P29 against E. coli are 32 and 64 μg/mL, respectively, but the MICs for S110–P34 and S270–P32 against E. coli are increased dramatically to 512 and 1000 μg/mL, respectively (Table 2). Given that the graft density of P4MVP brushes differs in NPPBs of different sizes (Table 1), the antimicrobial activities of NPPBs are further normalized to P4MVP brush concentrations for individual NPPBs with different specific surface areas, which too reveal a critical nanoparticle-size dependency, as examples of the normalized MICs shown for E. coli (Fig. 2d) and S. aureus (Fig. 2e), respectively.

This nanoparticle-size dependent antimicrobial potential is also evident in MBC assays (Supplementary Fig. 10 and Table 2) and in plain sight under confocal microscope: when the live and dead bacteria are stained in green and red²², respectively, the wellbeing of E. coli (Fig. 2g–j) and S. aureus (Fig. 2k–n) incubated with S10–P33 (Fig. 2g, k), S25–P31 (Fig. 2h, l), S110–P34 (Fig. 2i, m), and S270–P32 (Fig. 2j, n) at the same dose of NPPBs (i.e., 32 μg/mL) is visually stunning. At d_silica ≤ 50 nm, all E. coli (Fig. 2g, h) and S. aureus (Fig. 2k, l) are killed, whereas most E. coli (Fig. 2i, k) and S. aureus (Fig. 2m, n) are alive when NPPBs with d_silica > 50 nm are used. A threshold nanoparticle size (d_silica ~ 50 nm) appears to exist and roughly delineates the boundary between NPPBs of weak and strong antimicrobial potential. The presence of this boundary is not surprising for Gram+ bacteria because they are encapsulated by a thick nanoporous peptidoglycan layer that has a “mesh size” between ca. 5 to 50 nm^44,45. As we discovered before, this nanoporous peptidoglycan capsule acts as a selective filter that only allows small nanoantibiotics (d ≤ 50 nm) to cross and take actions but precludes large nanoantibiotics (d > 50 nm) from gaining access to the bacterial membrane^22,23. What remains puzzling is the existence of this boundary for Gram− bacteria as their thin peptidoglycan layer is sandwiched between the bacterial outer and inner membranes. Considering that the hydrophilic NPPB nanoparticles (d_silica~ 7–270 nm) would be too large to pass through bacterial membrane channels such as porins or LamB⁴⁶, they would have to disrupt the outer membrane of Gram− bacteria, which would kill the bacteria in the first place, before gaining access to their peptidoglycan layer. The observed nanoparticle size-dependent antimicrobial activities of the hydrophilic NPPBs against Gram− bacteria thus underscores a different yet unidentified mechanism at play.

Hydrophilic NPPBs are membrane-active antimicrobials that target bacteria with size-dependent membrane disruption activity while sparring mammalian cells

To gain more mechanistic insight on the different modes of actions that define the encounter between NPPBs and bacteria or mammalian cells, we used model giant unilamellar vesicles (GUVs) to mimic bacteria and mammalian cells, respectively. Given the dynamic membrane composition, membrane asymmetry, and cell wall differences among mammalian and bacterial cells, it is challenging to rely on simple GUVs to inform all aspects of responses ensued from mammalian or bacterial cells interacting with exogenous substances^47,48. We chose GUVs to shed light on the initial membrane remodeling event due primarily to their membrane lipid difference when bacteria or mammalian cells encounter NPPBs, and cross-checked the utility of this approach by comparing the data obtained from model GUVs with those from live cells. An important difference between mammalian and microbial membranes lies in their lipid compositions^{32,49,50,51,52}. Unlike mammalian membranes that consist predominantly of lipids with zero intrinsic curvature (e.g., phosphatidylcholine (PC) lipids), microbial membranes are laden with lipids of negative intrinsic curvature (e.g., phosphatidylethanolamine (PE) lipids)^51,52. Lipid bilayers enriched with zwitterionic PE and PC lipids, respectively, have been widely used as valuable models to mimic bacterial and mammalian membranes^53,54. Following this tradition and the pioneer works by Wong and colleagues^55,56, we chose model GUVs comprised of binary mixtures of anionic 1,2-dioleoyl-sn-glycero-3-phospho-(1′-rac-glycerol) (DOPG) and zwitterionic 1,2-dioleoyl-sn-glycero-3-phosphocholine (DOPC) or 1,2-dioleoyl-sn-glycero-3-phosphoethanolamine (DOPE). Both DOPG and DOPC have zero intrinsic curvature, whereas DOPE has a negative intrinsic curvature. We used 20/80 (molar ratio) DOPG/DOPE and DOPG/DOPC, respectively, to mimic the PE-rich microbial and PC-rich mammalian membranes by keeping the membrane charge density the same, hence eliminating any other variable in the model system.

To illustrate whether NPPBs encroach bacterial or mammalian cell membranes, we prepared bacteria-mimicking and mammalian cell-mimicking GUVs loaded with fluorescein and used confocal microscopy to examine membrane integrity by monitoring the dye leakage from GUVs exposed to individual NPPBs. Our previous study has shown that the P28 control doesn’t cause dye leakage from both types of GUVs²². Examples of time-lapse confocal microscopy images of GUVs interacting with model NPPBs are shown in Supplementary Fig. 11. No dye leakage is observed from mammalian cell-mimicking GUVs incubated with NPPBs of any size (Fig. 3a). This is consistent with hemolysis assays that revealed very weak hemolytic activity for all NPPBs (Fig. 2c), suggesting that the hydrophilic nanoparticles don’t rupture mammalian membranes. The compatibility of NPPBs to mammalian cells is further supported by the MTT assay (Fig. 2f): even though endocytosis may occur when HEK-293 cells encounter the hydrophilic nanoparticles, the viability of the cells still suggests that their membrane integrity is not compromised because loss of homeostasis would lead to cell death. In contrast, when bacteria-mimicking GUVs interact with the NPPBs, rapid and complete dye leakage indicative of membrane pore formation occurs when smaller NPPBs (d_silica ≤ 50 nm) are used, whereas small and slow dye release revealing mild deterioration of membrane integrity over time is observed for larger NPPBs (d_siilica > 50 nm) (Fig. 3b). Taken together, the dye leakage assays suggest that NPPBs remodel cell membranes with different modes of interaction depending on both the intrinsic curvatures of membrane lipids and the hydrophilic nanoparticles sizes.

The relevance of probing cell membrane integrity using model GUVs and dye leakage assays is further vindicated by the bacterial membrane permeation assay (Fig. 3c), MIC and MBC assays (Fig. 2 and Table 2), as well as the scanning electron microscopy (SEM) and cross-sectional TEM studies of both Gram− E. coli (Fig. 3d–k) and Gram+ S. aureus (Fig. 3l–q). Permeation of hydrophobic fluorescent probe 1-N-phenylnaphthylamine into disrupted outer membrane of Gram− bacteria causes a prominent increase of its fluorescent emission, which serves as a telltale sign to indicate whether the membrane disruption occurs⁵⁷. The membrane permeation assay (Fig. 3c) confirms that while E. coli membrane does not break in the presence of P28 control or larger NPPBs such as the S270–P32, membrane disruption does occur when E. coli cells encounter smaller NPPBs (i.e., d_silica ≤ 50 nm), as indicated by a dramatic increase of the fluorescent emission in the examples of S10–P33 and S25–P31. This nanoparticle size dependent bacterial membrane disruption of NPPBs is directly linked to their antimicrobial activities. As we discovered earlier in the MIC and MBC assays (Fig. 2 and Table 2), the smaller NPPBs (d_silica ≤ 50 nm) capable of disrupting bacteria-mimicking GUVs (Fig. 3b) and bacterial membranes (Fig. 3c) are the ones that show strong antimicrobial activities, whereas the larger NPPBs (d_silica > 50 nm) unable to disrupt either membrane are the ones that show weak antimicrobial activities. Clearly, the hydrophilic NPPBs are membrane-active antimicrobials that selectively kill bacteria by disrupting their membranes with size dependent activities while sparring mammalian cells.

The nanoparticle-size dependent activities of NPPBs on disrupting bacterial membranes are also directly exhibited under SEM and cross-sectional TEM. Examples of SEM (Fig. 3d–g) and cross-sectional TEM pictures (h–k) of the E. coli control (d, h) and that incubated with S25 (e, i), S25–P31 (f, j), and S270–P32 (g, k), respectively, are shown. We reported previously that the P28 control by itself shows no bactericidal activity against both Gram− E. coli and Gram+ S. aureus, and no bacterial membrane disruption was observed under SEM or TEM when P28 was introduced to either family of bacteria²². Likewise, when E. coli is incubated with the antimicrobially inactive silica nanospheres (e.g., S25), no morphological change is observed under SEM (Fig. 3e; S25 indicated by red arrow) as compared to the E. coli control (Fig. 3d), and no membrane disruption is observed under cross-sectional TEM either (Fig. 3i) just like the E. coli control (Fig. 3h; the bacterial membrane was stained with OsO₄ and appears as a continuous dark layer indicated by blue arrow). The S25 that appears to adhere to E. coli membrane under SEM (Fig. 3e) stays with the membrane under cross-sectional TEM (Fig. 3i) with no sign of membrane disruption.

Formation of NPPBs by grafting the hydrophilic linear-chain brush polymers onto silica nanospheres doesn’t alter their amicability with mammalian cells, but a fundamental transition occurs that transforms the hydrophilic “hairy” balls into potent antibiotics when the nanoparticles are small (i.e., d_silica ≤ 50 nm) (Fig. 2). This nanoparticle-size dependent transformation reveals itself clearly under electron microscopes: representative SEM shows that E. coli cells are crumbled with S25–P31 adhering to their surfaces (Fig. 3f, nanoparticles indicated by red arrow), and cross-sectional TEM shows that the bacterial membrane is completely obliviated (Fig. 3j) with some S25–P31 (indicated by red arrow) encroaching the used-to-be cell membrane and entering the cytosol of the bacteria, which is in sharp contrast to the S25 control that remains on the intact membrane (Fig. 3e, i). When E. coli cells encounter large NPPBs such as S270–P32, no bacterial morphological change is observed under SEM (Fig. 3g), and intact bacterial membrane (indicated by blue arrow) is shown under cross-sectional TEM (Fig. 3k) even when the large cationic NPPBs (indicated by red arrow) are bound to the anionic bacterial membrane (Fig. 3g, k) just like the small S25–P31 (Fig. 3f, j).

This nanoparticle-size dependent transformation of NPPBs into membrane-active antibiotics also shows up nicely for Gram+ S. aureus, which differs from the Gram− E. coli in that it has a thick nanoporous peptidoglycan encapsulation layer, and NPPBs need penetrate this capsule first in order to gain access to the bacterial membrane. Representative SEM (Fig. 3l–n) and cross-sectional TEM pictures (Fig. 3o–q) of S. aureus control (l, o) and that interacting with S25–P31 (m, p) and S270–P32 (n, q), respectively, are shown. Like E. coli (Fig. 3f, g), S. aureus cells are bound to the oppositely charged NPPBs due to the attractive charge interactions (Fig. 3m, n, NPPBs indicated by red arrow). Although it is difficult to tell the wellbeing of the bacteria under SEM (Fig. 3m) when compared to the S. aureus control (Fig. 3l), cross-sectional TEM clearly shows that while the S. aureus control (Fig. 3o) exhibits intact membrane (indicated by blue arrow) underneath the peptidoglycan encapsulation (indicated by black arrow), this bacterial membrane is mostly destroyed in the presence of small NPPBs such as S25–P31 (Fig. 3p). Some of the small NPPBs that cross the nanoporous peptidoglycan capsule to demolish the bacterial membrane are observed (indicated by red arrow). In contrast, when S. aureus encounter large NPPBs like S270–P32 (Fig. 3n, q), although the hydrophilic nanoparticles (indicated by red arrow) appear to land on the oppositely charged bacteria under SEM (Fig. 3n), they are too large to cross the nanoporous peptidoglycan encapsulation layer (indicated by black arrow) as revealed by cross-sectional TEM (Fig. 3q), and the structural integrity of bacterial membrane (indicated by blue arrow) is not compromised.

The fluorescent dye release and membrane permeation assays together with the SEM and cross-sectional TEM studies unambiguously show that while NPPBs with small nanoparticle sizes (d_silica ≤ 50 nm) kill both Gram+ and Gram− bacteria by disrupting their membranes, NPPBs with larger nanoparticle sizes (d_silica > 50 nm) are either precluded from gaining access to the membrane of Gram+ bacteria, or simply unable to cause membrane disruptions even when adhered to the outer membrane of Gram− bacteria. The reason that we are still able to observe MBCs for the larger NPPBs (Table 2), albeit at high nanoparticle concentrations, is not due to bacterial membrane disruptions. Rather, it is likely caused by the densely packed NPPBs bound on bacteria (Fig. 3g, k, n, q) that impede the normal homeostasis processes critical for bacterial survival. The membrane-disruption mode of bacteria killing exhibited by small NPPBs is a powerful antibiotic action because it evades the bacterial resistance mechanisms^58,59. As such, the small NPPBs also show potent activities against the clinical multidrug resistant PA14 and MU50 strains (Table 2).

Mechanistic insight on hydrophilic NPPB nanoparticles that selectively disrupt bacterial membranes while sparring mammalian cells: the role of lipid intrinsic curvature and nanoparticle size

Since NPPBs are cationic and the membrane potentials of live cells are negative inside, one may argue that bacterial membrane disruption is caused by electroporation, which was proposed as one possible mode of action for cationic antibacterial peptides⁶⁰. Although electroporation can’t explain the observation that NPPBs selectively disrupt bacterial membrane while sparring mammalian cells, we tested this hypothesis nevertheless using carbonyl cyanide m-chlorophenyl hydrazone (CCCP) to dissipate the potential and pH gradient across bacterial membrane before incubating the bacteria with NPPBs. Our SEM and cross-sectional TEM studies (Supplementary Fig. 12) clearly show that targeting and disruption of both E. coli and S. aureus membranes by small NPPBs are independent on bacterial membrane potential.

Despite the simplicity of model liposomes that lack some of the conspicuous structural features of bacterial and mammalian cells, such as the membrane asymmetry, the cytoskeleton network in mammalian cells that plays important roles for endocytosis, and the cell wall structures such as the lipopolysaccharides and lipoteichoic acids of Gram− and Gram+ bacteria, respectively, the dye release assays on model GUVs that mimic bacterial and mammalian cells still faithfully capture the fundamental membrane remodeling response when both types of cells encounter the hydrophilic NPPB nanoparticles: while mammalian membrane integrity is uncompromised regardless of the NPPB sizes, bacterial membrane can be ruptured and disintegrated when the nanoparticles are small (i.e., d_silica ≤ 50 nm). The different modes of actions are corroborated by various biological assays and microscopy studies on bacterial and mammalian cells (Figs. 2 and 3), suggesting that the intrinsic curvature of membrane lipids and nanoparticle size are the two important players that synergistically define how NPPBs remodel cell membranes. We probed the membrane remodeling process at the nanoscale by synchrotron SAXS. For mammalian cell-mimicking membrane consisting predominantly of lipids with zero intrinsic curvature, we observed two sets of scatterings (Fig. 4a), one of which (marked by black arrows) is equally spaced at 0.112, 0.223, and 0.335 Å⁻¹, respectively for all NPPBs, corresponding to a multilamellar structure with a lamellar periodicity of 56 Å, which fits nicely to a DOPG/DOPC bilayer thickness (i.e., ~44 Å^61,62) plus a hydration layer (i.e., ~12 Å), and is attributed to stacked membranes brought together by the oppositely charged NPPBs adhered in-between the membranes. Given that we know precisely the P4MVP brush sizes, graft density, the nanoparticle sizes (Table 1), and the fact that the charge state of P4MVP is independent on pH³⁹, the charge density of NPPBs is calculated (i.e., ~10–30/nm² depending on the P4MVP graft density and NPPB sizes). Because the charge density of NPPBs is much higher than that of the membrane (i.e., ~1.4/nm²)⁶², we don’t expect that NPPBs would form a continuous layer on the membrane because of the charge density mismatching⁶³. In fact, the lamellar spacing does not expand when the diameter of NPPBs increases. The diffusive nature of the lamellar harmonics also indicates membrane corrugations. Taken together, we believe that the nanoparticles are wrapped in the buckled membrane “bubbles” as shown in the schematic illustration (Fig. 4a, inset). This inclusion state of NPPBs within the membrane pockets increases the entropy gain through counterion release without incurring high energy cost to break and restructure the membrane despite of their mismatched charge density. In reality though, mammalian cells would proceed to engulf NPPBs via endocytosis³³. Nonetheless, the SAXS data still provide evidence that NPPBs do not rupture mammalian membranes regardless of their sizes.

The other set of scatterings (marked by red arrows) from the NPPB-membrane complexes varies as the NPPB size changes. Those scatterings are attributed to the correlations from the well-defined NPPBs because the same set of correlations show up for the same NPPBs interacting with either mammalian cell-mimicking (Fig. 4a) or bacteria-mimicking membrane (Fig. 4b). For the larger NPPBs (i.e., S110–P34 and S270–P32), only one barely discernible weak nanoparticle scattering feature shows up, likely because most of the scatterings from the larger nanoparticles are beyond the range of the SAXS. For the smallest NPPB (i.e., S7–P35), only one well separated nanoparticle peak at 0.021 Å⁻¹ shows up too, although it is possible that additional peaks at higher q are shadowed beneath the more prominent scatterings of the remodeled membranes (peaks marked by black arrows). For other small NPPBs (i.e., S25–P31 and S50–P29), multiple sharp nanoparticle diffraction peaks appear, which fit nicely to cubic structures indicative of the 3D assembly of NPPBs. For instance, the scatterings from S25–P31 are positioned at 0.033 and 0.054 Å⁻¹, respectively, which are related to each other by the ratio of (sqrt{3}):(sqrt{8}) and can be indexed as (1,1,1) and (2,2,0) of a 3D cubic structure with a lattice parameter a of 329.6 Å. This unit cell matches well to the size of S25–P31 (d_silica ~ 25 nm, Table 1) with additional spacing (~8 nm) to accommodate the bilayer and polymer brushes. Similarly, the diffractions from S50–P29 are positioned at 0.020, 0.034, 0.045, 0.058, and 0.068 Å⁻¹, respectively, which are related to each other by the ratio of (sqrt{2}):(sqrt{6}):(sqrt{10}):(sqrt{17}):(sqrt{24}) and fit nicely to the Miller indexes (1,1,0), (2,1,1), (3,1,0), (4,1,0), and (4,2,2), respectively (Fig. 4c), of the cubic structure (e.g., Pm3n) with a lattice parameter of 450 Å. The enlarged unit cell reflects the increased nanoparticle size of S50–P29 (i.e., d_silica ~ 45 nm, Table 1). Although it appears to be short of spacing for the bilayer and polymer brushes, this discrepancy could be attributed to the difference in measuring the nanoparticle size by TEM and SAXS.

For bacteria-mimicking membrane consisting predominantly of lipids with negative intrinsic curvature, besides the same set of scatterings coming from the 3D organized NPPBs that varies as the nanoparticle size changes (peaks marked by red arrows), two distinctly different sets of membrane scatterings (marked by solid and dashed black arrows, respectively) are observed (Fig. 4b). Within each set of the membrane scatterings, the SAXS pattern remain identical for different NPPBs, but there is a clear transition from one type of membrane remodeling to the other type when the NPPBs increase beyond a threshold size (i.e., d_silica ~ 50 nm). A summary of all SAXS peaks together with their Miller indices are listed in Table 3.

For small NPPBs (d_silica ≤ 50 nm), a set of six membrane scatterings (marked by solid black arrows) appear, which fit nicely to a 2D hexagonal structure with a lattice size of ~64 Å (Fig. 4e). With the exception of q₁₂ that is too weak to be clearly identified, those scatterings represent the first seven reflections from q₁₀ to q₃₁ of the 2D membrane lattice. For large NPPBs (d_silica > 50 nm), a set of fourteen membrane scatterings (marked by dashed black arrows) appear instead, which fit nicely to a cubic structure (e.g., Pn3m) with a lattice parameter of 193 Å (Fig. 4d). Interestingly, the two completely different membrane remodeling behaviors are parted by a threshold nanoparticle size (i.e., d_silica ~ 50 nm) that happens to delineate NPPBs of strong (d_silica ≤ 50 nm) and weak (d_silica > 50 nm) antimicrobial activities. It becomes clear that the ability to remodel the bacteria-mimicking membrane into 2D hexagonal rather than 3D cubic structures by the hydrophilic NPPBs is the harbinger of bacteria death, and this ability is critically dependent on the nanoparticle size.

To further understand the underlying mechanism of bacteria death caused by small NPPBs, we performed Fourier reconstruction of the 2D hexagonally ordered membrane using the method reported before^22,55,64. Based on the phase criteria developed by Turner and Gruner⁶⁴, our phase choices are (+−−+++). The reconstructed electron density map of bacteria-mimicking membrane remodeled by S7–P35 along the lattice plane reveals the 2D hexagonally packed membrane pores (Fig. 5a). Since the same 2D hexagonal lattice of remodeled membrane (i.e., a = 6.4 nm) is observed for NPPBs of different sizes ranging from S7–P35 to S50–P29 that are all larger than the lattice size, the nanoparticles themselves should not fit into the 2D lattice. Quantitative analysis of the electron density distribution (Fig. 5b) along the unit cell axis (e.g., x-axis) helps elucidate the cooperative molecular rearrangement leading to the membrane pore formation. The region in-between the pores has the lowest electron density, which can be attributed to the lipid hydrocarbon tails (ρ = 0.29 e/ų) because they have the lowest electron density in the membrane. Approaching the rim of those pores the electron density rises to 0.55 e/ų. It then drops to 0.46 e/ų inside the pores before a feature with the highest electron density (ρ = 0.65 e/ų) appears at the pore center. This electron density is much higher than that of any component in the membrane and can be only assigned to the polymer brushes because the quaternized P4MVP with associated iodide counterions has the highest electron density in the system. Notably, the high electron density confirms that even though polymer brushes are present in the pores, NPPBs are not part of the 2D lattice because the low electron density of silica nanospheres (i.e., 0.19 e/ų) doesn’t show up at all. The second highest electron density (ρ = 0.55 e/ų) at the rim of the pores can be subsequently assigned to phospholipid headgroups because they have the next highest electron density after the polymer brushes. This electron density is higher than that of a typical phospholipid headgroup (0.41 e/ų)⁶⁴, suggesting the presence of residue iodide ions, likely due to the unmatched charge density between the polymer brushes and membrane. The presence of residue iodide ions is further confirmed by the increased electron density inside the aqueous pores (ρ = 0.46 e/ų), which is much higher than water by itself (ρ = 0.33 e/ų)^22,64. Taken together, the Fourier reconstruction reveals a cooperative membrane remodeling pathway that starts by the attractive electrostatic interactions between polymer brushes and membrane surface, which induce a topological transition of the membrane by tilting the headgroups of surrounding lipids inward to encircle the polymer brushes in the center, and ends up with the formation of a honeycomb-like 2D inverted hexagonal phase (H_II), the hallmark of membrane pores. This mode of pore formation differs from any current mode of actions proposed for amphiphilic membrane-active antimicrobials, in which the hydrophobic moieties of those antimicrobials are expected to disrupt cell membranes by breaching into the hydrophobic membrane interior^65,66,67. It is also slightly different from what we observed previously with hydrophilic bottlebrush polymers, which induce pore formation by bending the membrane to encircle themselves^22,23. This difference is likely due to the increased sizes of NPPBs that are too large to fit into the membrane, as the polymer brushes are grown on silica nanospheres (d_silica ~ 7–270 nm) instead of the molecular backbones of bottlebrush polymers. As a subtle additional evidence, the lattice parameter of the H_II phase induced by the polymer brushes from NPPBs (i.e., a = 6.4 nm) is smaller than that by the bottlebrush polymers (i.e., a = 7.0 nm)^22,23. Given that only a portion of polymer brushes on NPPBs are involved in the formation of each membrane pore, individual NPPBs may induce the formation of ca. one to several pores on the membrane depending on the nanoparticle sizes, as schematically illustrated in Fig. 5c, d, respectively. However, there is a threshold size of NPPBs beyond which it is not energetically favorable for their polymer brushes to remodel the bacteria-mimicking membrane into the H_II phase anymore.

This threshold appears to be roughly 50 nm for the silica nanospheres. Large NPPBs (d_silica > 50 nm) remodel the bacteria-mimicking membrane into cubic structures similar to that observed when the membrane was remodeled by the linear-chain P28 control²². A unit cell of the bicontinuous double diamond Pn3m structure is schematically shown in Fig. 5e, where interactions between the bacteria-mimicking membrane and a blanket of oppositely charged polymer brushes on a quasi-flat surface of NPPBs induce the formation of 3D bicontinuous lipid cubic phase. For illustration the bilayer tubes in the unit cell are cut open at the boundary, and a cross-sectional view of a bilayer tube is shown (inset, Fig. 5e). Unlike the 2D hexagonal H_II phase that signifies membrane pore formation, continuous membrane still persists in the bicontinuous cubic phase. In reality though, bacterial membranes differ from the bacteria-mimicking model membrane in that the bacterial membranes are supported and constrained by bacterial cell walls. Although the polymer brush-induced formation of 2D H_II phase and pores on the model membrane is relevant for bacterial membranes, the 3D cubic structures are less relevant because bacterial membranes are unlikely able to organize freely into 3D structures. Nevertheless, the membrane integrity difference revealed in the phase behavior of model membrane incubated with hydrophilic brush polymer controls^22,23, bottlebrush polymers^22,23, and NPPBs still reflects faithfully the fate of bacterial membranes under the same encountering events: while the formation of 2D hexagonally packed pores in the model membrane bodes well for the observed disruptions of bacterial membranes, formation of 3D cubic phases is correlated with the observed intact bacterial membranes. This observation is again different from amphiphilic membrane-active antimicrobials, where the formation of bicontinuous cubic phases correlates with membrane pore formation⁶⁷.

It’s worthwhile to deliberate over the mechanism underlying the observation that no membrane disruption occurs for mammalian cells interacting with hydrophilic NPPBs of all sizes, whereas a nanoparticle size dependent membrane disruption exists for bacteria. This observation indicates that while nanostructures may help transform the non-toxic and non-bactericidal hydrophilic brush polymers into membrane-active antimicrobials, the successful outcome depends on both the intrinsic curvature of membrane lipids and the nanoparticle size. Unlike mammalian membranes that are rich in zero-intrinsic-curvature lipids, microbial membranes are laden with negative-intrinsic-curvature lipids that help generate the saddle-splay curvature critically needed for membrane pore formation⁶⁷. When encountering the hydrophilic NPPBs, mammalian membranes buckle locally to increase the contact areas with NPPBs by enclosing them in membrane “bubbles” (Fig. 4a), which increases the attractive electrostatic interactions and entropy gain through counterion release without incurring high energy cost to break the continuity of the membranes. In contrast, microbial membranes undergo a topological transition to form 2D inverted membrane pores that wrap around individual bundles of polymer brushes with much more enhanced contact areas (Fig. 5c, d). Although formation of membrane pores maximizes the attractive electrostatic interactions with polymer brushes and entropy gain through the counterion release, it only happens for microbial membranes because their abundant negative-intrinsic-curvature lipids help offset the energy cost to bend the membranes into nanopores via the spontaneous formation of the saddle-splay curvature. This energy cost would be prohibitively high for mammalian membranes rich in zero-intrinsic-curvature lipids. As further evidences to demonstrate the importance of lipid compositional difference that helps set apart their different paths toward different cellular fates when bacteria and mammalian cells encounter hydrophilic NPPBs, we performed SAXS studies of a series of NPPB-remodeled model membranes bearing the same charge density (i.e., 20% PG lipid) but systematically varied PE/PC ratios. As shown in Supplementary Fig. 13 for the small NPPBs (d_silica ≤ 50 nm), formation of the 2D H_II phase only occurs in the bacteria-mimicking membrane with a high PE content (i.e., ~80%). The ability to induce pore formation on the model membranes by the NPPBs subsides precipitately when the PE content drops below 70%, which includes representative membranes of eukaryotic cells.

In addition to that, there is a clear nanostructure size dependent transition on whether the hydrophilic NPPBs can induce the formation of pores on bacterial membranes. We observed previously a transition from antimicrobially-inactive hydrophilic polymers to antimicrobially-active bottlebrush polymers: while the hydrophilic linear-chain polymers interact with bacteria-mimicking membrane uniformly and induce the formation of 3D bicontinuous cubic structures without disrupting the membrane, the nanostructured bottlebrush polymers consisting of the same hydrophilic polymers grafted on molecular backbones induce the formation of 2D H_II phase and membrane pores^22,23. Here, we replace the soft molecular backbones with hard silica nanospheres of well-defined and systematically increased diameters, and witness a very similar transition where NPPBs of large and small diameters behave like the hydrophilic linear-chain polymers and bottlebrush polymers, respectively, in remodeling the bacteria-mimicking membrane. We attribute this transition to a change from concentrated to semidilute polymer brush regimes as the curvature of silica nanosphere increases. For NPPBs with large silica nanospheres (i.e., curvature → 0), their polymer brushes resemble those on a flat surface. At high graft density, the polymer brushes are in the concentrated regime characterized by uniformly distributed concentration along the brush height^68,69, which interact with bacterial membranes uniformly like a blanket of hydrophilic linear-chain polymers and induce the formation of 3D cubic structures on the model membrane (Fig. 5e). For NPPBs with small silica nanospheres, their curved surfaces give rise to a transition from concentrated to semidilute polymer brush regimes along the brush height⁷⁰. Within the outer layer of NPPBs, the polymer brushes are in their semidilute regime characterized by non-uniformly distributed concentration along the brush height and increasingly relaxed conformational freedom. Even though the diameters of NPPBs are too large to fit into the membranes themselves, their polymer brushes aided by increased conformational freedom are able to interact with bacterial membranes non-uniformly as individually grouped strands, which induce the formation of 2D H_II phase and membrane pores (Fig. 5c, d). Our observation of a size-dependent transition from concentrated to semidilute polymer brush regimes on NPPBs, or from intact bacterial membranes to disrupted bacterial membranes when the NPPBs encounter bacteria, occurs at a threshold nanoparticle size of d_silica ~50 nm. We expect that the threshold boundary may shift depending on the polymer brush size and graft density.

In summary, we aim to clarify the myth about nanoantibiotics and illuminate a path forward for their utility in the clinical battlegrounds by dissecting the antibiotic role of benign nanostructures from belligerent chemical moieties that indiscriminately target both bacteria and mammalian cells. Together with our previous studies on the bottlebrush polymers along this line^22,23, we come to a few preliminary conclusions.

First, nanostructures by themselves are not necessarily antibiotic, i.e., reducing the size of materials does not necessarily enhance their antibiotic activities, as witnessed by the MIC assays of bare silica nanospheres of different sizes against both Gram− E. coli and Gram+ S. aureus (Supplementary Fig. 9). It is the chemical function carried or enabled by the nanostructures rather than the physical size alone that has the potential to deliver a lethal blow to bacteria.

Second, when nanostructures come into play, benevolent chemical moieties that don’t ordinarily debilitate live cells may become active antimicrobials yet remain non-toxic to mammalian cells. Nanostructures are the linchpin of this transformation, as the acquired antimicrobial activity is lost when the nanostructures fall apart²³. We identified two important players that act synergistically underlying the transformation. On one hand, nanostructures give rise to multivalent interactions and structural rigidity needed to bend biological membranes locally upon contact, which could proceed to induce membrane pore formation; on the other hand, the intrinsic curvatures of the membrane lipids need to be conciliatory for this topological transition to happen. Unlike amphiphilic antimicrobial peptides and their synthetic mimics that use hydrophobic interactions to infiltrate and disrupt bacterial and mammalian membranes alike, the membrane disruption induced by hydrophilic nanoparticles occurs exclusively on bacteria instead of mammalian cells, because only microbial membranes laden with negative-intrinsic-curvature lipids are able to offset the energy cost to bend locally into nanopores via spontaneous formation of the saddle-splay curvature. Mammalian membranes rich in zero-intrinsic-curvature lipids can’t do that because the energy cost would be prohibitively high.

Third, the nanostructure-induced antimicrobial activity of hydrophilic nanoparticles is critically dependent on nanoparticle sizes. Taking hydrophilic NPPBs as examples, although large NPPBs may adhere to bacteria and impede bacterial homeostasis processes, they don’t kill bacteria by poking holes and the revealed antibacterial activity is not as potent as small NPPBs. As the size of NPPBs decreases, their polymer brushes are able to interact with bacterial membrane as individually grouped strands due to their increasing conformational freedom along the brush height, which induces pore formation and transforms the hydrophilic nanoparticles into potent antibiotics. We identified a threshold size of silica nanosphere (i.e., d_silica ~50 nm) that sets apart the strong and weak antimicrobial activities of NPPBs; this boundary may shift depending on the polymer brush size, size distribution, graft density, etc. that all affect the transition from concentrated to semidilute polymer brush regimes along the brush height.

Finally, the physical size and shape of membrane-active antimicrobials can be used to develop nanoantibiotics with high selectivity against the two different families of bacteria. Our previous studies demonstrated the role of size on defining the selectivity of nanoantibiotics: while small hydrophilic bottlebrush polymers are potent killers for both Gram+ and Gram− bacteria, long rod-like bottlebrush polymers that are still bactericidal against Gram− bacteria become inactive against Gram+ bacteria because of the selective filter effect of their nanoporous peptidoglycan capsule^22,23. Our study on NPPBs shed new light into this picture. Although small spherical NPPBs are indeed potent killers for both Gram+ and Gram− bacteria, large NPPBs (d_silica > 50 nm) are weak antimicrobials against both families of bacteria. We attribute the different antimicrobial selectivity between spherical NPPBs and rod-like bottlebrush polymers to their different shapes. Because spherical NPPBs have isotropic curvatures, their polymer brushes undergo a transition from antimicrobially-active to antimicrobially-inactive states when the nanoparticle size increases. In contrast, rod-like bottlebrush polymers have anisotropic curvatures along their longitudinal and transverse directions, respectively. As long as the rod diameter is small, polymer brushes on rod-like nanoantibiotics will remain in their antimicrobially-active state even when the rod length is long.

Although our benchtop biological assays revealed the biocompatibility and nanostructure-dependent antibacterial activity of model NPPBs, in vivo studies with animal models are ultimately needed to validate their efficacy for clinical applications. We are not there yet, but our understandings on the antibiotic role of nanostructures point toward many interesting new directions for antibiotics design. For example, membrane-active antimicrobials don’t have to be plagued by cytotoxicity. It’s possible to exploit the nanostructure-enabled multivalent interactions to turn a wide variety of non-toxic but antimicrobially inactive polymers into potent membrane-active antibiotics without deteriorating their cordiality with mammalian cells. Those nanoantibiotics would kill bacteria upon contact yet remain non-toxic when engulfed by mammalian cells. The nanostructure-induced transformation of antimicrobial activity is independent on specific chemical structures, as we demonstrated previously that both the non-bactericidal P4MVP²² and poly(N,N,N-trimethylamino-2-ethyl methacrylate) (PTMAEMA)²³ became potent antimicrobials following the same mechanism. Considering this modular design of nanoantibiotics, we would also envisage multifunctional nanoantibiotics in which the polymer corona helps crack the bacterial membranes while the nanoparticle core carries complementary therapeutic or diagnostic functions. The added benefits of nanostructures would also make it possible to develop antibiotics with triple selectivity, i.e., selectivity between bacteria and mammalian cells, between different families of bacteria, and between in-clinical-use and after-clinical-use states of the same antibiotics. The last concept was demonstrated recently with a nanoantibiotic design that can be dismantled and deactivated by enzymes existing exclusively in natural habitats²³. Given the rapid progress in both the bottom-up and top-down approaches for nanomaterials discovery, our findings suggest that nanoengineering may open a promising new path for the development of clinically viable candidates of nanoantibiotics.