Nanomaterials acquire a biomolecular corona upon introduction to biological media, leading to biological transformations such as changes in protein function, unmasking of epitopes, and protein fibrilization. Ex vivo studies to investigate the effect of nanoparticles on protein-protein interactions are typically performed in buffer and are rarely measured quantitatively in live cells. Here, we measure the differential effect of silica nanoparticles on protein association in vitro vs. in mammalian cells. BtubA and BtubB are a pair of bacterial tubulin proteins identified in Prosthecobacter strains that self-assemble like eukaryotic tubulin, first into dimers and then into microtubules in vitro or in vivo. Förster resonance energy transfer labeling of each of the Btub monomers with a donor (mEGFP) and acceptor (mRuby3) fluorescent protein provides a quantitative tool to measure their binding interactions in the presence of unfunctionalized silica nanoparticles in buffer and in cells using fluorescence spectroscopy and microscopy. We show that silica nanoparticles enhance BtubAB dimerization in buffer due to protein corona formation. However, these nanoparticles have little effect on bacterial tubulin self-assembly in the complex mammalian cellular environment. Thus, the effect of nanomaterials on protein-protein interactions may not be readily translated from the test tube to the cell in the absence of particle surface functionalization that can enable targeted protein-nanoparticle interactions to withstand competitive binding in the nanoparticle corona from other biomolecules.

The integration of inorganic nanoparticles into biomedicine has opened up new frontiers in areas such as photothermal cancer therapy (1,2), disease diagnosis (3,4), targeted drug delivery (5,6), and imaging (7)(8)(9). With their ability to overcome biological barriers due to their subcellular size scale (10), improved synthetic and functionalization design strategies of nanoparticles (NPs) to minimize side effects can position them as versatile tools for precise treatments (11,12). A better understanding of the nano-bio interface, particularly protein-nanoparticle interactions (13), can be leveraged to improve control of NP transport and clearance inside the body (14).

Nanoparticles can modulate protein assembly, depending on the nanoparticle type and experimental conditions, as has been observed in both buffer and in-cell studies. NPs were initially proposed as catalysts for protein fibrillation to develop treatments for amyloidosis (15). Later studies have shown that NPs can both inhibit and promote protein assembly (16)(17)(18) depending on the properties of the protein and the NP, as well as NP dosage (19). NPs can have a size-dependent effect on amyloid fibril formation where larger AuNPs accelerated Aβ (1-40) peptide fibrillation, and smaller AuNPs with the same surface chemistry inhibited formation of large oligomers in buffer studies (20,21). The surface charge of NPs has also been shown to dictate amyloid aggregation kinetics and structural polymorphism (22). Besides interactions with amyloidogenic proteins, NP uptake has been shown to disrupt microtubules (23) and actin filaments (24) in cells. This could be due to nanoparticles binding to polymerized protein structures and disrupting them or by sequestering free protein monomers from the cytoplasm to form a corona around the NPs. Hence, understanding the interplay between nanoparticles and protein assembly holds great promise for both advancing fundamental knowledge and developing nanotherapeutic strategies.

Most of the quantitative nanoparticle-induced protein aggregation or disassembly work has been carried out in buffer due to limited analytical techniques for characterizing the protein corona in live cells (25,26). To predict the in vivo behavior of protein corona-coated nanoparticles, interactions between nanoparticles and proteins under more complex biological or biomimetic conditions need to be probed (27). In this study, we use Förster resonance energy transfer (FRET)-based measurements to compare data from benchtop and in-cell experiments quantitatively for the same protein-nanoparticle combination. To study how the effect of NPs pnas.org on protein complex formation changes in different environments (buffer, lysate, and in living cells) when only nonspecific protein-nanoparticle interactions are present, we selected a pair of bacterial tubulin proteins that have a propensity to self-assemble and unfunctionalized silica nanoparticles with low background fluorescence that will not interfere with FRET measurements. By microinjecting known concentrations of proteins (bacterial tubulin A and B) (28)(29)(30)(31)(32)(33)(34)(35), and nanoparticles into cells, thus avoiding endosomal uptake, quantitative in-cell measurements could be made.

Silica nanoparticles-one of the most widely studied nanomaterials in biomedical applications-were reported to show a dose-dependent or surface ligand-dependent effect on the kinetics and morphology of amyloid fibril formation with macromolecules like amyloid-β fibrils (22), αsynuclein (36), and lysozyme (37). However, the interaction of silica nanoparticles (SiO 2 NPs) with polymerizing cytoskeletal proteins, specifically how they affect the association of tubulin proteins, has not been investigated quantitatively. This interaction is relevant to nanomedicine development because the interference of various NPs with the cell cytoskeleton has been reported to cause unintentional cytotoxicity (23,24,38,39).

In eukaryotes, microtubules are formed from polymerization of two monomers, αand βtubulin. A pair of bacterial tubulins, BtubA and BtubB, which are more similar in sequence to eukaryotic tubulins than to other bacterial proteins, were first identified in Prosthecobacter strains in 2002 (28,29). Similar to αand βtubulins in eukaryotes, BtubA and BtubB can dimerize and subsequently polymerize to form tubules with smaller dimensions than eukaryotic microtubules (30)(31)(32)(33)35). They also do not interact with tubulin in mammalian host cells. The BtubAB heterodimer has a K d in the μM concentration range, ideal for in-cell assembly experiments detected by fluorescence (34).

FRET has been rarely used to study the effect of nanoparticles on protein structure and function (40,41). Here, we use FRET to measure protein self-assembly in the presence of nanoparticles. Labeling of one of the Btub monomers with a donor fluorescent protein (mEGFP) and the other with an acceptor fluorescent protein (mRuby3) provides us with a quantitative tool to measure these protein-protein interactions at the dimer level (42).

Labeled Tubulins Can Still Associate, Even in the Presence of Colloidally Stable Nanoparticles. Unlabeled BtubA and BtubB bacterial tubulin monomers form heterodimers with an in vitro K d ≈ 7 μM in HMK (HEPES, magnesium acetate, and potassium acetate) buffer (34) and further assemble into minimicrotubules upon addition of GTP as observed by transmission electron microscopy (TEM) (30,32,33,35). In previous work, we have shown that labeling the C terminus of BtubAB with fluorescent proteins mEGFP and mRuby3, respectively (Fig. 1A and SI Appendix, Fig. S1), enables the use of mammalian U-2 OS cells as a platform to measure the effect of crowding and sticking interactions in cells on the dimer dissociation constant of the tubulin proteins, although they form dimers with a higher K d ≈ 24 μM in buffer than the unlabeled wild type. The distance between C termini of BtubA and BtubB within an adjacent heterodimer is 3.6 nm whereas between the C termini of BtubA and BtubB proteins in a cross-dimer is 10.1 nm (Fig. 1B). The mEGFP-mRuby3 FRET pair can reach 95% of FRET efficiency at 3.6 nm which drops off to less than 5% at fluorophore distances greater than 10 nm (42). BtubA labeled with mEGFP donor protein and BtubB labeled with acceptor protein mRuby3 are referred to as BtubA-mEGFP and BtubB-mRuby3, respectively, in the rest of the manuscript (Fig. 1A).

The fluorescently labeled BtubAB proteins retain their ability to self-assemble into tubules in buffer (Fig. 1C). We also observed formation of tubules from unlabeled BtubAB in mammalian cell lysate (Fig. 1D) collected from U-2 OS cells. A control experiment with cell lysate without BtubAB showed only protein aggregates (SI Appendix, Fig. S2), confirming that the tubules result from the polymerization of BtubAB.

Here, we use solid anionic SiO 2 NPs with 27 nm diameter, which cause no optical interference in the 450 to 700 nm region, to study the role of NPs in promoting or inhibiting self-assembly of BtubAB using FRET. The NPs were commercially purchased and used without further modification. NPs were characterized using inductively coupled plasma optical emission spectroscopy (ICP-OES) for concentration, TEM for size, and dynamic light scattering (DLS) for colloidal stability (SI Appendix, Fig. S3). FRET data, monitored as the sensitized emission intensity of acceptor fluorophore, were corrected for background scattering and monomeric protein association with the NPs (SI Appendix, Figs. S4-S10). Spectra were normalized to intensity instead of integrated areas (SI Appendix, Figs. S8 and S9) to maintain consistency with in-cell data analysis (Materials and Methods). FRET signal increases in the red channel are due to BtubA/BtubB dimerization, with some contribution from tubules and little contribution from oligomers (42).

Function of Concentration. The change in red emission intensity due to FRET between 7 μM BtubA-mEGFP and 7 μM BtubB-mRuby3 as a function of varying concentrations of silica NPs was measured for NP concentrations between 0 and 960 nM (Fig. 2A). The HMK buffer used has an ionic osmolarity that is similar to the Escherichia coli cytoplasm and has been used previously to study bacterial tubulin assembly in vitro (35,43). The order of addition of proteins to the NP solution does not matter, suggesting that the system is at equilibrium (SI Appendix, Scheme S1 and Fig. S11). The intensity of the FRET signal at 580 nm increases up to ≈ 230 nM NPs and then levels off. This indicates that SiO 2 NPs promote BtubAB dimerization over buffer at low NP concentrations, but additional nanoparticles (more than ~ 1 NP/60 proteins) do not further promote association (where the theoretical maximum surface coverage ratio of an ≈ 3 nm diameter protein adsorbing on a 27 nm diameter NP is ≈ 320 proteins/NP). Thus, in vitro FRET data establish that the NPs enhance BtubAB assembly under buffer conditions in a concentration-dependent manner. The nanoparticle surfaces act as concentrators in an otherwise dilute buffer environment that forces proximity between the tubulin monomers, enabling at least dimerization and possibly higherorder tubule assembly. The K d of FRET-labeled BtubAB does in fact decrease with the addition of a crowder like Ficoll, from 24 μM to 10 μM, because crowders increase the effective protein concentration in solution through the excluded volume effect (42).

We confirmed that exposure to SiO 2 NPs does not completely inhibit polymerization of labeled BtubAB proteins (Fig. 2B) using negative-stained TEM. Small tubules were observed for samples containing three different concentrations of NPs: 11.8 nM NP, 118 nM NP, and 236 nM NP. However, the abundance of tubules on the TEM grid was considerably lower for all the three samples containing NPs compared to a control sample of 7 μM equimolar mixture of BtubA-mEGFP and BtubB-mRuby3 without NPs (SI Appendix, Fig. S12). The NPs do associate with tubules (Fig. 2B) and the NPs also appear to form a protein corona around them as expected (SI Appendix, Fig. S12D). The concentration of free dimer in solution is presumably reduced due to protein adsorption onto the NP surfaces, which could lead to a diminished amount of tubule formation compared to the control sample. Our in vitro FRET data demonstrating promotion of BtubAB dimerization in the presence of NP surface corroborate this hypothesis. Specific binding of nanomaterials to the recognition motif of Aβ peptides has been shown to reduce aggregation and thus inhibit amyloid fibrillation (44). If adsorption to NP surface blocks off binding sites of the proteins to form protofilaments, this can also lead to a decrease in the number of tubules observed. Overall, in buffer, we conclude that protein heterodimerization is promoted by the presence of NPs if they are not in excess, while extended polymerization is reduced, but not eliminated, by the presence of NPs.

Assembly via FRET in Microinjected Cells. Although data from buffer measurements such as the above suggest a large nanoparticle effect, the in-cell effect may be different due to competitive binding of both nanoparticles and BtubAB to metabolites, inorganic ions, and macromolecules present in the cell. Hence, we compared the in vitro result to the association of BtubAB on nanoparticles in the cytoplasm of U-2 OS cells as a model for in-cell binding. We monitored protein assembly by using the increase in red fluorescence from labeled BtubB.

In order to ensure well-defined relative BtubAB/nanoparticle dosing in the cell, microinjection of fluorescently labeled tubulin proteins and nanoparticles was performed (as opposed to gene expression and NP incubation). Commercially available adherent human osteosarcoma (U-2 OS) cells were used to perform in-cell experiments for ease of microinjection, with a sufficiently large cytoplasmic area for microscopy. Introduction of fluorescently labeled proteins in U-2 OS cells has been successfully used for FRET measurements with an in-house fluorescence microscopy setup (SI Appendix, Fig. S13A) (45). Injecting labeled BtubAB proteins into the cytoplasm of this same cell line did not affect cell morphology or function. The difference in K d of BtubA-mEGFP and BtubB-mRuby3 heterodimerization in U-2 OS cell cytoplasm is significant, 3 ± 2 μM in cell versus 24 ± 1 μM in buffer (42). Thus, the Btub proteins bind more tightly in-cell than in buffer. Microinjection of the BtubAB+NP mixtures into cells, instead of incubating NPs with the cell media for cellular uptake and transfection of fluorescent BtubAB, was used to quantify the concentration of proteins and NPs in-cell (Fig. 3A and SI Appendix, Fig. S13). Compared to endocytosis, microinjection also gives better control over the time and location of introduction of the nanomaterials into cells (46). Microinjection of inorganic NPs into living cells has been reported before to study cytotoxicity (47), intracellular distribution and mobility (48,49), and autophagosome capture (50).

We were able to image healthy U-2 OS cells within an interval of 30 to 45 min after microinjecting a premixed protein + NP solution at a concentration of NP as high as 400 nM after filtration. The cells remained alive postinjection, as evidenced by cell morphology and adherence (Fig. 3B and SI Appendix, Figs. S13B and S14). The average height of a U-2 OS cell is 4 μm (42), and red channel intensities are calibrated using in vitro slides of known concentration of BtubB-mRuby3 so that the data from individual cells are consistent relative to one another (see Materials and Methods for data analysis). To compare with and without NP conditions for in-cell FRET, the FRET signal intensity as a function of BtubB-mRuby3 concentration alone using data points from only 1:1 ratio of BtubA-mEGFP: BtubB-mRuby3 was plotted. In Fig. 3C, data with (dark blue) and without NPs (pink) overlap indicating that NPs have little effect on protein association in-cell. Data from unhealthy cells, highly bright cells that saturated the FRET channel intensity, as well as cells with nonuniform cytoplasm pnas.org fluorescence (SI Appendix, Fig. S15) have not been included in Fig. 3C. Such cells may contain Btub-derived microtubules (microtubule formation from Btub monomers in cell has a lag phase (42)) or other aggregates or have altered cytoplasm composition.

A comparison of in-cell and in vitro experiments, then, reveals that conclusions drawn from in vitro cannot be always translated to in vivo behavior. In this case, protein dimerization is enhanced by NPs in vitro but NPs in the cell (microinjected, to avoid trafficking within endosomes that would happen by cellular uptake of nanomaterials) have no effect on protein dimerization in the cytoplasm. Various properties of the cellular milieu could be responsible for damping the NP effect on Btub association: the crowded cellular environment with its increased osmotic pressure; or the presence of many small molecules or macromolecules that interfere with NP activity by either binding to the NPs instead of the fluorescently labeled bacterial proteins, or interacting with BtubAB to shift in the binding equilibrium of the BtubAB proteins in-cell compared to in vitro.

Interactions. What constituents in the cell could be responsible for reducing the effect of nanoparticles on Btub assembly compared to HMK buffer: other proteins or small molecules? We started with bovine serum albumin (BSA), a model protein crowder used to improve the biocompatibility of materials (51), and known to adsorb readily onto inorganic nanomaterial surfaces (52,53). Addition of 0.5 mg/mL BSA in HMK buffer was initially tested as a proxy for the proteins in the cell to check whether it can compete with the BtubAB proteins for NP surface sites. BSA did not cause a significant decrease of FRET from BtubAB: The Btub proteins remained adsorbed onto the NP surface even after the solution was centrifuged and the NP pellet resuspended in solution (Fig. 4A), and at a 10x higher concentration (5 mg/mL) of BSA in solution (Fig. 4B), the NPs still induce a trend in FRET enhancement that is similar to results in 100% HMK buffer (SI Appendix, Fig. S16B). This indicates that BtubAB proteins are not displaced significantly from the NP surface due to competition from BSA. Individual cytoplasmic proteins have concentrations well below 5 mg/mL (equivalent to 100 μM for a 50 kDa protein), so any such protein would have to have a very high affinity for silica nanoparticles to dislodge Btub proteins from the nanoparticle corona.

Next, we turned to U-2 OS cell lysate (see Materials and Methods for preparation protocol of cell lysate) to see whether it could mimic the effect of the cytoplasm, as it contains the macromolecules in a U-2 OS cytoplasm, albeit at lower concentration. Experiments in cell lysate give insights into the evolution of the protein corona during NP cellular internalization (53-55) and provide an environment closer than buffer or BSA to the molecularly complex mammalian cellular environment where crowding, sticking, and quinary interactions come into play (56). U-2 OS cell lysate was prepared using Pierce™ IP lysis buffer (Materials and Methods). We prepared solutions containing a constant 7:7:0.23:500 μM concentration of BtubA-mEGFP:BtubB-mRuby3:NP:GTP with increasing concentrations of cell lysate. Adding cell lysate reduces red emission intensity due to FRET (Fig. 4C), so macromolecules and small molecules in the lysate are candidates for reducing the effect of nanoparticles on Btub assembly compared to HMK buffer.

However, lysate is dispersed in lysis buffer, which is used to break open the cell and extract the intracellular components. Lysis buffer by itself is known to mimic nonsteric in-cell interactions in a test tube (57,58). Thus, we performed control experiments with lysis buffer only. The lysis buffer we used to make lysate contains osmolytes and organic molecules, including EDTA (a metal ion chelator) and a neutral surfactant named NP-40 that contains both a poly(ethylene)glycol (PEG) chain and an alkyl chain (see Materials and Methods for full buffer composition). PEGylation can impart protein-resistance to the material surface (59,60). Furthermore, PEG is also known to associate with proteins as a protectant (61). Lysis buffer by itself causes the same reduction in nanoparticlemediated FRET as cell lysate dispersed in lysis buffer, at both 2.5% and 67% lysis buffer (% v/v mixed with HMK buffer) conditions (Fig. 4D). The labeled individual Btub monomers have lower affinity for the NP surface in 67 % lysis buffer than in 100 % HMK (SI Appendix, Fig. S16 D-F), hence heterodimerization is no longer promoted due to colocalization in the protein corona in the presence of lysis buffer. The nonionic NP-40 surfactant containing both the PEG chain and hydrophobic alkyl chain can possibly interact with the NP (SI Appendix, Fig. S17) (62) and the protein. Addition of small amounts of surfactants has indeed been shown to strip protein corona off nanoparticle surfaces (63,64). We conclude that the cell lysate does not displace BtubA or BtubB from the nanoparticle surface, whereas lysis buffer does. Thus, in the cell cytoplasm, a combination of crowding, and competitive binding from proteins with higher affinity for NP surface than BSA or small molecules disrupting Btub protein corona formation on the NPs remain as reasonable candidates for the reduction of the effect of nanoparticles on Btub assembly compared to that observed in HMK buffer.

FRET was used as a quantitative tool to measure how SiO 2 NPs influence bacterial tubulin association at a heterodimer level in different environments with increasing complexity, going from buffer to mammalian cell cytoplasm. The surface of the unfunctionalized SiO 2 NPs used in this work is expected to be overcoated with different types of biomolecular coronas depending on the suspension environment due to nonspecific interactions, and this study describes the zeroth order case. The results shown here reconcile seemingly contradictory conclusions in the literature (NPs inhibit, or promote, or have no effect on protein assembly). We find that in an HMK aqueous buffer, generic NPs do promote protein-protein association in a concentration-dependent manner: As the NP concentration increases, protein-protein association is increased up to a point, then plateaus, likely due to surface saturation of the NPs. Yet in the U-2 OS cytoplasm and in lysis buffer containing a PEGylated lipid-like detergent and EDTA, the NP promotion effects are damped, to the point of NPs having no apparent effect on tubulin protein assembly. In contrast, addition of BSA protein (at concentrations higher than Btub proteins) to HMK buffer, or cell lysate to lysis buffer, do not change nanoparticle-induced trends in tubulin protein association pnas.org significantly. As surface electrostatics change and there is competition from other molecules to interact with both the NP and Btub proteins in microenvironments containing more than just simple buffer salts, the Btub proteins are no longer able to adsorb readily on the NP surface, which leads to a loss in the enhancement of BtubAB association that the NPs could cause under laboratory buffer conditions.

Protein Expression and Purification. BL21 codon cells (Agilent) were transformed according to the manufacturer's protocol with pDream2.1 plasmid (GenScript Biotech). The expression vector is added to thawed competent E. coli cells before transformation. Heat-shocked cells are mixed with S.O.C. medium and shaken at 200 rpm and 37 °C. Transformed cell culture is then plated onto LB agar plates with ampicillin and incubated at 37 °C overnight. A few colonies of expression strain are added into 10 mL of LB with ampicillin, followed by shaking at 37 °C for ~4 h until the solution becomes sufficiently turbid. The culture is scaled up by adding the starter culture to a larger volume of LB with antibiotic. The large culture is incubated at 37 °C for 3 to 4 h until its optical density at 600 nm reaches 0.6 to 0.8. Protein expression is induced by adding IPTG to a final concentration of 0.1 to 1.0 mM. Cultures are shaken overnight (15 to 18 h) at 17 °C. Afterward, the cells are centrifuged at 5,000 rpm for 20 min at 12 °C. The pellet is resuspended in a lysis buffer (500 mM NaCl, 50 mM Na 3 PO 4 , and 20 mM imidazole, pH = 7.4) with protease inhibitor. After adding DNase, the solution is sonicated for 1 h to lyse the cells and degrade everything but the proteins. After sonication, the solution is centrifuged again at 1,000 rpm for 20 min at 12 °C. The supernatant containing protein is filtered with a 0.45 micron filter followed by a 0.22 micron filter. The filtered solution is then passed through a AKTA FPLC and a His-trap affinity chromatography column to purify protein from filtered cell lysate. The His tags in the Btub proteins allow them to adhere to the column which contains chelating Ni 2+ ions. The protein is then eluted out using a buffer with 200 mM concentration of imidazole. A SDS-PAGE gel is run using this solution to estimate the weight and fractions of proteins acquired. The protein solution is then dialyzed at 4 °C to perform buffer exchange where the exchange buffer is HMK (50 mM HEPES, 5 mM MgAc, 350 mM KAc, and 1 mM EGTA, pH = 7.5). Dialysis buffer is changed every 4 h for three times. After dialysis, the concentration of the protein is calculated by measuring the absorbance of the chromophore (mEGFP for BtubA, mRuby3 for BtubB, and tryptophan for unlabeled protein). SDS-PAGE and MALDI mass spectrometry are used to identify the purified protein. The protein is then split into aliquots, flash-frozen with liquid N 2 , and stored at -80 °C.

TEM for In Vitro Imaging of Tubulin Fibers. Protein solutions are drop-cast onto an EM carbon grid and imaged on the same day. The incubation time of the sample on the grid is 5 to 7 min. The excess sample is wicked off from the grid, followed by 2 to 5 min of incubation with staining solution (0.2% ammonium molybdate at pH 6.2). Excess stain is wicked off the grid which is left to air dry for 15 min before being put into the microscope. For sample preparations, 5-7 μM of Btub monomers in HMK buffer with 1 mM GTP were used. TEM images are taken with a 75 keV, 120 keV, or 200 keV electron beam.

Fluorescence Spectroscopy for In Vitro FRET Measurements. FRET data discussed in all figures are emission intensities from acceptor fluorophore of the FRET pair (mRuby3 labeled proteins). Changes in emission spectra of BtubA-mEGFP and BtubB-mRuby3 proteins with addition of increasing concentrations of NPs in HMK buffer (0.5 mM GTP added) are monitored using a fluorometer. Concentrations of the fluorescent proteins are calculated from their absorption spectra using 488 = 56,000 M -1 cm -1 for BtubA-mEGFP and 558 = 128,000 M -1 cm -1 for BtubB-mRuby3. Before each measurement, the sample in the cuvette is left to equilibrate at room temperature for 10 min after each addition and pipetted for better NP dispersion. The sample is excited at 450 nm, and the emission intensity from 460 to 700 nm is measured (bandwidth of excitation and emission = 5 nm).

In order to extract the FRET signal out of the donor-acceptor emission spectrum, we measure spectra of three different solutions for each concentration of NP: 1) spectrum of 7 μM of donor (green label) protein excited at 450 nm, 2) spectrum of acceptor (red label) protein by itself excited at 450 nm, and 3) spectrum of equimolar mixture (7 μM) of donor + acceptor protein pair excited at 450 nm (all measured at same PMT voltage on the same day on the fluorometer).

To get an intensity of FRET signal from spectrum 3, we first subtract spectrum 2 from spectrum 3 to remove a contribution to the peak at 580 nm from the direct excitation of red protein by excitation at 450 nm (refer to SI Appendix, Fig. S8 for additional data processing details). Spectrum 1 and spectrum 3 (after correcting for direct excitation of acceptor) show how signals vary as a result of energy transfer between acceptor and donor proteins following BtubAB dimer formation. Due to the presence of acceptor protein in spectrum 3, we observe a decrease in donor emission intensity at the peak centered around 507 nm and an increase in signal at peak centered around 580 nm compared to spectrum 1. To subtract the contribution from the donor emission to the peak at 580 nm, we first normalize the donor peaks on both spectra (SI Appendix, Fig. S8c) and then subtract the normalized spectrum 1 from the normalized spectrum 3. This normalization is done by using the intensity at 507 nm (almost exclusively donor emission). The obtained spectrum (SI Appendix, Fig. S8c) gives us the contribution to peak at 580 nm from FRET only. The background corrected spectrum thus obtained contains a small positive (480 to 510 nm)/negative (510 to 560 nm) baseline in the donor emission peak region because the spectrum of BtubA-mEGFP shifts slightly in the presence of FRET (SI Appendix, Fig. S8 C and D). For comparing FRET data of the same donor-acceptor pair at different NP concentrations, the acceptor emission intensity at 580 nm is corrected for changes in emission of acceptor protein due to any quenching and for scattering caused by nanoparticles at the particular concentration (SI Appendix, Figs. S7 and S9). The corrected red emission intensity plotted in Figs. 2A and 4D is defined as the emission intensity of acceptor at 580 nm corrected for cross talk and bleed through. In Fig. 2A, as data points are measured from samples containing different concentrations of NPs, the y-axis has also been corrected for quenching due to NPs (SI Appendix, Fig. S9).

Cell Culture. Human bone osteosarcoma epithelial cells (U-2 OS ATCC HTB-96, Manassas, VA) were cultured and grown to 70% confluency in DMEM (Corning, Corning, NY) + 1% penicillin-streptomycin (Corning) + 10% fetal bovine serum (FBS, ThermoFisher Scientific) media. Immediately prior to imaging, cells were placed in imaging chambers filled with Opti-MEM media (ThermoFisher Scientific) supplemented with 10% FBS.

Microinjection and Fluorescence Microscopy of Cells. Protein sample for microinjection is prepared by first switching the buffer from HMK to 0.5 M KCl+20 mM KPhos, pH 7.4 buffer. A 20 μL solution containing a total of 140 to 180 μM BtubA-mEGFP+BtubB-mRuby3 protein is prepared, and for the blue data points in Fig. 3C, which also contain silica NPs in the mixture, 470 to 930 nM NPs is added in. This mixture is passed through a 0.22 μm syringe filter before using for microinjection. NP concentrations post-filtration change to 260 to 410 nM as measured by ICP-OES. Three microliters of this solution is transferred into the glass micropipette tip, and bubbles are removed before loading onto a femtojet injector.

Cells for microinjection are cultured overnight in an Ibidi 60 μ-Dish (35 mm high, Grid-500 ibiTreat). Then, 0.3 mL of cells from a 10 mL, 90 % confluent flask is transferred to the imaging dish, making up to 2 mLwith DMEM +FBS +penicillin media, and is left to adhere to the bottom of the dish overnight in a 37 • C incubator. The media is replaced with Opti-MEM + 10 % FBS before bringing to the microscope to reduce background fluorescence. Cells are first focused for microinjection using the 40x objective. The injection pressure is 250 to 300 hPa, and the injection time is 300 ms. After microinjection, cell media is replaced with fresh imaging media and brought back to the microscope. Imaging two-color fluorescence from microinjected cells is done after focusing cells using the 63× objective. The donor label is excited using a blue LED and the acceptor label with a green diode laser. Under green LED excitation, the mRuby3 label was exposed for 20 s before imaging each cell to maintain consistency between experiments, as mRuby3 bleaches faster than mEGFP upon irradiation.

Cell imaging is conducted using a Carl Zeiss Axio microscope body (Zeiss). Excitation light was generated by a white UHP-T2 LED head (Prizmatix) passing through an ET470/40× bandpass filter (Chroma) with a T495lpxt dichroic for mEGFP excitation and FRET, and a T580/25× bandpass filter and T600lpxr dichroic for mRuby3 excitation. A Zeiss 63×/0.85 NA N-Achroplan objective focuses the excitation light on cells. Emission light, directed through an ET500lp filter (Chroma) and split by a T600lpxr dichroic (Chroma), is recorded by a CMOS camera (Lumenera, LT225 NIR/SCI CMOS detector) at a 60 Hz frame rate with 16 ms integration times.

In-Cell Data Analysis. Corrected red emission intensity in Fig. 3C is defined as the background subtracted intensity from acceptor fluorophore emission collected in the 600 to 750 nm region from microinjected cells. All data are analyzed using MATLAB. Images are taken in three different channels: green channel (excitation: 470 nm; emission: 500 to 600 nm), FRET channel (excitation: 470 nm; emission: 600 to 750 nm), and red channel (excitation: 580 nm; emission: 600 to 750 nm). Green and FRET channels are imaged simultaneously, while the red channel is imaged within 1 min to be used for concentration calibration. Cells microinjected with FRET labeled BtubAB proteins are generally brighter in the cytoplasm and darker in the nucleus, so the cytoplasm areas can be identified and cropped using MATLAB code. Average intensities of the cytoplasm area are calculated for each individual cells. The average cell height was estimated to be 4 μm for U-2 OS cells based on z-stack imaging using confocal microscopy (42). Intensities of the red channel are used for concentration calibration to obtain values for the x-axis and y-axis of Fig. 3C and SI Appendix, Fig. S15A (see SI Appendix, Fig. S13C for in vitro concentration calibration curve using 10 μm spacers). Intensities of the green channel are used for bleed-through correction. The intensity of the FRET channel corrected for bleed-through contributed to the z-axis of SI Appendix, Fig. S15A and the y-axes of Fig. 3C and SI Appendix, Fig. S15B.

Cell Lysate Preparation. U-2 OS cells are grown to 70% confluence in T-175 flasks (Thermo Fisher Scientific). Cell media is aspirated and washed with 30 mL PBS in each flask before trypsinizing the cells. Then, 3 mL trypsin is added to each flask and incubated for 5 min. Detached cells are collected in a 15 mL conical tube after neutralizing the trypsin by adding cell media. The collected cells are pelleted by centrifuging at 450 × g for 5 min. The supernatant is aspirated, and 10 mL ice-cold PBS is added to wash the cells. The cells are pelleted again by centrifuging at 450 × g for 5 min, and the supernatant is removed. One milliliter of cold lysis buffer is prepared by mixing 980 μL Pierce™ IP lysis buffer (Thermo Fisher Scientific, 87787) and 20 μL of 100× protease inhibitor cocktail (Thermo Fisher Scientific, 78442). Composition of Pierce™ IP lysis buffer is 25 mM Tris-HCl, 150 mM NaCl, 1 mM EDTA, 1% NP-40, and 5% glycerol with measured pH = 7.5. The cell pellet is resuspended to 750 μL of lysis buffer and incubated on ice for 30 min for lysing. The lysed cells are centrifuged at 13,000 × g for 5 min to separate out the cell debris from soluble proteins and small molecules. The supernatant is collected into an Eppendorf tube as cell lysate. Collected cell lysate dispersed in 100% lysis buffer is stored at -20 °C. Cell lysate stock contained 20 mg/mL protein concentration as determined using a BCA assay (Thermo Fisher Scientific, 23227).