Chaotropic Perturbation of Noncovalent Interactions of the Hemagglutinin Tag Monoclonal Antibody Fragment Enables Superresolution Molecular Census

Abstract

Antibody−antigen interactions represent one of the most exploited biomolecular interactions in experimental biology. While numerous techniques harnessed immobilized antibodies for nanoscale fluorescence imaging, few utilized their reversible binding kinetics. Here, we investigated noncovalent interactions of the monoclonal hemagglutinin (HA) epitope tag antibody, 12CA5, in the fixed cellular environment. We observed that the use of a chaotropic agent, potassium thiocyanate (KSCN), promoted the dissociation of the 12CA5 antibody fragment (Fab), which already displayed faster dissociation compared to its immunoglobulin G (IgG) counterpart. Molecular dynamic simulations revealed notable root-mean-square deviations and destabilizations in the presence of KSCN, while the hydrogen-bonding network remained primarily unaffected at the antigen-binding site. The reversible interactions enabled us to achieve a superresolution molecular census of local populations of 3xHA tagged microtubule fibers with improved molecular quantification consistency compared to single-molecule localization microscopy (SMLM) techniques utilizing standard immunofluorescence staining for sample labeling. Our technique, termed superresolution census of molecular epitope tags (SR-COMET), highlights the utilization of reversible antibody−antigen interactions for SMLM-based quantitative superresolution imaging.

Graphical Abstract

INTRODUCTION

The high specificity and affinity of antibodies render them an indispensable tool in biomedical research, diagnostic testing, and therapeutics.¹ One of the key aspects of antibody−antigen interactions is that they involve a plethora of weak noncovalent interactions. The interactions between an epitope, the binding site on the antigen, and the paratope, the site on the antibody that recognizes the antigen, are known to occur primarily in two steps.² First, long-range hydrophobic and electrostatic interactions attract an antibody to its cognate epitope on the antigen. The second step involves short-range hydrophobic interactions, van der Waals forces, and hydrogen bonding.^2,3 These noncovalent interactions render paratope−epitope interactions reversible. Hence, in standard immunofluorescence labeling, a postfixation step is commonly used to covalently link antibodies to their labeled targets.^4,5 While immunofluorescence labeling is widely used in single-molecule localization microscopy (SMLM), such as stochastic optical reconstruction microscopy (STORM)⁶ and DNA-based points accumulation in nanoscale topography (DNA-PAINT),⁷ the labeling technique introduces linkage artifacts at the nanoscale. In addition, relatively low labeling densities have hindered the full potential of quantitative SMLM.^8,9 Recent studies have developed labels against epitope or protein tags to achieve high labeling densities, which include antibody fragments,¹⁰ green fluorescent protein (GFP) nanobodies, and peptide tags.^11,12 Live imaging using Frankenbodies against epitope-tagged proteins has also been reported.¹³ The reversible nature of the paratope−epitope interaction may have unexplored potential to overcome artifacts inherent to immunofluorescence labeling in SMLM. A recent study has successfully used hybridoma technology to generate fast-dissociating antibody fragments (Fabs) against different molecular epitope tags for image reconstruction by integrating exchangeable single-molecule localization (IRIS).¹⁰ The work to screen for rapidly dissociating antibodies has begged the question of whether commercially available antibodies can attain similar characteristics. Despite the potential to overcome the immunolabeling artifacts, little is known about whether and to what extent transient antibody interactions are advantageous over existing SMLM techniques using immunofluorescence labeling.

Here, we introduce a superresolution census of molecular epitope tags, or SR-COMET, as a straightforward quantitative SMLM technique. We investigated the mouse monoclonal 12CA5 Fabs against the hemagglutinin (HA) epitope tag (YPYDVPDYA). We observed that the chaotropic perturbation of the noncovalent hydrophobic interactions promoted reversible interactions of the Fab. Molecular dynamics simulations further validated the destabilization introduced by the chaotropic agent. Leveraging transient 12CA5−Fab−HA interactions, we achieved consistent superresolution molecular census without immunolabeling artifacts associated with direct STORM (dSTORM) and DNA-PAINT.

RESULTS AND DISCUSSION

Fluorescence Decay Measurements Revealed Faster Dissociation of the 12CA5-Fab Compared to 12CA5-IgG at the Ensemble Level.

The paratope−epitope interactions are maintained by a combination of electrostatic interactions, hydrogen bonding, van der Waals interactions, and hydrophobic interactions (Figure 1a). These interactions are representative of the weak forces present in the 12CA5−HA interactions. The noncovalent nature of these interactions renders them reversible. We first investigated how the number of antigen-binding sites, or antibody valency, affected the antibody-binding dynamics (Figure 1b). We chose the HA tag antibody 12CA5 as a model system against the multimeric 3xHA epitope tag. Multimeric 3xHA tags are commonly used against HA antibodies in biological practices, as it enhances the frequency of interaction by tripling the effective concentration of the substrate.¹⁴ This design may increase the on-rate of the HA−antibody interaction similar to the strategy employed in a recent study.¹⁵ Furthermore, the 3xHA tag may improve the accessibility of the epitope tag. We generated a U2OS cell line expressing 3xHA tags on the N-terminus of the α-tubulin. For visualization, we conjugated the 12CA5-IgG with Alexa Fluor 647 (AF647) and performed papain digestion to produce the corresponding dye-labeled antibody fragment 12CA5-Fab-AF647 (Figure S-1, Methods). Figure 1c and d show the immunostained U2OS cells using the 12CA5-IgG-647 and 12CA5-Fab-AF647, respectively. Partial staining by the 12CA5-Fab-AF647 is notable in Figure 1d, indicating a possible faster dissociation of the Fab.

Figure 1. Noncovalent weak interactions govern antibody-antigen interactions and render them reversible for the 12CA5-IgG and Fab.

a. Schematic representation of antibody-antigen interactions via hydrogen bonds (H-bond), van der Waals interactions, hydrophobic interactions, and ionic bonds. b. Schematic representation of the reversible nature of noncovalent interactions for the full-length immunoglobulin G antibody (IgG) and antibody fragment (Fab). c. Immunofluorescence image of HA-tagged microtubules labeled with the 12CA5-IgG-AF647. d. Immunofluorescence image of HA-tagged microtubules labeled with the 12CA5-Fab-AF647. e. Normalized fluorescence intensity decay curves for the 12CA5-IgG-AF647 and 12CA5-Fab-AF647 showing different dissociation rates. Each curve represents the mean and standard deviation from three replicates. f. Comparison of the dissociation rates, k_off, of the 12CA5-IgG and 12CA5-Fab. Error bars represent standard deviation. Scale bars: 10 μm.

To quantify dissociation rates in the cell, we employed a fluorescence intensity decay model.¹⁶ The rationale for utilizing this approach over in vitro affinity assays is that the quantification directly relates to our single-molecule imaging condition in the cell. For instance, chemical fixation and the cellular environment may affect the dissociation of antibody molecules. However, fluorescence characterizations are subject to photobleaching. To factor in this phenomenon, we modeled the fluorescence intensity decay as a function of fluorophore photobleaching and dissociation of dye-conjugated antibodies and antibody fragments. The two processes are independent and described as

$$I\left(t\right)=B\left(t\right)\times D\left(t\right)=A{e}^{-(k_b+k_{off})t}+F_0$$ (1)

where I(t) is the overall intensity, B(t) and D(t) are functions of photobleaching and dissociation, respectively. A and F₀ are constant terms. F₀ accounts for the local fluorescence background, k_b represents the photobleaching rate, and k_off represents the dissociation rate. To obtain the dissociation rate, we first characterized the photobleaching rate.

To assess the photobleaching rate, we immunolabeled and postfixed 3xHA-expressing U2OS cells. Postfixation of the immunolabeled sample prevents antibody dissociation; therefore, the observed fluorescence decay is solely attributed to photobleaching. We acquired a time-lapsed video for 150 min using a 0.3 W cm⁻² illumination at 3 min intervals. The measured fluorescence intensity decay can be obtained from

$$B\left(t\right)=B_0+A{e}^{-k_{b}t}$$ (2)

where B₀ accounts for the local fluorescence background.

The fluorescence intensity decay analysis revealed similar photobleaching rates of 0.007 ± 0.001 min⁻¹ for both the 12CA5-IgG-AF647 and 12CA5-Fab-AF647 (Table S-1). Next, to characterize the dissociation rates, we repeated the previous analyses without postfixation. Thus, the fluorescence decay was associated with photobleaching and dissociation, described as

$$D\left(t\right)=D_0+A{e}^{-\left(k_b+k_{off}\right)t}$$ (3)

where D₀ similarly accounts for the local fluorescence background. Substituting k_b with the previously obtained value, we obtained dissociation rates of 0.012 ± 0.0004 min⁻¹ and 0.026 ± 0.001 min⁻¹ for the 12CA5-IgG and 12CA5-Fab, respectively (Table S-1). A related study reported a dissociation rate of 0.066 min⁻¹ for HA tag antibody 4B2 with the single HA.¹⁷ Our decay analyses indicated a slightly slower dissociation for the HA tag antibody 12CA5-IgG and 12CA5-Fab. Decay experiments displayed an approximate 2-fold increase in the dissociation rate for monovalent 12CA5-Fabs compared to the bivalent 12CA5-IgG (Figure 1e and f). We next investigated the dissociation of the 12CA5-Fab at the single-molecule level for SMLM.

Reversible Interactions Enabled SMLM Using the 12CA5-Fab.

We next characterized the fluorescent properties of single 12CA5-Fab molecules. Figure 2a shows a representative two-step photobleaching of an immobilized 12CA5-Fab-AF647 on a clean coverglass. NanoDrop measurements indicated an average Fab-to-dye-labeling ratio of approximately 1:2 (Figure S-1d). We next incubated U2OS cells expressing 3xHA-α-tubulin with 1 nM of the 12CA5-Fab-AF647 in the imaging buffer. Figure 2b shows single-molecule binding events detected from 2000 image frames acquired with 5 s intervals. Compared to the photobleaching of the immobilized Fab in Figure 2a, the distinct, sharp intensity profiles in Figure 2b indicate that a fraction of the 12CA5-Fab-AF647 constantly probed and dissociated from the 3xHA tag in the cell. Figure 2c displays a reconstructed SMLM image from 12 000 frames. Given the relatively slow antibody kinetics, we applied a camera exposure of 800 ms and achieved a signal-to-noise ratio (SNR) of approximately 10 (Figure 2d). Individual panels numbered 1−4 in Figure 2d show representative single-molecule localization events from four consecutive image frames.

Figure 2. Single-molecule localization microscopy utilizing transient noncovalent interactions of the 12CA5-Fab.

a. Representative stepwise photobleaching intensity trace for a 12CA5-Fab-AF647 immobilized on a clean coverglass. b. Representative time trace of single-molecule events detected from a 3×3 pixel area with 1 nM of the 12CA5-Fab-AF647 in a fixed U2OS cell stably expressing 3xHA-α-tubulin. Inset shows a representative single-molecule event. c. Reconstructed superresolution image from 12,000 frames using 1 nM of the 12CA5-Fab-AF647 in the imaging buffer. d. Representative single-molecule localization events. High-magnification panels 1–4 represent four consecutive image frames. The cross-sectional intensity profile across the indicated yellow bar shows the signal-to-noise ratio (SNR). e. High-magnification view of the boxed area in c. f. Gaussian fitted cross-sectional profiles of microtubule fibers marked in e. Scale bars: 10 μm (c) and 1 μm (d and e).

Figure 2e and f reveal superresolved, adjacent microtubules with a full-width at half-maximum (fwhm) of approximately 30 nm. Adjacent fibers were separated by approximately 65 and 50 nm at junction 1 and 2, respectively. While single-molecule imaging of the 12CA5-Fab-AF647 binding achieved super-resolution, we observed discontinuous coverage of the microtubule fibers (Figure 2e). Dark areas on the microtubule fibers likely correlated with photobleached 12CA5-Fabs that remained attached to the 3xHA tags during the SMLM experiment. A similar observation was made on slow dissociating Fab in a related study.¹⁰ To further improve the sampling rate of 3xHA tags, we destabilized the 12CA5−HA interactions.

KSCN Promoted Transient Interactions of the 12CA5-Fab.

We developed a chemical strategy to disrupt weak intermolecular forces and promote transient interactions between the 12CA5-Fab and HA tag (Figure 3a). Specifically, we utilized a chaotropic agent, potassium thiocyanate (KSCN). KSCN has been routinely used to solubilize phospholipids,¹⁸ evaluate serological neutralizing antibodies in antibody avidity assays,¹⁹⁻²⁴ elute proteins in nondenaturing conditions,²⁵ and dissociate protein polymers.²⁶ We evaluated how KSCN affected the dissociation rate of the 12CA5-Fab-AF647 at the ensemble level using the decay model described above (TableS-2). Figure 3b shows that the dissociation rate increases with the KSCN concentration. For instance, 200 mM KSCN resulted in an approximately 5-fold faster dissociation. We incorporated 200 mM KSCN in the imaging buffer containing 1 nM of the 12CA5-Fab-AF647 to perform SMLM. A noticeable increase in the single-molecule sampling rate was observed from the time trace of single-molecule events in the presence of KSCN (Figure S-2). This observation is in alignment with the increased dissociation characterized at the ensemble level by the fluorescence decay measurements (Figure 3b). Increased dissociation may be difficult to observe at the single-molecule level due to the limited photon budget and highly heterogeneous binding kinetics.

Figure 3. The chaotropic agent, KSCN, enhanced the dissociation of the 12CA5-Fab for SMLM.

a. Schematic representation of chaotropic perturbation of the 12CA5-Fab interaction with 3xHA tag using KSCN. b. Fold change in the dissociation rate (k_off) obtained with different KSCN concentrations. Error bars were determined by the propagation of error. c. Two selected regions of interest (ROI) of reconstructed images with and without the use of 200 mM KSCN, respectively. d. Linear regression fit of the cumulative number of localizations per frame with (squares) and without (circles) 200 mM KSCN. e. Computed root mean square deviation (RMSD) of 3xHA tag from the 12CA5 Fv-clasp with (KSCN+) and without (KSCN-) 200 mM KSCN. f. Computed corresponding interaction energy. g. Computed number of hydrogen bonds with 200 mM KSCN. h. Computed number of hydrogen bonds without KSCN. Scale bars: 1 μm.

Figure 3c and Figure S-3 further demonstrate the enhanced sampling rate for SMLM using the 12CA5-Fab-AF647 in the presence of KSCN. To quantify the sampling rate, we fitted the cumulative number of localizations detected per 1000 frames in the absence and presence of 200 mM KSCN to a linear regression model. The slope of the curve in Figure 3d represents the average number of single-molecule events detected per frame. We observed that the SMLM acquisition with 200 mM KSCN approximately doubled the single-molecule sampling rate. Furthermore, Figure 3d and Figure S-4 demonstrate a constant number of single-molecule events detected over an extended period, suggesting a significant turnover of the HA-bound 12CA5-Fab owing to the chaotropic disruption of noncovalent interactions. Since the frequency of interactions is directly related to the local population of HA molecules, we termed our technique superresolution census of molecular epitope tags, or SR-COMET.

To obtain mechanistic insights on the chaotropic effect for SR-COMET, we performed MD simulations for the 3xHA tag bound 12CA5 Fv-clasp with (KSCN+) and without (KSCN−) 200 mM KSCN (Figure S-5, PDB ID: 5XCU). First, we evaluated the root-mean-square deviation (RMSD) between the two systems (Figure 3e). We observed an average RMSD of 0.82 ± 0.13 nm for the KSCN+ system and 0.40 ± 0.09 nm for the KSCN− system. RMSD trajectories showed convergence of the protein structures during the simulation. The approximate 2-fold higher RMSD in the presence of KSCN suggests a strong conformational-perturbation of the 3xHA tag bound Fv-clasp. Second, we calculated the interaction energies of each system by taking Coulomb interactions and Lennard-Jones potentials into account. These models estimate the contributions from electrostatic interactions and van der Waals interactions to the system, respectively. We observed an average energy of −644.71 ± 54.92 kJ mol⁻¹ for the KSCN+ system and −713.31 ± 76.69 kJ mol⁻¹ for the KSCN− system (Figure 3f). This observation suggests an average energy perturbation of 68.60 kJ mol⁻¹ in the presence of 200 mM KSCN. We further performed a hydrogen bond analysis on the two systems. Our simulations showed an average hydrogen bond count of 7.64 ± 1.62 for the KSCN+ system and 7.58 ±1.83 for KSCN− systems (Figure 3g and h, respectively). Comparison of interaction energies of 3xHA with the single HA tag (Figure S-6a) showed similar results with chaotropic perturbation (Figure 3f vs Figure S-6b). The hydrogen-bonding networks remained mostly intact between Fv-clasp and the 3xHA tag in both KSCN+ and KSCN− systems. This observation suggests a significant disruption of hydrophobic interactions instead of hydrogen bonds for the destabilization of the interactions between the12CA5-Fab and 3xHA tag.

SR-COMET Improved the Molecular Quantification Consistency over dSTORM and DNA-PAINT.

In addition to achieving superresolution, SR-COMET improved molecular quantification consistency compared to dSTORM and DNA-PAINT. For dSTORM, we immunostained microtubules using the 12CA5-IgG-AF647; for DNA-PAINT, we immunostained microtubules using unconjugated mouse monoclonal 12CA5-IgGs followed by anti-mouse secondary antibodies conjugated with DNA docking strands. The reconstructed images were obtained from comparable numbers of frames for dSTORM, DNA-PAINT, and SR-COMET. In the reconstructed dSTORM superresolution image, the intensity of the probability histogram lacks a direct one-to-one correlation with the molecular quantity (Figure 4a). Figure 4b reveals the intensity fluctuation from line profiles of four adjacent microtubule fibers. Similarly, Figure 4c demonstrates the varying intensities at junctions of two crossing fibers. Figure 4d displays these variations pooled from various positions within the cell. Although the average intensities correlated with the number of fibers, the intensities from individual fibers varied substantially. DNA-PAINT displayed improved quantification consistency (Figure 4e–h). In particular, DNA-PAINT circumvents the photobleaching artifact and is independent of the photophysical property of the dye. Despite the improved quantification consistency (Figure 4h vs d), DNA-PAINT is still subject to the labeling artifact from immunofluorescence staining. In comparison, Figure 4i shows a reconstructed SR-COMET superresolution image harnessing transient 12CA5-Fab-HA interactions. Figure 4j–l demonstrate further improved quantification consistency over DNA-PAINT. The standard deviation of the intensity distributions for single and junctions of two microtubule fibers was found to be 2.25- and 2.88-fold lower than dSTORM and 1.5- and 1.42-fold lower than DNA-PAINT, respectively.

Figure 4. SR-COMET improved the molecular quantification consistency with superresolution compared to dSTORM and DNA-PAINT.

a. High-magnification view of a representative dSTORM superresolution image reconstructed from 20,000 frames. b. Representative line profiles across individual microtubule fibers marked by a yellow arrowed line in panel a. c. Representative line profiles across junctions of two microtubule fibers labeled with 1–3 in panel a. d. Distributions of 2D probability histogram values from line profiles for single microtubule fibers and junctions of two microtubules from dSTORM. Error bars represent standard deviation (n = 25). e. High-magnification view of a representative DNA-PAINT superresolution image reconstructed from 12,000 frames. f. A representative line profiles across individual microtubule fibers marked by a yellow arrowed line in panel e. g. Representative lines profiles across junctions of two microtubule fibers labeled with 1–3 in panel e. h. Distributions of 2D probability histogram values from line profiles for single microtubule fibers and junctions of two microtubules from DNA-PAINT. Error bars represent standard deviation (n = 25). i. High-magnification view of a representative SR-COMET superresolution image obtained from 12,000 frames. j. A representative line profiles across individual microtubule fibers marked by a yellow arrowed line in panel i. k. Representative line profiles across junctions of two microtubule fibers labeled with 1–3 in panel i. l. Distributions of 2D probability histogram values from line profiles for single microtubule fibers and junctions of two microtubules from SR-COMET. Error bars represent standard deviation (n = 25). m. Superresolution images from DNA-PAINT and SR-COMET reconstructed using a pixel size of 10 nm. n. Superresolution image using DNA-PAINT (red). o. Superresolution image using SR-COMET (grey) after performing DNA-PAINT on the same cell in panel n. p. High-magnification merged view of the indicated ROIs marked by 1–4 in n and o. Scale bars: 2 μm (a, e, and i), 0.20 μm (m), 5 μm (n and o) and 0.25 μm (p).

A noted advantage of SR-COMET is its immunity to artifacts associated with immunofluorescence staining. Antibody labeling introduces linkage errors that lead to increased widths of the microtubule fibers. We adjusted the experimental conditions to achieve a similar SNR of approximately 10 for both DNA-PAINT and SR-COMET, thus achieving a similar localization precision in the reconstructed image. While microtubule fibers from DNA-PAINT and SR-COMET displayed similar morphologies using a 20 nm pixel size in the reconstructed image (Figure S-7), SR-COMET revealed a noticeably finer microtubule structure compared to DNA-PAINT when the pixel size of the reconstructed image was reduced to 10 nm (Figure 4m). The broader appearance of microtubules in DNA-PAINT is likely due to size of the 12CA5-IgG and secondary antibody used for immunofluorescence staining. We further investigated additional HA sites accessible by the Fab after immunofluorescence staining (Figure S-8). To this end, we first labeled the cell sample using 12CA5-IgGs and a commercial DNA-PAINT kit. After performing DNA-PAINT (Figure 4n), we removed the imaging buffer, postfixed the sample to cross-link the 12CA5-IgG to the HA tag, and performed SR-COMET on the same sample by incubating the cell with an imaging buffer containing the 12CA5-Fab-AF647 (Figure 4o). The merged view in Figure 4p shows that SR-COMET captured a considerable fraction of previously undetected HA sites, possibly due to a combination of factors, including the limited labeling efficiency of immunostaining due to steric hindrance of antibodies, the smaller form factor of the Fab, and single-molecule labeling using transient antibody−antigen interactions. These combined characteristics render SR-COMET efficient in molecular quantification.

SR-COMET Provides Accurate Molecular Quantification in Sparse and Dense Subcellular Regions.

SR-COMET achieved a consistent molecular census of the 3xHA-tagged microtubule independent of the local molecular densities across an entire cell (Figure 5a). The intensity of the SR-COMET reconstruction correlated with the local HA population. Figure 5b captures a selected subcellular region with the line profiles taken at four different positions. The intensities from line profiles consistently reflected the marked number of fibers determined by superresolution. The mean SR-COMET intensity of single fibers across different subcellular regions was found to be 144.6 ± 24.5 (au). The mean intensities at junctions of two and three microtubule fibers were found to be 274.2 ± 44.9 and 352.1 ± 46.7 (au), respectively (Figure S-9 and Table S-3). Figure 5c plots the distribution of intensities for junctions up to five fibers. Figure 5d shows the linear regression fitting of mean intensities vs the number of fibers. The constant term in the fit model reflects the fluorescence background due to nonspecific interactions (Figure S-9d).

Figure 5: Transient 12CA5-Fab labeling enabled consistent superresolution molecular census of HA-tagged microtubules by SR-COMET.

a. Reconstructed SR-COMET superresolution image acquired in the presence of 200 mM KSCN. b. High-magnification view of the boxed region in panel a showing the merging of four microtubule fibers and corresponding intensity profiles along lines 1–4. Arrows point to the locations of individual fibers before merging at marked green lines 2, 3 and 4. c. Mean SR-COMET intensities from the line profile as a function of the number of fibers present. Error bars represent the standard deviation (n = 180, 51, 11, 7, 3, respectively). d. Linear regression fitting of the mean intensities vs. number of fibers from c. Scale bars: 10 μm (a) and 1 μm (b).

The linear model allowed us to obtain molecular census in “superresolution-limited” areas with dense fibers. In the perinuclear region, for example, SR-COMET revealed the number of fibers marked by red arrows in Figure S-10. While SR-COMET failed to spatially resolve dense fibers, the intensity value corroborated with the number of fibers predicted by the linear regression model in Figure 5d. The numerical values in subplots i−iv of Figure S-10 mark the inferred number of fibers in these dense areas.

We demonstrate that reversible antibody−antigen interactions can be utilized to bridge structural superresolution with molecular quantification in situ. The transient 12CA5-Fab interactions with the HA tag have enabled us to achieve superresolution using SMLM similar to DNA-PAINT. Transient Fab interactions occurred on a much longer time scale than those reported in standard single-molecule studies. To this end, we employed a 5 s interval between consecutive image frames to better capture the exchange kinetics. As such, a limitation of SR-COMET in its current form is the extended acquisition time (approximately 17 h to obtain Figure 5a).

We have enhanced the dissociation of an HA tag monoclonal antibody in two steps. First, we fragmented the antibody to Fab and achieved a 2-fold increase. Second, we used a chaotropic salt to perturb the noncovalent interactions and achieved a 5-fold increase in dissociation with 200 mM KSCN. Overall, the enhanced dissociation enabled us to achieve superresolution via PAINT imaging. A related study reports single-molecule screening of fast-dissociating antibodies directly from hybridoma cultures, identifying fast dissociating Fab with dissociation rates in the range of 2.1 to 30.6 min⁻¹.¹⁰ In comparison, our enhanced dissociation of the12CA5-Fab with 200 mM KSCN is still several-fold slower. While increasing the KSCN concentration may be a plausible strategy for even faster dissociation, too high of a concentration may disrupt the protein structure of the Fab and epitope. Nevertheless, our chemical approach is flexible, cost-effective, and readily adaptable for other antibody−epitope systems.

Nonspecific transient antibody interactions with the cellular environment were observed in our experiments (Figures S-9d and S-11). We found that blocking the sample with the 12CA5 isotype control significantly reduced nonspecific interactions (Methods). The number of specific interactions also outnumbered nonspecific interactions with the constant 12CA5-Fab exchange over an extended period. The consistent interactions detected throughout the acquisition also support the notion of the constant turnover of HA-bound Fabs (Figures S-4 and S-12).

Transient molecular interactions hold the potential to achieve unbiased labeling and detection. Recently, PAINT using transient coiled-coil interactions,²⁷ modified aptamers,²⁸ ligand-activated fluorescent proteins,²⁹ and exchangeable protein fragment-based probes³⁰ and antibody fragments¹⁰ has been developed. Despite these advances, consistent molecular quantification remains challenging. Our approach may provide an alternative approach to break this quantification barrier. While the use of a molecular epitope tag might restrain this approach to exogenously expressed protein targets, SR-COMET may prove advantageous for low abundance, poorly immunogenic proteins to which highly specific antibodies may not be commercially available.

We have shown that the chaotropic agent is effective in promoting transient interactions between the 12CA5-Fab and HA by disrupting hydrophobic interactions at the antigen-binding site. MD simulations showed that the hydrophobic interactions made between the 3xHA tag and Fv-clasp (Figure S-13) were weakened in the presence of 200 nM KSCN. Hydrophobic interactions are shown to be important to maintain the stability of the bound 3xHA. Weakening of these interactions has resulted in marked deviations in the RMSD values and interaction energies (Figure 3d and e). The effect of chaotropic disruption is likely specific to the amino acid environment in the antigen-binding site and the context of the epitope and therefore is antibody-dependent. We observed that the interaction energy calculations for the single HA system are in agreement with our 3xHA calculations, showing a slight but consistent lowering of affinity in the absence and presence of KSCN (Figure S-6b vs Figure 3f). Based on these observations, we expect similar results with the single HA tag. In addition, a spacer sequence may improve the tag accessibility when fused with different proteins. Similar to other computational studies, the 3xHA model we employed has limitations. The model may represent one of the possible structural predictions. Compared to the single HA system, the flanking amino acid sequence next to the HA tag may also affect the antibody−antigen interactions. With the increasing number of antibody structures solved by cryo-electron microscopy,^31,32 MD simulations could represent a cost-effective strategy to screen for antibody−antigen pairs that are sensitive to chaotropic or antichaotropic agents. Such insights can guide the experimental design of SR-COMET against other epitope tags for multiplexed imaging.

Our work also reveals that antibodies constantly undergo association and dissociation in the low-concentration region. The observation may appear counterintuitive in the context of immunofluorescence staining. Importantly, immunofluorescence staining employs antibodies at higher concentrations. Since the on-rate, k_on, scales with the antibody concentration,³³⁻³⁵ k_on is significantly higher than k_off, driving the reaction in the forward direction to form antibody−antigen complexes. When the antibody concentration decreases, dissociation becomes more apparent. We believe that it is important to recognize the noncovalent and reversible nature of these interactions, particularly in the designing of antibody-functionalized nanocarriers. The effective concentration of the antibody in the context of nanocarriers may fall within or below the nM concentration. Since weak intermolecular forces govern antibody−antigen interactions, the ionization of water molecules, pH, temperature, and salt conditions in the aqueous environment may all play critical roles in how antibody-laden nanocarriers interact with the intended target. It is also increasingly evident that high-affinity antibodies could display fast dissociation¹⁰ and context-dependent binding behaviors.^36,37 These factors have significant implications in a wide range of applications.

CONCLUSION

In summary, we investigated the noncovalent and reversible interactions between the 12CA5-Fab and 3xHA tag in the fixed cellular environment. We used the chaotropic agent KSCN to promote reversible interactions by perturbing weak hydrophobic forces. The method enabled us to obtain consistent molecular quantification with superresolution. Our study demonstrates the utilization of the noncovalent and reversible nature of antibody−antigen interactions as an alternative approach to circumvent labeling artifacts from immunofluorescence staining for quantitative SMLM.

METHODS

Materials and Reagents.

DMEM (11960069-500 mL), penicillin-streptomycin (15140-122-100 mL), L-glutamine (25030-081-100 mL), 1× DPBS (14190-144-500 mL), potassium chloride (P217-500), and glycine (BP381-500) were purchased from Fisher Scientific. FBS (F0926-500 mL), puromycin (puromycin dihydrochloride; P8833-25 mg), Triton X-100 (X100-1 L), MES (M3671-50 g), EGTA (E3889-25 g), sodium phosphate (342483-500 g), potassium thiocyanate (207799-100 g), magnesium chloride (M8266-100 g), dimethylformamide (227056-100 mL), sodium bicarbonate (S6014-500 g), glucose oxidase from Aspergillus niger (G7141-50KU), catalase from bovine liver (C40-100 mg), 2-mercaptoethanol (M6250-100 mL), cysteine-HCl (C1276), and Amicon Ultra-4 centrifugal filter units (30 K MWCO; UFC8030) were purchased from MillioreSigma. Immobilized papain (20341), mouse IgG2b kappa isotype control (eBMG2b) eBioscience (14-4732-85), TetraSpeck microspheres (T7280), EDTA (AM9261), protein-A IgG binding buffer (21001), Nunc Lab-Tek chambered cover glass (155411), Alexa Fluor 647 NHS ester (A20006), Zeba spin desalting column (89889), and Pierce Protein A columns (20356) were purchased from Thermo Fisher Scientific. Ethanol (2701G) was purchased from Decon Laboratories Inc. Paraformaldehyde (15710) and glutaraldehyde (16120) were purchased from Electron Microscopy Sciences. ReadyTag anti-HA (12CA5) (RT0268-25 mg) was purchased from Bio X Cell. DNA-PAINT kit “MASSIVE-AB 1-PLEX” was purchased from Massive Photonics.

Buffers.

The following buffers were used for sample preparation and imaging. Cytoskeleton buffer: MES (10 mM, pH 6.1), potassium chloride (90 mM), magnesium chloride (3 mM), and EGTA (2 mM). Fixation buffer: paraformaldehyde (3.7%), Triton-X-100 (0.5%), in cytoskeleton buffer. Postfixation buffer: paraformaldehyde (3.7%), in DPBS. Blocking buffer: mouse IgG2b kappa isotype control (10 μg mL⁻¹) in BSA (5% in 1× DPBS). Buffer A: Tris (10 mM, pH 8.0), sodium chloride (50 mM). Buffer B: Tris (50 mM, pH 8.0), sodium chloride (10 mM), glucose (10%). Glox solution: glucose oxidase (14 mg), catalase (17 mg mL⁻¹) in buffer A (200 μL). STORM buffer: Glox solution (7 μL), 2-mercaptoethanol (7 μL) in buffer B (690 μL). DNA PAINT used the following buffers provided with the ‘MASSIVE-AB 1-PLEX’ DNA-PAINT kit from Massive Photonics: 1x washing buffer, antibody incubation buffer, and imaging solution.

Cell Culture.

U2OS cells (ATCC HTB-96) were cultured in DMEM supplemented with 10% FBS, 2 mM L-glutamine, and 100 units mL⁻¹ penicillin−streptomycin. U2OS cells stably expressing 3xHA-α-tubulin were produced from a single cell clone and cultured in DMEM supplemented with 10% FBS, 2 mM L-glutamine, 100 units mL⁻¹ penicillin−streptomycin, and 0.25 μg mL⁻¹ puromycin. Both cell lines were maintained at 37 °C in a humidified atmosphere of 5% CO₂ and split at the confluence.

Sample Preparation.

Approximately 5000 cells were seeded in an eight-well chambered cover glass and grown in an incubator under controlled conditions at 37 °C and 5% CO₂. Following 36 h of incubation, cells were simultaneously fixed and permeabilized in the freshly prepared fixation buffer for 20 min at room temperature. Cells were washed three times for 5 min with DPBS and blocked for 2 h at room temperature on a rocker. The cells were postfixed in a freshly prepared postfixation buffer for 10 min at room temperature. Cells were washed three times and maintained in DPBS.

For fluorescence intensity decay analysis, the fixed cells were immunostained with 4 μg mL⁻¹ 12CA5-IgG-AF647 and 12CA5-Fab-AF647 overnight at 4 °C in a humid chamber. Cells were then washed three times with DPBS. The set of the samples required for photobleaching decay analysis were postfixed in postfixation buffer for 10 min and washed three times with DPBS. All samples were maintained in DPBS until imaging.

For SR-COMET, image acquisition was performed in 1 nM 12CA5-Fab-AF647 in blocking buffer or 1 nM 12CA5-Fab-AF647 in blocking buffer supplemented with 200 mM KSCN. For negative control experiments, image acquisition was performed in 1 nM 12CA5-Fab-AF647 in the blocking buffer on wild-type U2OS cells.

For dSTORM microscopy, fixed cells were immunostained with 4 μg mL⁻¹ 12CA5-IgG-AF647 for 1 h. Cells were washed three times with DPBS, and image acquisition was performed in the STORM buffer.

For DNA-PAINT, fixed cells were immunostained with 4 μg mL⁻¹ 12CA5-IgG overnight at 4 °C in a humid chamber. Cells were washed three times with the 1× washing buffer. The secondary DNA-PAINT antibodies were diluted to 1:500 in the antibody incubation buffer and incubated for 1 h at room temperature on a rocker. The well was then washed three times with the 1× washing buffer. The imaging strand (Imager 2 ATTO 655) was diluted to 250 pM in the provided imaging buffer, and image acquisition was performed.

For SR-COMET imaging on the same sample after DNA-PAINT, the DNA-PAINT imaging buffer was removed, and the well was washed with 1× DPBS three times. The sample was then postfixed for 10 min using the postfixation buffer at room temperature. The well was washed three times with 1× DPBS, and image acquisition was performed in 1 nM 12CA5-Fab-AF647 in blocking buffer supplemented with 200 mM KSCN.

Microscopy.

Fluorescence imaging was carried out on an inverted microscope (Nikon Instruments, Eclipse Ti2). Before image acquisition, the sample chamber was incubated with 1:1000 dilution of TetraSpeck beads in 1× DPBS to be used as fiducials. After a minimum of three beads settled in the field of view, beads were removed, and the sample was washed three times on-stage with 1 × DPBS. For SR-COMET image acquisitions, TetraSpeck beads were photobleached at full laser power to one-fifth of their initial intensity to avoid pixel saturation.

Movies for fluorescence intensity decay analysis were acquired with a 647 nm laser at a power density of 0.3 W cm⁻² and an integration time of 800 ms, with a 3 min nonilluminating interval between consecutive frames. For each analysis, 50 frames were collected.

SR-COMET movies of 12 000 frames were acquired with a 647 nm laser at a power density of 33 W cm⁻² and an integration time of 800 ms, with 5 s nonilluminating interval between consecutive frames. Periodic photobleaching at a laser density of 987 W cm⁻² was applied at every 20th frame with an integration time of 100 ms. The 100×/1.49 oil-immersion objective was used with 1.5× external magnification. The Prime 95B sCMOS camera was set at 16-bit with 100 MHz readouts without pixel binning.

dSTORM movies of 20 000 frames were collected with a 647 nm excitation laser at a power density of 1.974 kW cm⁻². The Princeton Instruments ProEM-HS 512 EMCCD camera was used at 16-bit with a 10 MHz readout, 30 ms exposure, EM gain multiplier of 200, conversion gain of 3, image size of 256 × 256, and no pixel binning. Acquisition of STORM movies was started once a low density of fluorescent molecules was attained by approximately 10−30 s illumination at the power density of 1.974 kW cm⁻². dSTORM movie durations were 20 000 frames.

DNA-PAINT movies of 12 000 frames were acquired with a 647 nm laser at a power density of 642 W cm⁻² and an integration time of 100 ms. The 100×/1.49 oil-immersion objective was used with 1.5× external magnification. The Prime 95B sCMOS camera was set at 16-bit with 100 MHz readouts with 2 × 2 pixel binning.

Data Processing.

Fluorescence Intensity Decay Analysis.

Z-axis profiles for the 50-frame image acquisition were obtained. The photobleaching rate was first determined by fitting the intensity decay into an exponential decay model, eq 2, using “cftool” of MATLAB to the samples that were postfixed after immunostaining. Dissociation rates (k_off) were obtained by applying a similar exponential fit, eq 3, to the samples that were not postfixed after immunostaining. A goodness of fit of at least 0.9 was maintained for all fitted data. A minimum of three trials was performed for each condition, and the average rate was calculated. Tables S-1 and S-2 summarize the obtained rate results.

Image Reconstructions.

dSTORM image reconstruction was performed with NIS elements analysis software from Nikon Instruments (version 5.20.01). The LUT “Fire” was applied using ImageJ for visual comparison. SR-COMET image reconstruction was performed using the open-source ImageJ plug-in ThunderSTORM.³⁸ Individual tiff files were binned 2 × 2 in ImageJ. Raw images were analyzed in the ThunderSTORM plug-in. The camera settings were set to pixel size 146 nm, photoelectrons per A/D count 1.2, and base level 120. Drift correction was done using fiducial beads that were present during the entire acquisition. For the visualization of the results, the “average shifted histogram” method was used at magnifications 2.92 and 7.3 as required, rendering a pixel size of 50 and 20 nm in the reconstructed image. Reconstructed SR-COMET images use the LUT “Fire” for visualization. DNA-PAINT image reconstruction was performed using the open-source ImageJ plug-in ThunderSTORM³⁸ in the same way as SR-COMET.

Evaluation of Nonspecific Binding.

In 3xHA-α-tubulin U2OS cells, areas with no microtubule fibers (Figure S-9d) were surveyed and compared with a negative control experiment on wild-type U2OS cells with no HA expression (Figure S-11). The SR-COMET image from 12 000 frames from 3xHA-α-tubulin U2OS cells (Figure S-9d) displayed low-intensity, scattered patterns of nonspecific events across the cell. The intensity profiles of nonspecific binding events, i.e., between 60 and 80 (au), were consistent with the constant term of 73.81 in the linear regression model. In contrast, SR-COMET on a wild-type U2OS cell manifested as a dense, sporadic distribution across the cell (Figure S-11).

MD Simulations.

A 3xHA (MEYPYDVPDYAGGEYPYDVPDYAGGEYPYDVPDYA) bound Fv-clasp of the 12CA5 model structure was built using the 5XCU (PDB ID) crystal structure. Molecular dynamics simulations were performed using the GROMACS software package for the system with and without KSCN (KSCN+ and KSCN−).³⁹ The CHARMM36 force field was used for the MD simulations.³⁷ Molecular modeling and structural analysis were carried out using the Chimera software package. The topology of the SCN⁻ ion was generated using the CGenFF server.⁴⁰ The prepared complex was immersed in a cubic box with a 1 nm distance between the protein surface and the boundary of the cubic box. The cubic box was solvated using SPC/E water molecules, and periodic boundary conditions were applied from all sides. For the KSCN+ system, K⁺ and SCN⁻ ions were added to reach a concentration of 200 mM. Na⁺ and Cl⁻ ions were added to maintain the ionic strength of the system close to the physiological conditions. Energy minimization was carried out using the steepest descent algorithm with 50 000 of 0.01 kJ mol⁻¹ minimization steps. The maximum force value was set to 1000 kJ mol⁻¹. NVT and NPT equilibriums were carried out for a total of 100 ps for each with 2 fs steps. Product MD runs were carried out for a total of 30 ns consisting of 2 fs steps. Temperature and pressure coupling were performed using a modified Berendsen thermostat and Parrinello−Rahman methods, respectively. The Ewald particle mesh method was used to calculate long-range electrostatic interactions. Results were analyzed using the VMD software package.⁴¹ The interaction energy between 3xHA and Fv-clasp was calculated using the built-in GROMACS energy calculation utility. Fv-clasp and 3xHA were selected as energy groups for the energy calculation by updating the “energygrps” keyword in the mdp file accordingly. The produced energy file (edr file) was used to extract the total interaction energy using the “gmx energy” command.

12CA5-Fab-Alexa^™ Fluor 647 Production.

Conjugation of Alexa Fluor NHS-Ester to 12CA5 Mouse IgG.

ReadyTag anti-HA (12CA5) was adjusted to pH 8.3 by adding sodium bicarbonate buffer (0.5 M, pH 8.3) to a final concentration of 0.15 M and 5 mg mL⁻¹ antibody concentration. Alexa Fluor 647 NHS ester was dissolved in DMF at 10 mg mL⁻¹. The antibody solution was added to the reactive dye solution at 10:1 dye to antibody molar ratio and was incubated for 2 h at room temperature in a rotator. The dye was deactivated by adding glycine (1 M, pH 7.4) to a final concentration of 30 mM. The reaction mixture was run through a Zeba spin desalting column to purify the dye-conjugated antibody from the excess dye. The antibody to dye labeling ratio was obtained using NanoDrop measurements.

Digestion of 12CA5-IgG-Alexa Fluor 647.

The papain digestion of12CA5-AF647 to generate Fab fragments is adapted from the manufacturer user guide at an enzyme to substrate ratio of 1:80. A well-suspended aliquot of immobilized papain slurry was equilibrated in freshly prepared digestion buffer (20 mM cysteine-HCl, 20 mM sodium phosphate, 10 mM EDTA, pH 7.0) to activate the enzyme. The papain gel beads were separated from the buffer by centrifugation at 1000g for 5 min, and the washings were discarded. Equilibration was performed twice, and papain gel beads were resuspended in a fresh volume of digestion buffer. The 12CA5-AF647 antibody was diluted (to 0.625 μg μL⁻¹) in the digestion buffer, added into the resuspended immobilized papain slurry, and incubated overnight (16 h) in a shaker water bath at 37 °C at high speed. The following day, 10 mM Tris-HCl (pH 7.5) was added to the digest and the supernatant was separated by centrifugation at 1000g for 15 min. Digestion was confirmed by performing an SDS PAGE.

Purification of 12CA5-Fab-Alexa Fluor 647.

All the steps were conducted in a dark room at room temperature. A Pierce Protein A column was equilibrated with Protein A IgG binding buffer (5 mL) and allowed to drain by gravity. The column was washed with another 15 mL of the Protein A IgG binding buffer. Supernatant from the previous step was diluted 1:1 Protein A IgG binding buffer and applied to the column. Since the Fab fragments do not bind the column and are eluted with the flow-through, fractions (of 0.5 μL) were collected immediately after sample introduction. Absorbance at 280 nm was measured for each fraction with a NanoDrop to monitor the protein elution with IgG binding buffer as the negative reference. Fractions with significant absorbances at 280 nm compared to the IgG binding buffer were checked for the presence of antibodies by running an SDS-PAGE. Confirmed fractions were combined and concentrated using an Amicon Ultra-4 centrifugal filter unit.