Research Article

Mechanically transduced immunosorbent assay to measure protein-protein interactions

Department of Biochemistry and Biophysics, The University of North Carolina School of Medicine, United States
UNC Lineberger Comprehensive Cancer Center, University of North Carolina, United States
Department of Chemistry, University of the Pacific, United States
Department of Biological Engineering, University of the Pacific, United States
Department of Mechanical Engineering, University of the Pacific, United States

Sep 28, 2021

https://doi.org/10.7554/eLife.67525

Version of Record

Accepted for publication after peer review and revision.

Download
Cite
Share
CommentOpen annotations (there are currently 0 annotations on this page).

Version of Record published: September 28, 2021 (This version)
Accepted: August 28, 2021
Received: February 18, 2021
Preprint posted: February 6, 2021 (Go to version)

1. Of interest
Autophagosome membrane expansion is mediated by the N-terminus and cis-membrane association of human ATG8s

Wenxin Zhang, Taki Nishimura ... Sharon A Tooze

Research Article Jun 8, 2023
Further reading

Article
Figures and data
Abstract
Introduction
Results
Discussion
Materials and methods
Data availability
References
Decision letter
Author response
Article and author information
Metrics

Abstract

Measuring protein-protein interaction (PPI) affinities is fundamental to biochemistry. Yet, conventional methods rely upon the law of mass action and cannot measure many PPIs due to a scarcity of reagents and limitations in the measurable affinity ranges. Here, we present a novel technique that leverages the fundamental concept of friction to produce a mechanical signal that correlates to binding potential. The mechanically transduced immunosorbent (METRIS) assay utilizes rolling magnetic probes to measure PPI interaction affinities. METRIS measures the translational displacement of protein-coated particles on a protein-functionalized substrate. The translational displacement scales with the effective friction induced by a PPI, thus producing a mechanical signal when a binding event occurs. The METRIS assay uses as little as 20 pmols of reagents to measure a wide range of affinities while exhibiting a high resolution and sensitivity. We use METRIS to measure several PPIs that were previously inaccessible using traditional methods, providing new insights into epigenetic recognition.

Introduction

Protein-protein interactions (PPIs) are essential to cellular biology and both high- and low-affinity interactions are required to maintain robust and dynamic responses in biological circuits (Nooren and Thornton, 2003; Kastritis and Bonvin, 2013; Evans, 2001). Low-affinity interactions are commonly leveraged, as is seen for multivalent recognition (Markin et al., 2010), readers of highly abundant proteins, and in protein allostery (Daily and Gray, 2009). In particular, recognition of the epigenome is recognized to rely on the interplay between post-translational modifications (PTMs), like methylation, phosphorylation, and ubiquitination (Yau et al., 2017; McGinty and Tan, 2016; Patel, 2016). Furthermore, multidomain-containing proteins are often regulated by allostery through weak interdomain interactions (Gladkova et al., 2018; Peng, 2015). Increasingly, the importance of weak interactions or relatively small changes in PPI affinity has been realized.

Despite the increasing sophistication of studying PPIs, biochemical characterization of these weaker and similar strength interactions remain a significant hurdle. Many techniques are useful for examining protein-binding strength, each with its own set of limitations (Rowe, 2011; Syafrizayanti et al., 2014; Pollard, 2010). However, virtually all the commonly used techniques to measure biological interactions, for example, like ELISA, FP, SPR, NMR, BLI, AUC, and ITC, rely on the law of mass action, and to measure protein binding affinities in the μM range and above, highly concentrated proteins or ligands are required (Xing et al., 2016). For many systems, obtaining such large quantities of materials can be unattainable. Furthermore, high protein concentrations leads to thermodynamic non-ideality and proteins can aggregate, self-associate, and non-specific interactions occur, thus obfuscating the binding signal (White et al., 2010; Saluja and Kalonia, 2008). NMR is the gold standard method to measure weak interactions; however, in addition to requiring copious amounts of materials, the proteins must also be isotopically labeled, a single affinity measurement requires substantial instrument time and complex data analysis, and of all the methods mentioned is the lowest throughput. BLI and SPR are methods that can measure interactions while using a small quantity of the immobilized partner, however, binding is still governed by mass action and the soluble analyte must be at concentrations above the K_d; for weak binders, this can still use large quantities of materials (Helmerhorst et al., 2012; Weeramange et al., 2020). Additionally, the signal is highly dependent on the mass change of the interaction, and for smaller ligands, binding to larger molecules this signal could be small. Moreover, discerning between background binding and specific binding can be difficult, especially for weak interactions which requires the analyte to be at a high concentration.

Another difficulty in determining the binding affinity of PPIs arises when measuring similar strength interactions, for example, two- to fivefold differences. Several factors contribute to this limitation, but determining the active fraction of protein is significant because, for most fitting techniques, the calculated affinity is a dependent variable of the protein concentration (Jarmoskaite et al., 2020; Hulme and Trevethick, 2010). Moreover, to achieve accurate fitting of a protein binding isotherm requires accurate determination of the end point of the saturation curve, which for weak interactions necessitates high concentrations of ligand. Another factor in differentiating similar strength interactions is that most binding measurements have low statistical power due to the resource intensiveness of performing multiple replicates. A method where binding strength can be measured independent of protein concentration, that uses small amounts of reagents, and that has high statistical power would be valuable.

Here, we present a novel approach to measuring the strength of biological interactions that is moderately high-throughput, requires a minimal amount of protein material, and can measure a wide range of K_d values from 10^-2 to 10^-15 M. This technique was initially inspired by the rolling of biological cells, like neutrophils exhibiting haptotaxis on endothelial cells. Neutrophil motion is driven by chemical or ligand gradients (Voisin and Nourshargh, 2013). The neutrophils roll on the endothelial cells due to PPIs between the cell surface receptors. The PPIs increase the effective friction between the two cells, allowing the rotational motion to be converted into translational displacement. We aimed to create a single particle biomimetic technique that leveraged this fundamental physical concept of friction to produce a mechanical signal to indicate binding events, the Mechanically Transduced Immunosorbent assay (METRIS). METRIS utilizes protein functionalized ferromagnetic particles to mimic the rolling cells. These ferromagnetic particles are made active via actuation of an externally applied rotating magnetic field and the particles proceed to roll, henceforth referred to as rollers, and translate across the surface using a similar mode of locomotion as the neutrophils. When the rollers are placed on a functionalized surface, the amount of rotational motion converted into translational motion depends on the effective friction between the rollers and the substrate. That effective friction scales with the strength of the binding interaction. Thus, a higher affinity PPI between the roller and the substrate will result in a larger translational displacement of the roller. Since both the roller and surface have immobilized proteins, the method is not dependent on mass action and requires approximately 20 pmols to measure PPIs regardless of their strength.

Using the METRIS assay, we reproduced well-characterized binding preferences for two different methyllysine histone reader domains (Gatchalian et al., 2013; Kuo et al., 2012) and weak interactions between the E2 Ube2D (Buetow et al., 2015) and UBL-domains (DaRosa et al., 2018). These affinities range between 10^-4 and 10^-6 M. However, we were also able to measure several weaker interactions between unmodified histone peptides, which allowed us to measure the ΔΔGs for the phospho/methyl switch phenomenon in DIDO1-PHD (Andrews et al., 2016). Finally, we also show that this method can be used to measure a weak interdomain interaction between the isolated UHRF1-UBL domain and SRA domain, which is known to control the E3 ligase specificity and epigenetic DNA methylation inheritance (Foster et al., 2018; DaRosa et al., 2018). Collectively, our results show that the METRIS assay can be a very powerful technique which has the potential to provide additional insight into PPI interactions that were difficult to measure using other methods.

Results

Rolling parameter scales with interaction affinity of PPI

In the METRIS assay, rollers are placed in a Helmholtz coil inspired apparatus (see Figure 1A and Figure 1—figure supplement 1) where an externally rotating magnetic field is applied at a constant frequency, $ω$ . The permanent magnetic moment of the roller, couples with the applied magnetic field, producing a magnetic torque and subsequent rotation of the ferromagnetic bead (Steimel et al., 2014; Sing et al., 2015). In the absence of effective friction, the rollers would rotate mostly in place with the frequency of the applied magnetic field; however, effective friction induced by binding between the rollers and the substrate will convert some of that rotational motion into translational displacement, $Δ x$ , thus indirectly measuring the effective friction between the substrate and the rollers. Since the magnetic field is many orders of magnitude stronger than the strength of noncovalent interactions, in this system the higher the effective friction between the rollers and the surface corresponds to larger translational displacement. The effective friction is determined by the strength and density of PPIs between the roller and the coated substrate. Thus, the translational displacement will scale with the density and affinity of the PPIs being measured, such that a higher $Δ x$ corresponds to a higher affinity. However, the displacement is also a function of several other parameters, specifically the diameter of the roller, $D$ , and the frequency of rotation of the applied magnetic field, $ω$ . Here we define a dimensionless parameter, which is the ratio of the observed translational displacement of the roller to the maximum theoretical translational displacement of a sphere that we refer to as the rolling parameter, RP (Figure 1B)

R P = \frac{Δ x}{π D τ ω}

Figure 1 with 3 supplements see all

Download asset Open asset

Experimental schematic of the Mechanically Transduced Immunosorbent Assay (METRIS) used to measure protein-protein interactions.

(A) General schematic of roller and surface functionalization. Both binder A and binder B are attached to the roller or the surface by biotin-streptavidin interactions. The direction of the rotating magnetic field is indicated by the curved arrow. (B) The rolling parameter (RP) is a dimensionless parameter that measures rolling. The RP is calculated by taking the ratio between the observed displacement of a roller, $Δ x$ , and the maximum theoretic rolling of a sphere, which is calculated from the circumference of each spherical particle ΠD, the frequency of the rotation of the magnetic field $ω$ , and the actuation time $τ$ . In the schematic three scenarios are depicted: (top) a RP of 1 where the roller moves the maximum theoretical displacement, (middle) a RP less than 1, and (bottom) a RP near 0 where the particle does not move. (C) Representative microscopy images of streptavidin rollers (black points) on an avidin (left) or a biotin surface (right). The scale bar in black is 100 μm and the images size are 1.28 mm × 0.96 mm. The position of the rollers prior to magnetic field actuation are indicated by the transparent spots and after actuation is opaque. The top panels are after a CW actuation and the bottom is after a CCW actuation. The magnification illustrates the difference between the null (streptavidin-avidin) and biotin-streptavidin interaction translational displacement. (D) Plot of a single roller from a streptavidin-biotin (black) and an avidin-streptavidin (red) experiment. The y-axis (X) represents the position of each roller in the field of view. The magnification above shows how Δx is calculated for each roller by subtracting the preactuation position from the postactuation position. CW and CCW actuations are repeated as described in the methods. Translational displacement is calculated as a vector. (E) The distribution of rolling parameters (RPs) from the streptavidin-avidin (N=8 rollers) and biotin-streptavidin (N=9 rollers) experiments. RP is calculated using the equation in (B) for each actuation period for each roller so the distributions contain N × 36 points. See Figure 1—source data 1 for the rolling parameter for each actuation.

Figure 1—source data 1 Rolling parameter from all rolls for either biotin-streptavidin and avidin-streptavidin.: https://cdn.elifesciences.org/articles/67525/elife-67525-fig1-data1-v1.xlsx
Download elife-67525-fig1-data1-v1.xlsx

where Δx is the translational displacement of the roller, D is the diameter of the roller, $τ$ is the actuation period of the magnetic field, and $ω$ is the rotational frequency of the magnetic field. The RP is a dimensionless parameter that varies from 0 to 1 where 0 is no translational displacement and one being a sphere perfectly rolling at a single hinge point and translating a distance equivalent to its circumference. Here, the sphere also undergoes a number of rotations given by product of the rotational frequency of the magnetic field and actuation time. The density of the interactions between the roller and the substrate are kept as constant as possible from experiment to experiment by fully saturating both the rollers and the substrate with proteins and peptides. As described in the Materials and methods, both the rollers and substrate are coated 50× the theoretical number of binding sites, so virtually all the sites should be occupied. Additionally, a series of washing steps are carried out to make sure no unbound protein or peptide remains on the surface. If the surface was not uniformly functionalized, the roller’s displacement in these regions would be detected by correlations to either the individual roller or areas on the substrate. However, no such anomalies were observed in these experiments.

To measure the translation displacement $Δ x$ and to calculate the RP of the rollers, a clockwise (CW) field was actuated at $ω = 1 H z$ for $τ = 5$ s. The field was then turned off for $τ = 5$ s. A counter-clockwise (CCW) field was actuated at $ω = 1 H z$ for $τ = 5$ s and then the field was turned off for $τ = 5$ s again. This process was repeated 18 times, and several example images of rollers and roller trajectories can be seen in Figure 1C and Figure 1D and in the supplemental videos. Appropriate parameters for the magnetic field strength and frequency were previously determined (Steimel, 2017). The rolling parameter is calculated from the observed roller translational displacement divided by the maximum theoretical translational displacement of a rolling sphere where all the rotational torque is converted into translation, so the rolling parameter varies from 0 to 1. A rolling parameter of 0 corresponds to a surface with no effective friction. Experimentally, a rolling parameter of 0 is never observed due to hydrodynamic friction between the roller and the substrate. A rolling parameter of 1 corresponds to the maximum theoretical rolling of a sphere.

We first measured the rolling of streptavidin rollers on an avidin surface or a biotin surface. Still images of a CW (top) and CCW (bottom) actuation (Figure 1C) show that on the streptavidin surface the rollers hardly move, while on the biotin surface the rollers translate well over 100 μm. A full trajectory for a roller on the biotin and streptavidin surfaces (Figure 1D) show the $Δ x$ for each roller remains relatively constant through each actuation (see Figure 1—video 1 and Figure 1—video 2 for movies of the experiment). $Δ x$ is calculated for each actuation for each roller and then converted to a rolling parameter (RP) (Figure 1D). The RPs have a gaussian distribution (Figure 1E) and the average RP on the avidin surface is 0.081 ± 0.004 while on biotin we observed a RP of 0.918 ± 0.002. The interaction between biotin and streptavidin is reported to be K_d=10^-15 M. While it is impossible to know the true K_d value for a null interaction, the weakest PPI measured are in the 10^-2 M range (Yoo et al., 2016) and enzymes with K_d values in the 10° M range have been reported (Bar-Even et al., 2011), so we assume that null interaction must be between 10° M and the concentration of water 5.5 × 10²M, and we settled on 10°M as an estimation of the null interaction, which based on our subsequent fitting seems like a suitable assumption. These two values provide an idea about the range of affinities that can be measured with METRIS.

DIDO1-PHD phospho/methyl switch characterized by METRIS

Next, we wanted to determine whether we can quantitatively correlate the measured RP to binding affinities for known PPIs and test the robustness of the METRIS assay as an experimental approach to measure PPIs. We focused our attention on weak interactions and interactions between several protein pairs that are similar in binding strength, given that these PPIs are typically difficult to accurately measure. We first examined the well-established interaction between DIDO1-PHD and H3K4 methylation. DIDO1 is responsible for interchanging between active and silent chromatin states in embryonic stem cells, and its chromatin localization is regulated through a phospho/methyl switch, where phosphorylation of H3T3 evicts DIDO1 from chromatin during mitosis (Fütterer et al., 2017; Di Lorenzo and Bedford, 2011; Liu et al., 2014). The affinities for mono-, di-, and trimethylated peptides are well described in the literature (Gatchalian et al., 2013) and interactions with the unmodified peptide and H3T3pK4me3 were too weak to be measured in the experiment setup. H3K4 peptides and DIDO1-PHD were both immobilized to the rollers and substrate through biotin-streptavidin interactions. The H3 N-terminus (a.a. 1–20) was biotinylated and immobilized on the roller, and biotinylated avi-tagged GST-DIDO1-PHD was attached to the substrate. DIDO1 has a preference for H3K4me3 > H3K4me2 > H3K4me1 (Gatchalian et al., 2013). The measured $Δ x$ and RP match this preference, with the largest rolling parameter for H3K4me3 (0.233 ± 0.012) > H3K4me2 (0.213 ± 0.010) > H3K4me1 (0.176 ± 0.005) and H3 and H3T3pK4me3 being the lowest, although still above the baseline rolling parameter value of 0.081 (Figure 2A and B). While the overall change to the RPs is small, these differences are all statistically significant because the data set has good statistical power and small percentage errors (<5%) (Figure 2—source data 2). Additionally, the distribution of rolling parameters can be found in Figure 2—figure supplement 1A, Figure 2—video 1 and 2, and Figure 2—animation 3, show the rolling for this family of interactions.

Figure 2 with 4 supplements see all

Download asset Open asset

DIDO1-PHD interactions with H3 peptides characterized using METRIS.

(A) Plot showing the average translational displacement per actuation for the rollers coated with the indicated H3K4 methylated peptide on a DIDO1-PHD surface. Streptavidin-biotin and streptavidin-avidin (PBS) are included for references. See Figure 2—source data 2 for results of statistical analysis; all comparisons are statistically significant (p<0.0001). (B) Plot showing the calculated average rolling parameter per interaction. (C) Log-Log plot of the rolling parameters (RP) from panel B with the reported K_ds. Extrapolated points for the unknown interactions are represented by unfilled markers, and the 95% confident interval for the fitting is depicted. (D) Table of rolling parameters and associated K_d estimates for the DIDO1-PHD interactions. Fold change is calculated as the ratio between the K_d values for the indicated peptide and for H3K4me3. These ratios are used to calculate $Δ Δ G$ at T=298K. The published values are from Gatchalian et al., 2013 using NMR (me1) and tryptophan fluorescence (me2/3); *ND = Not determined. (E) Image of the DIDO1-PHD crystal structure with H3K4me3 peptide, with the PHD surface electrostatic potentials shown (red = negative, blue = positive), the $Δ Δ G$ for K4me3, and the estimated $Δ Δ G$ for the rest of the peptide. The PTM reader sites are shown with greater detail to the right. Here, $Δ Δ G$ is calculated between the sequential methyl states, and the ratio of H3T3pK4me3 and H3K4me3 give the $Δ Δ G$ for T3p. (F) Results of the DIDO1-PHD histone peptide microarray assay against the indicated peptides (see Figure 1—figure supplement 1B for results of all peptides). Only H3K4me3 is statistically significant (P<0.05). (see Figure 2—source data 2 for results of statistical analysis). While these results indicate general binding trends, they cannot provide K_d estimates and do not have high enough resolution to distinguish between weaker binding interactions.

Figure 2—source data 1 Rolling parameter from all rolls for the indicated rollers on a DIDO1-PHD surface. Each row is a different roller and each column is an actuation.: https://cdn.elifesciences.org/articles/67525/elife-67525-fig2-data1-v1.xlsx
Download elife-67525-fig2-data1-v1.xlsx
Figure 2—source data 2 Statistical analysis of results from METRIS measurements and histone peptide microarray results for DIDO1-PHD. Results from statistical tests comparing the indicated pairs of rolling parameters (METRIS) or microarray results (Array).: https://cdn.elifesciences.org/articles/67525/elife-67525-fig2-data2-v1.xlsx
Download elife-67525-fig2-data2-v1.xlsx

In order to correlate binding affinity to RP, we developed an empirical fitting method based on available data. We noticed that the log-log plot of K_d vs. RP showed a linear relationship between the three known DIDO1-PHD binding interactions to the methylated peptides (R²=0.995). We also included a no-binding avidin-streptavidin interaction (RP=0.081) estimated to have a K_d = 1M and the streptavidin-biotin interaction where K_d = 10^-15M (DeChancie and Houk, 2007; Figure 2C). Overall, this experiment suggests that there is a linear dependence of the log of the RP to the log of K_d that spans roughly fifteen-orders of magnitude.

There is a clear correlation between RP and the measured K_d, the equilibrium constant for interactions, despite METRIS being a non-equilibrium technique. K_d is a ratio between the first-order dissociation rate (K_off) and the second-order association rate (K_on) (Sanders, 2004). For most PPIs, the K_on rates are very similar, and thus the K_d constant is mostly dependent on K_off. However, kinetic constants for binding interactions are rarely reported since few techniques can access this information, so for many interactions, only K_d is known. Since we do not have a theoretical model that relates RP to K_d, we sought to use an empirical fitting method based on the excellent correlation we observed between RP and K_d (Figure 2C). Using this fitting method, we could reproduce the literature K_d values with high accuracy; all of the predicted K_d values were roughly twofold tighter than the published values (Gatchalian et al., 2013) and the fold difference between the different methylation states is similar (Figure 2D). Remarkably, we were also able to estimate METRIS-K_d values for the weak interaction between the H3T3pK4me3 peptide (340 μM ± 90) and the unmodified H3 tail (1200 μM ± 440). While these are empirically derived estimates for K_d, it is clear from the RP measurements that these interactions are statistically distinct, and they represent a missing piece of data that is fundamental to a quantitative understanding of epigenetic recognition.

The utility of the METRIS data is exemplified when evaluating the $Δ Δ G$ $^{o}$ ( $Δ Δ G$ ) values, a common way to report the energetic contributions of individual amino acids for a set of related PPIs. $Δ Δ G$ is calculated by taking the natural log of the ratio of two K_d values (K_d1 and K_d2 in equation 2) in the Gibbs free energy equation, where R is the gas constant and T is the temperature in Kelvin (Sidhu and Koide, 2007).

Δ Δ G^{o} (Δ Δ G) = R T \ln \frac{K_{d 1}}{K_{d 2}}

This analysis allows for calculating the energetic contributions of the individual PTMs for binding to the DIDO1-PHD domain. For example, K4me3 is worth −4.2 $\frac{k c a l s}{m o l}$ while T3p is worth +3.4 $\frac{k c a l s}{m o l}$ (Figure 2D). To our knowledge, this is the first energetic analysis of the DIDO1 phospho/methyl switch. These values have more context when viewed with the crystal structure of DIDO1-PHD (Figure 2E; Gatchalian et al., 2013). The hydrophobic trimethyl-lysine binding site accounts for a significant amount of the total binding to the peptide, however, there are clearly other residues on H3 that interact with DIDO1-PHD, such as the N-terminus, R2, and T3, and therefore, it is not surprising that unmodified H3 can still bind and account for roughly −4 $\frac{k c a l}{m o l}$ when using 1 M K_d as the null reference. The deleterious effect of T3p is also resolved, since residue E308 of the PHD domain would clash and repel a T3p modified histone tail. Furthermore, this analysis also provides new insights into discrimination of methylation states by the DIDO1-PHD. For example, the greatest change in $Δ Δ G$ occurs between H3 from H3K4me1 (−1.9 $\frac{k c a l}{m o l}$ ), then H3K4me1 versus H3K4me2 (−1.6 $\frac{k c a l}{m o l}$ ), and H3K4me2 from H3K4me3 is the weakest (−0.7 $\frac{k c a l}{m o l}$ ). Thus, despite the DIDO1-PHD having the highest affinity for H3K4me3, it has the greatest discrimination between non-methylated H3K4 versus H3K4me1. The structure agrees with this observation, where two of the methyl binding sites are the most buried and the third is the most exposed one.

One of the significant advantages of the METRIS assay is that only 10 μl of 2 μM (20 pmol) is required to load the substrate and less is needed for the rollers, which is significantly less than any conventional method to measure PPI affinities. We compared METRIS to histone peptide microarrays, which is another methodology that can produce binding data with a minimal amount of protein (e.g. 500 μl of 0.5 μM [250 pmol] protein). While microarrays offer high-throughput screening, they lack the sensitivity to determine weak binding and small affinity differences. For DIDO1-PHD, we could observe a statistically significant difference between H3K4me3 and the other methylation states, but there were no other statistically significant differences (Figure 2F, Figure 2—figure supplement 1B, and Figure 2—source data 2). Given this result, METRIS is significantly more sensitive and quantitative than other common methods to measure protein affinities that use comparable amounts of reagents at low concentrations (i.e. ELISA and microarrays).

Determining ORC1-BAH methyl preferences using METRIS analysis

We further validated the METRIS assay using another methyllysine reader, the BAH domain of ORC1. ORC1 functions in licensing origins of replication by discriminating H4K20me2 from H4K20me1, a PTM on active chromatin, and H4K20me3 a repressive PTM (Bicknell et al., 2011a; Bicknell et al., 2011b; Kuo et al., 2012). We selected ORC1 because the reported affinities are within an order of magnitude, with a twofold difference reported between H4K20me1 and H4K20me3. The Δx and RP values we obtained matched the published binding preferences (Kuo et al., 2012) H4K20me2 ( $0.263 \pm 0.011$ ) > H4K20me1 (0.226 ± 0.008) >H4K20me3 (0.215 ± 0.005)> H4 (0.202 ± 0.005) (Figure 3A, Figure 3B, Figure 3—figure supplement 1A, Figure 3—video 1, Figure 3—video 2, and Figure 3—animation 1). Using the same fitting method, we observe a linear log-log dependence (R² = 0.967) and the METRIS calculated K_d values were between four- and eightfold tighter than the published values, yet there was good agreement between the fold-change and accordingly the $Δ Δ G s$ . (Figure 3C and Figure 3D). Thus, the METRIS assay is sensitive enough to measure changes that are 0.4 $\frac{k c a l}{m o l}$ .

Figure 3 with 4 supplements see all

Download asset Open asset

ORC1-BAH domain interactions characterized using METRIS.

(A) Plot showing the average translational displacement per actuation for the rollers immobilized with the H4K20 methylated peptide on a ORC1-BAH domain surface. Streptavidin-Biotin and Streptavidin-avidin (PBS) are included for references. See Figure 3—source data 2 for results of statistical analysis; all comparisons are significant (p<0.0001). (B) Plot showing the calculated RP for the indicated interactions (C) Log-Log plot of the rolling parameters, RP, from panel A. Extrapolated point markers are unfilled and the 95% K_d confident interval for the fitting is depicted. (D) Table of rolling parameters and associated K_d estimates for the ORC1-BAH. Fold change is calculated as the ratio between the K_d for the indicated peptide and the K_d for H4K20me2. These ratios are used to calculate $Δ Δ G$ at T=298K. The published values are from Kuo et al., 2012 using ITC; *ND = Not determined. (E) Image of the ORC1-BAH crystal structure with H4K20me2 peptide, with the BAH surface electrostatic potentials shown (red = negative, blue = positive) as well as the $Δ Δ G$ for K20me2 and the estimate for the rest of the peptide. The PTM reader site is shown with greater detail to the right. Here the $Δ Δ G$ is calculated between the sequential methyl states. (F) Results of the ORC1-BAH histone peptide microarray assay against the indicated peptides from panel A (see Figure 3—figure supplement 1B for complete peptide plot). Only H4K20me2 is statistically significantly different (p<0.05) from the other H4 peptides (Figure 3—source data 2). Again, we see that microarrays can indicate general binding trends but they cannot provide K_d estimates and do not have high enough resolution to distinguish between weaker binding interactions.

Figure 3—source data 1 Rolling parameter from all rolls for the indicated rollers on a ORC1-BAH surface. Each row is a different roller and each column is an actuation.: https://cdn.elifesciences.org/articles/67525/elife-67525-fig3-data1-v1.xlsx
Download elife-67525-fig3-data1-v1.xlsx
Figure 3—source data 2 Statistical analysis of results from METRIS measurements and histone peptide microarray results for ORC1-BAH. Results from statistical tests comparing the indicated pairs of rolling parameters (METRIS) or microarray results (Array).: https://cdn.elifesciences.org/articles/67525/elife-67525-fig3-data2-v1.xlsx
Download elife-67525-fig3-data2-v1.xlsx

Using the METRIS assay, we could also measure binding to the unmodified H4, which has previously not been detected. H4 unmodified binding was measured to be 44-fold weaker than H4K20me2 binding. With this value we could calculate that the $Δ Δ G$ for K20me2 is worth −2.2 $\frac{k c a l}{m o l}$ . When comparing this to the DIDO1-PHD, we find that DIDO1-PHD has a stronger interaction with the PTM (−4.2 versus −2.2 $\frac{k c a l}{m o l}$ ) however the ORC1-BAH domain has a stronger interaction with the unmodified histone than the DIDO1-PHD (−6 versus −4 $\frac{k c a l}{m o l}$ ). Examining the structure of ORC1-BAH domain bound to H4K20me2 (Kuo et al., 2012) shows the methyllysine binding pocket is more charged than DIDO1-PHD, and likely, in part, contributes to the higher affinity to the unmodified peptide (Figure 3D). The METRIS analysis also furthers our understanding of ORC1-BAH discrimination amongst methyl states. We find the greatest differentiation between H4K20me2 and H4K20me3 (1.6 $\frac{k c a l}{m o l}$ ) consistent with the biological role of ORC1 and this methyl sensing occurs through residue E93 (Figure 3E).

We also performed histone peptide microarrays on ORC1-BAH for comparison against the METRIS assay. The only statistically significant difference is between H4K20me2 and the other peptides (Figure 3F and Figure 3—figure supplement 1B), although the trends do match the literature and METRIS values, including the signal for the unmodified peptide when compared to the other peptides on the array, which support our findings with METRIS. However, due to the large standard deviation observed on the microarray, the assay would need to be repeated multiple times to achieve statistical significance. This highlights another advantage of METRIS assay, since it is a single particle method and the RP measurements are taken 36 times for each particle, this method has high statistical power.

Investigating noncovalent interactions between Ubiquitin-like domains and Ube2D1 utilizing METRIS

We next used METRIS to investigate interactions with the protein post-translational modification ubiquitin. Ubiquitin has an expansive cellular regulatory role that is controlled by weaker interactions with effectors (Oh et al., 2018; Cohen et al., 2017) including non-covalent interactions with E2s and E3 ligases (Brzovic et al., 2006; Zhang et al., 2016). Ubiquitin binding is wide-spread, and there are hundreds of UBLs in the human genome for these readers to discriminate amongst (Harrison et al., 2016b). For example, the E2 Ube2D1 binds to ubiquitin noncovalently with an affinity of 206 ± 6 and we have shown that a ubiquitin-like domain (UBL) on the E3 UHRF1 can bind with higher affinity (15 ± 1 μM with NMR or 29.0μM ±1 with ITC) (DaRosa et al., 2018). To probe this interaction with METRIS, both ubiquitin and the UHRF1-UBL domain were labeled using biotin-PEG-maleimide at an N-terminal cysteine installed for labeling, and Ube2D1 was labeled at native cysteines. The Δx and RP data match the affinity trend UHRF1-UBL (0.131 ± 0.005) > ubiquitin (0.108 ± 0.004) (Figure 4A, Figure 4B, Figure 4—figure supplement 1A, and Figure 4—animation 1) and fitting METRIS-K_ds produced values that were 5-fold weaker than the published values, but were in exact agreement with the 13-fold difference (1.4 $\frac{k c a l}{m o l} Δ Δ$ G) reported in the literature (Figure 4C). Therefore we have demonstrated that METRIS can measure and distinguish interactions in the 10^-4 M range without utilizing highly concentrated protein solutions, providing a simple method to measure weak interactions.

Figure 4 with 2 supplements see all

Download asset Open asset

Measuring the interaction of UBL domains with Ube2D1 or the UHRF1-SRA domain using METRIS.

(A) Plot showing the average translational displacement per actuation for the rollers coated with the indicated proteins on a surface containing UbeD21 (E2) or the UHRF1-SRA domain. Streptavidin-Biotin and Streptavidin-avidin (PBS) are included for references. All comparisons are statistically significant (See Figure 4—source data 2 for results of statistical analysis.) (B) Plot showing the calculated RP for the indicated rollers (C) Log-Log plot of the rolling parameters, RP, from panel A. Extrapolated point markers are unfilled and the 95% confident interval for the fitting is depicted. (D) Table of all rolling parameters and associated METRIS-K_d estimates. Fold change is calculated as the ratio between the indicated protein and the UHRF1-UBL domain and the $Δ Δ G$ is calculated using these ratios at T=298K. K_d values for ubiquitin are taken from Buetow et al., 2015 and UHRF1-UBL value taken from DaRosa et al., 2018. (E) Image of the UHRF1-UBL binding surface for the UHRF1-SRA and Ube2D1 shown with electrostatic surface potentials (red = negative, blue = positive) with insets highlighting the change of the UBL surface with the W2V mutation and the associated $Δ Δ G$ .

Figure 4—source data 1 Rolling parameter from all rolls for the indicated rollers on either a Ube2D1 surface or a UHRF1-SRA surface. Each row is a different roller and each column is an actuation.: https://cdn.elifesciences.org/articles/67525/elife-67525-fig4-data1-v1.xlsx
Download elife-67525-fig4-data1-v1.xlsx
Figure 4—source data 2 Statistical analysis of results from METRIS measurements for ubiquitin and UHRF1-UBL binding to Ube2D1 or UHRF1-SRA domain. Results from statistical tests comparing the indicated pairs of rolling parameters (METRIS).: https://cdn.elifesciences.org/articles/67525/elife-67525-fig4-data2-v1.xlsx
Download elife-67525-fig4-data2-v1.xlsx

Direct measurement of an interdomain interaction between UBL and SRA domains of UHRF1 using METRIS

For epigenetic readers/writers, there is an abundance of examples where interdomains interactions within a single polypeptide chain control allostery (Worden et al., 2019; Ruthenburg et al., 2007). For example, the role of UHRF1 in controlling DNA methylation requires interactions between its domains (Harrison et al., 2016a; Gao et al., 2018; Gelato et al., 2014), and specifically, our previous study provided evidence for an interaction between the UHRF1-UBL and the UHRF1-SRA domain, which is required for ubiquitylation of histone H3 (DaRosa et al., 2018). Studying interdomain interactions can be difficult, given the weak and transient nature of these interactions. We therefore thought METRIS is well-suited to measure this type of interaction. Accordingly, we tested the SRA and UBL interaction with METRIS by attaching biotinylated SRA to the substrate. For the SRA-UBL interaction, we measured an RP of 0.119 ± 0.004 for the particles, significantly higher than the 0.081 for an unmodified surface and the 0.085 we obtain with ubiquitin on the roller (Figure 4A and Figure 4B). This represents the first direct measurement of the interaction between the SRA and UBL domains of UHRF1. We also tested a mutation to the UBL (W2V) that previous biochemical assays suggested is critical for the interaction (DaRosa et al., 2018; Foster et al., 2018), and W2V had a significantly reduced RP to 0.098 ± 0.002 (Figure 4A, Figure 4B, and Figure 4—figure supplement 1A). Fitting METRIS-K_d shows the $Δ Δ G$ of the W2V variant is worth 1.5 $\frac{k c a l}{m o l}$ (Figure 4D) due to replacing the aromatic sidechain with the short aliphatic side chain (Figure 4E). This highlights another strength of METRIS; it is rare to assign $Δ Δ G$ values to mutations at binding hotspots because the mutated variant binds weakly (Pál et al., 2006). Therefore, we expect that METRIS will greatly enhance our understanding of PPIs.

Global fit of METRIS analysis

We sought to generate a global fit for all of the measurements from the three independent data sets. Overall, the log-log fit of the data remained linear ( $R^{2} = 0.89$ ) (Figure 5A), and even using this global fit, we observe agreement between fold changes and $Δ Δ G$ within a given set of PPIs (Figure 5B). However, the METRIS-K_d values were less accurate than with the individual fitting and we could not discriminate between similar strength binders in different sets of PPIs (e.g. between DIDO1 and ORC1). These results indicating that we cannot directly compare RP values obtained for different types of PPIs and that there is likely some structural difference in each system that is not yet accounted for. However, given that each set of values had similar systematic deviations from the experimentally determined values, which is why the $Δ Δ G$ remained accurate, we realized we could apply a simple scaling factor to the METRIS-K_d values to obtain measurements that matched the experimentally determined K_d. To determine the scaling factors for each interaction, we divided the published value against the METRIS-K_d, and averaged them, and then multiplied the METRIS-K_d by the scaling factor and could reproduce the literature values (Figure 5B). Indeed, this scaling can be applied to each of the individual fits to reproduce the literature values. It also provides a simple way to scale METRIS-K_d values to any experimentally determined K_d values.

Figure 5

Download asset Open asset

Global fit of binding partners for all METRIS experiments performed.

(A) Global Log-Log plot showing linearity between rolling parameter and dissociation constant for all interactions measured. (B) Table of binding constants of tested interaction partners when determined from the global fit. Fold change and $Δ Δ G$ are calculated in the same way as the previous example and the $Δ Δ G$ are similar to the previous reported value. Scaling factors are calculated by averaging the fold difference between METRIS-K_d and the published K_d for all interactions of the same type. Then the METRIS-K_d is multiplied by the scaling factor yielding the corrected K_ds, which match the published values.

Discussion

METRIS, which measures the effective mechanical friction induced by PPIs, is fundamentally different than current methodologies. Here, we have shown that METRIS can be advantageous to other approaches when measuring weak interactions, since both binding partners are immobilized. Thus, METRIS uses a very low concentration of proteins while maintaining high precision. These characteristics allow for the characterization of a vast array of PPIs, many of which were previously very laborious to measure. In this study, we demonstrate how METRIS can contribute to the study of epigenetics, by allowing us to assign $Δ Δ G$ for PTMs individually and in combination, including a phospho/methyl switch in DIDO1. These values are significant because they provide a quantitative measure for the interplay between concurrent PTMs, a central premise of the epigenetic code (Rothbart and Strahl, 2014). Furthermore, this study shows that even applying METRIS to characterized interactions can provide new insights into PPIs.

Another area where better characterization of weak interactions will contribute significantly to understanding is in studying interdomain interactions. These types of interactions can be difficult to quantify without very resource-intensive processes, and limitations with the proteins themselves (yield or solubility) may make these interactions unmeasurable. Currently, pulldown assays, chemical crosslinking, and proximity ligation are qualitative, rarely produce quantitative data, and require mass spectrometry (Richards et al., 2021). Here, we have measured a direct interaction between the UHRF1-UBL and SRA domains that we estimate to have a K_d 60μM (Figure 5B), however, the biological context for this interaction is between two tethered domains, so an absolute value is only partially relevant. More generally, we show that METRIS can be used to measure $Δ Δ G$ for hotspot mutations, which to our understanding, could previously only be measured indirectly using high-throughput selection strategies (Pál et al., 2006). Thus, METRIS will provide additional new data to the field of protein biochemistry and could aid in the parametrization of computational binding score functions.

METRIS can be conceptualized as a measure of the ability of converting rotational motion into translation motion. To translate, there must be sufficient magnetic torque applied to the rollers such that the PPI or other interactions between the roller and the substrate can be broken as they translate across the surface. Previous studies using force spectroscopy methods have measured the force of the biotin-streptavidin interaction to be 160pN (Florin et al., 1994). However, while there are some similarities with these techniques and METRIS (e.g. single particles in both cases Neuman and Nagy, 2008), there are distinct differences, particularly with the use of magnetic torque to impart this rolling motion in this nonequilibrium active system. Still, the effective friction is due to these intermolecular forces, but currently we do not have a model for how to relate the force directly to the rolling.

An essential advantage of METRIS is resolution, precision, and sensitivity, which allows for the differentiation of $Δ Δ G$ values as small as 0.4 $\frac{k c a l}{m o l}$ . Several factors likely contribute to this robustness: (1) the rolling parameter is not inherently dependent on the protein concentration, so long as the rollers and surface are saturated. (2) The measurements have high statistical power (≈10 particles each with 36 RP measurements) and very low percentage error, given the high accuracy of the measurement and low uncertainty of many of the variables in the rolling parameter calculation. (3) Multiple interactions between the bead and the surface amplify the friction, which may be necessary to measure weak interactions, and likely limits the impacts of inactive proteins on the roller and substrate. However, METRIS does have limitations, such as the reliance on literature values for extrapolating and scaling the METRIS-K_d and the need to biotinylate the binding proteins. We envision with future development, we will derive a better mathematical model that describes the relationship between protein affinity and rolling parameter as many factors will contribute to the friction, such as the number of interactions per bead or the size of the protein interaction. Despite these limitations, we expect that METRIS will be of great use to researchers studying PPIs.

Materials and methods

Key resources table

Reagent type (species) or resource	Designation	Source or reference	Identifiers	Additional information
Recombinant DNA reagent	BirA-GST-Orc1-BAH; Orc1-BAH	This Study		Strahl Lab, pGEX vector, Figure 2
Recombinant DNA reagent	BirA-GST-DIDO1-PHD; DIDO1-PHD	This Study		Strahl Lab, pGEX vector, Figure 3
Recombinant DNA reagent	N-cys Ubiquitin; ubiquitin	Kamadurai, Hari B et al. ‘Insights into ubiquitin transfer cascades from a structure of a UbcH5B approximately ubiquitin-HECT(NEDD4L) complex.’ Molecular cell vol. 36,6 (2009): 1095–102.		pGEX vector,
Recombinant DNA reagent	Ube2D1	DaRosa, Paul A et al. ‘A Bifunctional Role for the UHRF1 UBL Domain in the Control of Hemi-methylated DNA-Dependent Histone Ubiquitylation.’ Molecular cell vol. 72,4 (2018): 753–765.e6.		Pet15 vector,
Recombinant DNA reagent	UHRF1-SRA	This study		Harrison Lab, MBP-pQ80L, Figure 4
Recombinant DNA reagent	N-cys-UHRF1-UBL W2V	This study		Harrison Lab, MBP-pQ80L, Figure 4
Recombinant DNA reagent	N-cys-UHRF1-UBL	This study		Harrison Lab, MBP-pQ80L, Figure 4
Software	Able Particle Tracker	Mu Labs	Full Version	http://apt.mulabs.com/
Software	Mathematica	Steimel Labs
Software	OGG Video Converter	Ogg-converter.net	Version 6
Software	Pymol	Schrödinger	V2.4
Software	Prism	GraphPad
Other	Streptavidin Coated Ferromagnetic Beads	Spherotech	SVFM-100–4
Other	Avidin Coated Slides	ArrayIt	SMV
Other	Histone peptide array	Petell, Christopher J et al. ‘Improved methods for the detection of histone interactions with peptide microarrays.’ Scientific reports vol. 9,1 6265. 18 Apr. 2019
Antibody	Anti-GST (rabbit polyclonal)	Epicypher	13–0022	(1:1000)
Antibody	Anti-Rabbit AlexaFluor-647 (goat polyclonal)	Invitrogen	A21244	(1:10,000)

Magnetic probe and substrate functionalization

Request a detailed protocol

The streptavidin-coated ferromagnetic particles, provided by Spherotech with a nominal diameter of 10 μm, are composed of a core of polystyrene and CrO₂. 10 μL of the stock solution, 1.0% w/v, was extracted inserted into a micro-centrifuge tube. Biotinylated peptides were then inserted into the tube with the streptavidin-coated ferromagnetic particles. The amount of peptides was such to coach each bead 50× the theoretical limit, 1 mg of beads binds 0.18 nmole of biotin, to ensure all binding sites on the beads were covered. The bead and peptide solution was left to react at room temperature for at least 2 hr. The density of biotin binding sites per roller is on the order of 6 × 10⁸ biotin molecules per roller which corresponds to a density of biotin binding sites on the order of 5 × 10¹⁰ binding sites per mm². Assuming the footprint of the sphere on the substrate to be approximately 1% of the projected area that would result in approximately 10⁵ potential binding sites per bead, although the actual number of binders is likely much lower due to steric hindrance and other effects. The substrates are avidin-coated glass slides, provided by Arrayit, with a ligand density of 1.1 × 10¹⁰ ligands per mm². Microfluidic channels were created on this substrate using two pieces of double-sided tape, provided by 3M. The pieces of double-sided tape were cut to a width of several mms and a length of at least 25 mm. The pieces of tape were placed parallel to each other and at a distance of approximately 3–4 mm apart. Then a glass coverslip was placed on top of the tape to create channels approximately 22 × 5 mm. A solution of biotinylated proteins was then inserted into the channel. The amount of proteins inserted was enough to coat the channel surface 50× the theoretical limit to ensure that all of the sites on the substrate were coated. The substrate and solution was left in a container for 2 hr to allow the proteins time to bind to the substrate. After 2 hr, the solution was washed from the channel to remove any excess protein that was not attached to the substrate. Then the solution of peptide coated ferromagnetic beads was diluted approximately 2000× to reduce the probability of two ferromagnetic beads forming a magnetic dimer which cannot be analyzed in the rolling parameter analysis and dimer particles are excluded. The channel was sealed with epoxy and magnetized by an external permanent neodymium magnet. The substrate was placed in the slide holder at the center of the Helmholtz Coil Inspired Experimental Apparatus.

Helmholtz coil inspired experimental apparatus

Request a detailed protocol

The Helmholtz Coil Inspired Apparatus consists of three pairs of coils were secured in an apparatus, made of aluminum T-slots, and attached to an optical breadboard. The coils have an inner diameter of 7 cm and an outer diameter of 13 cm, as seen in Figure 1—figure supplement 1. Two sinusoidal signals, phase shifted by 90 degrees, were generated in Matlab. Those signals were sent to a National Instruments USB X Series DAQ and then passed through a 300W amplifier (150W/channel) before being sent through each pair of coils. The magnetic field is large enough (approximately 10mT) to ensure alignment of the rotational frequency of the particles with the frequency, $ω$ , of the magnetic field. The signal from the amplifier was routed to a Rigol Oscilloscope to measure the frequency and the voltage. The sample holder is made from 6061 aluminum and attached to a OMAX binocular microscope which functions as our 3D optical stage. Data acquisition was accomplished via a CMOS camera mounted on a C-mount DIN objective tube assembly. The camera was connected to a computer for visualization, video capture, and subsequent analysis. As mentioned, the magnetic field strength (B) Is approximately 10mT and the ferromagnetic particles exhibit a magnetic moment (m) on the order of 10^-11 Am². This combination of magnetic field strength and magnetic moment allows for sufficient magnetic torque, $τ$ to be applied to break the strongest non-covalent interaction, biotin and streptavidin.

τ = m \times B

For these experiments, once the sample was in the apparatus the actuation protocol was as follows: a five second actuation period, $τ$ in the RP equation, where the field rotates clockwise at a rotational frequency, $ω$ , of 1 Hz. The field is then shut off for 5 s to allow the particle to settle and bonds to equilibrate, and then the field rotated counter-clockwise for the same actuation period and at the same rotational frequency, and then the field is shut off again for 5 s. After this the process repeats 18 times, after which the video is post-processed and the particles are tracked.

Particle tracking

Request a detailed protocol

To analyze particle motion, we converted the video captured from the CMOS camera and converted it into an .avi file using ArcSoft Media Converter eight and then converted the .avi into a sequence of .jpeg images. To analyze the motion of the active particles we used Able Particle Tracker. The data was then imported into Mathematica for subsequent analysis. Particles that stick together, due to attraction between the ferromagnetic beads, are omitted from the analysis. The custom Mathmatica scripts measures the diameter of each roller and the distance that the particle travels per actuation period and calculates the rolling parameter according. The recorded images have a resolution of approximately 1 μm per pixel, and our image analysis software has subpixel resolution so we can measure differences in displacement on the order of 0.1 μm.

Protein purification and biotinylation

Request a detailed protocol

GST-[DIDO1-PHD/ORC1-BAH]-avi recombinant proteins we cloned into the pGEX-4T1 vector (GE, 27458001) to generate GST-[DIDO1-PHD/ORC1-BAH]-avi recombinant proteins. Recombinant proteins were purified as described in previous work . Briefly, the recombinant proteins were induced to express in SoluBL21 cells (Fisher, C700200) after reaching an OD600 of 0.4 with 0.2 mM IPTG and by shifting to 16°C for overnight growth. After induction, the cells were pelleted and resuspended in Lysis Buffer (50 mM HEPES, 150 mM NaCl, 1 mM DTT, 10% glycerol, pH 7.5) supplemented with protease inhibitors, then incubated in the presence of lysozyme (Sigma, L6876) and nuclease (ThermoFisher, PI88700) for 30 min. After this the cells were sonicated for six rounds consisting of 10 s continuous sonication at 50% intensity, 50% duty cycle followed by 60 s on ice. Lysates were centrifuged for 10 min at 10,000 rpm and the clarified lysates loaded onto a glutathione resin and purified by batch purification according to the manufacturer’s protocol (ThermoFisher, PI16101). Purified proteins were then dialyzed against Lysis Buffer to remove GSH and quantified using a Bradford assay per the manufacturer instructions (BioRad, 5000006) prior to being stored at −80°C. Ube2D1 is a his-tagged protein that was purified according to previous publications through standard Ni-NTA purification. The UHRF1-UBL, W2V-mutant, and UHRF1-SRA domain were cloned into a modified version of His-MBP-pQE80L vector that we have previously described. For the UHRF1-UBL domain and W2V mutant an N-terminal cystine was added using PCR for chemical conjugation with maleimide. These proteins were grown to O.D. 0.6 and induced with 0.6 mM IPTG. MBP was cleaved using TEV purified in house and removed using anion exchange. The ubiquitin with an N-terminal cystine was purified using a pGEX-4T1 expression system described previously. The ubiquitin was removed from the resin by cleavage with TEV. Purified proteins with an avi-tag were biotinylated by using BirA following the BirA500 kit’s protocol (Avidity, BirA500). Biotinylation was confirmed by performing a Coomassie gel shift assay according to Fairhead and Howarth, 2015. Cysteine Biotinylation was carried out using Poly(ethylene glycol) [N-(2-maleimidoethyl)carbamoyl]methyl ether 2-(biotinylamino)ethane (Sigma 757748) (Biotin-maleimide). Typically, small volumes were biotinylated such that very little biotin-malamide was needed (below a mg) so we added some powder and confirmed biotinaylation with SDS-page gel. For UHRF1-UBL variants and ubiquitin there is only a single engineered cysteine available for modification. For the Ube2D, UHRF1-SRA domain, and GST-PHD-DIDO1, we labeled native cysteines which resulted in heterogenous labeling. Excess biotin-maleimide was removed using size-exclusion or anion exchange for the UHRF1-UBL and ubiquitin, and dialysis for the SRA and GST-PHD-DIDO1. Proteins were typically aliquoted and frozen before METRIS analysis. Both labeling methods (N-terminal BirA tag versus biotin-maleimide) were evaluated for their ability to return RP values within error, which is shown in Figure 2—figure supplement 1C.

Histone peptide microarrays

Request a detailed protocol

Histone peptide microarrays were performed and analyzed as described in Petell et al., 2019. In brief, 500 nM of the avi- and GST-tagged DIDO1-PHD or ORC1-BAH constructs in 1% milk 1x PBST (10 mM Na₂HPO₄, 1.8 mM KH₂PO₄, 2.7 mM KCl, 137 mM NaCl, pH 7.6, 0.1% Tween-20) were incubated overnight at 4°C with shaking. The following day, the arrays were washed by submerging in 1x PBS briefly, then submerged in 0.1% formaldehyde in 1x PBS for 15 s to cross-link, formaldehyde was then quenched by submerging in 1 M glycine in 1x PBS for 1 min, after which the arrays were submerged in 1x PBS and inverted five times to remove remaining glycine. Next, the arrays were washed three times with high-salt 1 X PBS (1x PBS with 497 mM NaCl rather than 137 mM NaCl) for 5 min each at 4°C with shaking. Then, the arrays were incubated with a 1:1000 dilution of anti-GST (EpiCypher, 13–0022) in 1% milk 1x PBST for two hours at 4°C with shaking. After incubation with anti-GST antibody the arrays were washed with 1x PBS, three times for five minutes at 4°C with shaking. Next, they were exposed to a 1:10,000 dilution of anti-Rabbit AlexaFluor-647 (Invitrogen, A21244) for 30 min at 4° with shaking. Lastly, the arrays were washed three times for 5 min with 1x PBS as in the previous wash step, then submerged in 0.1x PBS prior to imaging. The arrays were imaged using a Typhoon (GE) and quantification was carried out using ImageQuant TL software. Analysis of the data was done by first averaging the triplicate intensities for a given peptide on the array; the values for an arrays’ dataset were then linearly scaled from 0 to 1 by applying a min-max formula such that the minimum value became 0 and the maximum 1. After, this all the scaled array values were combined to derive a single average and standard deviation for each peptide and the averages used for the graphs; see plots for what peptide modification states are shown. Results for the DIDO1-PHD and ORC1-BAH domains showing all peptides carrying the specified modifications, alone and in combination with other PTMs is shown in Figure 2—figure supplement 1C and Figure 3—figure supplement 1B.

Data collection and statistical analysis

Request a detailed protocol

All experiments for METRIS measured at least 10 different rollers that were rolled 36 times and all array data consist of at least three replicates, and averages with standard deviation are shown in the tables for each figure. All statistical analyses were done by using the Student’s T-Test (unpaired, two-tailed distribution) using Graphpad Prism. The results of this statistical analysis are reported in Figure 2—source data 2, Figure 3—source data 2, and Figure 4—source data 2.

Data availability

We have included all of the calculated rolling parameters for each roll as source data and movies and animations of the experimental results are included in the manuscript.

References

(2016) Regulation of methyllysine readers through phosphorylation
ACS Chemical Biology 11:547–553.
https://doi.org/10.1021/acschembio.5b00802
- PubMed
- Google Scholar
1. Bar-Even A
2. Noor E
3. Savir Y
4. Liebermeister W
5. Davidi D
6. Tawfik DS
7. Milo R
(2011) The moderately efficient enzyme: evolutionary and physicochemical trends shaping enzyme parameters
Biochemistry 50:4402–4410.
https://doi.org/10.1021/bi2002289
- PubMed
- Google Scholar
1. Bicknell LS
2. Bongers EM
3. Leitch A
4. Brown S
5. Schoots J
6. Harley ME
7. Aftimos S
8. Al-Aama JY
9. Bober M
10. Brown PA
11. van Bokhoven H
12. Dean J
13. Edrees AY
14. Feingold M
15. Fryer A
16. Hoefsloot LH
17. Kau N
18. Knoers NV
19. Mackenzie J
20. Opitz JM
21. Sarda P
22. Ross A
23. Temple IK
24. Toutain A
25. Wise CA
26. Wright M
27. Jackson AP
(2011a) Mutations in the pre-replication complex cause Meier-Gorlin syndrome
Nature Genetics 43:356–359.
https://doi.org/10.1038/ng.775
- PubMed
- Google Scholar
1. Bicknell LS
2. Walker S
3. Klingseisen A
4. Stiff T
5. Leitch A
6. Kerzendorfer C
7. Martin CA
8. Yeyati P
9. Al Sanna N
10. Bober M
11. Johnson D
12. Wise C
13. Jackson AP
14. O'Driscoll M
15. Jeggo PA
(2011b) Mutations in ORC1, encoding the largest subunit of the origin recognition complex, cause microcephalic primordial dwarfism resembling Meier-Gorlin syndrome
Nature Genetics 43:350–355.
https://doi.org/10.1038/ng.776
- PubMed
- Google Scholar
(2006) A UbcH5/ubiquitin noncovalent complex is required for processive BRCA1-directed ubiquitination
Molecular Cell 21:873–880.
https://doi.org/10.1016/j.molcel.2006.02.008
- PubMed
- Google Scholar
1. Buetow L
2. Gabrielsen M
3. Anthony NG
4. Dou H
5. Patel A
6. Aitkenhead H
7. Sibbet GJ
8. Smith BO
9. Huang DT
(2015) Activation of a primed RING E3-E2-ubiquitin complex by non-covalent ubiquitin
Molecular Cell 58:297–310.
https://doi.org/10.1016/j.molcel.2015.02.017
- PubMed
- Google Scholar
(2017) Analysis of structural features contributing to weak affinities of ubiquitin/Protein interactions
Journal of Molecular Biology 429:3353–3362.
https://doi.org/10.1016/j.jmb.2017.09.003
- PubMed
- Google Scholar
1. Daily MD
2. Gray JJ
(2009) Allosteric communication occurs via networks of tertiary and quaternary motions in proteins
PLOS Computational Biology 5:e1000293.
https://doi.org/10.1371/journal.pcbi.1000293
- PubMed
- Google Scholar
1. DaRosa PA
2. Harrison JS
3. Zelter A
4. Davis TN
5. Brzovic P
6. Kuhlman B
7. Klevit RE
(2018) A bifunctional role for the UHRF1 UBL Domain in the Control of Hemi-methylated DNA-Dependent Histone Ubiquitylation
Molecular Cell 72:753–765.
https://doi.org/10.1016/j.molcel.2018.09.029
- PubMed
- Google Scholar
1. DeChancie J
2. Houk KN
(2007) The origins of femtomolar protein-ligand binding: hydrogen-bond cooperativity and desolvation energetics in the biotin-(strept)avidin binding site
Journal of the American Chemical Society 129:5419–5429.
https://doi.org/10.1021/ja066950n
- PubMed
- Google Scholar
1. Di Lorenzo A
2. Bedford MT
(2011) Histone arginine methylation
FEBS Letters 585:2024–2031.
https://doi.org/10.1016/j.febslet.2010.11.010
- PubMed
- Google Scholar
1. Evans E
(2001) Probing the relation between force--lifetime--and chemistry in single molecular bonds
Annual Review of Biophysics and Biomolecular Structure 30:105–128.
https://doi.org/10.1146/annurev.biophys.30.1.105
- PubMed
- Google Scholar
1. Florin EL
2. Moy VT
3. Gaub HE
(1994) Adhesion forces between individual ligand-receptor pairs
Science 264:415–417.
https://doi.org/10.1126/science.8153628
- PubMed
- Google Scholar
1. Foster BM
2. Stolz P
3. Mulholland CB
4. Montoya A
5. Kramer H
6. Bultmann S
7. Bartke T
(2018) Critical role of the UBL domain in stimulating the E3 Ubiquitin Ligase Activity of UHRF1 toward Chromatin
Molecular Cell 72:739–752.
https://doi.org/10.1016/j.molcel.2018.09.028
- PubMed
- Google Scholar
(2017) DIDO as a switchboard that regulates Self-Renewal and differentiation in Embryonic Stem Cells
Stem Cell Reports 8:1062–1075.
https://doi.org/10.1016/j.stemcr.2017.02.013
- PubMed
- Google Scholar
1. Gao L
2. Tan XF
3. Zhang S
4. Wu T
5. Zhang ZM
6. Ai HW
7. Song J
(2018) An intramolecular interaction of UHRF1 reveals dual control for its histone association
Structure 26:304–311.
https://doi.org/10.1016/j.str.2017.12.016
- PubMed
- Google Scholar
(2013) Dido3 PHD modulates cell differentiation and division
Cell Reports 4:148–158.
https://doi.org/10.1016/j.celrep.2013.06.014
- PubMed
- Google Scholar
1. Gelato KA
2. Tauber M
3. Ong MS
4. Winter S
5. Hiragami-Hamada K
6. Sindlinger J
7. Lemak A
8. Bultsma Y
9. Houliston S
10. Schwarzer D
11. Divecha N
12. Arrowsmith CH
13. Fischle W
(2014) Accessibility of different histone H3-binding domains of UHRF1 is allosterically regulated by phosphatidylinositol 5-phosphate
Molecular Cell 54:905–919.
https://doi.org/10.1016/j.molcel.2014.04.004
- PubMed
- Google Scholar
(2018) Mechanism of parkin activation by PINK1
Nature 559:410–414.
https://doi.org/10.1038/s41586-018-0224-x
- PubMed
- Google Scholar
1. Harrison JS
2. Cornett EM
3. Goldfarb D
4. DaRosa PA
5. Li ZM
6. Yan F
7. Dickson BM
8. Guo AH
9. Cantu DV
10. Kaustov L
11. Brown PJ
12. Arrowsmith CH
13. Erie DA
14. Major MB
15. Klevit RE
16. Krajewski K
17. Kuhlman B
18. Strahl BD
19. Rothbart SB
(2016a) Hemi-methylated DNA regulates DNA methylation inheritance through allosteric activation of H3 ubiquitylation by UHRF1
eLife 5:e17101.
https://doi.org/10.7554/eLife.17101
- PubMed
- Google Scholar
(2016b) UbSRD: the ubiquitin structural relational database
Journal of Molecular Biology 428:679–687.
https://doi.org/10.1016/j.jmb.2015.09.011
- PubMed
- Google Scholar
(2012)
Real-time and Label-free Bio-sensing of molecular interactions by surface plasmon resonance: a laboratory medicine perspective

The Clinical Biochemist. Reviews 33:161–173.
- PubMed
- Google Scholar
1. Hulme EC
2. Trevethick MA
(2010) Ligand binding assays at equilibrium: validation and interpretation
British Journal of Pharmacology 161:1219–1237.
https://doi.org/10.1111/j.1476-5381.2009.00604.x
- PubMed
- Google Scholar
(2020) How to measure and evaluate binding affinities
eLife 9:e57264.
https://doi.org/10.7554/eLife.57264
- PubMed
- Google Scholar
1. Kastritis PL
2. Bonvin AMJJ
(2013) On the binding affinity of macromolecular interactions: daring to ask why proteins interact
Journal of the Royal Society Interface 10:20120835.
https://doi.org/10.1098/rsif.2012.0835
- Google Scholar
1. Kuo AJ
2. Song J
3. Cheung P
4. Ishibe-Murakami S
5. Yamazoe S
6. Chen JK
7. Patel DJ
8. Gozani O
(2012) The BAH domain of ORC1 links H4K20me2 to DNA replication licensing and Meier-Gorlin syndrome
Nature 484:115–119.
https://doi.org/10.1038/nature10956
- PubMed
- Google Scholar
1. Liu Y
2. Zhou J
3. White KP
(2014) RNA-seq differential expression studies: more sequence or more replication?
Bioinformatics 30:301–304.
https://doi.org/10.1093/bioinformatics/btt688
- PubMed
- Google Scholar
(2010) Mechanism for recognition of polyubiquitin chains: balancing affinity through interplay between multivalent binding and dynamics
Journal of the American Chemical Society 132:11247–11258.
https://doi.org/10.1021/ja103869x
- PubMed
- Google Scholar
1. McGinty RK
2. Tan S
(2016) Recognition of the nucleosome by chromatin factors and enzymes
Current Opinion in Structural Biology 37:54–61.
https://doi.org/10.1016/j.sbi.2015.11.014
- PubMed
- Google Scholar
1. Neuman KC
2. Nagy A
(2008) Single-molecule force spectroscopy: optical tweezers, magnetic tweezers and atomic force microscopy
Nature Methods 5:491–505.
https://doi.org/10.1038/nmeth.1218
- PubMed
- Google Scholar
1. Nooren IM
2. Thornton JM
(2003) Diversity of protein-protein interactions
The EMBO Journal 22:3486–3492.
https://doi.org/10.1093/emboj/cdg359
- PubMed
- Google Scholar
1. Oh E
2. Akopian D
3. Rape M
(2018) Principles of Ubiquitin-Dependent signaling
Annual Review of Cell and Developmental Biology 34:137–162.
https://doi.org/10.1146/annurev-cellbio-100617-062802
- PubMed
- Google Scholar
1. Pál G
2. Kouadio JK
3. Artis DR
4. Kossiakoff AA
5. Sidhu SS
(2006) Comprehensive and quantitative mapping of energy landscapes for protein-protein interactions by rapid combinatorial scanning
Journal of Biological Chemistry 281:22378–22385.
https://doi.org/10.1074/jbc.M603826200
- PubMed
- Google Scholar
1. Patel DJ
(2016) A structural perspective on readout of epigenetic histone and DNA methylation marks
Cold Spring Harbor Perspectives in Biology 8:a018754.
https://doi.org/10.1101/cshperspect.a018754
- PubMed
- Google Scholar
1. Peng JW
(2015) Investigating dynamic interdomain allostery in Pin1
Biophysical Reviews 7:239–249.
https://doi.org/10.1007/s12551-015-0171-9
- PubMed
- Google Scholar
1. Petell CJ
2. Pham AT
3. Skela J
4. Strahl BD
(2019) Improved methods for the detection of histone interactions with peptide microarrays
Scientific Reports 9:6265.
https://doi.org/10.1038/s41598-019-42711-y
- PubMed
- Google Scholar
1. Pollard TD
(2010) MBOC technical perspective: a guide to simple and informative binding assays
Molecular Biology of the Cell 21:4061–4067.
https://doi.org/10.1091/mbc.e10-08-0683
- Google Scholar
(2021) Mass spectrometry-based protein-protein interaction networks for the study of human diseases
Molecular Systems Biology 17:e8792.
https://doi.org/10.15252/msb.20188792
- PubMed
- Google Scholar
1. Rothbart SB
2. Strahl BD
(2014) Interpreting the language of histone and DNA modifications
Biochimica Et Biophysica Acta (BBA) - Gene Regulatory Mechanisms 1839:627–643.
https://doi.org/10.1016/j.bbagrm.2014.03.001
- Google Scholar
1. Rowe AJ
(2011) Ultra-weak reversible protein-protein interactions
Methods 54:157–166.
https://doi.org/10.1016/j.ymeth.2011.02.006
- PubMed
- Google Scholar
1. Ruthenburg AJ
2. Li H
3. Patel DJ
4. Allis CD
(2007) Multivalent engagement of chromatin modifications by linked binding modules
Nature Reviews Molecular Cell Biology 8:983–994.
https://doi.org/10.1038/nrm2298
- PubMed
- Google Scholar
1. Saluja A
2. Kalonia DS
(2008) Nature and consequences of protein-protein interactions in high protein concentration solutions
International Journal of Pharmaceutics 358:1–15.
https://doi.org/10.1016/j.ijpharm.2008.03.041
- PubMed
- Google Scholar
Software
1. Sanders CR
(2004) Biomolecular Ligand-Receptor Binding Studies: Theory, Practice, and Analysis
Biomolecular Ligand-Receptor Binding Studies: Theory, Practice, and Analysis.

http://csb.vanderbilt.edu/sanders/Binding_Principles_2010.pdf
1. Sidhu SS
2. Koide S
(2007) Phage display for engineering and analyzing protein interaction interfaces
Current Opinion in Structural Biology 17:481–487.
https://doi.org/10.1016/j.sbi.2007.08.007
- PubMed
- Google Scholar
Software
(2015) Systems and Methods for Detecting Molecular Interactions Using Magnetic Beads
Systems and Methods for Detecting Molecular Interactions Using Magnetic Beads.

https://patents.google.com/patent/US9977015B2/en
(2014) Artificial tribotactic microscopic walkers: walking based on friction gradients
Physical Review Letters 113:178101.
https://doi.org/10.1103/PhysRevLett.113.178101
- PubMed
- Google Scholar
Software
1. Steimel JP
(2017) Investigating Non-Equilibrium Phenomena in Active Matter Systems
Investigating Non-Equilibrium Phenomena in Active Matter Systems.

https://dspace.mit.edu/handle/1721.1/111339
(2014) Methods for analyzing and quantifying protein-protein interaction
Expert Review of Proteomics 11:107–120.
https://doi.org/10.1586/14789450.2014.875857
- PubMed
- Google Scholar
1. Voisin MB
2. Nourshargh S
(2013) Neutrophil transmigration: emergence of an adhesive cascade within venular walls
Journal of Innate Immunity 5:336–347.
https://doi.org/10.1159/000346659
- PubMed
- Google Scholar
(2020) The strengths and limitations of using biolayer interferometry to monitor equilibrium titrations of biomolecules
Protein Science 29:1004–1020.
https://doi.org/10.1002/pro.3827
- PubMed
- Google Scholar
1. White DA
2. Buell AK
3. Knowles TP
4. Welland ME
5. Dobson CM
(2010) Protein aggregation in crowded environments
Journal of the American Chemical Society 132:5170–5175.
https://doi.org/10.1021/ja909997e
- PubMed
- Google Scholar
(2019) Mechanism of Cross-talk between H2B ubiquitination and H3 methylation by Dot1L
Cell 176:1490–1501.
https://doi.org/10.1016/j.cell.2019.02.002
- PubMed
- Google Scholar
(2016) Techniques for the analysis of Protein-Protein interactions in vivo
Plant Physiology 171:pp.00470.2016.
https://doi.org/10.1104/pp.16.00470
- PubMed
- Google Scholar
1. Yau RG
2. Doerner K
3. Castellanos ER
4. Haakonsen DL
5. Werner A
6. Wang N
7. Yang XW
8. Martinez-Martin N
9. Matsumoto ML
10. Dixit VM
11. Rape M
(2017) Assembly and function of heterotypic ubiquitin chains in Cell-Cycle and protein quality control
Cell 171:918–933.
https://doi.org/10.1016/j.cell.2017.09.040
- PubMed
- Google Scholar
1. Yoo J
2. Lee T-S
3. Choi B
4. Shon MJ
5. Yoon T-Y
(2016) Observing Extremely Weak Protein–Protein Interactions with Conventional Single-Molecule Fluorescence Microscopy
Journal of the American Chemical Society 138:14238–14241.
https://doi.org/10.1021/jacs.6b09542
- Google Scholar
1. Zhang W
2. Wu K-P
3. Sartori MA
4. Kamadurai HB
5. Ordureau A
6. Jiang C
7. Mercredi PY
8. Murchie R
9. Hu J
10. Persaud A
11. Mukherjee M
12. Li N
13. Doye A
14. Walker JR
15. Sheng Y
16. Hao Z
17. Li Y
18. Brown KR
19. Lemichez E
20. Chen J
21. Tong Y
22. Harper JW
23. Moffat J
24. Rotin D
25. Schulman BA
26. Sidhu SS
(2016) System-Wide Modulation of HECT E3 Ligases with Selective Ubiquitin Variant Probes
Molecular Cell 62:121–136.
https://doi.org/10.1016/j.molcel.2016.02.005
- Google Scholar

Decision letter

Raymond E Goldstein

Reviewing Editor; University of Cambridge, United Kingdom
Cynthia Wolberger

Senior Editor; Johns Hopkins University School of Medicine, United States
Erika Eiser

Reviewer; University of Cambridge, United Kingdom

In the interests of transparency, eLife publishes the most substantive revision requests and the accompanying author responses.

Acceptance summary:

This article introduces a new experimental method that enables the direct measurements of weak interactions between proteins. It is based on densely attaching one type of protein (antigen) to a ferromagnetic microsphere, which can then bind to another protein that has been attached to a flat surface, and applying an external, rotating magnetic field to force the particle to roll across the surface, where its motion is slowed by binding and unbinding of antigens. The sensitivity of the method suggests that it may prove useful in the study of weak protein-protein interactions.

Decision letter after peer review:

Thank you for sending your article entitled "Mechanically Transduced Immunosorbent Assay To Measure Protein-Protein Interactions" for peer review at eLife. Your article is being evaluated by 2 peer reviewers, and the evaluation is being overseen by a Reviewing Editor and Cynthia Wolberger as the Senior Editor. The following individual involved in reviewer of your submission has agreed to reveal their identity: Erika Eiser (Reviewer #3).

As you can see from the reviews below, the reviewers agree that the method you have described has interesting potential, particularly in the small amount of material needed for the assay. However, there are concerns about the fact that the method itself has been published already some time ago, lessening the novelty of the publication. Focusing on the applications of the method would in principle be acceptable, but the concerns about the dynamic range of the method, and the important need for validation need to be addressed (perhaps requiring new experiments to do so) before we can reach a decision on your paper.

Reviewer #1:

The authors are seeking to develop a method for measuring the binding affinities between biomolecules using low amounts of material. The method described in this manuscript uses a magnetic field to cause a ferromagnetic particle to roll along a surface. When the particle is modified with one biomolecule and the surface is modified with another, there is an increase in friction if the two molecules interact. The degree of interaction is proportional to the amount of friction.

Strengths

– The method requires a minimal amount of material.

– The rolling parameter can be measured for many ferromagnetic particles simultaneously, leading to robust statistical analyses.

Weaknesses

– The method itself has already been described in a paper published by one of the corresponding authors in 2014.

– Insufficient detail is provided on the method and output. Raw data showing that the rolling particle is displaced in the presence of a binding partner on the surface is missing.

– The Kd cannot be directly obtained from the rolling parameter.

– Unlike methods such as surface plasmon resonance and biolayer interferometry, which use a comparably low amount of material, this method requires both binding partners to be affinity-tagged. The tags and the immobilization could obscure native interactions.

– Biomolecular interactions have a wide range of affinities that span over 15 orders of magnitude (1 M to 10-15 M). Yet, the dynamic range of the rolling parameter is only one order of magnitude (0.081 for Kd of 1 M and 0.918 for Kd of 10-15M). As a result, the rolling parameter values cluster together when correlated with Kds on a logarithmic scale. Subtle differences in binding affinity can therefore be lost.

– For any given system, the Kd must be measured using an orthogonal technique in order to establish binding affinities that can be correlated with the measured rolling parameter. More specifically, the rolling parameter values are not generalizable from system to system.

– The schemes presented in Figure 1 are difficult to interpret. Perhaps it would be better show a video of the particles rolling along a surface with no binding partner and another video when there is a binding partner. In the absence of these raw data for each system, the binding measurements are not compelling.

– The authors spend a great deal of time comparing their method to NMR, why? Other measurements such as SPR and BLI would be more appropriate comparisons.

– In general, I found the way in which the manuscript was written to be very misleading. After reading the Intro, I was expecting the rest of the paper would focus on the development of the method. Then, in lines 184-185 on page 6, it becomes clear that the method was already published 7 years ago by one of the corresponding authors. Thus, the paper is less about the method and more about the application.

– Please remove sentences like "It is difficult to overstate how transformative this will be for the study of PPIs." The method is not a direct measurement of binding affinity. The dynamic range of the measurement does not correlate with the range of biomolecular binding affinities. Multiple affinity tags must be used and orthogonal binding measurements must be made in order to interpret the rolling parameter values.

Reviewer #3:

In this article the authors introduce a new experimental method, named mechanically transduced immunosorbent (METRIS) assay, enabling the direct measurements of weak interactions between proteins. Such protein-protein interactions (PPT) are typically very weak (on the order of a fraction of the thermal energy, kBT). This new technique is based on densely attaching one type of protein (antigen) to a ferromagnetic micron bead, which can then bind to another protein, that has been attached to a flat surface. By applying an external, rotating magnetic field, the probe particles will be driven to perform a rolling motion on the flat surface. In order to see any displacement of the probe particle the antigens need to detach on one side and bind on the opposing side in the direction of the motion induced by the magnetic field. This apparent friction informs us on the strength of the PPT. The authors have tested a number of important PPT's – the non-covalent Interactions between Ubiquitin-like Domains and Ube2D1 is one example.

An important strength of the new technique the authors introduce is the sensitivity of their new method – it exceeds that of common methods by far, as they were able to measure protein-protein interactions at much lower concentrations (2 orders of magnitude) than is typically done. This allows a systematic determination of many interaction potentials between the complex configurations of proteins as function of solvent parameters and I guess also as function of temperature, which could be interesting new the denaturation region. In addition, the affinity or attractive interactions between specific proteins or RNA and proteins may change as function of concentration: As the reaction constant or KD value depends directly on the total concentration in the system the up- and down regulation processes of proteins may be better understood. The authors demonstrate that they can achieve such a sensitivity, which is well explained in the results and discussion.

The only point that needs to be addressed is the interpretation of the data. While the systems and references are very well presented and researched it would be helpful to the reader to explain in more detail the actual measurements and in particular their interpretation and relation between the displacements measured and the KD and Gibbs free energy differences.

My main suggestions are of technical nature concerning the interpretation of the results and the underlying theory.

In the present description (line 1567) the authors relate the displacement of the probe bead to a parameter they introduce as rolling parameter RP: zeta = δ x/(pi times the diameter of the bead times the frequency of the oscillating magnetic field). First I would call RP = zeta, or only RP unless I have not understood the difference. However, my main questions is how did the authors derive this expression in equation 1. It is clear that zeta must be dimensionless, hence it is presented as displacement divided a velocity that is multiplied with the actuation time. This velocity is the rolling velocity = angular velocity times the radius of the probe colloid. It is confusing what π is doing there. The authors should double check the equation and also explain to the reader why you introduce this parameter. Moreover, the radius of the colloids sured here should be given.

If I understand right, this zeta or rolling parameter is at the same time normalised against the streptavidin-biotin binding, which is known to be the strongest non-covalent binding in the system. If the probe particle is brought in contact and assuming that there is full coverage of avidin and biotin on the opposing surfaces, the effective contact area will depend on the size of the particle, and the binding strength with be more than 100 kBT. This means there is full traction and now slip or friction between the probe particle and the support surface. But with such strong binding energies, this means the fields applied must be very large and as the particle roller over the surface the biotin or avidin on one of the surfaces must be ripped off. Such behaviour was observed previously in AFM contact experiments, e.g. by the group of Herman Gaup in the 90's. First, the authors need to give details about the reproducibility of these reference measurements, in particular, the applied Forces (e.g. Magnetic field) grafting density and effective number of bonds in the contact area (which also may be influenced by the roughness of the colloidal bead surface) need to be detailed. Secondly, the authors do not discuss or present any calibration of the friction or rolling parameter. It is not necessarily a linear function between zero (no interaction and thus no friction) and one (full sticking/traction like when riding a bicycle).

In this context, it is important to look also into the displacement velocity. In these measurements the friction relies on the fact whether the protein-protein interactions are easily pealed off on one side and possibly re-established on the front of the rolling motion. But this will depend on how fast the rolling motion is. Again, AFM work by Gaup and in particular by Evan Evans (Annu. Rev. Biophys. Biomol. Struct. 2001. 30:105-28 – which should be referenced) showed that the force or interactions measured depend on how fast one pulls on the protein-protein link. Hence, the authors should give much more experimental detail and relation to known measurements. The numbers presented here for the Binding energies and their relation to the rolling parameters must be clarified as well, so that it would be possible for others to reproduce these measurements.

To summarise, the technique and the measured parameters need better elucidation and validation to be useful to others. In Particular friction is not necessarily the right terminology used in these measurements.

https://doi.org/10.7554/eLife.67525.sa1

Author response

Reviewer #1:

The authors are seeking to develop a method for measuring the binding affinities between biomolecules using low amounts of material. The method described in this manuscript uses a magnetic field to cause a ferromagnetic particle to roll along a surface. When the particle is modified with one biomolecule and the surface is modified with another, there is an increase in friction if the two molecules interact. The degree of interaction is proportional to the amount of friction.

Strengths

– The method requires a minimal amount of material.

– The rolling parameter can be measured for many ferromagnetic particles simultaneously, leading to robust statistical analyses.

We thank the reviewer for these comments and agree that these are some of the practical strengths of this new technique, in addition to the high resolution, simplicity, and ability to measure weak interactions using minimal amount of material. Furthermore, by immobilizing both binding partners, the assay is not dependent on mass action, and therefore, interactions of any strength can be measured with a small quantity of protein.

Weaknesses

– The method itself has already been described in a paper published by one of the corresponding authors in 2014.

We appreciate the reviewer for pointing out this potential concern with the manuscript. We understand that the previous wording of the manuscript, particularly the following phrasing, might have accidently been misleading:

"Previous studies have demonstrated the utilized of the METRIS assay to measure rolling parameters for a variety of interactions (e.g., Protein A-Fc, histidine-metal, and Protein-PIP lipid interactions) [42-44]."

While the statement could be misinterpreted to mean that the assay was previously developed, the PRL paper was actually focused on trying to achieve artificial chemotactic directed motion based on differences in the effective friction coefficient across a substrate. This was accomplished by creating a gradient in the density of binding sites by passivating a biotin substrate with a droplet of avidin and allowing it to evaporate and leave coffee-ring-like gradients in the density of binding sites. This paper was a physics paper, and we weren't interested or aware of the possibility of deducing binding affinity. It was only after this paper that we recognized the potential of using the concept of effective friction to detect differences in binding affinity by measuring translational displacement. There are also several critical differences between the 2014 PRL paper and this manuscript. The 2014 PRL paper utilized superparamagnetic walkers composed of two 3 μm diameter polystyrene and iron oxide beads which align their magnetic moments in the presence of a rotating magnetic field produced by two coils that were stochastically actuated. We were not focused on measuring the biotin-streptavidin interaction, and in fact, we could not measure the biotin-streptavidin interaction without heavy passivation due to the limited magnetic torque that could be applied due to the small magnetic moment of our superparamagnetic particles. The current technique utilizes a single roller 10μms in diameter, a glass substrate functionalized with avidin, and a 3D Helmholtz coil like apparatus to drive walker motion. Additionally, this current study is exclusively focused on correlating differences in the translational displacement (via introduction of a rolling parameter) to binding affinity and extrapolating an estimated K_d for PPIs. Given these significant differences, we have revised the manuscript to indicate that the foundations for this technique to be applied were previously published. The other citations in this section were to a thesis and a patent, which are not peer reviewed publications, which have some of the experimental optimization, but were not able to measure protein-protein affinities. However, this previously cited works were investigating a primarily physics phenomenon, whereas here we have further developed this technique to measure differences in binding affinity.

We would like to thank the reviewer for pointing out this potential area of confusion, and we have significantly changed the manuscript, spending more time discussing the experimental results of the streptavidin-avidin streptavidin-biotin experiment in Figure 1, since this is the first reporting of that experiment. Overall, this new format really improves the readability and clarity of the manuscript.

– Insufficient detail is provided on the method and output. Raw data showing that the rolling particle is displaced in the presence of a binding partner on the surface is missing.

We appreciate this comment and agree that being able to visualize and work with the raw data would be very helpful to understand the technique and potential applications. To that end, we have made significant changes to the manuscript to illustrate more raw data in the presence of non-binding, extremely strong binding (biotin-streptavidin), and other binding PPI partners with differing affinity.

1. We have made raw videos from some of the conditions we tested available as a supplemental video. These raw videos should be very helpful in visualizing the differences in the mechanical signal observed, translational displacement due to the presence of the binding interaction.

2. For figure 2,3,4, we have included a panel which shows the average distance that a roller translates per actuation.

3. We have also included some animations that illustrate the translational displacement of the beads, in micrometers, for the different interactions to illustrate even small differences in binding affinity can be deduced visually by the differences in the translational displacement of the rollers.

4. We have included histograms of the rolling parameters of the beads for all interactions, and these show clear differences. As seen for all the interactions (shown in Author response image 1), the distributions are Gaussian, but we can see some stark differences in the width of the interactions, particularly the PPI interactions.

Author response image 1

Download asset Open asset

Distribution of rolling parameters for different PPI interactions.

5. We have substantially revised figure 1 to include clockwise and counter-clockwise actuation for Streptavidin-Biotin and Streptavidin-Avidin (Figure 1C), More detail about the roller trajectory (Figure 1D), and included the rolling parameter distribution

We believe this additional raw data will help clarify how this technique's mechanical signal output is utilized to differentiate between binding affinities, and we hope that these videos illustrate the novel nature and ease of use of this technique. If there are any other raw data the reviewers think should be included in the manuscript, we are happy to include it.

– The Kd cannot be directly obtained from the rolling parameter.

While it is true that the K_d cannot be directly obtained in isolation at this time, this fact does not take away from the importance of this technique and possible implications. With further work and more data, it will be possible in the future to derive a mathematical model to calculate K_d directly or develop a fundamental theoretical model that will enable one to directly measure binding affinity from the rolling parameter value. Nonetheless, this is the first report of a technique that can be utilized to deduce differences in binding affinity from a mechanical signal, the translational displacement. Also, even without a K_d based on a standard curve, related molecules' binding strengths can still be measured with reasonable accuracy using the ΔΔG method as described in the manuscript.

– Unlike methods such as surface plasmon resonance and biolayer interferometry, which use a comparably low amount of material, this method requires both binding partners to be affinity-tagged. The tags and the immobilization could obscure native interactions.

We agree that METRIS requires both proteins be immobilized with high-affinity tags, which could obstruct some biological interactions. However, this issue is not unique to METRIS. Both SPR and BLI have one protein immobilized, usually through an affinity tag (not necessarily a high-affinity tag) or surface adsorption, which can also block interactions. In this study, we present two methods of tagging the protein at different locations, through a BirA tag and with maleamide-biotin, which in this study showed comparable RPs, (Supplemental Figure 3) so this gives the user flexibility in how to achieve immobilization, and there are many other possible immobilization strategies that we are testing. Importantly, this also indicates that we can see similar results from both approaches. We also have added some additional text in the Discussion section of the manuscript about this. For the interactions we measured in our study, we have no indication that tagging blocked any of the interactions, given the strong correlation of our data to other studies that measured these protein-protein interactions.

Additionally, immobilizing both proteins affords advantages to SPR and BLI because we use significantly less material than either BLI and SPR. However, in BLI and SPR, the soluble analyte is still dependent on mass action and thus can require high concentrations of protein to measure weaker affinities, still consuming a lot of materials and also leading to potential thermodynamic non-ideality of the soluble analyte. We have a more detailed comparison between SPR/BLI and METRIS below and directly compare BLI and METRIS for low-affinity binders.

– Biomolecular interactions have a wide range of affinities that span over 15 orders of magnitude (1 M to 10-15 M). Yet, the dynamic range of the rolling parameter is only one order of magnitude (0.081 for Kd of 1 M and 0.918 for Kd of 10-15M). As a result, the rolling parameter values cluster together when correlated with Kds on a logarithmic scale. Subtle differences in binding affinity can therefore be lost.

We believe that this is one of the key properties that makes the METRIS technique valuable as a bioassay platform, and can clarify this point. While the rolling parameter only spans one order of magnitude, our manuscript and other preliminary data illustrate that METRIS can measure affinities spanning 15 orders of magnitude, but is also sensitive enough to measure differences as low as 2-fold like we observed when measuring the Orc1-BAH interaction with methylated peptides (See Figure 3). Thus, we do not believe we are losing subtle differences in binding. This accuracy is due to the very small uncertainty or error that we report and this may be because of some confusion with the definition of the rolling parameter, which we have clarified in the manuscript in response to one of the comments by Reviewer #3 (see below). The rolling parameter is the ratio of the observed, experimental translational displacement of the roller, Δx, to the maximum theoretical displacement of a rolling sphere. The maximum displacement of a rolling sphere that hinges perfectly on a single point will simply be the circumference of the sphere (πD, where D is the diameter of the sphere typically 10 μm for our rollers, but we explicitly measure the diameter for every roller) times the number of rotations which is governed by the rotational frequency of the magnetic field, ω, which is 1Hz for this manuscript, times the actuation period in seconds, τ, which is 5s for this experiment. Thus, the maximum theoretical displacement is simply πDωτ. A rolling parameter of 1 would equal 157 µm, where a rolling parameter of 0.082, the null binding measurement in this manuscript, would correspond to a distance of 13 µm. We can measure the translational displacement and the diameter of the particle with high accuracy using a custom-written Mathematica analysis script in conjunction with Able Particle tracker, and there is virtually no uncertainty for the actuation period and the rotational frequency of the magnetic field in this technique. The recorded images have a resolution of approximately 1μm per pixel, and our image analysis software has subpixel resolution so we can measure subtle differences in displacement on the order of 100s of nanometers. Couple this with the fact that we measure on average 20-30 rollers. Each roller rolls approximately 36 times, 5 seconds each roll, meaning we roll the entire sphere 5 complete rotations, we have incredible statistical averaging. This METRIS technique and probes are extremely sensitive and can distinguish forces on the order of 10s of femtonewtons, as previously reported (PRL 2014), and we have added this to the main text of the manuscript.

This is also somewhat akin to thermodynamics, while biological interactions span 15 orders of magnitude, the ΔG/mol between the bound and unbound states only span ~2 orders of magnitude (~20 kcal/mol). Moreover, we would like to highlight that for these interactions, which are very weak and can be impossible to measure with other traditional techniques like BLI, METRIS is indeed sensitive enough to distinguish and measure a wide range of binding affinities with high resolution.

– For any given system, the Kd must be measured using an orthogonal technique in order to establish binding affinities that can be correlated with the measured rolling parameter. More specifically, the rolling parameter values are not generalizable from system to system.

We agree with the reviewer's comment. As we acknowledge in the text, this is one of the limitations of METRIS at this point, as we do not have a robust theoretical model that allows for the direct calculation of binding affinity from the rolling parameter. We are continuing to work on this aspect of the project. Still, at this point, even an empirical and orthogonal calculation of binding affinities using METRIS is valuable at this point for the following reasons:

(1) This technique does not utilize a large amount of material, so even at this stage, METRIS can be very useful as a quick diagnostic technique for binding partners that are difficult to produce in large quantities.

(2) METRIS is relatively simple to perform, and one can quickly obtain at the very least a qualitative comparison between several different binders within a family of binders/interactions.

(3) METRIS is well suited to measure very weak interactions, and as we demonstrated in this paper – even when studying characterized systems, we can still produce biological insights into interactions. Furthermore, even without an empirically determined K_d using an orthogonal system, you can still derive relatively accurate ΔΔG using the standard curve from our aggregate data for a series of family members.

(4) If a user establishes a baseline, the results can be highly quantitative compared to other methods for the reasons described.

(5) METRIS can still be very useful even as a qualitative bioassay screener, especially as we improve throughput.

Additionally, similar calibration curves are required to get quantitative results from ELISA measurements; however, our sensitivity and reproducibility is well beyond the ability of any quantitative ELISA with significantly less material utilized.

– The schemes presented in Figure 1 are difficult to interpret. Perhaps it would be better show a video of the particles rolling along a surface with no binding partner and another video when there is a binding partner. In the absence of these raw data for each system, the binding measurements are not compelling.

We really appreciate this comment from the reviewer, and we have made the recommended changes in Figure 1 and included more supplemental videos and supplemental figures.

– The authors spend a great deal of time comparing their method to NMR, why? Other measurements such as SPR and BLI would be more appropriate comparisons.

In the manuscript, we focused on NMR because it is the standard for measuring very weak interactions. However, the reviewer is correct that SPR and BLI can use small quantities of materials and may be more similar to METRIS, given that one of the interacting partners is immobilized. We have added some text to the manuscript to address this point.

Additionally, METRIS has significant advantages to SPR and BLI, especially when examining very weak binders. Both SPR and BLI are dependent on mass action, so the soluble analyte still must be above the Kd, and for weak binding, this still requires a significant amount of protein. For example, in another unpublished study where we used METRIS in combination with BLI to measure weak binding to ubiquitin for three mutants, we needed 5.5 mLs of ubiquitin at 5mM (~200 mg) to measure affinities in the 10^-4-10^-3 range (see Author response image 2). For very weak binders, it is not possible to measure the K_d. When fitting BLI data, it is critical to establish the endpoint, i.e., saturate the binding curve, and one often needs very high concentrations of proteins. With BLI, we could not accurately measure the K_d values for the weakest binders (compare results from METRIS and BLI for mutants 2 and 3 in Author response image 2C) because we did not saturate the binding curves, and at higher ubiquitin concentrations, thermodynamic non-ideality occurs (Author response image 2C and D). With METRIS, we could accurately measure these affinities with 30 pmols of protein total, and we could even measure the affinity for mutant 3, which we couldn't fit with BLI (Author response image 2 A-C). For these BLI measurements, we needed 23 µmols of material – ~100,000 times more material than we needed for METRIS. For most proteins, it is impossible to obtain that much protein at that high of a concentration. Moreover, the BLI signal is noisy and is dependent on accurately measuring the ubiquitin concentration (Figure 2D).

Additionally, for BLI and SPR, it is difficult to distinguish between background binding to the chip when studying weak interactions, where for METRIS, we have good sensitivity between the null and weak binding range.

Author response image 2

Download asset Open asset

Comparison Between BLI and METRIS.

(A) Rolling parameter measurements for 3 mutants that bind ubiquitin. (B) Fit of METRIS to Kd values. (C) Kdvalues obtained from METRIS and BLI. (D) BLI data for each of the mutants.

For SPR, sometimes you can derive binding values below the K_d, but in the scenarios, only a fraction of the chip is bound, and depending on your signal, you may not be able to detect binding. Also, the signal for BLI and SPR is dependent on a significant size change (i.e., change the refractive index), so for some binding experiments, like a peptide/small molecule to a protein, the signal can be very small. Given these issues, we think that METRIS has significant advantages to BLI and SPR. We have incorporated this discussion into the manuscript and highlighted the utility of METRIS as a new technique.

– In general, I found the way in which the manuscript was written to be very misleading. After reading the Intro, I was expecting the rest of the paper would focus on the development of the method. Then, in lines 184-185 on page 6, it becomes clear that the method was already published 7 years ago by one of the corresponding authors. Thus, the paper is less about the method and more about the application.

We appreciate the reviewer's comment, and this is due to some confusion about how the manuscript was written. Above, we described the key differences between the 2014 PRL paper, and we have provided more background about how this method is different from previously described techniques. We have also made changes to the manuscript to avoid this potential source of confusion.

– Please remove sentences like "It is difficult to overstate how transformative this will be for the study of PPIs." The method is not a direct measurement of binding affinity. The dynamic range of the measurement does not correlate with the range of biomolecular binding affinities. Multiple affinity tags must be used and orthogonal binding measurements must be made in order to interpret the rolling parameter values.

We have removed the claims. We were attempting to highlight the importance of the METRIS technique but appreciate this comment and have made subsequent changes to the manuscript to ensure the claims made are reasonable, accurate, and not overstated

Reviewer #3:

In this article the authors introduce a new experimental method, named mechanically transduced immunosorbent (METRIS) assay, enabling the direct measurements of weak interactions between proteins. Such protein-protein interactions (PPT) are typically very weak (on the order of a fraction of the thermal energy, kBT). This new technique is based on densely attaching one type of protein (antigen) to a ferromagnetic micron bead, which can then bind to another protein, that has been attached to a flat surface. By applying an external, rotating magnetic field, the probe particles will be driven to perform a rolling motion on the flat surface. In order to see any displacement of the probe particle the antigens need to detach on one side and bind on the opposing side in the direction of the motion induced by the magnetic field. This apparent friction informs us on the strength of the PPT. The authors have tested a number of important PPT's – the non-covalent Interactions between Ubiquitin-like Domains and Ube2D1 is one example.

An important strength of the new technique the authors introduce is the sensitivity of their new method – it exceeds that of common methods by far, as they were able to measure protein-protein interactions at much lower concentrations (2 orders of magnitude) than is typically done. This allows a systematic determination of many interaction potentials between the complex configurations of proteins as function of solvent parameters and I guess also as function of temperature, which could be interesting new the denaturation region. In addition, the affinity or attractive interactions between specific proteins or RNA and proteins may change as function of concentration: As the reaction constant or KD value depends directly on the total concentration in the system the up- and down regulation processes of proteins may be better understood. The authors demonstrate that they can achieve such a sensitivity, which is well explained in the results and discussion.

We truly appreciate the comments made by the reviewer and appreciate the insight into the METRIS technique. We agree that METRIS will allow scientists to study interactions in different ways than other methods and we are excited to test some of the experiments that the reviewer mentions above.

The only point that needs to be addressed is the interpretation of the data. While the systems and references are very well presented and researched it would be helpful to the reader to explain in more detail the actual measurements and in particular their interpretation and relation between the displacements measured and the KD and Gibbs free energy differences.

My main suggestions are of technical nature concerning the interpretation of the results and the underlying theory.

In the present description (line 1567) the authors relate the displacement of the probe bead to a parameter they introduce as rolling parameter RP: zeta = δ x/(pi times the diameter of the bead times the frequency of the oscillating magnetic field). First I would call RP = zeta, or only RP unless I have not understood the difference. However, my main questions is how did the authors derive this expression in equation 1. It is clear that zeta must be dimensionless, hence it is presented as displacement divided a velocity that is multiplied with the actuation time. This velocity is the rolling velocity = angular velocity times the radius of the probe colloid. It is confusing what π is doing there. The authors should double check the equation and also explain to the reader why you introduce this parameter. Moreover, the radius of the colloids sured here should be given.

This is a great comment. We have made changes in the text that correct the description of the rolling parameter, which initially and mistakenly stated normalization by the theoretical velocity, which was not correct. We have also changed the labeling of the zeta to the RP. The RP is the ratio of the actual translational displacement of the roller, Δx, divided by the maximum theoretical displacement of a rolling sphere on a substrate. One can imagine that the maximum distance a sphere can roll on a substrate would be to hinge on a single point and then roll perfectly on this point, whereby the sphere would translate a distance equivalent to the sphere's circumference. In this scenario, our sphere or roller undergoes 5 complete rolls per actuation period as we kept the rotational frequency of the magnetic field constant at 1Hz and the time of rolling was 5 seconds, so the roller undergoes 5 complete rotations per roll. So, the maximum theoretical translational displacement of a sphere would be the circumference of our sphere π D, multiplied by the actuation time τ, and the rotational frequency of the magnetic field ω. We introduce this parameter because, as the reviewer points out, the translational displacement of the roller depends on several factors beyond that of binding partners, and those include D, τ, and ω. We introduced the rolling parameter given this dimensionless number allows us to provide the reader with an apples-to-apples comparison of how the translational displacement varies across each PPI. Additionally, and pragmatically, although our particle diameters are nominally approximately 10 μm, they tend to vary slightly, but these differences affect the rolling parameter (which is why we explicitly measure each roller diameter) and thus the extrapolated binding affinity. The introduction and creation of the rolling parameter is essential to provide an accurate comparison of the effective friction induced by binding and, therefore, the translational displacement of the rollers. Without such a parameter, the extrapolation to a binding affinity would be impossible or, at the very least, much less robust.

If I understand right, this zeta or rolling parameter is at the same time normalised against the streptavidin-biotin binding, which is known to be the strongest non-covalent binding in the system. If the probe particle is brought in contact and assuming that there is full coverage of avidin and biotin on the opposing surfaces, the effective contact area will depend on the size of the particle, and the binding strength with be more than 100 kBT. This means there is full traction and now slip or friction between the probe particle and the support surface. But with such strong binding energies, this means the fields applied must be very large and as the particle roller over the surface the biotin or avidin on one of the surfaces must be ripped off. Such behaviour was observed previously in AFM contact experiments, e.g. by the group of Herman Gaup in the 90's. First, the authors need to give details about the reproducibility of these reference measurements, in particular, the applied Forces (e.g. Magnetic field) grafting density and effective number of bonds in the contact area (which also may be influenced by the roughness of the colloidal bead surface) need to be detailed. Secondly, the authors do not discuss or present any calibration of the friction or rolling parameter. It is not necessarily a linear function between zero (no interaction and thus no friction) and one (full sticking/traction like when riding a bicycle).

We appreciate the reviewer’s keen insight. Indeed, we calibrate our METRIS platform by looking at the RP of a non-interacting binding pair and that of the strongest non-covalent interaction biotin and streptavidin. As we mention in the manuscript, we fully coat our probe and substrate with more than 50X the theoretical binding capacity to ensure a fully coated substrate and particle. We do this to ensure that the density of binders remains as constant as possible, but as the diameter of the particles is slightly different, the number of binders will change slightly, hence why we also introduce the rolling parameter to account for these differences. The density of biotin-binding sites per roller is on the order of 6 x 10⁸ biotin molecules per roller, which corresponds to a density of biotin-binding sites on the order of 5 x 10¹⁰ binding sites per mm², which is of similar density to the density of streptavidin binding sites is reported in the manuscript. The actual footprint of the sphere on the substrate is not a trivial calculation, but assuming that footprint is 1% of the projected area, that would result in approximately 10⁵ potential binding sites, but due to the size of tagged PPI, this number of binding interactions is likely less. Regardless to break these interactions will require an applied magnetic torque, τ, which is defined as the m x B where m is the magnetic moment, and B is the applied magnetic field strength. As described in the manuscript, the applied magnetic field strength is approximately 10mT, and the magnetic moment is on the order of 10^-11 A M^-2 this would produce a torque large enough to break the biotin-streptavidin bond. In our previous study in 2014, we studied a system composed of walking streptavidin-coated superparamagnetic dimers, 3 μm in diameter each, the biotin-streptavidin bond was not able to be broken without heavily passivating the biotin substrate with avidin particles and coating the walkers with biotin-PEG. This combination reduced the number of biotin-binding sites, and due to the PEG brush, the biotin-streptavidin interaction strength was reduced because compressing the polymer PEG brush is entropically unfavorable. The superparamagnetic particles were unable to break the biotin-streptavidin interaction even when using approximately the same magnetic field strength because the magnetic moment of the walkers is more than an order of magnitude less than that of the ferromagnetic rollers. Thus, the torque is more than an order of magnitude less.

However, despite this increase in the magnetic torque, we are not stripping the biotin from the substrate in this system. This is simply not possible in the system that we are using because the biotin or the streptavidin is covalently linked to the glass substrate, and we are not breaking covalent bonds, just breaking the biotin-streptavidin interaction. If the biotin or avidin was being ripped off from the substrate and remained stuck on the roller, there is no way for these ligands to re-attach to the substrate. Thus, if this is indeed the mechanism when the rollers are actuated in the reverse direction, we should see a drastic decrease in the rolling parameter as one of the binding partners has been removed. However, this is not observed, as seen in Author response image 3 and in our supplemental movies. This is distinct from scenario where binding partners are being ripped from the surface. Take for example the data from a pre-print of a system consisting of a supported lipid bilayer that has biotinylated lipid bilayers. The supported lipid bilayer is not covalently attached to the substrate and in here the binding affinity of the biotin-streptavidin interaction is enough to rip the lipids from the substrate and you can see in Author response image 3B as the lipids are ripped from the bilayer the rolling parameter decreases as the roller continues to roll, thus illustrating that in this system in the manuscript we are not ripping off binding partners from the substrate.

Author response image 3

Download asset Open asset

Rolling parameter of biotin-streptavidin interaction on avidin substrate (A) and a biotinylated supported lipid bilayer (B).

Finally, in terms of calibrating the RP to friction, we clarify that the RP gives us an indication of the effective friction induced by binding interactions, and we correlate RP to the binding affinity. We appreciate the reviewer for this comment and have updated the manuscript to make this distinction more clear. We do not assume that the RP and friction scale linearly and have changed the manuscript. We also want to thank the reviewer for the comments and hope that these experimental details the distributions of the rolling parameter values given in Figure 1 illustrate the robust and reproducible nature of the technique.

In this context, it is important to look also into the displacement velocity. In these measurements the friction relies on the fact whether the protein-protein interactions are easily pealed off on one side and possibly re-established on the front of the rolling motion. But this will depend on how fast the rolling motion is. Again, AFM work by Gaup and in particular by Evan Evans (Annu. Rev. Biophys. Biomol. Struct. 2001. 30:105-28 – which should be referenced) showed that the force or interactions measured depend on how fast one pulls on the protein-protein link. Hence, the authors should give much more experimental detail and relation to known measurements. The numbers presented here for the Binding energies and their relation to the rolling parameters must be clarified as well, so that it would be possible for others to reproduce these measurements.

We thank the reviewer for pointing us to the excellent manuscript by Evan Evans and also pointing us to relevant papers from the field of force spectroscopy. We have added these citations to the manuscript and have also added a section in the discussion about force spectroscopy and compared it to METRIS. We would also like to point the reviewer to the previous comments regarding the interactions being peeled off the substrate. In terms of the protein-protein interactions begin broken, as the roller rolls, this does indeed depend on the rotational frequency of the magnetic field. In this study, the rotational frequency is kept fixed at 1Hz as this has been determined as an appropriately slow rotational frequency to allow for binding events. If one increases the magnetic field's rotational frequency, there will be less time for binding, and the effective friction will decrease, as you can see in Author response image 4. Here we find that at 5Hz, most interactions are no longer measurable as the bead is rotating too fast, and binding cannot occur because there is not sufficient time for binding. In fact, only the biotin-streptavidin interaction is observed at 10Hz, but as you can see, the rolling parameter has decreased drastically. We hope this new information makes it easier for comparisons.

Author response image 4

Download asset Open asset

Rolling parameter as a function of rotational frequency for interactions.

To summarise, the technique and the measured parameters need better elucidation and validation to be useful to others. In Particular friction is not necessarily the right terminology used in these measurements.

We have made the appropriate changes to the manuscript and included the information suggested above. We have also changed the term friction to effective friction induced by binding to be more accurate. However, the reviewer is correct that the particle is essentially slipping along and what we are measuring is how effectively the rotational torque is converted to rotational motion. Thus, we have simplified this concept and describe it as effective friction. Again, we thank the reviewer for these comments, which have improved the quality of the manuscript.

https://doi.org/10.7554/eLife.67525.sa2

Article and author information

Author details

Christopher J Petell
1. Department of Biochemistry and Biophysics, The University of North Carolina School of Medicine, Chapel Hill, United States
2. UNC Lineberger Comprehensive Cancer Center, University of North Carolina, Chapel Hill, United States
Contribution
Conceptualization, Resources, Data curation, Formal analysis, Investigation, Visualization, Writing - original draft, Writing - review and editing

Competing interests
No competing interests declared
Kathyrn Randene

Department of Chemistry, University of the Pacific, Stockton, United States

Contribution
Resources, Data curation, Formal analysis, Investigation

Competing interests
No competing interests declared
Michael Pappas

Department of Biological Engineering, University of the Pacific, Stockton, United States

Contribution
Resources, Data curation, Formal analysis, Investigation

Competing interests
No competing interests declared
Diego Sandoval

Department of Biological Engineering, University of the Pacific, Stockton, United States

Contribution
Data curation, Formal analysis, Investigation

Competing interests
No competing interests declared
Brian D Strahl
1. Department of Biochemistry and Biophysics, The University of North Carolina School of Medicine, Chapel Hill, United States
2. UNC Lineberger Comprehensive Cancer Center, University of North Carolina, Chapel Hill, United States
Contribution
Conceptualization, Resources, Formal analysis, Supervision, Funding acquisition, Methodology, Writing - original draft, Writing - review and editing

Competing interests
No competing interests declared

"This ORCID iD identifies the author of this article:" 0000-0002-4947-6259
Joseph S Harrison

Department of Chemistry, University of the Pacific, Stockton, United States

Contribution
Conceptualization, Resources, Data curation, Formal analysis, Supervision, Investigation, Visualization, Methodology, Writing - original draft, Project administration, Writing - review and editing

For correspondence
joseph.scott.harrison@gmail.com

Competing interests
No competing interests declared

"This ORCID iD identifies the author of this article:" 0000-0002-2118-6524
Joshua P Steimel

Department of Mechanical Engineering, University of the Pacific, Stockton, United States

Contribution
Conceptualization, Resources, Data curation, Software, Formal analysis, Supervision, Funding acquisition, Validation, Investigation, Visualization, Methodology, Writing - original draft, Project administration, Writing - review and editing

For correspondence
jsteimel@pacific.edu

Competing interests
is a cofounder of Tribosense Technologies

"This ORCID iD identifies the author of this article:" 0000-0003-2437-8545

Funding

University of the Pacific

Joseph S Harrison
Joshua P Steimel

National Institute of General Medical Sciences (GM126900)

Brian D Strahl

American Cancer Society (PF-19-027- 459 01-DMC)

Christopher J Petell

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

This work was supported by NIH grant GM126900 to BDS, an ACS Postdoctoral Fellowship (PF-19-027-01–DMC) to CJP University of the Pacific startup funds to JPS and JSH and an undergraduate research grant to KR.

Senior Editor

Cynthia Wolberger, Johns Hopkins University School of Medicine, United States

Reviewing Editor

Raymond E Goldstein, University of Cambridge, United Kingdom

Reviewer

Erika Eiser, University of Cambridge, United Kingdom

Publication history

Preprint posted: February 6, 2021 (view preprint)
Received: February 18, 2021
Accepted: August 28, 2021
Version of Record published: September 28, 2021 (version 1)

Copyright

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.