Article pubs.acs.org/Langmuir
Diffusing Colloidal Probes of Protein−Carbohydrate Interactions Shannon L. Eichmann, Gulsum Meric, Julia C. Swavola, and Michael A. Bevan* Chemical and Biomolecular Engineering, Johns Hopkins University, Baltimore, Maryland 21218, United States ABSTRACT: We present diffusing colloidal probe measurements of weak, multivalent, specific protein−polysaccharide interactions mediated by a competing monosaccharide. Specifically, we used integrated evanescent wave and video microscopy methods to monitor the three-dimensional Brownian excursions of conconavilin A (ConA) decorated colloids interacting with dextran-functionalized surfaces in the presence of glucose. Particle trajectories were interpreted as binding lifetime histograms, binding isotherms, and potentials of mean force. Binding lifetimes and isotherms showed clear trends of decreasing ConA−dextran-specific binding with increasing glucose concentration, consistent with expectations. Net potentials were accurately captured by superposition of a short-range, glucoseindependent ConA−dextran repulsion and a longer-range, glucose-dependent dextran bridging attraction modeled as a harmonic potential. For glucose concentrations greater than 100 mM, the net ConA−dextran potential was found to have only a nonspecific repulsion, similar to that of bovine serum albumin (BSA) decorated colloids over dextran determined in control experiments. Our results demonstrate the first use of optical microscopy methods to quantify the connections between potentials of mean force and the binding behavior of ConA-decorated colloids on dextran-functionalized surfaces.
■
INTRODUCTION Interactions of proteins and carbohydrates, two essential building blocks of life, mediate a vast number of processes in biology and are therefore of great significance in medicine.1 When considering such interactions to include all saccharides (i.e., mono-, di-, oligo-, and polysaccharides) and glycoconjugates (e.g., glycoporetins, proteoglycans, glycolipids), it is obvious that protein−carbohydrate interactions are ubiquitous. For example, protein−carbohydrate interactions mediate cell− cell, cell−matrix, cell−virus, cell−protein, and cell−drug interactions important to tissue morphogenesis, cell proliferation and differentiation, signal transduction, infection, inflammation, immunity, and therapeutics. Despite the importance of protein−carbohydrate interactions, the interaction of individual proteins and carbohydrates is generally weak, which produces relatively low binding affinities. The net interaction can be made stronger through multiple parallel protein−carbohydrate interactions or multivalency. However, the net interaction between a given protein−carbohydrate pair can also be weakened by the presence of other carbohydrates that compete to interact with the protein. Understanding weak protein−carbohydrate interactions in the presence of multivalency and competing interactions is necessary to answer basic questions in biology and medicine. Although the low affinity of protein−carbohydrate interactions is not synonymous with low specificity or a lack of biological significance, it does complicate the direct measurement of such interactions. Although computational tools provide a promising route toward understanding protein− carbohydrate interactions,2 experimental measurements have been the dominant approach to date and are essential for exploration and validation. Some of the earliest assays for © 2013 American Chemical Society
measuring protein−carbohydrate interactions were based on detecting complex formation through turbidity.3 This general approach has evolved into modern measurements involving various solution-based spectroscopy methods (e.g., nuclear magnetic resonance,4 fluorescence5). More recently, surface spectroscopies have been used to detect protein−carbohydrate binding (e.g., surface plasmon resonance,6 interferometry7) and further extended to combinatorial microarray measurements.8 By nonintrusively probing large ensembles of molecules, spectroscopic methods can measure weak interactions in a statistically significant manner. However, the indirect nature of such measurements limits spatial information, requires interpretive models, and does not readily provide information on single-molecule interactions. Furthermore, small signal changes due to low protein−carbohydrate complex concentrations, relative to a background of uncomplexed molecules due to weak association, can also complicate spectroscopic measurements. Direct mechanical methods, of which atomic force microscopy (AFM) is probably the most common, provide another approach to measure protein−carbohydrate interactions. AFM can measure separation-dependent recognition events between proteins and carbohydrates immobilized on surfaces and the cantilever tip, which provides the potential capability to interrogate single molecules. For example, AFM has been used to measure interactions of tip-immobilized proteins with oligosaccharides,9 tethered monosaccharides,10 and cell-surface polysaccharides.11 Other AFM measurement Received: November 2, 2012 Revised: January 17, 2013 Published: January 18, 2013 2299
dx.doi.org/10.1021/la304355t | Langmuir 2013, 29, 2299−2310
Langmuir
Article
Figure 1. Schematic of ConA and dextran attachment to particles and surfaces. (A) Dextran-functionalized slide (green brush) obtained by GPTMS linkages to produce a layer of thickness δdex. The schematic depicts how a single dextran tail (or loop) of contour length LT might form a stretched tether when bonded to a ConA immobilized on a colloid surface. (B) ConA- (blue circles) decorated particle above a dextran-functionalized surface. Labels show ConA layer thickness, δConA, and slide-to-particle surface distance, h. (C) Silica particle with sequential adsorption of physisorbed biotinylated BSA (light blue ellipsoids), NeutrAvidin linker (Av, green squares), and mixed biotinylated ConA/biotinylated BSA outer layer.
spectroscopic methods based on the relatively small ensemble size currently accessible with our method. Theoretical models are not yet available to quantitatively interconnect the separate pieces of information quantified from three-dimensional trajectory data. This is particularly the case for the competitive and multivalent interactions studied in this work, where multivalency is present both at the macromolecular scale (i.e., ConA and dextran both have multiple binding sites) and at the particle-scale (i.e., the ConA-decorated colloid and dextranmodified substrate also have multiple binding sites). However, it was a goal of this work to generate quantitative binding isotherm, lifetime, and potential data that could indeed be connected to each other in a self-consistent manner in future modeling efforts. By measuring binding data similar to that obtained by other methods but also simultaneously measuring interaction potentials, our results provide a basis to better understand the microscopic mechanisms of binding between surface-immobilized proteins and carbohydrates. Ultimately, our results are able to quantify weak protein−carbohydrate interactions in the presence of a competing interaction with high spatial resolution and energetic sensitivity and in a direct and statistically significant manner.
configurations have included studies of tip-immobilized glycoproteins with patterned proteins12 and tip-immobilized oligosaccharides with cell-surface proteins.13 Beyond quantifying rupture forces, such measurements often include information on polysaccharide chain mechanics and blocking of specific interactions through competitive interactions. By directly probing single molecules, AFM can measure separation-dependent interactions between protein−carbohydrate binding partners. However, the mechanically intrusive nature of AFM can perturb interactions from equilibrium, and the weakest accessible force is determined by the spring constant and smallest measurable deflection (in the presence of noise). A large number of important protein−carbohydrate interactions, and weak biomolecular interactions in general, might be better understood through measurements that are more direct than spectroscopic methods but also more sensitive than the approximately piconewton limit of AFM. In this work, we employ diffusing colloidal probes to measure kT- and nanometer-scale interactions (i.e., femtonewton to piconewton forces) between surface-immobilized proteins and carbohydrates. The goal is to overcome the limitations of mechanical and spectroscopic methods, or achieve a union of their most desirable traits, to enable measurements of weak, specific protein−carbohydrate interactions that are both direct and nonintrusive. The specific approach is to use total internal reflection microscopy (TIRM) and video microscopy (VM) to measure three-dimensional trajectories of micrometer-sized colloids decorated with immobilized conconavilin A (ConA) above dextran covalently attached to glass microscope slides (see Figure 1). From these raw data, ConA−dextran interactions in different glucose concentrations (0−100 mM) were quantified in terms of equilibrium binding isotherms, binding lifetime histograms, and potentials of mean force. ConA and dextran were chosen as a well-established model system representative of weak, multivalent protein−carbohydrate interactions that we used to develop diffusing colloidal probe measurements. This model system is probably most relevant to specific biological examples such as the lectin complement activation pathway in the immune response to various pathogens14 and cell adhesion to glycosylated extracellular matrix components in stem-cell engineering and cancer metastasis.15 Our results demonstrate a unique capability to directly resolve potentials of mean force while simultaneously obtaining statistical measures of binding (i.e., isotherms, lifetime histograms), although with lower statistical certainty than for
■
THEORY Net Interaction Potentials. For protein-coated colloids interacting with a carbohydrate-coated flat surface in physiological-ionic-strength media, the net particle−wall potential energy of each particle, uN(h), is given by uN(h) = uG(h) + uV (h) + uS(h) + uH(h)
(1)
where h is surface−surface separation of the particle and wall. Gravity. The gravitational potential energy of each particle depends on its height, h, of the particle above the wall, multiplied by its buoyant weight, G, as given by uG(h) = Gh = mgh =
4 3 πa (ρp − ρf )gh 3
(2)
where m is the buoyant mass; g is the acceleration due to gravity; and ρp and ρf are the particle and fluid densities, respectively. van der Waals Attraction. The van der Waals attraction between flat plates at a separation l is given as E V (l ) = − 2300
A (l ) 12πl 2
(3) dx.doi.org/10.1021/la304355t | Langmuir 2013, 29, 2299−2310
Langmuir
Article
where A(l) is the Hamaker function, which can be computed from Lifshitz theory16 to include retardation and screening effects.17 To obtain the particle−wall potential energy, uV(h), the Derjaguin approximation can be used as18 uX (h) = 2πa
∫h
of two symmetric layers interacting at the same force, given by24,25 h(F ) =
∞
E X (l ) d l
⎛ h ⎞ FA(h) = 8πa Λ1λ1/(λ1+ λ2)Λ 2 λ2 /(λ1+ λ2) exp⎜ − ⎟ ⎝ λ1 + λ 2 ⎠
(5)
where p is a noninteger power, δV is a surface roughness correction factor, and A is an effective Hamaker constant. Steric Repulsion. To understand the repulsion between adsorbed macromolecular layers, it is useful to start by considering the free energy change of a single brush layer, given by22,23 5⎤ ⎡ ⎛ δ ⎞2 f (δ ) 5 ⎢ δ0 1⎛ δ ⎞ ⎥ = +⎜ ⎟ − ⎜ ⎟ f0 9 ⎢⎣ δ 5 ⎝ δ0 ⎠ ⎥⎦ ⎝ δ0 ⎠
uA (h) = 8πa Λ1λ1/(λ1+ λ2)Λ 2 λ2 /(λ1+ λ2)(λ1 + λ 2) ⎛ h ⎞ exp⎜ − ⎟ ⎝ λ1 + λ 2 ⎠
(6)
⎛Γ⎞ uA (h) = 8πa⎜ ⎟f0,1δ0,1/(δ0,1+ δ0,2) f0,2 δ0,2 /(δ0,1+ δ0,2) (δ0,1 + δ0,2) ⎝γ⎠ ⎛ ⎞ hγ ⎟⎟ exp⎜⎜ − ⎝ δ0,1 + δ0,2 ⎠
(7)
⎛ 3kT ⎞ FH(h) = ⎜ ⎟h ⎝ 2PL T ⎠
(15)
26
(8)
a freely jointed chain FF(h) =
kT −1 h 3 2P LT
(16)
or a wormlike chain27 FW (h) =
(9)
−2 ⎤ ⎡ kT ⎢ 1 ⎛ h ⎞ 1 h ⎥ ⎟ − − ⎜1 − P ⎢⎣ 4 ⎝ LT ⎠ 4 L T ⎥⎦
(17)
where k is the Boltzmann constant, T is the absolute temperature, P is the persistence length (which is half the Kuhn length, b; i.e., P = 0.5b), LT is the tether contour length, and 3−1 is the inverse Langevin function [i.e., the Langevin function is 3(x) = coth(x) − x−1]. For small extensions, all three models converge to that of a Hookean spring, so that the potential for a single tether is
which can be integrated to give the potential between two symmetric macromolecular layers as ⎛ hγ ⎞ ⎛Γ⎞ uS(h) = 16πaf0 δ0⎜ ⎟ exp⎜ − ⎟ ⎝γ⎠ ⎝ 2δ0 ⎠
(14)
which reduces to eq 10 for symmetric layers when δ0 = δ0,1 = δ0,2 and f 0 = f 0,1 = f 0,2. Tether Potential. The relationship between force, F, and extension, h, for macromolecular tethers can be modeled as that of a Hookean spring
where h = h1 + h2; h1 and h2 correspond to the compressed layer thicknesses of layers 1 and 2, respectively; and δ1,0 and δ2,0 are the uncompressed thicknesses of layers 1 and 2, respectively. In the case of symmetric layers, h1 = h2 = h/2, which can used to simplify eq 8 to ⎛ hγ ⎞ FS(h) = 8πaf0 Γ exp⎜ − ⎟ ⎝ 2δ0 ⎠
(13)
For the case of two asymmetric layers with the same architecture (i.e., Γ = Γ1 = Γ2 and γ = γ1 = γ2) this becomes
where Γ and γ are dimensionless constants. Different Γ and γ values can be used to generalize eq 7 to other adsorbed macromolecular architectures with different decaying density profiles at their periphery. Using eq 7, the force, F, to compress two noninterpenetrating macromolecular layers between a sphere and wall (by the Derjaguin approximation) at a separation of h is given, for h < δ1,0 + δ2,0, as FS(h) = 4πa[f (h1) − f (δ1,0) + f (h2) − f (δ2,0)]
(12)
which can be integrated to give the separation-dependent potential between asymmetric layers as
where f(h) is the free energy per unit area of a brush compressed to a height of δ < δ0, δ0 is the uncompressed brush layer thickness, and f 0 is the free energy of an uncompressed brush. For compressed brushes with δ/δ0 > 1/2, an exponential nearly identical to eq 6 is given by ⎛ f (δ ) δ⎞ = 1 + Γ exp⎜ −γ ⎟ f0 ⎝ δ0 ⎠
(11)
The interaction between asymmetric macromolecular layers can be obtained by (1) solving eq 9 for the heights of each layer, h1 and h2, at a given compressive force, F, and inserting them into eq 11; (2) defining λi = δ0,i/γi and Λi =Γi f 0,i (i = 1, 2), and (3) performing some algebra to solve for F(h). This gives the force between two asymmetric layers as
(4)
where the subscript X indicates this expression can be used for any potential between flat plates. For convenience, the theoretical potential resulting from eqs 3 and 4 can be approximated with a fitted inverse power law as19−21 uV (h) = −Aa(h + δ V )−p
1 [2h1(F ) + 2h2(F )] 2
(10)
where, again, Γ and γ can be adjusted to represent different single-layer density profiles, free energy changes under compression (eq 7), force versus distance profiles (eq 9), and energy versus distance profiles (eq 10). For two interacting asymmetric macromolecular layers, the bisection rule states that the total thickness of the two different layers under compression at a given force is the average of that
uH(h) =
∫0
h
FH(x) dx =
3kT 2 h 4PL T
(18)
and for a number of tethers, N, of different lengths acting in parallel, the potential becomes 2301
dx.doi.org/10.1021/la304355t | Langmuir 2013, 29, 2299−2310
Langmuir
Article N
uH(h) =
3kT 2 h ∑ L T, i−1 4P i=1
track particles and integrate the evanescent wave scattering intensity of each particle. To ensure a fair comparison between measured and predicted potentials, the theoretical potentials from eq 1 were convoluted with Gaussian noise characterized in control measurements of irreversibly deposited particles (that exhibited signal fluctuations not because of Brownian motion but only as the result of the net measurement noise).38,39 ConA−Dextran Binding. Previous work has shown that evanescent-wave- and video-microscopy-based tracking of colloidal particle trajectories can be used to determine binding lifetimes.40 Briefly, the number of frames during which each particle remained localized was compiled for each particle. Single colloids were considered to be associated with the surface in a given image if their height excursions in the two preceding and following images (five total) had a standard deviation of σh
