1,2-Dioleoyl-sn-glycero-phospho-ethanolamine-3-N-(cap biotinyl) (biotin-DOPE) and 1,2-dioleoyl-sn-glycero-3-phosphocholine (DOPC) were purchased from Avanti Polar Lipids (Alabaster, AL, United States). De-glycosylated neutravidin was purchased from Invitrogen (Tokyo, Japan). Unless stated otherwise, all chemicals, purchased from FUJIFILM Wako Pure Chemical (Osaka, Japan), Nacalai Tesque (Kyoto, Japan), Sigma-Aldrich (St. Louis, MO, United States), or Japan Alcohol Trading (Tokyo Japan), were used without further purification.

The chemical structures of polymers synthesized are shown in Figure 1. pAA-CysX-biotin (X = 0, 5, and 20) were synthesized through copolymerization of acrylic acid (AA for pAA) and S-trityl-cysteine acrylamide (S-Tri-Cys-AAm). Here X denotes mol% content of S-Tri-Cys-AAm over all monomers. For the synthesis of pPEGMA-Cys5-biotin, acrylic acid was replaced with poly (ethylene glycol) methyl ether acrylate (PEGMA). 4,4’-((E)-diazene-1,2-diyl)bis (4-cyano-N-(2-(5-((3aR,4R,6aS)-2-oxohexahydro-1H-thieno [3,4-d]imidazol-4-yl)pentanamido)ethyl)pentanamide) (ACVA-biotin) and 2-(dodecylthiocarbonothioylthio)-2-methylpropionic acid (DDMAT) were used as the initiator and chain transfer agent, respectively. After the copolymerization, the trityl group was deprotected with trifluoroacetic acid (TFA). See Supplementary Material S1 for details of the synthesis and basic characterization of the polymers.

For QCM-D experiments, AT cut quartz crystals (5 MHz) coated with SiO₂ (QSX 303, QSense, Gothenburg, Sweden) were soaked in 2 wt% sodium dodecyl sulfate (SDS) solution for 1 h, followed by rinsing with ultrapure water, drying under a N₂ stream, and treatment in a UV-ozone cleaner for 20 min (Makky and Tanaka, 2015). For RICM experiments, the glass substrates (Menzel, Braunschweig, Germany) were cleaned using a modified RCA method (Kern and Puotinen, 1970), intensively rinsed with water, dried at 70 °C, and stored in a vacuum chamber. The cleaned substrates were bonded to a bottomless plastic round dish from Matsunami Glass (Osaka, Japan). A CHCl₃ solution of DOPC doped with DOPE-biotin ( χ biotin = 0.02 ) was exposed to a nitrogen stream and then kept in a vacuum chamber at 25°C overnight. After suspending the dry lipid films in 100 mM NaCl aqueous solution buffered with 10 mM Tris-HCl (pH = 7.4), the suspension was sonicated with a tip sonicator (Misonix, New York, United States) for 30 min at 1.0 W, yielding small unilamellar vesicles (SUVs). To prepare supported membranes, SUV suspensions were deposited on cleaned glass substrates, and the remaining vesicles were washed off by intensive rinsing. The supported membrane was incubated in neutravidin solution (5 μg/ml) for 1 h, followed by intensive rinsing. Finally, the sample was incubated with an aqueous solution of the biotinylated polymer (0.1 mg/ml) overnight, and the unbound polymers were washed off.

The vertical Brownian motion of a particle in the interfacial potential V(∆h) is given by a Langevin equation: m ∂ 2 Δ h ∂ t 2 + γ ∂ Δ h ∂ t + ∂ V ∂ Δ h = F stoch . (2)

Here, m is the mass of a particle, γ is the friction coefficient, Δh is the relative height of the particle position, and F stoch is the stochastic force. The first acceleration term is negligible because particle movement is overdamped (Pincus, 1991). The second term is the frictional force given by the Reynolds equation (Reynolds, 1883): γ = 6 π η eff R 2 h . (3)

Here, η _eff is the effective viscosity and ∂ Δ h ∂ t is the relative velocity. If one assumes ∂ Δ h ∂ t ≈ 10 nm / 0 . 03   s = 3×10 - 7 m / s , the Reynolds number Re of the system estimated from the density of the latex particle ρ = 1.04 g/cm³, particle radius R ≈ 5 µm, and viscosity of water η eff = 1 m   N m − 2 s is R e ≤ R ( ∂ Δ h ∂ t ) ρ η eff ∼ 10 − 4 , (4) which is far below 1. Note that the effective interfacial potential near the minimum can be approximated as a harmonic (Derjaguin, 1987): V ( Δ h ) = 1 2 V ″ Δ h 2 . The autocorrelation function of height fluctuation can be calculated within the framework of the fluctuation–dissipation theorem: 〈 Δ h ( δ t ) Δ h ( 0 ) 〉 ≈ 〈   Δ h 2 ( 0 ) 〉 e − δ t τ . (5)

The mean squared amplitude of the fluctuation Δ h 2 ( 0 ) = k B T / V ″ is given by the equipartition theorem and τ = γ / V ″ . Under thermodynamic equilibrium, the effective interfacial potential V(Δh) can also be calculated from the probability function of the fluctuation amplitude P(Δh), following the inverse work functional theorem (Albersdörfer and Sackmann, 1999; Monzel et al., 2015; Higaki et al., 2017): V ( Δ h ) = − k B T ⁡ ln ⁡ P ( Δ h ) + const . (6)

As shown in Scheme 1, the membrane was first incubated with a neutravidin solution prior to the grafting of polymer brushes. Figure 2A shows the change in ∆f and ∆D caused by successive deposition of (I) a supported lipid membrane (II) neutravidin crosslinkers, and (III) pAA-Cys5-biotin (Figure 1A, X = 5). Following the establishment of a baseline in the buffer, the sample was incubated with a suspension of DOPC vesicles doped with 2 mol% DOPE, corresponding to 〈d〉 = 5.5 nm. The obtained shift in frequency and dissipation after vesicle fusion and neutravidin binding are in good agreement with the previous reports, confirming the deposition of a planar lipid membrane and the binding of neutravidin (Keller et al., 2000; Körner et al., 2013; Amadei et al., 2018). The layer parameters obtained by the fitting of QCM-D response using Voigt-Voinova model (Vogt et al., 2004) were presented in Supplementary Material Table S2. After the incubation with pAA-Cys5-biotin solution and the successive rinsing (III), we observed a further decrease in ∆f and an increase in ∆D by the grafting of polymer brushes, ∆f ≈ –21 Hz and ∆D ≈ 4.1 × 10⁻⁶. The brush layer thickness, shear elasticity and viscosity calculated with the same model are d _pAA-Cys5 ≈ 13 nm, µ ≈ 13 × 10³ Pa, and η = 1.0 × 10⁻³ Pa⋅s, respectively. Homogeneous grafting of polymer brushes onto a supported membrane was also confirmed by grafting a fluorescently labeled polymer, Fluor-pAA-Cys5-biotin (Figure 1B). Figure 2B shows the fluorescence microscopy image of a supported membrane functionalized with Fluor-pAA-Cys5-biotin. The fluorescence signal was continuous all over the surface, and the constant intensity profile across the whole field of view confirmed that the membrane surface was uniformly coated with polymer brushes.

Figure 3A shows the change in Δf (solid line) and ΔD (broken line) under a step-wise increase in [Cd²⁺] over time. In the following, both Δf and ΔD values at [Cd²⁺] = 0 M were set as the reference levels to compare the response of polymer brushes to Cd²⁺ ions. As presented in Figure 3B, Δf monotonically increased and ΔD monotonically decreased with increasing [Cd²⁺]. An increase in Δf and decrease in ΔD was clearly observed at [Cd²⁺] ≥ 10⁻⁵ M. The increase in Δf suggests that the binding of Cd²⁺ to pAA-Cys5 caused the loss of water molecules associated with the brushes, while the decrease in ΔD suggests that the binding of Cd²⁺ makes the pAA-Cys5 layer becomes less dissipative and hence more elastic. Notably, the fitting results of QCM-D response exhibited large errors at [Cd²⁺] ≥ 10⁻⁴ M (Supplementary Table S3), implying that Voigt-Voinova model is no longer valid at high [Cd²⁺] due to the damping of dissipation from ΔD ≈ 0 (0 M, reference level) to ΔD ≈ –3.7 × 10⁻⁶ (10⁻⁴ M). Under these conditions, pAA-Cys5 brushes are very poorly hydrated and thus water is no longer a good solvent. We observed a very minor shift in Δf and ΔD values in the opposite direction when [Cd²⁺] was elevated from 10⁻³ M to 10⁻² M, suggesting that the response already reached to the saturation level. Note that both Δf and ΔD came back very close to the original values (0 M) once pAA-Cys5 brushes were rinsed with the buffer containing 10 mM EDTA and 100 mM NaCl. As the deviations of both Δf and ΔD are minor (+1.9 Hz and +0.1 × 10⁻⁶), we presume that the response of pAA-Cys5 to Cd²⁺ ions is reversible.

In the next step, we addressed the following question: what is the structural origin of the prominent [Cd²⁺]-dependent viscoelastic response of pAA-Cys5? Structurally, pAA-Cys5 possesses both –COOH and –SH side chains. It is possible that these two side chains play key roles in the [Cd²⁺]-dependent response because this is a common feature shared with naturally occurring proteins, such as metallothionein and phytochelatin. Additionally, it is well established that cysteine forms solid-state complexes with Zn²⁺ and Cd²⁺ ions (Shindo and Brown, 1965), whereas polysaccharides with –COOH motifs, such as alginate, tend to form egg-box like complexes with Ca²⁺ (Morris et al., 1978).

To unravel how the –SH group affects the interaction with Cd²⁺ ions, we monitored the interaction of pAA-Cys0 (no –SH groups, Figure 1A, X = 0) with Cd²⁺ ions. Figure 4A shows the Δf (solid symbols) and ΔD (open symbols) of pAA-Cys0 brushes as a function of [Cd²⁺]. Compared with pAA-Cys5 (Figure 3B), pAA-Cys0 showed a markedly lower response. In particular, Δf increased by + 4.9 Hz and ΔD decreased by –0.5 × 10⁻⁶ as [Cd²⁺] increased from 0 to 10⁻² M. Moreover, the direction of shift was opposite for both Δf and ΔD, suggesting that increasing [Cd²⁺] led to a slight increase in the amount of water molecules associated with pAA-Cys0 brushes. The opposite trend in response as well as the lower changes in both Δf and ΔD suggest that the mode of interaction between pAA-Cys0 (–COOH only) and Cd²⁺ ions is distinct from that between pAA-Cys5 (both –COOH and –SH) and Cd²⁺ ions. Next, to study the influence of –COOH groups on the interaction with Cd²⁺ ions, we monitored the interaction of pPEGMA-Cys5 (no –COOH groups, Figure 1C) with Cd²⁺ ions. Figure 4B shows the Δf and ΔD of pPEGMA-Cys5 attached to a supported membrane in the absence and presence of 1 mM Cd²⁺ ions, showing almost no response to Cd²⁺ ions. The obtained data clearly indicate that the co-existence of both –COOH and –SH groups are necessary to respond to Cd²⁺ ions.

We further investigated how the molar fraction of –SH side chains affect the viscoelastic response by grafting pAA-Cys20-biotin (Figure 1A, X = 20) onto the supported membrane doped with 2 mol% DOPE-biotin. Figure 4C shows the change in Δf (solid symbols) and ΔD (open symbols) of pAA-Cys20 brushes as a function of [Cd²⁺]. It is notable that pAA-Cys20 exhibited qualitatively the same tendency as pAA-Cys5 (Figure 3B). Increasing [Cd²⁺] led to a monotonic increase in Δf and ΔD, and the response became prominent at [Cd²⁺] ≥ 10⁻⁵ M. It is notable that the change in both Δf and ΔD was lower for pAA-Cys20 than for pAA-Cys5, suggesting that the response did not increase monotonically with increasing molar fraction of the –SH moiety X. Figure 4D shows a comparison of the changes in Δf and ΔD between [Cd²⁺] = 10⁻³ M and 0 M for pAA-Cys0, pAA-Cys5, pAA-Cys20, and pPEGMA-Cys5. Our data demonstrated that the presence of both –COOH and –SH groups is prerequisite for [Cd²⁺]-dependent modulation of the interfacial viscoelasticity, and the sensitivity does not increase with increasing molar fraction of the –SH group. The changes in Δf and ΔD recorded for pAA-Cys0 and pAA-Cys20 are presented in Supplementary Figure S4.

A distinct [Cd²⁺]-dependent change in the viscoelastic properties of pAA-Cys5 brushes (Figure 3) strongly suggests that binding of Cd²⁺ ions modulates the interfacial potential in a [Cd²⁺]-dependent manner. For this purpose, we monitored the vertical Brownian motion of latex particles (diameter of 10 µm) hovering on the surface of polymer brushes using RICM (Figure 5A). Changes in the particle height can be detected by analyzing interference patterns obtained from RICM images (Fröhlich et al., 2021). Figure 5B is an RICM snapshot of a particle hovering on the surface of pAA-Cys5 brushes at [Cd²⁺] = 0 M. The concentric rings reflect the shape of the particle. Figure 5C shows the intensity profile calculated from radial integration of the concentric interference patterns in panel b. The mean radial intensity profiles after azimuthal integration are indicated by black symbols. The radial integration was introduced to improve the accuracy for Δh determination (Fröhlich et al., 2021). Figure 5D shows the height profile of the particle calculated from the intensity profile shown in panel c using Eq. (1).

Figure 6A shows the fluctuation of the particle height recorded over time and the corresponding interfacial potential measured at [Cd²⁺] = 0 M (top), 10⁻⁵ M (middle), and 10⁻³ M (bottom). The plot of Δh vs. t clearly indicated that the amplitude of height fluctuation was strongly damped at [Cd²⁺] = 10⁻³ M. The effective interfacial potential V(Δh) calculated using Eq. 6 (Figure 6B) was well approximated by parabolic functions (solid line) described within the framework of Derjaguin’s approximation for weak interactions (Derjaguin, 1987). The obtained potential curvature V″ and hence the spring constant of harmonic potential showed an increase from 0.8 × 10⁻²¹ J/nm² (0 M) to 5.9 × 10⁻²¹ J/nm² (10⁻³ M), suggesting that the particle height was sharply confined in the presence of Cd²⁺ ions. As shown in Figure 6C, the potential curvature V″ was plotted as a function of [Cd²⁺]. The potential curvature V″ exhibited a clear increase at [Cd²⁺] ≥ 10⁻⁴ M. It is notable that Δf and ΔD showed significant deviation from the swollen brushes at this concentration.

Because changes in the potential curvature may correlate with changes in hydrodynamic friction, we calculated the autocorrelation of the height fluctuation for different [Cd²⁺] (Figure 6D). The correlation functions were well fitted by single exponential functions (solid lines, Eq. 5). Increasing [Cd²⁺] led to a distinct decrease in the mean squared amplitude of fluctuation Δ h 2 ( 0 ) and thus the characteristic relaxation time τ. The increase in [Cd²⁺] from 0 through 10⁻⁵ to 10⁻³ M resulted in a decrease in Δ h 2 ( 0 ) from 5.1 through 1.9 to 0.7 nm², reflecting the suppression of vertical Brownian motion. As presented in Figure 6D, the relaxation time τ at [Cd²⁺] = 0 M, τ ≈ 61 ms, showed a distinct decrease by adding [Cd²⁺] = 10⁻⁷ M, close to the detection limit (τ < 30 ms). The data suggest that the hydrodynamic friction γ significantly increases even at the very low [Cd²⁺] not detected by the change in the potential curvature.

Naturally occurring proteins possessing –SH and –COOH moieties show high specificity towards heavy metal ions. Therefore, we further examined the ion specificity of pAA-Cys5 brushes by performing experiments in the presence of Ca²⁺ ions, which are similar to Cd²⁺ ions in ionic radius. Figure 7A shows the change in Δf (solid symbols) and ΔD (open symbols) of pAA-Cys0 brushes as a function of [Ca²⁺], showing distinct differences. First, the change in QCM-D response was undetectable until [Ca²⁺] = 10⁻⁴ M, an order of magnitude lower than the onset level for Cd²⁺. Moreover, the change in Δf and ΔD was less pronounced compared with the response to [Cd²⁺]. The differences in Δf and ΔD between [Ca²⁺] = 10⁻³ M and 0 M (Δf = 5.5 Hz and ΔD = –2.5 × 10⁻⁶) were less significant compared with the response to Cd²⁺ ions (Figure 4D). The interfacial potentials V(Δh) calculated for [Ca²⁺] = 10⁻⁶ and 10⁻³ M were plotted, as shown in Figure 7B. Compared with the data presented in Figure 6D, no clear differences in the global shape of the potential as well as the curvature V” were detected, suggesting that the less pronounced change caused by Ca²⁺ binding was not enough to alter the confinement of particle position. Figures 7C, D show the response of pAA-Cys0 and pPEGMA-Cys5, respectively. The response of pAA-Cys0 to Ca²⁺ was weaker than that to Cd²⁺ (Figure 4A). The differences in Δf and ΔD between [Ca²⁺] = 10⁻³ M and 0 M were Δf = –2.0 Hz and ΔD = –0.3 × 10⁻⁶, respectively. pPEGMA-Cys5 showed no detectable changes even in the presence of [Ca²⁺] = 10⁻³ M (Figure 7D). Changes in Δf and ΔD monitored over time are presented in Supplementary Figure S5. These data suggest that the response of pAA-Cys5 to Ca²⁺ is significantly enhanced by the coexistence of –SH and –COOH groups, which is similar to the data presented in Figure 4. However, the one order of magnitude lower sensitivity and very minor change in V” observed here suggest ion-specific interaction of pAA-Cys5, that is, with Cd²⁺ ions, which may be attributed to the formation of a chelator complex, such as the one reported for N-acetylcysteine in aqueous solution (Jalilehvand et al., 2011).

Ample studies have demonstrated that the conformation of polyelectrolyte brushes strongly depends on the pH and salt concentration (Ahrens et al., 1998; Rehfeldt et al., 2006b). So far, to highlight the influence of [Cd²⁺] and to compare the effects of Cd²⁺ and Ca²⁺ ions, we performed all experiments in the presence of 10 mM Tris-HCl and 100 mM NaCl. This enables us to sustain pH constant (7.4) and minimize crosstalk with electrostatic screening (Debye screening length κ ⁻¹ < 1 nm). In the next step, we removed the supporting salt (100 mM NaCl) and examined if the ion-specific response of pAA-Cys5 brushes was enhanced. Figure 8 shows the height fluctuation over time (top) and the interfacial potentials V(Δh) (bottom) measured in 10 mM Tris-HCl (pH 7.4) in the presence of [Na⁺] = 10 mM (a) [Cd²⁺] = 10 mM (b), and [Ca²⁺] = 10 mM (c). The corresponding potential curvature values were V” = 3.7 × 10⁻²² J/nm² (Na⁺), 125 × 10⁻²² J/nm² (Cd²⁺), and 66 × 10⁻²² J/nm² (Ca²⁺). It is notable that the potential curvature V” increased by an order of magnitude by changing monovalent Na⁺ ions to divalent ions, and the effect was more pronounced with Cd²⁺ ions than with Ca²⁺ ions. Distinct sharpening of the potential confinement is also a reflection of a significant increase in the hydrodynamic friction γ and hence a decrease in the relaxation time τ. As presented in Supplementary Figure S6, the relaxation time τ = 75 m s could be calculated in the buffer containing NaCl, although τ was not detectable in the buffers containing CaCl₂ and CdCl₂. This seems reasonable because τ was not detectable at [Cd²⁺] = 10⁻⁶ M even in the presence of 100 mM NaCl.

To highlight how the presence of 100 mM NaCl affects the response of pAA-Cys5 brushes to divalent cations, we measured the zeta potential. Here, we deposited supported membranes doped with 2 mol% DOPE-biotin on porous SiO₂ microparticles (diameter 3 µm) and grafted pAA-Cys5 brushes via neutravidin crosslinkers (Figure 9A, see Supplementary Information S7 for the detail of preparation). As shown in Figure 9B, grafting of Fluor-pAA-Cys5 brushes enabled uniform coating of all particles with brushes. Figures 9C, D show the zeta potential ζ plotted as a function of [Cd²⁺] (red) and [Ca²⁺] (blue), measured in the presence and absence of 100 mM NaCl, respectively. As a general tendency, we observed a monotonic increase in ζ with increasing divalent cation concentration, and the response was more pronounced with Cd²⁺. Remarkably, the change caused by increasing [Cd²⁺] from 10⁻⁶ M to 10⁻² M showed a distinct difference: Δζ _100mM ≈ + 15 mV and Δζ _0mM ≈ + 50 mV. The stronger modulation of surface potential in the absence of supporting salt indicates that the more pronounced change in V(Δh) presented in Figure 8 can be interpreted as the additional contribution of electrostatics.

During the past several decades, our understanding of how polyelectrolytes interact with ions in aqueous solution has advanced significantly, from both theoretical and experimental aspects. In particular, ion-specific interactions have drawn increasing attention owing to the potential applications of polyelectrolytes in microencapsulation, controlled release, and fabrication of switchable and adaptable surfaces. Hofmeister’s pioneering work on the ion-specific precipitation of proteins (Hofmeister, 1888a; Hofmeister, 1888b) was extended to explain experimental observations as a manifestation of the hydration strength, known as kosmotropes and chaotropes. However, this simple description lacks details of proteins and other biopolymers, which makes it very difficult to understand many exceptions of the Hofmeister’s series (Zhang and Cremer, 2009; Jungwirth and Cremer, 2014; Gibb, 2019).

In this study, we found that the response of pAA-Cys5 brushes to Cd²⁺ was structure dependent and attributed this finding to common structural motifs, such the –SH and –COOH groups. Intriguingly, the interaction of pAA-Cys5 brushes with Ca²⁺ was approximately one order of magnitude weaker than that with Cd²⁺, and, more remarkably, Ca²⁺ did not cause a remarkable change in the curvature of interfacial potentials. This suggests that the interaction of pAA-Cys5 with Cd²⁺ is not only stronger than but also different from that with Ca²⁺. One possible scenario is the formation of coordination complexes with Cd²⁺. For example, Jalilehvan et al. demonstrated that a complex formed from N-acetylcysteine and Cd²⁺ ions in aqueous solutions using NMR (Jalilehvand et al., 2011). Although these experiments with block copolymers are non-trivial owing to the presence of two different monomers, molecular-level understanding of ion-specific interactions would help us rationally design polymeric materials for the selective capture of toxic ions in our aquatic environments.

To gain molecular-scale understanding of ion-specific interactions involving biopolymers or bio-inspired polymers, another experimental challenge is to localize specific ions with high spatial resolution, that is, Δr < 1 nm. By the combination of ultrafiltration and inductively coupled plasma optical emission spectrometry (ICP-OES), it is possible to determine the amount of Cd²⁺ ions bound to 1 g of pAA-Cys5 polymer in solution (Supplementary Information S8). However, it is extremely difficult to determine how ions are associated with certain units using commonly used optical spectroscopy, such as vibrational spectroscopy. This can be overcome partially if one confines polyelectrolytes or charged biopolymers near the interface and illuminate the interface with evanescent fields, because the depth of penetration and hence the illumination volume can be modulated by adjusting the angle of incidence. Unfortunately, widely used X-ray and neutron scattering and reflectivity cannot discriminate ions, such as K⁺ vs. Ca²⁺, although these techniques detect contrast in scattering density with high spatial resolution. Previously, we reported that grazing incidence X-ray fluorescence is able to determine the density profile of specific elements/ions with an accuracy of several Å (Schneck et al., 2010; Abuillan et al., 2013). Element specific, surface sensitive techniques combined with advanced computational methods would help us not only understand the electrostatics of soft matters near the interface but also rationally design new bio-inspired materials by optimizing the molecular-level structure to gain maximum functional output.