Binding Mechanism of the Model Charged Dye Carboxyfluorescein to Hyaluronan/Polylysine Multilayers

Biopolymer-based multilayers become more and more attractive due to the vast span of biological application they can be used for, e.g., implant coatings, cell culture supports, scaffolds. Multilayers have demonstrated superior capability to store enormous amounts of small charged molecules, such as drugs, and release them in a controlled manner; however, the binding mechanism for drug loading into the multilayers is still poorly understood. Here we focus on this mechanism using model hyaluronan/polylysine (HA/PLL) multilayers and a model charged dye, carboxyfluorescein (CF). We found that CF reaches a concentration of 13 mM in the multilayers that by far exceeds its solubility in water. The high loading is not related to the aggregation of CF in the multilayers. In the multilayers, CF molecules bind to free amino groups of PLL; however, intermolecular CF–CF interactions also play a role and (i) endow the binding with a cooperative nature and (ii) result in polyadsorption of CF molecules, as proven by fitting of the adsorption isotherm using the BET model. Analysis of CF mobility in the multilayers by fluorescence recovery after photobleaching has revealed that CF diffusion in the multilayers is likely a result of both jumping of CF molecules from one amino group to another and movement, together with a PLL chain being bound to it. We believe that this study may help in the design of tailor-made multilayers that act as advanced drug delivery platforms for a variety of bioapplications where high loading and controlled release are strongly desired.


■ INTRODUCTION
The layer-by-layer deposition of polyelectrolyte multilayers has become very popular in the last decades due to its simplicity, low cost, automation, and reproducibility. 1−7 The method is based on sequential deposition of oppositely charged polymers onto a solid surface. 1,3−7 Physical−chemical properties of the multilayers can easily be adjusted to meet various requirements by proper choice of preparation conditions, variation of polyelectrolyte chain length and nature, and postmodification. 8−11 Furthermore, the multilayers have attracted significant attention, since they have been shown to serve as good candidates for a number of bioapplications. 2,12−24 The planar multilayers (films) and curved structures (capsules) mostly play the role of hosting reservoirs and releasing numerous biologically active molecules, such as nucleic acids, proteins, and peptides. 2,25−30 Recently, a number of reports appeared to demonstrate the high capacity of multilayers to host small molecules, including drugs and dyes. 31−33 This makes the multilayers very attractive as biocoatings (e.g., implant coatings) that host bioactive drugs (antibiotics, anticancer drugs, small biofactors such as growth factors, cytokines, etc.) and release them in a controlled manner, for instance, through erosion or biodegradation.
One of the most commonly used methods for loading biomolecules of interest into the multilayers is the incubation of the as-prepared multilayers in a solution of the bioactive agent. Simple dyes are often used as model molecules to mimic the behavior of a real drug. Additionally, model dyes such as carboxyfluorescein (CF) or rhodamine can be directly monitored by fluorescence or absorbance. In some cases, the capacity of a multilayer to store a dye is determined by the nature of the last polyelectrolyte layer deposited. 34 Fluorescently labeled paclitaxel was homogeneously distributed throughout multilayers composed from hyaluronic acid (HA) and poly-L-lysine (PLL). 33 Likewise, it was possible to load the anti-inflammatory drug diclofenac into cross-linked multilayers, and the amount of the stored drug can be tuned by the multilayer thickness/number of layers. 35 Furthermore, the same study showed that paclitaxel loaded into multilayers was effective in killing cells with an efficiency of about 90%. Variation of the number of layers affects both the loading and release rate of the loaded molecules. 36 High water content in the biopolymer-based multilayers, along with their structural properties, allows one to load large macromolecules such as proteins (albumin, lysozyme, others) 28,37 and nucleic acids. 29,30 Full exploitation of these films for bioapplications (cell culture and tissue engineering) requires that these systems should offer a controlled release of the stored molecules. The trigger for the release can be a change in the physical−chemical properties of the multilayers or environment changes: temperature, pH, composition, ionic strength, hydrolysis, and light. 38−43 These stimuli can be employed for both short-term 31 and long-term 44 release. Though the possibilities for applications and control over loading and release performance are countless, the mechanism of storage/release of small molecules/drugs in the multilayers is still not well understood and a number of reports on loading and release of small drugs show high interest for this topic. [32][33][34]36,45,46 It has recently been demonstrated that small charged dyes can be loaded into the biopolymer-based multilayers in enormous concentrations. 47 This concerns not only model dyes such as CF and rhodamine but also biologically relevant small charged molecules like adenosine triphosphate (ATP), which is of particular interest due to its well-known role in bioenergy storage and conversion. A number of reports focus on investigation of ATP loading and release into/from surface films employing the energy of ATP hydrolysis, 48 electrochemically stimulated release, 49 ATP-induced self-assembly, 50 and loading into layered double hydroxides through intercalation of ATP and other nucleotides. 51 More complex self-assembled ATP-containing structures based on metal nanoparticles and block copolymers with bioinspired binding units have also been reported. 52,53 The multilayers that have been given enormous attention as a model system are assembled from model biopolymers HA and PLL. These films can easily reach up to a few micrometers in thickness or even tens of micrometers due to their exponentiallike growth character, 54−57 and thus they offer considerable reservoir capacity. 47 HA/PLL multilayers can be used as supports for cell culture. 58,59 Bearing in mind their high storage capability, these multilayers are perfect candidates as drug delivery systems for cellular applications. However, in order to comprehend their delivery potential, it is of immense importance to understand the mechanism by which a drug is withheld in the multilayers.
In this paper we focus on the binding mechanism of the small, charged, fluorescent dye CF to the HA/PLL multilayers assembled by the dipping technique. Spectral characteristics of CF along with its mobility and adsorption equilibrium capacity in the multilayers are analyzed and used as major indicators for assessment of CF binding to multilayers. We firmly believe that a better understanding of the binding mechanism of small dyes such as CF could be used to design novel, effective, polymerbased systems with tuned loading and release characteristics required for cutting-edge drug delivery and tissue-engineering applications.
Assembly of HA/PLL Multilayers. HA/PLL films were assembled on 12 mm round glass slides using an automated dipcoating machine according to the protocol described elsewhere. 59 Briefly, after washing with Hellmanex II, 14 mm round glass slides were immersed for 10 min in 1 mg/mL PEI, followed by washing in 10 mM TRIS buffer containing 15 mM NaCl, pH 7.4 (referred to as TRIS-buffer) three times. After that slides were immersed for 10 min in 0.5 mg/mL HA, washed three times in the buffer, immersed in 0.5 mg/mL PLL, and again washed three times in the buffer. The HA/ PLL deposition step was repeated 24 times. The films were always terminated with PLL as the outermost layer.
Loading the Multilayers with CF and Analysis of CF Fluorescence. Loading of HA/PLL multilayers was achieved by incubating the HA/PLL in 1 mL of a 1 μM solution of CF in TRISbuffer in darkness for 24 h. The loading efficiency and released amount of the loaded compound were determined by analysis of the fluorescence of the supernatant.
Fluorescence emission spectra of CF solutions in pure buffer and in the buffer containing PLL were taken at 492 nm excitation. The concentration of CF was kept fixed at either 5 or 300 μM, and the PLL concentration was varied to obtain different ratios between the number of lysine units (M r = 209) in PLL and the CF molecules (molar charge PLL:CF ratio). Furthermore, fluorescence emission spectra of CF loaded in the (HA/PLL) 24 multilayers were determined; the multilayers were loaded with CF by incubation in 1 μM CF. The measurement was done at 1 and 60 min after immersion of the multilayers into the TRIS-buffer in the cuvette; the moment of immersion is considered as the start of the release.
Real-time observation of the loading was done by online measurement of CF fluorescence in solution where the (HA/PLL) 24 multilayers were incubated. The experiment was done in quadruplicate. The amount of CF loaded in the multilayers was determined as the difference in the amount of CF measured in aliquots taken from the solution with the multilayers and the control one (without the multilayers; bar coverslip used instead).
Analysis of CF Diffusion in Multilayers by Fluorescence Recovery after Photobleaching. The diffusion coefficient and the amount of immobile fraction were assessed by evaluation of data resulting from the fluorescence recovery after photobleaching (FRAP) experiment. In general, the photobleaching may have an effect on the chemical structure of a fluorescent dye that may affect its interaction with its surroundings. However, FRAP is well-accepted method to assess the mobility of fluorescent probes like CF, and this is why we used this method in this study. The experimental procedure and the data evaluation procedure were similar to procedures described elsewhere 60 but using planar multilayers. Glass slides with deposited multilayers containing CF were mounted into a holder, covered by a layer of TRIS-buffer to protect them from drying, and placed under a confocal laser scanning microscope (Zeiss LSM 510 meta/Axiovert 200M). Using a 63×/1.4 objective the sample was bleached by the 488 nm laser line of an Ar ion laser at its full power and scanned by highly attenuated laser power. The scanning area spanned 73.12 × 73.12 μm (512 × 512 pixels) and the bleached area spanned 73.12 × 2.14 μm (512 × 15 pixels). The proper z-position was determined as the position of the highest intensity of the fluorescence signal. The sample was primarily scanned twice, afterward bleached (by scanning the laser 50 times over the bleached area, resulting in a dip of the fluorescence signal), and then again scanned 30 times. The time span between following images was 2 s, and the scanning time per pixel was 3.2 μs. The evaluation procedure of the acquired data was based on an analytical solution of Fickian diffusion. 61 The evaluation of the immobile fraction amount is based on analysis of the dip depth as a function of time. The measurements were repeated six times. The details of the evaluation procedure are given below.

Research Article
The raw images from the FRAP experiment were primarily cropped (to limit edge effects) and averaged (ImageJ, NIH) in the direction perpendicular to the direction of diffusion. The resulting onedimensional FRAP profiles were transferred into a spreadsheet template (MS Excel) for data corrections and analysis. The set of profiles resulting from a single FRAP experiment were normalized to the prebleach fluorescence intensity, corrected to the inhomogeneity of the local fluorescence, and corrected for unwanted bleaching and edge effects in the other dimension.
Every FRAP profile is fitted by a Gaussian curve defined as where I(x,t) is the fluorescence intensity at a distance x from the dip and a time point after bleaching t, I 0 (t) is the fluorescence intensity of the background (ideally its value is constant and close to 1), A(t) is the depth of the dip at the time point t, and w describes the width of the Gaussian between inflection points. All four parameters are free for fitting with every separate FRAP profile. The fitting is performed as a minimization of the sum of squared residuals. After fitting the Gaussian function into FRAP profiles, the diffusion coefficient D is evaluated as half of the slope of the plot of w 2 versus t: Only the first three points of this plot were used for slope evaluation, because the plot tends to deviate when more than one fraction (e.g., immobile fraction) is present. No change over time of the w 2 (not equal to zero) would indicate the presence of the immobile fraction because of no recovery of the stable fluorescent profile. The slope is influenced already from the onset of the recovery process; thus, the resulting D rather represents an average diffusion coefficient.
The dimensionality of diffusion d may be evaluated as twice the slope of the plot log(A) versus log(t + t 0 ) where the newly appeared parameter t 0 refers to a time shift correcting deviations from the approximation of instantaneous bleaching of an infinitely thin region. The time shift can be found by varying its value until the plot log(A) versus log(t + t 0 ) becomes linear (here implemented as a maximization of the coefficient of determination of an expected linear function). Because also this plot tends to deviate when more than one fraction is present, only the first 10 points of the plot were used for evaluation of d.
The relative amount of the immobile fraction K rel is evaluated from the time development of the dip depth A(t) by fitting the function where A(t) is the depth of the dip at time point t, M is the overall reduction of the fluorescence intensity caused by bleaching, D is the diffusion coefficient, t is time after bleaching, t 0 is the time shift correction, d is the dimensionality of diffusion, and K is the partial depth of the dip attributed to the immobile fraction. D and d were taken from the previous evaluation and fixed during fitting. These values tend to be underestimated due to the influence of the immobile fraction, but their presence is essential during fitting for a good estimation of the amount of immobile fraction. The d value used here is bottom-limited to a value of 1, because values lower than 1 do not reflect reality. M, t 0 , and K are free for fitting. The input values A(t) are acquired directly from the FRAP profiles, not from fitting the Gaussian function, because the evaluation of the immobile fraction is intended to be independent from the evaluation of D as much as possible. Afterward, K rel is obtained by relating K to the depth of the dip just after bleaching where K rel stands for the relative amount of the immobile fraction, K is the depth of the dip attributed to the immobile fraction resulting from eq 4, and A 0 (t) is the depth of the dip just after bleaching (t = 0). Such an approach to the evaluation of the immobile fraction amount expects that only two fractions are present in the sample: a single mobile fraction with certain D and an immobile one. In other cases, the evaluation of the immobile fraction amount gives approximate results. When the amount of the immobile fraction is known, a contribution of this fraction can be subtracted from raw FRAP profiles. The evaluation procedure can then be repeated with the FRAP profiles free of immobile fraction. Results of this repeated evaluation describe the pure mobile fraction rather than a mixture of the mobile and the immobile fractions.
■ RESULTS AND DISCUSSION Spectral Characteristics of CF in Solution and in the HA/PLL Multilayers. Our previous study has demonstrated that CF can be loaded into (HA/PLL) 24 multilayers in TRISbuffer giving high concentrations in the multilayers of at least a few millimolar. 47 Incubation of excess CF with the multilayers (at concentrations up to 100 μM) under the same conditions has revealed that the saturation concentration of CF in the multilayers is 13 mM. This is far more than the CF solubility in the same buffer (about 0.5 mM).
In this work, we focus on the mechanism of binding of CF to multilayers in order to explain the strong accumulation of CF in the multilayers. The molecular structures of CF and PLL and HA biopolymers used to build the multilayers are presented in Figure S1 of the Supporting Information (SI). The dye CF possesses a carboxylic group that is deprotonated at the physiological pH and potentially allows the dye to interact with charged polymers through electrostatic attractive forces. Thus, throughout all experiments in this study, TRIS-buffer containing a rather low concentration of added salt (15 mM) and possessing the physiologically relevant pH 7.4 has been used. A low salt concentration is needed to ensure that electrostatic interactions between CF and permanent charges on the polymer backbone are not screened by salt counterions.
In order to investigate the interaction of CF with multilayers, fluorescent spectra of CF have been taken in the presence of the polymers from which the multilayers are made, PLL and HA. Figure S2A (SI) demonstrates that HA has no effect on the maximum of CF fluorescence and that the presence of PLL results in a shift of the maximum of fluorescence from 512 nm (without PLL) to 514.5 nm (0.5 mg/mL PLL solution). This can be explained by electrostatic binding of the oppositely charged CF to amino groups of PLL and no electrostatic interaction between CF and HA, which both carry a negative charge. A variation of the PLL:CF molar charge ratio has shown that the red-shift of the fluorescence maximum (from 512 to 515.5 nm) takes place at a PLL:CF molar charge ratio of more than 5 ( Figure S2B, SI). The conditions of a high excess of PLL can simulate those in the multilayers, where PLL may be in much higher excess compared to CF because of the high concentration of PLL in the multilayers, which can be estimated as 250 mM taking into account that the multilayers have about 80% of water, that the molar charge ratio between PLL and HA is about 2, and that 46% of amino groups of PLL is free, i.e., unpaired with HA. 62 Apart from the binding of CF to amino groups of the PLL, CF is prone to aggregation, due to its core hydrophobic part of

ACS Applied Materials & Interfaces
Research Article the molecule. Considering this, one can propose three potential models of CF binding to multilayers, as shown in Figure 1. The model A suggests that only CF−CF interactions take place in the multilayers. This may be reasonable because the CF concentration in the multilayers by far surpasses the solubility of CF in buffer, as discussed above. Such a high concentration may be explained by formation of CF aggregates if one assumes that the CF−CF interaction is stronger than the CF−PLL interaction. Contrary to the first model, model B assumes that the energy gain for formation of CF−PLL is higher than that of the CF−CF interaction and suggests that no aggregate formation occurs and that CF is solely bound to PLL and is also homogeneously distributed throughout the film. The third model C is basically a combination of the previous two models A and B and is based on the assumption that both CF−CF and CF−PLL interactions take place.
Further, the CF-loaded multilayers and CF solutions have been analyzed to reveal which model can better describe the mechanism of binding of CF to the multilayers. Fluorescence spectra of CF in the buffer, CF loaded in the multilayers, and CF released from the multilayers are shown in Figure 2. One can compare the emission maximum of the spectra and take it as a measure of CF aggregation and/or CF−PLL interaction. About 70% of CF stored in the multilayers is released after 1 h incubation in a fresh buffer solution. 47 Therefore, we have analyzed the CF spectra in the multilayers at the beginning of the release process (1 min incubation in the buffer), when almost all CF molecules should still be present in the multilayers, and after 1 h of incubation, when the majority of CF molecules have already left the multilayers (Figure 2A). One can see that CF molecules stored in the film have an identical emission maximum, regardless of the incubation time point, and it is about 516 nm. In comparison, CF released to the buffer has the maximum emission of 512 nm. This wavelength is similar to that of free CF in solution at a rather low CF concentration of 1 μM and is the same as in the stock solution used in this experiment for CF loading into the multilayers ( Figure 2B). However, a CF solution at the higher concentration of 500 μM gives an emission maximum of 525 nm, indicating an effect of the CF concentration on the fluorescence emission maximum.
These findings allow one to conclude that CF released from the multilayers is in a free state (nonaggregated); however, in the multilayers its emission maximum is shifted to higher wavelengths of about 516 nm. To understand the reason for the shift, we have considered the effect of CF concentration on its emission maximum more in detail. Figure 2C shows that an increase of CF concentration in solution results in a progressing red-shift of emission maximum up to 525 nm. This is obviously because of CF−CF interactions, i.e., formation of CF aggregates in solution. 63,64 CF is becoming insoluble in TRIS-buffer at concentration of more than 500 μM.
The results discussed above and presented in Figure 2 clearly show that although the concentration of CF in the film is in the range of a few millimolar, it has an emission maximum at 516 nm, much below 525 nm, which is the emission maximum of less concentrated 0.5 mM CF saturated solution in buffer. Additionally, there are no other peaks identified for CF stored in the film, apart from that at 516 nm, meaning that only one CF population is present. It would then not be misleading to conclude that there are no pure CF aggregates in the multilayers. This means that the first model A (Figure 1) is not feasible and that it is either model B or C.
In order to assess the strength of CF−PLL and CF−CF interactions and make it reflective of the CF microenvironment in the multilayers, we have performed experiments in solution

Research Article
where highly concentrated CF solution (0.3 mM, a concentration comparable to that in the multilayers) has been titrated with PLL ( Figure 3). With no PLL present (at zero PLL:CF molar charge ratio), the maximum emission of CF is about 525 nm, caused by CF aggregation as discussed above ( Figure 2C). The wavelength maximum of emission progressively increases to about 547 nm until the molar charge ratio of 5. At a higher molar ratio, the emission maximum decreases back to about 530 nm (at the molar charge ratio of 42). The measured fluorescence intensity behaves oppositely, having a minimum at the PLL:CF molar ratio of 5.
The results obtained are explained schematically in the figure above the graph in Figure 3. Addition of positively charged PLL induces an interaction between the negatively charged CF aggregates, making the aggregates larger. This increases the number of CF−CF interactions, which results in a significant reduction of fluorescence due to self-quenching of CF molecules and a significant increase of the red-shift of the fluorescence emission maximum. Both the red-shift and selfquenching are phenomena related to intermolecular CF−CF interactions, i.e., between π-electrons of CF molecules. 63,64 At the PLL:CF molar charge ratio of more than 5, the high excess of PLL results in destruction of the formed aggregates and formation of the PLL−CF complex. This, in turn, enhances the fluorescence and reduces the red-shift due to disruption of CF aggregates. A PLL:CF molar charge ratio higher than 42 has not been tested because of the need for a highly concentrated PLL solution, but we hypothesize that at a higher PLL concentration the red-shift will be further reduced down to the emission maximum of 516 nm, as has been found for PLL−CF complex at low CF concentration but high PLL excess ( Figure  S2B, SI).
The results above allow one to conclude that PLL−CF interaction should dominate with an excess of charged amino groups of PLL compared to carboxylic ones for CF. This is what most likely happens in the multilayers loaded with CF, because in the multilayers the concentration of overall charges of PLL (about 250 mM) significantly exceeds the concentration of CF (13 mM). According to the results presented above, it can be concluded that the CF binding in the multilayers can be described by models B or C (Figure 1). At the same time, the CF environment in the PLL solution and inside the multilayers may be different (e.g., due to the presence of HA), and further analysis of CF−multilayer binding has been performed to better understand the binding mechanism and reveal the model of the binding (Figure 1).

Research Article
Analysis of CF Diffusion into the Mutilayers. A series of FRAP experiments have been conducted to check the mobility of CF in the film. This may help to explain the interaction of CF and PLL. The mean CF diffusion coefficient (D CF ) and the amount of immobile CF fraction were estimated from these experiments (Figure 4). The resulting diffusion coefficient is an average diffusion coefficient of all CF fractions presented in our sample and was found to be 1.3 ± 0.6 μm 2 /s. The amount of immobile fraction refers to the CF fraction that seems to be immobile on the time scale of our experiment (D < 0.01 μm 2 / s). The immobile fraction was found to be about 9 ± 10% of the total CF presented in the multilayers. One can compare the found D CF in multilayers and D of PLL analyzed by a similar FRAP approach. 65 The mean D CF is similar to the D PLL for the fastest PLL fraction, which is about 1 μm 2 /s. Three PLL diffusive fractions were found in the multilayers. 65 Slower fractions of PLL have also been identified, but their contribution into the mean D PLL measured will not be significant because the second diffusive fraction is much slower (about 0.1 μm 2 /s) than the fastest one. The third fraction of PLL is immobile (D below 0.001 μm 2 /s), but its content in the multilayers (30−40%) is much higher than that for CF found here (9%).
The results above suggest that CF diffusion in the multilayers is most probably a result of both (i) free diffusion of CF through jumping of CF molecules from one PLL backbone to another and (ii) diffusion of CF molecules together with PLL chains (bound to the PLL backbone). These results corroborate well with our previous findings based on analysis of CF release from multilayers as a function of the multilayer properties. 47 To further understand the binding of CF to the multilayers, the kinetics of CF interaction with the multilayers has been assessed at increased CF concentrations (Figure S3, SI). From

ACS Applied Materials & Interfaces
Research Article these results, the adsorption isotherms have been constructed and fitted by six well-known adsorption models based on equilibrium adsorption process, i.e., Langmuir, Freundlich, Langmuir−Freundlich, Temkin, sigmoidal Langmuir, and Brunauer−Emmett−Teller (BET) models 66 ( Table 1). The main characteristics of the models are summarized in Table S1 (SI). Briefly, only the Langmuir model is based on the assumption of formation of a monolayer (monoadsorption), and all other models assume that each adsorbed molecule provides a new binding site for another molecule to be adsorbed. All other five polyadsorption models are based on the assumption that either adsorbate (CF) molecules do not interact with each other or the interaction of adsorbate (CF) and adsorbent (HA/PLL film) has a cooperative or competitive nature. It is important to note that for these models only Langmuir and BET equations (Table 1) can be derived on the basis of statistical thermodynamics, while the other models are empirical. The data used for fitting was the average of the quadruplicate measurement. As the measure of the fit we used the coefficient of determination (R 2 ) in order to reveal the model that better fits the experimental results (R 2 closest to a unit). R 2 was calculated using the equation for the general case where SS res is the sum of squared residuals, It is a parameter to be minimized during the fitting procedure. y stands for input data values, and f stands for predicted values. SS tot is the total sum of squares, SS tot = ∑ i (y i − y av ) 2 . y stands for input data values, and y av stands for the average value of variable y in our sample. This parameter is proportional to the variance of y.
It should be noted that the experimental adsorption isotherm has a clearly defined region of saturation ( Figure 5). This was taken into account by fixing the parameter C film max at 13.1 ± 1.2 mM. By the same token, Freundlich, Temkin, and BET equations were fitted until the saturation plateau. Figure 5A presents the fitting of experimental results of the CF loading into the multilayers by the three above-mentioned adsorption models that provided the best fitting results (correlation coefficient R > 0.90), i.e., BET, Freundlich, and Langmuir−Freundlich models. Adsorption isotherms for the three other models (Langmuir, sigmoidal Langmuir, and Temkin), which show R < 0.90, are given in Figure 5B.
The best fit was found for the BET model (R 2 = 0.966). This indicates that the interaction of CF with (HA/PLL) 24 multilayers is likely according to assumptions on which the BET model is based. Considering the five general types of BET adsorption isotherms that are known (Table S2, SI), CF loading into the multilayers can be described in the best way by the S-shaped type V adsorption isotherm. The S-type of isotherm suggests cooperative adsorption, when the binding of adsorbate molecules to the adsorbent depends on the concentration of adsorbed molecules; in other words, binding of two adsorbate molecules to the adsorbent is not independent. 67 The binding of one molecule increases the affinity for the binding of the next molecule. 67 At the same time, the type V BET isotherm refers to the polyadsorption of CF molecules to equivalent binding cites. This means that each CF molecule adsorbed to a PLL backbone provides a new site for the adsorption of the molecule in the layer above it, and the anchorage to two types of binding cites (NH 3 + group or CF molecule) is characterized by the same or close adsorption energies; i.e., binding sites within the film are equivalent. The proposed integrated mechanism of adsorption is schematically represented in Figure 6. This proves that the most suitable model to describe binding of CF to HA/PLL multilayers is model C proposed in Figure 1.
According to the schematics in Figure 6, first molecules of CF loaded into the multilayers bind to amino groups of PLL and do not interact with each other (Figure 6, case 1). This type of binding is driven through the PLL−CF interaction and is apparently predominant at a low concentration of CF in solution (zone a). As CF molecules keep entering the film (zone b), apart from binding to PLL, they interact with each other. CF molecules that are anchored to the film either strengthen the binding of upcoming CF molecules via a CF− CF interaction (cooperation, case 2) or act as new binding cites for CF molecules that are coming from the solution (case 3). At the same time, there is no clustering of CF molecules to each other (aggregation), as proven earlier (Figure 2), the interaction is presumably between two CF molecules. At high concentrations of CF (zone c), all possible binding sites will be occupied, resulting in saturation of the multilayers with CF molecules and reaching a plateau on the adsorption isotherm. Figure 6 illustrates two extreme cases, where either only the cooperative mechanism (case 2) or only polyadsorption (case 3) is realized. However, we suppose that both cases take place simultaneously. Both ion pairs and intermolecular CF−CF interactions play a role; however, not all the amino groups can be bound to CF, due to either a sterical factor or the reduced number of charged amino groups due to the significantly lower apparent pK a of PLL in the multilayers compared to that in solution.

Research Article
Interestingly, the number of occupied binding sites is less than the total number of free amino groups of PLL in the multilayers. One can notice that the molar charge ratio PLL:CF is about 20 [CF maximum concentration in the multilayers is 13 mM ( Figure 5) and the total concentration of amino groups of PLL is about 250 mM, as we calculated above]. This can be explained either by sterical factors or the reduced number of charged amino groups available for interaction with CF in the multilayers. A sterical factor may mean that the CF−PLL interaction can only be realized if CF can find a PLL amino group properly oriented in the space to make an ion pair with it ( Figure 6B). One cannot expect significant sterical limitations for CF molecule to reach a backbone of PLL because the HA/ PLL multilayers are highly hydrated (about 80% of water) and about half (46%) of the amino groups of PLL are unpaired with HA (free to interact with CF). 62 In addition, CF is a small, compact molecule, allowing free diffusion in the hydrated and bulky multilayers. However, the PLL is a highly flexible polymer, and it can intertwine, making the access of CF to the amino groups of PLL difficult.
A limited number of charged amino groups can be another reason to be considered. The reduced pK a of PLL in HA/PLL multilayers has been reported. 68 Both HA and PLL are becoming weaker acid and weaker base in the multilayers compared to their strength in solution. The changes in the pK a are significant and may be a few pH units. The authors 68 have reported that in the multilayers assembled at pH 7.0, HA changes its pK a from 3.8 to 4.9 and the pK a of PLL is significantly decreased from 9.4 to 6.8. This is related to the ability of the polymers to change their secondary structure and increase a degree of conformational order upon adsorption to the multilayers. At pH 7.0 in the multilayers HA remains still fully negatively charged (pK a much below 7.0) but the charge density of PLL can be significantly reduced due to its much lower pK a value in multilayers, i.e., 6.8. This may explain why CF does not interact with all the available amino groups of PLL in the multilayer but only with a rather small number of the groups that still carry a charge. One can speculate that both the sterical factor and reduction of the apparent pK a of PLL can play a role.
We believe that the physical−chemical approaches employed here to understand CF interaction with HA/PLL multilayers may be helpful to assess the interaction of real drugs with multilayer films and thus open new perspectives to make advanced biocoatings with tuned loading and release of drugs. One would expect that some real drugs, such as the anticancer drug doxorubicin, may possess stronger intermolecular interaction between drug molecules due to the hydrophobic nature of the drug and lack of charge. The same approach used here may be used for doxorubicin loaded into the multilayers. The aspects of drug or biomolecule (e.g., ATP, doxorubicin) release from HA/PLL multilayers will be referred to in our upcoming research.
Understanding the drug−multilayer interaction is a key for the design of new multilayer-based drug delivery systems, because this interaction will drive drug release performance. The proper choice of the multilayer constituents (two or more polymers such as blend multilayers) will dictate the release rate, allowing one to get the desired release profile, e.g., burst or prolonged release or a more complicated profile. In addition, knowledge of the molecular binding to the multilayers may help one to engineer free-standing multilayer shells enabling protective characteristics 69 through controlled shell perme-ability governed by the interaction of molecules from outside with the capsule shell. We believe that the results of this study performed at a salt concentration below the physiological salt concentration will be helpful for the design of novel multilayerbased drug delivery systems, because the salt concentration in the multilayers shall be defined by uncompensated charges in the polymer network of the multilayers if enough salt (to form counterions) is provided, as in this study. However, interesting aspects related to the salt concentration inside and outside the multilayers and the effect salt can have on drug loading/release will be considered separately in our future research.

■ CONCLUSIONS
In this study we focused on the binding mechanism of the small, charged dye CF to biopolymer-based multilayers, namely, (HA/PLL) 24 multilayers. CF is a popular fluorescent dye often used as a model drug. CF can spontaneously be loaded into the multilayers with a very high partitioning coefficient of about 10 4 , reaching a dye concentration in the film of up to 13 mM, which is much more than the dye solubility in solution (0.5 mM). The CF loading is driven by binding of CF molecules to free amino groups of PLL in the multilayers. In the multilayers CF−CF interactions take place but do not dominate over CF− PLL interactions, giving no aggregates of CF, as found by analysis of CF fluorescence spectra in the multilayers and in solution. The binding of CF to multilayers is cooperative, as indicated by the S-shaped adsorption isotherm. This means that binding of one CF molecule to the multilayer affects the binding of another molecule, making the binding of new coming molecules more favorable. At the same time, anchorage of CF takes place in accordance with the BET type V polyadsorption model, meaning that CF molecules that are stored in the HA/PLL film play a role as new binding sites for new CF molecules. Both mechanisms of adsorption are driven by CF−CF interactions that enhance the binding affinity and strengthen the CF−multilayer interaction. Analysis of CF mobility in the multilayers by FRAP has revealed that more than 90% of CF molecules diffuse very fast (diffusion coefficient D of about 1 μm 2 /s) and the rest is immobile. As reported in the literature, mobile PLL has a similar D in the multilayers but a much higher content of immobile fraction (30−40%). 65 On the basis of this, one can conclude that CF diffuses in the multilayers through both (i) jumping from one amino group to another (at one or different PLL chains) and (ii) diffusion together with PLL chains being bound to the chains. This study provides new insights into the multilayer−drug interactions using the model fluorescent dye CF. We believe that the developed approach and knowledge received can be further extended and utilized for understanding the mechanism of loading and release of biologically relevant drugs. Elucidation of the storage mechanisms would open a new way that would provide one with knowledge to precisely design tailor-made multilayers well-suited for applications in drug protection, separation, and controlled release.