Molecular Dynamics Simulations of Pregelification Mixtures for the

ARTICLE pubs.acs.org/Langmuir

Molecular Dynamics Simulations of Pregelification Mixtures for the Production of Imprinted Xerogels Manuel Azenha,*,† Borys Szefczyk,‡,§ Dianne Loureiro,† Porkodi Kathirvel,† M. Nat
alia D. S. Cordeiro,‡ and Ant
onio Fernando-Silva† †

CIQ-UP/‡REQUIMTE-ICETA, Department of Chemistry and Biochemistry, Faculty of Sciences, University of Porto, Rua do Campo Alegre 687, 4169-007, Porto, Portugal § Institute of Physical and Theoretical Chemistry, Wroclaw University of Technology, Wybrzeze Wyspianskiego 27, 50-370 Wroclaw, Poland

bS Supporting Information ABSTRACT: A series of molecular dynamics (MD) simulations of different pregelification mixtures representing intermediate stages of the sol
gel process were set up to gain insight into the molecular imprinting process in xerogels, namely, to assess the template
gel affinity and template self-aggregation. The physical plausibility of the parametrization was checked, confirming the reliability of the simulations. The simulated mixtures differed in the water/methanol ratio (1:3, 5:3, and 5:1) and in the absence/presence of an organic functional group (phenylaminopropyl
) in the silicate species. The simulation results, expressed mainly by the radial distribution functions and respective coordination numbers, showed that the affinity of the template molecule, damascenone (a hydrophobic species), for the gel backbone would not be attained without the tested functional group, phenylaminopropyl
. The affinity, related to the capability to trap the template within the gel network, was derived mostly from the hydrophobic interaction. It was also inferred from MD simulations that lower water contents (methanolricher mixtures) would facilitate a better dispersion of both the functional group and the template within the final gel, therefore favoring the imprinting process. From the experimental counterparts of the simulated mixtures, a series of imprinted and nonimprinted xerogels were obtained. There was only one xerogel exhibiting the imprinting effect, namely, the one containing the organic group obtained at the lower water/methanol ratio (1:3), in agreement with predictions from the MD simulations. Such congruence demonstrates the ability of MD simulations to provide information regarding the fine aspects of molecular interactions in pregelification mixtures for imprinting.

1. INTRODUCTION Analytical chemistry requires the development of materials that are able to differentiate among a diversity of different molecules on the basis of size, charge, shape, and/or chemical affinity.1 One promising method of fabricating materials with improved molecular specificity involves molecular imprinting or templating.2
4 In this approach, a polymer network is assembled around a suitable template molecule or structure. Upon removal of the template, microcavities with a specific size, shape, and/or organic group remain in the cross-linked host. Commonly, simple commodity organic monomers, such as methacrylic acid, are radically copolymerized with a cross-linking monomer (e.g., ethyleneglycol dimethacrylate) to form good binding sites for a large variety of template structures containing hydrogen bond- or proton-accepting organic groups.5,6 In the past few years, sol
gel polymerization has been increasingly employed with success as r 2011 American Chemical Society

an alternative approach to the preparation of molecularly imprinted polymers7 with relatively hydrophilic templates. Sol
gel technology in fact provides advantages for the preparation of selective materials. It constitutes a relatively straightforward means of preparing inorganic or organic
inorganic hybrid glasses through the hydrolysis and condensation of suitable metal alkoxides.1 The most widely used precursors for preparing sol
gel materials for use in chemical analysis applications have been the silicon alkoxides, particularly tetramethoxysilane (TMOS) or tetraethoxysilane (TEOS). These reagents can be readily hydrolyzed and condensed under relatively mild aqueous conditions, Received: January 4, 2011 Revised: February 28, 2011 Published: March 18, 2011 5062

dx.doi.org/10.1021/la200032e | Langmuir 2011, 27, 5062–5070

Langmuir enabling the introduction of water-soluble molecules within a highly cross-linked porous host without problems associated with thermal or chemical decomposition. Additional organic groups (introduced via the so-called functional monomers) can be combined with the inorganic precursor into the framework, aiming at the improvement of imprinting features. However, organic group-template interactions relying on hydrogen bonding or proton transfer in aqueous media tend to be greatly weakened, contrary to what happens in the imprinting of radically polymerized acrylics that takes place in nonpolar or weakly polar solvents. The sol
gel imprinting of hydrophobic compounds constitutes a challenge because of the low water solubility. A hydrophobic molecule dissolved in an essentially polar environment such as a silicate network/water gel is expected to exhibit an affinity for hydrophobic organic groups introduced into the gel network and be well dispersed within the network. However, this may not be the case, and phase segregation or microscopic template aggregation may occur.8 As recently reviewed for radically polymerized imprints by Nicholls et al.,9 molecular dynamics (MD) simulations with an explicit representation of all imprinting-mixture constituents may provide important information related to multiple kinds of interactions. For example, O’Mahony et al.10 found that the functional monomer was capable of effectively competing with self-association and the cross-linker in complexing the template molecules. However, the prevalence of cross-linker
template interactions and the importance observed for interactions other than functional monomer
template interactions made these authors conclude that further studies should be carried out because they may provide for some insight into important aspects such as binding-site homogeneity, the role of the crosslinker, or the identification of instances where a relatively weak prepolymerization complex results in a successful imprint. If such insight is considered to be valuable in the relatively mature field of organic polymer imprinting, we judge that in the case of sol
gel imprinting it will be even more valuable because this is still in an incipient stage. Clues to which aspects of pregelification mixtures, other than the more obvious functional monomer
template interaction, can induce selectivity in the resulting material would then be of great utility. Thus, we decided to undertake what constitutes, to the best of our knowledge, the first study on the application of MD simulations to pregelification sol
gel mixtures in the context of imprinting. In agreement with what is recognized for the process of radically polymerized imprints, currently the complex mechanisms involved in the sol
gel processes cannot be modeled by MD, so ideal solutions representing stages of the sol
gel process have to be considered. Important pioneering MD simulations of such ideal solutions have been carried out by Price et al.,11,12 who developed a consistent force field (CFF) applicable to the common sol
gel reagents (TMOS, TEOS, water, methanol, and ethanol) and some of the initial sol
gel products, Si(OH)4, Si2O(OH)6, and Si3O3(OH)6. Their results have shown the trend for silica clusters to aggregate, with the trend becoming more pronounced between the larger clusters. This means that the intermolecular interactions leading to the condensation reactions were systematically, for long run times, stronger than the interactions leading to the reverse reactions. These were essential results in constructing a theoretical model able to perform realistic simulations of silica-based sol
gel processes. On the basis of many of the parameters from Price’s work, we set up a number of simulations of sol
gel solutions

ARTICLE

Table 1. Composition of All Six Systems under Study components (number of molecules) model water/methanol ratio SI3

SIPA

SI3

SIPA

DAM

water

methanol

1:3

136

12

396

1200

5:3

136

12

1272

764

5:1

136

12

2000

400

1:3

108

40

12

400

1200

5:3

108

40

12

1272

764

5:1

108

40

12

2000

400

with compositions based on previous experimental work13 related to the sol
gel imprints of damascenone. From the available silica clusters, we chose Si3O3(OH)6 (SI3 from now on) as the one providing the best representation of a stage between initial mixing and gelification. Because an organotrimethoxysilane [(C6H5NHC3H6
Si(OCH3)3, (3-propylaminophenyl)-trimethoxysilane, PAPTMOS)] was also employed as a functional monomer, in the corresponding simulations a new Si cluster (trimer), Si3O3(OH)5C3H6NHC6H5, was included (SIPA from now on). Both of these small condensation products were regarded as providing the simulation with a certain resemblance to the final xerogel backbone (same surface groups). The simulations proved to be a useful tool for studying the solvation environment of the template and the silica clusters by direct observation of the atomic arrangements around them. In particular, the information collected about the competing interactions that take place in different water-to-methanol ratios either in the absence or presence of the functional group is in excellent agreement with the experimental observations of the performance of different prepared xerogel imprints of damascenone.

2. MATERIALS AND METHODS 2.1. Computational Details. The molecular dynamics simulations were performed with the GROMACS 4.014 package. It is an open source package enjoying a good reputation concerning speed and reliability and has been used by the authors to study other systems.15 However, it does not provide a straightforward CFF implementation. Instead, we used OPLS-AA16 implementation. Most of the force field parameters related to the silicon species, namely, cyclic trimers Si3O3(OH)6 and Si3O3(OH)5CH2CH2CH2NHC6H5 have no available OPLS-AA implementation, so angle and dihedral torsion potentials were taken from Price’s force field11,12 and converted to an OPLS-AAcompatible form in the same manner as was done by other authors17,18 (details in the Supporting Information). Charges on the silica trimers, PAPTMOS, and damascenone were calculated in an OPLS-AA-compliant manner: the geometry was first optimized at the HF/6-31G(d) level, and then the charges were computed from a single-point run using the CHelpG scheme at the MP2/aug-cc-pVTZ(-f) level of theory (results available in the Supporting Information; the General Atomic and Molecular Electronic Structure System (GAMESS) software package19 was used for this purpose). The parameters included in the OPLS-AA implementation of GROMACS were used to describe van der Waals interactions. Two kinds of models were studied: those containing only SI3 units (further referred to as SI3 models) and those containing a mixture of SI3 and SIPA units (further referred to as SIPA models). Both SI3 and SIPA models also contain damascenone (DAM) molecules and methanol and water in different ratios. Three different water/methanol molar ratios were studied: 1:3, 5:3, and 5:1 (Table 1). The composition of the 5063

dx.doi.org/10.1021/la200032e |Langmuir 2011, 27, 5062–5070

Langmuir

ARTICLE

simulated mixtures was based on the following assumed stoichiometries for the formation of silica trimers: SI3 model: H2 O=MeOH

3SiðOCH3 Þ4 sf Si3 O3 ðOHÞ6

permittivity of εr = 1 was used. Statistical and trajectory analyses were performed with the programs included in the GROMACS package, and visualizations of the simulations were made with VMD.26 The analysis consisted essentially of the comparison of the radial distribution functions (g(r) or RDF) between different types of molecules, with g(r) being defined as

SIPA model: gAB ðrÞ ¼

H2 O=MeOH

101SiðOCH3 Þ4 þ 10SiðOCH3 Þ3
R sf 27Si3 O3 ðOHÞ6 þ 10Si3 O3 ðOHÞ5
RðR ¼ phenylaminopropyl
Þ The assumed stoichiometry for the SIPA model was based on an initial TMOS/PAPTMOS molar ratio of 10, which was used experimentally. The simulated mixtures then corresponded to the composition obtained after an idealized process of complete precursor hydrolysis and trimerization yielding all silica in the SI3 and SIPA forms. Although obviously unrealistic (the condensed silica content at a certain moment of the reaction is in reality distributed between many different species, including linear, branched, and cyclic), this mixture was justified by the fact that we were interested in stages as close as possible to gelification when the imprints are actually formed. Also, theoretical work indicated that this is the prevalent intermediate.20 Computational power and time effectiveness limitations prevented the consideration of many other condensed forms, especially larger aggregates that would certainly be better representatives of a nearly gelified system. The SI3 and SIPA models pose the advantages that there are previous molecular dynamics studies (for SI312) that allow for comparison while containing the basic units of the final gel backbone, namely, the
O
Si
O
,
Si
OH, and, in the case of the SIPA model, the
Si
R chemical environments. It is thus possible with such mixtures to formulate a good picture of the preferred interactions between the template and the other chemical units, for example. Silicates in the ionic form were not considered in these simulations. In fact, the experiments were carried out at pH below silica’s isoelectric point (∼2), hence the abundance of the ionic forms is negligible. That is also the most common pH range used in this kind of mixture. The initial boxes were constructed by inserting the respective number of molecules at random positions. Each box was initially optimized using the steepest-descent method and then equilibrated for 1 ns in the NVT ensemble. After that, the production run was performed in the NPT ensemble. The total simulation time varied, depending on the system. Because of the strong interaction and low mobility of the SI3 and SIPA species (diffusion coefficients in Table 3) and the relatively small number of DAM molecules, very long simulations were needed to reach equilibrium, ranging from 70 to 200 ns. As we found, the most critical and stringent factor controlling whether equilibrium has been reached is the intermolecular interaction between different groups in the system (e.g., between SI3 and DAM); therefore, the simulations were performed until all intermolecular interactions had reached stable values. Additionally, to enhance the sampling and improve the statistics, each of the six model systems presented in the Table 1 was simulated twice, starting from different random initial distribution of molecules (these parallel simulations are called throughout the paper run 1 and run 2). The time step used in the simulations was 2 fs. Constant-temperature conditions were ensured by applying the velocity rescaling thermostat,21 with the reference temperature set to 298 K; constant-pressure conditions were maintained using a Nose-Hoover barostat,22,23 with the reference pressure set to 1 bar. For the water molecules, the transferable intermolecular potential (four-point) model, TIP4P,24 was applied. The nonbonded electrostatic interactions were calculated using a sixth-order particle mesh Ewald (PME) method25 beyond the cutoff radius of 0.9 nm. The Lennard-Jones interactions were calculated using a pair list within a cutoff of 1.0 nm and updated every 10 steps. A dielectric

ÆFBðrÞæ ÆFBæloc

where ÆFB(r)æ refers to the average density of particle B at a distance of r, around particle A, and ÆrBæloc refers to the density of particle B averaged over the whole sphere of radius rmax, centered on particle A. The g(r) function is additionally averaged over all spheres centered on all particles of type A present in the system and also averaged over the trajectory (simulation time). Unless otherwise stated explicitly in the Figure captions, RDFs were calculated by taking the centers of mass of the whole molecules. In selected instances, the coordination numbers NB around centers A were calculated by integrating the RDF gAB(r) Z rm NB ¼ 4πFB gAB ðrÞr 2 dr 0

between center A, limited to the distance to the first local minimum rm. rB refers to the density of species B (expressed in molecules per volume unit). The mobility of each species is assessed on the basis of the diffusion coefficient D, calculated from the Einstein relation (root-meansquare displacement, rmsd): D¼

1 Æj r ðtÞ
B r i ð0Þj2 æ 6t Bi

The rmsd is averaged over molecules, and in order to improve the statistics, several restarts r(0) are used along the trajectory. 2.2. Synthesis and Evaluation of Xerogels. Two sets of molecularly imprinted xerogels (MIX) were prepared. The set called SI3-MIX corresponded to the experimental counterpart of simulation model SI3 (no functional monomer, three different water/methanol molar ratios). For that purpose, 10 mmol of TMOS (Aldrich, Steinheim, Germany), methanol (Merck, Darmstad, Germany), water (Milli-Q quality, Millipore, Italy), 102 μL of catalyst TFA (Aldrich, Steinheim, Germany), and 60 μL (0.3 mmol) of the template β-damascenone (a kind offer from Firmenich, Geneve, Switzerland) were added to 20 mL glass vials. Three different MIXs were prepared in set 1 by varying the amounts of water and methanol to obtain approximate molar ratios of 1:3 (10 mmol water, 30 mmol MeOH), 5:3 (32 mmol water, 19 mmol MeOH), and 5:1 (50 mmol water, 10 mmol MeOH) employed in the simulations. The three corresponding nonimprinted xerogels (NIXs) were also prepared for reference from similar starting mixtures with the omission of the template. The set called SIPA-MIX corresponded to the experimental counterpart of the simulation model SIPA (PAPTMOS as the functional monomer, three different water/methanol molar ratios) and was obtained in a similar manner as SI3-MIX but with the addition of 1 mmol of PAPTMOS (Fluka, Buchs, Switzerland). All of the described mixtures were stirred for 6 h and after this period were transferred to closed 12-cm-diameter flat plastic containers whose covers were punctured with a few small holes so as to allow slow solvent evaporation, leading to the formation of gels after 2 to 3 days. The xerogels were crushed in a mortar and sieved to promote the selection of particles with sizes in the range of 25
45 μm. Target molecules present in MIPs were removed by Soxhlet extraction with successive volumes of methanol until the extract monitoring by HPLC showed undetectable levels of β-damascenone. Finally, the gels were dried at 60 C for 4 h and stored at room temperature. A prompt evaluation of performance was carried out by packing 500 mg of the prepared MIXs and NIXs into cartridges of 5 mL capacity that are used as SPE sorbents for the target molecule β-damascenone. The 5064

dx.doi.org/10.1021/la200032e |Langmuir 2011, 27, 5062–5070

Langmuir

ARTICLE

Table 2. Calculated and Experimental Diffusion Coefficients and Densities at 20 C and 1 atm calculated D/10
5 cm2 s
1 methanol

experimental and literature D/10
5 cm2 s
1, 2.10a

2.8

calculated density/g cm
3

experimental density/g cm
3

0.77

0.78

3.51
9.03b

a

1.73
5.54b

TMOS

0.96

1.09

1.02

PAPTMOS

0.60

1.05

1.07

DAM

0.60

0.93

0.95

Experimental. b From ref 12.

Table 3. Calculated Diffusion Coefficients in Models SI3 and SIPAa D/10
5 cm2 s
1 model SI3

SIPA

a

water/methanol ratio

SI3

1:3 5:3

SIPA

DAM

water

methanol

0.012

0.31

0.84

0.83

0.005

0.18

0.91

0.79

5:1

0.004

0.12

1.33

0.92

1:3 5:3

0.009 0.006

0.005 0.005

0.11 0.077

0.76 0.91

0.74 0.71

5:1

0.005

0.004

0.054

1.24

0.77

Mean values from runs 1 and 2.

SPE procedure was as follows: sorbents were conditioned with 20 mL of acetonitrile (ACN), and then 1 mL of 2.5  10
5 mol of β-damascenone in 60% ACN was eluted very slowly, followed by 1 mL of water for washing. Both the eluate and washing water were collected for the HPLC analysis of the nonretained fraction. The sorption percentage was calculated as [load
(eluate þ cleaning with water)]/load. To allow the reuse of the sorbents, these were then washed with pure ACN for cartridge cleaning. A Perkin-Elmer high-performance liquid chromatographic system (Perkin-Elmer 200 Series) consisting of a pump, ultraviolet detector, and port injection valve provided with a 20 μL injection loop was used for this study. All separations were achieved on an analytical reversed-phase Nucleosil ODS C18 column (25  0.46 cm2) of 5 μm particle size at a mobile flow rate of 1.8 mL min
1. The mobile phase consisted of ACN and water (60:40 v/v). The UV detector was operated at 226 nm. TotalChrom Navigator software was used to acquire and process chromatographic data.

3. RESULTS AND DISCUSSION 3.1. Foreword. The simulations comprised a few molecules that had never before been modeled using the OPLS-AA force field: SI3, SIPA, and damascenone. It was necessary to check somehow for the physical plausibility of their parametrization by running single-component simulations and calculating some properties known experimentally or computed by other authors. SI3 and SIPA are sol
gel intermediate species that have not been isolated in purified form and therefore no physical
chemical properties are available. PAPTMOS was included in this preliminary study because it was used as a model of the organic branch existing in SIPA with respect to the charges, with bonded and nonbonded parameters being transferred from the PAPTMOS model to the SIPA model. TMOS was also included as a probe for the force field parameters developed by Price and coworkers.11,12 The obtained results are presented in Table 2. Density was chosen as one of the test properties because of its importance and extensive data availability. It is also a good indicator of how well the nonbonding interactions are reproduced. In fact, for PAPTMOS and damascenone, density was the

Figure 1. Snapshot of the simulation of a SIPA model (water/methanol 1:3, run 1) showing the DAM molecules (van der Waals representation) to be well dispersed within the box and in close proximity to the SIPA rings. SI3 and SIPA displayed in green and gold, respectively. Water and methanol molecules were omitted from the representation.

only useful property found in the literature. In general, the agreement between calculated and experimental values was good (maximum relative error of 6.8% for TMOS). These are acceptable results for a general-purpose force field using the same potentials for different species. The diffusion coefficient (D) is a dynamic property of liquids that is of great importance in the present work because of expected large differences between the dynamic properties of the smaller molecules and those of the larger silica rings. The calculation of D is not straightforward because it is very sensitive to modeling conditions such as the force field, time step, and NPT algorithm.11 The calculated D for methanol in the present study (2.8  10
5 cm2 s
1) is satisfactorily comparable to the experimental value (2.1  10
5 cm2 s
1). On the contrary, the authors of ref 12, in applying the CFF, calculated much higher values than the experimental values for methanol (included in Table 2), ethanol, and water. Considering that our values of D are systematically improved for those compounds for which the corresponding experimental values exist, we assume that the parametrization procedure is correct and therefore the presented results are trustworthy. 3.2. Precolloidal Character of the Simulated Mixtures. The analysis of trajectories of the simulations with both SI3 and SIPA models has shown the clear precolloidal character of the mixtures, expressed by the observation of considerable aggregates of silica trimers (Figure 1). Such aggregates exhibited greatly diminished mobility (D = (0.004
0.012)  10
5 cm2 s
1) 5065

dx.doi.org/10.1021/la200032e |Langmuir 2011, 27, 5062–5070

Langmuir

ARTICLE

Figure 2. Radial distribution functions calculated for the silica rings, g(SIPA
SIPA), g(SI3
SI3), and g(SI3-SIPA) from the simulations of the SIPA model. Results are from run 1 and are representative of both runs.

during the simulation (Table 3) as compared to the molecules present in what we might call the continuous aqueous/methanolic phase (D = (0.71
1.33)  10
5 cm2 s
1). This fact had the effect that long equilibration times and analysis windows had to be allowed so as to achieve statistical significance for the cluster dynamics and interactions. The pronounced trend of silica trimer aggregation was expected on the basis of the experimental knowledge of the sol
gel process and had also been observed in previous molecular dynamics simulations (for SI3 trimers only12). The observed precolloidal character was regarded as a favorable circumstance in view of our interest in simulating a scenario as close to gelification as possible. The RDFs in Figure 2A,B reflect the trend in SI3 and SIPA trimers to self-aggregate, respectively, and Figure 2C shows that mixed SI3/SIPA aggregates are comparatively less favored (also observable from the snapshot in Figure 1). These results show that the formation of a gel with nonhomogeneous microstructure bearing hydrophobic pockets is likely to occur. Such a possibility of subclustering derived from the utilization of organofunctionalized silanes is supported by experimental observations by Guardia and co-workers,27 who were able to show electron microscope images of a crushed gel exhibiting particles of two well-defined morphologies when a functionalized monomer/ TMOS molar ratio of 11.5 was applied. The RDFs show that SI3
SI3 and SI3
SIPA aggregations are not significantly affected by the water/methanol relative abundance whereas the

SIPA
SIPA aggregation appears to become unfavored in the richest methanol mixture, in agreement with the better solvation ability of methanol toward the hydrophobic group. 3.3. Interaction of the Template with the Silica Aggregates and Other Components in the Mixture. In both the SI3 and SIPA models (both runs), the template (DAM) exhibited no appreciable affinity for the SI3 clusters as shown in Figure 3A. This means that in the case of the SI3 model the template molecules were mostly located away from the silica network (not interacting with it), in the continuous medium, because of their good affinity for methanol (as shown by the corresponding RDF, Figure 4A) and despite the hydrophobicity of DAM (confirmed by the DAM
water RDF, Figure 4B). It is reasonable to admit that under such circumstances, after gelation takes place, especially when syneresis (additional cross-linking/shrinkage of the network) provokes the expulsion of liquid from the pores, a large fraction of the template molecules will likely be found in the expelled liquid or at the surface of the gel after drying. On the contrary, the DAM-SIPA RDFs (Figure 3B) show, to a higher or lower extent depending on the water/methanol ratio, the affinity between the template and the organic branch of the introduced functional monomer. A comparison of the RDFs of DAM-SI3 and DAM-SIPA pairs indicates that binding to the phenylaminopropyl-rich fragments of the growing gel matrix would be preferred. By looking at the RDF obtained when considering only the aromatic part of SIPA (SIPA-ar, used as a 5066

dx.doi.org/10.1021/la200032e |Langmuir 2011, 27, 5062–5070

Langmuir

ARTICLE

Figure 3. Radial distribution functions calculated between DAM and the silica rings (or subgroups), g(SI3-DAM), g(DAM-SIPA), g(DAM-SIPA-ar), and g(DAM(O)-SIPA(N), from the simulations of the SIPA model. SIPA-ar represents the center of mass of the aromatic ring. DAM(O) and SIPA(N) represent damascenone’s oxygen and SIPA’s nitrogen atoms, respectively. The results are from run 1 and are representative of both runs.

Figure 4. Radial distribution functions calculated between DAM and methanol, g(DAM-MeOH), or DAM and water, g(DAM-H2O), from the simulations of the SIPA model. Results are from run 1 and are representative of both runs with both SI3 and SIPA models. 5067

dx.doi.org/10.1021/la200032e |Langmuir 2011, 27, 5062–5070

Langmuir

ARTICLE

Table 4. Coordination Numbers (N) Calculated for DAMSIPA and DAM-DAM Pairs by Integrating the Corresponding RDFs in Figures 3B and 5, Respectivelya coordination number (runs 1 and 2) water/methanol ratio

DAM-SIPA

DAM-DAM

SI3 Model 1:3

0.46, 0.43

5:3

0.85, 0.70

5:1

1.37, 2.74 SIPA Model

1:3 5:3

0.37, 0.40 2.90, 3.27

0.53, 0.50 1.17, 2.01

5:1

3.41, 4.04

0.95, 1.76

a

For each model, two numbers obtained from two independent runs (1 and 2) are given.

hydrophobic interaction indicator, Figure 3C) and the RDF obtained between the oxygen of damascenone and the nitrogen atom of SIPA (used as an indicator of the
CdO 3 3 3 H
N
hydrogen bond, Figure 3D), it appears from the peak heights for the first coordination sphere that the hydrophobic interaction is largely dominant (as expectable for polar media), although the sharp peak at 0.3 nm in Figure 3D clearly shows that the formation of the hydrogen bond also played a role in the DAM-SIPA interaction. The water/methanol ratio had a significant influence on the DAM-SIPA interactions. Most noticeably, from the analysis of RDFs, it can qualitatively be observed that the interactions became stronger as the continuous medium became richer in water. A more meaningful (i.e., quantitative) assessment of the influence of the water/methanol ratio may be performed by the analysis of the coordination numbers (N) of DAM-SIPA and DAM-DAM pairs (Table 4). For the 1:3 ratio of

water/methanol, isolated DAM molecules and pairs of DAM molecules are observed. As the content of water increases, clusters are formed, and for the 5:1 ratio, the typical cluster of DAM has three to four molecules in the SI3 model and two to three molecules in the SIPA model with a hydrophobic functional group. Although the DAM-SIPA N increases with the content of water, other factors such as the clustering of DAM molecules may prevail and the higher water/methanol ratio might not be necessarily profitable. Both qualitative and quantitative analysis comply with the strengthening of the hydrophobic interactions and at the same time with the decreased solvation capability of the continuous medium as a result of the diminished amount of methanol. Accordingly, in this case, contrary to the SI3 system, one may expect a major fraction of the template molecules to become effectively trapped within the aged gel, especially if the chemical environment of the gel is richer in water. 3.4. Template Self-Aggregation. The RDFs obtained for DAM against other DAM molecules may provide us with the trend in self-aggregation and are shown in Figure 5, whereas the respective N values (in Table 4) provide a more quantitative measurement of the same phenomenon. Contrary to all other RDFs, in this case run 1 and run 2 curves are both shown because the results of the two runs, in most situations, were not in agreement. Not surprisingly, the same occurs with N, with as much as 100% between-run disparity being observed. This is probably caused by the relatively small number of DAM molecules included in the simulations (unfavorable to good statistics), not by a convergence issue (details available in Supporting Information). Additionally, the between-run variation in template dimer/trimer formation, found to be stable during considerable fractions (up to 30%) of production time, also contributed to the RDF and N differences. Nevertheless, important qualitative and semiquantitative information may be collected from the analysis of the RDFs and N, respectively. In the SI3 model, a continuous rise in N with increasing water/methanol

Figure 5. Radial distribution functions calculated between DAM molecules, g(DAM-DAM), from simulations of the (A) SI3 and (B) SIPA models and (C) a snapshot taken from run 1 of the SIPA model at the 5:1 water/methanol ratio, showing a hydrophobic pocket with three DAM molecules (van der Waals representation) trapped. 5068

dx.doi.org/10.1021/la200032e |Langmuir 2011, 27, 5062–5070

Langmuir

Figure 6. Retention efficiency of DAM (in 60% ACN) observed with SPE columns filled with the different MIXs and NIXs prepared.

ratio is clearly observable, meaning that self-aggregation is favored by a medium richer in water, which is explained by the decreased methanol abundance and the increased hydrophobic effect. For the SIPA model, because the N values for the 5:3 and 5:1 ratios are undistinguishable but clearly higher than the N values for the 1:3 ratio, it can be stated only that for the methanol-rich medium the lowest aggregation trend was observed, in congruence with the good solvation capability of methanol and the minimized hydrophobic effect. A careful observation of the trajectories shows that the disparity observed between runs with 5:3 and 5:1 water/ methanol ratios results from different episodes of DAM dimer/ trimer formation within hydrophobic pockets formed from SIPA aggregates, as illustrated in Figure 5C. 3.5. Consolidated Analysis of the Simulations. Here we focus and summarize the molecular-scale aspects of the simulations that may likely reflect the real imprinted xerogel features. First, the presence of the phenylaminopropyl group was critical to observing the template/gel backbone affinity, and thus the simulations predict that a good imprint will not be likely found with TMOS-only mixtures, regardless of the water/methanol ratio. They also predict that a less heterogeneous hybrid organic
inorganic xerogel with a minor propensity to entrap template aggregates in hydrophobic pockets (i.e., better template dispersion within the gel) will be more easily obtained under methanol-rich conditions. This should then contribute to the formation of better imprints,8 if considering also that under methanol-rich conditions the DAM-SIPA N, ∼0.4, is still significant in terms of DAM affinity toward the SIPA group. 3.6. Xerogel Evaluation. By aiming at the assessment of the consistency of the predictions drawn in the previous section, we prepared the experimental counterparts of the simulated mixtures and obtained the corresponding xerogels as described in the Materials and Methods section. The different materials were then studied with respect to their role as SPE damascenone sorbents, with the imprinting effect being evaluated by the MIX/ NIX comparison in sorption efficiency. The results, depicted in Figure 6, show that in the case of the SI3-MIXs a moderate (∼30
60%) sorption efficiency was observed. However, this could not be ascribed to an imprinting effect because the respective control NIXs exhibited statistically identical efficiencies. Therefore, the observed sorption was mainly due to nonselective binding. In the case of SIPA-MIXs, the sorption efficiencies were considerably higher (∼60
90%) and an imprinting effect could be identified for one of the materials, the one obtained within the methanol-rich medium. These observations

ARTICLE

were exceedingly congruent with the expectations driven from the simulation results that a gel backbone bearing the hydrophobic functional group and a good dispersion of the template, favored in the methanol-richer medium, would seemingly constitute better conditions for successful imprinting. Such congruence reinforces our confidence in the potential of MD simulations as a means to provide important information regarding fine aspects of molecular interactions in pregelification mixtures for imprinting. The fact that the sorption was higher for the SIPA-MIXs, even in cases when it was nonselective, could probably be attributed to the higher affinity provided by the presence of the functional monomer that was randomly distributed or formed localized hydrophobic pockets. The fact that the nonselective binding decreased with the decrease in water content is most probably related to the different gel microstructures obtained at different water/silicon ratios. This is in fact compatible with BET measurements that gave surface areas of