Manipulating ligand–nanoparticle interactions and catalytic activity through organic-aqueous tunable solvent recovery

Abstract

Dispersed gold nanoparticles demonstrate much higher catalytic activities compared to their supported counterparts. Current methods for nanoparticle recovery are difficult to scale up, can alter the morphology of the nanoparticles, require large amounts of energy or solvents, and/or need highly specialized syntheses. Herein, we propose a facile system for the recovery of nanoparticles from reaction mixtures using Organic-Aqueous Tunable Solvents (OATS). OATS are homogeneous mixtures of an organic solvent and water which have the inherent property to phase separate into a heterogeneous mixture under moderate CO₂ pressure. A 60 vol% acetonitrile and 40 vol% water mixture was chosen to disperse gold nanoparticles for these experiments. Poly(vinylpyrrolidone) (PVP) was selected because it is water soluble and is believed to allow access to the entire metal surface for reactive compounds. We synthesized gold nanoparticles of average size 9.4 ± 1.4 nm and performed four different thermal treatments (no thermal treatment, 40, 50, 60 °C). The phase separation of the OATS mixture was determined to occur between 9.6 and 11.3 bar of absolute CO₂ pressure. Complete recovery of gold nanoparticles was achieved for all thermal treatments based on UV-vis absorbance of the localized surface plasmons collected from each phase. Catalytic activity of the nanoparticles was benchmarked using the hydrogenation of 4-nitrophenol. Reduction of catalytic activity occurred after the nanoparticles underwent thermal treatments or were subjected to pressure separation in OATS but still remained highly active. The decrease in catalytic activity can be attributed to additional PVP functional groups having passivated the surface of the nanoparticles and blocking active sites as the system minimized free energy indicating that PVP does not allow access to the entire nanoparticle surface. The additional surface-bound ligands also prevented precipitation and growth in OATS, but slowed dynamic surface restructuring, ultimately decreasing the apparent catalytic activity.

---

Introduction

The catalytic activity of dispersed nanoparticles has been shown to be over 400 times greater than supported catalysts.¹ However, recovery and purification of dispersed nanoparticles from reaction mixtures is difficult. Previous methods of nanoparticle recovery include magnetic separation,^2–4 temperature induced phase separation,^5,6 nanoparticle destabilization due to pH manipulation,⁷ precipitation using gas-expanded liquids,^8–10 reverse micelles,^11,12 and microfiltration.^13,14 Many of these recovery methods are difficult to scale up, require a large amount of energy, and/or can change the morphology of the nanoparticles. A widely applicable method for nanoparticle synthesis and separation that is economical and would maintain the catalytic activity of the nanoparticles is highly desirable.

A potential solution for recovery and reuse of nanoparticles is through the use of Organic-Aqueous Tunable Solvents (OATS).^15–19 OATS are single phase mixtures of water and a water-soluble organic selected specifically for its ability to dissolve large amounts of CO₂ (e.g., acetonitrile, tetrahydrofuran, or 1,4-dioxane). CO₂ is added via pressurization and dissolves primarily into the organic, inducing a phase separation between the water and CO₂-expanded organic. The single phase mixture would serve as the reaction media and the heterogeneous mixture can be used to selectively recover dispersed catalysts in one phase and the products in the other phase. Addition of CO₂ causes a shift in the critical solution temperature (e.g., adding CO₂ to a mixture of water and acetonitrile increases the upper critical solution temperature, UCST, while adding CO₂ to a mixture of tetrahydrofuran decreases the lower critical solution temperature, LCST).^20,21 Phase separation occurs between the water and CO₂-expanded organic solvent when the critical solution temperature traverses the ambient temperature. The more-dense, aqueous bottom phase can be decanted from the less-dense, CO₂-expanded organic. A benefit of this system is that CO₂ can be recovered and reused via simple depressurization, the aqueous phase containing the catalyst can be completely reused without further purification, and final purification of the product is easier from a volatile organic solvent. Previous work using OATS has focused on separating palladium-based,¹⁶ rhodium-based,^17–19 and enzymatic homogeneous catalysts¹⁷ after completing C–O coupling, hydroformylation, and hydrolysis reactions, respectively. Separation and recovery of nanoparticles have yet to be demonstrated using OATS due to the complex interactions between the solvent, the stabilizing ligand, and the nanoparticle surface. Previous work with OATS have relied on simply changing the solubilization between the two solvents to facilitate catalyst extraction. Here, we attempt to control the solvation of a polymeric stabilizing ligand in the organic-aqueous tunable solvents to manipulate the dispersion.

In this study, we demonstrate a facile method for separating poly(vinylpyrrolidone)-stabilized gold nanoparticles dispersed in a mixture of acetonitrile and water using OATS. We determined that the primary factor affecting the recovery and catalytic capability of the nanoparticles is the PVP ligand surface coverage. Studies of the adsorption of a small molecule (2-mercaptobenzimidazole) onto the nanoparticle surface and a solvatochromic characterization of OATS provide new insights into the interactions between PVP, the surface of gold nanoparticles, and this complex solvent mixture.

Experimental section

Materials

Poly(vinylpyrrolidone) (PVP, MW ∼ 29 000), hydrogen tetrachloroaurate trihydrate (HAuCl₄·3H₂O, 99.9%) and 2-mercaptobenzimidazole (2-MBI, 98%) were obtained from Sigma-Aldrich. Sodium borohydride (NaBH₄, 97+%) was obtained from Alfa Aesar. Acetonitrile (reagent grade, ≥99.9%) was obtained from JT Baker. 4-Nitrophenol (98%) was obtained from Acros. 9-Diethylamino-5-benzo[α]phenoxazinone (Nile Red, 100%) was obtained from Chem-Impex International Inc. Clinical laboratory reagent water (CLRW) – Type 1 was obtained from Thermo Scientific. All materials were used as received without any further processing.

Gold nanoparticle synthesis and characterization

Gold nanoparticles were synthesized based on a modified procedure.²² Briefly, an aqueous solution of PVP (100 mL, 7.66 mM macromolecule basis) was combined with an aqueous solution of HAuCl₄·3H₂O (3.50 mL, 62.3 mM) under vigorous magnetic stirring at room temperature for approximately 5 minutes. An aqueous solution of NaBH₄ (0.300 mL, 0.41 M) was added to the PVP–HAuCl₄ mixture as a reducing agent and to induce nanoparticle nucleation and growth. The reaction mixture was stirred for 10 hours. The PVP-stabilized nanoparticles dispersed in water were used as synthesized.

The nanoparticles were imaged using a FEI Technai G2 20 Twin transmission electron microscope at a voltage of 200 kV and their sizes were determined by counting more than 1000 nanoparticles using ImageJ software. TEM grids were prepared via drop casting onto a Ted Pella 300 mesh formvar/carbon copper grid. Mass concentrations of gold for all solutions were determined using a TA Instruments Q50 Thermogravimetric Analyzer. An aliquot of the nanoparticle dispersion (1 mL) was dried to remove excess water, the remaining material was crushed, and then placed in the platinum sample pan. In an air environment, the TGA temperature was ramped to 1000 °C to remove all surface bound ligand from the gold.

Nanoparticle recovery using OATS

A custom built stainless steel pressure vessel, shown in Fig. S1 of the ESI,† was used to measure the UV-vis absorbance spectrum of the nanoparticles solutions when pressurized with CO₂. The vessel is fitted with two sets of parallel quartz windows to allow light to pass through the mixture at different vertical positions. A pressure transducer and thermocouple were used to measure the pressure and temperature, respectively, inside the high pressure vessel. CO₂ was controllably delivered into the vessel and constant pressure maintained with a Teledyne Model 500D syringe pump.

For a typical experiment, a single phase solution of 6.00 mL of acetonitrile, 3.25 mL of water, and 0.75 mL of the nanoparticle dispersion was prepared to provide a solvent mixture of 60 vol% acetonitrile and 40 vol% water. This mixture was selected because this concentration would require the smallest change in the UCST as it is near the apex of the phase envelope.²³ The gold loading in the final mixture was typically 0.57 mg (0.057 mg mL⁻¹). The solution was then loaded into the pressure vessel and flushed with CO₂ to remove all air from the head space of the vessel. The intensity of the localized surface plasmon resonance (LSPR) was monitored to determine the amount of gold nanoparticles dispersed in each phase with a Cary 100 UV-vis spectrophotometer at pressures of CO₂ ranging from 0.9 to 18.2 bar absolute at equilibrium.^9,24,25 The LSPR for gold nanoparticles is approximately 530 nm (ref. 25) and the intensity is related to nanoparticle concentration via the Beer–Lambert law. The phase separation was reached after approximately 72 hours and the entire experiment required approximately 120 hours due to the time required to reach equilibrium at each intermediate pressure.

A second apparatus, shown in Fig. S2,† comprised of two high pressure vessels (Jerguson 17R20) was used to perform nanoparticle recoveries. The vessels were oriented vertically and separated by isolation valves, such that the aqueous phase could be decanted into the bottom vessel at elevated applied CO₂ pressure and the phases collected separately. In a typical experiment, 30 mL of a mixture that is 60 vol% acetonitrile and 40 vol% water with a typical gold loading of 1.93 mg (0.0644 mg mL⁻¹) was placed into the top vessel and both vessels were flushed with CO₂ to remove all air from the headspace. Both vessels were pressurized with CO₂ simultaneously. The phase separation was observed through the front viewing window and equilibrium was established when the syringe pump stopped delivering CO₂ to maintain a constant pressure. After phase separation, the bottom aqueous phase was decanted slowly into the bottom vessel by opening the isolation valve between the two vessels. Both vessels were slowly depressurized simultaneously overnight. The aqueous phase was collected for catalytic testing. The volume of each phase was determined by measuring the height change of each phase. Calibrations were performed by adding a known volume of water and measuring the height change.

The concentration of gold in solutions was also determined via Inductively Coupled Plasma – Mass Spectrometry (ICP-MS). 250 μL of a nanoparticle solution was digested in 14.75 mL of 5.6 vol% aqua regia. ICP-MS was performed by the WSU GeoAnalytical Laboratory on an Agilent 4500 ICP-MS.

Reduction of 4-nitrophenol

Reactions conducted at room temperature were performed in a 1 cm path length quartz cuvette. Dispersed gold nanoparticles (67 μL, gold loadings varied due to changes in concentrations and volumes but were determined by TGA) were first added to the quartz cuvette, followed by an aqueous solution of 4-nitrophenol (100 μL, 0.30 mM), and water (2.733 mL). Aqueous NaBH₄ (100 μL, 0.41 M) was added last to initiate the reaction. The final concentration of 4-nitrophenol and NaBH₄ was 10 μM and 14 mM, respectively. The cuvette was immediately placed in the UV-vis spectrophotometer and the absorbance measured at 400 nm (the wavelength of maximum absorbance of 4-nitrophenolate).

Reactions were performed in a 25 mL three-neck flask at various temperatures to calculate the activation energy of the reaction. The volume of each component in the reaction was increased by four times and added in the same order into the flask under magnetic stirring. The concentration of NaBH₄ was increased to 16 mM in the reaction mixture to ensure a high excess throughout the reaction under heat and stirring. Aliquots of the reaction mixture were collected periodically and the absorbance spectrum measured.

2-MBI adsorption onto nanoparticle surface

The adsorption of 2-MBI onto the nanoparticle surface was used to measure the surface availability. The procedure was developed from literature.²⁶ Time resolved absorbance spectra were collected immediately after gold nanoparticle solution was added to aqueous 2-MBI. Final concentrations were 33.6 μg mL⁻¹ and 10 μM, respectively. The absorbance peak at 300 nm wavelength was monitored for changes over time.

Results and discussion

Volume measurements

A mixture of 60 vol% acetonitrile and 40 vol% water was loaded into the top vessel of the high pressure apparatus (see Fig. S2†) and the height change of the liquid was recorded as a function of applied CO₂ pressure. The volume of each phase is presented in Fig. 1. The volume change is needed to accurately calculate concentrations before and after the phase separation. Acetonitrile was selected as the organic solvent for these OATS experiments because it is water soluble and has the ability to absorb a large amount of CO₂ relative to water.²⁷ Furthermore, mixtures of acetonitrile and water are routinely used for many industrial reactions.²⁸ Initially, the volume of the liquid decreases as it is compressed due to the increased pressure in the head space. A phase separation is observed above 9.6 bar. The volume of the aqueous phase decreases as pressure is increased because it contains less acetonitrile and the volume of the organic phase increases as CO₂ absorption increases.

Fig. 1 Volume measurements as a function of CO₂ pressure of a solution that is 30.0 mL of a 60 vol% acetonitrile and 40 vol% water at atmospheric pressure.

Gold nanoparticle synthesis

Poly(vinylpyrrolidone) (PVP) was selected as the stabilizing ligand because it is polar and water soluble and should allow for recovery of the nanoparticles in the aqueous phase. A representative TEM micrograph and size distribution of the synthesized nanoparticles is shown in Fig. 2. The average diameter of the PVP stabilized gold nanoparticles can be described by a log-normal distribution with a mean diameter of 9.4 nm and a standard deviation of 1.4 nm. While this sample of nanoparticles is polydisperse, it is representative of typical catalytic mixtures. Monodisperse populations would allow for an investigation into the role of nanoparticle size, however this is beyond the scope of these studies.

Fig. 2 Transmission electron micrograph of gold nanoparticles stabilized with PVP. Inset: histogram of gold nanoparticles fitted with a normal distribution curve.

Preliminary extraction

The percentage of nanoparticles dispersed in a particular phase was calculated from absorbance data before and after pressurization with the Beer–Lambert law. Nanoparticle recovery can be calculated by a simple mass balance normalized by the initial absorbance measurement at atmospheric pressure. The extinction coefficient is constant as it is not a function of the solvent or the ligand, and it is constant for nanoparticles of constant average size.²⁹ An expanded explanation of the calculation can be found in the ESI.† Fig. 3 shows the percentage of nanoparticles dispersed in each phase as a function of applied CO₂ pressure. CO₂ was allowed to dissolve via diffusion during this experiment which accounted for the long time period to reach equilibrium and the solution is a single phase at and below 9.6 bar. The percent of nanoparticles dispersed in this single phase is the average of the two sets of parallel windows. In the single phase region, the nanoparticles precipitate as CO₂ is added, indicated by a significant reduction in the intensity of the LSPR. Extinction spectra are available in Fig. S3 of the ESI.† No change in the LSPR band occurred nor any additional bands indicating no major changes in size or shape. The first appearance of phase separation occurred at 9.6 bar, and the nanoparticles spontaneously re-disperse in solution by 11.3 bar. At 11.3 bar, 33.6% of the gold nanoparticles spontaneously re-disperse in the aqueous phase after the improved solvent conditions provided by the phase separation. The remaining nanoparticles precipitated and attached to the inner walls of the vessel. The data suggests that some nanoparticles remain dispersed in the acetonitrile phase, but a distinction between nanoparticles precipitated on the windows and nanoparticles dispersed in solution is not possible with the method employed here. We suspect that the actual concentration of nanoparticles dispersed in the acetonitrile phase is less than what is represented in Fig. 3. The percentage of nanoparticles dispersed in the aqueous phase increases as CO₂ pressure increases because the aqueous phase contained less acetonitrile.^20,23 Partition coefficients are the ratio of the concentration of dispersed nanoparticles in the aqueous phase to the concentration of dispersed nanoparticles in the acetonitrile/CO₂ phase. Larger partition coefficients indicate greater preference to disperse in the desired aqueous phase. Partition coefficients (K_P) after the phase separation can be viewed in Table 1. As above, we believe the actual K_P are larger than the measured K_P due to precipitated nanoparticles on the windows. Above 14.8 bar the solution became cloudy and absorbance could not be measured.

Fig. 3 Recovery of nanoparticles as a function of absolute pressure of CO₂ for a 60 vol% acetonitrile, 40 vol% water and nanoparticle solution.

Table 1 Partition coefficients after phase separation for both the non-thermal treated and 40 °C thermal treated nanoparticle solutions

Absolute pressure (bar)	Partition coefficient (KP)
	No thermal treatment	40 °C thermal treatment
9.6	1.0	42.9
11.3	9.1	36.0
14.7	12.6	33.8
16.5	—	39.3

---

Low recovery was not expected as prior results demonstrate extremely high recoveries of water-dispersible catalysts.^15–19 We hypothesized two possible reasons to explain the low recovery: (1) the precipitation of the nanoparticles prior to the induced phase separation provides an opportunity for the nanoparticle to grow to larger sizes through close contact as the nanoparticle surface is not sufficiently stabilized and (2) the nanoparticles precipitate at low pressures because of a small amount of CO₂ dissolved into the single phase mixture worsening the solvent–ligand interactions sufficiently causing precipitation of the nanoparticles. If these two hypotheses could be confirmed and remedied, then high recoveries of the nanoparticles should be achieved. Preventing the growth could be accomplished by lowering the surface energy of the nanoparticle by improving stabilization via thermal treatments of the nanoparticles. We hypothesize that if the temperature of the nanoparticle dispersion is raised prior to recovery, the rotational energy barrier about the carbon–carbon PVP backbone will be reduced allowing more carbonyl oxygen groups to reach the surface, thereby reducing the surface energy and minimizing growth to larger sizes. Secondly, if the overall mixture is becoming more nonpolar because of the addition of CO₂, rapid pressurization to reach the two-phase region should prevent the precipitation of nanoparticles and lead to an increased recovery.

Thermal treatments

Thermal treatment (TT) of the nanoparticles prior to OATS separation at 40 °C was performed overnight to increase the stability of the nanoparticles during the OATS process. The nanoparticles were allowed to cool to room temperature prior to use. Nanoparticle separations were carried out in the UV-vis pressure vessel using the thermal-treated dispersion; the results are presented in Fig. 4. Prior to the phase separation, the nanoparticles were more stable at lower pressures than the non-thermally treated sample, but the same precipitation behaviour was observed below 9.6 bar. After the phase separation above 9.6 bar, 98.2% of gold nanoparticles spontaneously re-disperse; the nanoparticles in the non-thermally treated sample did not re-disperse 9.6 bar and only re-dispersed once higher pressures were reached. Recovery decreased slightly as pressure was further increased, however the nanoparticles remained dispersed above 14.8 bar. Partition coefficients for the nanoparticle dispersions after phase separation can be viewed in Table 1. The partition coefficient increased approximately 6 fold indicating that recovery of the thermal treatment is more thermodynamically favorable. Applying a thermal treatment to the solution prior to separation increased stability, partially confirming our hypothesis.

Fig. 4 Solution of 60 vol% acetonitrile and 40 vol% water and nanoparticles that was thermally treated at 40 °C overnight and allowed to cool to room temperature prior to use.

Effect of thermal treatments on nanoparticle size

Nanoparticle size was measured before and after four thermal treatments (no thermal treatment, 40 °C, 50 °C, 60 °C) were applied to separate samples having the same origin to determine the impact of the thermal treatments on the size and shape. Results can be seen in Table 2. Corresponding histograms are available in Fig. S4† and TEM micrographs are available in Fig. S5 of the ESI.† The temperature of the thermal treatments did not alter the sizes or shapes of the nanoparticles within error, likely due to an increase in surface stability.

Table 2 Effect of thermal treatment on nanoparticle size, total surface area-normalized rate constants, and induction times

Sample	Size (nm)	kSA (mL min^−1 cm^−2) × 10^−4	Induction time (min)
No thermal treatment	9.7 ± 1.4	6.2 ± 3.7	11.1 ± 1.2
40 °C	9.2 ± 1.4	4.1 ± 2.4	14.8 ± 1.5
50 °C	9.4 ± 1.5	3.8 ± 2.4	18.1 ± 2.2
60 °C	9.1 ± 1.5	4.9 ± 3.2	21.1 ± 3.5

---

Effect of thermal treatments on kinetics

The hydrogenation of 4-nitrophenol to 4-aminophenol is an ideal demonstration reaction because it produces no side products, cannot proceed in the absence of a catalyst, and can be modelled as a pseudo-first order reaction in the presence of a large excess of NaBH₄.^30–33 Fig. 5 presents typical pseudo-first-order reaction kinetics used to determine the apparent rate of reaction constant (k_app) for mixtures containing nanoparticles that have undergone the four thermal treatments. A control experiment was performed in the absence of gold nanoparticles to demonstrate that the reaction does not proceed in the absence of a catalyst. This reaction has been shown to be fit well by a Langmuir–Hinshelwood kinetic model.^1,32–38 The first implication of Langmuir–Hinshelwood model is that k_app is proportional function of the surface area of the nanoparticle and thus k_app should be normalized by the total surface area of all nanoparticles in the reaction which can be calculated from the average diameter and mass loadings. The second implication is that if the entire surface area is available for the reaction, the ligand must be rapidly desorbing and re-adsorbing to maintain the surface stability of the nanoparticle throughout the reaction. Table 2 presents the total surface area normalized rate constants (k_SA) for each thermal treatment calculated using nanoparticle loading of 1.68 μg (0.558 μg mL⁻¹) and sizes also presented in Table 2. The large uncertainties are the result of error propagation of the square of the radius in the nanoparticle surface area calculations. All reactions experienced an induction time or a period of time between the addition of all reactive species and the first disappearance of the reactants. Average induction times for each thermally treated sample can be found in Table 2. Induction time for the non-thermally treated nanoparticles is greater than other values reported in literature because 4-nitrophenol is approximately an order of magnitude less concentrated in our system. The concentration of 4-nitrophenol has been reported³⁶ to have the greatest influence on induction time. We chose a low concentration for our studies to amplify the induction time to study the effects of the thermal treatments. The observed effect of the thermal treatments was an increase of induction times as the temperature of the treatments was increased. Furthermore, the total surface area normalized rate constants decreased after the nanoparticles had been thermally treated.

Fig. 5 Typical reaction rate data based on first-order rate law for the hydrogenation of 4-nitrophenol to 4 aminophenol using the four samples of thermally treated gold nanoparticles.

This decrease of the total surface area normalized rate constant could be attributed to an actual loss of activity at the catalytic surface or could be due to an over-prediction of the available surface area. In all cases, the surface area available to the reactants was calculated as the total surface area of the nanoparticle as it is assumed that the PVP–surface interaction is highly dynamic with carbonyl oxygen groups rapidly desorbing from the surface and re-adsorbing, therefore the entire surface area would be available.^1,32–38 However, if the PVP–surface interaction was not a stochastic on–off modality and carbonyl oxygen groups effectively blocked active sites, an over prediction of the available surface area would lead to decreased total surface area normalized rate constants. Therefore, in order to better understand the nature of the PVP–surface interaction and relative surface area availability, adsorption studies were conducted using 2-mercaptobenzimidazole (2-MBI) as a probe.

Evaluation of surface availability of thermally treated nanoparticles

2-MBI has been used in previous studies to probe the available surface area of gold nanoparticles.^26,39,40 2-MBI absorbs light at 300 nm when in solution but becomes quenched due to charge transfer when it binds to the surface of a nanoparticle.⁴¹ These studies were leveraged to measure the number of potential reactive sites in situ in the liquid phase without having to dry the sample. Fig. 6 presents the concentration of 2-MBI adsorbed onto the surface of the gold nanoparticles as a function of time for the four thermal treatments. In all cases, there is a sharp initial increase as 2-MBI quickly binds to the most easily accessible sites followed by a slower adsorption as 2-MBI adsorbs to less accessible sites or displaces PVP carbonyl oxygen groups from the surface. Simulations of PVP-stabilized nanoparticles have demonstrated that not all of the PVP carbonyl oxygen will adsorb to the surface of the nanoparticle.⁴² This means that when the nanoparticles are dispersed in solution, unbound carbonyl oxygen groups are present. The increase in temperature of the thermal treatments leads to a decrease in both the number of easily accessible surface sites (as seen by the decrease in the initial spike in adsorption) and the number of less accessible sites or the adsorption rate. We interpret this as the thermal treatment leads to previously unbound carbonyl oxygen groups occupying surface sites as PVP can rotate about its backbone more easily during the thermal treatments. Furthermore, this indicates that the PVP–surface interaction is not a highly dynamic, stochastic on–off modality and carbonyl oxygen groups remain on the surface once adsorbed until there is a driving force for them to be removed.

Fig. 6 Surface volume concentration of adsorbed 2-MBI versus time for each of the four thermal treatments performed on the nanoparticles.

The adsorption of 2-MBI can further be used to explain the induction times seen with ligand stabilized nanoparticle catalysts. The induction time could be attributed to the dynamic surface rearrangement of gold atoms induced by reactants nearing and adsorbing to the surface,^36,43–45 the diffusion of reactants to the surface,^46–48 or the leaching of gold surface atoms.⁴⁹ The initial rapid binding of 2-MBI to the surface of the nanoparticles for all four thermal treatments shows that diffusion to the surface is not limited even as the surface coverage of ligand increases. Our experiments do not allow us to identify whether dynamic surface restructuring or leaching of surface atoms are responsible for the induction time. The thermal treatments have been established to increase the amount of PVP carbonyl oxygen groups on the surface with no effect on the size of the nanoparticle. We can therefore attribute the increase in induction time to increased PVP coverage. The higher coordination from the increase in carbonyl oxygen groups on the surface lower the nanoparticle surface energy producing a greater energy barrier for either dynamic surface rearrangements or atom leaching to overcome. The increased energy barrier slows both of these processes, thereby increasing the induction time. Furthermore, the increase in nanoparticle stability in OATS after the thermal treatment can now be directly attributed to an increase in carbonyl oxygen groups on the surface lowering the surface energy and slowing the rate of nanoparticle growth and precipitation.

Solvatochromic characterization

We hypothesized that the initial precipitation of nanoparticles seen in Fig. 3 and 4 is due to a small amount of dissolved CO₂ worsening the PVP–solvent interactions. A characterization of the polarity of the solvent mixture as a function of applied CO₂ pressure is necessary to describe the solvent conditions. Nile Red was used as a solvatochromic dye to qualitatively describe the polarity of the solution.⁵⁰ The wavelength of maximum absorption (λ_max) of Nile Red is larger in polar solvents than in non-polar solvents. Nile Red (30 μM) was dissolved in a mixture of 60 vol% acetonitrile and 40 vol% water and the λ_max of Nile Red was monitored as the pressure of CO₂ was increased as shown in Fig. 7. Extinction spectra were collected simultaneously through both windows and can be viewed in Fig. S6.† In all cases, a single λ_max is seen indicating that a single phase is analysed through each respective window. The λ_max is identical through each window between 0.9 and 7.8 bar indicating no phase separation has occurred. At 9.6 bar, phase separation occurs but the λ_max shift is within error indicating that the polarity in each phase is nearly the same. Above 9.6 bar, the λ_max decreases with increasing pressure indicating the solution becomes increasingly less polar as the non-polar CO₂ partitions into the liquid phase. The λ_max deviation between the two windows above 9.6 bar indicates the phase split between the two solvents. The organic phase continues to shift toward shorter wavelengths because it contains a large amount of dissolved, nonpolar CO₂. The water phase becomes more polar as evident by the shift of λ_max to higher wavelengths because it contains less acetonitrile and CO₂. Fig. 7 shows that the small decrease in polarity is likely the cause of the precipitation of the nanoparticles. Rapid pressurization to reach more favorable solvent conditions would likely increase nanoparticle dispersion because the poor solvent conditions which cause the nanoparticles to precipitate would be avoided.

Fig. 7 The wavelength of max absorption λ_max of Nile Red in a 60 vol% acetonitrile, 40 vol% water mixture when pressurized with CO₂.

Nanoparticle dispersibility was again measured to investigate whether recoveries were related to the amount of time required to reach the phase separation pressure, and the effect of temperature of thermal treatments. Four thermal treatments were performed prior to pressurizing the mixture rapidly with CO₂ to create a two-phase system as quickly as possible. Equilibrium was reached in approximately 15 minutes. Pressure was held for 72 hours to provide a direct comparison to previous results. Fig. 8 shows the percentage of gold nanoparticles dispersed versus time for each thermal treatment. The sample that did not undergo a thermal treatment resulted in improved initial dispersibility but the nanoparticles were not thermodynamically stable and precipitated over time. Thermal treatments lead to thermodynamically stable dispersions resulting in 100% nanoparticle recovery.

Fig. 8 Percent nanoparticle recovery as a function of time for four differently thermally treated solutions of 60 vol% acetonitrile and 40 vol% water. Initial measurements were collected at atmospheric pressure of CO₂ before rapidly increasing the pressure to 11.3 bar. Data were collected every 24 hours up to 72 hours.

To confirm this behaviour, a larger-scale study was performed using the apparatus shown in Fig. S2.† A nanoparticle solution of 60 vol% acetonitrile and 40 vol% water was rapidly pressurized to 11.3 bar, allowed to equilibrate, the phases physically separated, depressurized, and the two phases recovered. Gold concentrations were determine via ICP-MS and the percentage recovered and partition coefficient at 11.3 bar were calculate via a mass balance accounting for the volume changes. The percent recovered for no thermal treatment, 40 °C thermal treatment, 50 °C, and 60 °C was determined to be 91.2 ± 8.5%, 96.7 ± 7%, 88.1 ± 2.3%, and 83.6 ± 3.3%, respectively. The partition coefficients (K_P) were 23.3 ± 2.4, 16.7 ± 1.4, 36.5 ± 1.9, and 21.6 ± 8.5. These results generally agree with our in situ UV-VIS measurements with the difference likely being processing limitations.

Larger-scale recovery and reuse

A 30 mL solution of 60 vol% acetonitrile and 40 vol% water and nanoparticles was placed in the top vessel of the high pressure apparatus to investigate whether the OATS process affects catalytic activity of the nanoparticles. The nanoparticle solution did not undergo any thermal treatments because high recovery was achieved via rapid pressurization, and retention of full catalytic ability was desired. The catalytic activity of the nanoparticles was tested prior to the separation using only the aqueous portion that contained nanoparticles with the hydrogenation reaction of 4-nitrophenol to 4-aminophenol. The gold loading in the reaction cuvette was 10.8 μg (3.6 μg mL⁻¹). The aqueous portion was then separated into three separate solutions and each combined with acetonitrile. Each solution was individually loaded into the top vessel of the high pressure apparatus. Our previous results showed that recovery is path dependent, therefore all CO₂ required for phase separated was pumped into the head space leading to a rapid dissolution into the mixture in order to avoid the poor solvent conditions present at pressures below 11.3 bar. After the phase separation was complete, the aqueous phase was decanted from the organic phase into the bottom vessel and depressurized slowly overnight. The bottom aqueous phase was used in the 4-nitrophenol hydrogenation reaction as recovered. Gold loading into the reaction mixture was 7.75 μg (2.58 μg mL⁻¹) after averaging all three trials. The gold loading is a lower concentration because the recovered aqueous phase contains a small amount of acetonitrile, increasing the volume. Table 3 summarizes the size of the nanoparticles and the surface normalized rate constants for pre- and post-OATS nanoparticles. The small decrease in size can be explained by the largest nanoparticles precipitating under pressurization and not re-dispersing into the aqueous phase after phase separation. TEM images and their respective histograms can be seen in Fig. S7.† The narrowing of the histogram after pressure separation supports the explanation that the largest nanoparticles precipitated during pressurization.

Table 3 Summary of nanoparticle sizes, total surface area normalized rate constants, and induction times after phase separation and collection of the aqueous phase

	NP size (nm)	kSA (mL min^−1 cm^−2) × 10^4	Induction time (min)	Activation energy (kJ mol^−1)
Pre-OATS	10.6 ± 1.5	6.2 ± 3.3	2.1 ± 0.2	90.8
Post-OATS	9.4 ± 1.4	3.6 ± 2.6	10.3 ± 3.2	91.8

---

The available surface area of the nanoparticles was analysed using the same 2-MBI procedure used for the thermal treatments. The concentration of 2-MBI adsorbed onto the surface as a function of time can be seen in Fig. S8.† The phase separation causes a decrease in the available surface area of the nanoparticles presumably because the phase separation results from the minimization of the system free energy, including the surface energy of the nanoparticle through the adsorption of previously unbound carbonyl groups. It then follows that the phase separation would cause additional PVP to bind to the surface, causing the increase in induction time seen for the post-OATS nanoparticles resulting in higher coordination of surface atoms. The activation energy of the hydrogenation of 4-nitrophenol before and after recovery show the same value indicating the fundamental catalysis is not altered on the surface of the nanoparticle (see Table 3 and Fig. S9†). This demonstrates that we can recover nanoparticles with a slight reduction to catalytic activity because of the reduced number of catalytic sites. A slight size decrease in average nanoparticle size is observed after the phase separation which can be explained as the largest nanoparticles not being recovered, reducing the overall nanoparticle population size on average. Nanoparticles of small size tend to have enhanced kinetics because their higher surface energy due to less coordinated number of surface atoms.^1,43,51 Therefore the reduced kinetics from the smaller population of nanoparticles after their recovery can be explained as having the same underlying cause as the reduction in activity with the thermal treated nanoparticles; additional ligand reached the surface during the separation process to reduce the total energy of the system, including the surface energy resulting in more coordinated nanoparticles. Additionally, a control experiment was performed in the absence of acetonitrile to determine whether CO₂ or carbonates formed in situ are capable of poisoning the surface of the nanoparticles. The pseudo-first order kinetic plots shown in Fig. S10† show conclusively that the decrease in the rate constant is not due to a deactivation of the surface from CO₂ or carbonates.

Conclusion

A facile method for complete recovery and reuse of dispersed gold nanoparticles has been presented. This method is an improvement over previous methods because it is a general technique that does not require a specialized nanoparticle synthesis and does not produce any waste products. In this process, gold nanoparticles stabilized with PVP were placed in a 60 vol% acetonitrile, 40 vol% water solution and recovered completely using CO₂ at 11.3 bar of absolute pressure. The nanoparticles exhibited complete recovery when the solution underwent a thermal treatment of at least 40 °C prior to separation. Both the thermal treatments and the pressure separation were shown to increase the amount of ligand bound to the surface of the nanoparticles. The additional surface bound ligand improved stability in solution but decreased total surface area-normalized reaction rates and increase induction times because high coordination of PVP to the surface slowed the dynamic surface restructuring of the nanoparticles or leaching of surface atoms. Additional ligand binding to the surface of the nanoparticles shows that PVP binding is not a highly dynamic, stochastic on–off modality.

Acknowledgements

The authors would like to thank the Gene and Linda Voiland School of Chemical Engineering and Bioengineering at Washington State University for providing funding on this project along with the Franceschi Microscopy Center at Washington State University.

References

1. S. Panigrahi, S. Basu, S. Praharaj, S. Pande, S. Jana, A. Pal, S. K. Ghosh and T. Pal, J. Phys. Chem. C, 2007, 111, 4596–4605.
2. P. D. Stevens, J. Fan, H. M. R. Gardimalla, M. Yen and Y. Gao, Org. Lett., 2005, 7, 2085–2088.
3. Z. Xu, Y. Hou and S. Sun, J. Am. Chem. Soc., 2007, 129, 8698–8699.
4. J.-M. Yan, X.-B. Zhang, T. Akita, M. Haruta and Q. Xu, J. Am. Chem. Soc., 2010, 132, 5326–5327.
5. S. Kanaoka, N. Yagi, Y. Fukuyama, S. Aoshima, H. Tsunoyama, T. Tsukuda and H. Sakurai, J. Am. Chem. Soc., 2007, 129, 12060–12061.
6. J. Zhang, Y. Yuan, K. J. Kilpin, Y. Kou, P. J. Dyson and N. Yan, J. Mol. Catal. A: Chem., 2013, 371, 29–35.
7. Y. Yuan, N. Yan and P. J. Dyson, Inorg. Chem., 2011, 50, 11069–11074.
8. S. R. Saunders and C. B. Roberts, Nanotechnology, 2009, 20, 475605.
9. S. R. Saunders and C. B. Roberts, J. Phys. Chem. C, 2011, 115, 9984–9992.
10. S. R. Saunders and C. B. Roberts, Curr. Opin. Chem. Eng., 2012, 1, 91–101.
11. X. Feng, J. Zhang, S. Cheng, C. Zhang, W. Li and B. Han, Green Chem., 2008, 10, 578–583.
12. R. Zhang, J. Liu, J. He, B. Han, W. Wu, T. Jiang, Z. Liu and J. Du, Chem.–Eur. J., 2003, 9, 2167–2172.
13. P. G. N. Mertens, M. Bulut, L. E. M. Gevers, I. F. J. Vankelecom, P. A. Jacobs and D. E. De Vos, Catal. Lett., 2005, 102, 57–61.
14. R. Sablong, U. Schlotterbeck, D. Vogt and S. Mecking, Adv. Synth. Catal., 2003, 345, 333–336.
15. J. Lu, M. J. Lazzaroni, J. P. Hallett, A. S. Bommarius, C. L. Liotta and C. A. Eckert, Ind. Eng. Chem. Res., 2004, 43, 1586–1590.
16. V. M. Blasucci, Z. A. Husain, A. Z. Fadhel, M. E. Donaldson, E. Vyhmeister, P. Pollet, C. L. Liotta and C. A. Eckert, J. Phys. Chem. A, 2009, 114, 3932–3938.
17. P. Pollet, R. J. Hart, C. A. Eckert and C. L. Liotta, Acc. Chem. Res., 2010, 43, 1237–1245.
18. A. Z. Fadhel, W. Medina-Ramos, A. Wu, J. Ford, V. Llopis-Mestre, R. Jha, P. Pollet, C. L. Liotta and C. A. Eckert, J. Chem. Eng. Data, 2011, 56, 1311–1315.
19. J. P. Hallett, J. W. Ford, R. S. Jones, P. Pollet, C. A. Thomas, C. L. Liotta and C. A. Eckert, Ind. Eng. Chem. Res., 2008, 47, 2585–2589.
20. M. J. Lazzaroni, D. Bush, R. Jones, J. P. Hallett, C. L. Liotta and C. A. Eckert, Fluid Phase Equilib., 2004, 224, 143–154.
21. J. Szydłowski and M. Szykuła, Fluid Phase Equilib., 1999, 154, 79–87.
22. M. Habib Ullah, T. Hossain and C.-S. Ha, J. Mater. Sci., 2011, 46, 6988–6997.
23. D. N. Oence and T. Gu, Sep. Purif. Technol., 1996, 6, 261–264.
24. M. Anand, M. C. McLeod, P. W. Bell and C. B. Roberts, J. Phys. Chem. B, 2005, 109, 22852–22859.
25. M. C. McLeod, M. Anand, C. L. Kitchens and C. B. Roberts, Nano Lett., 2005, 5, 461–465.
26. K. Siriwardana, M. Gadogbe, S. M. Ansar, E. S. Vasquez, W. E. Collier, S. Zou, K. B. Walters and D. Zhang, J. Phys. Chem. C, 2014, 118, 11111–11119.
27. P. G. Jessop and B. Subramaniam, Chem. Rev., 2007, 107, 2666–2694.
28. N. E. Leadbeater, Chem. Commun., 2005, 2881–2902.
29. X. Liu, M. Atwater, J. Wang and Q. Huo, Colloids Surf., B, 2007, 58, 3–7.
30. N. Pradhan, A. Pal and T. Pal, Langmuir, 2001, 17, 1800–1802.
31. S. K. Ghosh, M. Mandal, S. Kundu, S. Nath and T. Pal, Appl. Catal., A, 2004, 268, 61–66.
32. Y. Mei, G. Sharma, Y. Lu, M. Ballauff, M. Drechsler, T. Irrgang and R. Kempe, Langmuir, 2005, 21, 12229–12234.
33. P. Hervés, M. Pérez-Lorenzo, L. M. Liz-Marzán, J. Dzubiella, Y. Lu and M. Ballauff, Chem. Soc. Rev., 2012, 41, 5577.
34. S. Wunder, F. Polzer, Y. Lu, Y. Mei and M. Ballauff, J. Phys. Chem. C, 2010, 114, 8814–8820.
35. S. Gu, S. Wunder, Y. Lu, M. Ballauff, R. Fenger, K. Rademann, B. Jaquet and A. Zaccone, J. Phys. Chem. C, 2014, 18618–18625.
36. S. Wunder, Y. Lu, M. Albrecht and M. Ballauff, ACS Catal., 2011, 1, 908–916.
37. P. Zhao, X. Feng, D. Huang, G. Yang and D. Astruc, Coord. Chem. Rev., 2015, 287, 114–136.
38. M. Li and G. Chen, Nanoscale, 2013, 5, 11919–11927.
39. K. Vangala, F. Ameer, G. Salomon, V. Le, E. Lewis, L. Yu, D. Liu and D. Zhang, J. Phys. Chem. C, 2012, 116, 3645–3652.
40. S. M. Ansar and C. L. Kitchens, ACS Catal., 2016, 6, 5553–5560.
41. S. Franzen, J. C. W. Folmer, W. R. Glomm and R. O'Neal, J. Phys. Chem. A, 2002, 106, 6533–6540.
42. A. Kyrychenko, O. M. Korsun, I. I. Gubin, S. M. Kovalenko and O. N. Kalugin, J. Phys. Chem. C, 2015, 119, 7888–7899.
43. X. Zhou, W. Xu, G. Liu, D. Panda and P. Chen, J. Am. Chem. Soc., 2010, 132, 138–146.
44. R. S. Das, B. Singh, R. Banerjee and S. Mukhopadhyay, Dalton Trans., 2013, 42, 4068–4080.
45. M. A. Mahmoud, B. Garlyyev and M. A. El-Sayed, J. Phys. Chem. C, 2013, 117, 21886–21893.
46. K. Kuroda, T. Ishida and M. Haruta, J. Mol. Catal. A: Chem., 2009, 7–11.
47. J. Zeng, Q. Zhang, J. Chen and Y. Xia, Nano Lett., 2010, 10, 30–35.
48. H. Yamamoto, H. Yano, H. Kouchi, Y. Obora, R. Arakawa and H. Kawasaki, Nanoscale, 2012, 4, 4148–4154.
49. M. M. Nigra, J.-M. Ha and A. Katz, Catal. Sci. Technol., 2013, 3, 2976–2983.
50. A. J. Carmichael and K. R. Seddon, J. Phys. Org. Chem., 2000, 13, 591–595.
51. H. Zhang, W. Li, Y. Gu and S. Zhang, J. Nanosci. Nanotechnol., 2014, 14, 5743–5751.

---

Footnote

† Electronic supplementary information (ESI) available. See DOI: 10.1039/c6ra11475j

---