Article pubs.acs.org/Biomac
Asymmetric Flow Field-Flow Fractionation with Multiangle Light Scattering Detection for Characterization of Cellulose Nanocrystals Xia Guan,† Rafael Cueto,‡ Paul Russo,‡ Yadong Qi,§ and Qinglin Wu*,† †
School of Renewable Natural Resources and Macromolecular Studies Group, LSU AgCenter, Baton Rouge, Louisiana 70803, United States ‡ Department of Chemistry and Macromolecular Studies Group, Louisiana State University, Baton Rouge, Louisiana 70803, United States § Urban Forestry Program, Southern University, Baton Rouge, Louisiana 70813, United States S Supporting Information *
ABSTRACT: Cellulose nanocrystals (CNCs) were analyzed by asymmetric flow field-flow fractionation (AF4) coupled with multiangle light scattering (MALS) detection. Small fractions were collected from the output of the AF4 apparatus for investigation by transmission electron microscopy (TEM). The influence of CNC injection amount, the number of passes through a high-pressure homogenizer, and different CNC sources on the elution behavior and particle size distribution was investigated. The AF4-MALS results on crystal length were compared with those from TEM. Peak distortion and variation in elution profiles with the increase in sample load were observed. Good resolution was obtained when the injection mass varied from 20 to 40 μg, corresponding to injections of 4−8 μL at a starting concentration of ∼5 μg/μL; concentrations during the separation process and at the detector were significantly lower. As the number of homogenization treatments increased, the peak shape became narrower and more symmetrical. This indicates a narrowed crystal length distribution, but regardless of source or homogenization treatment, no CNC preparation was as uniform as tobacco mosaic virus, a well-known rigid rod model structure, whose length was found by AF4-MALS to be in agreement with literature values. CNCs derived from cotton contained longer crystals than those derived from microcrystalline cellulose, as shown by both AF4-MALS and TEM techniques. An advantage of AF4-MALS compared to TEM is the ability to sample large numbers of rodlike particles, which is challenging and time-consuming for TEM image analysis, especially without the presorting afforded by AF4. The good TMV results suggest a high degree of accuracy will pertain to the CNC size distribution measurements.
1. INTRODUCTION Synthesis and application of novel renewable nanoparticles are important parts of the sustainability initiative. Cellulose nanocrystals (CNCs) have a wide application in manufacturing nanocomposite materials such as solid composites based on natural (e.g., starch, cellulose, poly(hydroxyalkanoate)) and synthetic (e.g., polyvinyl alcohol, polyurethane, epoxy resin) polymers and polymer hydrgogels.1 CNCs with different morphologies can be prepared by varying the source of cellulose and processing conditions. For example, it has been shown that whisker-like crystals with a length on the order of micrometers are obtained by hydrolyzing highly crystalline cellulose samples from tunicates2 and green algae.3 Cellulose from cotton and wood yields shorter rod-shaped particles, a few hundreds of nanometers long.1 Particle size is one of the principal parameters impacting all properties of nanoscale materials; it determines diffusivity and viscosity, which are important for understanding processing conditions. Phase relations such as the location and sharpness of the isotropicto-liquid crystalline boundary also depend sensitively on size and size distribution. Design and synthesis of CNCs with a © XXXX American Chemical Society
controlled size distribution and novel physicochemical features are key aspects in the development of current nanotechnology.4 Transmission electron microscopy (TEM)5 and atomic force microscopy (AFM)6,7 are among the most common techniques to determine CNC size and shape. However, the observed size and morphology may significantly differ from the native size and morphology CNCs display in a liquid dispersion. Light scattering (LS) methods have also been used to size-analyze CNCs.8,9 The CNCs were shown to be elongated, a few hundreds of nanometers long, 10−20 nm wide, and a few nm thick.5 Static, multiangle light scattering (MALS) gives independent information on molar mass and root-mean-square radius; in principle, the relation between these quantities may provide information on the conformation and structure of the particles. The MALS detection accuracy is reduced for nanoparticles with a complex and multimodal particle size distribution. Coupling of MALS detection with size-based Received: April 17, 2012 Revised: July 19, 2012
A
dx.doi.org/10.1021/bm300595a | Biomacromolecules XXXX, XXX, XXX−XXX
Biomacromolecules
Article
separation methods can then enhance the accuracy of size analysis of complex nanoparticle samples. Flow field-flow fractionation (FFFF or F4)10−12 is a family of methods able to separate macromolecules and particles over a wide size range (from a few nanometers to a few micrometers). Analytes can be separated by different mechanisms in F4. The “normal” mode of separation is the most widely used mechanism13 and usually applies to analyte sizes smaller than ∼1 μm. The “steric” mode is applicable for particles larger than ∼1 μm.14 A subtechnique of F4 called asymmetric flow fieldflow fractionation (AF4), introduced by Wahlund and Giddings,15 is the most-developed and applied F4 variant. AF4 is emerging as a powerful tool for obtaining highresolution information on size, molecular weight, composition, and stability of nanoscale particles in a liquid media. AF4-MALS technique has been found effective for the analysis and characterization of polysaccharides,16,17 liposomes,18 and different types of functional nanoparticles.19 AF4 can also be performed with samples of nonspherical shapes. Separation of carbon nanotubes (CNTs) by length was achieved through AF4.20,21 For AF4 analysis of rod-like CNTs, the rotational dynamics was considered in addition to the vertical position of the CNT center of mass. A Brownian dynamics simulation based on an ellipsoidal particle model was developed to model the separation of nanotubes in AF4.22 The simulation shows that the nanotube particles eluted by a normal mode mechanism up to aspect ratios of about 1000. The CNT lengths were calculated by converting the hydrodynamic volumes from polymer latex-particle calibration to rod lengths with an appropriate hydrodynamic model.20 Based on these results, size separation and characterization of rod-shape cellulose crystals by AF4-MALS could be a promising methodology. MALS was previously used to detect length distribution of rodlike cellulose;8,9,23 however, no AF4 separation has been applied to CNCs to perform a complete size-characterization. The combination of AF4 and MALS to analyze CNCs promises convenient assessment of the size distribution, but from the outset it was anticipated that the rodlike shape of CNC particles would pose special challenges. At a given mass, rods are more likely than spherical objects to interact with each other, experience long-range interactions with the walls of the AF4 separation system, or violate the usual assumptions of light scattering size analysis. The objectives of this work were (1) to establish a test protocol for analyzing size-distribution of rod-shape CNCs with an AF4-MALS system, (2) to study effect of cellulose source and processing conditions on measured crystal length distribution, and (3) to compare the AF4-MALS length data with TEM results.
Figure 1. Schematic representation of the separation principle in a normal mode AF4 separation channel.
form diffusional clouds in the channel and the thickness of the clouds depends on the diffusion coefficient of the particles and the applied field force. The larger the diffusion coefficient, the larger the resulting cloud’s thickness. Smaller particles have higher diffusivity, so their cloud is higher above the accumulation wall. As a result, they travel at a faster mean velocity and elute before the larger ones. Through balancing diffusion and applied field forces, particles are separated into fractions according to size.10 2.2. Form Factor Model for Rodlike Particles. Data collected by a MALS system following AF4 separation can be used to derive molecular parameters of the materials. In particle mode, concentration is not measured; instead, only the angular variation of the scattered light is measured and used to characterize particle shape. For completeness, we begin with the basic Zimm equation relating the Rayleigh factor Rθ (cm−1) to molar mass, form factor, and other parameters. It has the following form:25 Rθ = MP(θ ) − 2A 2 cM2P 2(θ ) ([1]) K *c where c is the mass concentration of the solute molecules in the solvent (g/mL), M is the weight average molar mass (g/mol), A2 is the second virial coefficient (mol·mL/g2), and K* = [4π2no2(dn/dc)2 λo−4NA−1] is an optical constant with no as the refractive index of the solvent at the incident radiation (vacuum) wavelength, λo. NA is Avogadro’s number, and dn/ dc is the specific refractive index increment of the solution with respect to a change in solute concentration, expressed in mL/g (this factor must be measured separately if M is to be obtained). The function P(θ) = I(q)/I(q = 0) reflects the decline of scattered intensity as q increases; it decreases from 1 to 0 in a manner that depends on the molecular size, shape, and internal structure. P(θ) can be described by the theoretically derived form factor, but at low angles it is approximately equal to [1 − 2q2⟨r2⟩/3! + ...], where q = (4πno/λo) sin(θ/2) is the scattering vector magnitude and ⟨r2⟩ represents the z-average of the squared, mass-weighted particle radius about the center of mass. For thin rods of length L, ⟨r2⟩ = L2/12. In the absence of a concentration detector, or in case the available concentration detector lacks the sensitivity required at concentrations low enough to ensure good chromatographic separation, the molar mass (M) is not determined and dn/dc is not required. Only the angular dependence of the scattered intensity is available to determine particle size parameters. The theoretical form factor P(θ) for the specific model being analyzed is fit to the q-dependent intensities. The maximum intensity at q = 0 is obtained either by extrapolation from the lowest reliable scattering angles or left as a parameter of the fit.
2. THEORETICAL BACKGROUND 2.1. Basic Principle AF4. The basic theory of AF4 is well documented in many publications,12,24 but a short description is provided to enhance the coherence and completeness of this work. Figure 1 shows a general schematic of the separation principle in a normal mode AF4 system. The separation channel consists of an impermeable top wall and a bottom block with a semipermeable ultrafiltration membrane. The semipermeable membrane sits atop a porous frit, which constitutes the accumulation wall of the fractionation channel.12 The hydrodynamic field is generated by splitting the longitudinal flow to generate a second flow stream (crossflow), which permeates the accumulation wall. The particles B
dx.doi.org/10.1021/bm300595a | Biomacromolecules XXXX, XXX, XXX−XXX
Biomacromolecules
Article
For the purpose of comparison to another well-studied colloidal rod, tobacco mosaic virus (TMV) from a previous study28 was used. 3.2. AF4-MALS Measurements. The AF4 instrument was an Eclipse 2 Separation System (Wyatt Technology Corp., Santa Barbara, CA, hence, referred to as Wyatt). A pump with an in-line vacuum degasser and an autosampler (1100 series, Agilent Technologies, Palo Alto, CA) delivered the mobile phase (water with 200 ppm sodium azide) and injected samples onto the AF4 channel. Water was chosen as carrier after comparing it with several aqueous salt and buffer solutions with no noticeable difference in the separation results. The channel was assembled with a Mylar spacer of 490 μm thickness and had a trapezoidal shape with a tip-to-tip length of 24 cm, a width at inlet of 2.15 cm, and a width at outlet of 0.6 cm. The membrane forming the accumulation wall was made of regenerated cellulose with a cutoff of 10 kDa (Wyatt). The channel flow rate was kept constant at 1 mL/min. A focus flow of 3 mL/min was first applied for 2 min prior to injection. During the injection step, all samples were injected at an injection flow of 0.2 mL/ min with a focus flow of 3 mL/min for 2 min, and then samples were focused at 3 mL/min for 5 min. During the separation process, a threestage cross-flow sequence was used: (1) 0.8 mL/min flow, hold constant for 15 min; (2) ramped flow from 0.8 to 0.01 mL/min in 80 min; and (3) 0.0 mL/min flow, hold constant for 20 min. Cross flow rates ranging from 0.3 to 1.5 mL/min were tried, but the flow profile just described gave a good compromise reflecting the competing factors of good resolution, adequate concentrations without obvious aggregation, and acceptable run times. No claim is made that these conditions are truly optimized. In selected cases, samples of 2 mL volume were collected from the detector outlet every 5 min and used for the subsequent TEM analysis. The detector set (all from Wyatt) include an Optilab rEX differential refractive index (DRI) detector operating at 658 nm, a Heleos multiangle light scattering (MALS), and a QELS (single-angle quasielastic light scattering, i.e., dynamic light scattering at a single angle, 100.29 degrees adjusted for solution refractive index). Not all these detectors would prove useful, for reasons discussed in a later section. The source for scattering measurements was a GaAs 50 mW laser operating at λo = 658 nm. Calibration of the MALS detector to obtain Rayleigh factors was done using filtered toluene. Normalization of the individual detectors was performed using an AF4 run of bovine serum albumin (BSA, Pierce Biotechnology, Rockford, IL). Under the conditions used, BSA eluted as several peaks, representing various aggregates. The peak for monomeric BSA was selected for normalization, and also used to compute the interdetector delay volumes. The finite size of BSA was taken into consideration during the former calculation. Data acquisition and processing were performed using Astra V (5.3.4.13) software (Wyatt). Lengths were obtained by fitting the angular dependence of the scattered light intensities to a rod form factor (eq 2). The exact diameter specified has no effect, but a diameter of 10 nm (based on TEM) was input to the software when analyzing CNC results, while for TMV the accepted diameter of 18 nm was selected. Cumulative sample length distributions were obtained from calculated rod length data for each sample set, using Wyatt’s Astra software. For completeness, an attempt was made to fit the angular data to the form factor for spheres; this was unsuccessful (not shown). 3.3. TEM Measurements. The morphology of the CNC samples collected from AF4 was characterized using TEM (JEOL 100CX, JEOL U.S.A., Inc.) with an acceleration voltage of 80 kV. A drop (5 μL) of a diluted suspension of CNCs was deposited on a 400-mesh carbon-coated copper grid. The material was allowed to dry, then stained with uranyl acetate to improve the contrast. The distribution of nanocrystal dimensions was obtained from the analysis of TEM images using Adobe Photoshop software. A total of 50 nanocrystals were randomly selected and the length and width of each were measured using the ruler tool.
Form factor models have been derived for many shapes, and they are covered in the text by van de Hulst.26 The theoretical form factor for a rod, P(θ ) = (1/u)
∫0
2u
sin t sin 2 u dt − t u2
(2)
can be used to determine the length of the rod. In eq 2, u = [(2πno/λo)L sin(θ/2)] (it is assumed that L ≫ rod diameter). The integral appearing eq 2 (called the sine integral) cannot be solved analytically, but its numerical values can be found in standard mathematical tables. The form factor appearing in eq 2 is only valid in the Rayleigh−Gans−Debye (RGD) limit that the particle does not reflect light efficiently and does not cause light passing through it to undergo a very different phase shift compared to light passing a similar distance directly through the solvent. Respectively, these requirements are met by the simultaneous conditions |m − 1| ≪ 1 and 2ka|m − 1| ≪ 1, where m = nparticle/ no is the ratio of particle to solvent refractive index, a is a characteristic dimension of the particle,26 and k = 2π/λ (λ = wavelength in the medium ≈ λo/no for dilute suspensions). The validity of the RGD approximation for CNCs is considered in the Results and Discussion section, along with a second constraint on eq 2, which is that it applies only to solutions sufficiently diluted that intermolecular interference effects can be neglected.
3. EXPERIMENTAL SECTION 3.1. Preparation of Cellulose Nano Crystal Suspension. Microcrystalline cellulose (MCC; Avicel FD-100 MCC, FMC Biopolymer, Philadelphia, PA) and cotton fabric were selected as raw materials for producing the cellulose nanocrystals. Sulfuric acid (95−98 wt %, VMR, West Chester, PA) was analytical grade and used as received. To make MCC-based CNCs, 40 g of MCC were mixed with 64 wt % sulfuric acid aqueous solution (700 mL) and the mixture was stirred vigorously at 45 °C for 3 h. A 5-fold dilution was then applied to the mixture to stop the hydrolysis reaction. The suspension was centrifuged at 12000 rpm for 10 min (Sorvall RC-5B Refrigerated Superspeed Centrifuge, DuPont Instrument) to separate the crystals in the suspension. The crystals were then washed with distilled water; the mixture was centrifuged and the crystals were separated again. The process was repeated five times for each sample. The precipitate was finally placed in regenerated cellulose dialysis tubes (Fisher Scientific Inc., Pittsburgh, PA) having a nominal molecular weight cutoff of 12000−14000 and dialyzed against distilled water for several days until the water pH level reached neutral. The product was designated as acid-hydrolyzed MCC.4 To produce cotton-based CNCs, the cotton fabric sample was first cut into 5 × 25 mm pieces using a fabric cutter, and then was processed with a Wiley mill (Arthur H. Thomas Co.) to pass a 100-meshscreen.The same procedures used for digesting MCC were employed to process cotton fiber, again beginning from 40 g of starting cotton fiber.27 The total yield of CNCs was about 35% by weight. To further reduce the size of the cellulose crystals, mechanical treatment was applied for the acid-hydrolyzed crystal samples. The suspension of cellulose crystals was processed through a high-pressure homogenizer (Microfluidizer M-110P, Microfluidics Corp., Newton, MA) equipped with a pair of Z-shaped interaction chambers (one 200 μm ceramic, and one 87 μm diamond) under an operating pressure of 207 MPa. Liquid CNC suspension samples from MCC (M) and cotton (C) were collected after completing a designated number of homogenization passes, labeled as CNC-material-pass number (e.g., CNC-M-1, CNC-C-1, etc.). CNC concentrations from MCC and cotton were controlled at 1 and 1.5 wt %, respectively. C
dx.doi.org/10.1021/bm300595a | Biomacromolecules XXXX, XXX, XXX−XXX
Biomacromolecules
Article
Figure 3. Effect of homogenization passes of MCC CNCs on AF4MALS fractograms and TEM length distribution. (a) LS intensity (dashed lines) and calculated rod length (full lines) for CNC-M-1, CNC-M-3, CNC-M-5, and CNC-M-12; (b) rod length distributions from TEM; and (c) a typical TEM image of the CNC-M-12 sample. Figure 2. Effect of injection mass of CNC-M-12 samples on the elution behavior (a), rod length distribution (b), and cumulative length distributions (c) from the AF4-MALS.
injecting the 5 μg/μL sample at increased injection volumes. Figure 2a shows measured LS intensity of the CNCs versus the elution volume. Peak distortion and variation in elution profiles with the increase in the sample load were observed. Good fractionation was obtained when the injection mass was less than 40 μg. Unfortunately, the DRI signal for such injections was too weak to use, meaning that the full Zimm equation (eq [1]) could not be applied. Lengths were instead obtained using the rodlike form factor, eq 2. Figure 2b exhibits calculated rod length distribution of CNCs under different injection mass. The rod length distribution approximately spanned from 75 to 300 nm. Figure 2c shows the cumulative fractions versus their
4. RESULTS AND DISCUSSION 4.1. Effect of CNC Injection Mass. To identify acceptable chromatographic conditions, which requires high dilution, and at the same time acquire a measurable detector signal, the effect of CNC sample injection mass on the separation process was investigated. The 1 wt % CNC-M-12 suspension was diluted 2fold to achieve a final concentration of CNC of ∼5 μg/μL. The experiment was carried out with an increased sample mass by D
dx.doi.org/10.1021/bm300595a | Biomacromolecules XXXX, XXX, XXX−XXX
Biomacromolecules
Article
size. At first, the particles are diluted during injection, but then they are concentrated again during focusing. They are primarily concentrated in a direction orthogonal to the transport, and we estimate from the dimensions of the focused cloud that the concentration is slightly reduced compared to that at injection. Even so, a portion of the particles may be jammed against the membrane or otherwise hindered from reaching random orientation. As a consequence, such particles may end up in a layer that migrates at a different velocity within the parabolic flow profile. As already mentioned, the DRI detector was not sufficiently sensitive to give concentrations accurately where it matters, after the MALS detector, but Figure 3 shows the peak spans about 50 mL (50000 μL) of elution volume. For an injection of 40 μg, near the upper limit for good chromatographic performance, we have injected 8 μL at 5 μg/μL so the dilution is 8/50000 = 0.00016 for the peak as a whole. Near the top of the peak, where L ∼ 150 nm, the concentration is greater but the breadth of the peak ensures that very substantial dilution applies to each slice analyzed and that even rods longer than 150 nm are measured in the dilute limit. Based on the above analysis, a 20−40 μg injection mass range was chosen for the tested samples to obtain usable detection signals. Readers interested in seeing a typical scattering envelope (intensity as a function of scattering angle) are referred to in the Supporting Information. The next question is the validity of the RGD approximation. Choosing as a worst case the largest rods, ∼300 nm, a reasonable particle refractive index nparticle ≈ 1.47, and no = 1.33 for water, the parameter |m − 1| = 0.105, which is ≪1, as required. For the wavelength used (658 nm) the parameter 2ka|m − 1| = 0.2 if we take as the characteristic dimension a = L/2. This parameter is also substantially
