11148
Langmuir 2006, 22, 11148-11157
Three-Dimensional Pore Structure of Chromatographic Adsorbents from Electron Tomography Yan Yao,†,‡ Kirk J. Czymmek,§ Rajesh Pazhianur,†,| and Abraham M. Lenhoff*,† Department of Chemical Engineering, UniVersity of Delaware, Newark, Delaware 19716, and Delaware Biotechnology Institute Bioimaging Center, Newark, Delaware 19711 ReceiVed May 10, 2006. In Final Form: August 30, 2006 The pore structure of chromatographic adsorbents directly influences macromolecular partitioning and transport in chromatography. Quantitative structural characterization of chromatographic media has generally been performed in terms of the mean pore size or, at best, the pore size distribution (PSD), but more detailed information on, e.g., connectivity has been lacking. We have applied electron tomography, a 3D TEM technique that views a sample from multiple perspectives and allows reconstruction of the volumetric structure, to capture the internal details of microporous chromatographic media with nanometer-scale resolution. Visualization of reconstructions of three adsorbents, Toyopearl SP-650 C, SP-550 C, and CM Sepharose FF, provides thorough and direct information on the geometry and the interconnectivity of the pore network. The structures are qualitatively consistent with in situ AFM images, and quantitative data for the porosities and PSDs from the analysis of tomographic data agree reasonably well with inverse size-exclusion chromatography results. For a more straightforward representation of the networking and size features of the disordered pore space, a 3D thinning algorithm was used to derive pore skeletons and consequently quantitative data on distributions of local path lengths, widths, tortuosities, and connectivities. Such enriched structural information can be instrumental in more discriminate structural evaluation and construction of engineered pore models for the study of solute intraparticle transport.
Introduction Pore geometric and topological properties are of primary importance in solute transport in chromatographic adsorbents.1 The pore size distribution (PSD) provides statistical information on size composition, and the interconnectivity2 describes the extent of communication among pores in the 3D space. Pore structure also affects adsorption capacity due to its determining the accessible surface area for adsorption. To date quantitative structural characterization of chromatographic adsorbents has relied mainly on indirect methods that measure the functional behavior, such as inverse size-exclusion chromatography (ISEC).3-5 Derivation of the PSD from ISEC measurements is based on a priori assumptions regarding pore geometry and a size distribution function. Without prior knowledge of the porelevel structure, there are uncertainties in defining a physical model that adequately represents the real structure. Furthermore, the model of parallel pores inherent in the ISEC data analysis fails to address connections among pores,6 and consequently, no topological information is obtainable from this technique. Network models have been used to interpret the bulk behavior of porous media measured by functional characterization methods, such as gas sorption, mercury intrusion, and ISEC, from which * To whom correspondence should be addressed. E-mail: lenhoff@ udel.edu. Fax: (302) 831-4466. † University of Delaware. ‡ Present address: Biotechnology Development, Schering-Plough, 1011 Morris Ave., Union, NJ 07083. § Delaware Biotechnology Institute Bioimaging Center. | Present address: Rhodia Inc., George Patterson Blvd., Bristol, PA 19007. (1) Dullien, F. A. L. Porous Media, Fluid Transport and Pore Structure; Academic Press Inc.: New York, 1992. (2) DeHoff, R. T.; Aigeltinger, E. H.; Craig, K. R. J. Microsc. (Oxford) 1972, 95, 69-91. (3) Hagel, L. In Aqueous Size-Exclusion Chromatography; Dubin, P. L., Ed.; Elsevier: Amsterdam, 1988; Vol. 40, p 119. (4) Gorbunov, A. A.; Solovyova, L. Y.; Pasechnik, V. A. J. Chromatogr. 1988, 448, 307-332. (5) Yao, Y.; Lenhoff, A. M. J. Chromatogr., A 2004, 1037, 273-282. (6) Loh, K. C.; Wang, D. I. C. J. Chromatogr., A 1995, 718, 239-255.
a connectivity parameter can be estimated.7-9 However, the connectivity derived in this way is model-dependent, and its relation to the actual topological characteristics is not well-defined. To overcome the limits of the indirect characterization approaches, experimental investigation of the detailed microscopic structure is necessary. Advances in more rigorous characterization of adsorbent structure are helpful for a better understanding of chromatographic performance, as well as for evaluating the relevance of models used in pore structural analysis based on functional characterization. Direct transmission electron microscopy (TEM) imaging can provide information on internal structure with nanometer resolution, the scale of interest for chromatographic separations. EM of chromatographic adsorbents3,10-14 has provided valuable information on pore morphology. Detailed geometric parameters, such as porosity, PSD, and accessible surface area, can be determined from 2D images15,16 by quantitative stereology,17,18 but a complete description of the topological properties in 3D is not obtainable by this technique.17 A volumetric representation of a 3D medium can be generated by reconstruction of porous media from serial sections,19-23 from (7) Tsakiroglou, C. D.; Payatakes, A. C. AdV. Water Resour. 2000, 23, 773789. (8) Portsmouth, R. L.; Gladden, L. F. Chem. Eng. Sci. 1991, 46, 3023-3036. (9) Tsakiroglou, C. D.; Payatakes, A. C. J. Colloid Interface Sci. 1993, 159, 287-301. (10) Tanaka, N.; Kimata, K.; Hosoya, K.; Araki, T.; Tsuchiya, H.; Hashizume, K. J. High Resolut. Chromatogr. 1991, 14, 40-47. (11) Afeyan, N. B.; Fulton, S. P.; Regnier, F. E. J. Chromatogr. 1991, 544, 267-279. (12) Nash, D. C.; Chase, H. A. J. Chromatogr., A 1998, 807, 185-207. (13) Hunter, A. K.; Carta, G. J. Chromatogr., A 2000, 897, 65-80. (14) Spencer, M. J. Chromatogr. 1982, 238, 317-325. (15) Fredrich, J. T.; Greaves, K. H.; Martin, J. W. Int J. Rock Mech. Min. 1993, 30, 691-697. (16) Krohn, C. E. J. Geophys. Res., Solid Earth 1988, 93, 3286-3296. (17) Russ, J. C.; DeHoff, R. T. Practical Stereology, 2nd ed.; Kluwer Academic: New York, 2000. (18) Underwood, E. E. QuantitatiVe Stereology; Addison-Wesley: Reading, MA, 1970.
10.1021/la0613225 CCC: $33.50 © 2006 American Chemical Society Published on Web 11/07/2006
Structure of Adsorbents from Electron Tomography
which the geometric and topological properties can be extracted.2 The accuracy of 3D information assembled in this way depends on whether the sectioning is fine enough to capture the structural changes in the z direction.22 Satisfactory volumetric data constructed from serial sections22,24 have been obtained for macroporous materials, such as reservoir rocks and soil.7 However, for most chromatographic media with pore dimensions spanning the nanometer range, adequate resolution cannot be obtained from stacks of serial slices approximately 50 nm in thickness, the limit of microtoming.25 The loss of z directional information within one physically sliced section can have a significant influence on the accuracy of the reconstruction of these microporous adsorbents. To go beyond the resolution threshold set by physical sectioning, alternative nondestructive approaches can be explored. Those used for characterizing porous media include confocal microscopy,26 X-ray computerized tomography,27-29 and nuclear magnetic resonance imaging,19 the resolutions of which are much coarser than the desired nanometer scale for our systems.26,30 Characterization of the fine structure of biological complexes has been realized by electron tomography,31 a technique that collects projections of a relatively thick sample at different tilt angles using TEM and reconstructs them to obtain a structure of the entire object. By exploring a finely spaced tilt series via TEM, electron tomography has the capability of resolving nanometer-scale features in the entire 3D space and has contributed greatly to progress in understanding 3D biological structures.32-34 More recent efforts have extended this technique to the characterization of porous catalysts.35-37 The demonstrated capabilities of electron tomography prove its great potential to reveal pore-level information of chromatographic adsorbents, which typically contain abundant and diverse pores. The complete 3D structure can be visualized to facilitate a qualitative overview of the internal structure. Partitioning of the complex pore space into well-defined pore components19,38-40 (19) Baldwin, C. A.; Sederman, A. J.; Mantle, M. D.; Alexander, P.; Gladden, L. F. J. Colloid Interface Sci. 1996, 181, 79-92. (20) Kwiecien, M. J.; Macdonald, I. F.; Dullien, F. A. L. J. Microsc. (Oxford) 1990, 159, 343-359. (21) Lin, C.; Pirie, G.; Trimmer, D. A. J. Geophys. Res., Solid Earth 1986, 91, 2173-2181. (22) Lymberopoulos, D. P.; Payatakes, A. C. J. Colloid Interface Sci. 1992, 150, 61-80. (23) Macdonald, I. F.; Kaufmann, P.; Dullien, F. A. L. J. Microsc. (Oxford) 1986, 144, 277-296. (24) Lin, C.; Cohen, M. H. J. Appl. Phys. 1982, 53, 4152-4165. (25) Yao, Y. Ph.D. Dissertation. Department of Chemical Engineering, University of Delaware, Newark, DE, 2005. (26) Fredrich, J. T. Phys. Chem. Earth, Part A 1999, 24, 551-561. (27) Wellington, S. L.; Vinegar, H. J. J. Pet. Technol. 1987, 39, 885-898. (28) Coker, D. A.; Torquato, S.; Dunsmuir, J. H. J. Geophys. Res., Solid Earth 1996, 101, 17497-17506. (29) Auzerais, F. M.; Dunsmuir, J.; Ferreol, B. B.; Martys, N.; Olson, J.; Ramakrishnan, T. S.; Rothman, D. H.; Schwartz, L. M. Geophys. Res. Lett. 1996, 23, 705-708. (30) Al-Raoush, R. I. Ph.D. Dissertation. Department of Civil and Environmental Engineering, Louisiana State University, Baton Rouge, LA, 2002. (31) Frank, J. Electron Tomography: Three-Dimensional Imaging with the Transmission Electron Microscope; Plenum Press: New York, 1992. (32) Ladinsky, M. S.; Kremer, J. R.; Furcinitti, P. S.; Mcintosh, J. R.; Howell, K. E. J. Cell Biol. 1994, 127, 29-38. (33) Sperling, R.; Koster, A. J.; Melamed-Bessudo, C.; Rubinstein, A.; Angenitzki, M.; Berkovitch-Yellin, Z.; Sperling, J. J. Mol. Biol. 1997, 267, 570583. (34) Mannella, C. A.; Buttle, K.; Rath, B. K.; Marko, M. Biofactors 1998, 8, 225-228. (35) Koster, A. J.; Ziese, U.; Verkleij, A. J.; Janssen, A. H.; de Jong, K. P. J. Phys. Chem. B 2000, 104, 9368-9370. (36) de Jong, K. P.; Koster, A. J. Chemphyschem 2002, 3, 776. (37) Ziese, U.; de Jong, K. P.; Koster, A. J. Appl. Catal., A 2004, 260, 71-74. (38) Liang, Z.; Ioannidis, M. A.; Chatzis, I. J. Colloid Interface Sci. 2000, 221, 13-24. (39) Spanne, P.; Thovert, J. F.; Jacquin, C. J.; Lindquist, W. B.; Jones, K. W.; Adler, P. M. Phys. ReV. Lett. 1994, 73, 2001-2004.
Langmuir, Vol. 22, No. 26, 2006 11149 Table 1. Comparison of the Porosity, E, PSD Parameters, rp and sp, and the Mean Pore Radius, rm, of the Cation Exchangers Determined by Electron Tomography and ISECa adsorbent
rp (nm)
sp
rm (nm)
EM ISEC EM ISEC EM ISEC EM ISEC
SP-550 C 0.59 0.63 14.0 6.8 SP-650 C 0.60 0.63 49.0 51.2 CM Sepharose FF 0.80 0.84 29.6 27.3
0.3 0.2 0.2
2.1 0.4 1.0
14.4 8.8 50.0 76.0 30.3 27.3
a The PSD function is defined by eq 1, in which rp and sp represent the mean and the width of the size distribution, respectively. rm is defined as rm ) ∫0∞rf(r)dr/∫∞0 f(r)dr.
allows further quantitative analysis of the pore structure. A medial axis algorithm has been applied in mapping the skeleton of the pore space,40,41 with skeleton elements categorized as connecting paths or as nodes where multiple paths converge. By providing a clear presentation of topology and a detailed record of pore geometries at nanometer resolution, medial axis analysis of the pore space allows determination of the statistics of a variety of network parameters,41 such as path length and width, tortuosity, and connectivity. These parameters are valuable additions to structural characterization for better differentiated descriptions of porous materials that may have similar bulk properties. Here we apply electron tomography for a more thorough investigation of internal pore features in microporous chromatographic media. With knowledge of two-phase spatial distribution at a fine resolution, further analysis is carried out to locate the pore skeleton and to extract statistical descriptors of pore network properties. The structural descriptors can serve as valuable inputs for the construction of engineered pore models for structural evaluation and modeling of molecular transport in porous materials. Materials and Methods Materials. Three commercial cation exchangers, Toyopearl SP550 C and SP-650 C (Tosoh Biosep, Montgomeryville, PA) and CM Sepharose FF (GE Healthcare, Piscataway, NJ), were included in the study (Table 1). Both Toyopearl SP materials have sulfonated propyl functional groups on a methacrylate copolymeric base matrix, with SP-650 C having a larger mean pore size than SP-550 C. CM Sepharose FF is a highly cross-linked agarose matrix derivatized with weak cation-exchange (carboxylate) groups. The porosities and PSDs of the adsorbents have been determined by ISEC (Table 1).42 Sample Preparation and Electron Microscopy. The adsorbents were equilibrated overnight in 5 mg/mL lysozyme solution to allow binding of protein molecules to the adsorbent. The adsorbed protein can increase the electron density of the pore region and consequently enhance the image contrast under EM. The adsorbent particles were transferred into a 1.2 mm × 200 µm high-pressure freezing (HPF) flat specimen carrier (Leica Microsystems, Cat. No. 16706898), and 3 µL of a 20% solution of dextran (MW 40 000; Sigma, St. Louis, MO) was added as a space filler. HPF of the adsorbents was carried out at a cooling rate of at least 14 000 °C/s in a Leica EMPACT (Leica Microsystems, Vienna, Austria). The samples were then freezesubstituted with 1% osmium tetroxide in acetone in a Leica EM Automatic Freeze Substitution system (Leica Microsystems), which was programmed to maintain -90 °C for 72 h, warmed at 5 °C/h to -30 °C and held for 2 h, and then warmed at 5 °C/h to room temperature. Upon reaching room temperature, samples were rinsed three times in 100% anhydrous acetone and then slowly infiltrated with EMbed 812 using 1:3, 1:1, and 3:1 (v/v) solutions of resin/ (40) Thovert, J. F.; Salles, J.; Adler, P. M. J. Microsc. (Oxford) 1993, 170, 65-79. (41) Lindquist, W. B.; Lee, S. M.; Coker, D. A.; Jones, K. W.; Spanne, P. J. Geophys. Res., Solid Earth 1996, 101, 8297-8310. (42) DePhillips, P.; Lenhoff, A. M. J. Chromatogr., A 2000, 883, 39-54.
11150 Langmuir, Vol. 22, No. 26, 2006 acetone and two changes of 100% resin at 3 h intervals. Polymerization in EMbed 812 was carried out at 60 °C for 3 days. Electron tomography involves collection of projections at different tilt angles, Fourier transformation of each projection, and Fourier synthesis of all projections back into volume data.31,43 To include sufficient information in the third direction, sample sections about 300 nm thick were prepared on a Leica Ultracut UCT ultramicrotome at the Resource for Visualization of Biological Complexity (RVBC) at the Wadsworth Center (Albany, NY). The sections were stained in an aqueous 4% uranyl acetate solution diluted 1:1 with 0.3% Tween for 1 h at 60 °C and subsequently stained with Reynolds lead citrate for 20 min. Radiation from the electron beam can lead to specimen shrinkage in the direction normal to the grid plane, usually referred to as collapse or flattening, which complicates accurate reconstruction of the volumetric data in electron tomography. To counteract this problem, the samples were pre-irradiated to allow collapse to occur before the tomographic collection, thus limiting compression of the specimen when tilting views were collected. Although structural change occurs to a certain extent, the shrinkage is uniform in the collapsing direction and nominal in the plane.44,45 With the initial section thickness known, a simple scaling of the 3D tomographic data along the z direction can be applied to recover the original depth characteristics. Accurate alignment of projections, which is essential for a successful reconstruction, can be accomplished using correlation functions46 or fiducial markers.47 15 nm colloidal gold particles were applied to one side of the section as fiducials for later alignment by least-squares analysis to determine the precise locations of images at various tilts. Single-axis tilt, double-axis tilt, and conical tilt are among the data collection methods48 with increasing coverage of the Fourier space and hence higher precision, but at the same time these also introduce a need for more extensive data processing and a greater total electron dose. Single-axis tilt series were recorded using the Albany AEI EM7 MkII HVEM operating at 1000 kV. The angular range was 120°, and the tilt interval was 2°. Atomic Force Microscopy (AFM) Imaging of Chromatographic Adsorbents. A comparison between AFM images of adsorbents in liquid and EM results can provide some information on the extent of artifacts introduced by sample preparation in EM. AFM imaging was performed in liquid tapping mode on a Nanoscope III Multimode SPM (Veeco Metrology, LLC, Santa Barbara, CA). Epoxy resin was used to attach adsorbent particles to glass coverslips (Ted Pella, Inc., Redding, CA) 15 mm in diameter. The epoxy was almost dry and care was taken to avoid the epoxy creeping onto the particle surface. Silicon nitride probes (NP series) (Veeco Metrology, LLC) with force constants ranging from 0.06 to 0.58 N/m were used to image the surface of the adsorbent particles. An optical microscope (1500×) was used to locate the attached particles so that the AFM tip would scan the particle surface once engaged. Since the scan areas imaged in this study were typically less than 1.5 µm × 1.5 µm (compared to particle diameters of order 100 µm), artifacts arising due to curvature effects should be negligible. Three-Dimensional Reconstruction and Volume Rendering. Alignment of the tilt-series and 3D reconstruction by weighted backprojection31 were carried out using SPIDER (RVBC, Wadsworth Center). Presentation of distinct properties and quantitative analysis of the volumetric data can be achieved by segmentation of the original data into a binary representation via an image processing algorithm. The original tomographic reconstruction was first processed by the segmentation utilities in SPIDER to obtain a preliminary binary volume. For samples with a large number of intricate features with (43) DeRosier, D. J.; Klug, A. Nature (London) 1968, 217, 130-134. (44) Bennett, P. M. J. Cell Sci. 1974, 15, 693-701. (45) Luther, P. K.; Lawrence, M. C.; Crowther, R. A. Ultramicroscopy 1988, 24, 7-18. (46) Frank, J. In Computer Processing of Electron Microscope Images; Hawkes, P. W., Frank, J., Eds.; Springer-Verlag: New York, 1980. (47) Lawrence, M. C. In Electron Tomography: Three-Dimensional Imaging with the Transmission Electron Microscope; Plenum Press: New York, 1992; pp 197-204. (48) Valdre`, U. J. Microsc. (Oxford) 1979, 117, 55-75.
Yao et al. low contrast, the preliminary two-phase data from automated processing are usually unsatisfactory, and extensive user intervention was necessary to obtain an accurate binary volume. Local contrast enhancement and binary conversion were performed using SPIDER and Fovea Pro 3.0 (Reindeer Graphics, Inc.). Hand-tracing of the pore-solid borders was performed using STERECON49 and Photoshop to produce the final binary reconstruction. A 3D rendering of the binary adsorbent volume was realized in IRIS Explorer 5.0 (Numerical Algorithm Group, U.K.) on a SGI Onyx 3200 visualization server. Analysis of Pore Structure. The binary 3D image was converted into a voxel model that tessellates the sample into elements corresponding to the resolution of tomographic reconstruction. Each voxel was defined as pore or solid according to the binary reconstruction. The porosity of the sample was easily obtained by recording the fraction of voxels defined as pore phase. The PSD was determined using tools of mathematical morphology,50,51 i.e., erosion of the pore volume, which imitates the size-exclusion effects that solutes experience in ISEC in the PSD characterization of adsorbents.42 The partition coefficient, which represents the fraction of pore space accessible to a certain probe, can be measured experimentally by ISEC.42 Here it was determined from the reconstructed structure using an image-oriented approach: the accessible space is estimated as the pore region remaining after erosion by a probe of a specific size. By application of the erosion algorithm using probes with a range of sizes, the calibration curve, with partition coefficients as a function of probe size, can be constructed. The derivation of the PSD from the distribution coefficients is analogous to the analysis used in ISEC,42,52 which assumes that the pores are cylindrical with radii that follow a lognormal distribution, f(r) )
[ ( )]
A 1 ln(r/rp) exp r 2 sp
2
(1)
Topological information was obtained by skeletonization/thinning of the volumetric data using the topology-preserving algorithm 3DMA,41,53 which is analogous to the burning of the pore space layer by layer to reduce a 3D data set into a collection of geometric centers, i.e., the medial axis. Connectivity can be defined on a cubic grid, whether just along the three principal coordinate axes or also allowing for connections with voxels sharing edges or vertexes with the central voxel. The latter (26-connected) medial axis is more realistic than the former (6-connected) one and was used in the skeletonization process here. The burning algorithm produces the initial medial axis, consisting of voxels defined as path elements, i.e., voxels with only two immediate skeleton voxel neighbors and nodes, where multiple paths meet. The paths are categorized into three types: a branch-branch path connecting two nodes; a branch-leaf path involving only one node; a leaf-leaf path that is essentially isolated. All the leaf-leaf paths and those branch-leaf paths not in contact with the volume boundaries were removed to yield the final percolating skeleton. This facilitates evaluation of the extent of connectivity of the pore region that is effective for solute transport. The Inventor data (a 3D data format) of the medial axis was generated for viewing in IRIS Explorer. In this way the void space is partitioned into nodal pores and pore channels, which allows more structural descriptors to be defined.41 The local constrictions on each pore path, referred to as the pore throats, may result in bottleneck effects on solute partitioning and transport, as envisioned in some network models, with pore throats defined as the network bonds.54 (49) Marko, M.; Leith, A. J. Struct. Biol. 1996, 116, 93-98. (50) Vogel, H. J.; Roth, K. AdV. Water Resour. 2001, 24, 233-242. (51) Serra, J. P. Image Analysis and Mathematical Morphology; Academic Press: New York, 1982. (52) Yao, Y.; Lenhoff, A. M. J. Chromatogr., A 2006, 1126, 107-119. (53) Ma, C. M.; Sonka, M. Comput. Vision Image. Und. 1996, 64, 420-433. (54) Bryntesson, L. M. J. Chromatogr., A 2002, 945, 103-115.
Structure of Adsorbents from Electron Tomography
Langmuir, Vol. 22, No. 26, 2006 11151
Figure 1. z slices of the tomographic reconstruction of (a) Toyopearl SP-550 C, (b) Toyopearl SP-650 C, and (c) CM Sepharose FF. In (a) and (b) the lightest areas in the image are the methacrylate base matrix, the darkest lines are pore walls with bound protein, and the dark clusters enveloped by the darkest walls are pore channels. In (c) the black bundles are the cross-linked agarose base matrix with bound protein and the intervening light regions are pore space. (1)-(4) in each set are slices 50 nm apart along the z-axis of the sample. The interconnectivity is characterized by the coordination number, which corresponds to the number of paths sharing a common node. The path tortuosity is defined as the shortest path between two medial axis voxels divided by the Euclidean distance51 between the two points.55 Details of the path tortuosities and the local constrictions in the paths have been recognized as critical factors in determining macromolecular diffusion in chromatography.
Results and Discussion Tomographic Reconstruction. The tomographic reconstruction contains the information necessary for a complete 3D characterization of the adsorbent, which can be presented in various ways. The volumetric data can be sliced in any direction to reveal the respective planes. For example, the z slices of the tomographic reconstructions of the three adsorbents (Figure 1a(55) Prodanovic, M. Personal communication. Applied Mathematics and Statistics Department, Stony Brook University, Stony Brook, NY, 2004.
c) provide information on pore geometry features similar to that seen in 2D TEM, but the computational slices of the reconstructions can be obtained at any arbitrary z value. The individual images in the tilt series were collected using a pixel size of approximately 3 nm, but the procedures for generating the 3D reconstructions give rise to a lower overall resolution. This is estimated to be approximately 8 nm according to the Crowther formula,56 which is generally assumed to represent an upper bound on the actual resolution. The Toyopearl materials (Figures 1a, b) contain pore channels that have proteins fixed to their surface due to high-pressure freezing with freeze substitution; thus, interpretation of features on the slices is that medium to sharp dark regions are pore space and the lightest regions correspond to the solid methacrylate polymer. The image for SP-550 C features sharp dark edges with faint threads, but a (56) Crowther, R. A.; DeRosier, D. J.; Klug, A. Proc. R. Soc. London, Ser. A 1970, 317, 319-340.
11152 Langmuir, Vol. 22, No. 26, 2006
Yao et al.
Figure 3. Binary z slices of 3D reconstruction of SP-550 C (a2), SP-650 C (b2), and CM Sepharose FF (c2), compared to the original tomographic data (a1-c1).
Figure 2. Height channel AFM image of adsorbents from liquid tapping mode imaging: (a) Toyopearl SP-550 C (scan size 1.59 µm square, z range 600 nm); (b) Toyopearl SP-650 C (scan size 1.78 µm square, z range 661 nm); (c) SP Sepharose FF (scan size 2.19 µm square, z range 130 nm).
more smeared-out image is obtained for pores in SP-650 C. Considering the staining effects, the interpretation of the sliced images is that the two Toyopearl adsorbents have similar pore geometries, with SP-650 containing pores larger than SP-550. The tomographic reconstructions of the two Toyopearl materials are also consistent with the AFM images obtained in situ by liquid tapping mode, shown in Figures 2a, b. These confirm the similar pore geometries but also show more clearly the 3D
methacrylate structure, most notably the irregular surface comprising multiple polymer clusters. The resulting pore structures are similarly irregular, and these correspond qualitatively to the pore shapes seen in cross section in the tomographic planar slices. The interpretation of images in CM Sepharose, which contains voids formed between cross-linked agarose bundles, is different from that of the Toyopearl materials containing pore channels. CM Sepharose has an agarose backbone with bound proteins appearing dark, and a large volume of light areas corresponding to pore space (Figure 1c). The cross-linked agarose bundles in this Sepharose-based material leave abundant pore space, thus resulting in a larger porosity, as also found by ISEC (Table 1). The thickness of the agarose fibers shown in the tomographic slices (Figure 1c) is about 10 nm, which is similar to the value found in an SEM study of Sepharose 5B.14 In situ inspection of the pore microstructure (Figure 2c) using AFM suggests that the overall structure of the adsorbent is well preserved in EM. A “walk-through” of xy planes along the z direction shows the tortuosity of the pore space, as demonstrated in the changes of pore features in different planes. The gray scale images reflect various subtle structural features, but for the purposes of further analysis, binary images were produced using image segmentation to delineate the pore/solid boundaries (Figure 3). For ease of comparison, the color schemes
Structure of Adsorbents from Electron Tomography
Langmuir, Vol. 22, No. 26, 2006 11153
Figure 4. 3D reconstruction of Toyopearl SP-550 C (a1 and a2), Toyopearl SP-650 C (b), and CM Sepharose FF (c). (a2) is a magnified view of the SP-550 C structure. Within the frame, black represents the void space, and the colored region is the solid base matrix in the resin.
in the binary presentations are similar to those in the original tomographic data, with Toyopearl materials containing black pores and white solid and CM Sepharose FF retaining the black agarose bundles and white void space. The section of SP-650 C shows large patches of pore and solid. The layout of the porespace clusters vs solid phase in SP-550 C is similar to that in SP-650 C, the distinction being the fine divisions within the pore clusters that divide the whole space into small pore units in SP-550 C. The generation of binary data is the limiting step in the visualization and quantitative characterization of the 3D structure. For materials with intricate structural details, significant effort was expended in using hand tracing and computer recognition to complete the rendering, but the identification of material property distributions remains subjective and dependent on detailed processing strategies. Nevertheless, the essence of the structure should be conserved by the methods used. Volume rendering of the reconstruction (Figure 4) allows direct observation of the internal structure. SP-650 C has large open pores and a fairly simple architecture, while SP-550 C has a large number of small pores clustered into discrete pore compartments. As indicated by the volume rendering, SP-550 C can provide a much higher adsorption surface area for accessible solutes, as supported by experimental measurements of the high binding capacity of small molecules.57 The void space in the Sepharose material is organized differently: instead of taking
the form of well-defined pore channels, the voids are formed among the intertwined agarose bundles. Porosity and Pore Size Distribution. Good agreement was obtained between the porosities and PSDs obtained from tomography analyses and ISEC measurements (Table 1). The partition coefficients estimated for solutes of different sizes from volume image analysis of the tomographic reconstructions are compared with those from ISEC measurements in Figure 5. The calibration curves for CM Sepharose FF from the two methods overlap with very minor deviations, suggesting that the tomographic reconstruction captures the characteristics of the pore space in this Sepharose material quite well. For SP-550 C and SP-650 C, however, noticeable discrepancies between the tomography results and ISEC are apparent in the permeation extents of small probes and are quite pronounced for SP-550 C. A significant portion of the small pores (radius < 3 nm) may be sampled in ISEC but not in the tomographic analysis due to the marginal resolution in this limit, and this effect is compounded by the use of adsorbed lysozyme (approximately 3-4 nm in diameter) to enhance contrast. SP-550 contains predominantly small pores; consequently, the tomographic analysis of this material has the larger deviation from the ISEC result in the small-pore range. (57) Staby, A. Personal communication. Protein Purification, Novo Nordisk A/S, Gentofte, Denmark, 2004.
11154 Langmuir, Vol. 22, No. 26, 2006
Yao et al.
Figure 5. Calibration curves from 3D image analysis of tomographic reconstruction (lines) compared with the ISEC measurements (4): (a) SP-550 C; (b) SP-650 C; (c) CM Sepharose FF.
The tomographic technique therefore provides an appreciable capability to capture fairly fine structural characteristics that otherwise would remain inaccessible. The likelihood of obtaining a complete characterization of the pore structure would be reasonably good for materials with pore dimensions mostly above the resolution limit. For materials with more challenging pore content, i.e., small pores that cannot be clearly contrasted in electron microscopy, it is advisable to combine microscopy with other characterization techniques for evaluative purposes. Another caveat is that structural analysis based on the tomographic reconstruction of a limited portion of an adsorbent particle may yield results different from those from functional characterization methods such as ISEC, which investigate the bulk behavior of a relatively large quantity of adsorbent particles. Information from microscopic imaging is representative of the overall properties only if the adsorbent is macroscopically homogeneous;1 i.e., the imaged section contains most of the features representative of the material. Extensive sampling of imaged regions of the adsorbent is necessary to determine the applicability of tomographic data to the description of overall structure and the rationalization of macroscopic properties. Medial Axis Analysis of Adsorbents. The complete 3D data sets provide the raw information for extraction and compilation of a plethora of quantitative data regarding each of the adsorbents. The percolating medial axis of the adsorbents (Figure 6a-c) gives a clear presentation of the pore architecture in a condensed and informative form, with the color coding indicating the local pore cross section. Overall the pore connectivity characteristics extracted from the complete tomographic reconstruction indicate the unrealistic nature of cubic network models. In addition, though, pronounced differences among the three materials are apparent. SP-550 C contains abundant pore paths in a fairly dense skeleton, including a large population of red skeleton voxels, viz. regions of small local pore radius (Figure 6a). Wide pores and a much simpler pore architecture are seen within a sample block of SP650 of similar size (Figure 6b). The maximum pore width in SP-650 is about 90 nm, although there are narrow constrictions along the pore paths, as seen from the color distribution (Figure 6b). The intertwined agarose bundles in CM Sepharose FF (Figure 6c) leave an intermediate number of skeleton paths, with a maximum local path width of approximately 70 nm. Local constrictions (throats) in the pore channels were identified in the pore networks. A wide range of throat sizes were found in SP-650 C, while CM Sepharose FF has more evenly distributed throat dimensions. As the z-dimensions of the samples from which the tomographic data were obtained were relatively small due to concerns about tomographic resolution, the edge effects are more pronounced for such data, with a relatively large
proportion of unknown components at the boundaries. This severely impairs the capability of the analysis software to identify the throats, and consequently, throats on a number of paths that exchange information with the region outside the imaged block cannot be defined. Nevertheless, the skeleton view of the volumetric data, along with the throats of the completely defined pores, provides an effective organization of the detailed pore information, including a number of properties with significance for transport and adsorption. Pore Structure Statistics. Visualization of the color-coded skeletons of the pore space provides a qualitative depiction of the geometry, interconnection extent, and tortuosity of the pore phase in the adsorbents. The decomposition of the heterogeneous pore space into paths and nodes allows identification and measurement of a number of structural descriptors. As information on the lengths of the branch-leaf type pores that extend through the edges of the sample is lost, the statistical analysis here is performed using only the branch-branch paths. The averaged properties and the detailed distributions of the pore structure descriptors are shown in Table 2 and Figure 7a-l. The pore path length determines the distance that a molecule must migrate between junction points or nodes, at which multiple pore paths meet. In narrow-pored materials, such as SP-550 C, the pore space is divided into a large number of paths, while for the macroporous SP-650 C, the layout of only 335 branchbranch paths in the equidimensional sample block is rather simple (Table 2). Paths shorter than 20 nm are prevalent in SP-550 C, a more uniform distribution of path lengths in the range of 0-500 nm is observed in SP-650 C, and a unimodal distribution with a mode of about 100 nm is found for CM Sepharose FF (Figure 7a). The tortuosity denotes the extent of deviation between an idealized model assuming a straight pore channel and the more irregular path found in reality. It is significant in determining the effective rate of progress of a diffusing molecule from the surface to the center of the particle and can indicate the likely efficacy of using a simplified network model to characterize bulk transport through the material. The tortuosities were found to be in the range 1-4.5 for these adsorbents (Figure 7b). For all branchbranch paths, SP-650 C has the highest average tortuosity (Table 2), but the detailed statistics can shed additional light on features in a microscopic picture. For SP-650 C and CM Sepharose FF there is one major peak in the distribution of over 40% frequency, around tortuosity values of 1.4. For both the Toyopearl materials a significant population of pores with tortuosities >4 is calculated. Distributions of the burn distances of path voxels, i.e., path widths (or pore radii), for these materials follow a unimodal distribution of a form similar to the log-normal, with a median
Structure of Adsorbents from Electron Tomography
Figure 6. Skeleton of the pore space in (a) Toyopearl SP-550 C, (b) Toyopearl SP-650 C, and (c) CM Sepharose FF. Colored lines within the bounding box are the medial axis of the pore space, with the color denoting the local size (radius) of the pore space. The color bars define the ranges of the size distributions.
value of 9 nm for SP-550 C, a broad distribution up to 90 nm with a median value of 31 nm for SP-650 C, and a median value of 25 nm for CM Sepharose FF (Figure 7c). In network modeling the pore radius is sometimes assumed to be proportional to pore length.1,58,59 However, experimental (58) Imdakm, A. O.; Sahimi, M. Chem. Eng. Sci. 1991, 46, 1977-1993.
Langmuir, Vol. 22, No. 26, 2006 11155
measurements on a real random sphere packing indicate that neither this assumption nor that of constant pore length is valid for that system.60 It is thus informative to compare the lengths and widths of pores in real structures. The maximum and minimum widths (burn distances) on each path are compared with the corresponding pore length in Figure 7d,e. SP-650 C has fairly widely scattered ratios of length and width, including a number of very long and wide pores, up to 500 nm in length and 80 nm in width. These may be macropores that span significant distances within the material and that in so-called perfusive materials11 are thought to allow intraparticle convection that reduces the intraparticle transport resistance. In all three adsorbents, there are individual cases where the local constrictions in the pore path are quite narrow. The distributions of the length/the maximum burn distance, the length/the minimum burn distance, and the maximum/minimum burn distance on each path are presented in Figures 7f, g, h, respectively. Concentrations of values in the distributions of L/Dmax in the range 1-10, with the peak position around 2.5-4, are observed on all three adsorbents. Approximately 10-15% of pores are shorter than the maximum burn distance, and all the adsorbents have a considerable fraction of pores with low aspect ratios, L/Dmin < 5. SP-550 C has almost 20% pores with relatively uniform diameters, Dmax/Dmin < 1.5, and overall no Dmax/Dmin values are high than 8. Maxima in the distributions are found at Dmax/Dmin ∼ 1-1.5 for SP-650 C and CM Sepharose FF. SP-650 C contains pores with the most roughness or diameter variation; Dmax/Dmin > 25 is observed. The equivalent throat radii (Figure 7j) show that a major constriction of less than 6 nm is prevalent in SP-550 C, while SP-650 C and CM Sepharose FF have wider distributions of throat sizes. Wide throats greater than 60 nm are observed in both SP-650 C and CM Sepharose FF. The maximum burn distance on the cluster defines the dimension of the junction of pore paths (Figure 7k), while the connectivity is a key feature of the topology of the pore space. Coordination numbers of 3 are dominant in the distributions of all three materials (Figure 7l). Larger fractions of higher coordination numbers are seen in the Sepharose material, which consequently also has the highest average coordination number (Table 2). A statistical analysis of the kind presented here provides a fine-grained quantitative characterization of the complex structure of the pore space with considerably greater detail than is available from ISEC. The statistics of the pore dimensions agree with our qualitative knowledge of the characteristics of the void space from microscopy, with SP-550 containing mostly short and narrow pores, the relatively macroporous SP-650 displaying a wider sampling of sizes, and CM Sepharose FF being intermediate in the mean pore size and the width of the distribution. Additional detailed comparisons of this kind relating to 3D characteristics, such as the tortuosity, the aspect ratio of the pore path, and the connectivity, cannot be inferred easily from either functional characterization or qualitative microscopy, so the results obtained here are unique in their quantitative structural detail. Considering the properties of the adsorbent base matrixes, it was initially speculated that CM Sepharose FF, based on cross-linked agarose bundles, would have significantly higher connectivity than the Toyopearl adsorbents, which contain dispersed pore channels. However, our skeleton analysis of the tomographic data yields only a slightly higher connectivity for the Sepharose material (average coordination number of 5.3 for CM Sepharose FF vs (59) Rege, S. D.; Fogler, H. S. Chem. Eng. Sci. 1987, 42, 1553-1564. (60) Bryant, S. L.; King, P. R.; Mellor, D. W. Transp. Porous Media 1993, 11, 53-70.
11156 Langmuir, Vol. 22, No. 26, 2006
Yao et al.
Table 2. Pore Structure Statistics of SP-550 C, SP-650 C, and CM Sepharose FF from Medial Axis Analysisa path no./µm3
average path length
tortuosity
adsorbent
tot.
b-b
b-l
tot.
b-b
b-l
tot.
b-b
b-l
cluster no./µm3
P
SP-550 C SP-650 C CM Sepharose FF
21902 827 1983
9102 335 978
12800 492 1006
45.9 152.9 110.5
45.8 146.9 123.9
46.0 157.0 97.5
1.63 1.68 1.46
1.97 2.04 1.55
1.40 1.44 1.37
4504 200 463
4.54 4.64 5.29
a
b-b refers to the branch-branch paths, and b-l, to branch-leaf paths. P is the coordination number averaged over all the clusters.
Figure 7. Pore structure statistics derived from the medial axis analysis of the tomographic data of (1) SP-550 C, (2) SP-650 C, and (3) CM Sepharose FF: distributions of (a) path lengths, (b) pore tortuosities, (c) path burn distances, (d) path lengths, L, vs the maximum burn distances of the paths, Dmax, (e) path lengths, L, vs the minimum burn distances of the paths, Dmin, (f) the ratio of path length, L, to the maximum burn distance of the path, Dmax, (g) the ratio of path length, L, to the minimum burn distance of the path, Dmin, (h) Dmax and Dmin, maximum vs minimum burn distances of paths, (i) the ratios of Dmax and Dmin, maximum and minimum burn distances, (j) the equivalent circular radius of pore throats, Rth, (k) the burn distances of clusters, and (l) the coordination numbers.
4.6 for SP-650 C). Possible reasons for the smaller difference in connectivity than expected include the higher porosity of CM
Sepharose FF (80% for CM Sepharose FF vs approximately 60% for SP-650 C and SP-550 C) and the different void
Structure of Adsorbents from Electron Tomography
architecture in Sepharose vs the Toyopearl materials. The abundant void space in CM Sepharose, spanned by the intertwined agarose bundles, is partitioned into a large number of pore paths, but the connectivity is not very high because the connecting nodes are distributed throughout the skeleton instead of converging at a limited number of nodes. The structural descriptors obtained enrich the micropore-level knowledge of the adsorbent that directly influences solute adsorption and transport behavior. The 3D pore dimensions at voxel resolution can be used to estimate the extents of size exclusion and the available surface for adsorption. The population of pore widths, coupled with the overall skeleton structure, determines the hindrance that the solute experiences at the pore level. The local length and width of a pore section define the time that a molecule migrates before reaching intersections, at which the connectivity plays a significant role in directing the solute toward a transport-efficient route. Overall, a preliminary qualitative analysis of transport and adsorption in adsorbents can be discussed using these statistics. However, an investigation of the relation between the measured transport properties, structural statistics, and the complete tomographic data containing the detailed spatial arrangement of pore features is required to obtain a quantitative description of performance properties and to define structural descriptors that govern adsorbent performance. Results of this aspect of the investigation are reported elsewhere.61
Conclusions TEM imaging of adsorbents provides pore-level knowledge that is invaluable for a mechanistic understanding of the chromatographic process. However, tomographic reconstruction goes significantly further in revealing the quantitative, 3D structure of chromatographic adsorbents with nanometer resolution. The subsequent structural characterization, which yields PSDs in reasonably good agreement with those from ISEC, demonstrate the quantitative capability of tomography. Analysis (61) Langford, J. F.; Schure, M. R.; Yao, Y.; Maloney, S. F.; Lenhoff, A. M. J. Chromatogr., A 2006, 1126, 95-106.
Langmuir, Vol. 22, No. 26, 2006 11157
such as the medial axis analysis shown here then produces extensive sets of structural statistics that can be used in comparisons of different adsorbents. Specifically, the availability of this information obviates the need for fitting parameters in quantitative transport models such as network models, which limits the predictive value of the models. Major challenges in 3D structural imaging of chromatographic media remain, including optimal landmarking for image alignment and the complicated and laborious data processing. Where detailed quantitative information is not required, simpler 2D TEM remains a valuable tool for applications such as identifying structural variations at various levels, not all of which are detectable by functional characterization, e.g., ISEC. This can be useful to identify lot-to-lot variability in the pore structure, which can be manifested in process unpredictability in preparative scale operations. Related questions remain regarding how representative the particles studied are of the commercial materials and how extensive structural variations are within particles. All of these characteristics can directly influence macroscopic chromatographic performance, as well as its predictability and scale-up. Acknowledgment. This work was supported by the National Science Foundation (Grant Nos. CTS-9977120 and CTS0350631). We are thankful to the Wadsworth Center’s Resource for Visualization of Biological Complexity (RVBC), an NIH National Biotechnological Resource supported by Grant RR 01219 from the National Center for Research Resources (DHHS/ PHS), for making the electron tomography study possible. We are grateful to Karolyn Buttle, Dr. Ardean Leith, and Dr. Carmen Mannella at RVBC for technical support and helpful discussions. We thank Dr. Masa Prodanovic and Dr. W. Brent Lindquist at Department of Math and Statistics, State University of New York, Stony Brook, NY, for kindly providing the 3DMA package and helpful discussions on 3D skeletonization. High-pressure freezing was performed in the Delaware Biotechnology Institute BioImaging Center, which is partially supported by the NIH/NCRR under INBRE Grant 2P20 RR016471. LA0613225
