Size distribution of CNCs from various sources of cellulose.
Cellulose nanocrystals (CNCs) are high aspect ratio nanomaterials readily obtained from cellulose microfibrils via strong acid hydrolysis. They feature unique properties stemming from their surface chemistry, their crystallinity, and their three-dimensional structure. CNCs have been exploited in a number of applications such as optically active coatings, nanocomposite materials, or aerogels. CNC size and shape determination is an important challenge and transmission electron microscopy (TEM) is one of the most powerful tools to achieve this goal. Because of the specifics of TEM imaging, CNCs require special attention. They have a low density, are highly susceptible to electron beam damage, and easily aggregate. Specific techniques for both imaging and sampling have been developed over the past decades. In this review, we describe the CNCs, their properties, their applications, and the need for a precise characterization of their morphology and size distribution. We also describe in detail the techniques used to record quality images of CNCs. Finally, we survey the literature to provide readers with specific examples of TEM images of CNCs.
- cellulose nanocrystals
- transmission electron microscopy
- particle size
- size distribution
- sample preparation
1.1. Native cellulose and the production of cellulose nanocrystals
Cellulose, the most abundant biopolymer on Earth, has been extensively used by man, in the form of macro- and microstructures, as a traditional resource in many aspects of daily life, notably to produce textiles and papers. This polymeric material is biosynthesized by a wide variety of living species such as plants, animals, bacteria, and some amoebas. Glucose is polymerized by enzymes in a continuous fashion. The resulting cellulose chains are homopolymers of β-1,4-linked anhydro-D-glucose units which associate to form microfibrils, further assembled into macrofibers and fibers. Crystalline and disordered regions alternate along the microfibrils (Figure 1a) [1-4].
In the late 1940s, cellulose crystallites were isolated for the first time by chemical treatment of a cotton substrate in hot concentrated sulfuric acid . Soon after, Rånby showed that stable colloidal suspensions of negatively charged cellulose particles could be obtained [6, 7]. During the extraction process, the cellulosic fibrous structure is broken down in the presence of concentrated sulfuric acid (other mineral acids can be used such as HCl). After diffusion of the acid within the substrate, the glycosidic linkages in disordered regions, more accessible and reactive, are preferentially broken. An additional mechanical or ultrasound treatment results in the release of rodlike cellulose crystallites, so-called cellulose nanocrystals or CNCs (Figure 1b). Consequently, as the hydrolysis proceeds, the degree of polymerization of the cellulose macrostructures decreases, while the crystallinity of the nonsoluble particles increases .
1.2. Cellulose nanocrystal properties
In the early 1950s, detailed characterizations of the shape and size of various CNCs (cotton, ramie , and bacterial cellulose ) were proposed from transmission electron microscopy (TEM) images. Since then, CNCs have been produced from a wide (and expanding) variety of sources  such as wood (Figure 2a), cotton (Figure 2b), bamboo (Figure 2c) , bacteria (Figure 2d) , algae (Figure 2e), tunicates (Figure 2f) , eucalyptus , spruce bark , corncob , etc. The CNCs are rodlike objects with a length typically ranging from 50 to 1000 nm and a width varying from 3 to 50 nm (Table 1). CNCs thus have high length-to-width (aspect) ratios (10-100) [17-22]. Their morphology depends on the cellulose source and the conditions of preparation (type and concentration of acid, acid-to-cellulose ratio, reaction time, and temperature) (Figure 2). The nanoscopic features of the resulting CNCs considerably influence their colloidal and macroscopic properties such as suspension rheology, phase separation concentration, liquid crystal behavior, orientation under electric or magnetic field, and mechanical reinforcement ability in nanocomposites [14, 16].
CNCs can be prepared in different forms. First, CNCs can be manipulated in the form of acidic aqueous suspensions resulting from the strong acid hydrolysis of microfibrils or as neutral suspensions after neutralization. Subsequent surface modifications can be achieved via chemical treatment [23, 24]. Alternatively, different drying methods may be used to afford a fluffy material, with the aspect of white sugar candy .
In 1959, Marchessault et al. revealed that the chemical reaction of sulfuric acid with hydroxyl groups at the surface of cellulose CNCs formed sulfate ester groups, resulting in electrostatic repulsions between the particles and ensuring colloidal stability. Moreover, the authors showed that CNCs could self-organize into liquid crystalline phases .
The sulfuric acid hydrolysis of cellulose has remained confined to academic research laboratories until a big step was made in the mid 1990s, when it was shown that CNCs were efficient reinforcing fillers in latex-based polymer matrices, opening a new potential market for this high-end material [14, 27]. Since then, CNCs have become intensively studied with an accelerating rate of publications released each year [17-22]. Beside mechanical properties, colloidal properties of CNCs in suspension generated a series of studies investigating their ability to form liquid-crystal phases. The colloidal suspensions of CNCs spontaneously organize into a chiral nematic phase above a certain critical concentration. As a consequence, CNCs have been used to produce iridescent and birefringent films [28-31], chiral mesoporous silica [32, 33] and carbon , gold nanoparticle films with chiral plasmonic properties , enantioselective amino acid hydrolysis catalysts , hydrogels , and aerogels . Moreover, the high surface area and controllable surface chemistry of CNCs make them a unique support for metal nanoparticles (NPs)  CNC supported NPs, including Pd  and Au , were used as nanocatalysts. Pd NPs distributed onto CNCs were used for hydrogenation and carbon-carbon bond coupling reactions [40, 42-44], while the Au counterparts were used for 4-nitrophenol reduction . Ag NPs were also deposited onto CNCs and the resulting material featured antibacterial activity .
This large panel of high-end applications strongly encouraged industry to achieve large-scale production of CNCs. The forest-based industry in North America, Northern Europe, and Japan is currently looking into renewing and reinventing itself to extend its activities and guarantee its survival, while renewable materials are being increasingly appealing as fossil-based material replacement. In addition, CNCs are inherently safe, practically nontoxic materials . Nanocellulose-based, value-added materials definitely constitute promising vectors to “turn wood into gold” and revive the forest-based industry. These combined factors led to the opening of the first commercial plant by CelluForce Inc. (Windsor, Québec, Canada), producing 1 ton of CNCs per day .
1.3. Need for CNC characterization
CNCs possess appealing properties in direct relation with their three-dimensional (3D) structure, including well-defined shape, size, and aspect ratio. From an industrial perspective, it is essential to collect reliable data on CNCs, especially for quality control, toxicology assessment, R&D and applications.
1.3.1. Quality control
The large-scale production of CNCs became a reality when CelluForce Inc. started up its demonstration plant. Getting reliable, fast, and accurate measurement of the particle size is a key to guarantee a consistent production of high-quality CNCs. Several methods exist (vide infra) to determine the size, size distribution, and shape of CNCs. So far, they are typically used in academic studies and provide a fairly consistent understanding of the material structure. However, for the moment, no systematic and streamlined method exists for size determination and evaluation of polydispersity for anisometric nano-objects such as CNCs. This question is particularly acute for large-scale quality control, for instance, in pilot plant production and beyond.
1.3.2. Toxicology assessment
Manufactured nanomaterials have recently caused societal concerns about their possible adverse effects on health and safety. Properties of nanomaterials typically differ from those of their parent bulk materials because of their larger surface area, leading to a greater activity, their smaller size, resulting in their ability to cross natural barriers, and intrinsic effects caused by nanometric size, including electronic and plasmonic effects. Since such particles may have a negative effect on biological systems and ecosystems, their toxicological risks must be evaluated and an accurate description of the product in terms of dimensions, chemistry, and toxicity is required by the authorities. CNCs have extensively been evaluated using standard ecotoxicological and mammalian test protocols and have, to date, been shown to be practically nontoxic in each of the individual tests . In addition, CNCs have recently obtained regulatory approval under Canada’s New Substances Notification Regulations (NSNR) for unrestricted use in Canada and is the first organic nanomaterial to be added to Canada’s domestic substance list.
1.3.3. R&D and applications
To maximize the reinforcing or liquid crystal behavior potential of CNCs, the particles have to be as well-dispersed as possible, especially in nanocomposite applications. An aspect that can lead to aggregation/agglomeration and then affect the further redispersion of the particles is mainly the final drying stage. Drying is a critical process for the large-scale industrial production of CNCs. Aqueous suspension leads to bacterial contamination and precludes long-term storage. In addition, the cost associated to the transportation of suspensions containing up to 90% water is not economically viable. Various drying processes can be used such as freeze-drying, supercritical drying, spray-drying, oven drying, and air drying. Spray-drying methods are used in the industrial production of CNCs. The resulting solid, hornified macrostructure may be difficult to further redisperse, even in water. Neutralization of the acidic sulfate ester groups prior to drying helps with redispersion .
1.3.4. Challenges regarding CNC particle size measurement
The aspect ratio of CNCs dictates the percolation threshold, a key parameter to control mechanical properties in nanocomposites. Knowing the size and morphology of CNCs with precision plays a key role in the development of numerous applications where these features directly impact the properties of the final product. CNCs prepared by sulfuric acid hydrolysis of cellulose substrates are rigid, acicular-shaped and highly crystalline nanoparticles. The geometrical dimensions of these rodlike crystallites vary exceedingly with the source of cellulose (Table 1) and with the hydrolysis reaction conditions. For example, CNCs extracted from wood are 3-7 nm in width and 100-200 nm in length, while CNCs derived from tunicate are 10-20 nm in width and 500-2000 nm in length . These values are indicative as different reaction conditions result in different sizes and size distributions. Indeed, size distributions were published as early as 1944 for ramie and cotton CNCs . Lists of the different sizes of CNCs extracted from various sources can be found in recent reviews [17, 18, 20, 21] and in Table 1. The size distribution may have an impact on the properties and thus the applications of CNCs. For instance, it has been shown that polydispersity influenced the phase separation behavior of liquid crystalline suspensions [51, 52]. CNCs are usually obtained as colloidal aqueous suspensions and the production process induces batch-to-batch particle size variability. The rheological properties of the suspension and the state of individualization of the particles are strongly affected by external parameters such as nanoparticle concentration, pH, ionic strength, temperature, or the presence of an additional compound or impurities [53-55]. The stability of the suspension is due to the electrostatic repulsion forces created by the negatively charged sulfate ester groups located at the surface of the crystals. More generally, their propensity to agglomerate is driven by their surface chemistry. For example, the addition of electrolyte screens the surface charges of the particles, reducing the electrostatic repulsion that prevents CNCs from agglomerating, which results in either an increase or a decrease of the measured particle size by photon correlation spectroscopy (PCS) depending on the amount of electrolyte introduced. Since CNC particles do not undergo swelling or compaction upon the addition of electrolyte, this increase/decrease of the measured particle size is a pure artifact driven only by the laws of physics and thermodynamics. Consequently, for a given sample, particle size variability depends on the sample preparation conditions and the measurement techniques.
Particle size is a good indicator of the quality of the CNC dispersion but the direct observation of nanoparticles remains a challenge and high-resolution direct imaging or light scattering techniques are required.
1.4. Microscopy and spectroscopy techniques used for CNC characterization
As previously mentioned, CNCs come in a wide variety of length, width, and shape depending on the cellulose sources. The cross sections of CNCs also display a variety of shapes, e.g., square, rectangular, or parallelogram, that are dictated by the arrangements of enzymatic terminal complexes extruding cellulose chains during the biosynthesis. As shown in different cases, the edges of the crystals can be eroded during the acid hydrolysis yielding hexagonal or octagonal cross sections exposing small surface area of hydrophobic (200) planes .
Many instruments are commonly used for the determination of particle size distribution (PSD) and particle dimension(s), each instrument detecting size through its own physical principle. Numerous techniques based on microscopy, light interactions, electrical properties, sedimentation, sorting, and classification allow access to particle size. Depending on the technique, the results are more or less accurate and are related to the nanoparticle shape and its physicochemical features such as chemical composition, heterogeneity, topography, surface charge density, dispersing medium, viscosity, etc. The morphology of CNCs can be accurately revealed either by microscopic methods including transmission electron microscopy (TEM), cryo-TEM, atomic force microscopy (AFM), field emission gun scanning electron microscopy (FEG-SEM), or by scattering techniques such as polarized and depolarized light scattering (DLS and DDLS, respectively) and small- and wide-angle neutron or X-ray scattering (SANS, WANS, SAXS, and WAXS, respectively). Each of these techniques has its own advantages and limitations in their application to the study of CNCs. Consequently, care should be taken while comparing particle size data obtained from different techniques.
Light scattering techniques rely on the interaction of light with the hydrodynamic volume of a particle. With anisometric nanoparticles, the response varies as a function of the orientation of each individual particle. The collected data thus need to be mathematically processed to extract meaningful information. Typically, for acicular particles, such as CNCs, light scattering techniques have not been able to match the precision of microscopy. Microscopy, on the other hand, provides direct images of individual particles and allows characterization of its morphology and size (length, width and thickness). Microscopy techniques which have the nanometer scale resolution capabilities to image CNCs are electron microscopy and AFM. Electron microscopy enables the direct observation of the dimensions (i.e., length and width) of a given particle. AFM provides information on morphology, surface topography, mechanical properties, and adhesion of CNCs under ambient conditions [15, 16, 57, 58]. While AFM provides reliable information on the thickness of the particles deposited on a flat substrate, the lateral resolution is limited by the convolution of the tip end, whose size and curvature are significantly larger than the dimensions of the nano-objects. However, this so-called tip-broadening effect can sometimes be deconvoluted . On the contrary, TEM images provide good nanometric (and often subnanometric) lateral resolution, allowing to rapidly screen a large population of particles, thus avoiding major sampling issues. However, as TEM images are projections of the objects along the incident beam direction, it may be difficult to accurately measure the particle thickness. Moreover, the low density of CNCs calls for the use of staining methods. These limitations may be overcome by some of the recent developments described in the following sections, including low-dose microscopy and 3D imaging. When possible, the combination of data from AFM and TEM images, and scattering analyses provides an optimal description of the particle morphology. For instance, a combination of imaging and scattering data was used by Elazzouzi-Hafraoui et al.  and Brito et al.  to precisely describe the morphology of CNCs extracted from a variety of sources.
It has to be noted that due to significant technical progress in instrumentation (low-voltage and beam-decelerated imaging, variable pressure) and detector variety and sensitivity, FEG-SEM imaging has become a perfectly valid approach to visualize the surface of CNC assemblies from dried systems  or fractured chiral nematic films . CNC suspensions dried on TEM grids can be observed as well in an SEM equipped with a detector located below the specimen. With this so-called scanning transmission electron microscopy (STEM) mode, transmitted and scattered electrons can be collected to reconstitute an image of the specimen very similar to those obtained with a TEM. Although the accelerating voltage is lower (10-30 kV) than that used in a TEM (100-300 kV), the resolution is sufficient to see the fine details of CNCs .
This chapter focuses on TEM approaches. We describe several methods to prepare CNC specimens and emphasize the specific conditions to observe these highly beam-sensitive nanoparticles in order to provide images of CNC dispersions and determine reliable size distributions. A thorough description of sample preparation procedures and observation techniques is followed by a review of the literature on CNC imaging by TEM.
2. TEM techniques used for the observation of CNCs
2.1. Selected milestones in the characterization of nanocellulose by TEM
TEM imaging and electron diffraction played a significant role in the morphological and structural studies of cellulose microfibrils and nanocrystals, complementing data from spectroscopic (Fourier-transform infrared spectroscopy and solid-state NMR) and scattering (of X-rays or neutrons) analyses. Important information was obtained at a local scale, allowing to identify mechanisms at work during the biosynthesis and biodegradation of cellulose crystals and characterize some physical properties, with or without additional chemical modification, inside cell walls or, after extraction, in suspension or incorporated in nanocomposite materials.
In 1940, H. Ruska (brother of E. Ruska, inventor of the TEM) published what can be considered to be the first images of microfibrils obtained after HCl hydrolysis of cotton cellulose . During the following two decades, the progress in TEM imaging of cellulose was mostly driven by the motivation to characterize the submicrometer structure of natural cellulose fibers. However, the identification of smaller constituting elements required the disintegration of cell walls and fibers using strong mechanical and/or chemical treatment [26, 64-68]. In particular, images of individual crystalline “fragments” (not yet called CNCs) were recorded after strong sulfuric acid hydrolysis of celluloses from various sources [9, 10, 69, 70].
In parallel to these morphological studies, important structural information was obtained using electron diffraction. Fiber diffraction patterns were collected from Valonia cell wall fragments and microfibrils, at room  and low  temperature, confirming at a local scale the longitudinal orientation of the cellulose chains. The size and distribution of crystallites were determined in Valonia microfibrils using diffraction contrast imaging . Later on, electron diffraction was notably used to validate the existence of separate regions of Iα and Iβ allomorphs of cellulose in microcrystals from Microdictyon cell wall fragments . TEM imaging and diffraction methods were also combined to demonstrate that the biosynthesis of bacterial cellulose ribbons occurred by addition of the cellulose precursors at the nonreducing end of the chains . A further striking demonstration of the molecular order in cellulose crystallites was provided when Sugiyama et al. successfully recorded the first high-resolution images directly showing longitudinal and axial projections of the crystal lattice of Valonia microfibrils [76, 77].
2.2. Sample preparation
The sample for TEM imaging has to be extremely thin in order to be transparent to electrons. Sample preparation is thus a crucial step. The limiting thickness depends on the atomic number and density of the observed material and on the energy of the incident electrons. Typically, for polymers, the thickness should remain well below 1 µm. For bulk materials, preparing such a thin specimen requires specific sectioning procedures. However, as the thickness of individual CNCs is well below this limit, the preparation of TEM specimens, for instance from dilute suspensions, is a lot easier.
2.2.1. Grids, supporting films and homogeneous distributions of nano-objects
In order to be observed by TEM, nanoparticle suspensions are deposited on 3.05-mm-large thin circular metallic grids with typical meshes around a few tens of micrometers, which will be placed in the sample holder for microscopy. There is a huge variety of TEM grids available and the selection of the right kind of grid is a crucial step for acquiring good images. Copper grids are the cheapest and the most widely used. However, when the pH of the suspension is low or high, copper can be degraded, resulting in the formation of artefactual crystallites or dendrites upon drying. In that case, gold or nickel grids have to be preferred. Carbon is by far the most commonly used supporting film. Carbon-coated grids are available commercially but carbon films can also be prepared in the laboratory by evaporating carbon on sacrificial collodion films or cleaved mica. The resulting films are then “fished” on copper grids. Carbon films have a typical thickness of 5-20 nm, depending on whether transparency or mechanical stability is favored. Formvar is also used as supporting film but it is less electron-transparent than carbon and can drift when heated by the electron beam at high magnification.
The main challenge in preparing a specimen suitable for TEM observation is to ensure a satisfying dispersion on the various objects. This is particularly critical when one wants to determine the size distribution of a population of CNCs. The supporting carbon film is initially rather hydrophobic, so the deposition of aqueous suspensions and subsequent air-drying generally result in locally accumulated material. Glow discharge is considered to be the easiest and most efficient pretreatment of carbon films before sample deposition. The TEM grids then are placed in the chamber of a glow discharge unit (e.g., automated commercial systems like easiGlow from Pelco or ELMO from Agar Scientific). A pumping system allows to reach a low pressure and the residual air is ionized. The grids are submitted to this mild plasma cleaning during a few seconds, which results in a hydrophilic carbon surface (at least during several minutes) onto which aqueous suspensions easily spread. As charges are also generated on carbon during the treatment, the nano-objects tend to adsorb on the surface. The excess of liquid can thus be gently blotted out without waiting for complete drying and the objects should be homogenously distributed on the grid. An example of a distribution of cotton CNCs prepared after glow discharge of the carbon film is shown in Figure 3a. The factory settings of the commercial systems generally promote negative charges and hydrophilic surfaces but, depending on the applications and systems, positive charges and/or hydrophobic surfaces can be produced using additional agents.
Different materials have been tested as supporting films. While hydrophobic formvar promoted aggregation (Figure 4a), satisfactory dispersions of individual CNCs were achieved on hydrophilic silicon monoxide (Figure 4b) . Silica (SiO2) and silicon nitride (Si3N4) films, with or without additional functionalization, are also commercially available (NanoGrids from Dune Sciences).
In addition to the choice of supporting film and the use of an additional pretreatment, the use of very dilute suspensions is recommended, typically 0.001-0.01 % (w/v) for CNC suspensions. The nanocrystals will adsorb more efficiently onto hydrophilic films and particularly those that have been glow-discharged and this effect can be compensated by a higher dilution.
2.2.2. Impact of the CNC suspension on the TEM observation
After acid hydrolysis, CNCs occur in the form of suspensions that generally require further purification steps to remove the residual acid and salts. Indeed, residual soluble salts will likely crystallize upon drying on the supporting film of the TEM grid, resulting in the formation of artifactual dendrites. The suspensions are dialyzed against water to reach a pH around 2.5-3. Then CNCs may be used as such – i.e., in the form of an acidic aqueous suspension – or may be neutralized with NaOH to reach pH~7, and be used in the form of neutral aqueous suspension . The drying step of an acidic CNC suspension prevents the CNCs to be later redispersed in water because CNC aggregates are created by hydrogen bonds via a phenomenon called hornification. Adjusting the suspension to a neutral pH prior to drying allows the dried CNCs to be spontaneously redispersed in water, while the specimens dried from acid pH suspensions cannot be properly resuspended .
The impact of the drying step of the CNCs prior to resuspension has been demonstrated by comparing three TEM specimens: 1) “never-dried” CNCs – CNC suspensions resulting directly from cellulose hydrolysis, 2) “freeze-dried” CNCs – CNC suspensions made from dispersing pH-neutralized freeze-dried CNCs, and 3) “spay-dried” CNCs – CNCs suspension made from dispersing pH-neutralized spray-dried CNCs . For all three specimens, the pH of the redispersed suspension was adjusted to 5-6 and the suspensions were deposited on carbon-coated grids. As seen in Figure 5, dispersions from spray-dried samples featured more CNC aggregates than those prepared from freeze-dried and never-dried samples. More single rods were observed in the latter than in any of the other two. On Figure 5a, the typical width of the observed bundles in the three samples is marked and is indicative of the degree of aggregation. This conclusion is actually consistent with the fact that spray-dried CNCs are more difficult to disperse than freeze-dried or never-dried counterparts . In addition, a mushroom-like morphology has been exclusively reported for the spray-dried CNCs, which look like patches, with a size ranging from 100 to 200 nm, with a much higher contrast than the rods [79, 80]. However, this morphology was proved to be an artifact from improper dispersion of the sample on the TEM grid . Additionally, the suspensions of previously dried CNCs could be briefly sonicated to promote greater dispersion before preparing TEM grids.
The pH of CNC suspensions before TEM sample preparation has a clear influence on the CNC dispersion on the grid. When CNCs are redispersed in dionized water, the pH is around 5-6. If never-dried CNCs are observed at this pH, on carbon-coated TEM grids, the images show both individual and aggregated CNCs (Figure 6). Theimproved CNC dispersion on the grid is obtained at pH 3.5. Indeed, at pH below 4-5, CNCs are more negatively charged as the half-sulfate ester groups are in their acidic form.
The particle morphology and aggregation state observed by TEM depends on how the dried CNC sample has been redispersed in water (use of sonication to promote de-agglomeration), which in turn depends upon the drying history of the CNCs (i.e., the pH of the suspension before drying). The optimal conditions to image CNCs is to prepare a TEM specimen from a suspension of never-dried CNCs or a pH-neutralized suspension.
2.2.3. Contrast enhancement techniques
18.104.22.168. Negative staining
Since the contrast of cellulose specimens is rather low, and since using high magnification to see more details will result in a rapid degradation of the particles (hence a loss of contrast), several techniques may be used to enhance the contrast in the images. The most widely used is commonly called negative staining. A drop of an aqueous solution of heavy atom salts is deposited on the specimen. Upon drying, a thin layer of heavy atoms concentrate around the nanoparticles, creating an electron dense outline . The CNCs thus appear as clear objects on a dark background, hence the negative effect on the image. The preparations can be observed at a higher magnification as, although cellulose is indeed damaged by the electron beam, the heavy atom “imprint” is resistant, revealing fine details of the surface topography. So far, 2% (w/v) uranyl acetate solutions have been the most commonly used stain but phototungstic acid or ammonium molybdate are sometimes used as well. Uranyl acetate will likely disappear from the catalogues as recent regulations have been enforced regarding the handling of radioactive products, resulting in a significant price increase. Consequently, new commercial ready-to-use stain solutions have been recently proposed, such as Uranyless (that contains lanthanides, from Delta Microscopies) or NanoVan (methylamine vanadate, from Nanoprobes). Practically, a homogenous negative staining of CNCs will be achieved under two conditions: i) the supporting carbon film has to be glow-discharged before the nanoparticles and the stain are deposited; ii) the negative stain has to be deposited on the specimen before drying. After a few minutes, the stain in excess can be blotted away with filter paper and the residual thin stain film allowed to dry. It has to be noted that negative staining can partly promote the local flocculation of the CNCs on the supporting film, as can be seen in Figure 3b.
22.214.171.124. Metal shadowing
The technique has been used very early on to observe nanocellulose. Indeed, in the 1950s, Rånby published images of metal-shadowed preparations of wood and cotton microfibrils , as well as tunicate  and bacterial CNCs . Images of ramie and cotton CNCs can also be found in Mukherjee and Woods’ paper . Metal (generally gold/palladium or platinum) is vaporized on the sample with a given incidence angle. It thus accumulates on one side of the nanoparticles (electron-dense region) and is absent on the other side (electron-transparent region). This results in a shadow-and-light effect that enhances the topography details of the specimen with a very high contrast. Again, it is the metallic layer that is observed even if the cellulose particles are damaged by the incident electron beam. The resolution is limited by the granularity and thickness of the metal film that increases the apparent width of the particles.
In the specific case of concentrated CNC suspensions, freeze-fracture can be used. A drop of the suspension is fast frozen and, under vacuum, a sharp knife breaks the frozen specimen into two fragments. The resulting surfaces are then directionally shadowed with a thin layer of evaporated metal, consolidated with an additional layer of carbon. The specimen is then warmed up and the metallic replica washed, deposited on a carbon-coated grid, and observed in the TEM, providing a high-contrast image of the topography of the fractured surface. An example of freeze-fracture replica of a concentrated suspension of cotton CNCs is shown in Figure 7a.
Cryo-TEM consists in observing nanoparticles embedded in a thin film of transparent vitreous ice, thus avoiding the detrimental effects of drying and/or staining (degradation, deformation, aggregation, agglomeration, coalescence, uniplanar orientation, buffer salt crystallization, etc.). This technique is particularly helpful with dilute suspensions of nano-objects that are soft or liquid at room temperature and whose morphology or structure may be affected by air-drying (deformation due to capillary forces, decrystallization). Droplets of suspensions are typically deposited on holey or lacey (e.g., NetMesh from Pelco) carbon films supported by TEM copper grids. Perforated support foils with predefined hole size and arrangements (e.g., Quantifoil from Quantifoil Micro Tools or C-flat from Protochips) can also be used. The liquid in excess is blotted with filter paper and the thin remaining film is quench-frozen in liquid ethane . The frozen specimen is mounted in a cryo-specimen holder precooled with liquid nitrogen, transferred in the microscope, and maintained at low temperature during TEM observation. Temperature- and humidity-controlled chambers with automated plungers are commercially available (Vitrobot from FEI, EM-GP from Leica or Cryoplunge from Gatan), ensuring a higher reproducibility of the fast-freezing procedure. Cryo-TEM is not limited to aqueous suspensions and organic solvents can also be used, provided that they can be properly vitrified and that they are not dissolved in liquid ethane.
In the case of CNCs which are rigid crystalline nanoparticles, using cryo-TEM is helpful to prevent the aggregation that may happen upon drying, resulting in unwanted particle superimposition. Indeed, cryo-TEM has only been used in a small number of cases, to observe unstained nonflocculating cotton CNCs (Figure 7b) [60, 82] and characterize by electron tomography (see § 2.3.3) the 3D shape of CNCs wrapped with dendronized-polymers  or covered with Pd patches .
This preparation technique should be used for bulk materials like plant tissues, cell wall fragments, nanocomposite materials, fibers, etc. When the specimen is affected by air-drying, it has to be first fixed in paraformaldehyde/glutaraldehyde, post-fixed with osmium tetroxide, dehydrated by exchange with ethanol and embedded in hardening resins (Epon, LR White, Spurr, etc.) . In the case of nanocomposite materials incorporating cellulose microfibrils (CMFs) or CNCs as reinforcing fillers, small fragments can be used without resin-embedding, provided that the material is sufficiently hard. Ultrathin (50-150 nm) sections of the specimens are then cut at room temperature with a diamond knife in an ultramicrotome . The sections are collected on bare or carbon-coated TEM grids and, if necessary, in order to enhance contrast, may be post-stained with uranyl acetate/lead citrate. When the sample is soft at room temperature, ultrasectioning has to be performed in cryogenic conditions with a dedicated unit.
2.3. Observation techniques
Beside the constraints directly related to sample preparation (drying, staining, etc.), the TEM observation of CMFs and CNCs should result in the recording of images with a good signal-to-noise ratio and showing enough fine details of the objects. The various contrasts observed in the images are generated by physical interactions of the incident electrons with the material  and these contrasts have to be properly recorded by a sensitive “detector.” Although for many years, micrographs were recorded on films, those have been progressively replaced by digital cameras typically equipped with a 1k×1k, 2k×2k or 4k×4k pixel-large CCD or CMOS detector. These cameras have wide dynamics, good linearity, and are very sensitive, allowing to record images with short exposure times and low electron doses. In addition, the software that pilots the camera can process the signal in real time, allowing, for instance, to calculate Fourier transform and correct some aberrations.
Three main phenomena contribute to the overall contrast of CNCs in an image. Mass/thickness contrast (also called amplitude or diffusion contrast) is generally low for polymer particles as they are mostly composed of light elements that weakly scatter electrons. The proportion of electrons transmitted and scattered by the specimen depends on its density and thickness. By inserting an aperture located in the back focal plane of the objective lens, the operator blocks a certain amount of scattered electrons and form an image with the transmitted (or weakly scattered) electrons. In this so-called bright field image, the dark regions thus correspond to the ones scattering electrons the most. As CMFs and CNCs are generally very thin, they generate a low amplitude contrast which, as described in § 2.2.3, can be significantly enhanced by using specific staining techniques involving heavy atoms. However, for unstained preparations, large differences in contrast can be observed in hybrid organic-inorganic systems, allowing, for instance, to locate metallic or oxide nanoclusters distributed along cellulose nanoparticles .
Diffraction contrast occurs when the specimen is crystalline or semicrystalline, independently of its chemical nature (light or heavy atoms). Depending on the orientation of the crystal with respect to the incident beam, the electrons will be diffracted away from the optical axis. The set of discrete diffraction angles is defined by the unit cell of the crystal and can be determined using Bragg’s relation . When a diffracted beam is stopped by the objective lens aperture, in images with negative defocus, the regions of the particle from which the diffracted beam originates are very dark against the clear amorphous carbon background. This effect is thus important to visualize CNCs in bright field mode as it compensates somewhat the low amplitude contrast. It depends on the crystallinity index of cellulose (low for wood and high for tunicate CNCs, for instance). However, as cellulose is highly sensitive to beam damage, the diffraction contrast lasts for a limited time and disappears as soon as the material becomes amorphous.
Phase contrast is crucial in the case of electron-transparent nano-objects. It results from sharp differences in scattering properties between regions of the specimen. This is the case for CNCs spread on a carbon film (particles in vacuum), and CNC suspensions observed by cryo-TEM (particles embedded in vitreous ice). Phase contrast generates interference Fresnel fringes whose amplitude and distribution depend on the defocus of the objective lens . The effect of phase contrast on the image of unstained CNCs is illustrated in Figure 8a and 8c. High defocus values (positive or negative) increase contrast but also generate larger Fresnel fringes and increase the apparent size of the nanoparticles. Note the complete reverse of contrast between underfocused and overfocused images. In overfocused images, the CNCs are clear with a dark outline which reminds the aspect they would have after negative staining. Close to zero defocus, the contrast is minimal and the specimen is hardly visible. Paradoxically, the operator will be satisfied by images recorded with some defocus as the contrast is higher and as the Fresnel fringes around the particles create an impression of “sharpness.” However, this effect is associated with a loss of details. A satisfactory image is thus obtained by balancing the opposing requirements of contrast and ultrastructural details. By convention, the images are recorded in underfocused conditions, the amount of applied defocus increasing with decreasing magnification (typically about -5 µm at 3000x and -1 µm at 10000x). The importance of using underfocused conditions is clear in the case of negatively stained specimens, as seen in Figure 8b.
In the routine observation of unstained CNCs, three types of contrast contribute: a weak amplitude contrast due to the small size and organic nature of the particles; a significant diffraction contrast that depends on the crystallinity index of cellulose and orientation of the particles with respect to the incident electron beam; a phase contrast that generate Fresnel fringes along the CNCs controlled by the amount of defocus. As detailed in the following section, all these contributions will be affected by radiation damage.
2.3.2. Radiation damage
The most important limitation to observe crystalline polymers by TEM is the significant damage created by the electron beam which rapidly affects the resolution and decreases the contrast of the objects. In such low-density materials, the inelastic scattering of the electrons generates molecular excitations and ionization phenomena, resulting in local heating of the material, breaking of covalent bonds, diffusion of free radicals, and emission of volatile species . The consequences can be mass loss, fusion, vaporization, loss of crystallinity, all resulting in a significant decrease of amplitude, diffraction, and Fresnel contrasts. For crystalline materials, one can define a “lethal” or “total end point” dose by monitoring the disappearance of the electron diffraction pattern and calculating the electron dose at which crystallinity is irreversibly lost.
Even if the selected microscope has the ability to visualize details down to a resolution of 0.1-0.2 nm, the sample itself imposes drastic constraints. Several solutions exist to limit the detrimental effects of beam damage during the observation and image recording. First, contrary to what is done with materials made of heavy atoms, increasing the accelerating voltage of the electrons decreases the interactions with the polymer and thus increases the lethal dose (typically about a factor of 2 between 100 and 200 kV). However, the consequence of a higher accelerating voltage is a decrease in contrast. For instance, individual CNCs observed at 200 kV or a higher voltage tend to become transparent.
Second, the operator has to work with electron doses much lower than the lethal dose, which requires both to use low magnifications and significantly decrease the illumination (e.g., by changing the “spot size” and spreading the surface of illumination). The observation of unstained cellulose specimens thus requires for the operator to make a compromise between image magnification (to see fine details with a sufficient contrast) and electron dose (to have a sufficient signal-to-noise ratio for the detector recording the image). Chanzy has quantified the lethal dose of cellulose by submitting Valonia CMFs to increasing electron doses at various accelerating voltages and monitoring the decay of diffracted beam intensity . The average value for the lethal dose was about 4 electrons.Å-2, which illustrates the extreme sensitivity of polysaccharides.
Third, the radiation damage can be significantly slowed down (but not suppressed) when the diffusion of free radicals is reduced by maintaining the specimen at low temperature, using a specific specimen holder cooled down with liquid nitrogen. In that case, an increase of the lethal dose by a factor of 3 has been reported in the literature.
The images in Figure 9 qualitatively illustrate the cumulative effect of beam damage during the observation of CNCs. Cotton and tunicate CNCs (Figures 9a and 9b, respectively) have been submitted to increasing electron doses. The three contributions to contrast previously mentioned are affected: diffraction contrast by disruption of the crystallinity; amplitude contrast by mass loss; and phase contrast by thinning of the particle and surface damage. The CNCs become transparent and their shape less well-defined. These effects are very fast and irreversible, in particular if the specimen is readily observed at magnifications above 10000x or the beam is inadvertently focused into a smaller disk (Figure 9c).
2.3.3. 3D imaging
TEM images are two-dimensional (2D) projections of the 3D objects along the beam direction. Information along this direction is thus lost. One 2D image is generally not enough for the viewer to have a clear idea of the 3D shapes and several images, recorded at different tilt angles of the specimen, are necessary to make a reliable morphological analysis. Electron tomography (ET) can be performed in modern microscopes, thanks to the use of digital cameras and software that can precisely control the specimen orientation and image acquisition. Briefly, series of 2D images containing the data for mass-density distribution in the specimen are automatically recorded with small angular increments over a large angular range (usually from +70° to -70°). Using specific software, the collected images are aligned with respect to one another and a back-projection is carried out to obtain a 3D reconstruction of the specimen volume [88-90]. While this technique is now widely used to study the morphology and structure of biological systems (proteins, viruses, membranes, etc.), So far, only two cases of CNC analysis can be found in the literature and will be described in § 3.4 [44, 62]. As explained in § 2.3.2, unstained cellulose is highly sensitive to beam damage and will be rapidly degraded. The amplitude contrast will thus decrease during the recording of the tilt series. This necessitates working at low magnification, with extremely low electron doses and a sensitive digital camera. In addition, due to the crystalline nature of CNCs, the contribution of diffraction contrast may fluctuate depending on the tilt angle of the particles.
3. Review of cellulose nanocrystal imaging and size analysis
3.1. TEM images and size distributions of CNCs
Although AFM is still often used to perform size measurements of CNC populations [15, 16, 57-59], TEM has been a method of choice to visualize the shape and structure of individual or bundled CNCs [16, 60, 91, 92]. The images are consistent and show that CNCs are rodlike and have a high aspect ratio. Some authors have observed spherical CNC particles  but these have been shown to be aggregation-induced artifacts . As seen in Figure 2, the CNC aspect ratio and structure greatly vary with the cellulose source. CNCs obtained from plant sources like cotton [60, 93, 94], rice [95, 96], ramie [97-99], sisal [99, 100], or sugar beet  have smaller aspect ratios than those of bacterial cellulose [101, 102], microcrystalline cellulose (MCC) [51, 98, 103-106] and tunicin  (Table 1). The significant heterogeneity of the length and width distributions of CNCs, for a given source of cellulose, can be attributed to the method of preparation: type and concentration of the acid, reaction time, temperature, sonication, method of drying, etc.
To a first approximation, the size measurement of CNCs should not be particularly difficult, provided that the objects are fairly well individualized on the supporting films, not severely damaged by the electron beam (§ 2.3.2), and that a reasonable underfocus was used to record the image (§ 2.3.1). Their length and width can be measured from general views recorded at intermediate magnifications (typically 5000 to 10000×) by using a software like ImageJ . Automated procedures are generally difficult to implement with TEM micrographs as the preliminary binarization of the images requires to clearly distinguish the CNCs from a continuous background. As seen in many images presented in this article, this is rarely the case. Thus, one directly uses the measurement tools of the software or outlines each CNC so that a semiautomatic measurement routine can be run. In addition, a review of the literature shows that the number of counted particles greatly varies (typically from 100 to 1000), mostly depending on the number of individual particles in the images and the patience of the operator. Of course, the number of objects taken into account will influence the reliability of the population statistics.
When higher magnifications are used (typically, 10000 to 50000×), keeping in mind the potential beam damage of unstained specimens, a better resolution is achieved and more details are revealed, particularly after negative staining. As confirmed by many images found in the literature, the majority of CNCs prepared from various sources are not single crystallites. They are often composed of a few laterally joined rodlike crystallites, as illustrated by Figure 10 in the case of cotton CNCs. The same feature has been reported for tunicate , Avicel , or bamboo  CNCs, to give just a few examples. The images in Figure 10 clearly show that cotton CNCs have a high variability in shape and dimensions. They are all different in size and structure, being composed of different numbers of unit crystallites, joined in a different fashion. This association inside individual particles does not result from artifactual flocculation but rather from the fact that the hydrolysis was performed on dry cellulosic substrates in which microfibrils are strongly linked by hydrogen bonds. Neither acid hydrolysis nor sonication (nor any high-shearing device) could fully separate the constituting crystallites.
The composite nature of most CNCs, which was validated by a combination of imaging and scattering techniques [12, 60], will thus influence the measurement of the lateral dimensions which depend on the cross-sectional shape and its regularity along the whole particle, the existence of a longitudinal twist, and the specimen preparation method. Considering the acicular morphology of CNCs, their distribution on a supporting film or inside thin vitreous ice favors their planar orientation, allowing to easily measure the length of the rodlike nanoparticles. However, while the particle thickness can be accurately measured from tapping mode AFM images, a choice has to be made regarding the definition of the width in TEM micrographs. As many types of CNCs exhibit a spindle-like shape, the width is generally defined as the largest dimension perpendicular to the particle long axis. Moreover, when the CNCs are ribbon-like, because of the rectangular cross section of individual crystallites (tunicate CNCs, for instance) or the lateral association of crystallites (cotton CNCs), the drying on a supporting film will generate a strong uniplanarity, the particles tending to lie on their widest face. This propensity to uniplanarity, which brings very useful information of the crystallite orientation in the CNCs, can be easily evaluated by performing complementary WAXS analysis on air-dried films [12, 60]. In addition, some CNCs are longitudinally twisted, which can be detected in TEM images by variations in contrast or stain distribution along the particle (an example on tunicate CNCs can be seen in Figure 9b). Therefore, the apparent width of the particle will vary, making the choice of a relevant region to measure more difficult.
Cryo-TEM has been shown to be helpful in preventing particle aggregation/agglomeration due to drying [60, 82]. If the suspension concentration has been chosen with care, and considering the fact that CNCs have charges on their surface, they are conveniently individualized in the embedding film of vitreous ice. As their lateral dimensions are smaller than the typical thickness of the liquid film prior to fast-freezing, they could freely rotate around their long axis. Consequently, the measured width indeed corresponds to the projection of the CNC in a given orientation. Therefore, the width distribution of the population from a cryo-TEM image provides a rotational average of the lateral dimension of the particle and can thus be different from the distribution that is obtained from a specimen prepared by air-drying on a supporting film. The effect will be more pronounced for flat particles, such as cotton CNCs, as shown in Figure 10. Using a simple model, Elazzouzi-Hafraoui et al. have described the relation that can be made between the distributions obtained with both preparation techniques .
Determining a size distribution from TEM images thus implies clearly stating the choices that were made in relation with the particle morphology. In addition, length and width distribution can be determined independently (Figure 11a) [78, 82] or assuming that there is a correlation between longitudinal and lateral dimensions of the CNCs. In that case, length and width are tabulated for each particle, resulting in a 2D size distribution histogram (Figure 11b) which validates (or not) the correlation .
The relevance of the size distribution of a given population of CNCs depends on several factors in relation with the problem that is addressed or the specific property that is investigated. For acicular (and possibly ribbon-like) CNCs, the influence of the statistical errors on the measurements will differ depending on the considered dimension (smaller for the length and larger for the width and thickness). Moreover, the size distributions are always wide, with large standard deviations and asymmetrical shapes (Figure 11) [16, 82, 91, 92]. Average lengths, widths, and thicknesses provide a cylindrical average view of the particles that can be useful to build a preliminary model. However, it is likely that a more complete description of the population, made by fitting the data with suitable functions, will be more relevant for comparisons with X-ray or neutron scattering results , or, for example, to understand the percolation behavior of CNCs inside nanocomposite materials.
3.2. TEM of CNC-hybrid composites
Hybrid composites of CNC with metal nanoparticles (NPs) such as Ag [39, 46, 112], Au [39, 41], Cu , Pd [40, 44, 113], Pt [39, 114], or Se  have been reported. TEM, along with its analytical tools, was used to characterize such composites (Figure 12). Due to their high atomic number and possible crystallinity, the metal NPs appear as dark spots in underfocused images. The difficulty to observe such hybrid organic-inorganic systems comes from the fact that CNCs and metal NPs have very different sensitivity to beam damage. While metallic NPs are very stable, cellulose is rapidly damaged, resulting in a significant loss of contrast and recognizable morphology. Consequently, to have a better visibility, a large defocus is usually applied which causes an overestimation in the size of metal NPs . Additional negative staining can be effective to outline the CNCs more precisely but care must be taken to prevent the layer of stain from overshadowing the contrast of the metallic nanoclusters. Indeed, if the metallic NPs are smaller than 3-4 nm, they will not be distinguished from the granularity of the dry stain. As shown in Figure 12f, for larger NPs, CNCs can benefit from the negative staining while the metallic clusters still appears as dark spots. Finally, recent results demonstrated that highly sensitive direct detection device (DDD) cameras, associated with phase plates, allowed to observe simultaneously CNCs and subnanometer-thick palladium plates at their surface (Figure 13) . In this experiment, the organic and inorganic phases were observed together at near-focus conditions.
Alongside visualization of metal NPs, it is very important to validate the chemical nature of the NPs seen in the images. This analysis can be done in a TEM by using energy-dispersive X-ray spectroscopy (EDX) which allows for quantitative measurement of the chemical composition .
3.3. TEM of CNC-polymer nanocomposites
Due to their nanoscale dimensions, high surface area, low density, and high mechanical strength, CNCs have been incorporated as reinforcing component into various polymer matrices: styrene-butyl acrylate [14, 27] polylactic acid [116, 117], poly(vinylalcohol) , polyurethane [119, 120], epoxy , and polyvinylacetate [100, 122]. In order to explain the variation in mechanical properties, these bulk materials have often been characterized by SEM after fracturing, allowing to visualize the distribution of CNCs that emerge at the fracture surface. However, SEM remained limited to the characterization of the topography of the fracture surfaces. TEM was also used to visualize the CNCs inside the material via preparation by ultramicrotomy of ultrathin sections of the composites. Figure 14 shows examples of images of nanocomposites of CNCs in matrices of styrene-butyl acrylate  and polyurethane . CNCs can be visualized using their intrinsic diffraction contrast  but, as they are very sensitive to beam damage, their contrast rapidly decreases to the point where they become hardly visible with the polymer matrix due to the lack of density difference. Additional staining of the matrix can be used, for instance with OsO4, creating a negative staining effect revealing the embedded CNCs .
3.4. 3D imaging of CNCs and polymer-wrapped CNCs
Electron tomography (described in § 2.3.3) has only very recently begun to be used to characterize cellulose-based architectures or individual particles. Ciesielski et al. have recently published ET reconstructions of treated plant cell walls, allowing to get 3D information on the microfibril morphology . Majoinen et al. studied the morphology of individual CNCs embedded in vitreous ice. Using cryo-TEM conditions, they recorded tilt series and reconstructed tomograms from pristine CNCs (Figure 15a) and from CNCs wrapped up with dendronized polymers (Figure 15b). This 3D visualization of the CNCs revealed the presence of an helicoidal twist along the rodlike particles. These twists have been proposed to be significant in explaining the cholesteric liquid crystalline assembly [18, 19] and their use in chiral templating [32, 33]. More recently, tomograms of ice-embedded Pd modified CNCs were also useful to gain insight into the nature of the material under close-to-catalysis conditions .
4. Conclusion and perspectives: Challenges and solutions in imaging CNCs
CNCs have been extensively studied and many reports can be found in the literature. The particle shape and size vary as a function of the cellulose source and the methodology used to produce the CNCs. In all cases, the CNC populations are polydisperse, featuring wide and asymmetrical size distributions. An accurate knowledge of the CNC dimensions and size distribution is crucial to characterize the properties of individual particles and their assemblies in suspension, in dry films or after incorporation into matrices. For instance, morphology and size will influence the phase separation behavior and the formation of liquid crystalline organizations or the percolation properties in nanocomposites. The development of chemical modifications strategies also requires a good knowledge of the accessible surface. In addition, CNCs have recently become a new type of industrial nanomaterial with numerous exciting properties, and an accurate characterization is important to optimize processes and control their reproducibility.
TEM is a key imaging technique to achieve this goal, thanks to the development of specific techniques to observe beam-sensitive polymers. Cellulose is without doubt one of the most beam-sensitive materials that can be observed by TEM. Consequently, special care must be taken to record reliable images of CNCs (low dose, low temperature, additional staining). In this article, we have described several sample preparation procedures and observation techniques adapted to the recording of images of CNC populations and have carried out reliable size distribution analyses. New approaches have recently been proposed, taking advantage of technical improvements of the microscopes, such as a better control of the stage movements, the use of highly sensitive digital cameras, and automatized corrections of several optical aberrations, allowing to record electron tomography image series and reconstruct the volume of the specimens. Although these sophisticated approaches are not routinely used yet in the field of cellulose science, they will certainly rapidly develop. In addition, at a time when the research on CNCs is booming and the range of commercial applications is expanding, TEM imaging will continue to play a central role. Automated particle measurement procedures using TEM image analysis software will be needed to streamline quality control.
The TEM images in Figures 3, 8, and 9 were specifically recorded for this chapter by J.-L. Putaux on a Philips CM200 microscope operating at an accelerating voltage of 200 kV and equipped with a TVIPS TemCam F216 camera. We thank the Electron Microscopy Platform of Institut de Chimie Moléculaire de Grenoble for granting access to this equipment.