Introduction

Cellulose is a linear polymer composed of glucose units interconnected via the β-1,4 bond. In the cellulose structure, anhydrous d-glucose units are packed in a fibrillar arrangement with alternating amorphous and crystalline regions that form fibrils. The packing of the fibrils leads to the formation of cellulosic fibers that are commonly found in plants and bacteria (Habibi et al. 2010; Eichhorn 2011; de Oliveira et al. 2012).

Nanocellulose (NC) such as cellulose nanofibers (CNFs) or cellulose nanocrystals (CNCs) are obtained from the chemo mechanical degradation of cellulose fibers and are widely employed as reinforcement agents. CNFs are usually obtained through mechanical processing of cellulosic fibers, while CNCs are usually obtained by acid hydrolysis. CNFs show diameter ranging from 10 to 100 nm and length in the order of hundreds of microns. CNCs present a needle-like morphology with an approximate diameter of 5–10 nm and 100–500 nm length range (Siqueira et al. 2010).

Nanocellulose materials are renewable fillers with low density, high elastic modulus and stiffness that might be used in a wide range of polymer composite applications, such as coatings and gels, as well as sensing and optoelectronic devices (Habibi et al. 2010; Golmohammadi et al. 2017; Zanata et al. 2018). CNCs are particularly interesting fillers due to the substantial number of hydroxyl groups on the surface, which generate particle–particle hydrogen bonding and ultimately an interconnected CNC network within the polymer matrix. According to the percolation theory, CNC networks promote significant strengthening in polymer composites due to preferential load bearing and tensile dissipation (Dufresne 2008).

Particle–particle hydrogen bonding is also a useful strategy for the preparation of CNC-based gels and foams, which are of growing interest in the biomedical field (Capadona et al. 2007; Yang and Cranston 2014; Sapkota et al. 2014; Yang et al. 2015). However, CNC aggregation in solution is a major drawback of the high hydrogen bonding yield. For this reason, many studies have focused on the CNC surface modification to reduce the number of hydroxyl groups per area, which in turn reduces the strength of particle–particle interactions as well as increases compatibility with hydrophobic polymer matrices (Wang et al. 2007; Dufresne 2013; Eyley and Thielemans 2014; Yang and Cranston 2014).

CNC size and morphology are widely dependent on the cellulose source as well as extraction conditions. Since size and morphology are key aspects for the tailoring and prediction of CN properties, reaching reliable and reproducible methods to determine these parameters is of paramount importance (Azizi Samir et al. 2005; Kaushik et al. 2015). In particular, the determination of the CNC length, L, and the diameter, d, leads to the determination of the CNC aspect ratio (L/d), which gives the minimum particle concentration needed to reach gelation as well as the percolation network formation. Nevertheless, the shape determination cannot be overlooked, since it might have a profound influence over the estimation of the number of hydroxyl groups per surface area, as well as hydrophobic matrix compatibility. As an example, Lin and Dufresne (2014) have shown that CNCs with a hexagonal or octagonal cross-section are better dispersed in a hydrophobic polymer matrix than the remaining CNC structures due to the exposure of a hydrophobic (200) plane on the crystal edges.

Several characterization techniques might be used to determine CNC shape and size. The use of scattering techniques, such as small-angle neutron scattering (SANS), small-angle X-ray scattering (SAXS) and dynamic light scattering (DLS), was recently investigated by Mao and co-workers (Mao et al. 2017). Consistent information concerning particle cross-section was obtained from SANS and SAXS whereas length measurements cannot be obtained by these techniques. In contrast, DLS data can only provide apparent sizes and not actual length and diameter values (Foster et al. 2018).

As a consequence, microscopy techniques such as scanning (de Oliveira et al. 2012) and transmission (Taipina et al. 2013) electron microscopies (SEM and TEM) as well as atomic force microscopy (AFM) (Reifenberger et al. 2008) are generally better suited to the size and morphology determination of nanocellulose. AFM has a good thickness resolution, which is measured by particle height; however, it is a time-consuming technique with poor lateral resolution, which impairs length measurements. Moreover, AFM specimen preparation is also still challenging (Jakubek et al. 2018). SEM has the advantages of speed and reduced beam damage; however, it also has the disadvantage of poor resolution, even when a transmission detector (STEM mode) is considered (Jakubek et al. 2018). Cryo-TEM might prevent drying induced aggregation; however, it is an intrinsically low contrast technique due to the low dose requirements. In contrast, TEM is readily available, cost effective and allows fast analysis of nanocellulose in solution as well as embedded in soft matter. All of these advantages make TEM the most useful technique for CNC size and morphology characterization.

With all this in mind, over the last few years at LNNano EM facility (Leite et al. 2016; Pinheiro et al. 2017, 2019; Tibolla et al. 2017, 2018; Zanata et al. 2018; Ferreira et al. 2018; Germiniani et al. 2019) we found the need to optimize the uranyl acetate staining protocol for the TEM analysis of cellulose-based nanomaterials. Uranyl acetate staining is a widespread methodology in the biological field that is also used for the analysis of cellulose-based nanomaterials. Nevertheless, issues concerning particle aggregation and grid conditions have led to poor quality images and unreliable size measurements. Some authors have already discussed these issues. However, most of the studies available in the literature propose alternative, and sometimes expensive, solutions to these problems rather than concentrate on establishing an optimized uranyl acetate negative staining protocol that could be promptly used anywhere.

Ogawa and Putaux (2019) claim that the use of high-resolution cameras and diffraction imaging are necessary. This is in fact true if the goal is achieving crystallographic or detailed morphological information. However, it might be unnecessary for routine bright field TEM imaging and aspect ratio measurements. Stinson-Bagby et al. (2018) investigated in detail the possibility of using several stains, grids types and dispersing agents (surfactants). However, the authors end up with a protocol that requires expensive SiO2-coated grids and Bovine Serum Albumin. Both Jakubek et al. (2018) and Foster et al. (2018) rely solely on pointing out the variables to consider during specimen preparation but fail to provide straightforward guidelines. Jakubek et al. (2018) showed that aspect ratio measurements must be performed manually; however, the authors highlight the significant influence of analyst bias on the aspect ratio values.

Herein we introduce an optimization of the uranyl acetate staining protocol for the preparation of CNC specimens for TEM analyses as well as guidelines for aspect ratio measurements using conventional image processing software. The present method considers the calculated aspect ratio for each condition as a measurable parameter for protocol optimization. Dispersion state and concentration, glow discharge application, drying method, as well as the need to remove CNC suspension excess on the TEM grid were the main variables considered in this optimization. Images of the specimen under different preparation conditions were analyzed by TEM and length and width values were measured using Image J software and statistically compared. Finally, this protocol was validated using a commercial CNC sample as well as CNCs embedded in soft matter.

Results and discussion

Cellulose nanocrystals chemically isolated from cotton fiber de Oliveira et al. (2012) were chosen as a model specimen. Initially, we qualitatively evaluated the sample preparation variables that are usually considered in biology to determine if these were significant to CNC imaging. Variables such as uranyl acetate concentration, staining time, number of applications and others were found to be less significant. Therefore, we selected typical values used in negative staining protocols for biological samples: uranyl acetate concentration was fixed at 2% (w/v), staining time at 30 s and the number of applications was set at 2. Further experiments were concentrated solely on investigating the interaction between the five most significant variables, which were described in Table S1 (ESI). These variables, in fact, concern sample preparation and application as well as grid preparation rather than the staining itself. The detailed steps of the optimized protocol for the preparation of cellulose nanocrystal suspensions and TEM specimens are also described in ESI.

The first variable evaluated (Glow Discharge) was related to the use of a glow discharge treatment on the grid, prior to sample application, in which the surface of the grid is exposed to a plasma environment. Current values were selected according to the equipment manual recommendation. The second variable (Concentration) was the CNC suspension concentration, before applying it to the grid, which was in the 0.1–2 mg mL−1 range, according to previous work of our group (de Oliveira et al. 2012). The third variable (Grid Side) referred to the use of the shiny or dull side of a lacey ultrathin grid (TedPella Prod. # 01824) for sample application. The fourth variable (Dry) was related to the use (or not) of a filter paper to dry the specimen in between sample and stain applications. The fifth variable (Wash) indicated the use (or not) of water droplets to wash the grid before the drying step.

For each individual variable, all samples (see Table S1 in ESI) were prepared and analyzed on the same day using the preparation conditions shown in Table 1. Some duplicate or triplicate samples were prepared throughout the study. The protocol described in ESI for the measurement of CNC length diameter and aspect ratio using FIJI (ImageJ®) (Schindelin et al. 2012) provides quantitative parameters for protocol optimization. Once all variables were optimized, a new sample with unknown aspect ratio was prepared and analyzed in triplicate on different days. T-tests were performed on these replicate pairs and showed that the average length, diameter and aspect ratios were statistically equal, which validated this protocol.

Table 1 Specimen preparation conditions in which each micrograph in Fig. 1 was taken
Full size table

Images of the negatively stained samples described in Table 1 are shown in Fig. 1. The average length and diameters measured for each sample, as well as their respective aspect ratios are shown in Table 2. Initially, the effect of the glow discharge treatment on the CNC spreading and attachment on the TEM grid was investigated. Glow discharge with a 15 mA (Fig. 1a) or − 15 mA (Fig. 1a) electrical current were applied on TEM grids and, subsequently, a drop of a 2 mg mL−1 CNC suspension was deposited. The results were compared to the control (0 mA, Fig. 1b) and showed that the glow discharge significantly improved nanoparticle dispersion. On the glow discharged grids, isolated and well stained CNCs are observed, while on the untreated grid accumulated and poorly stained material is verified. Another micrograph of sample b is presented in ESI to highlight CNC aggregation.

Fig. 1
figure 1

Representative TEM micrographs showing each parameter effect on the image quality. Samples a–c show the effect of glow discharge (a = -15 mA, b = none, c = 15 mA). Samples d and e show CNC concentration effect (d = 2 mg mL−1, e = 0.1 mg mL−1). Samples f and g show grid side effect (f = shiny, g = dull); h and i show the dry effect (h = filter paper; i = none) and finally, samples j and k show the wash effect (j = x3; k = none). Sample preparation conditions are shown in Table 1, while the average length, diameter and aspect ratio are shown in Table 2

Full size image
Table 2 CNC aspect ratio and dimension values as well as number of particles measured (counts)
Full size table

These results are expected, since glow discharge is a well-known method to improve the wettability by aqueous suspensions of the hydrophobic ultrathin carbon film on the TEM grids. The improved wettability is obtained from the effect of the air plasma generated inside the glow discharge chamber on the ultrathin carbon coating. This plasma creates temporary charges on the ultrathin surface, which improves water wettability. These charges also allow hydrophilic particles, such as CNCs, to interact with the surface and retain their dispersed state upon drying, without accumulating when the excess liquid is blotted out (Kaushik et al. 2015; Grassucci et al. 2007). In the glow discharge equipment, a positive or negative electron charge indicates the discharge head polarity, which does not dictate the net surface charge. Both positive and negative electrical charge provided good CNC dispersion. However, the use of positive charge resulted in better staining.

CNC suspension concentration was also investigated (Fig. 1d, e). Literature reports the use of different concentration values for the TEM analysis of CNC suspensions (Mao et al. 2017; Lu and Hsieh 2010; Oechsle et al. 2018). It is a common agreement that the lower the concentration, the better the CNC dispersion on the TEM grid. Figure 1d shows a micrograph of a micrograph of a 0.1 mg mL−1 CNC suspension, both of which were prepared without the use of glow discharge. Even though the more concentrated sample has a significant number of visible particles, the CNC morphology is not distinguishable due to the high particle density and poor staining. As already reported, (Kvien et al. 2005) CNCs tend to be aggregated in parallel; consequently, the staining molecules are not restricted to the CNC edges, covering most part of the aggregates as a continuous film. This phenomenon is detrimental to the contrast enhancement, which in this case was so severe that no particles could be measured, as shown in Fig. 2. Conversely, a good number of particles are visible and well stained on the diluted sample. In sample e, more than 300 individual CNC particles were identified and measured (See Table 2 and Fig. 2). This large number of particles led to a representative aspect ratio value. These results show that the combination of a concentrated suspension with an untreated grid is the worst-case scenario for CNC specimen preparation.

Fig. 2
figure 2

Particle counting (top left), CNC average aspect ratio (top right), length (bottom left) and diameter (bottom right) for each sample (sample description shown in Table 1), showing the effect of glow discharge conditions (purple); CNC concentration (gray); grid side (orange); drying (green) and washing (blue) methods. Samples marked by an asterisk are statistically equal

Full size image

The TEM grid side was also evaluated during protocol optimization. Figure 1f shows a sample prepared by the deposition of a 0.1 mg mL−1 CNC suspension on the shiny side of the grid, while Fig. 1g shows the deposition on the dull side of the grid. It is easily observed that sample deposition on the dull side of the grid provides better staining and particle dispersion. The explanation for this is the fact that the ultrathin carbon support film is applied on the shiny side of the grid. Therefore, if the suspension (or dye) is applied on the dull side, it is constrained by the copper mesh, which allows better contact with the surface, leading to better results. This same principle could also be applied to handmade carbon support grids or even cryoultramicrotomed sections. However, the use of handmade carbon coated grids could be detrimental to image contrast, since the usually thicker support film increases the background noise.

To further illustrate the grid side effect, ultrathin sections (≈ 60 nm thick) of a polycaprolactone/CNC nanocomposite were cryoultramicrotomed and deposited on the shiny side of an empty TEM grid. To reveal CNC distribution within the polymer matrix, the ultrathin sections were stained from the ultrathin section side (Fig. 3a) and also from the opposite side (Fig. 3b). In both cases, a glow discharge treatment was performed immediately prior to staining, which was carried out analogously to the CNC suspensions: two 30 s applications of a 2% (w/v) uranyl acetate solution. At this point it might be interesting to comment that the glow discharge or UV treatment might be performed on either side of the grid and will increase the water wettability on both sides equally.

Fig. 3
figure 3

TEM micrographs of cellulose-based nanomaterials embedded in polymer matrix: melt-extruded polycaprolactone matrix with 5 wt % CNC and stained from the section side (a) or the opposite side (b); solvent-cast polycaprolactone matrix with 25 wt % CNC (c) (Germiniani et al. 2019); photopolymerized poly(acrylic acid) matrix with 5 wt % CNC (d)

Full size image

The micrographs in Fig. 3 show that the morphology definition of CNCs embedded in polymer matrix is significantly improved when the uranyl acetate solution is deposited on the opposite grid side to which the ultrathin section was deposited (Fig. 3b). Individual CNC particles with sharp edges are observed in Fig. 3b, while in Fig. 3a only the diffuse edges of CNC aggregates are distinguished. The contrast enhancement in Fig. 3b is a result of the improved contact (or interaction) between the staining solution and the polymer surface, which in turn is promoted by damming of the solution by the copper mesh.

In this context, it is also interesting to comment that CNC nanocomposites should be preferentially analyzed on the TEM a few days after staining, in order to allow uranyl acetate diffusion through the polymer bulk. The micrographs in Fig. 3c, d were taken 7 days after staining and illustrate well the effect of uranyl acetate diffusion on hydrophobic (Fig. 3c) and hydrophilic (Fig. 3d) polymer matrices. In the hydrophobic polymer matrix (polycaprolactone), CNCs appear as dark needle-like structures due to the uranyl acetate migration to the hydroxyl-rich CNC surface. In the hydrophilic polymer matrix, poly(acrylic acid), the overall matrix becomes darker, since uranyl acetate mostly stains the matrix.

Once it was established that diluted CNC suspensions should be deposited on the dull side of a glow discharged grid (positive electron current), the effect of sample drying and washing prior to stain application was investigated. Typically, after sample deposition the excess liquid is removed by using filter paper and, afterwards, the grid is washed a few times with water to remove contaminants. In some cases, this procedure could have a significant impact on the particle size measurement due to the removal of smaller particles. To investigate this effect, filter paper dried (Fig. 1h) samples, in which excess water was removed 60 s after sample deposition, were compared to air dried (Fig. 1i) samples, in which the sample drop was left to evaporate naturally. The results presented in Fig. 2 show that filter paper drying reduced approximately 90% of the measured particle number, without changing significantly the average size parameters. This is due to the fact that the removal of the excess liquid also removed most of the particles.

Finally, Fig. 1j shows the micrograph of an air-dried sample washed three times with water, while Fig. 1k shows the micrograph of an unwashed air-dried sample. Both images show that adequate CNC dispersion and staining was attained. As shown in Table 2 and Fig. 2, these samples showed statistically equal length and average aspect ratio values. Nevertheless, washing led to a 30% average increase in the number of measured particles due the removal of contaminants.

To validate this protocol, an unknown sample of commercial CNCs was prepared and analyzed in triplicate (Germiniani et al. 2019). The results are shown in Fig. 4. The unknown sample was diluted to 0.1 mg mL−1, deposited on the dull side of glow discharged grids (15 mA) and left to dry naturally. Afterwards sample was washed with water three times. In spite of being prepared on different days, all replicates provided statistically equal length, diameter and aspect ratio average values, which validate this protocol. To summarize, the protocol optimization showed that good particle dispersion on the grid is key to obtaining homogeneous staining and reliable measurements.

Fig. 4
figure 4

The length, diameter and aspect ratio values of the commercial CNC triplicates, as well as the number of particles counted for each specimen, are presented on the left. A representative TEM micrograph of the commercial cellulose nanocrystals used to validate this protocol is on the right

Full size image

Conclusions and perspective

The protocol presented herein provides cost-effective and readily available guidelines to successful TEM imaging and aspect ratio calculation of cellulose-based nanomaterials. Notably, few alterations in the standard uranyl acetate protocol currently used for staining macromolecules are needed. These alterations revolve around the sample application on the grid and grid preparation, not the staining itself. Even though some authors have already shown that CNC concentration and dispersion are key to achieving good quality images, to the best of our knowledge none has attempted to provide clear and concise guidelines for both, specimen preparation and aspect ratio measurements. The step-by-step procedure that we present herein (supplementary information) provides straightforward guidelines to achieving reliable and comparable aspect ratio measurements with reduced analyst bias. Therefore, this protocol represents a significant contribution to this field of research allowing not only reliable and reproducible CNC aspect ratio calculation, but also high-quality imaging of cellulose nanocrystals embedded in different polymer matrices.