Cryostat Slice Irregularities May Introduce Bias in Tissue Thickness Estimation: Relevance for Cell Counting Methods

Abstract Stereological techniques using the optical disectors require estimation of final section thickness, but frozen tissue irregularities may interfere with this estimation. Cryostat slices from rodent nerve tissues (dorsal root ganglia, spinal cord, and brain), cut at 16, 40, and 50 μm, were digitized with a confocal microscope and visualized through 3D software. Geometric section thickness of tissue (T geom) was defined as tissue volume/area. Maximal section thicknesses (T max), from the top to the bottom of the section, were measured in a random sample of vertical ZX planes. Irregularities were mostly related to blood vessels traversing the tissue and neuronal somas protruding over the cut surfaces, with other neuron profiles showing a fragmented appearance. Irregularities contributed to increasing the distance between the tops and bottoms of slices sectioned in different laboratories. Significant differences were found between T max and T geom for all thickness studies and counting frames (p<0.01). The T geom/T max average rate was 68.4–85.7% in volumes around cell profiles (∼600–1,200 μm2) and 83.3–91.8% in subcellular samples (∼25–160 μm2). Confocal microscopy may help to assess tissue irregularities, which might lead to an overestimation of tissue volume if section thickness is estimated by focusing on the top and bottom of the sections.


INTRODUCTION
Estimation of cell populations in optical microscopy is potentially affected by the splitting of cells and their corresponding particles (nuclei or nucleoli) that may result when sectioning the tissue. If this occurs, the final number of visible profiles counted will be greater than the true number of cells, resulting in an overestimation of the studied population. Stereological techniques were designed to deal with this problem (see Dorph-Petersen & Lewis, 2011 for a review).
Among modern stereological methods, the most popular is the optical disector (Gundersen et al., 1988), which is now automatic and counts particles in a known 3D fraction of tissue inside a thick slice (Howard et al., 1985). This fraction does not include the upper and lower surfaces (Williams & Rakic, 1988) where irregularities and lost caps are thought to occur. In the optical disector, it is necessary to know the final section thickness of the studied slice. Section thickness has been commonly estimated by visualizing the upper and lower surface of the slice and measuring the difference by means of the microcator (Konigsmark, 1970;Coggeshall et al., 1990;Andersen & Gundersen, 1999;Dorph-Petersen et al., 2001;Bermejo et al., 2003;Gardella et al., 2003;Baryshnikova et al., 2006;Rafati et al., 2013;Zhao et al., 2013). However, this way of estimating thickness might be impaired by the different focusing capacity of the observer's eye (Guillery, 2002) and excessively early or late recognition of the top or bottom of a section (Dorph-Petersen et al., 2001) if the tissue is markedly irregular. Intra-section thickness variations of only 0.3 μm for individual glycol methacrylate sections and 1-3 μm for individual paraffin sections have already been described (Helander, 1983). However, the marked, uneven shrinkage of frozen sections (Bonthius et al., 2004;Ward et al., 2008;Carlo & Stevens, 2011), resulting in section deformations, has been reported to introduce substantial noise in the number estimates when unbiased stereological methods are used (Negredo et al., 2004). Frozen sections are commonly used in many nervous system studies and have been proposed for peripheral selective regeneration assessment with several fluorescent dyes (Puigdellívol-Sánchez et al, 2006;Ruiter et al., 2014).
This confocal microscopy-based study was designed to assess the irregularities in cryostat sections, and how they affect stereological counts.
All the animals of the present study were intracardially perfused with 4% paraformaldehyde in 0.1 M phosphate buffer. Details of tissues examined, postfixation, cryoprotection, knives/blades, section thickness, section collection on the slide and immersion medium are found in Table 1.

Confocal Microscopy
Slices were examined at different postsectioning periods (t0-t30 days, Fig. 5) under a Leica TCS SP5 laser scanning confocal microscope (Leica Microsystems Heidelberg GmbH, Mannheim, Germany), equipped with a DMI60/00 inverted microscope and XCS PL APO CS 63 × glycerol objective (NA 1.3) for slices (1.466 refractive index) with aqueous mounting medium, using a 20% H 2 O-80% glycerol immersion media. Images of 512 × 512 pixels with a pixel size of 0.48 × 0.48 µm (246 × 246 µm 2 ) were taken at 8 bits for randomly selected unlabeled tissue and at 12 bits for randomly selected slices with visible FB-DY-labeled cells, with a 0.3 µm interval in Z. An APO lambda blue 63× oil immersion objective lens (NA 1.4) and oil immersion medium (1.518 refractive index) were used for DPXmounted slices in comparable acquisition conditions. Focusing through sections was performed against gravity. Point spread function (PSF) was assessed routinely by the confocal staff using 170 nm microspheres (PS-Spect TM microscope point source kit; Molecular probes, Eugene, OR USA), analyzed using the ImageJ program (Schneider et al., 2012) and Metroloj plugin (Matthews & Cordelieres, 2010), resulting in a measured resolution at the full-width at half maximum of 331 nm (lateral resolution) and 881 nm (axial resolution) for the glycerol objective and 270 nm (lateral resolution) and 680 nm (axial resolution) for the oil objective. Refraction was assessed calculating the ratio of the visible distance between two coverslips separated by tested mounting media and immersion oil, resulting in 1.060-1.057 for 10-20% PBS in 90-80% glycerol, respectively. Addition of antifading 1% paraphenylendiamine resulted in ratios of 1.052 and 1.058, respectively. Maximum "smart gain" was set just below image saturation levels, while minimal "smart offset" was set just below emptiness to ensure that the grayscale range included the whole range of the visible parts of the tissue. Images were acquired sequentially using 458 nm Argon laser lines and DPSS 561 nm laser lines, AOBS as beam splitter, emission detection ranges 415-515 nm and 570-700 nm and the confocal pinhole set at 1 Airy unit.

Three-Dimensional Analysis and 3D Reconstruction
Confocal files were analyzed using Amira 5.2 software (Mercuri Co. Boston, USA). Projection views and XZ or YZ planes were examined ( Fig. 1). Tissues, fibers and neuron profiles were visible due to their labeling or autofluorescence. A systematic random sample of ZX planes of all digitizations (n = 5 for each sample of tissue) was selected for measurement of maximum (T max ) and minimum (T min ) visible thicknesses in each plane (Fig. 2). Labeled cells and nuclei profiles were visualized in the three planes ( Fig. 3). A three-dimensional reconstruction of the digitized tissue sample was performed by semiautomatically choosing segmentation thresholds that select existing tissue and applying additional smoothing filters. The resulting reconstruction was examined in all spatial planes to ensure that the final reconstruction reliably reproduced the tissue shape.
The total volume of the reconstructed structure was then obtained. An estimated "Mean geometric section thickness" (T geom ) was calculated for each 3D reconstruction or digitized tissue (246 × 246 μm, 60,516 μm 2 ) and for smaller samples around visible cell bodies (n = 6 per tissue sample; 600-1,200 μm 2 ) and subcellular samples within each cell sample (25-160 μm 2 Fig. 4 and Table 3), and was defined as Mean geometric section thickness (T geom ) = volume (mm 3 )/area (mm 2 ).

RESULTS
Irregularities in the surfaces of the frozen sections were seen in all samples and laboratories (Fig. 1), corresponding  in some instances to blood vessels traversing the tissue (Fig. 1a) and also to some protruding neuronal cell bodies (Figs. 1c, 1d). The surface facing the glass slide tended to appear more flattened.
Even when projection views show an apparent homogenous shape of the tissue slice (Fig. 2a), 3D reconstruction of the digitized tissue confirms the presence of tissue irregularities (Fig. 2b). These irregularities (Figs. 1 and 2) increase the distance (red lines) between tops and bottoms of the section (yellow lines) in all the samples from the different laboratories.
Individual visualization of nuclei profiles in 16 µm thick DRG or SC slices (98 DRG nuclear profiles and 27 SC motoneuron profiles) revealed undivided nuclei and cells, others with a fragmented appearance (12 and 59%, respectively, Figs. 3a, 3c) and some with an identifiable complementary profile in the adjacent slice (2 and 23% respectively; Fig. 3b). Visualization of parvalbumin-labeled brain neurons also showed cells without any apparent fragmentation of the cell body (Fig. 1c).
Differences between T max and T min , measured in the same XY or XZ planes (Fig. 2b), were significant (p range = 0.00001-0.049) for all the section thicknesses (Table 2), mounting media and postsectioning periods compared. Measurements digitized on the same day as tissue sectioning are summarized in Table 2 while those digitized later showed T geom reductions of up to 51% at t30 days compared with t0 measurements (Fig. 5).
Significant differences were found between T max and T geom for all thickness studies and counting frames, when quantified in smaller samples around visible cell bodies and also in corresponding subsamples (Figs. 4b,4c, respectively) selected within the soma sample (p < 0.00005-0.01). The area in samples and subsamples was comparable to the counting frame used in stereological studies, where the T geom /T max ratio showed a range of 68-92% (Table 3).

Tissue Irregularities
Irregularities are known to occur in frozen slices (Dorph-Petersen et al., 2001;Gardella et al., 2003;Bonthius et al., 2004;Ward et al., 2008), but no previous confocal studies have directly assessed this phenomenon. These irregularities, including neuronal somas protruding over the surfaces of the cut tissue or blood vessels traversing the tissue were seen in DRG, SC, or brain slices from the different cryostats, blades, and laboratories. Optical disector measurements in the central part of a thick section were introduced to avoid such interference (Williams & Rakic, 1988) while modified Figure 3. Cell splitting and complementary adjacent profiles. 16 μm cut slices: a: Diamidino Yellow-labelled nuclei in the DRG tissue boundary (UB-Anat) with a slightly fractioned appearance (yellow circles) while another appears complete (green circle). b: Aligned planes of consecutive sections suggest that the nuclear profile is the complementary adjacent part of the same nucleus. c: Projection view of a sample of spinal cord tissue (UdL-Anat) and of subsamples including CT-488 blue-labelled motoneurons. One is fractioned (yellow circle) in the upper surface of the slice (arrowhead) and the other is complete (green circle). Scale bar is 50 μm. T geom is calculated for the whole piece of sampled tissue (60,516 μm 2 ). T max and T min are measured in a systematic random sample of 5 ZX planes (X: 246 μm) inside the digitized tissue piece. Means ± standard deviation are presented. Significant differences (p) between those parameters and the relationship T geom /T max are detailed. disector including sampling of the whole section has also been proposed (Hatton & Von Bartheld, 1999;Carlo & Stevens, 2011). The presence of some fractioned labeled nuclei, without identification of the fractioned nuclei portion in the adjacent DRG slice, might suggest a certain fragment loss but also the impossibility of visualizing small fractions in the consecutive section (suggested by Floderus, 1944;cited in Haug, 1986 and others). The possible occurrence of lost caps in slide surfaces has been proposed (Andersen & Gundersen, 1999) and it has been suggested that they may interfere in the different counting methods analyzed (Hedreen, 1998;Heller et al., 2001). However, the visualization of protruding cell bodies in the frozen slices may also explain the existence of corresponding "holes" in the adjacent section.

Section Thickness Estimation
The current way of estimating section thickness visualizing the upper and lower surface of the slice (Konigsmark, 1970;Coggeshall et al., 1990;Andersen & Gundersen, 1999;Dorph-Petersen et al., 2001;Bermejo et al., 2003;Gardella et al., 2003;Baryshnikova et al., 2006;Rafati et al., 2013;Zhao et al., 2013) may be comparable to estimating the distance between lines located at the top and bottom of the sections in the different figures, here considered as T max , that would be above final averaging section thickness, here proposed as T geom . Although the variable extent of such differences may be due to the different tissue sources and cutting knives and blades, the difference between T max and T geom is significant in the different types of tissues examined. Previous studies have assessed the axial shrinkage by means of confocal microscopy (Janáček et al., 2012). Here, we introduce the T geom proposal to estimate an average section thickness.
The absolute measures of T geom and T max must be considered with caution, as they are affected by the refractive phenomena of the different immersion media (Hell et al., 1993;Kuypers et al., 2005) and of the tissue studied (Franze et al., 2007)-about 6% here, maximum 3 μm in wider samples in aqueous mounting media. The recommendation is that it should be directly assessed in each situation. Furthermore, the PSF effect is negligible in those samples (causing <1 μm of bias). Thus, since refraction of lasers would have an equal effect on the absolute estimation of T max and T geom , their relationship would be consistent, and indicative of the maximal possible bias when using T max in the disector formula. It is suggested that the combined measurements should be performed on the same day when using aqueous mounting media, due to the progressive shrinkage that may occur after a certain degree of drying during no-frost freezer storage. Future work is needed to  T max and T geom were measured in 3D cubes sampled around visible neuron cell bodies (n = 6 per tissue) inside the digitized pieces of tissue. Additional cubes were subsampled inside. The area studied is indicated, together with the relationship between T geom and T max and their significant differences (p) in all samples of tissue.
assess the known deformations in other embedding techniques (Helander, 1983) and to determine the right sample of confocal measurements needed to "calibrate" an automated microcator approach.

Counting Frames and the T geom /T max Rate
Different microscopy counting frames (Howard et al, 1985), the sample area where cells are estimated, have been used in stereological studies. The area of the subcellular samples ( Fig.  4 and Table 3) was similar to the counting frames used in Bergman & Ulfhake (1998) Fig. 2 and Table 2). Although the T geom /T max ratio is often above 90%, it may reach 84% in the subsample area, 66% in samples around cell bodies and 55% in 246 × 246 µm tissue samples digitized at t0. The area of the counting frame is related to the likelihood of including any irregularities affecting T max . Significant differences between T max and T geom have been found in all kinds of tissue samples and in cellular and subcellular samples of the different laboratories (Table 3 and Fig. 4). Thus, if the aim is to achieve higher precision and to reduce the potential overestimation of section thickness, small counting frames (<200 μm 2 ) could be used in samples with greater irregularities. This procedure may allow a higher precision in section thickness estimates if desired. The current subjective estimation of thickness visualizing the top and bottom of the slice using an optical microscope is uncertain and is affected by the different focusing ability of the observers (Guillery, 2002).
Recommendations arising from this study would depend on the rate of cell splitting, the tissue shrinkage, and the degree of irregularity. Low splitting rates (knife-cut DRG samples) could explain the similarities between cell estimations from profile counts (Puigdellivol-Sánchez et al., 2006) and from systematic reconstruction of sciatic cell populations (Sweet et al., 1986(Sweet et al., , 1991. As decreases in thickness have been detected even 1 day after sectioning when mounted in aqueous mounting medium, it is important to perform counts on the day of tissue sectioning, especially in certain samples (here DRG and SC), although some unlabeled brain samples directly mounted on the slide showed less thickness shrinkage (Fig. 5, Table 2) while preliminary observations suggest that free-floating slices may also show differences from cut thickness at t0, and more irregularities were detected in other brain-labeled samples ( Fig. 1 and Table 3). The long time needed for digitization of high resolution 246 × 246 μm-labelled samples-about 10 min per each 12 bit acquisition-limits the number of pieces of tissue that may be studied in a single day of combined stereological estimations. If the degree of irregularity (T max -T min ) is homogeneous between the samples acquired for any specific experimental design and tissue, the T geom /T max rate could be representative of the maximal bias for thickness estimation, and a range including the real data could be estimated. Volume quantification may be performed through free software (see 3D Image J Suite, for instance). Future work is needed to assess the tissue shrinkage during freezer versus refrigerator storage, with or without no-frost technology, depending on the mounting medium and tissue types.

CONCLUSION
Confocal microscopy may enhance the assessment of cell splitting and frozen slice irregularities. These irregularities may affect the estimation of section thickness that is needed in the disector method. If high precision is desired, taking into account refractive aspects, the use of confocal microscopy and 3D software allows the calculation of mean geometric section thickness.