A systematic assessment of goblet cell sampling of the bulbar conjunctiva by impression cytology.

Experimental Eye Research 136 (2015) 16e28

Contents lists available at ScienceDirect

Experimental Eye Research journal homepage: www.elsevier.com/locate/yexer

A systematic assessment of goblet cell sampling of the bulbar conjunctiva by impression cytology Michael J. Doughty Glasgow Caledonian University, Department of Vision Sciences, Cowcaddens Road, Glasgow G4 OBA, Scotland, UK

a r t i c l e i n f o

a b s t r a c t

Article history: Received 22 October 2014 Received in revised form 30 March 2015 Accepted in revised form 14 April 2015 Available online 14 April 2015

The purpose of this study was to assess the apparent goblet cell density (GCD) from conjunctival impression cytology (CIC) samples in relation to the number of conjunctival cells collected onto the filters. CIC specimens were collected from the superior-temporal bulbar conjunctiva of 16 pigmented rabbits onto Biopore (Millicell-CM) membranes, fixed with buffered glutaraldehyde and stained with Giemsa. Different numbers of microscope fields of view in each of the specimens were imaged by light microscopy using a 20 magnification objective lens (200 final magnification), and the goblet cells marked and counted. The GCD values/sq. mm were calculated. The same conjunctival region of 3 other rabbits was also prepared for transmission electron microscopy (TEM) by fixation, in situ, with the same buffered glutaraldehyde. Mean values for GCD estimates were found to vary from 399 to 1576 cells/sq. mm, depending on the image sampling and analysis strategy chosen, with the lowest inter-sample variance of around 10% being found if a maximum goblet cell count was taken on substantially multilayered regions of the CIC specimens. Counts of the number of goblet cells per 1000 visible conjunctival epithelial cells yielded a value of close to 90 (range 36e151), with modest inter-sample variability of around 30%. A three or ten 200 microscope field and random sampling strategy yielded mean GCD values between 542 and 670 cells/sq. mm, but with very high intra- and inter-sample variance of at least 60% and sometimes higher than 100%. TEM confirmed the multilayered organization of the conjunctiva and the deeper lying goblet cells. The general use of a goblet cell count as an objective marker for conjunctival normality or health is likely to be highly variable unless a more specific strategy is adopted. Beyond providing details of exactly the counting strategy used, it would be very useful to provide full details of the actual microscope field size used as well as information on the intra-sample variability in goblet cell counts. © 2015 Elsevier Ltd. All rights reserved.

Keywords: Impression cytology Bulbar conjunctiva Electron microscopy Rabbit Goblet cells

1. Introduction Conjunctival impression cytology (CIC) has become a widely used and accepted method to assess the cells of the ocular surface (Calonge et al., 2004; Lopin et al., 2009; Baudouin et al., 2014). It provides a method of obtaining cells from the bulbar (or palpebral) conjunctiva without the need of the surgical intervention that would be required in taking a biopsy sample of the same cells (Doughty, 2012a; Calonge et al., 2004; Lopin et al., 2009). In a recent European assessment of diagnosis options for severity of dry eye, it was concluded that impression cytology is a conjunctival epithelium sampling method which is “Well validated with published scoring systems' and which includes “Goblet cell count as an

E-mail address: [email protected]. http://dx.doi.org/10.1016/j.exer.2015.04.007 0014-4835/© 2015 Elsevier Ltd. All rights reserved.

objective marker’ (Baudouin et al., 2014). Assessments of the conjunctiva and its goblet cells by CIC have been widely used in clinical studies (Doughty, 2012a; Calonge et al., 2004; Lopin et al., 2009), but the method has also been undertaken on rabbits, usually as part of toxicology or disease-related studies that constitute an important element of understanding the clinical applications of CIC (Hatchell and Sommer, 1984; Shellans et al., 1989; Driot and Bonne, 1992; Hareuveni and Maurice, 1994; Nakamura et al., 1998; Martinez et al., 1995; Murakami et al., 2003; Marino et al., 2005; Oh et al., 2007; Altinors et al., 2007; Toshida et al., 2007; Xiong et al., 2008; Qiu et al., 2010; Kim et al., 2011; Liang et al., 2011; Cai et al., 2011; Li et al., 2012; Pauly et al., 2012; Yu et al., 2013; Li et al., 2013; Hwang and Kim, 2014). The use of CIC in such animal studies provides an opportunity, not routinely available for human studies, to further investigate the structural organization of the bulbar conjunctiva in relation to the

M.J. Doughty / Experimental Eye Research 136 (2015) 16e28

CIC sampling. The basic principle of the CIC technique is generally credited to Egbert and colleagues (Egbert et al., 1977) who reported that goblet cells as well as non-goblet conjunctival cells could be removed onto the CIC filter. Using rabbit conjunctiva as source material, consideration has been given to the issue of cell yield (Martinez et al., 1995), and also as to how comparable repeated impressions might be (Hareuveni and Maurice, 1994). It has also been noted that in some regions of the CIC filters taken from rabbits there could be multilayers of cells (with numerous accompanying goblet cells) while other parts of the filter only a monolayer of epithelial cells could be evident (Driot and Bonne, 1992) or a ‘homogeneous cell sheet’ present (Pauly et al., 2012). The goblet cell ‘density’, especially in relation to the non-goblet (epithelial) cells is something that needs further validation as there has been no obvious standardization of the way in which this might be assessed. Using light microscopy of rabbit CIC specimens, investigators have usually reported average values for the goblet cell numbers that can be seen in what have usually been stated to be high power (microscope) fields (HPF) of view, rather than just assigning a grade based on approximate numbers of goblet cells (Liang et al., 2011). In one such approach, five different HPF were selected for the counting, but only from the central region of the filter surface where multilayering of cells was evident (Driot and Bonne, 1992). Other investigators appear to have usually used three microscope fields for counting, although not necessarily at high power (Shellans et al., 1989; Nakamura et al., 1998; Murakami et al., 2003; Xiong et al., 2008; Kim et al., 2011; Li et al., 2012; Yu et al., 2013; Hwang and Kim, 2014), or 10 fields (Oh et al., 2007; Toshida et al., 2007; Qiu et al., 2010; Li et al., 2013). In none of these studies, however, has any indication been provided of the basis of the visual selection of the microscope fields for counting, nor on the actual variability in the counts across different fields. Since the numbers of goblet cells across CIC samples can differ, perhaps quite substantially (Driot and Bonne, 1992; Hareuveni and Maurice, 1994), part of the objective of this present study was to illustrate and systematically assess this variability in the goblet cell numbers that can be counted. While numerous investigators have reported using multiple microscope fields of view to undertake counts of goblet cells in CIC samples (reviewed in Doughty, 2012b), details have not be provided on the variability between fields nor on whether only fields that included goblet cells were actually counted. For the present studies, this variability is that which could be visualized across CIC samples from normal healthy conjunctiva, using a rabbit model. The normality of the specimens was underpinned by morphometric analyses of the epithelial (non-goblet) cells and then the outcome of using various strategies to count goblet cells objectively compared. Finally, using ultrastructurebased cell and tissue assessments of the same region of the rabbit bulbar conjunctiva, the aetiology of the differences in goblet cell counts is considered. 2. Materials and methods 2.1. Animal source, care and collection of eye tissue The animal use and all protocols were in accord with the ARVO resolution on the Use of Animals in Research, and were approved by a local animal care committee at the University of Waterloo, Canada (Doughty, 2013, 2014). Female Dutch Belt pigmented rabbits, aged between 9 and 11 weeks (1.9e2.4 kg body weight) were either being used for teaching purposes (Biology and Optometry) or for National Sciences and Engineering Research Council (NSERC) funded research to the author in the Optometry department. All animals were initially quarantined for 7e10 days to check that there

17

were no obvious clinical signs of ocular, nasal or GI tract disease. Individual animals were then euthanized with i.v. pentobarbital sodium. 2.2. Impression cytology Immediately after euthanasia, the upper eyelid of the left eye of 16 rabbits was carefully lifted and a 10 mm diameter Millicell®-CM filter unit 0.4 PICM 012550 with the Biopore® membrane (Millipore, Co. Cork, Ireland) was applied to the superior-temporal bulbar conjunctiva, with the edge of the filter clear of the limbus. The filter was removed, allowed to air dry for 2e5 min and then a single drop of buffered glutaraldeyde solution applied, and left for 15 min at room temperature (RT). The filter unit was then returned to its packaging and stored in a 4 C refrigerator in opaque containers for later processing. With fixation, the filter units can be stored for years without any obvious deterioration in the cells. The fixative was a 2% solution of electron microscopy grade glutaraldehyde, freshly prepared in 80 mM sodium cacodylate buffer, pH 7.2 to 7.4, 320e340 mOsm/L. The filter units were later stained by immersion in 99% methanol for 2 min, then in distilled water for 1 min and finally in Giemsa solution (product number G032, Sigma, Kingstonupon-Thames, UK) for 2 min at RT. The Giemsa option was selected so as to allow visualization of epithelial cells and their nuclei (which usually stain substantially) as compared to the essentially non-staining goblet cells which appear as ghostly, balloon-like oval-shaped features (Doughty, 2011a, 2012a) The same appearance can be seen with haematoxylin (Rath et al., 2006) and a similar appearance following use of certain immunocytochemical staining (Krenzer and Freddo, 1997; Danjo et al., 1998). Following previous protocols, the filters were photographed (to allow for measurements of the total area covered with cells), and then examined by light microscopy using a 20 microscope stage objective lens (medium power field, 200 final magnification) (Doughty, 2012b). For general assessment of the non-goblet (epithelial cells), regions across the filters containing only a monolayer or very slightly multilayered regions were photographed, and a 100 mm scale marked fixed to prints made of the images. These images were graded based on cell size and nucleus-to-cytoplasm ratio using Nelson-type guidelines (Nelson, 1988) as reported using a set of schematic diagrams (Blades and Doughty, 2000) for subjective assessment of the relative area occupied by the nucleus compared to the surrounding cell cytoplasm. The prints were then scanned at 400 d.p.i., a JPEG file generated and projected onto a wall screen at a magnification of close to 1000 so that an overlay could be generated to show the cell and cell nucleus outlines and which included the scale bar. These overlays were then used to measure cell and cell nucleus dimensions and areas, as well as for counts of the total number of cells/200 field of approximately 0.14 mm2 area (Doughty, 2012b). Areas included in images as well as specific cell and nucleus dimensions and areas were measured by planimetry undertaken on the much enlarged overlays using a commercial digitizer pad (DigiPro, Elestree Computing, London). From the dimension measurements, the length of the nucleus in relation to the cell length was calculated (as nucleus length/cell length) and from this an estimate made of the nucleo-cytoplasmic (N/C) ratio. The calculation was (nucleus length/cell longest dimension) (Doughty, 2011b). Since nucleo-cytoplasmic ratios are usually presented as ratios rather than fractions (Nelson, 1988), the fractions (i.e. 0.4 for the nucleus) were also used to calculate an N:C ratio, i.e. would be 1 (for the nucleus) to 1.5 for the cytoplasm so yielding 1: 1.5 etc. On all these overlays, any discrete regions not covered with cells were also outlined and their areas measured. With the total area included in a 200 field known, the relative area (%) actually covered with cells was estimated.

18


For assessments of goblet cell density, two methods were used according to the overall density. In the first, the same projection method used to assess the epithelial cells was used with both the nuclei of the epithelial cells being outlined as well as the goblet cell outlines. From these overlays made from the projected images, the total number of epithelial cell nuclei was counted as well as the total number of goblet cells. This allows for calculations not only of the number of goblet cells/mm2, but also the number of goblet cells per 1000 epithelial cells (Yeo et al., 2003). Alternatively, the number of goblet cells/200 field were simply counted for calculation of goblet cells/mm2 from the known field size. The strategy for selection of suitable images for this method will be described as part of the results. For repeat assessments of goblet cell density (GCD), the images off the CIC filter surface were displayed through a camera-attached video camera (JVC TK-1280E) onto 14 inch analog colour TV monitor (nominally 560 lines/inch), i.e. the 200 field of view was displayed as a region measuring 29 cm 22 cm. A plastic sheet was placed across the screen and a marker pen used to outline the location of all the goblet cells that could be discerned, even if slightly out of focus (Doughty, 2012b). The final size of the goblet cells was routinely several mm in diameter. The plastic sheet was then removed for later counting (where each of the outlined goblet cells was then marked with a dot to indicate it had been counted so avoiding errors in counting), as well as calculation of the goblet cell density (GCD) as the number of goblet cells/mm2. The strategy for selecting the fields of view using the monitor display will be outlined in detail in the results. In brief, for repeat assessments of 3 fields, a region close to the center of the filter was first selected and then two adjacent fields, several mm apart then assessed. For 10 fields, an image close to one edge of the filter was first selected and then the stage moved in a sequence to include successive nonoverlapping fields. 2.3. Preparation and fixation of the bulbar conjunctiva for electron microscopy From three different rabbits to the ones used for CIC, the superior-temporal bulbar conjunctiva of the right eye was prepared for transmission electron microscopy (TEM) using a special dissection and tissue preparative method to preserve the tissue in as smooth (fully extended) condition as possible (Doughty, 2013, 2014). In brief, this involves gently stretching the bulbar conjunctiva back over the enucleated eyeball prior to the fixative application post-mortem. The fixative was the same as that used for the impression cytology filters. The net result is a piece of tissue (approximately 8 10 mm) extending from the corneo-limbus edge, most of which was used for scanning electron microscopy (SEM) (Doughty, 2013, 2014). However, prior to the specific preparation for SEM, a narrow strip (approximately 1 mm wide) was cut from the temporal edge, and two small blocks of tissue (approximately 1 2 mm) processed and examined by TEM as essentially as previously detailed (Doughty, 2012c, 2015). Both blocks were processed for light microscopy (with thick sections being stained with toluidine blue), and then thin sections prepared from one block and stained with lead citrate then uranyl acetate. 2.4. Statistical analyses All data were entered into spreadsheets in Systat v. 11 (Syatat, Evanston, IL) for calculation of global statistics and generation of graphical output. The global statistics included average ± SD, as well as the coefficient of variation (COV) as based on (average/SD) x 100. Data sets of goblet cell counts and GCD were checked for normality using the ShapiroeWilk option, and sets of GCD

estimates compared by non-parametric Friedman rank order test because of the substantial heterogeneity (non-normality), with the level of significance set at p < 0.05. 3. Results 3.1. General assessment of epithelial cells and goblet cells collected onto the CIC filters In Fig. 1 are shown a series of images that reflect the very different appearances that the cells on the CIC filters can have. The images all include a rectangular box as the scale marker, the length of which is 100 mm. In Fig. 1A is shown a typical example off a small region of a filter where only a monolayer, or nearly a monolayer, of conjunctival epithelial cells was evident. The 200 magnification was sufficient for not only individual epithelial cells to be clearly seen, but also having a sufficient size for morphometry to be undertaken. Some epithelial cells were more clustered together (see right hand side of Fig. 1A), while others appeared to be a little more spread out (see left hand side of Fig. 1A). That all the cells in the monolayer are not all the same could be due to slight differences in pressure that could be exerted with the manual application of the filter so that some cells are slightly more flattened out than others. It addition, it should be noted that parts of the filter surface in such regions had no adherent cells with just the grainy texture of the filter forming the background. It would, of course, have been possible on a field of the type displayed in Fig. 1A to have used the higher magnification objective (400) to just select a much smaller region where the cells were slightly more spread out or just those fairly densely packed together. The 200 field was, however, deliberately chosen to illustrate both that the extent of cell spreading is not always the same and also that some spaces between conjunctival epithelial cells on the CIC filters is not uncommon. Notwithstanding, the overall feature of the 200 field regions was that numerous individual epithelial cells with their nuclei could be clearly discerned, and even where the actual boundaries of individual cells are not clearly visible, the densely staining nuclei were very apparent. An image of the type shown in Fig. 1A was taken from each of the 16 filters, one from each of the rabbits. In Fig. 1B is shown a typical appearance that could also be fairly readily found across the same filters, but unlike the cells shown in Fig. 1A, the overall staining intensity of the cells was usually less even though the actual number of cells was considerably higher. In such regions it was extremely hard to discern any individual epithelial cells but the image clearly contains numerous Giemsastained nuclei which can be counted. In addition, there were a fairly large number of faint-staining, balloon-like cells that have much smaller nucleus; these are the goblet cells (labelled as GC on the Figure). In images of the type shown in Fig. 1B, the entire microscope field was filled with cells, i.e. there were no gaps. It was possible to find images of this type from most of the CIC filters examined (12/16) with all having an intermediate cell density (between a monolayer and a moderately high density) with some goblet cells evident. In Fig. 1C is shown a region of a filter surface where the entire 200 field is essentially filled with the balloon-like goblet cells. These balloon-like cells were even easier to see if the background of stained epithelial cell nuclei was substantial. Regions showing this type of appearance were located on 13 of the 16 filters, although the overall density of the background staining was somewhat variable. In Fig. 1D is shown an example where the background staining is much more intense but the goblet cells are revealed because of their lack of staining. Similar to Fig. 1C, almost all of the 200 field is clearly filled with goblet cells.


19

Fig. 1. Representative images of medium power microscope fields (MPF) of the surface of impression cytology filters taken from the superior-temporal bulbar conjunctiva of rabbits to illustrate regions of epithelial cells without goblet cells (A), regions dominated by large numbers of epithelial cells and some goblet cells (B), and regions where the individual epithelial cells are no evident and just the balloon-like goblet cells are evident (C, D). Giemsa stain. Goblet cells are marked as GC. The length of the vertically-oriented rectangular boxes represents the scale bar of 100 mm.

3.2. Objective assessment of the epithelial cells in impression cytology specimens The first assessment was to verify that the epithelial (nongoblet) cells had morphological features that were consistent with them being considered normal and healthy, with assessments made both of cell density on the filter surfaces and of the actual cell size. On images of the type shown in Fig. 1A (or the overlay shown in Fig. 2A), no goblet cells were routinely evident. This was the case for 14 of the images of epithelial cells in a monolayer. A solitary goblet cell was included in one of the other images and just two in another (not shown), and in both of these examples the goblet cells were found where there was slight multilayering of the epithelial cells.

The number of epithelial cell nuclei were counted and then also taking into consideration the actual area covered by the cells. These assessments were all made from the overlays (e.g. Fig. 2A) to minimize counting errors. The overlay also serves to highlight those regions where there were no cells to calculate cell densities. Using the actual nuclei counts as generated from each overlay, the values ranged from 197 to 398 (group mean 290 ± 54/200 field, ± SD) to give an estimated group mean epithelial cell density of 2024 ± 373 cells/mm2. It should be noted that the inter-specimen coefficient of variation (COV) on these estimated cell density values was 18.4%. Measurements of the actual area on the overlays occupied by cells indicated that the coverage over the 200 field was between 69 and 91% (mean 80 ± 8%). Adjustment of the cell density estimates to take these cell-free regions into account yielded a revised value of

Fig. 2. Examples of overlays generated from projecting MPF images at 1000 final magnification to mark some of the individual cell borders (shown in black) and their nuclei (shown in red) as well as the nuclei from the rest of the epithelial cells (A) or just the epithelial cell nuclei (marked in red) and the goblet cells (marked in green) (B). Scale bar is 100 mm. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

20


2450 ± 446 cells/mm2. The inter-specimen COV for the estimated cell density was very similar to that obtained without the adjustment, being 18.2%. As an indicator of the normality of these CIC specimens, the areas of at least 35 cells (average 56 ± 12/overlay) were measured from each of the 16 images (from the 16 specimens from the 16 different rabbits), and yielded average values between 207 and 350 mm2, with a group mean of 284 ± 42 mm2. A histogram of the distribution of the epithelial cell areas across the 16 specimens (Fig. 3) shows there was reasonable consistency (or low interspecimen variability) in the epithelial cells in the CIC samples. The overall distribution was slightly skewed and not normal (p ¼ 0.001), but the inter-specimen COV for epithelial cell size (based in the set of average area values for the epithelial cells) was just 14.8%. The average cell length values ranged from 16.1 to 21.9 mm (group mean 18.8 ± 1.6 mm). Calculations of the relative size of the nuclei compared to the cell length yielded average N/C ratio values between 0.374 and 0.519 (group mean 0.434 ± 0.042). These values yield N:C ratios ranging from 1:1.05 to 1: 1.67 (for a group mean of 1: 1.32). Using a modified Nelson scheme, all epithelial cells in the monolayer regions were therefore considered to be grade 0, with N:C ratios of between 1:1 and 1:2. Using the average cell area values from each overlay for the calculations, the cell density estimates had values ranging from 2857 to 4831 cells/mm2, for a group mean 3605 ± 571 cells/mm2). 3.3. Objective assessment of goblet cell numbers in relation to overall epithelial cell density In systematically searching across the surface of the CIC filter, regions could be found where there was some multilayering with just a few goblet cells (not shown) or, more often, regions where there was obvious multilayering and quite a few goblet cells (see Fig. 1B and overlay Fig. 2B). Regions of the latter type were found in 12 of the 16 filters and were selected for objective analyses. The overlay (Fig. 2B) highlights the numerous and small-sized epithelial cell nuclei that could be fairly easily identified (and shown as small (red) circles on the on-line version of the overlay). Such nuclei essentially completely filled the 200 field with no notable gaps where cells were absent. The nuclei were usually slightly less

Fig. 3. Histogram to illustrate distribution of area values for superficial epithelial cells arranged in a monolayer on CIC filters. Bar heights indicate average values from 16 specimens ± SEM.

densely staining as compared to those for cells in a monolayer, but were always very close together and in some cases actually touching each other or even slightly overlapping. This indicates that the epithelial cells were multilayered. From the 12 selected image overlays, the number of epithelial cell nuclei ranged from 496 to 802/200 field (group mean 643 ± 88 cells/200 field), with the group mean epithelial cell density being calculated to be 4496 ± 619 cells/mm2. This intermediate cell density, with some or quite substantial multilayering of the cells, was between 24.7 and 86.9% higher than cell density estimates off monolayered regions of the filter surface. On the same 12 images, the number of goblet cells (shown as the larger (green) oval-shaped features in Fig. 2B), ranged from 26 to 91/200 field (mean 58 ± 19/200 field), which translates to GCD estimates of between 182 and 566/mm2 (group mean 399 ± 124 goblet cells/mm2). With the higher numbers of epithelial cells at the ‘intermediate’ cell density, the number of goblet cells/1000 epithelial cells was calculated, yielding a mean of 92 ± 32 goblet cells/1000 epithelial cells (range 36e151). Using these two slightly different approaches, the variability (as COV based on the SD values) on the goblet cell densities was between 30 and 35%. 3.4. Objective assessment of maximum goblet cell numbers/unit area Ten of the 16 images had features of the type shown in Fig. 2C with very high epithelial cell density, and a further three more resembled that shown in Fig. 2D, i.e. as having extremely high epithelial cell density. In the former cases, the overall characteristic was that the 200 field was largely filled with the very weak staining ovoid goblet cells although it was extremely hard to discern the nuclei of many of the individual epithelial cells which formed the background to the goblet cells. Images of this type (13 in total) yielded ‘maximum’ goblet cell counts between 186 and 272/200 field (mean 226 ± 27 goblet cells/200 field) to give GCD estimates of 1576 ± 192/mm2 (range 1300e1838/mm2). These values are from single fields selected from each specimen. The averaged inter-sample variability in maximum GCD, as the COV, was just 12.2% (Table 1). 3.5. Systematic assessment of goblet cell counts and goblet cell density estimates based on three microscope fields of view The region of the surface of the filter covered with Giemsastained material was usually close to circular in shape, and centered on the middle region of the filter or slightly off center. Such a stained region was at least 6 mm in diameter so the cells covered an area of at least 25 mm2. In more than half the filters, the diameter of this stained zone was actually equal or greater than 7.5 mm so giving an area of cells of at least 30 mm2 considered suitable for large numbers of repeat counts of goblet cells. For all the filters, the area covered by Giemsa-stained material ranged from 24.4 to 55.4 mm2. With just three fields being that more commonly selected in published studies on rabbit CIC samples, repeat counts were made off just three randomly selected fields for all 16 specimens. This was done by finding a field roughly in the middle of the filter surface with goblet cells in it (regardless of the epithelial cell density or intensity of the background staining produced by the epithelial cells), counting the number of goblet cells (by marking on a plastic overlay) and then moving the microscope stage by a fairly large distance (i.e. a mm or so) to locate another field and then repeating this again. If the new field encountered did not contain any goblet cells, another was selected. Using this approach, the number of goblet cells in any particular


21

Table 1 Relative numbers of goblet cells, minimum and maximum goblet cell density estimates.

Epithelial cell area (mm2) Epithelial cell density based on average area (cells/mm2) Goblet cell counts at intermediate epithelial cell density GCD at intermediate epithelial cell density GCD/1000 epithelial cells Goblet cell counts at very high epithelial cell density GCD at very high epithelial cell density

Mean

SD

Range

COV (%)

284 3605 58 399 92 226 1576

42 571 19 124 32 27 192

207e350 2857e4831 26e91 182e566 36e151 182e272 1300e1838

14.8 15.8 32.8 31.1 34.8 11.9 12.2

200 field ranged from 7 to 246/200 field. Comparisons of the first, second and third counts across the 16 specimens revealed goblet cell numbers ranging from 8 to 194, 17 to 211 and 7 to 246/ 200 field respectively, with the average value (±SD) from the first, second and third counts being 85 ± 57, 97 ± 63 and 100 ± 79 goblet cells (Table 2). There was no measurable statistical difference between each set of successive counts (p > 0.5). A box plot of this data (Fig. 4A) shows the substantial inter- and intra-sample variability in GCD counts/200 field with the differences in the median values (the horizontal lines in the boxes) and the very large ± 25% intervals (indicated by the height of the boxes). For the sets of three counts/ specimen, the coefficient of variation values were between 12.7% and 116.7% for an overall mean of 63.6%. There were similar mean COV values of 67.3%, 64.9% and 78.8% derived from the first, second and third counts respectively (bottom line, Table 2). If the just the average goblet cell counts (from three/200 fields) were used to calculate an overall value from the 16 specimens, then this was 96 ± 24 goblet cells (±SD)/200 field, which could make the inter-sample COV estimate as low as 25.2%. It could be argued that this lower SD value (across 16 specimens) could be used as an indictor of lesser inter-specimen variability (with the ‘mean’ COV being just 25.2%), but this ignores the use of three 200 fields for each estimate. Using the average counts (from the three fields taken from the 16 different CIC samples), the average GCD values were between 106 and 981 cells/mm2 from which a group mean GCD of 670 ± 169 cells/mm2 can be calculated. As with the goblet cell counts from which the GCD estimates originated, the inter-sample COV was again be calculated to be only around 25% (Table 2), but this is clearly not a good indication of the broad ± 25% inter-quartile intervals on the GCD estimates obtained from successive 200 field evaluations (Fig. 4B). 3.6. Systematic assessment of goblet cell counts and goblet cell density estimates based on ten microscope fields of view For this approach, only 11 of the filters were used where the area covered by Giemsa-stained material was higher (being at least 7.5 mm in diameter with covered areas averaging 41.4 mm2, range 31.5e55.4 mm2) so as to minimize the chance of the same or slightly overlapping regions being counted twice. The same overall

Table 2 Goblet cell counts and density estimates off three medium power microscope fields (MPF).

Min Max Av SD COV

C1

C2

C3

GCD1

GCD2

GCD3

8 194 85 57 67.3

17 211 97 63 64.8

7 246 100 79 78.8

56 1356 595 400 67.1

119 1474 720 459 63.7

49 1719 696 551 79.2

C1, C2, C3 are the goblet cell counts/MPF. GCD1, GCD2, GCD3 are the goblet cell density estimates (GCD)/mm2. COV ¼ coefficient of variation.

strategy was adopted as for three fields, except that ten randomlyselected 200 fields were used. The process was started on one side of the filter and the microscope stage progressively moved horizontally across the filter to 2, 3 or 4 more different locations roughly in a straight line, then moving the stage vertically and repeating the process. If, in the process of moving the microscope stage, a region of just epithelial cells with no goblet cells was encountered, this was noted and another selected by now moving the microscope stage just slightly (guided by the video display) to get a nearadjacent field that did include goblet cells. After such an encounter and slight adjustment of the field of view, the broader strategy of moving the stage by about 0.5e1 mm for each successive field was resumed until 10 fields with goblet cells had been counted. This meant that up to 14 microscope fields were actually examined/filter. The extreme edges of the stained material were avoided, mainly because the coverage with cells was usually rather sparse. The goblet cell outlining process, for 10 fields/specimen, usually took 15e20 min. In assessments of the goblet cell numbers, the data for all 10 fields where goblet cells were encountered was first considered (i.e. ignoring the fact that a few fields with no goblet cells were randomly encountered), and then a separate analysis was undertaken of the just the first 10 fields so including those where no goblet cells were encountered. Overall, it can be noted that at least some goblet cells were frequently encountered in such a pseudorandom sampling approach of the CIC filter surfaces. Overall, from the sequential analyses of ten 200 fields/specimen, goblet cells were encountered in 96 of 110 fields (i.e. 87.3%). For the first set analyses, that only included fields with goblet cells, the actual numbers of goblet cells/200 field differed substantially, ranging from just 3 to 261, with there being no substantial differences between specimens, i.e. all showed a wide range of goblet cell counts (Table 3). A step-wise assessment of the number of goblet cells for each specimen also indicated random heterogeneity in that these numbers ranged from 11 to 244 (mean 100) in the first 200 field assessed (for each specimen), from 7 to 211 (mean 88) in the second 200 field viewed, from 6 to 207 (mean 78) in the third 200 field etc (see Table 3). Overall, the range of average numbers of goblet cells/200 field (from 10 fields/ specimen) ranged from 54 to 134, for a group mean of 87 ± 26/ 200 field ± SD). The coefficient of variation (COV) estimates on the ten 200 field counts for each specimen were high to very high with values between 50.5 and 94%, with an overall group mean COV of 71.7% (Table 3, sixth line, COV). As with the calculations made off three 200 fields, it might be argued that because the SD that could be calculated from the sets of average values from 10 fields was only 26% (about the group mean of 87 goblet cells/200 field), that the overall variability (as the COV) was only modest (at 29.6%). This however was much larger than similar estimates off 3 fields (of c. 25%) and ignores the fact that 10 fields have been assessed. The heterogeneity in the goblet cell counts can be illustrated as a box plot (Fig. 5A) where most of the inter-quartile intervals are broad, and where relatively narrow these data sets can still include

22


Fig. 4. Box plots to illustrate the variability in goblet cell counts (A) and calculated goblet cell density (GCD) in cells/mm2 (B) from three successive medium power microscope fields of view of the CIC filters.

Table 3 Goblet cell counts and density estimates off 10 medium power microscope fields (MPF) ignoring fields without goblet cells. Sample ID

C1

C2

C3

C4

C5

C6

C7

C8

C9

C10

Min Max Median Mean SD COV

11 244 100 100 66 65.6

7 211 72 88 69 78.6

6 207 44 78 72 92.5

24 196 101 96 48 50.2

14 236 82 96 72 74.5

12 261 62 85 71 83.9

7 240 77 86 75 87.2

7 212 100 94 69 73.6

3 186 44 70 66 94

3 207 41 77 72 93.6

GCD1

GCD2

GCD3

GCD4

GCD5

GCD6

GCD7

GCD8

GCD9

GCD10

77 1706 699 698 458 65.7

49 1475 503 615 483 78.6

42 1447 308 546 506 92.7

168 1370 706 671 337 50.2

98 1650 573 679 505 74.3

84 1825 433 597 499 83.5

49 1678 538 602 525 87.1

49 1482 699 654 484 74.0

21 1300 308 489 460 94.0

21 1447 287 536 504 94.0



Fig. 5. Box plots to illustrate the variability in goblet cell counts (A) and calculated goblet cell density (GCD) in cells/mm2 (B) from 10 successive medium power microscope fields of view of the CIC filters. Asterisks indicate outliers.

notable outliers (asterisks, Fig. 5A). The goblet cell counts/200 field were again used to calculate GCD estimates (bottom part of Table 3), the net results of which are also shown as a box plot (Fig. 5B). The goblet cell numbers in Table 3 were determined from using

only fields with goblet cells. If instead the strategy adopted was to use ten successive 200 fields, so included some fields without goblet cells, the overall group mean for goblet cell numbers was reduced to 78 ± 31/200 field (±SD), but the mean COV increased. With these being sequential field assessments, the spread of values


23

layer from the surface. While this was not always the case, the example was chosen as being the most representative, and most commonly encountered so as to allow for illustration of the possible cell sampling process in CIC taken off the superior bulbar conjunctival surface. In a few sections, a goblet cell was occasionally found almost in contact with the surface, or in the second layer or, rarely, be in reasonably close contact with the basement membrane. As illustrated, with the special preparative method used, the goblet cells were ovoid in shape with the width slightly exceeding the thickness. If a CIC filter was placed in contact with a portion of the conjunctival surface of the type shown in Fig. 6B and four to five cell layers were then removed onto the surface of the filter, the net result would be that shown in the top part of Fig. 7 (with the filter in the upper part of the schematic derived from the cell outlining shown in Fig. 6B. A scenario where just the most superficial epithelial cells were removed is shown in Fig. 7 (as one layer of cells), where two layers of epithelial cells were removed and finally where three or more layers of cells were removed. For the cell layers, the schematic has been inverted to align with the light microscope view of looking down on the filter surface. Where one or even two layers of epithelial cells were removed, the light microscope view would be expected to show well spaced or closely grouped epithelial cells respectively. As deeper lying cells were also removed (in this case three layers), some goblet cells would also be expected in the CIC sample, but it could be just one of them that adhered to the filter (cf. Fig. 1B) or both (or more) goblet cells (cf. Fig. 1C and D respectively).

in each successive set (first, second, third etc) from the 11 specimens was slightly different from when fields without goblet cells were ignored. The mean values were lower in 7 of the 11 sets of fields (Table 4). This difference in the mean goblet cell numbers (to when zero counts were ignored) was statistically significant (paired t-test, p ¼ 0.004). Overall, the variability in each successive set increased, with the range of COV values being from 71.1 to 112.3 %, with the overall group mean COV now being 86.9% (Table 4, sixth line, COV). The average GCD values from the ten 200 fields that included goblet cells, ranged from 375 to 936/mm2, for an overall group mean of 608 goblet cells/mm2. As with the goblet cell number counts, the COV on the GCD estimates were again high (mean 87.1%). With inclusion of zero goblet cell counts as part of the calculation of GCD estimates, the range of average GCD values changed to 277 to 936 cells/mm2, for a group mean of now 542 ± 219 cells/mm2. As with the goblet cell counts used to derive the GCD estimates, this difference was highly statistically significant (p ¼ 0.005). The COV on the mean GCD estimate was 40%, which must also be considered as a substantial underestimate of variability because ten 200 fields were used off each specimen. In comparisons of the outcomes (Tables 3 and 4), there are some slight differences arising from the combined impacts of using a multiplication factor and rounding off numbers to integer values (i.e. a partial goblet cell was not computed).

3.7. Ultrastructure of rabbit bulbar conjunctival epithelium and the aetiology of variable goblet cell counts on impression cytology samples

4. Discussion In Fig. 6A is shown a representative, lower power TEM image taken from the superior-temporal region of rabbit bulbar conjunctiva. In Fig. 6A is the original micrograph with one of the goblet cells (GC) marked, while in Fig. 6B the cellecell borders and the goblet cells are outlined. Only three rabbits were examined using this assessment but the results were consistent across all 6 blocks of tissue that were examined, showing 4 to 6 layers of cells interspersed with some goblet cells with their characteristic mucin granules. The epithelium was lying on top of a very thin basement membrane that separated it from the underlying parenchyma. This collagen-based connective tissue also included some fibroblast-like cells (see lower right of image) but no inflammatory cells were observed in the examination of numerous different sections. Both blood vessels and lymphatic vessels were noted slightly deeper in the parenchyma (not shown). In the example illustrated, both goblet cells were in the third

CIC has been used on rabbits for at least 30 years, and even more widely used in human studies (Doughty, 2012a), with goblet cell assessments being provided in many of the published reports. If ‘goblet cell count’ is to be considered as an ‘objective method’ (Baudouin et al., 2014), then it is surely important to have data that would define what ‘normal’ numbers of goblet cells should be and their expected variability, as well as an agreed method by which such counts might be undertaken. While numerous investigators have reported using multiple microscope fields of view to undertake counts of goblet cells in CIC samples (reviewed in Doughty, 2012b), details have not be provided on the variability between fields nor on whether only fields that included goblet cells were actually counted. The present studies provide, for the first time, a detailed assessment of the difficulties that can be encountered in trying to determine what normal goblet cell counts should be, as

Table 4 Goblet cell counts and density estimates off 10 medium power microscope fields (MPF) including fields without goblet cells. Sample ID

C1

C2

C3

C4

C5

C6

C7

C8

C9

C10


9 244 100 92 71 76.7

0 211 42 73 72 98.4

0 207 44 82 76 92.5

0 152 96 65 52 80.6

0 236 34 71 77 108.2

0 261 62 94 79 83.8

0 240 102 92 65 71.1

0 201 17 57 64 112.3

7 212 17 79 80 101.0

0 207 41 71 72 102.0

GCD1

GCD2

GCD3

GCD4

GCD5

GCD6

GCD7

GCD8

GCD9

GCD10

63 1076 699 642 494 76.8

0 1475 294 508 502 98.8

0 1447 308 571 530 92.8

0 1062 671 456 366 80.2

0 1650 238 498 536 107.8

0 1825 433 654 551 84.2

0 1678 713 642 457 71.2

0 1405 119 399 447 112.1

49 1482 119 553 558 101.0

0 1447 287 499 506 101.4



24


Fig. 6. Lower-power transmission electron microscope image taken from the superior-temporal region of rabbit bulbar conjunctiva (A) and the same image with the epithelial cell borders and goblet cells (GC) marked. The length of the rectangular box (lower left) indicates 5.5 mm.

well as providing data on goblet cell counts when different methods are used. The studies detailed in this paper are presented to indicate that CIC-based goblet cell counts off rabbit bulbar conjunctiva will likely be highly variable (even unpredictable) unless a specific strategy for cell sampling is adopted. Simply counting goblet cells in a microscope field of view will likely yield numbers that can easily differ 100-fold, even if some averages from repeated counts might appear to only vary by 2- to 3-fold. The box plots are presented to emphasize the heterogeneity in each set of 200 fields examined, and also show how this can change quite substantially in successive sets of counts. The cell sampling strategies that could be adopted involve selecting a minimum goblet cell count in relation to non-goblet cells, or a maximum goblet cell count from very high density yield (multilayered) specimens. Each of these will be discussed in more detail later. Overall, this variability in what have been referred to as random counts of goblet cells in CIC specimens has been noted previously, albeit indirectly. The present studies provide a systematic confirmation of this variability, i.e. direct evidence that such variability is real and likely to be encountered. In human CIC studies, Egbert and colleagues noted that the apparent density of intact goblet cells (based on the number of PASstained spots seen on a filter) was ‘roughly equivalent to the density of the impressions’, but provide no further details (Egbert et al., 1977). It is this outcome of CIC that has been routinely obtained by numerous investigators since the introduction of the technique, i.e. the CIC sample can be expected to include both goblet cells and non-goblet (epithelial) cells. The overall character of CIC samples and the proportions of each type of cell, however, have not been well defined. This overall conclusion has also made in a study on rabbits, where it was noted that ‘the impressions left by cells on successive filters varied widely in intensity’ (Hareuveni and Maurice, 1994). Nelson (Nelson and Wright, 1986), also reporting on goblet cell numbers but in human CIC samples, does comment that “marked variations in goblet cell densities do occur …”, and a similar conclusion can be drawn from rabbit CIC studies (Driot and Donne, 1992) where a strategy was adopted to only use the center part of the filter with multilayered cells. Albietz (2001) acknowledges that ‘variability in superficial conjunctival ‘histology’ across a given (human CIC) sample occurs' but, as with other reports, provides no further detail of the magnitude of the ‘variability’ that was encountered. The current studies are presented as a start to providing some of this detail, especially when different numbers of microscope fields are used for goblet cell counts. These differences or the apparent inconsistency of goblet cell densities in the CIC samples could be because the goblet cell distribution across the conjunctiva is inherently heterogeneous, with

substantial variability from location to location. Alternatively, and the explanation offered in earlier rabbit studies (Driot and Bonne, 1992) and considered to be more likely, is that the CIC sample itself is inherently variable with some regions of multilayering and others of lesser cell densities. This leads to notable differences in the goblet cell numbers at different locations across the specimen. Light microscope evaluation of sections taken through CIC filters (from human subjects) can be used to show how the cell sampling can include just epithelial cells or a mixture of goblet cells and epithelial cells (Doughty, 2012a). The schematics developed for Fig. 7 are designed to further illustrate how these regional differences in CIC specimens could arise. As should hopefully be apparent from the examples presented in this study, if significant numbers of goblet cells are to be seen, then this is associated with a high or very high density of the impressions which are composed of multilayers of cells. The present studies used the mounted Biopore membrane (Millicell-CM) unit because, based on experience, its use can be expected to provide high yields of cell material. This yield, essentially as the area covered by cells, is especially important if objective strategies using counts off multiple fields of view are to be adopted, but with minimum chance of assessing the same or slightly overlapping fields. As illustrated, significant numbers of goblet cells were invariably located within multilayered regions of the cell samples. There will also be other regions across the specimen where there are far fewer goblet cells as a result of the lesser extent of the cell multilayering. Many issues could be considered relevant to any assessment of the numbers of goblet cells in a particular region of a CIC filter. Three inter-related issues will be considered in more detail here. These are to consider what differences in outcome might arise if different sizes or numbers of microscope fields of view are randomly sampled (for undertaking goblet cell counts), the tissue structure-related aetiology of variability in goblet cell numbers (across a discrete region) and the potential use of more specific sampling strategies. Previous investigators using CIC have used a wide range of numbers of different fields of view, and a summary of these numbers can be found elsewhere, where the issue of field size and within-field variability on GCD estimates was empirically considered (Doughty, 2012b). The first issue for goblet cell sampling is that of microscope field size as determined partly by the magnification (objective lens) chosen to undertake and report on goblet cell counts. The final field size, in real terms, will also depend on the optics of the particular microscope imaging system which could produce a round, square or rectangular field. Several rabbit studies have reported using a high magnification (40 objective, final magnification presumed to be 400) to thereby use what might be termed high power fields


Fig. 7. Set of cell profiles expected in a typical sagittal image of bulbar conjunctiva overlaid with a conjunctival impression cytology (CIC) filter (top) and the expected appearances of cells on the filter (now turned up the other way) for when one layer, two layers or three layers of cells were removed from the bulbar conjunctiva onto the filter. GC ¼ goblet cells.

(HPF) to undertake goblet cell counts (Hatchell and Sommer, 1984; Driot and Bonne, 1992; Altinors et al., 2007; Oh et al., 2007; Kim et al., 2011; Li et al., 2012, 2013). However, there are few indications, directly or indirectly, of the actual final size of the socalled HPF. This is a very important issue if the data is only presented as goblet cells/HPF without details of the actual field size since comparisons between different studies could be misleading. It should be considered a necessity to provide details of the actual field size, perhaps indirectly by simply including a scale bar on a representative image used for cell counts. While knowing the numbers of goblet cells/microscope field is useful, the differences in

25

field size can make inter-study comparisons very difficult, which is why data is best presented as GCD/mm2. That the image that is selected is representative of the field assessed is also important since it is also unclear in some published studies whether the images presented include the entire field of view or just part of it (Shellans et al., 1989; Li et al., 2012; Yu et al., 2013). Others have stated using an ‘HPF’ yet elsewhere indicate a 20 objective was used (Hwang and Kim, 2014), or that they used a 200 final magnification (Marino et al., 2005) stated by others (as in the present study as well) to be a medium power field (200 field) (Cai et al., 2011; Qiu et al., 2010). In some reports this information is not obviously provided (Toshida et al., 2007). A 250 magnification has also been used (Shellans et al., 1989), with the latter report also highlighting the issue of the final field size where, despite the rectangular image provided, the field was stated to be 0.14 0.14 mm square. As reviewed elsewhere, there has also been considerable variability for human CIC studies in the reported size of the ‘HPF’ examined (Doughty, 2012b), a variability which makes it extraordinarily difficult to even try to calculate what the standardized goblet cell densities, as cells/mm2, might actually be. Across the various reports on rabbit conjunctiva, the average numbers of goblet cells/‘HPF’ have been reported to be from 55 to 170 (Oh et al., 2007; Xiong et al., 2008; Kim et al., 2011; Li et al., 2012, 2013; Yu et al., 2013), yet very similar numbers have also been reported when ‘MPF’ have been used (Qiu et al., 2010; Hwang and Kim, 2014) and just 12/field when an intermediate magnification was used (Shellans et al., 1989) (although in the latter study, the rabbits had been treated with an artificial tear for 1 day prior to CIC). Even allowing for the ambiguities in what constitutes an HPF or MPF (in terms of the actual area over which goblet cells were assessed), and allowing for the specific GCD reports (in cells/mm2) (Marino et al., 2005; Altinors et al., 2007; Toshida et al., 2007; Cai et al., 2011), the various reports on rabbit conjunctiva could yield GCD estimates of anywhere between 60 and approximately 4500 goblet cells/mm2. This very substantial range in apparent GCD values, similarly found in human CIC studies (Doughty, 2012a), is likely because of the inherent differences and variability in the CIC sampling and assessment methods. A related issue is how consistent any counts of goblet cells might be for a single observer versus two or more observers. A systematic study of the resultant errors that could arise, at least when using marked images, indicates that the net result of any such errors can be expected to be greater if 400 fields are used (Doughty, 2012b), but a specific study comparing the counts from different masked observed could be useful. Overall, the point that can be made is that such variability in random GCD estimates cannot be considered as a useful basis for an objective marker. The second, and closely related, issue is that of the number of microscope fields used to generate goblet cell counts or GCD estimates. As noted in the introduction, an indication of these numbers has been provided in most publications on rabbit CIC studies, but these are neither the same nor has there been any data provided on the variability between the counts. There have also been no obvious indications provided on the impact of using different numbers of microscope fields for the assessments. It is hopefully very evident from the present analyses that the numbers of goblet cells/field can easily be highly variable and that no obvious advantage is gained by undertaking what would likely be presented as ‘random sampling’ of 10 vs. 3 microscope fields. Using very specific strategic objectives for image selection, the variability can actually be much less. From the most optimistic perspective, a case could be made that the inter-sample variability (based on comparing average GCD estimates from ten or three 200 fields) could be somewhere between 25 and 30% if a strategy were adopted (and clearly stated) to only consider a microscope field that included goblet cells. It might well

26


be argued that such a strategy is, of course, the one that has been used in previous studies (despite there being no statement to this effect in the methods provided in the different studies). An equally strong argument could also be made that including microscope field areas without goblet cells in the GCD assessments is a more accurate reflection of the distribution of goblet cells across a CIC filter. However, if this is done then the variability in goblet cell counts will likely exceed 40% and the averaged number of goblet cells will be decreased (based on the probability of encountering ‘0’ fields). This was found, in the present studies, to be around 13% when assessing filters that had substantial coverage with cells, with at least 30 mm2 for repeat counts on ten 200 fields. With such considerations, the possible variability in goblet cell counts can now be objectively addressed even better. The averaged estimates of statistical variability (as the COV) are likely to increase substantially from c. 12e25 % (Tables 1 and 2) to at least 40%, and likely even higher at close to 70% (compare Tables 3 and 4). As calculated from the data obtained in the present studies, the average COV estimates can easily range from 63.7 to 112.1 %. The unequal distributions of goblet cells on the CIC filters obtained from rabbit conjunctiva are likely similar to those that can be found in human CIC specimens (Doughty, 2011b, 2012b), with the net result that a 200 field can have much higher or lower numbers of goblet cells than adjacent regions on the filter. With a ‘HPF’ field being only a small part of a ‘MPF’, it should not be surprising if HPF fields without goblet cells were encountered more frequently (Hatchell and Sommer, 1984), so introducing even further variability in the so-called objective assessments. The selection of the 200 fields for the basis of goblet cell counts in the present study was, in part, based on systematic assessments errors and variability of GCD estimates from high, medium and low power fields in human CIC samples (Doughty, 2012b). Given the variability in goblet cell numbers across a CIC filter, it is appropriate to further consider the aetiology of this phenomenon before outlining possible specialized strategies for trying to deal with it. The TEM-based assessments undertaken for the present study generally confirm the expected ultrastructural features of rabbit conjunctiva with their goblet cells (Aragona et al., 1998; Hatchell and Sommer, 1984; Doughty and Bergmanson, 2004). Using similar CIC and TEM assessments, it can be concluded that the conjunctival cell layers constitute a ‘cuboidal’ and essentially non-stratified (mucous) epithelium and that the more superficial cells of the normal bulbar conjunctiva are very substantially different in ultrastructure from the squamous cells of the adjacent corneal epithelium (Doughty, 2015). Based on such verification, in Fig. 7 is shown a series of diagrams to illustrate the removal of successive layers of cells off the bulbar conjunctiva onto the surface of a CIC filter. For the set of schematics shown, based on a real example from superior bulbar conjunctiva of a rabbit, it is only in the third layer of cells that goblet cells are shown to be removed. It could, however, be considered just as likely that one of the goblet cells became attached with the second layer of epithelial cells. Overall, with just a single layer of superficial cells removed, these could be present essentially in a monolayer when the filter surface is examined from the coronal perspective using a 200 field. Since, as illustrated (Figs. 1A and 7), the conjunctival surface is not perfectly smooth, there could easily be some smaller gaps in the layer of cells where the filter surface was not in uniform intimate contact with the actual conjunctival (cell) surface. Stated another way, it should not be surprising if some spaces are observed between the superficial cells collected onto a CIC filter from a normal healthy conjunctiva, although nearly complete monolayers (in a single field) can be encountered in rabbit sampling as is shown for this study. The presence of some spaces between the epithelial cells can have quite a substantial impact on any estimates of cell density,

even though most cells were of a similar small size. A recent study on rabbits also provides images showing small cells approximately 20 mm in diameter with an N:C ratio stated to be 1:2 (Kim et al., 2011). Some earlier studies also note that normal rabbit epithelial cells to about 10e15 mm in diameter (Hatchell and Sommer, 1984) so it is the cell size that is important, not simply the cell density. Some of these epithelial cells should be in contact with one another (cf. Fig. 1A) and, overall, the cell nuclei should be reasonably well spaced. At the next level of sampling, the most superficial layer of cells is collected and now with some or all of a second layer of cells lying on top of it (as viewed down the microscope). With this extent of collection of material onto the filter, the epithelial cells can be expected to be overlapping each other (i.e. for two layers of cells, Fig. 7). Images taken of sections through CIC filters embedded into resin for conventional histology can be used to show how small regions of the adherent material can be composed exclusively of epithelial cells in one or two layers (Doughty, 2012a). When viewed from above, and with this cellecell overlap (seen in the sagittal perspective in the diagram), the epithelial ell nuclei will be expected to be closer together (or not spaced out as much). If, as just noted, some of the third layer of cells also adhered to regions of the CIC filter, then the cell overlapping and crowding of the cell nuclei will be even greater and some goblet cells will likely be seen on that particular region of the filter. This is essentially what is shown in Fig. 2B. If all three cell layers were collected, or even more layers, the nucleus crowding (and net uptake of stain) might be approaching a limit at which the individual nuclei may or may not be resolved when examining the filter from above. CIC specimens with such multilayers of cells and several goblet cells can also be seen using histological methods (Doughty, 2012a). If further layers of cells were collected onto the CIC filter, with various numbers of goblet cells, then all the stained nuclei of the epithelial cells will merge into an almost confluent (Fig. 1C) or completely confluent (Fig. 1D) background to the non-staining goblet cells. From these perspectives, different strategies for sampling the goblet cells can be better considered. These strategies could be to undertake what some might consider to be random sampling across the filter (regardless of the extent of multilayering), or to select an image suitable for estimates of goblet cells in relation to the numbers of epithelial cells. The latter could be for what might be considered as a minimum count, where the goblet cell density is assessed at an intermediate epithelial cell density level where the individual epithelial cell nuclei can be counted. Alternatively, what might be considered as a maximum count could be undertaken where the goblet cell density is assessed at a very high epithelial cell density where individual epithelial cells cannot be discerned simply because the sample is so substantially multilayered. The present evaluations of goblet cells were undertaken on tissue from young adult rabbits verified as being normal based on overall external eye appearance and objective assessments of the epithelial cells. These cells were all of ‘small’ size with a ‘high’ nucleocytoplasmic ratio. From this perspective, any differences in the goblet cells in the CIC specimens cannot be simply dismissed as being the result of variability in the tissue source. Overall, three slightly different approaches to random sampling yielded mean goblet cell counts for the superior bulbar conjunctiva of rabbits of between 78 and 96/200 field. It should be noted that these are from 200 fields, and not 400 fields, and so are notably different from quite a few previous reports. The equivalent mean GCD estimates ranged from 542 to 670 cells/mm2, which are also numbers that are substantially lower than some estimates that might be made from published values for goblet cell counts off HPF. Notwithstanding, a similar value of c. 500/mm2 has been reported for upper bulbar conjunctiva of rabbits (Toshida et al., 2007). However, while the present estimates might seem reasonably


consistent across a modest number of CIC specimens, the overall inter-sample variability was between 60 and 80%, i.e. a GCD estimate of c. 600/mm2 for superior bulbar conjunctiva cannot be considered as being any better than between 150 and 1050 cells/ mm2. It has been previously noted that only about 75% of HPF fields examined (from control rabbits) included goblet cells, although no specific indication was provided of the actual numbers of goblet cells/‘HPF’ (Hatchell and Sommer, 1984). However, in further assessments made on CIC samples from both normal young adult rabbits and humans, the results indicated that if 400 fields were used ‘to try to assess goblet cell numbers then the probability of encountering goblet cells in a randomly selected field was quite often less than 50% (Doughty, unpublished analyses). As the present studies are designed to illustrate, for the first time, the impact of these zero fields on the estimates of GCD (and its variability) would be expected to be even greater and with the overall averaged numbers of goblet cells being lower. The first strategic alternative to random sampling of microscope fields was to consider goblet cell numbers in relation to a measured epithelial cell density. For this, the nuclei of the epithelial cells had to be just sufficiently separated such that counts can be made. For the present studies, these epithelial cell nuclei counts were made from overlays, but counts of cells or nuclei can also be facilitated using public domain software such as Image J on the uploaded and enlarged images, i.e. an overlay does not need to be drawn but is simply one option, and just undertaking manual counts on an enlarged print could be done as well. Notwithstanding despite this labour-intensive approach (since counting several hundred nuclei can take a considerable time) and the fact that a subjective judgement needs to be made in selecting an image where an ‘intermediate’ epithelial cell density is present, the net outcome appears promising in that the goblet cell counts/ 200 field (mean 58) and GCD estimates (at 399 cells/mm2) are not substantially different from those obtained from random sampling of multiple fields. Furthermore, and of greater importance, is that the averaged inter-sample variability on these goblet cell counts or GCD estimates is closer to 30% (rather than 60e80 %). Images similar to those shown in Fig. 1B can be found in other rabbit studies as well (Murakami et al., 2003; Qiu et al., 2010; Cai et al., 2011). Based on experience with these rabbit samples (and also human CIC specimens, Doughty, unpublished assessments), it should be possible to undertake such estimates of goblet cells/ 1000 epithelial cells on most specimens, and even if the overall yield of cells is modest (i.e. a smaller area on the filter covered with cell material). The method has been reported for human CIC (Yeo et al., 2003). In some specimens however, there may not be a suitable region of ‘intermediate’ epithelial cell density, and these would simply have to be considered as unusable for further assessments. The second strategy was to only count goblet cells where very obvious and substantial multilayering of the cells was present, a method also reported for rabbit CIC (Driot and Bonne, 1992). This previous study (Driot and Bonne, 1992), however, did not provide any GCD estimates but rather just the percentage differences from controls. In the present studies, the goblet cell counts/200 field were about 225 and the GCD estimate was nearly 1600 goblet cells/ mm2, both very substantially higher than the numbers generated from the other methods. As with assessments at intermediate epithelial cell density, these estimates with a very high background of epithelial cells (i.e. very high epithelial cell density) require a subjective judgement to be made that the background staining is indeed confluent. For high yield specimens, this would appear to be possible in many samples but, as with the intermediate density estimates, there may be some specimens that simply do not contain such regions. The same applies to human CIC specimens obtained

27

with the Millicell-CM filter (Doughty, 2011b, 2012b), and it remains largely unknown as to why some samples yield substantially multilayered regions and others do not. Notwithstanding, the present analysis indicate that this approach can provide much less variable estimates of the number of goblet cells in a conjunctiva. The inter-sample variability in these estimates was closer to 10% (see Table 1). With this apparent variability being the lowest when this image selected strategy is used, it is the one for which it would be most useful to undertake further studies on. This also needs to include human CIC samples. It is acknowledged, that while these goblet cell estimates from when there are very substantial numbers of epithelial cells are presented as ‘densities’ (cf. Table 1), the goblet cell counts are clearly from multiple layers of cells and so, to be strictly correct, should be presented as a fraction of a volume of cell material. This issue equally applies to goblet cell counts made from whole mounts of rabbit conjunctival tissue (Waheed and Basu, 1970). Overall, this issue of the density of the CIC specimen in terms of the number of epithelial cells needs to be much more substantially considered. From published images from rabbit CIC studies, there are several very obvious examples where lower numbers of goblet cells (PAS-positive material) were observed when the background number of epithelial cells was lower (and vice versa) (Murakami et al., 2003; Oh et al., 2007; Cai et al., 2011; Li et al., 2012, 2013; Yu et al., 2013). While the reasons(s) for different numbers of epithelial cells being sampled (for a particular region of a filter or the whole filter surface) is really unknown, a change in (apparent) GCD could simply be due to the cell sampling density encountered. Differences in such sampling could also arise simply because a particular region of the conjunctiva was being re-sampled by CIC as has been done in some studies (Shellans et al., 1989; Xiong et al., 2008). In summary, the present study was undertaken to try to generate data that would indicate why goblet cell sampling of the bulbar conjunctiva can be so variable. The identification method chosen for the goblet cells was that of identifying them by their lack of uptake of a cell stain (Giemsa in this study) so as to show as ghost-like, ovoid cells sometimes with a very small nucleus evident (Krenzer and Freddo, 1997; Danjo et al., 1998; Rath et al., 2006; Doughty, 2011a). With this unique appearance, it has not been found difficult to identify the goblet cells whether present at low, intermediate or high numbers. PAS staining has been tried as well (Doughty, unpublished) but the mucin stain is sometimes rather irregular or diffuse (Doughty, 2012a). When very high numbers of PAS-positive overlapping goblet cells are present it becomes very difficult to make reliable counts, and the GCD estimates are likely to be even more variable (Doughty, 2012b). Slight or substantial goblet cell overlap is, however, very much discernable using the Giemsa stain. Overall, the results clearly show that with random sampling of CIC specimens for GCD estimates, the data are unlikely to have any obvious reliability and that specific strategies need to be adopted. Regardless of the strategy for goblet cell assessments that is adopted, the present analyses indicate that it would be very useful, of not considered essential, for investigators to provide unambiguous details of the final magnification and microscope field size for which CIC images were assessed as well as provide information on the intra-sample variability in goblet cell counts made from CIC specimens. It should also be a requirement to state whether the multi-field counts included those with zero goblet cells. The strategies that could be adopted should be to either count the lower or highest goblet cell numbers to yield minimal or maximum GCD estimates from any particular sample. This could be fairly easily accomplished providing some specific criteria are applied to select suitable images from CIC samples.

28


Acknowledgements The author has no proprietary interests in any of the products or procedures outlined in this article. References Albietz, J.M., 2001. Conjunctival histologic findings of dry eye and non-dry eye contact lens wearing subjects. CLAO J. 27, 35e40. Altinors, D.D., Bozbeyoglu, S., Karabay, G., Akova, Y.A., 2007. Evaluation of ocular surface changes in a rabbit dry eye model using a modified impression cytology technique. Curr. Eye Res. 32, 301e307. Aragona, P., Puzzolo, D., Micali, A., Ferreri, G., Britt, D., 1998. Morphological and morphometric analysis on the rabbit conjunctival goblet cells in different hormonal conditions. Exp. Eye Res. 66, 81e88. Baudouin, C., Aragona, P., Van Setten, G., Tlando, M., Ikec, M., Benitez del Castillo, J., Geerling, G., Labetoulle, M., Bonini, S., ODISSEY European Concensus Group members, 2014. Diagnosing the severity of dry eye: a clear and practical algorithm. Br. J. Ophthalmol. 98, 1168e1176. Blades, K., Doughty, M.J., 2000. Comparison of grading schemes to quantitative assessments of nucleus-to-cytoplasmic ratios for human bulbar conjunctival cells collected by impression cytology. Curr. Eye Res. 20, 335e340. Cai, R.R., Wang, Y., Xu, J.J., Zhang, C.R., 2011. The effects of hyperosmotic stress on rabbit ocular surface and mucin 5AC expression (in CHINESE). Zhonghua Yan Ke Za Zhi 47, 252e259. Calonge, M., Diebold, Y., Saez, V., Enriquez de Salamanca, A., Garcia-Vazquez, C., Corrales, R.M., Herreras, J.M., 2004. Impression cytology of the ocular surface: a review. Exp. Eye Res. 78, 457e472. Danjo, Y., Watanabe, H., Tisdale, A.S., George, M., Tsumura, T., Abelson, M.B., Gipson, I.K., 1998. Alternation of mucin in human conjunctival epithelia in dry eye. Investig. Ophthalmol. Vis. Sci. 39, 2602e2609. Doughty, M.J., 2011a. Contact lens wear and the goblet cells of the human conjunctiva e a review. Contact Lens Anterior Eye 34, 157e163. Doughty, M.J., 2011b. Objective assessment of conjunctival squamous metaplasia by measures of cell and nucleus dimensions. Diagn. Cytopathol. 39, 409e423. Doughty, M.J., 2012a. Goblet cells of the normal human conjunctiva and their assessment by impression cytology sampling. Ocul. Surf. 10, 149e169. Doughty, M.J., 2012b. Sampling area selection for the assessment of goblet cell density from conjunctival impression cytology specimens. Eye Contact Lens 38, 122e129. Doughty, M.J., 2012c. Observations on the ultrastructure of equatorial scleral collagen fibrils in sheep eyes. Vet. Ophthalmol. 15, 71e80. Doughty, M.J., 2013. Assessment of goblet cell orifice distribution across the rabbit bulbar conjunctiva based on numerical density and nearest neighbours analysis. Curr. Eye Res. 38, 237e251. Doughty, M.J., 2014. Functional morphology of mucosal goblet cells based on spatial separation of orifice openings to the surface e application to rabbit bulbar conjunctiva. Tissue Cell 46, 241e248. Doughty, M.J., 2015. Assessment of size and nucleo-cytoplasmic characteristics of the squamous cells of the corneal epithelium. Clin. Exp. Optom. http:// dx.doi.org/10.1111/cxo.12253 (in press). Doughty, M.J., Bergmanson, J.P.G., 2004. Heterogeneity in the ultrastructure of the mucous (goblet) cells of the rabbit palpebral conjunctiva. Clin. Exp. Optom. 87, 377e385. Driot, J.-Y., Bonne, C., 1992. Beneficial effects of a retinoic acid analog, CBS-211 A, on an experimental model of keratoconjunctivitis sicca. Investig. Ophthalmol. Vis. Sci. 33, 190e195. Egbert, P.R., Lauber, S., Maurice, D.M., 1977. A simple conjunctival biopsy. Am. J. Ophthalmol. 84, 798e801. Hareuveni, T., Maurice, D.M., 1994. Short-term reproducibility of impression cytology. Cornea 13, 250e252. Hatchell, D.L., Sommer, A., 1984. Detection of ocular surface abnormalities in experimental vitamin A deficiency. Arch. Ophthalmol. 102, 1389e1393.

Hwang, H.B., Kim, H.S., 2014. Phototoxic effects of an operating microscope on the ocular surface and tear film. Cornea 33, 82e90. Kim, J.R., Oh, T.H., Kim, H.S., 2011. Effects of benzalkonium chloride on the ocular surface of the rabbit. Jpn. J. Ophthalmol. 55, 283e293. Krenzer, K.L., Freddo, T.F., 1997. Cytokeratin expression in normal human bulbar conjunctiva obtained by impression cytology. Investig. Ophthalmol. Vis. Sci. 38, 142e152. Li, C., Song, Y., Luan, S., Wan, P., Li, N., Tang, J., Han, Y., Xiong, C., Wang, Z., 2012. Research on the stability of a rabbit dry eye model induced by topical application of the preservative benzalkonium chloride. PLoS One 7, e 33688. Li, N., Deng, X., Gao, Y., Zhang, S., He, M., Zhao, D., 2013. Establishment of the mild, moderate and severe dry eye models using three methods in rabbits. BMC Ophthalmol. 13, 50. Liang, H., Brignole-Baudouin, F., Pauly, A., Riancho, L., Baudouin, C., 2011. Polyquadpreserved travoprost/timolol, benzalkonium chloride (BAK)-preserved travoprost/timolol, and latanoprost/timolol in fixed combinations: a rabbit ocular surface study. Adv. Ther. 28, 311e325. Lopin, E., Deveney, T., Asbell, P.A., 2009. Impression cytology: recent advances and applications in dry eye disease. Ocul. Surf. 7, 93e110. Marino, C., Paladino, G.M., Scuderi, A.C., Trombetta, F., Mugridge, K., Enea, V., 2005. In vivo toxicity of netilmicin and ofloxacin on intact and mechanically damaged eyes of rabbit. Cornea 24, 710e716. Martinez, A.J., Mills, M.B., Jaceldo, K.B., Tio, F.O., Aigbivbalu, I.B., Hilsenbeck, S.B., Yee, R.W., 1995. Standardization of conjunctival impression cytology. Cornea 14, 515e522. Murakami, T., Fujihara, T., Nakamura, M., Nakata, K., 2003. P2Y(2) receptor elicits PAS-positive glycoprotein secretion from rabbit conjunctival goblet cells in vivo. J. Ocul. Pharmacol. Ther. 19, 345e352. Nakamura, M., Endo, K.-I., Nakata, K., Hamano, T., 1998. Gelfarnate increases PAS positive cell density in rabbit conjunctiva. Br. J. Ophthalmol. 82, 1320e1323. Nelson, J.D., 1988. Impression cytology. Cornea 7, 71e81. Nelson, J.D., Wright, J.C., 1986. Impression cytology of the ocular surface in keratoconjunctivitis sicca. In: Holly, F.J. (Ed.), The Preocular Tear Film in Health, Disease, and Contact Lens Wear. Dry Eye Institute, Lubbock, TX, pp. 140e156. Oh, J.Y., In, Y.S., Kim, M.K., Ko, J.H., Lee, H.J., Shin, K.C., Lee, S.M., Wee, W.R., Lee, J.H., Park, M., 2007. Protective effect of uridine on cornea in a rabbit dry eye model. Investig. Ophthalmol. Vis. Sci. 48, 1102e1109. Pauly, A., Roubeix, C., Liang, H., Brignole-Baudouin, F., Baudouin, C., 2012. In vitro and in vivo comparative toxicological study of a new preservative-free latanoprost formulation. Investig. Ophthalmol. Vis. Sci. 53, 8172e8180. Qiu, X.D., Gong, L., Chen, M.J., 2010. Research on the effects of vitaminA palmitate on repair of mechanical corneal epithelial defects and conjunctival goblet cells in rabbits (in CHINESE). Zhonghua Yan Ke Za Zhi 46, 151e160. Rath, R., Stave, J., Guthoff, R., Giebel, J., Tost, F., 2006. In-vivo-Darstellung des Bindehautepithels. Ophthalmologe 103, 401e405. Shellans, S., Rich, L.F., Louiselle, I., 1989. Conjunctival goblet cell response to vasoconstrictor use. J. Ocul. Pharmacol. 5, 217e220. Toshida, H., Nguyen, D.H., Beuerman, R.W., Murakami, A., 2007. Evaluation of novel dry eye model: preganglionic parasympathetic denervation in rabbit. Investig. Ophthalmol. Vis. Sci. 48, 4468e4475. Waheed, M.A., Basu, P.K., 1970. The effect of air pollutants on the eye. I. The effect of an organic extract on the conjunctival goblet cells. Can. J. Ophthalmol. 5, 226e230. Xiong, C., Chen, D., Liu, J., Liu, B., Li, N., Zhou, Y., Liang, X., Ma, P., Ye, C., Ge, J., Wang, Z., 2008. A rabbit dry eye model induced by topical medication of a preservative benzalkonium chloride. Investig. Ophthalmol. Vis. Sci. 49, 1850e1856. Yeo, A.C.H., Carkeet, A., Carney, L.G., Yap, M.K.H., 2003. Relationship between goblet cell density and tear function tests. Ophthal. Physiol. Opt. 23, 87e94. Yu, F., Liu, X., Zhong, Y., Guo, X., Li, M., Mao, Z., Xiao, H., Yang, S., 2013. Sodium hyaluronate decreases ocular surface toxicity induced by benzalkonium chloride-preserved latanoprost: an In vivo study. Investig. Ophthalmol. Vis. Sci. 54, 3385e3393.

Comparison of morphology of bulbar conjunctival cells assessed by impression cytology versus scrape and smear methods.

Impression cytology: a novel sampling technique for conjunctival cytology of the feline eye.

Clinicopathological correlation of hard contact lens related changes in tarsal conjunctiva by impression cytology.

Goblet cells of the conjunctiva: A review of recent findings.

Trichilemmal cyst of the bulbar conjunctiva: a rare presentation.

Sjogren's syndrome: a comparative study of impression cytology of the conjunctiva and buccal mucosa, and salivary gland biopsy.

Mucoepidermoid carcinoma of the bulbar conjunctiva - an interventional case report.

Conjunctival impression cytology.

Impression cytology implicates cell autophagy in aqueous deficiency dry eye.

Optimal surface for impression cytology.

Impression cytology diagnosis of ulcerative eyelid malignancy.

Impression cytology of conjunctival melanosis and melanoma.

Sebaceous carcinoma of the eyelid with Pagetoid involvement of the bulbar and palpebral conjunctiva.

Mascara pigmentation of the bulbar conjunctiva associated with rigid gas permeable lens wear.

Impression Cytology in Different Types of Contact Lens Users.

Cervical cytology: a randomized comparison of four sampling methods.

Goblet cell carcinoid of the appendix.

Goblet cell carcinoid of the appendix.

Laboratory assessment of impression accuracy by clinical simulation.

Spindle Cell Lipoma of the Conjunctiva.

Invasive squamous cell carcinoma of the conjunctiva.

Primary Transitional Cell Carcinoma of the Conjunctiva.

Basaloid squamous cell carcinoma of the conjunctiva.

Conjunctival impression cytology in non-proliferative and proliferative diabetic retinopathy.