Mutagenesis vol. 29 no. 1 pp. 63–71 Advance Access publication 16 December 2013

doi:10.1093/mutage/get064

An automated new technique for scoring the in vivo micronucleus assay with image analysis

Aya Shibai-Ogata, Haruna Tahara, Yusuke Yamamoto, Masaharu Fujita, Hiroshi Satoh, Atsuko Yuasa, Takanori Hioki and Toshihiko Kasahara* Safety Evaluation Center, Fujifilm Corporation, 210 Nakanuma, Minamiashigara-shi, Kanagawa 250-0193, Japan

Received on August 15, 2013; revised on November 15, 2013; accepted on November 19, 2013

The mammalian erythrocyte micronucleus assay is frequently used to assess chemical-induced damage to the chromosomes or the mitotic apparatus of erythroblasts. Because quantitative analysis of micronuclei by microscopy is time consuming and laborious, several automatic scoring methodologies with image analysis have been reported. However, there have been cases in which it was difficult to examine the proportion of polychromatic erythrocytes (PCEs) among total erythrocytes as an index for bone marrow (BM) toxicity, and sample slide preparation has proven to be laborious with existing automatic methods. We developed an automatic scoring system with image analysis for the rodent erythrocyte micronucleus assay using 12-well plates employing high-content screening analyser. In our method, micronucleated PCEs (MNPCEs), PCEs and erythrocytes were identified from three kinds of images: bright field image, fluorescence image with Hoechst 33342, and fluorescence image with propidium iodide. The frequencies of MNPCEs and PCEs were subsequently calculated. A comparison of automatic and manual scoring was carried out using BM and peripheral blood (PB) obtained from mice treated with stepwise doses of mitomycin C. The scores obtained by automatic analysis corresponded to those obtained by manual scoring; the frequencies of MNPCEs in BM and PB obtained by automatic scoring were 132 and 113%, respectively, of those obtained by manual scoring, and the corresponding frequencies of PCEs were 95 and 120%, respectively. Furthermore, we performed five repeats of the examinations of mouse BM and PB treated with mitomycin C or vinblastine sulphate and showed that automatic scoring was equivalent to manual scoring in reproducibility. Meanwhile, the scoring data obtained by manual scoring tended to vary among observers. These results suggest that our automatic scoring system with image analysis is superior to manual microscopy scoring in terms of speed and objectivity, comparable in reproducibility and useful for the in vivo micronucleus assay.

Introduction The in vivo rodent erythrocyte micronucleus (MN) test is a widely accepted assay for assessing the potential of chemicals to induce cytogenetic damage (1,2). The assay evaluates the formation of MN-containing lagging chromosome fragments

© The Author 2013. Published by Oxford University Press on behalf of the UK Environmental Mutagen Society. All rights reserved. For permissions, please e-mail: [email protected].

63

Downloaded from http://mutage.oxfordjournals.org/ at Belgorod State University on December 20, 2013

*To whom correspondence should be addressed. Safety Evaluation Center, Fujifilm Corporation, 210 Nakanuma, Minamiashigara-shi, Kanagawa 250-0193, Japan. Tel: +81 465 73 7440; Fax: +81 465 73 7975; Email: [email protected]

or whole chromosomes in polychromatic erythrocytes (PCEs, immature erythrocytes), as a consequence of chemical-induced damage to the chromosomes or the mitotic apparatus of erythroblasts in bone marrow (BM). When a BM erythroblast develops into a PCE, following the expulsion of the nucleus, the MN that is formed remains in the cytoplasm. The micronucleated polychromatic erythrocytes (MNPCEs) can be observed and quantified in either BM or peripheral blood (PB), and the PCE population among total erythrocytes can be determined for each animal as an indicator of BM toxicity. The conventional approach for quantitative analysis of MN formation by microscopy is laborious to perform and is subjective, leading to variability in results among scorers (3). Furthermore, the procedure limits the number of cells evaluated, particularly in PB, because the prevalence of PCEs is very low. To overcome these disadvantages, automatic techniques have been developed using flow cytometry (2,4–7) and computerized image analysis (8–11). The advantages of both types of automatic systems are speed (which allows the analysis of a large number of cells), objectivity and nonreliance on individual skill for scoring. Image analysis also is advantageous because it allows for the possibility of repeated scoring with stored image data after acquisition and visual checking of results. In the mid-1990s, Parton et al. demonstrated an automatic scoring system that differentiates PCEs and normochromatic erythrocytes (NCEs) based on colour, size and shape using Wright-Giemsa staining (10). They reported that there was no statistically significant difference between MNPCE frequencies in mice BM obtained by manual scoring and their automatic scoring system, although differences in the PCE/ NCE ratios were observed in the group treated with mitomycin C (MMC), which resulted in an increase in the population of transitional cells. Manually scored cells were considered either pink (NCE) or slate blue (PCE) by the observer, and it was a challenge to classify transitional cells as PCE or NCE using manual scoring; they explained that the variability between the manual and automatic systems was derived from human limitations of colour perception. Asano et al. reported an automatic image analysis system for the rodent PB MN assay using acridine orange (AO) supravital staining (11). They showed a good correlation between the MNPCE frequencies obtained by the automatic system and those obtained by manual scoring; however, it was hard to distinguish NCEs because NCEs do not fluoresce in AO supravital staining. For the evaluation of BM toxicity, they classified PCE type I through type IV by scoring red fluorescing pixels, which indicate the level of maturity of PCEs, and the proportion of type I–III PCEs among the total PCE population was used as an indicator of BM toxicity. Each automatic system with image analysis still requires the labour associated with the preparation of the slide specimens. Furthermore, there have been cases in which it has proven difficult to examine the proportion of PCEs among total erythrocytes. To solve these problems, we developed a new automatic image analysis scoring system using 12-well plates for the rodent erythrocyte MN assay using the IN Cell

A. Shibai-Ogata et al. prepared for microscopic scoring. From the remaining BM cell suspensions, the BM erythrocytes were harvested using the cell separation procedure with the cellulose column, as previously described (6,8,9). Approximately 1 ml of BM cell suspension was added to the cellulose column using a pipette. As soon as the cell suspension was fully soaked into the column, 8.3 ml of HBSS was gently added to the column surface and eluted; this was repeated three times. Approximately 25 ml of eluate containing the BM erythrocytes was then centrifuged at 300 × g for 5 min, the supernatant was discarded and the resulting cell pellet was resuspended in FBS–25 mM EDTA (~30–100  µL). The cells were then smeared on 12-well plates (Corning, NY, USA) for image analysis and on glass slides for microscopic observation, fixed with 100% methanol for 10 min, air-dried and stored at room temperature. The IN Cell Analyzer can analyse a glass slide. However, use of a glass slide requires additional time and effort because it is necessary to cover the slide with a cover glass and inclusion; moreover, only four slides can be loaded at one time. Since a 12-well plate does not require these extra time-consuming steps, and 12 samples can be loaded at one time, we chose to use a 12-well plate in this study.

Materials and methods

Staining The BM and PB samples on 12-well plates were stained with 10  µg/ml of Hoechst 33342 (Life Technologies Corporation, CA, USA) and 0.25 µg/ml of PI (Wako Pure Chemical Industries, Ltd) in 1 ml PBS (–) for 30 min. Hoechst 33342 was used for staining of MNs and nuclei of nucleated cells, and PI was used for staining of RNA. After staining, samples were washed with PBS (–) once and 1 ml of PBS (–) was added.

Animals Male ICR mice of 7 weeks of age were purchased from Charles River Laboratories Japan, Inc. (Kanagawa, Japan). Food and water were available ad libitum throughout the experiment. The animal care and experimental procedures were performed in accordance with the Guideline for Animal Experimentation issued by the Japanese Association for Laboratory Animal Science (1987), and the protocol was approved by the Institutional Animal Care and Use Committee of FUJIFILM Corporation. Chemicals and treatment Mitomycin C (MMC; titre: 2 mg/vial) formulation for injection was purchased from Kyowa Hakko Kirin Co., Ltd (Tokyo, Japan) and was evaluated in two independent studies. Vinblastine sulphate (VB; CAS, 143-67-9; purity: 97%) was purchased from Wako Pure Chemical Industries, Ltd (Osaka, Japan). Animals in the test chemical group were treated once a day for 2 consecutive days, then euthanised ~24 h after the previous treatment. Dose levels were based on the findings of a literature review. For comparative image analysis and microscopic analysis of individual mouse BM and PB, groups of six mice were intraperitoneally treated with MMC dissolved in saline at a dose of 0.5, 1.0 or 2.0 mg/kg/day for 2 days. Six mice of the same age were treated with saline to serve as negative controls. For the evaluation of reproducibility of measurements, groups of four mice were intraperitoneally treated with MMC at a dose of 2.0 mg/kg/day or VB at a dose of 1.0 mg/kg/day for 2 days. As negative controls, five mice were treated with saline. Cellulose column preparation The cellulose column was prepared as previously described, with several modifications (6,8,9). A  20-mm disc of microscope-cleaning tissue (type 105, Whatman, Maidstone, UK) was placed at the inside bottom of a 20-ml disposable plastic syringe (Terumo, Tokyo, Japan). To avoid parts of the cellulose mixture escaping through the syringe, a 10-µm disc filter (PALL, NY, USA) was attached to the outlet of the syringe. Equal parts by weight of microcrystalline cellulose—Sigmacell type 50 (Sigma-Aldrich, MO, USA)—and α-cellulose fibre (Sigma-Aldrich) were mixed, and 100 mg/ml cellulose mixture was suspended in HBSS (Wako Pure Chemical Industries, Ltd). Then, 10 ml of the cellulose suspension was added to the 20-ml disposable plastic syringe and washed with 20 ml HBSS (Wako Pure Chemical Industries, Ltd). PB samples Approximately 100 µL of blood was collected into a heparinised syringe from the inferior vena cava under anaesthesia. For microscopic observation, a portion of the blood sample was diluted 2-fold with foetal bovine serum containing 25 mM EDTA (Wako Pure Chemical Industries, Ltd; FBS–25 mM EDTA) and smeared on glass slides. The remaining blood was diluted 16-fold with FBS–25 mM EDTA and smeared on 12-well plates for imaging analysis. The cells were fixed with 100% methanol for 10 min, air-dried and stored at room temperature. BM samples The proximal end of the femur was cut and a 24-gauge needle was inserted at the distal end. The BM cells were flushed with FBS–25 mM EDTA and thoroughly mixed by gentle pipetting. Glass slide specimens of the BM cells were

64

Image acquisition and automatic counting The bright field and fluorescence images of cells stained with Hoechst 33342 (excitation: 350 nm, emission: 461 nm) and PI (excitation: 530 nm, emission: 625 nm) were acquired with the IN Cell Analyzer 2000 using a ×40 objective lens. Typical images are shown in Figure 1. The stored images were analysed using IN Cell Developer Toolbox software (GE Healthcare, UK). Automatic counting is a five-step process: (i) cell recognition, (ii) nuclei recognition, (iii) erythrocyte recognition, (iv) PCE recognition and (v) MN recognition. Each step includes the following procedures.   (i) Cell recognition: identify cellular area with the bright field images. Remove cells within a distance of 50 pixels from the borders of the field. Divide the areas of cells in contact with each other using ‘clump breaking’ processing. Recognise the objects as cellular area based on an area of 20–120 µm2.   (ii) Nuclei recognition: identify nuclei with Hoechst-staining areas. Divide the areas of nuclei in contact with each other using ‘clump breaking’ processing. Recognise objects exceeding 7 µm2 as nuclei. (iii) Erythrocyte recognition: exclude the objects including nuclei identified by process (ii) in cellular area as nucleated cells from the cells identified by process (i), then recognise and count the remaining cells as erythrocytes.  (iv) PCE recognition: identify RNA with PI-staining areas. Divide the areas of RNA in contact with each other using ‘clump breaking’ processing. Recognise and count the objects including RNA–PI staining areas exceeding 10 µm2 as PCEs.    (v) MN recognition: identify Hoechst 33342-stained areas using the ‘vesicle segmentation’ method, which identifies vesicles with diameters of 2000 PCEs. The PCE appearance ratio (PCE%) was calculated as the number of PCEs versus a little >1000 erythrocytes. In our image analysis method, we predefined the number of images taken per well, and the analysis starts automatically following the end of image acquisition for each plate. Since the frequencies of MNPCEs are calculated from the accumulated number of cells that are classified and counted per image until the number of PCEs reaches 2000, the number of PCEs accumulated to calculate the MNPCE frequency often just slightly exceeds 2000.

Downloaded from http://mutage.oxfordjournals.org/ at Belgorod State University on December 20, 2013

Analyzer 2000 (GE Healthcare, UK). Hoechst 33342 was used for DNA staining of the nucleus and MN, and propidium iodide (PI) was used for RNA staining of PCEs. In our method, MNPCEs, PCEs and erythrocytes were identified from three types of images: bright field image, fluorescence image with Hoechst 33342 and fluorescence image with PI; the frequencies of MNPCEs and PCEs were then calculated. To evaluate this automatic system for its capability as an alternative to manual evaluation, we carried out a comparison between scoring results obtained by our automatic system with those obtained by manual scoring and further assessed the reproducibility of both scoring methods. This is the first report of an in vivo MN assay using a high-performance imaging instrument to allow for microplate-based high-content screening.

In vivo micronucleus assay with image analysis

Microscopy manual scoring Glass slides of BM or PB samples were stained with 40  µg/ml AO (Wako Pure Chemical Industries) in PBS (–). Two thousand PCEs were examined for MNPCE% and 1000 erythrocytes were examined for PCE% under a fluorescent microscope (BX51, OLYMPUS, Tokyo, Japan). Some purified BM samples were also examined under AO staining. The reproducibility of image and microscopic analysis At the time of sample preparation for the evaluation of reproducibility, individual mouse BM suspensions flushed from femurs were pooled and mixed well, divided into five aliquots and processed by column purification. The purified BM erythrocyte suspensions were repooled and smeared on 12-well plates for image analysis and on glass slides for microscopy scoring. Individual mouse PB samples were also pooled and smeared. For manual scoring, five specimens were coded separately from the BM or PB pools of vehicle-, MMC- or VB-treated groups, but the observers were not aware that they were identical. The pooled BM and PB samples were examined five times for MNPCEs and PCEs with the automatic system and microscopic manual scoring. Manual scoring with blinding was carried out by four observers. Two thousand PCEs were scored for examination of the frequencies of MNPCEs, and 1000 erythrocytes were scored for those of PCEs by microscopy scoring. In the case of image analysis, a little over 2000 PCEs and 1000 erythrocytes were scored.

Results Two types of tests were conducted: in the first, we compared the data generated by image analysis with the data obtained by standard microscopy analysis, and in the second, we evaluated the reproducibility of the measurements obtained using image analysis and traditional slide scoring. Comparison between automatic and manual scoring of BM and PB The frequencies of MNPCEs and PCEs in the BM and PB samples from individual mice treated with MMC at doses of

0.5, 1.0 or 2.0 mg/kg/day or saline were examined with image analysis and microscopy manual scoring (Table I). Overall, the mean frequency of MNPCEs obtained by image analysis in the BM and PB were 132 and 113%, respectively, of those obtained by microscopy. Individual MNPCE frequency obtained by image analysis in the BM and PB samples were 74–198% and 51–250%, respectively, of those obtained by microscopy. The MNPCE frequencies obtained by image analysis were slightly higher than those obtained by microscopy. The mean frequencies of PCEs obtained by image analysis in the BM and PB were 95 and 120%, respectively, of those obtained by microscopy. Individual PCE frequencies obtained by image analysis in the BM and PB were 68–133% and 65–313%, respectively, of those obtained by microscopy. The mean frequencies of PCEs obtained by image analysis were higher than those obtained by microscopy in the samples from mice treated with 2.0 mg/kg/day MMC. Linear approximation of the individual MNPCE and PCE frequencies between the two methods including controls and MMCtreated samples is shown in Figure 2. The concordance rates of MNPCE frequency for the two methods were 0.84 in BM and 0.78 in PB and those of PCE were 0.85 in BM and 0.81 in PB. The reproducibility of measurements of MNPCEs and PCEs The results of five repeated examinations using automatic and manual scoring are shown in Figure  3 and Table II. The coefficient of variation (CV%) values for MNPCE and PCE frequencies by each method were similar; however, there was variability in scores among observers in manual scoring. Furthermore, for PCE% in PB in the VB-treated group, there was variability in scores within each observer. 65

Downloaded from http://mutage.oxfordjournals.org/ at Belgorod State University on December 20, 2013

Fig. 1.  Bright field and fluorescence microscope images obtained with the IN Cell Analyzer 2000. A: bright field image, B: fluorescence image with Hoechst 33342 staining (arrowhead indicates an MN), C: fluorescence image with PI staining, D: fused bright field and fluorescence images with Hoechst 33342 [blue: micronucleus (MN)] and PI (red: MN and RNA) staining, E: automatic identification of areas of cell (red outline), MN (yellow outline) and RNA (green outline). Automatic identification of types of erythrocytes is as follows. PCE: double outlines of red and green, NCE: single outline of red, MNPCE: small yellow outline inside double outlines of red and green.

MMC (2 mg/kg)

MMC (1 mg/kg)

MMC (0.5 mg/kg)

Control

1 2 3 4 5 6 Mean SD 7 8 9 10 11 12 Mean SD 13 14 15 16 17 18 Mean SD 19 20 21 22 23 24 Mean SD

0.35 0.25 0.25 0.25 0.35 0.35 0.28 0.05 1.30 0.95 1.05 0.50 1.30 0.70 0.97 0.32 2.15 3.25 2.30 3.80 3.45 4.55 2.88 0.79 5.85 4.60 4.80 6.10 5.80 5.30 5.41 0.61

0.44 0.35 0.35 0.30 0.58 0.60 0.36 0.06 1.71 1.00 1.77 0.99 1.60 1.14 1.37 0.37 3.84 5.13 4.46 3.34 3.35 3.35 4.19 0.77 6.68 4.82 3.68 5.33 7.63 6.48 5.77 1.43 Mean

126.67 139.10 138.89 119.11 166.92 171.26 143.66 – 131.91 105.11 168.73 197.82 122.77 163.06 148.23 – 178.53 157.71 193.76 87.89 97.05 73.65 131.43 – 114.24 104.79 76.66 87.36 131.57 122.30 106.15 – 132.37

71.20 68.80 53.20 73.80 69.20 64.80 66.75 9.26 53.40 66.20 68.20 60.60 55.20 51.80 59.23 6.88 49.80 44.80 46.60 57.80 31.00 49.80 49.75 5.75 17.80 19.20 16.40 20.20 27.00 52.60 25.53 13.76 –

64.92 63.29 56.59 66.41 65.10 61.09 62.80 4.33 52.10 44.75 61.40 47.25 44.47 43.14 48.85 6.92 50.20 39.26 38.18 67.13 31.12 54.17 48.69 13.44 19.93 15.59 21.76 17.14 32.55 50.98 26.32 13.48 Mean

Automatic scoring

Automatic scoring/manual scoring (%)

100.80 87.62 81.94 116.14 100.40 108.78 99.28 – 111.94 81.18 132.71 84.84 120.55 96.93 104.69 – 95.36

91.17 92.00 106.38 89.99 94.07 94.28 94.65 – 97.57 67.60 90.03 77.98 80.56 83.28 82.83

0.15 0.10 0.40 0.30 0.10 0.20 0.21 0.12 1.20 0.70 1.15 1.05 1.00 1.25 1.06 0.20 2.70 2.45 2.55 3.25 2.95 3.50 2.90 0.41 5.30 5.10 5.80 6.70 4.40 3.75 5.18 1.04 –

Manual scoring 0.15 0.15 0.45 0.45 0.25 0.20 0.27 0.14 1.05 0.40 1.14 0.75 1.60 1.44 1.06 0.44 5.04 3.10 2.69 3.00 3.55 4.21 3.60 0.88 5.74 4.19 6.29 3.84 4.95 1.90 4.49 1.57 Mean

Automatic scoring

Manual scoring

Manual scoring

Automatic scoring/manual scoring (%)

MNPCE (%)

PCE (%)

MNPCE (%) Automatic scoring

PB

BM

99.85 149.93 111.55 149.55 249.75 99.95 143.43 – 87.37 56.80 99.45 71.36 159.68 115.25 98.32 – 186.66 126.34 105.41 92.17 120.28 120.41 125.21 – 108.38 82.23 108.46 57.38 112.39 50.57 86.57 – 113.38

Automatic scoring/manual scoring (%)

Downloaded from http://mutage.oxfordjournals.org/ at Belgorod State University on December 20, 2013

66

Animal No.

Table I.  MNPCE and PCE scores for mouse BM and PB obtained by automatic and manual scoring

Automatic scoring 4.38 2.58 4.28 3.28 2.52 3.28 3.39 0.80 2.13 3.63 3.24 1.68 1.61 2.66 2.49 0.83 1.59 0.59 1.46 1.54 1.17 3.87 1.70 1.12 0.63 0.24 0.40 0.23 0.70 1.27 0.58 0.39 Mean

Manual scoring 5.60 4.10 5.00 3.80 3.80 3.20 4.25 0.88 2.10 2.00 3.80 1.60 2.20 2.30 2.33 0.76 1.40 0.90 1.20 1.20 1.40 4.40 1.75 1.31 0.20 0.10 0.50 0.10 0.60 0.80 0.38 0.29 –

PCE (%)

78.30 62.86 85.56 86.23 66.40 102.45 80.30 – 101.40 181.58 85.23 105.25 73.36 115.73 110.42 – 113.22 65.48 121.98 128.67 83.30 87.86 100.08 – 313.36 237.33 79.94 228.15 116.27 158.99 189.01 – 119.95

Automatic scoring/manual scoring (%)

A. Shibai-Ogata et al.

In vivo micronucleus assay with image analysis

Discussion We developed a novel technique for scoring the in vivo MN assay using a high-performance image analyser for high-content screening. Image analysers, such as the IN Cell Analyzer 2000, can provide multicolour imaging of the phenomena arising in individual cells, measure biochemical information, simultaneously analyse morphological parameters and measure the expression and localization of multiple markers within the same cell (12–14). Methods using these types of image analysers have enabled in vitro MN assays using cultured cells to be run accurately and quickly (15,16). Our method allows the identification of MNPCEs, PCEs and total erythrocytes (PCEs + NCEs) by the use of a bright field image, DNA-stained image with Hoechst 33342 fluorescence and RNA-stained image with PI fluorescence. An image of the identification and classification is shown in Figure 1, panel E. In our approach, the cell is identified from the bright field image, the nucleus and MN are identified from the DNA-stained image with Hoechst 33342 fluorescence based on size and shape, and the cell without nucleus is identified as an erythrocyte. Erythrocytes with or without PI staining in PI-RNA fluorescence imaging are classified as PCEs or NCEs, respectively, and the PCEs including MNs in the cell-recognising area are classified as MNPCEs. The MNPCEs, PCEs and total erythrocytes are quantified, and the numbers in each field are displayed automatically. The analysis of the bright field image and the two fluorescence images enables the correct differentiation and quantification of PCEs and NCEs, a process that had been difficult to accomplish using conventional image analysis with fluorescent staining.

AO is a widely used fluorescent dye for microscopic observation of MNs. To select a fluorescent dye, we carried out automatic counting using AO staining, but in the evaluation of PB samples, it was difficult to detect a difference in PCE frequencies between controls and groups treated with a BM-toxicityinducing agent (data not shown). AO molecules intercalating into double-stranded DNA emit green fluorescence and those binding to single-stranded RNA cause stacking and emit red fluorescence (6). Because interaction between neighbouring AO molecules causes red fluorescence, cells with low concentrations of RNA emit low levels of fluorescence. Since the PCEs in PB have low concentrations of RNA due to RNA loss through the process of maturity, the low levels of induced AO– RNA fluorescence showed little difference in intensity relative to the background, and it was thought that the correct identification and counting of PCEs had failed. Harada et al. showed that the in vivo MN assay with Hoechst 33258 and PI staining using flow cytometry is able to clearly distinguish MNPCEs, PCEs and NCEs (6). We confirmed that an automatic image analysis system using Hoechst 33342 and PI staining and using an image analyser is able to classify them clearly. Table I and Figure 2 show the frequencies of MNPCEs and PCEs in BM and PB samples obtained from mice treated with stepwise doses of MMC as determined by manual and automatic scoring; good correspondence and high correlation can be seen. However, the scores obtained with the automatic system tended to be slightly higher, in terms of MN frequency, than those obtained by the manual method. In particular, the mean frequencies of PCEs obtained by our image analysis were higher than those obtained by microscopy in the samples of mice treated with 2.0 mg/kg/day MMC. Although the exact reason for this is not 67

Downloaded from http://mutage.oxfordjournals.org/ at Belgorod State University on December 20, 2013

Fig. 2.  Linear approximation of MNPCE and PCE frequencies in mouse BM and PB: automatic image analysis scoring compared with manual scoring.

A. Shibai-Ogata et al.

68

and were incorrectly identified as MNPCEs. Although these mistakes occurred infrequently, they may have led to the differences in MNPCEs found between the automatic and manual scoring systems in the control group, in which the frequency of MNPCEs was low. We believe that it is not a problem for the reliability of the results, because the frequencies of mistakes are in fact not more than a few hundredths of a per cent (data not shown). Instead, flow cytometry may have been a greater factor for the reliability than the image analysis because it is difficult to check whether mistakes occurred in detection of MN owing to the inability to confirm them by imaging. In the evaluation of the variability of data for repeated scorings, CV% values of frequencies of MNPCEs and PCEs were similar for the manual and automatic scoring methods (Figure 3 and Table II). Dertinger et al. performed an evaluation of inter- and intra-laboratory variability for established microscopy-based scoring methods and the flow cytometric technique through the analysis of replicate specimens of BM and PB obtained from rats treated with vehicle or cyclophosphamide (CP) as a genotoxic substance (4). In their report, intra-laboratory CV% values for MNPCE% for controls and the CP-treated group by conventional microscopy scoring with AO staining were 31.5–173.2% and 5.4–28.4%, respectively, in BM

Downloaded from http://mutage.oxfordjournals.org/ at Belgorod State University on December 20, 2013

clear, it may be because there were too few PCEs for BM toxicity. Asano et al. indicated that the MN scores obtained with manual scoring tended to be higher than those obtained by automatic image analysis (11). They suggested that the manual analysis might be subject to bias due to scoring of MNPCEs rather than that of mature erythrocytes. We assume that the bias pertains to the influence of the AO staining conditions and individual judgement, which may make it more difficult to distinguish between NCE and PCE. Meanwhile, Harada et al. showed that the MN values obtained by automatic scoring using flow cytometry were higher than those obtained by manual scoring (6). They suggested that this effect was a consequence of the higher sensitivity of flow cytometry and that this is generally to be expected for flow cytometry scoring. The following two causes may account for the higher MN frequencies observed by our automatic image analysis. First, it might be suggested that automatic scoring has better sensitivity and can detect very small MNs that manual scoring fails to detect. Second, we visually checked the images of individual cells that were scored as MNPCEs for any mistakes and confirmed that a few cases of false scoring had occurred. A small number of erythroblasts or particles that remained after erythrocyte purification overlapped with PCE recognition areas

In vivo micronucleus assay with image analysis

Downloaded from http://mutage.oxfordjournals.org/ at Belgorod State University on December 20, 2013

Fig. 3.  The frequencies of MNPCEs and PCEs repeatedly examined five times in pooled (A) BM and (B) PB samples by automatic image analysis scoring and microscopy manual scoring. BM and PB samples were obtained from mice treated with saline or MMC at a dose of 2.0 mg/kg/day or VB at a dose of 1.0 mg/ kg/day for 2 days. Manual scoring was carried out by four observers (A–D on the x-axes), and individual results are shown. Bars indicate the means of the five measurements.

Table II.  The variability in MNPCE% and PCE% values obtained by manual and automatic scoring BM

PB

MNPCE (%)

Control

MMC

VB

Mean SD CV% Mean SD CV% Mean SD CV%

PCE (%)

MNPCE (%)

PCE (%)

Manual scoringa

Automatic scoring

Manual scoringa

Automatic scoring

Manual scoringa

Automatic scoring

Manual scoringa

Automatic scoring

0.14 0.07 50.26 9.24 0.65 7.35 3.19 0.38 12.98

0.28 0.13 48.61 9.07 0.51 5.57 1.94 0.46 23.80

62.82 2.27 3.67 23.18 2.02 8.76 15.36 2.78 17.43

52.21 2.52 4.82 25.22 3.15 12.51 17.71 2.42 13.64

0.20 0.08 39.36 5.18 0.70 14.16 3.67 0.57 16.61

0.25 0.14 56.37 2.88 0.63 21.83 2.10 0.39 18.43

3.90 0.36 9.74 0.60 0.13 22.54 0.41 0.27 63.20

2.97 0.33 11.02 0.34 0.18 52.48 0.40 0.10 24.71

Mean, SD and CV% for manual scoring are shown as the means of individual scores taken by four observers.

a

69

A. Shibai-Ogata et al.

70

consider flow cytometry to be superior in terms of counting speed and more suitable than image analysis for PB. The relationship between sample size (the number of PCEs ranging from 2000 to 1 000 000 cells per animal) and statistical power of the flow cytometry-based PB MN assay was examined in a study (18). Based on that analysis, ≥20 000 PCEs were sufficient to obtain reliable data for the evaluation of MN induction by the flow cytometry-based analysis. In our study, the variability of data were similar for the manual scoring and the image analysis when 2000 PCEs were scored for the examination of the frequencies of MNPCEs by both scoring methods (Figure 3 and Table II). To evaluate the capability of image analysis for the detection of the minimal MN response, it will be required to determine the minimum sample size to ensure that scoring error is maintained below the level of variation among individual animals. And we have not evaluated MNPCEs in rat PB and subchronic exposure using our image analysis system. We will validate whether our method can be used to accurately evaluate the frequencies of MNPCEs and PCEs in BM and PB in subchronic exposure tests of mice and rats. Validation studies using a wide variety of chemicals are necessary for further evaluation of our automatic image analysis system. Finally, the results of this study suggest that automatic image analysis scoring can be used for scoring the in vivo MN assay. Acknowledgements We thank Mr Gen Takata and Mr Taro Nakazawa (GE Healthcare) for technical support with the IN Cell Analyzer 2000. We also thank Ms Miyuki Shimoda, Ms Hiroe Yoshizawa, Ms Nanako Aihara, Dr Hirokazu Kurihara, Mr Masataka Komatsu, Dr Chie Emuta and Dr Taisuke Baba for their cooperation and contributions. Conflict of interest statement: None declared.

References 1. Hayashi, M., MacGregor, J. T., Gatehouse, D. G., et  al. (2000) In vivo rodent erythrocyte micronucleus assay. II. Some aspects of protocol design including repeated treatments, integration with toxicity testing, and automated scoring. Environ. Mol. Mutagen., 35,234–252. 2. Fiedler, R. D., Weiner, S. K. and Schuler, M. (2010) Evaluation of a modified CD71 MicroFlow method for the flow cytometric analysis of micronuclei in rat bone marrow erythrocytes. Mutat. Res., 703, 122–129. 3. Fenech, M., Bonassi, S., Turner, J., et al.; HUman MicroNucleus project. (2003) Intra- and inter-laboratory variation in the scoring of micronuclei and nucleoplasmic bridges in binucleated human lymphocytes. Results of an international slide-scoring exercise by the HUMN project. Mutat. Res., 534, 45–64. 4. Dertinger, S. D., Bishop, M. E., McNamee, J. P.et  al. (2006) Flow cytometric analysis of micronuclei in peripheral blood reticulocytes: I.  Intraand interlaboratory comparison with microscopic scoring. Toxicol. Sci., 94, 83–91. 5. MacGregor, J. T., Bishop, M. E., McNamee, J. P.et al. (2006) Flow cytometric analysis of micronuclei in peripheral blood reticulocytes: II. An efficient method of monitoring chromosomal damage in the rat. Toxicol. Sci., 94, 92–107. 6. Harada, A., Matsuzaki, K., Takeiri, A., Tanaka, K. and Mishima, M. (2013) Fluorescent dye-based simple staining for in vivo micronucleus test with flow cytometer. Mutat. Res., 751, 85–90. 7. Elhajouji, A. and Lukamowicz-Rajska, M. (2013) Flow cytometric determination of micronucleus frequency. Methods Mol. Biol., 1044, 209–235. 8. Romagna, F. and Staniforth, C. D. (1989) The automated bone marrow micronucleus test. Mutat. Res., 213, 91–104. 9. Frieauff, W. and Romagna, F. (1994) Technical aspects of automatic micronucleus analysis in rodent bone marrow assays. Cell Biol. Toxicol., 10, 283–289.

Downloaded from http://mutage.oxfordjournals.org/ at Belgorod State University on December 20, 2013

and 0–100% and 10–16.6%, respectively, in PB. Intra-laboratory CV% values of PCE% in the control and CP-treated groups by conventional microscopy scoring with AO staining were 1.5–4.9% and 1.9–6.1%, respectively, in BM and 4.2–13.5% and 8.2–27.8%, respectively, in PB. Intra-laboratory CV% values for MNPCE% of the control and CP-treated groups in PB determined by the flow cytometric technique with anti-CD71fluorescein isothiocyanate and PI staining were 24.8–32.9% and 6.5–6.8%, respectively, and in terms of PCE%, CV% values were 1.18–2.42% and 0.75–9.88%, respectively. As shown in Table II, CV% values for MNPCE% of the control, MMCtreated and VB-treated groups by automatic image scoring were 48.61, 5.57 and 23.80%, respectively, in BM and 56.37, 21.83 and 18.43%, respectively, in PB. The CV% values for PCE% for the control, MMC-treated and VB-treated groups by automatic image scoring were 4.82, 12.51 and 13.64%, respectively, in BM and 11.02, 52.48 and 24.71%, respectively, in PB. The comparison between the CV% values of our image analysis scoring and the manual scoring performed by Dertinger et al. suggests that the scoring reproducibility of automatic image analysis is comparable to that of manual scoring. The acquisition by the flow cytometric technique was set such that 20 000 PCEs were analysed per blood sample, whereas our system analysed 2000 PCEs for MNPCEs and 1000 erythrocytes for PCEs; therefore, lower standard deviation and CV% were observed in the flow cytometry analysis. Furthermore, there was variability in scores among observers for manual scoring, which was possibly due to variation in subjective judgements. Moreover, for PCE% for the VB-treated group in PB, there was variability among scores for each observer. It was thought that the BM toxicity of VB induced an increase in the population of AO–RNA weak fluorescent cells in PCEs, and therefore it was difficult to distinguish PCEs from NCEs. In analysis of these samples, automatic scoring had better reproducibility and lower variability. Our image analysis method is available for in vivo MN assays of BM and PB; however, in situations where severe BM toxicity induces a dramatic decrease in PCEs in PB, as well as in aging rats, the small number of PCEs in each field would require the acquisition of many images and substantial time for analysing. Flow cytometry might be more suitable for analysis of such samples, because it allows analysis of a large number of cells continuously and at high speed. For MN assays using PB, flow cytometry has been validated internationally (17) and is recognized as a method that is more mainstream than image analysis. Our image analysis method made it possible to analyse several thousand cells per minute by introducing a high-content screening analyser; however, the analysis is not as fast as flow cytometry. However, image analysis has some inherent advantages. The acquisition images can be stored, can be subsequently checked for accuracy and are acceptable for good laboratory practice. In our method, samples are smeared onto multiwell plates, undergo automatic and continuous loading to the analyser via an automation robotics system and again undergo automatic image acquisition and analysis. Therefore, our image analysis method requires less time for scoring than traditional slide counting. Scoring with flow cytometry using an autosampler does not take any more time than our method using an image analyser; however, sample preparation is more time consuming and costly because the CD71-antibody reaction and centrifugation are absolutely essential for sample preparation in flow cytometry. We set the instrument to count 2000 PCEs with reference to OECD TG474. If the number of cells counted is >2000, the statistical power will increase. We

In vivo micronucleus assay with image analysis 10. Parton, J. W., Hoffman, W. P. and Garriott, M. L. (1996) Validation of an automated image analysis micronucleus scoring system. Mutat. Res., 370, 65–73. 11. Asano, N., Katsuma, Y., Tamura, H., Higashikuni, N. and Hayashi, M. (1998) An automated new technique for scoring the rodent micronucleus assay: computerized image analysis of acridine orange supravitally stained peripheral blood cells. Mutat. Res., 404, 149–154. 12. Thomas, N. (2010) High-content screening: a decade of evolution. J. Biomol. Screen., 15, 1–9. 13. Krylova, I., Kumar, R. R., Kofoed, E. M. and Schaufele, F. (2013) A versatile, bar-coded nuclear marker/reporter for live cell fluorescent and multiplexed high content imaging. PLoS One, 8, e63286. 14. Chan, G. K., Kleinheinz, T. L., Peterson, D. and Moffat, J. G. (2013) A simple high-content cell cycle assay reveals frequent discrepancies between cell number and ATP and MTS proliferation assays. PLoS One, 8, e63583.

15. Shibai-Ogata, A., Kakinuma, C., Hioki, T. and Kasahara, T. (2011) Evaluation of high-throughput screening for in vitro micronucleus test using fluorescence-based cell imaging. Mutagenesis, 26, 709–719. 16. Diaz, D., Scott, A., Carmichael, P., Shi, W. and Costales, C. (2007) Evaluation of an automated in vitro micronucleus assay in CHO-K1 cells. Mutat. Res., 630, 1–13. 17. Hayashi, M., MacGregor, J. T., Gatehouse, D. G. et  al.; In Vivo Micronucleus Assay Working Group, IWGT. (2007) In vivo erythrocyte micronucleus assay III. Validation and regulatory acceptance of automated scoring and the use of rat peripheral blood reticulocytes, with discussion of non-hematopoietic target cells and a single dose-level limit test. Mutat. Res., 627, 10–30. 18. Asano, N., Torous, D. K., Tometsko, C. R., Dertinger, S. D., Morita, T. and Hayashi, M. (2006) Practical threshold for micronucleated reticulocyte induction observed for low doses of mitomycin C, Ara-C and colchicine. Mutagenesis, 21, 15–20.

Downloaded from http://mutage.oxfordjournals.org/ at Belgorod State University on December 20, 2013

71

An automated new technique for scoring the in vivo micronucleus assay with image analysis.

The mammalian erythrocyte micronucleus assay is frequently used to assess chemical-induced damage to the chromosomes or the mitotic apparatus of eryth...
3MB Sizes 0 Downloads 0 Views