versión impresa ISSN 0301-732X
Arch. med. vet. v.37 n.2 Valdivia 2005
| Arch. Med. Vet. 37, Nº 2, 2005, págs. 147-54 |
Di- and triploid erythrocyte identification by multi-parameter image analysis: A new method for the quantification of triploidization rates in rainbow trout (Oncorhynchus mykiss)
Identificación de di- y triploidización por análisis multiparamétrico de imágenes: Un nuevo método para la cuantificación de la tasa de triploidización en trucha arcoiris (Oncorhynchus mykiss)
S Härtel1*, R Rojas2, C Räth3, M I Guarda4, O Goicoechea5
1 Centro de Estudios Científicos (CECS). 2 Xperts Ltda., Yerbas Buenas. 3 Max-Planck-Institut für Extraterrestrische Physik. 4 Institute of Physics. 5 Institute of Embryology, Universidad Austral de Chile.
La creciente competencia internacional ha forzado a la industria del salmón a la incorporación de técnicas innovadoras. El cultivo de hembras triploides tiene múltiples ventajas sobre poblaciones diploides. En la actualidad, no existe un método simple, exacto y de bajo riesgo para la cuantificación de tasas de triploidización. En este trabajo presentamos un método que combina microscopía de campo claro convencional (con marcación GIEMSA) con el análisis multiparamétrico de imágenes, denominándolo como microscopía morfológica cuantitativa (QMM). Se utilizó citometría de flujo (FC) como un método de referencia para determinar el contenido de ADN en eritrocitos diploides y triploides extraídos de truchas arco iris inmaduras (Oncorhynchus mykiss). Además, se aplicó microscopía de fluorescencia cuantitativa (QFM), usando los marcadores de ADN: 4,6-diamidina-2-fenilindol (DAPI), Yoduro de Propidio (PI), y Naranja de Acridina (AO). Nuestros datos muestran que QMM posee una capacidad discriminante comparable o incluso superior a FC o QFM. El método desarrollado ofrece una nueva perspectiva para clasificar objetos microscópicos, con muchas posibles aplicaciones.
Key words: triploidization, cytometry, microscopy, image processing.
Palabras clave: triploidización, citometría, microscopía, imagen procesada.
Growing international competition is forcing salmon farmers to incorporate innovative techniques into the production process. The use of triploid, all-female breeding populations offers multiple advantages over diploid populations. Currently, an exact, simple, and non- hazardous method for the quantification of diploid- and triploid salmon erythrocytes does not exist. We present a method that combines a standard microscopic bright field technique (contrast staining with GIEMSA) with multi-parameter image analysis and termed it quantitative morphologic microscopy (QMM). We used flow cytometry (FC) as the reference method to determine the DNA content of di- and triplod erythrocytes from immature rainbow trout (Oncorhynchus mykiss). Additionally, we applied quantitative fluorescence microscopy (QFM), using the DNA stains 4',6-diamidino-2-phenylindole (DAPI), propidium iodide (PI), and acridine orange (AO). Our data show that QMM possess comparable or even superior discriminating capacities than FC or QFM. The developed method opens a perspective for the classification of microscopic objects with many possible applications.
The economic sitution has forced the salmon industry to maximize production rates, improve fillet quality, and minimize breeding costs. The use of all-female, triploid lines provides advantages in comparison to non-sterile diploid females, male, or sexually mixed populations (Utter et al 1983, Wlasow et al 2004, Friars et al 2001). Female triploid salmons do not develop ovaries, nor do they express the characteristic phenotypic changes of the sexual maturation of the diploid species (Happe et al 1988). Methods for triploidization include the application of high pressure or hyperthermia after the fertilisation procedures. Presently, high pressure treatment is preferred because it yields better triploidization and surviving rates. Since young di- and triploid fishes cannot be distinguished by their phenotypes, triploidization rates are determined on a cellular basis with different protocols (Thititananukij el al 1996, Boron 1994, Child and Watkins 1994, Wattendorff 1986, Ewing and Scalet 1991, Cozier and Moffett 1989, Al-Sabti 1995, Thomas and Morrison 1995). Due to its precision, the reference method for the quantification of the nuclear DNA content is based on fluorescent staining of the cellular chromatin in combination with flow cytometry (FC) (Utter et al 1983). The main disadvantages of FC are high costs and the handling of hazardous DNA binding fluorescent dyes, such as acridine orange (AO), propidium iodide (PI), or 4',6-diamidino-2-phenylindole (DAPI). So far, an easy, precise, and non-hazardous method for the quantification of triploidization rates has not been developed. Applied methods are based on the fish to fish determination of triploidy in erythrocytes by trained specialists who classify up to 100 samples in order to obtain statistically significant results. In this work we present a microscopic bright field technique for the classification of di- and triploide erythrocytes of O. mykiss. This technique takes advantage of image processing routines in combination with multiple component analysis. The results obtained with quantitative morphologic microscopy (QMM) or quantitative fluorescent microscopy (QFM) are comparable or even superior to the FC-data. QMM substitutes fluorescence parameters for pure morphologic features of the erythrocyte nuclei which can be contrasted by simple GIEMSA staining procedures. QMM offers a precise, non-hazardous, inexpensive, and relatively simple method for classification of di- and triploid erythrocytes of O. mykiss.
MATERIAL AND METHODS
Triploidization method & collection of blood samples: Triploidization of rainbow trout (Oncorhynchus mikiss, Smith and Stearly 1989) was induced by hyperthermia (28ºC for 10 min), 40 min after fertilisation. Blood samples (1 ml) were collected from immature fishes (350 ± 50 g) by piercing the caudal vein. Heparin (Sigma) was added to inhibit blood coagulation.
Discrimination of di- and triploid erythrocytes by flow cytometry (FC): Following the method described by Darzynkiewicz et al (1979), 100 ml blood suspension (105 erythrocytes) was mixed with 200 ml of acid detergent (0.1 % Triton X-100, 0.08 N HCl, 150 mM NaCl). After 30 s, 600 ml of the acridine orange (AO) staining solution was added (100 ml contained 1.2 mg AO (Sigma), 29 mg ethylene-diamine-tetra-acetic acid (EDTA), 150 mM NaCl, 0.1 M C6H8O7H2 O, and 0.2 M Na2HPO4). All steps were performed in a 0°C ice water bath. After 10 min, the DNA content of 5000 cell nuclei was determined by FC as described by Ojeda et al (1992). Excitation of DNA-bound AO was performed with a 50 mW laser at 488 nm, while the emission intensity was recorded at 515-575 nm for each individual cell. Additionally, side scattering (90°) and forward scattering were recorded for each cell. The setup of the FC, designed at the Institute of Physics of the Universidad Austral de Chile by Flavio Ojeda, is described in Ojeda et al (1992).
Discrimination of di- and triploid erythrocytes by quantitative fluorescence microscopy (QFM): Blood suspensions (5 ml) were pipetted onto different microscopic slides. Erythrocyte nuclei were stained with AO staining solution (see above), propidium iodide (PI) (Molecular Probes, Eugene, OR, USA) staining solution (50 mg/ml), or 4',6-diamidino-2-phenylindole dihydrochloride (DAPI) staining solution (50 mg/ml). For AO and PI staining, 5 ml of acid detergent was added to the slides in order to permeabilize the plasma membranes of the erythrocytes. After 5 min, 5 ml of the corresponding staining solutions were added. For DAPI staining, 5 ml staining solution was added directly to the erythrocytes, since the small dye diffuses freely into the chromosomal region. After an incubation time of 10 min, samples were covered with slide covers and observed on a Leica epifluorescence microscope DM LB, equipped with a type A filter set and a CS 50 W-4/L2 lamp for the fluorescent and the bright field mode. Gray scale images (8-bit, 701 ´ 480 pixels) were recorded using a CCD camera (CE, Model OS70D HR, Taiwan) and GrabIT frame grabber soft- and hardware (AIMS Lab., USA).
Discrimination of di- and triploid erythrocytes using bright field quantitative morphological microscopy (QMM): Small aliquots of blood samples were spread out on microscopic slides and dried at room temperature. Samples were covered with May Grünwalds solution (SIGMA) for 1 min, covered with aqdest for 3 min, washed with aqdest, covered with GIEMSA solution (SIGMA, 50 ml/ml aq) for 15 min, and washed with aqdest. Finally, samples were covered with cover slides and images were recorded in the bright field mode with the microscopic setup described above.
Segmentation and parameterization of erythrocyte nuclei: Interactive routines for cellular image analysis and multivariate data processing (SCIAN) were written in IDL® 5.4 (Interactive Data Language, Research Systems, CO, USA). SCIAN (www.scian.cl) contains a library of different filters which allow the segmentation of complex structures (Fanani et al 2002, Härtel et al 2005), cells (Jessel et al 2002, Härtel et al 2003), or higher biological samples (Alvarez et al 2004/2005). For fluorescent images, threshold segmentation in the intensity-histogram yielded excellent results. For bright field images, either threshold segmentation in the intensity-histogram (figure 1) or a modified scaling index method (SIM) was applied. SIM was originally introduced by Räth and Morfill (1997) as a pixel-wise non linear transformation. We introduced the weighting vector p in  and  in order to optimize the segmentation results:
Vector is defined at each picture position (x,y) as = [x, y, I(x,y)]. Q Is the Heaviside Function and || the Euclidean Norm. We defined r1 and r2 with 4 parameters [r'1 < r'2, f1 £ f2 ], yielding (r1/2)2 = (r'1/2)2 + (f1/2)2. S Is defined for (xi - x)2 + (yi - y)2 £ r'2. Nuclei are segmented by selecting a threshold value in the transformed picture a(x,y). Consecutive erosion and dilation operators correct remaining segmentation defects. Debris and nuclei which touch the image borders are removed by morphological filters which identify the pure nuclei population (figure 1b/c). For each segmented nucleus, morphological and intensity dependent parameters were calculated (table 1). Ten pictures containing 1500 to 4500 nuclei were used for QFM and QMM.
Principal Component Analysis (PCA) of FC-, QFM-, and QMM-data: Fluorescent and morphologic parameters were determined for di- and triploids erythrocyte populations as described above. The parameters Pi of the di- and triploid population were centred and normalised by . Eigen Values (l) and Eigen Vectors (V) were calculated for each cell population (table 2). The probability that the members of each cell population are sorted correctly into the corresponding group (a and b) was quantified by discriminating capacities (CD, D = X/X,Y/X,Y,Z for 1/2/3 dimensions, figure 2):
(da/b)D represent normalised density distributions for group a and b. CD becomes 1 for entirely separated density distributions and 0.5 for identical distributions. We further calculated the signal to noise ratios (SNRDI/TRI) of the most discriminant parameters of FC, QFM, and QMM:
For the classification of unknown cell populations, parameters were transformed into the coordinate system of each calibration set (a and b). Then, the Mahalanobis Distance (MD) (Taguchi and Jugulum, 2002) was calculated for each cell in respect to each coordinate system and the minimum distance criteria min(MDa,MDb) was used to classify the cell into group a or b.
Segmentation of erythrocyte nuclei of O. mykiss from digital images was achieved in the bright field mode and in the fluorescent mode of the microscopic setup. Figure 1 shows the basic steps of a representative segmentation of GIEMSA-stained triploid nuclei which are also valid for the segmentation of diploid nuclei (results not shown). First, a threshold-interval is selected interactively in the intensity-histogram of the digital image (figure 1a and 1b, inset). The segmented regions of interest (white pixels in figure 1b) include erythrocyte nuclei, but also cellular debris and membrane residues. Two morphologic filters based on object size and circularity (shape sensitive parameter P2/A, see table 1) separate the nuclei from the undesired structures (figure 1c). The insets of figure 1c show that cellular debris and membrane residues are smaller and less circular (high P2/A-values) than the nuclei.
Morphological and intensity dependent parameters of segmented nuclei were calculated and the data were analysed by multi-parameter techniques (table 1 and 2, figure 2). As figure 2 shows, di- and tri-ploid (*) erythrocyte populations were separated by FC (figure 2a), QFM (figure 2b), and QMM (figure 2c). The results are plotted in bivariate histograms with area normalised density distributions. All three methods clearly distinguish between the diploid and the triploid erythrocyte population. The corresponding CD-values were calculated for 1-, 2-, and 3-dimensions and plotted next to each histogram. The CX-values are highest for the FC-data (AO-intensity), followed by the QMM-data (area), and the QFM-data (DAPI-intensity). For QFM, staining with DAPI, PI, or AO yielded comparable results (not shown). For the 2- and 3-dimensional analysis, the CD-values are almost identical. The calculation of the SNR with respect to the most discriminant parameters yielded: SNRDI/TRI = 4.5/3.1 (AO-intensity for FC), SNRDI/TRI = 2.5/2.1 (DAPI-intensity for QFM), and SNRDI/TRI = 4.0/2.0 (area for QMM).
Additional insight in the interdependence between the parameters is obtained by multi parameter analysis of the FC-, QFM-, and QMM-data (table 2). The correlation coefficients (r) between the X-, Y-, and Z-axis parameters (figure 2) are plotted in the first row of table 2. The correlation coefficient for the X/Y-axis of the FC-data is higher than the correlation coefficients of the QFM- and the QMM-data. This explains why the CXY-values for the QFM- and the QMM-data finally vary little from the CXY-value of the FC-data, although the CX- and the CY-values of the FC-data are higher than the corresponding values for the QFM- and the QMM-data. The second column of table 2 shows the correlation coefficients calculated between fluorescence intensity-dependent parameters (I and I-Ecc) and pure morphological parameters (A, P, D, m'22, and Ecc) of the QFM-data. Since r-values are close to one, fluorescence dependent parameters can be substituted by pure morphological parameters; the latter can be obtained by ordinary light microscopy and QMM. The third column of table 2 shows the interdependence among fluorescence intensity independent parameters extracted for GIEMSA-stained erythrocytes with QMM. Parameter groups (A-M) with r > 0.95 are formed in order to reduce the dimensions of the matrix for PCA-analysis to 13 x 13 (grouped parameters contain redundant information). PCA can be applied to the QMM-data since the extracted parameters yield a gauss-like distribution. Due to the different nuclear morphology, parameter groups formed by the diploid cell population differ in part from the groups formed by the triploid set. In consequence, the compositions of the first three principal axes which result from the PCA (table 2, bottom) are similar but not identical for the di- and the tri-ploid erythrocyte populations (only parameter groups with a participation > 10% were plotted). For both erythrocyte populations, the calculated Eigen Values (l) for the first principal axis (l1) are more than twice as high as those for the second principal axis (l2). The ratio of the standard deviation of the diploid nuclei in relation to the first two principal axis defines the relative importance of the major in respect to the minor principal axis for the classification of sample erythrocytes. This outlines the advantage of the Mahalanobis Distance over the simple Euclidean Distance: The weights compensate for the nonhomogeneous distributions of the cellular parameters in relation to each principal component. Accordingly, the Mahalanobis Distance represents a reliable measure for the classification of unknown objects into well-known populations.
In 1991 Kamentsky and Kamentsky reported that quantitative cytomeric data obtained with fluorescence microscopy is comparable to data obtained with FC. Many different applications have emerged since then which take advantage of quantitative fluorescence microscopy or quantitative morphological microscopy, QFM or QMM (e.g. Fanani et al 2002, Härtel et al 2003/2005, Alvarez et al 2004/2005). As we report here, the relative DNA content of erythrocytes of O. mykiss can be quantified with FC (AO-staining) or QFM (AO-, DAPI-, or PI-staining). The discriminating capacities CD indicate that the relative DNA content in erythrocytes can be determined by QFM with a precision close to the FC-data (the reference method in quantitative cytometry). Unfortunately, determination of morphological parameters with FC is not reliable; side- and the forward-scattering of the laser beam resemble surface properties or the dimension of the objects in a very rough manner. As we show, the correlation coefficients between the parameters of the FC-data are higher than the coefficients of the QFM- or QMM-data, and we observe that the CXY- and CXYZ-values for QFM and QMM catch up to or are even greater than the FC-data. In consequence, the strength of the image processed microscopic methods QFM and QMM lies in the precise determination of more than one parameter coupled to the fluorescence (QFM) or to the morphology (QMM) of objects.
To our surprise, the integral fluorescence intensities of the di- and triploid erythrocyte nuclei correlate almost perfectly with basic morphologic properties (r > 0.99 for I and A). The correlation between the relative DNA content and the nuclear size suggests a homogeneous packing of the chromatin inside the nuclear membrane. In consequence, the relative DNA content in erythrocytes can be estimated indirectly by the size of the nuclei. This conclusion reveals why cell size and morphology have been used as discriminant parameters for the identification of di- and triploid salmon populations (Phillips et al 1986, Wlasow et al 2004). In addition to the correlation between relative DNA content and the nuclear size, further intensity coupled parameters can be substituted by their morphologic counterparts (r > 0.99 for I-Ecc and Ecc). Since morphologic parameters are calculated directly from the binary mask of the objects, fluorescence staining of the nuclei becomes unnecessary and can be substituted by simple, inexpensive, and non hazardous contrast staining procedures like GIEMSA in combination with bright field microscopy. For the morphological discrimination between erythrocyte populations with QMM, size, eccentricity (a measure for circularity), and invariant moments were shown to be the most accuratet parameters (table 2, figure 2). This result is supported by the visual impression: diploid nuclei are small and circular in comparison to the curved, bean-like triploid nuclei; see Small and Benfey (1987) for diploid erythrocytes and figure 1 or Wlasow et al (2004) for triploid samples.
Without PCA, visual multi-parameter analysis becomes a difficult task when more than two parameters are considered. For n parameters, n(n-1)/2 two-dimensional histograms have to be monitored in order to find an optimal representation. With increasing n, the task becomes time- consuming if not impossible. As shown, QMM in combination with PCA performs excellent feature extraction for di- and triploid erythrocyte populations (table 2, bottom). PCA bundles redundant information and reduces the dimension of the parameter space to a minimum. Besides PCA, alternative methods like principal component regression, partial least square regression, and artificial neuronal networks have been tested by Davey et al (1999) for FC classification of different micro-organism with up to 7 parameters. The authors reported that artificial neuronal networks provided the best discriminating capacities in most experiments. For QMM, we preferred PCA in combination with the Mahalanobis Distance, because it presents a very clear classification criterion.
In conclusion, QMM can quantify triploidization rates in rainbow trout in a reliable, fast, and innovative way. The results are comparable or even superior to the existing reference method, FC. Since di- and tri-ploid rainbow trout do not show significant variations of erythrocyte density in the blood (data not shown), aliquots of blood samples can be mixed into a single sample. This way, the time and cost-intensive analysis of many individual blood samples can be reduced to a single step. The developed method opens a perspective for the classification of microscopic objects with many possible applications.
S.H. is supported by Fondecyt 3030065 (Chile). Institutional support of the Centro de Estudios Científicos (CECS) from Empresas CMPC is gratefully acknowledged. CECS is a Millennium Science Institute and is funded in part by grants from Fundación Andes and the Tinker Foundation. RR thanks the DAAD, the University of Bremen, and the Universidad Austral de Chile for a travel-funds. C.R. thanks the DLR for financial support. O.G. thanks DID-UACH 160404-01/EN for financial support.
Al-sabti K. 1995. Detection of triploidy in fish using the cytokinesis-blocked method for erythrocyte and hepatic cells. Cytobios 82, 181-187. [ Links ]
Alvarez M, R Godoy, W Heyser, S Härtel. 2004. Surface bound phosphatase activity in living hyphae of ectomycorrhizal fungi of Nothofagus obliqua. Mycologia 96, 479-487. [ Links ]
Alvarez M, R Godoy, W Heyser, S Härtel. 2005. Anatomical-physiological determination of surface-bound phosphatase activity in ectomycorrhiza of Nothofagus obliqua based on image processed confocal fluorescence microscopy. Soil Biol Biochem 37, 125-132. [ Links ]
Boron A. 1994. Use of erythocyte measurements to detect natural triploids of spined loach Cobitis taenia (L.). Cytobios 78, 197-202. [ Links ]
Castleman K R. 1996. Digital Image Processing. Prentice Hall, Englewood Cliffs, NJ, USA. [ Links ]
Cozier W W, I J Moffett. 1989. Application of an electrophoretically detectable genetic marker to ploidy testing in brown trout (Salmo trutta L.) triploidized by heat shock. Aquaculture 80, 231-239. [ Links ]
Child A R, H P Watkins. 1994. A simple method to identify triploid molluscan bivalves by the measurement of cell nucleus diameter. Aquaculture 125, 199-204. [ Links ]
Danielson P E. 1978. A New Shape Factor. Comp Image Process 7, 292-299. [ Links ]
Davey H M, A Jones, A Shaw, D Kell. 1999. Variable selection and multivariate methods for the identification of micro-organisms by flow cytometry. Cytometry 35, 162-168. [ Links ]
Ewing R R, C G Scalet. 1991. Flow-cytometric identification of larval triploid walleyes. Progress in fish culture 53, 177-180. [ Links ]
Fanani M L, S Härtel, R G Oliveira, B Maggio. 2002. Bidirectional control of sphingomyelinase activity and surface topography in lipid monolayers. Biophys J 83, 3416-3424. [ Links ]
Friars G W, I McMillan, V M Quinton, F M Oflynn, S A McGeachy, T J Benfey. 2001. Family differences in relative growth of diploid and triploid Atlantic salmon (Salmo salar L.). Aquaculture 192, 23-29. [ Links ]
Happe A, E Quillet, B Chevassus. 1988. Early life history of triploid rainbow trout (Salmo gairdneri Richardson). Aquaculture 71, 107-118. [ Links ]
Härtel S, M Zorn-kruppa, S Tikhonova, P Heino, M Engelke, H Diehl. 2003. Staurosporine-induced Apoptosis in Human Cornea Epithelial Cells in Vitro. Cytometry 08, 15-23. [ Links ]
Härtel S, M L Fanani, B Maggio. 2005. Shape transitions and lattice structuring of ceramide-enriched domains generated by sphingomyelinase in lipid monolayers. Biophys J 88, 287-304. [ Links ]
Gonzales R C, R E Woods. 2002. Digital Image Processing. Addison-Wesley, MS, USA. [ Links ]
Jessel R, S Härtel, C Socacio, S Tykhonova, H Diehl. 2002. Kinetics of apoptotic markers in exogeneously induced apoptosis of EL4 cells. J Cell Mol Med 6, 82-92. [ Links ]
Kamentsky L A, L Kamentsky. 1991. Microscope-based multi-parameter laser scanning cytometer yielding data comparable to flow cytometry data. Cytometry 12, 381-387. [ Links ]
Ojeda F, M I Guarda, C Maldonado, H Folch. 1992. A flow-cytometric method to study DNA fragmentation in lymphocytes. J Immunol Methods 152, 171-176. [ Links ]
Phillips R B, K D Zajicek, P E Ihssen, O Jonson. 1986. Application of silver staining to the identification of triploid fish cells. Aquaculture 54, 313_319. [ Links ]
Räth C, G Morfill. 1997. Texture detection and texture discrimination with anisotropic scaling indices. J Optical Soc Amer 14, 171-176. [ Links ]
Small S A, T J Benfey. 1987. Cell size in triploid salmon. The J Exp Zool 241, 339-342. [ Links ]
Taguchi G, R Jugulum. 2002. The Mahalanobis-Taguchi Strategy. Wiley, New York, USA. [ Links ]
Thomas P, R Morrison. 1995. A method to assess triploidy in swim-up rainbow trout. Aquaculture 9, 62-63. [ Links ]
Thititananukij S, R Vejaratpimol, T Pewnim, A W FAST. 1996. Ethidium bromide nuclear staining and fluorescence microscopy: An alternative method for triploidy detection in fish. J World Aquac Soc 27, 213-217. [ Links ]
Utter F M, O W Johnson, G Thorgaard, P S Rabinovitch. 1983. Measurement and potential applications of induced triploidy in pacific salmon. Aquaculture 35, 125-135. [ Links ]
Wlasow T, H Kuzminski, P Woznicki, E Ziomek. 2004. Blood Cell Alteration in Triploid Brook Trout Salvelinus fontinalis (Mitchill) Acta Vet Brno 73, 115-118. [ Links ]
Wattendorff R J. 1986. Rapid identification of triploid grass carp with a Coulter Counter and Channelyzer. Prog Fish Cult 48, 125-132. [ Links ]
Young I T, J E Walker, J E Bowie. 1974. An Analysis Technique for Biological Shape. Inf Control 25, 357-370. [ Links ]
Young I T, J J Gerbrands, L J Van Vliet. 1997. Fundamentals of Image Processing, Cip-Data-Koninklijke Bibliotheek Den Haag, ISBN 90-75691-01-7, 19-32. [ Links ]
* Corresponding author: Dr. Steffen Härtel, firstname.lastname@example.org, Centro de Estudios Científicos (CECS), Arturo Prat 514, Valdivia, Chile, Tel/Fax: (56) + 63 234589/17.