Abstract
The study of genome size variation is important from a number of practical and theoretical perspectives. For example, the long-standing “C-value enigma” relating to the more than 200,000-fold range in eukaryotic genome sizes is best studied from a broad comparative standpoint. Genome size data are also required in detailed analyses of genome structure and evolution. The choice of future genome sequencing projects will be dependent on knowledge regarding the sizes of genomes to be sequenced, and so on. To date, genome size data have been acquired primarily by Feulgen microdensitometry or flow cytometry. Each has several advantages but also important limitations. In this review, we provide a practical guide to the new technique of Feulgen image analysis densitometry. The review is designed for those interested in genome size measurements but not extensively experienced in histochemistry, densitometry, or microscopy. Therefore, relevant historical and technical background information is included. For easy reference, we provide recipes for required reagents, guidelines for cell staining, and a checklist of steps for successful image analysis. We hope that the accuracy, rapidity, and cost-effectiveness of Feulgen image analysis demonstrated here will stimulate further surveys of genome sizes in a variety of taxa.
T
In 1950, Hewson Swift developed the concept of the “C-value” in reference to the haploid “class” of DNA in plants, and 1 year later Alfred Mirsky and Hans Ris (1951) carried out the first broad survey of animal genome sizes, including representatives of all five vertebrate superclasses as well as several invertebrates. On the basis even of these preliminary results, it became clear that DNA contents varied greatly among species and that this variation bore no relationship to intuitive notions of organismal complexity. This observation had become no less confusing 20 years later and was dubbed the “C-value paradox” (Thomas 1971). The discovery of non-coding DNA a short time later dissolved the “paradox,” but several puzzles remain to this day as part of a larger “C-value enigma” (Gregory 2001a). For example, a great deal remains to be discovered regarding the mechanisms that account for the spread and loss of non-coding DNA, the reasons for its differential maintenance and/or deletion among species, and the cytological, physiological, and evolutionary implications of its presence (Gregory and Hebert 1999; Gregory 2001a, b).
Eukaryotic genome sizes vary more than 200,000-fold, with this entire range found among protists. In animals, the range is roughly 2500-fold and in vertebrates it is more than 350-fold (Gregory 2001c). Genome size is strongly positively correlated with cell size in each of the vertebrate classes (Olmo 1983; Gregory 2000, 2001b, in press). In homeotherms, there is a negative relationship between genome size and metabolic rate (Vinogradov 1995, 1997; Gregory 2002). In plants, amphibians, insects, and fish, genome size appears to be directly relevant to the rate and complexity of development (Gregory in press). In many invertebrates, genome size correlates positively with body size (e.g., Gregory et al. 2000 and references therein). Therefore, the prediction made half a century ago by Vendrely and Vendrely (1950), that “the systematic study of the absolute nuclear DNA content across numerous animal species will, without doubt, provide many interesting suggestions concerning the question of evolution” [our translation], now seems a charmingly profound understatement.
Knowledge of species' genome sizes not only is relevant to a host of important general biological questions but it may also be useful in the classification of organisms in the way that chromosome numbers have been (e.g., Manfredi Romanini 1972; Ohri 1998). Perhaps most importantly in today's climate of molecular biology, genome size is an important consideration in the development of future genome sequencing projects and for comparative studies of genome structure and evolution. To date, genome sizes have been reported for roughly 3000 animals and nearly 4000 plants (Bennett and Leitch 2001; Gregory 2001c), as well as for many fungi, protists, and bacteria (although detailed databases have yet to be compiled for the latter groups). Although these data have allowed many intriguing comparative analyses, some of which were outlined briefly above, there remain enormous gaps in the present genome size database. In plants, roughly 1.4% of angiosperms, 16% of gymnosperms, 0.4% of pteridophytes, and 0.1% of bryophytes have been analyzed (Bennett and Leitch 2001). The 2100 or so vertebrate genome sizes published similarly cover only small fractions of the total species in each class: birds 2%, fish 3%, reptiles 4%, mammals 7%, and amphibians 8% (Gregory 2001c). Invertebrates, which make up the majority of the multicellular life on the planet, have thus far been examined to the tune of a mere 800 species (Gregory 2001c). As the relevance of genome size information continues to increase, so also should the intensity of genome quantification efforts. However, in many respects such work has been hindered by difficulties of measurement imposed by the high cost of the equipment required for such analyses and a tedious methodology. Here we outline a new method for rapid and inexpensive genome size measurements, based on image analysis densitometry, which largely eliminates these barriers to genome size study. We recognize that not everyone interested in genome size quantification is a histochemist or a microscopist (ourselves included). Therefore, we have aimed to provide relevant historical and technical information, to present the various protocols in a clear fashion, and to outline the most likely sources of error encountered at each step. In short, this is the review we wish we had had when we began our own animal genome size measurements.
DNA Quantification: Past and Present
Several methods have been employed to quantify nuclear DNA. Some of the earliest studies involved bulk biochemical DNA extraction techniques to estimate the total DNA content of a preparation, which was then divided by an estimate of the number of nuclei present. Although imprecise, these methods were sufficient to demonstrate the constancy of DNA across tissues and among conspecific individuals, and to hint at the pronounced variation of genome sizes in comparisons of different species (e.g., Vendrely and Vendrely 1948, 1949, 1950). Analyses of reassociation kinetics similarly provided usable but somewhat questionable estimates of DNA content. A short time after the publication of these initial exploratory studies, methodologies began to shift towards densitometric techniques. This approach remains prominent and underpins the methodology described here. Therefore, it is important to discuss some of the fundamental physical and chemical aspects of densitometry.
Densitometry
Densitometric methods were first employed for relative nucleic acid quantifications by Torbjörn Caspersson in the 1930s, although actual genome size measurements were not made with these techniques until some time later. The landmark surveys of both Swift (1950) and Mirsky and Ris (1951) made use of densitometric techniques to quantify nuclear DNA, as have countless studies since (for a discussion of the utility of Feulgen densitometry, see Rasch 1985). This process involves staining fixed (air-dried) tissue preparations on microscope slides by the Feulgen reaction, most often using Schiff's leucofuchsin sulfurous acid reagent (although occasionally other chemical stains have been used, e.g. acriflavine, gallocyanin chromalum). Details of the Feulgen reaction are discussed below, but Feulgen densitometry relies on the simple premise that the amount of stain bound is directly proportional to the amount of DNA present. The quantity of stain is itself determined based on the amount of light it absorbs (i.e., its density). Detailed reviews of the physics of cytophotometry are available elsewhere (e.g., Fukuda et al. 1978), and therefore only a basic outline is provided here.
Two issues complicate the quantification of stain molecules bound to DNA. The first is that it is not possible to measure absorbance directly; absorbance is the lack of emitted light, and therefore represents non-information. Instead, absorbance (optical density, OD) must be calculated indirectly from measurements of the amount of light passing through the object (transmittance, T). Transmittance, in turn, is measured as the difference between the intensity of incident light entering the object and that of the transmitted light leaving it. In Feulgen DNA densitometry, measurements are taken both within the nucleus and outside the nucleus in a clear area of the slide. The difference in light intensity between the two areas represents the transmittance. Optical density and transmittance are related to one another as follows:
Because the relationship is not simple, this calculation is usually done automatically by the densitometry equipment.
In the simplest physical terms, the absorbance of monochromatic light by a uniform solution is proportional to both its concentration (Beer's law) and its thickness (Lambert's law). These rules hold for measurements of stained nuclear DNA, but whereas the absorbances of solutions can be ascertained by a single measurement (e.g., with a spectrophotometer), the heterogeneous nature of DNA stain in the nucleus means that any single point measurement will not be representative of the nucleus as a whole. Moreover, a single density measurement ignores variation in the sizes of individual nuclei. To solve these problems, it is necessary to take a series of point densities covering the entire nuclear area. The sum of these individual optical densities is the integrated optical density (IOD) of the nucleus:
It is the mean (or modal) IOD of all nuclei measured for an unknown species that is compared against the IOD of nuclei from a known “standard” (more on choice of standards below). Historically, these individual point densities have been obtained in two ways: by moving a narrow beam of light through the nucleus and taking density measurements at each point (e.g., with a “flying spot” densitometer) or by moving the nucleus itself through a narrow beam (e.g., with a “scanning stage” densitometer). Both approaches have been successfully employed in traditional Feulgen densitometry methods. However, in each case the cumulative measurement of individual point densities substantially slows the analysis and limits measurements to a single nucleus at a time. As discussed below, image analysis-based methods suffer neither of these constraints.
Fluorometry
As an alternative to measurements of stain absorbance, it is also possible to quantify the fluorescence of a DNA-specific stain. This is accomplished by using an appropriate light source to stimulate the emission of light of a specific wavelength by the stain molecules. In some cases, fluorescence studies have used Feulgen staining similar to densitometric studies (“static fluorescence cytophotometry”; e.g., Böhm and Sprenger 1968; Prenna et al. 1974), or have measured the fluorescence of a bulk nuclear preparation and divided by nuclear counts (e.g. Hinegardner 1968; Hinegardner and Rosen 1972). More recently, confocal laser microscopy has been used to accurately determine DNA contents by performing repeated “slices” through the three-dimensional structure of fluorescently-stained nuclei (e.g., Erlandsen and Rasch 1994). However, most fluorescence work has now moved from “static” approaches such as these to the more dynamic process of flow cytometry.
Developed in the late 1970s primarily as a means of detecting the anomalous DNA contents of cancer cells, flow cytometry has since become a staple of genome size research. Briefly, this method involves treating samples of nuclei in suspension with a DNA-specific fluorochrome (e.g., propidium iodide or DAPI) and measuring their fluorescence against that of a known standard included in the sample. This is accomplished by passing the stained nuclei through the path of a laser of a specific wavelength, which stimulates the emission of light by the fluorochrome. The technique is rapid and accurate but it has some important limitations related to the large number of nuclei required for analysis and the need to place them in suspension.
When leaf nuclei or blood cells are used, very large cell populations can be sampled from a single individual. However, when small organisms are to be studied, the large cell numbers required for flow cytometry can present a problem. For example, a recent study of genome sizes in the crustacean genus Daphnia used 20–160 individuals for each measurement (Korpelainen et al. 1997). In many cases specimens are either not available in these numbers, or the issue under investigation requires genome size estimates for single individuals, or even tissue-specific estimates. Flow cytometric measurements of blood tend to require freezing or other preservation methods (Gold et al. 1991), which are much less convenient in the field than the preparation of simple air-dried smears. Repeated measurements of samples are not possible because the procedure does not produce permanent preparations and because fluorescence fades quickly. This method is also typically limited to the inclusion of only one (or sometimes two) standard(s), and problems with staining can therefore be difficult to detect (e.g., Vindeløv et al. 1983). The fluorochromes used in flow cytometry are base pair-specific, such that differences in GC/AT content can affect the measurements if only one stain is used (e.g., Vinogradov 1994, 1998).
Although flow cytometry is currently the most efficient and accurate method available for genome quantification, the cost of equipment is a major barrier to its broad use. The constraints of standard densitometric approaches make them even less appealing. However, a method that combines the advantages of Feulgen densitometry (e.g., permanent and easily prepared specimens, tissue-specific measurements, multiple standards, low cost) without the immense time consumption of traditional densitometric techniques would provide an attractive alternative to flow cytometry. Fortunately, advances in computing and imaging technology have facilitated the development of such a method.
Image Analysis Densitometry
As with flow cytometry, the use of image analysis technology in DNA quantification began in cancer diagnosis (e.g., Borgiani et al. 1994; Fischler et al. 1994). Given the extreme importance of accuracy in such an application, it is not surprising that the fidelity of the technique has been scrutinized with exceptional rigour (e.g., Böcking et al. 1995; Thunnissen et al. 1996, 1997; Reeder et al. 1997; Puech and Giroud 1999). Comparisons have shown that flow cytometry and image analysis provide similar efficacy of DNA quantification for diagnostic purposes (e.g., Bertino et al. 1994; Borgiani et al. 1994; Marcos et al. 1998). In fact, image analysis has occasionally been considered superior to flow cytometry in this regard (e.g., Pindur et al. 1994; Yamamoto et al. 1994). More recently, Feulgen image analysis has been used for quantitative DNA studies in plants (e.g., Venora et al. 1995a, b; Cremonini et al. 1998; Voglmayr and Greilhuber 1998; Voglmayr 2000), and its utility in the determination of plant genome sizes has been examined in some detail (Vilhar et al. 2001). However, this technique has thus far been used in only a very small number of genome size studies in animals (e.g., Dorward and Wyngaard 1997; González-Tizón et al. 2000) aside from humans, despite the obvious applicability of this technique to easily prepared cells such as vertebrate erythrocytes. In this review our focus relates primarily to cells of this type, although several other animal cell types are also considered. As will be seen, Feulgen image analysis provides a rapid, cost-effective, and user-friendly method for the measurement of animal genome sizes.
Basic Concepts
The heterogeneity of DNA staining within nuclei is a problem faced by any densitometric technique, and Feulgen image analysis is no exception. However, the means by which this difficulty is overcome differs crucially in image analysis vs flying spot or scanning densitometry. In image analysis densitometry, the microscope field is captured by a microscope-mounted CCD (charge-coupled device) or digital camera connected to a computer via a “frame-grabber” board. As with all digital images, these photos are displayed as a series of pixels, each of which is of a specific color and intensity. The different intensities of the various nuclear pixels represent ready-made point intensities that can be converted to absorbance values by the image analysis software.
A color image of stained nuclei can be made into a single linear scale of pixel intensities by converting the image to grayscale, by using monochromatic incident light (as with an interference filter in front of the light source), or by analyzing only one of the three constituent “channels” (red, green, or blue) that make up the color pixel. In each case, pixel intensities vary along a scale from 0 (black) to 255 (white). A measurement of a section of the slide lacking nuclei provides the measure of incident light, just as it does in traditional Feulgen densitometry methods. However, in this case integrated optical densities are calculated from pixel values along the 256-value scale:
where n = total number of pixels in the nucleus, IFi = intensity of “foreground” (nuclear) pixel, and IBi = intensity of “background” (clear area) pixel. Thus, the image analysis approach uses individual pixel values to instantaneously calculate IOD from the image as a whole. This approach not only avoids the necessity of acquiring individual point densities one at a time, as in standard methods, but it also allows the simultaneous tabulation of IODs for all of the nuclei within a microscope field. By way of comparison, it may take more than an hour to measure 50 nuclei with a scanning stage or flying spot densitometer, but an image analysis system can measure 500 nuclei in less than five minutes.
Outline of Methodology
Hardware and Software. The technique outlined below is based on measurements of Feulgen-stained animal nuclei (primarily nucleated vertebrate erythrocytes) conducted with the Bioquant True Color Windows 98 v3.50.6 image analysis software package (R&M Biometrics; Nashville, TN). Hardware consisted of an Optronics DEI-750 CE three-chip CCD camera connected via a BQ6000 frame-grabber board to a Pentium II 300 MHz PC running Windows 98. To reinforce the cost-effectiveness of the image analysis approach, we employed an inexpensive Leica DM LS compound microscope in this study. Higher-quality optics do improve the measurements slightly but are not generally necessary for accurate genome size estimates.
The particular imaging equipment used in this study represents just one of a large number of hardware and software packages available. Choice of equipment will be dependent on price, preference for user interface, and other features of the package in addition to densitometric tools (e.g., fluorescence or morphometric capability). In any case, the key components of the system are a camera with a linear response (e.g., a doubling of density registers as such) and a software package capable of accurate IOD measurements. Linearity of camera response can be tested with the use of stepped density filters (“density wedges”), which are available from optics suppliers (e.g., Edmund Industrial Optics; Burlington, NJ). Any system should be tested for accuracy using a broad series of known genome size standards before purchase. A second system (Hitachi HV-C20 three-chip CCD camera, Pro Series Capture 128 frame-grabber, and Image-Pro Plus 3.0 image analysis software) was tested, and did not perform satisfactorily. Caveat emptor!
The Feulgen Reaction. In its initial formulation, the histochemical reaction developed by Robert Feulgen was used simply for the detection of DNA in the nucleus (Feulgen and Rossenbeck 1924), but since the demonstration that it is both specific and stoichiometric for DNA it has become the most important means of staining nuclear DNA for densitometric quantification. The protocol has been modified frequently and substantially since its early development, but the basic components have not been altered.
It is important to note that “Feulgen” is an ordered series of chemical reactions, not a stain. The most commonly employed stain in the Feulgen reaction is Schiff's reagent, developed by Hugo Schiff in the 1860s (Schiff 1866). Schiff carried out extensive research involving reactions of aldehydes with aromatic amines, one of which was the triphenylmethane basic dye rosanaline (also known as fuchsin). The Feulgen reaction uses strong acid to generate free aldehyde groups in the DNA molecule (specifically, by splitting off the purine bases A and G to produce “apurinic acid”; note that it is therefore not base pair-specific as are many fluorochromes), to which a fuchsin molecule decolorized with SO2 can bind and regain its pink color. The exact chemistry of this reaction is understood only in general terms, but its efficacy is well-established (for reviews see, e.g., Kasten 1960; Kjellstrand 1980; Schulte 1991).
Schiff reagent has been prepared in a variety of ways, including bubbling SO2 gas through a solution of dissolved fuchsin, but many of these proved difficult to standardize (for review see Kjellstrand 1980). Most modern preparations follow a modified version of the protocol developed by Lillie (1951), which uses sodium (or potassium) metabisulfite as a decolorizing agent. Although Schiff reagent is available commercially, it is not as reliable as freshly prepared solutions and is therefore not recommended. The recipe for Schiff reagent used in the present study, and the one recommended for image analysis densitometry of animal nuclei, is outlined in Appendix 1.
Almost every step in the Feulgen reaction procedure has been varied among studies, and each can affect the efficacy of staining. Notable examples include sensitivity to the choice of fixative and fixation time, concentration and temperature of acid and hydrolysis time, preparation of Schiff reagent and staining time, and so on. In previous studies, hydrolysis conditions have varied greatly, from “hot hydrolysis” in relatively weak acid (e.g., 1 N HCl at 60C) to “cold hydrolysis” in strong acid (e.g., 5 N HCl at 20–30C). The recommended duration of hydrolysis has also varied greatly, and must be altered according to cell type, fixative used, and so on (e.g., Deitch et al. 1968; Schulte 1991; see below). Today, most hydrolysis regimens involve relatively short (30–60-min) exposures to strong acid (5 N HCl) at room temperature (RT). We have further optimized the staining protocol with specific reference to Feulgen image analysis of animal cells, and have investigated many sources of error related to variables in the staining procedure (these are discussed in detail below). Our recommended protocol for Feulgen staining of animal nuclei is outlined in Appendix 2, and involves much longer fixation and hydrolysis times than are usual. The various steps and their roles are described in Table 1. Detailed reviews of the sensitivities of the Feulgen reaction are available elsewhere (e.g., Schulte 1991), and we therefore discuss only factors directly relevant to genome size measurements in animals.
Once a Feulgen protocol is decided upon (see Appendix 2 and below for reasons to vary it slightly), material on prepared microscope slides can be stained in groups. Traditionally, slides have been stained in small numbers, usually not more than 20 at a time. With slow measurement techniques these small batches were not a problem. However, with the measurement lag eliminated by the rapidity of image analysis techniques, it is desirable to stain slides in large collections to allow multiple standards to be included in each staining run and to reduce the error associated with comparisons across runs. With the use of 1L glass staining boats and large-capacity metal racks (both available from Fisher), up to 100 slides can be stained simultaneously (and in fact, we routinely perform three such runs together for a total of 300 slides per batch). If stored in the dark (e.g., in a slide box), stained slides remain measurable for long periods of time, although different mounting media and exposure to intense light shorten their lifespan (Dewse and Potter 1975).
Outline of steps in the Feulgen reaction and their respective roles; see Appendix 2 for details of the recommended protocol
Microscope Set-up. The first step in the measurement of DNA contents by image analysis involves setting up the microscope (Appendix 3). Lenses must be clean (because even small marks will appear in the image). Proper Köhler illumination should be established according to the microscope manufacturer's instructions (briefly, this involves ensuring that the light path is direct from source to ocular and that contrast is optimized). The microscope should be placed on a microscopy bench to minimize vibrations. The microscope light source, camera, and computer should all be plugged into a voltage regulator to eliminate fluctuations in the current, which degrade the image quality. Standard surge suppressors and “line conditioners” are not sufficient for this purpose; a proper voltage regulator is required, such as the Sola Minicomputer Regulator (MCR) (Elk Grove Village, IL) employed in this study.
It is customary in Feulgen densitometry to use monochromatic light of a wavelength near the absorption maximum for Schiff reagent (~560 nm). Flying spot densitometers (e.g., Vickers M85) can have their light source set to a specific wavelength, whereas scanning densitometers employing modified compound microscopes have traditionally been equipped with interference filters to produce the desired wavelength of incident light. This is not strictly necessary, because measurements at off-peak wavelengths give the same ratios as those performed around 560 nm (Fand and Spencer 1964). In image analysis, only one of the three light channels (green) is used, and measurements can be made directly from color images (Kamma et al. 1992). As a result, measurements can be made with an unfiltered light source. To test this directly, we compared mean IODs of chicken and rainbow trout nuclei and their ratios measured at ×100 magnification with and without the use of a 546-nm interference filter (20-nm bandwidth), and found them to differ by less than 1%. Monochromatic light may reduce problems associated with chromatic aberration in lower-quality lenses and may increase contrast between the background and lightly stained objects, but the density measurements are not otherwise affected by the use of unfiltered light. Neutral density filters similarly help to homogenize the incident light but are not required.
The most significant optically based source of error is glare (i.e., the loss of light caused by refraction), and cell membranes represent the primary source of unwanted glare in an image. In some cases, as with very old slides or when nonspecific background staining has occurred, it will be impossible to eliminate the cell membranes from the captured image. Under most conditions, however, a properly matched refractive oil placed between the specimen and a coverslip can remove this problem. Once the appropriate refractive liquid is determined, it should be consistently applicable to cells of the same type. For example, most vertebrate erythrocytes can be examined with the use of oil of nD = 1.540. Kits containing oils of several different refractive indices are available (e.g., from Cargille Laboratories; Cedar Grove, NJ). The following steps can be taken to select an oil with the appropriate refractive index for the cell type being analyzed:
Begin by placing a drop of an oil with a mid-range refractive index (e.g., nD = 1.540) on the slide, and add a coverslip.
Focus on a nucleus, which should appear as a pink jewel suspended in space. If membranes are visible, the refractive index liquid used is incorrect. (Note that ×100 objectives require a second drop of immersion oil, usually nD = 1.515, between the coverslip and the lens.)
To determine whether a higher or lower refractive index is needed, focus up and down through the nucleus. In one direction, a bright outline of pink light will be seen on the perimeter of the nucleus. In the other direction, the nucleus will appear to enlarge as the focus moves through it, without any bright light. Take note of which direction of focus (i.e., “up,” moving the lens away from the slide, or “down,” towards the slide) produces which effect.
If the bright light appears when focusing up, then a higher refractive index is required. If the bright light appears during the down focus, then a lower-nD liquid is needed.
Before capturing an image (and indeed, before setting up Köhler illumination), it is necessary to choose the magnification level that will be used. Most animal nuclei are too small to measure with objectives lower than ×40. The use of a ×40 objective (actually ×400, since there is normally an additional ×10 lens in the body of the microscope) eliminates the need for a second immersion oil and also provides more individual nuclear measurements per field, and is therefore considerably faster than higher magnification lenses. However, lower magnifications mean smaller individual nuclear images, i.e., fewer pixels (points densities) per nucleus. This generally produces higher coefficients of variation and less accurate IOD ratios, and can also result in greater eye strain because fewer pixels in each nucleus makes it more difficult to define the appropriate thresholds (see below). In general, ×100 oil immersion objectives should be used unless prohibitively large nuclei are being measured.
Image Capture and Analysis. Image analysis software packages and CCD cameras differ somewhat in their use, but their principles are shared. Details of how to carry out these general instructions should be available in the user's manual. In any case, the first step involves establishing the appropriate conditions for capturing an image. First, a field containing nuclei must be located and brought into focus (on the computer screen, not the ocular). Once focus is optimized, an area of the slide should be selected that is free of nuclei. This blank area can be used to adjust the brightness and color balance of the microscope and camera. The exposure level of the camera can be used to increase or decrease the brightness of the image, but it is best to make this adjustment with the microscope light source. In the present study, we ordinarily used an exposure time of 1/125 sec (longer exposures provide brighter but “shaky” images, and very short exposures are susceptible to the effects of high-frequency oscillations in microscope light output). Even at a suitable exposure, repeated measurements of the same nucleus can show minor variations in IOD, although these are usually small (<1%).
After a choice of exposure, the camera should be “white balanced” on a blank area of the slide. If an interference filter is used, this should be done before the filter is added (the brightness will have to be increased after the filter is put into place). Once completed, a field containing nuclei for measurement can be relocated. Final adjustments to brightness should then be performed; a maximal pixel intensity on the field of 190 (in the green channel) is a good general guideline for how bright the image should be (a histogram of pixel intensities in the entire image can usually be generated by image analysis programs).
Some image analysis packages allow measurements to be performed from “live” images, whereas others require images to be saved to the hard drive before analysis. In either case, the green channel should be used for the IOD measurements because it includes the absorption peak for the Feulgen-DNA dye complex, and therefore gives the highest IODs and the most accurate estimations of genome size. Some packages allow the green channel to be measured directly from the image, but others require it to be extracted from the color image to give a derived grayscale image. Once a measurable image is obtained, it is necessary to select the threshold of pixel values that are to be included in the measurement. In most cases it will be possible to zoom in on a nucleus to carefully select the pixels within it. It is not strictly necessary to highlight all pixels within a nucleus (this is especially true if there are very light areas within it, which could cause large non-nuclear areas of the image to be selected). The threshold is used only for outlining the objects to be measured; so long as the nuclei are properly outlined, all pixels within them will be included in the measurement. As a rule, it is less problematic to over-select pixels (i.e., to include pixels outside the perimeter of the nucleus) than to under-select, because these extra pixels will not contribute to the resulting IOD measurement when the background pixel value is subtracted from them.
Once thresholding is complete, all objects falling within the density of the thresholds should be highlighted. Any non-nuclear objects that are highlighted should be omitted, as should misshapen, broken, overlapping, or otherwise anomalous nuclei. Most software packages include data filters to automatically eliminate objects touching the edge of the field or falling outside a specified size range. Consistency in the criteria employed for exclusion is important for minimizing variability in results obtained by different investigators (e.g., Thunnissen et al. 1996, 1997; Reeder et al. 1997; Vilhar et al. 2001).
To measure the IOD of green-channel pixels, it is necessary to provide a measure of background (incident) light for comparison. This is not the same as a “background correction,” which must not be used in IOD measurements. Most software packages allow a non-nuclear area of the field to be selected manually, and the background intensity (IBi in equation 3) is taken as the average of all pixels in the selected area. This single value is then compared against all foreground (nuclear) pixels. Therefore, it is important that an area free of nuclei and located as centrally as possible be used in determining the background. Estimating specific backgrounds (e.g., individually next to each nucleus) does not greatly improve results and greatly slows data acquisition. So long as the illumination is homogeneous and the field does not otherwise contain light and dark areas, a single background measurement per field is sufficient. Using even an inexpensive microscope, we find that the repeated measurement of the same nucleus captured in different areas of the field produces an error of only ~1% (a maximal CV of 3% is recommended for this test; Böcking et al. 1995), and that building a new background next to it each time does not improve this significantly.
An important consideration in the choice of field to measure is the number of nuclei contained within it. Obviously, a very dense field with overlapping nuclei is not suitable, and very scanty fields with only a few nuclei each will substantially slow measurement times (although in smears from many amphibians, sparse nuclei are a fact of life). This problem can be largely alleviated, with vertebrate blood samples at least, by using a “flame tip” method of smear preparation (Figure 1). However, as shown in Figure 2, there is a strong negative relationship between the mean IOD of nuclei and their number in a given field. This variance can be explained by a combination of factors, including slower drying and/or crowding in dense areas of the slide that prevents nuclei from flattening out and later reduces their uptake of stain, darker background measurements that result in lower calculated IODs, and increased glare from membranes. In our analysis of varying the number of nuclei per field by more than an order of magnitude, the error in mean IOD exceeded 10%. Field density should therefore be kept as constant as possible, including across slides. When nuclei of different sizes are used, a measure of percent field covered by nuclear pixels can be employed in place of a simple count of nuclei present. Thus, a typical field density used in our analyses at ×100 would be 20–30 nuclei for chicken and 10–20 nuclei for trout. With larger nuclei and smaller nuclear counts per field, the variance among fields becomes higher relative to within-field variance (ANOVA, 20 fields each for chicken and trout), although in each case the maximal error in mean IOD among fields is only about 4%.
Some variation among individual nuclei is also to be expected, based on slight differences in staining, orientation, and local background conditions. In cancer detection applications, a coefficient of variation (CV = 100% × standard deviation/mean) of 6% or less is considered acceptable (e.g., Böcking et al. 1995), and this convention has been adopted by botanists using image analysis for plant genome size estimates (Vilhar et al. 2001). This level of variation is also easily achievable for vertebrate blood cells. Using chicken, trout, and other species of similar DNA contents, we routinely observe CVs of about 3%. (Because of the accuracy of the instrument, the analysis of only 50 erythrocyte nuclei provides nearly identical CVs and mean IODs to measurements of more than 1000 nuclei on the same slide.) However, this standard may need to be relaxed somewhat for more difficult preparations, such as squashes of small invertebrates. Nevertheless, CVs of 10% or less can usually be obtained even with the difficult cell types.
Recommended technique for producing undamaged monolayers of vertebrate blood cells. (
Calculations of Genome Size
As with all densitometric and fluorometric methods, image analysis-based techniques involve the conversion of unitless IOD values to absolute genome sizes by the comparison of ratios with standards of previously estimated DNA contents. In some recent image analysis studies, the peak IOD values of standards and unknowns have been used for these ratios, as is typically done with flow cytometry (e.g., Vilhar et al. 2001). Traditional densitometric methods have generally used comparisons of mean IODs rather than peak values (modes). Insofar as the distribution of nuclear IODs is normal (as it usually is), there is no difference between these methods. In the present study we have preferred to retain the practice of comparing mean IODs.
The choice of standards used in the calculation of absolute genome sizes is crucial. The rapidity of image analysis and the feasibility of staining large numbers of slides make it possible to use several (five to ten) standard species. It has sometimes been suggested that standard cells (e.g., chicken blood) should be placed on the same slide as the unknown specimen, a protocol that would limit the number of standards that could be used. However, this is unnecessary so long as the series of standard slides is included in each staining run. It is desirable to choose standards of the same cell type as the unknowns, although this convention has often been ignored. As will be seen below, differences in the characteristics of standard vs unknown cells can represent a substantial source of error.
It is best to include a range of standards broader in both directions than the expected range of unknowns. Wherever possible, standards should be of commonly used species (e.g., chicken) or ones that have previously been measured using a reliable method such as flow cytometry.
Before calculation of mean IOD, it is good practice to use a spreadsheet to “sort” the data in ascending/descending order. This will enable extremely anomalous values (such as those caused by small bits of debris overlooked during manual exclusion) to be identified and excised. In addition, it is advisable to clip the top and bottom 5% of the measured IOD values (e.g., five values on either end of a 100-value sample), because the extremes of the range may reflect improperly oriented nuclei or other minor problems with staining and/or analysis.
Genome sizes can be calculated in two ways. In the simplest approach, a standard curve (IOD vs known C-value) is generated and used primarily as a “check” that the stain was accurate across the range of standards included. A single primary standard (preferably chicken at 1 C = 1.25 pg) can then be used with confidence to calculate genome sizes:
The effects of field density on the mean IOD of rainbow trout nuclei. The more nuclei present in the field, the lower the mean nuclear IODs (r = −0.73; p<0.0005). Field density should be controlled as much as possible. See text for details.
Standard curve generated by image analysis densitometry of Feulgen-stained animal nuclei. (
where CVu = C-value of unknown, CVs = C-value of standard, IODu = mean IOD of unknown, and IODs = mean IOD of standard.
Alternatively, the regression equation of the standard curve can itself be used to calculate genome sizes:
where CVu = C-value of unknown, IODu = mean IOD of unknown, with y-intercept (y0 ) and slope referring to those of the regression of mean IODs of standards vs C-values of standards. It is worth noting that although this may help to distribute the error more evenly among standards, most of the C-values used in the regression will have been measured against a chicken standard, so that they may in fact introduce an additional level of error. Moreover, this approach is statistically problematic if the unknown genome size falls outside the range of the standards included. In either case, a standard curve is important because it demonstrates that the staining procedure was successful and that the analysis equipment is performing in a properly linear fashion. Figure 3A presents a very broad standard curve covering a roughly 400-fold range in DNA contents. It is evident from this figure that the imaging system and staining protocol are accurate across this range in DNA contents, as well as within the more limited range encountered in most animal genome size estimates (Figure 3B).
Additional Sources of Error
By following the procedures outlined above and in Appendix 2, accurate and rapid measurements can be made on large numbers of nuclei. However, no amount of care can provide accurate genome size estimates if there are fundamental discrepancies in the amount of stain present in the nuclei. The inclusion of several standards can help to identify such problems, but they must ultimately be addressed before the cells are stained. Surprisingly, each of the sources of error discussed below has been largely overlooked by previous densitometric studies of animal genome sizes. It is our hope that they will be properly addressed in future surveys.
Staining Protocol. Not all stains are created equal. For one, even certified dye lots of “basic fuchsin” are invariably mixtures of pararosaniline along with traces of rosaniline, magenta II, and other impurities (Schulte 1991). The content of pararosaniline in different dyes can vary from 90 to 99%. We evaluated dyes from both Fisher and Sigma sold as “basic fuchsin (pararosaniline hydrochloride),” “pararosaniline,” “pararosaniline acetate,” and “basic fuchsin: special for flagella,” and found marked differences in their efficacy. It is also worth noting that some dye lots purchased many years ago (courtesy of Ellen Rasch) tested very well, whereas recent versions of the same product were poor. Moreover, different lots of the same product purchased at the same time can give discordant results (which is why we have not listed a favorite type of dye). It is not enough to purchase the same dye product listed in the methods of a previous publication, nor is it sufficient to simply repurchase the same product. All stains should be tested by comparing the ratios of known standards before any valuable unknowns are stained with it. Once a suitable dye lot is found, it is not necessary to retest it before each use.
It is common practice to re-use Schiff reagent, but its efficacy diminishes with each use (i.e., intensity of staining fades but ratios of standards may not change). If Schiff reagent is to be re-used, it should be stored in a refrigerator, because cooling increases the solubility of SO2 and prevents its dissociation from the stain molecules. The storage container should also be filled completely and capped tightly to prevent dissolution of SO2. Eventually a white precipitate forms, and the reagent should then be discarded. Under ideal circumstances, Schiff reagent is best prepared fresh before each staining run and used only once. Recall, however, that stain preparation should begin a few days before an intended staining run to allow the recommended period of decolorization (see Appendix 1). Other minor sources of variation in stain quality are addressed in Appendix 1.
Age of Slides. Several fixatives are routinely employed in the preparation of blood smears. The most common in the genome size literature include methanol-glacial acetic acid (3:1 MeOH:GAA), methanol-formalin-glacial acetic acid (85:10:5 MFA), and formalin (usually 10%). Cells are typically fixed immediately after smear preparation, or (as in this study) are stored after air-drying and then postfixed before hydrolysis. In our experiences, these two protocols have little effect on the resulting IOD measurements of nucleated blood cells, although the latter is much more convenient when smears are prepared in the field. “Wet-fixing” before air-drying can significantly affect densitometric measurements in tissues such as liver (e.g., Schulte and Wittekind 1990; Schulte 1991), but this is not usually feasible with liquid-suspended cells such as erythrocytes. A surprisingly strong fixative, however, is time. Whether slides are fixed immediately or postfixed later, the age of slides has a significant effect on the intensity of staining. In fact, new (24 hr old) blood smears have mean IODs as much as 35% lower than those that are 2 years old (Figure 4). Comparing newly collected samples against old standards, or vice versa, can therefore lead to major errors in the resulting genome size estimates. Conditions of slide preparation are not usually reported, so the extent to which this error has infiltrated the animal genome size literature is not known.
The effect of slide age on the mean IOD of rainbow trout nuclei. New slides take up significantly less stain than older slides. Error bars represent SD. See text for details.
Effects of fixation time on the acid lability of nuclei containing different amounts of DNA. Species included are as follows: Betta splendens (circles), Gallus domesticus (triangles), Oncorhynchus mykiss (squares), and Notophthalmus viridescens (diamonds). Duration of fixation in MFA was for either 30 min (filled symbols) or 24 hr (open symbols). Hydrolysis was conducted in 5 N HCl at RT as part of a standard Feulgen staining protocol. Although long fixation necessitates increased hydrolysis times for cells with more DNA, it prevents DNA loss in smaller nuclei. Note logarithmic scale, which was used to facilitate comparisons of cells with greatly different optical densities.
This error is best and most simply corrected by using slides of approximately the same age. When this is not possible, error can be reduced (to an extent) by using lengthier fixation regimens. We have tested fixation times ranging from 30 min to 1 week and have determined that a 24-hr fixation is the most suitable for use with erythrocytes. However, long fixations can pose their own problems if an inappropriate fixative is used (different fixatives also tend to alter the kinetics of Feulgen hydrolysis; e.g., Schulte and Wittekind 1990). Very long fixation in 3:1 MeOH:GAA causes significant expansion and distortion of the nuclei. Long fixation in formalin results in very darkly staining nuclei and substantial nonspecific background staining (formalin is, after all, an aldehyde). MFA is well suited for prolonged fixation protocols, but a longer fixation necessitates a longer hydrolysis time for the complete staining of larger nuclei. Figure 5 compares the results of a hydrolysis time series for a collection of standards fixed in MFA for 30 min and overnight. Fortunately, after overnight fixation even small nuclei, such as those of Betta splendens, can tolerate long hydrolysis times without losing DNA. Therefore, although slides of very different ages should not be analyzed together, slides of slightly different ages can readily be standardized by overnight fixation followed by a long (e.g., 2-hr) hydrolysis.
Cell Types. A hydrolysis curve (i.e., IOD vs hydrolysis time) should be prepared when nuclei of different sizes are compared. Even if cells of the same type and age are being used, there can still be significant differences among cells in the optimal time between full DNA hydrolysis and subsequent depolymerization and loss of DNA. In some cases, no fixation or hydrolysis regimen can correct for differences in the affinity of nuclei for stain. This is particularly true when comparisons are conducted across cell types. Again, this is a source of error that has not generally been addressed in previous genome size studies, but it is one that can affect the validity of the resulting estimates.
The main difficulty in comparisons across cell types (and indeed, with different cell ages) lies in differences in the level of DNA compaction. This problem has long been recognized in Feulgen densitometry and can even be seen among different types of white blood cells taken from the same individual (e.g., Hale 1963; Bedi and Goldstein 1974). An extreme example of this effect is given in Figure 6, which shows the large difference in density of sperm vs buccal epithelium nuclei. To an extent, deviations from the expected ratios may result from the higher degree of glare produced by densely stained nuclei (e.g., Bedi and Goldstein 1976), or by problems with the accurate measurement of very high point densities (Allison et al. 1981). Several physical, optical, and chemical methods have been suggested to correct for these effects [e.g., glare correction (Chieco et al. 1994; Kindermann and Hilgers 1994); “cell crushing” (Davies et al. 1954); off-peak measurements or shortening of the staining times (Allison et al. 1981)]. Nevertheless, differences in the actual uptake of stain exist among cells of different DNA compaction levels which cannot be corrected in any simple way. (Feulgen fluorescence studies have encountered similar problems even within cell types, indicating that discrepancies in the amount of stain, and not the densitometric quanitification of it, are partly to blame; e.g., Fujita 1973.)
Discrepancies in IOD measurements caused by different levels of DNA compaction in human sperm (1 C = 3.5 pg, IOD = 500) and buccal epithelium (2 C = 7.0 pg, IOD = 1370). The expected ratio between them should be 1:2 but it is measured as 1:2.7.
In vertebrates, the most reliable way to correct for these errors is to simply be consistent in the types of cells selected for measurement. For example, a comparison of mouse liver cells against chicken erythrocytes gave an inflated mouse genome size estimate of 3.45 pg, whereas a comparison of mouse liver vs chicken liver gave a C-value of 3.23 pg, or approximately 0.5% lower than the most recent flow cytometric estimate (Vinogradov 1998). Liver and other mitotically active cells may be particularly unsuitable for comparison with non-dividing cells such as erythrocytes because the mean IOD of the former will inevitably contain measurements of cells with a range of DNA contents from 2 C to 4 C, depending on the extent to which DNA replication has occurred in each cell. It is notable in this regard that chicken liver gives significantly higher mean IODs than do chicken erythrocytes. (Liver is also potentially problematic because it contains polyploid cells. Endopolyploidy is an even more relevant concern in studies of invertebrate tissues, in which locating diploid cells can sometimes be a challenge.) Comparison of human lymphocytes with nucleated non-mammalian erythrocytes can be similarly problematic because of the high level of compaction of DNA in the former (e.g., Schulte 1991). Interestingly (and fortunately!), invertebrate hemocytes and sperm do appear suitable for comparison with chicken erythrocytes. The classic example of this is the remarkably accurate estimate of the genome size of Drosophila melanogaster by Ellen Rasch, using sperm and hemocyte nuclei against a chicken erythrocyte standard, made 30 years before genome sequencing confirmed its size with certainty (Rasch et al. 1971; Mulligan and Rasch 1980; Adams et al. 2000).
The suitability of a cell type for use as a standard can be demonstrated only through independent confirmation by other techniques. For example, if insect sperm or hemocytes measured densitometrically against chicken erythrocytes give the same values as insect neural tissue measured by flow cytometry (or better yet, complete genome sequencing), then the chicken standard can be considered reliable for this type of analysis. In cases such as these, it may be advisable to use only the proven standard in the calculation of genome size, although a standard curve should still be included as a check of staining accuracy, as outlined above. What should no longer be considered acceptable is the use of a single standard of one cell type compared against an unknown of another cell type, a protocol employed in many previous studies. With the use of rapid and accurate methodologies such as that described here, there is no reason to use small numbers of untested and potentially unreliable standards.
Concluding Remarks
Although the accurate measurement of genome sizes is not simple, it is of enough practical and theoretical importance to justify the effort. Recent advances in computing and image processing technology, combined with proven methods of histochemical staining, allow the reliable and rapid estimation of nuclear DNA contents. Image analysis densitometry has been accepted as an accurate means of quantifying DNA for clinical applications (Bertino et al. 1994; Borgiani et al. 1994; Fischler et al. 1994; Pindur et al 1994; Yamamoto et al. 1994; Böcking et al. 1995; Thunnissen et al. 1996, 1997; Reeder et al. 1997; Marcos et al. 1998; Puech and Giroud 1999), and its broader utility in the measurement of both plant and animal genome sizes has been demonstrated (Vilhar et al. 2001; and the present study). In this review we have provided the necessary background information, as well as a straightforward guide to the use of image analysis technology for genome size determination using this method. We have addressed the most significant pitfalls encountered with this technique, specifically in regards to genome size measurements across species, and have provided suggestions for how they can be minimized. The problems discussed are those faced by all densitometric methodologies, but the speed and accuracy of image analysis allows many of them to be addressed directly in ways not previously possible. It is our hope that the existence of a fast, accurate, and cost-effective alternative to time-consuming densitometric and expensive fluorometric techniques will stimulate the collection of genome size data so that the profound gaps in the current genome database can be closed. Perhaps most importantly, the accuracy and quantity of genome size determinations generated with this method may permit broad comparative analyses of genome size diversification among taxa and thereby illuminate the mechanisms and time scales involved. Only when this is accomplished will a truly comprehensive analysis of eukaryotic genome evolution be possible.
Footnotes
Appendix 1
Recipes for the preparation of Schiff reagent and bisulfite water to Feulgen stain 100 slides per run in 1 L staining boats. Note that three runs can easily be done simultaneously, for a total of 300 slides per batch. Stains made with different dye lots should be tested before use by staining standards of known DNA content ratios. Bisulfite water must be prepared fresh for each staining run.
To prepare 1100 ml of Schiff reagent:
1. Dissolve dye in acid: in a glass bottle, completely dissolve 11 g of basic fuchsin in 100 ml of 1.0 N HCl.
2. Decolorize with bisulfite: completely dissolve 10 g sodium metabisulfite (Na2S2O5 in 1000 ml) dH2O and add to the solution prepared in step 1.
4. Allow fuchsin to decolorize: cap tightly and store at RT in the dark for 72 hr, shaking occasionally to dissolve any pink precipitate. The solution will turn from a dark purple to a wine-red.
5. Remove organic impurities: add 8.25 g (i.e., 0.75 g per 100 ml-up to 1.0 g per 100 ml is acceptable if needed) of activated charcoal, shake vigorously for 2 min, then filter through #40 Whatman filter paper using a Buchner funnel. Use the minimum amount of charcoal necessary to decolorize the stain because this also removes dye molecules. The stain should be water-clear, but minor pink or yellow tinting is not a problem.
6. Storage: Schiff reagent can be stored in a full, tightly capped bottle in a refrigerator (~4C). Exposure to light should be minimized and the stain should be discarded when a white precipitate begins to form. However, note that the use of freshly prepared stain is recommended.
To prepare ~3300 mL of bisulfite solution:
1. Add 150 ml of 10% sodium metabisulfite solution to 3000 ml of distilled water. This can be prepared as a stock solution. 10% potassium metabisulfite (K2S2O5) solution is also acceptable.
2. Add 150 ml of 1.0 N HCl and mix.
Appendix 2
Outline of the Feulgen staining protocol recommended in the present study. Choice of fixative, fixation time, hydrolysis time, and stain time will all depend on the genome size, cell type, slide age, and other factors. This protocol is based on air-dried smears of nucleated vertebrate erythrocytes. Fixation and hydrolysis times can be as short as 30 min, depending on the nature of the cells being stained. A hydrolysis curve (IOD vs hydrolysis time) should be plotted to determine the best protocol for the specimens to be measured. All steps should be conducted at RT (20–23C). A weak HCl solution can be used to wash glassware and to clean spills of Schiff reagent (known in our lab as “the pink menace”).
Appendix 3
Checklist of steps in a generalized Feulgen image analysis densitometry protocol.
Acknowledgments
Supported by a University of Guelph Alumni Doctoral scholarship to TRG, Natural Sciences and Engineering Research Council of Canada (NSERC) postgraduate scholarships to TRG and DCH, and an NSERC research grant to PDNH.
We wish to thank all those who provided samples or technical advice: Marc Freeman, Jean Joss, Lloyd Kinzer, Denis Lynn, John Phillips, Kate Sheridan, Adrian Sumner, Phil Wiebe, Tony Wood, and the staff of the Arkell Poultry Research Facility. Our most sincere thanks to Ellen Rasch and Grace Wyngaard for their invaluable input and generous hospitality.