0 1992 Wiley-Liss, Inc.

Cytometry 13:846-852 (1992)

Quantification of Fluorescence In Situ Hybridization Signals by Image Cytometry' P.M. Nederlof: S. van der Flier, N.P. Verwoerd, J. Vrolijk, A.K. Raap: and H.J. Tanke Sylvius Laboratory, Department of Cytochemistry and Cytometry, University of Leiden, 2333 A1 Leiden, The Netherlands Received for publication June 28, 1991; accepted March 18, 1992

In this study we aimed at the development of a cytometric system for quantification of specific DNA sequences using fluorescence in situ hybridization (ISH) and digital imaging microscopy. The cytochemical and cytometric aspects of a quantitative ISH procedure were investigated, using human peripheral blood lymphocyte interphase nuclei and probes detecting high copy number target sequences as a model system. These chromosome-specific probes were labeled with biotin, digoxigenin, or fluorescein. The instrumentation requirements are evaluated. Quantification of the fluorescence ISH signals was performed using an epi-fluorescence microscope with a multi-wavelength illuminator, equipped with a cooled charge couple device (CCD) camera. The performance of the system was evaluated using fluorescing beads and a homogeneously fluorescing specimen. Specific image analysis programs were developed for the automated segmentation and analysis of the images provided by ISH. Non-uniform background fluo-

Different specific staining procedures as well as flow and image cytometric equipment have been developed to determine the total nuclear DNA content of single cell nuclei. However, even with a precision of 2% coefficient of variation (CV) obtainable by these methods (21) the loss or gain of a n averaged size chromosome, containing thousands of genes, may remain undetectable. To be able to detect much smaller DNA aberrations, quantification of specific DNA sequences may be required. In situ hybridization (ISH) procedures allow the detection of specific DNA and RNA sequences in individual cells, but have been optimized for visual interpretation and are in general not quantitative. Quantitation of large DNA and RNA target se-

rescence of the nuclei introduces problems in the image analysis segmentation procedures. Different procedures were tested. Up to 95% of the hybridization signals could be correctly segmented using digital filtering techniques (min-max filter) to estimate local background intensities. The choice of the objective lens used for the collection of images was found to be extremely important. High magnification objectives with high numerical aperture, which are frequently used for visualization of fluorescence, are not optimal, since they do not have a sufficient depth of field. The system described was used for quantification of ISH signals and allowed accurate measurement of fluorescence spot intensities, as well as of fluorescence ratios obtained with double-labeled probes. o 1992 Wiley-Liss, Inc. Key terms: Quantification, CCD camera, image analysis, chromosome polymorphism

quences by fluorescence ISH and flow cytometry has been reported (5, 7, 22). However, flow cytometry was not successful for detection and quantification of

'This study was sponsored in part by the Netherlands Organization for Scientific Research (NWO) grant 534-060,and Pioneer Subsidie N W O grant 90-129.90. 'Present address: Dr. P.M. Nederlof, Department of Laboratory Medicine, Division of Molecular Cytometry, MCB 230, University of California, San Francisco, San Francisco, CA 94143-0808. 3Address reprint requests to Dr. A.K. Raap, Sylvius Laboratory, Department of Cytochemistry and Cytometry, University of Leiden, Wassenaarseweg 72, 2333 A1 Leiden, The Netherlands.

QUANTITATIVE ISH BY IMAGE CYTOMETRY

smaller target sequences. The main reason is that in flow cytometry the total nuclear fluorescence is measured whereas the volume of the specific signal is small compared to the total nuclear volume. The detection is therefore mainly determined by the signal to noise ratio of the specific signals and the background fluorescence of the nucleus. Another approach uses image cytometry, which has the advantage that image processing techniques can be used to select and measure only objects of interest. ISH can be visualized using different types of light microscopy (brightfield, reflection contrast, or fluorescence microscopy) depending on the properties of the cytochemical label. For quantitative ISH, fluorescence is preferred for a number of reasons. Firstly, fluorescence provides higher sensitivity. Secondly, fluorescence has the advantage that it can deal with the large variations in intensities that are the result of biological variations, and variations introduced by the procedure. Finally, selective spectral separation of excitation and emission of dyes allows simultaneous detection of multiple targets. Quantification in absorbance cytophotometry is hampered by the broad absorption spectra and spectral overlap of the dyes, and is especially difficult when dyes are co-localized (8, 9, 15). In fluorescence microscopy spectral separation of fluorochromes is obtained by differences in excitation and emission properties of the dyes. At present at least three different fluorochromes are available for fluorescence ISH: fluorescein-isothiocyanate, tetramethyl rhodamine isothiocyanate, and aminomethyl coumarin acetic acid (FITC, TRITC, and AMCA, respectively). Other potential fluorochromes, such as the phycobiliproteins, are less suitable for ISH because of their high molecular weight, which decreases their ability t o reach the target in the preparation. Laser scanning microscopes can be used to visualize ISH, but are generally equipped with one laser, which limits the possibility of performing multiple (color) ISH. Recent improvements of excitation and emission filters and development of highly sensitive video cameras such as intensified silicon-intensified targets (ISIT) and solid-state cameras (CCD: charge coupled device), together with the increasing computer performance, now open the way for a quantitative approach for ISH using digital imaging microscopy (3-5, 14, 18, 20). The procedure for quantitative ISH is complex, and consists of many individual steps a t the cytochemical and cytometrical level. To obtain meaningful quantitative data from ISH experiments, each of the procedural steps should be evaluated for its influence on the final quantitative outcome. Optimization of the ISH procedure for visual interpretation is not necessarily optimal for quantification. To determine the optimal hybridization protocol and the best instrumentation setup, one should keep in mind that the two are not independent: the characteristics of the signals provided by the cytochemical procedure will determine the instrumentation needed, and vice versa.

847

The instrumentation required consists of a microscope system, a detector for collection of images, and computer hardware and software for image storage and analysis. It is common practice that for fluorescence microscopy objective lenses are used with high magnification and high numerical aperture, since these characteristics determine the spatial resolution and light collecting properties. However, a drawback of these objectives is their limited depth of field, which may limit their usefulness for quantitative microscopy. The detector used for imaging the ISH results should have sufficient sensitivity to detect the relatively weak fluorescing signals obtained by ISH. Ideally, the detector also should provide a linear response to a wide range of wavelengths, have a high signal to noise ratio, and have wide dynamic range, and little geometric aberrations. The cooled CCD camera fulfills these requirements (2,ll) and was therefore used in this study. In this paper we aim a t a systematic investigation of procedures for quantitative ISH. For this purpose, specific hard- and software were developed for the automatic analysis of the ISH images. The system described was used for quantification of ISH signals and allowed accurate measurement of fluorescence spot intensities (16), as well as of fluorescence ratios obtained with double-labeled probes (17).

MATERIALS AND METHODS Preparation of the Slides In situ hybridization was performed as described in ref. 16. Preparations used contained human lymphocytes that were hybridized with probes specific for the centromeric region of chromosomes 1 and 7 labeled with biotin or digoxigenin and detected with FITC or TRITC. Slides were mounted in phosphate-buffered saline (PBS)/glycerol (1: Sviv) containing diamidinophenylindole (DAPI, 0.5 pg/ml) as a blue fluorescing total DNA counterstain, and 2.3% (wiv) 1,4-diazobicyclo(2,2,2)-octane (DABCO) (Sigma, St. Louis, MO) as an anti-fading agent (13).

Microscopy The epi-fluorescence microscope (Leitz Orthoplan) was equipped with a 100 W mercury-arc lamp. The results were obtained with a 40x oil 1.3-0.8 NA objective lens with iris diaphragm (CF Fluor, Nikon), which proved to have sufficient depth of field for the preparations used. For the visualization of the different f luorochromes, specific combinations of selection filters were inserted to obtain narrow selection of both the excitation and emission wavelengths. The filtersets used were a compromise between optimal excitation of the fluorophores, maximal light collection, and minimal crosstalk between the fluorophores at the different wavelengths, since the fluorochromes showed overlap in excitation and emission spectra. The combinations of excitation filters, dichroic mirrors, and emission filters used are given in Table 1.

848

NEDERLOF ET AL.

Table 1 Filter Sets Used for Multiple Fluorescence ISH Analysis"

DAPI FITC TRITC

Wavelength h m i Dichroic Excitation filters mirror DM400 LP340-SP380 LP450-SP490 DM510 DM580 LP530-SP580

Emission filters LP420-SP560 LP515SP560 LP580

"DM,dichroic mirror; ISH, in situ hybridization; LP, long wave pass filter; SP, short wave pass filter.

Image Collection and Analysis Images were recorded using a cooled CCD camera (Photometrics, Tucson, Arizona) with a Kodak chip with 1,024 x 1,345 elements of 6.8 x 6.8 pm2 each. The 12 bit images were transferred using a DMA-VME interface to a memory board installed in a SUN SPARC 330 workstation. The center of the camera target with a size of 512 x 512 pixels was used. Image sizes were reduced to 256 x 256 during readout of the chip by using the binning option. The binning procedure reduced the spatial resolution, but for the experiments described this was not crucial. An advantage of the binning was the increased image processing speed and reduction of the required image storage space. Cell density on the preparations was such that with the objective used (40 X ) an average of four nuclei were present in each image. Image Collection. First, series of images were collected and stored on (optical) disk, after which they were analyzed by the program in batch, without user interaction. For the selection of FITC, TRITC, and DAPI, different filter blocks were used that contained various fluorescence selection filters, as listed in Table 1. The slightly different position (angle) of the interference filters of different filter blocks caused shifting of the images. The image displacements could not, in practice, be corrected routinely by adjustment of the filter position. However, the image shifts were reproducible and so could be corrected by a pixel-wise shift of the digitized images. The image shift was determined using small fluorescing spheres that were excited a t various wavelengths. Selection of a new microscope field and focusing were performed visually using the DAPI fluorescence of the stained nuclei with UV excitation. The FITC and TRITC images were collected without additional visualization. For each microscope field three images of 256 x 256 pixels were recorded, a FITC, TRITC, and a DAPI image. The exposure time for each slide and fluorophore was determined empirically aiming at maximal gray level resolution (selecting for high intensities with restriction of the maximum pixel values t o 4,095), and minimal bleaching (6). The average exposure times were 1 to 3 s for the different fluorophores.

Image analysis. All programs were developed using the TCL-image software package developed at the Technical University of Delft (Multihouse, Amsterdam, the Netherlands). 1. Correction of image shift. As a first step, the image shifts that resulted from the slightly different position (angle) of the interference filters of different filter blocks were corrected by a pixel-wise shift of the images according to the measured image shift with fluorescent spheres. 2. Selection of the nuclei. For the segmentation of the DAPI-stained nuclei from the background, a global histogram-based segmentation method [Isodata (1911 was applied. This resulted in a binary mask image. Objects touching the border and dirt particles were removed on the basis of, their position and their (smaller) size, respectively. The area of the lymphocyte was on the order of a few thousand pixels. Dirt particles on the slides usually had an area smaller than 100 pixels, and this value was used to remove these dirt particles. 3. Selection of the hybridization spots. For correct segmentation of the hybridization spots from the background fluorescence on the nucleus, first the local background was estimated using digital filtering techniques [minimum/maximum filter (23)]. This estimated background image was subtracted from the original image and thereafter a fixed threshold procedure was performed.

The program used a fixed value for the size of the (round) structuring element of the min-max filter. The optimal size of this filter for a specific set of images was determined before the analysis, by minimizing the size to that of the (largest) ISH spot, in order to obtain an optimal estimation of the local background. The size of the structuring element depended on the probe and magnification used, and was in general between 25 and 169 pixels. For the experiments described the filter size was 49 pixels. For each nucleus within an image the corresponding hybridization spots were determined. For each spot the size of the area in number of pixels, the integrated intensity of all pixels within this area corrected for background, the highest pixel intensity, and the average intensity per pixel were determined. These data were combined with information on the size and background intensity of each nucleus. All values were stored in data files.

RESULTS AND DISCUSSION Performance of the Imaging System The uniformity of illumination of the microscope was measured using a homogeneous fluorescent preparation that consisted of uranyl glass. A decrease in intensity of 5% was measured within the central 9002 pixels of the CCD ship, comparing the central pixels of this area with the most outward pixels. The decrease in intensity was 2.5% for the central area of 512' pixels. This 512' area was used in the experiments.

QUANTITATIVE ISH BY IMAGE CYTOMETRY

The accuracy of the imaging system was evaluated using fluorescent microspheres with a mean diameter of 2.99 pm and FITC fluorescence (Becton Dickinson), and a coefficient of variation in intensity of 2% according to the manufacturer. These spheres showed a coefficient of variation of 2.8% when analyzed on a flow cytometer (Facstar, Becton Dickinson). The same microspheres were analyzed with the imaging system, using the same microscope and detector settings as used in the experiments described. A number of images were collected and the integrated intensities of 120 spheres were measured. The coefficient of variation was 3.2% when no correction for uneven illumination was applied (2, 11).

849

original image, as is illustrated in Figure 1C and D. The resulting image still contains some background noise in addition to the specific ISH signals. To obtain a binary mask with the boundaries of the spots for the measurement of the integrated spot intensities, a fixed threshold value could be used. The heterogeneous background of the nuclei not only causes problems for the segmentation procedures, it also introduces errors in quantification. In Figure 2 this is illustrated for the same two ISH spots, which have different local backgrounds. The min-max filter size was varied and a fixed threshold value was used to obtain a binary mask. The binary mask was used to determine the inkegrated intensity of the spots and the integrated intensity corrected for background. Three Image analysis effects can be seen. Firstly, when the filter size is too For the analysis of in situ hybridization images, spe- small the binary mask obtained will be smaller than cific application programs were developed using gen- the actual ISH spot (Fig. 2A) and therefore the inteeral image analysis software. To prevent subjective se- grated intensity will be reduced (Fig. 2B) Secondly, the lection of nuclei and spots and to increase speed, the estimated background will be too high for small minimage analysis procedure was automated (without user max filter sizes (see also Fig lA,B) and will reduce the interaction). intensity when used for correction. Thirdly, even when The segmentation procedure is a critical step within the filter is of the right size, correction for local backthe quantitative procedure. It was observed that het- ground intensity is necessary since the two spots in erogeneous background of the nuclei caused major Figure 2 have different local background, as illustrated problems in the segmentation procedure. Different al- by the gap between the two lines for the uncorrected gorithms for the automated segmentation of the spots intensities in Figure 2B. Subtraction of the local backwere examined, It frequently appeared that the seg- ground results in the integrated intensity of the spementation of the spots could not be performed by any cific signal. This illustrates that the accuracy of quanfixed threshold method based on the gray-level histo- tification of the ISH spots is highly determined by the gram of the total image or of the nucleus alone. The accuracy of the local background estimation. The size main reason was the heterogeneous background f luo- of the spots and therefore the size of the structuring rescence of the nucleus, caused by auto-f luorescence element of the min-max filter will depend on the size of the target sequence detected by the probe, the microand non-specific staining. The use of digital filtering techniques to estimate the scopic magnification, and the detector used. Therefore, local background intensities, and in particular the the optimal size of the structuring element of the m i d min-max filtering technique (23), increased the num- max filter should be determined for different experiber of successfully segmented spots to over 95%. The ments. determination of the optimal size of the structuring Objects at the border of the image, and also small element of this filter was found to be very important. dirt particles, could be easily removed by automatic The size of the filter should be larger than the largest routines. In some cases, in addition t o the specific ISH object (spot) to be selected, whereas, on the other hand, signals, minor binding sites were observed. They could the size should be minimized to obtain a more precise usually be removed on the basis of their (smaller) size estimation of the local background. It was found that, if and intensity. However, when the minor binding sites the size of the chosen filter was too small, part of the were located in close proximity to the specific spots, signal was interpreted as background, resulting in an correct separation was not possible. Therefore, much attention was given to performing the cytochemical underestimation of the spot intensity. In Figure 1the effect of the filter size on the estima- procedures under high stringency conditions to prevent tion of the local background is illustrated. The figure as much as possible the formation of minor binding shows intensity profiles along two lines of a digital sites. Different stringency conditions were tested for image that cut through two ISH spots within the same the hybridization and for the wash steps. At 60% fornucleus. Two different filter sizes are shown with sizes mamidi2 x SSC, 37"C, usually no minor binding sites of 15 and 31 pixels, respectively. For both ISH signals were present. (Fig. 1A,B) a filter size of 15 pixels was obviously too Following the analysis of all images, the correctness small, since the estimated background can be seen to of the automatic segmentation of the spots could be contain part of the specific ISH signal. The filter size of examined by analyzing the stored mask images. Data 31 pixels, on the other hand, followed the contour of the from improperly segmented nuclei or ISH spots, or nunucleus smoothly. clei with ISH spots that were out of focus (as judged by The estimated background was subtracted from the eye), were not included for further analysis.

NEDEKLOF ET AL.

850

1800 -

A

!I

% 1 4 0 0 --

.-g

c,

Q,

c,

.-C

"

1 2 0 0 -1000

B

d

1 6 0 0 --

--

II

Q. 400

2oo

t

2oo

t pixel

pixel 1800

1800

C

1600

1600

a

>.

1400

CI

.-g

Q)

Q)

.-

gJ 1 2 0 0

CI

.--C

1400

c,

1000

C .-

c,

800

1200 1000 800

Q,

.x

600

Q

200

,

,

,

0

. ' : I

i.-,r

,a,_

: - ' II

,

, .."

,,,

Quantification of fluorescence in situ hybridization signals by image cytometry.

In this study we aimed at the development of a cytometric system for quantification of specific DNA sequences using fluorescence in situ hybridization...
647KB Sizes 0 Downloads 0 Views