DE29824468U1 - Chromosome classification device - Google Patents

Description

# Applicant: Applied Spectral Imaging Ltd.

^f ' Our reference: 50411 DE

CHROMOSOME CLASSIFICATION DEVICE
FIELD AND BACKGROUND OF THE INVENTION

The present invention relates to the classification of in situ stained chromosomes into a staining (spectral) karyotype. In particular, the present invention relates to a method for identifying prominent internal reference spectra to perform such classification.

The use of fluorescent dyes (i.e. fluorescent probes or markers, fluorophores, fluorochromes, all used interchangeably in this document) represents one of the most powerful and common tools for analyzing tissues and cells. Fluorescence microscopy is therefore one of the most important experimental techniques used in light microscopy [Lakowicz, Principles of fluorescence spectroscopy, Plenum Press, New York, London (1983)].

The advantage of fluorescent probes is mainly due to the wide variety of biological structures to which certain dyes can be bound [Waggoner, Applications of fluorescence in the biomedical sciences, Taylor et al. (eds.), New York: Alan Liss, Inc., (1986)

3-28]. For a detailed review of fluorescent probes, see Mason ^ (ed.) Fluorescent and Luminescent Probes for Biological Activity, Biological Techniques Series, edited by Sattelle, Academic Press Limited, London (1993), and Ploem and Tanke, Introduction to Fluorescence Microscopy, Oxford University Press, Royal Microscopical Society (1987).

The rapid development of new and more sophisticated multicolor fluorescent dye molecules continually creates a need for more advanced fluorescence imaging techniques that can exploit the full potential of these dyes. For an explanation of the revolutionary impact that fluorescent dyes have had and will continue to have on the research currently being conducted, see Taylor et al., The New Vision of Light Microscopy, American Scientist, Vol. 80 (1992) 322-335.

An important example where the detection of multiple fluorescent probes can provide a significant advantage is FISH (fluorescent-irt-j/ta hybridization) [Emanuel, Growth Genetics and Hormones 9 (1993) 6-12], which is used to analyze genes at the chromosomal level and to detect possible genetic lesions, such as gene/chromosome amplification, deletion, translocation, rearrangement, and other gene and/or chromosome-associated abnormalities.

Certain diseases and disorders, including many cancers and congenital defects, are genetic disorders caused by lesions (i.e., mutations) in one or more genes. Many other diseases are known or suspected to have a genetic component(s), i.e., there is a genetic lesion(s) that does not cause the disease on its own, but contributes to it or increases the likelihood of developing the disease later in life. These phenomena are known in the art as multifactorial diseases and genetic predispositions.

Correlating visible genetic damage with known diseases would enable doctors to make meaningful diagnoses and allow early detection and treatment of many diseases. Genetic counseling can alert prospective parents and at-risk individuals to the possibility of potentially serious medical problems in the future, allowing appropriate interventions to be made:

More than 8,000 genetic disorders have been identified to date, many of which are associated with multiple genetic lesions. Knowing that chromosomes carry genetic information, scientists concluded that it should be possible to document visible lesions in chromosomes that are responsible for certain diseases.

In the 1960s, staining techniques for classifying metaphase chromosome spreads based on microscopy were developed. For several decades, visual analysis of chromosome banding patterns has been used to correlate human genetic disorders with observed structural abnormalities in metaphase chromosomes. To reveal distinctive light and dark bands along their length, chromosomes are usually examined by Giemsa staining (G-banding) using brightfield microscopy or by fluorescence staining (R-banding) using fluorescence microscopy. Careful comparison of a patient's banding pattern with that of normal chromosomes can reveal abnormalities,

such as translocations (exchange of genetic material between or within chromosomes), deletions (missing chromosome(s) or fragment(s) thereof), additions, inversions and other malformations and damages causing genetic diseases. To date, conventional chromosome banding techniques or techniques for the experimental generation and staining of chromosome bands are limited in terms of resolution.

Fluorescent in situ hybridization (FISH) has evolved over the past 25 years through the improvement of a number of complementary techniques. Its emergence was driven by the desire of cell geneticists to develop better tools to map the precise location of genes on chromosomes and to detect very small genetic lesions that are not visible when large areas of chromosomes are stained.

The Human Genome Project (HGP), a broad initiative to identify and map all human genes, showed interest in FISH and accelerated the development of much-needed probes. Common FISH techniques were also made possible by the competitive development of powerful immunological probes, by a growing variety of excellent fluorescent dyes for microscopy and spectroscopy, and by dramatic improvements in the objectives, illuminators, and filters used for fluorescence microscopy.

The power and usefulness of FISH is determined by many factors: (1) FISH can be used not only on isolated chromosomes and nuclei, but also on whole cells in fixed, paraffin-embedded tissue sections, (2) relatively small lesions can be detected (the ability to detect small lesions is constantly increasing), (3) results can be made available relatively quickly, (4) the moderate cost allows use in most diagnostic and research laboratories, (5) adaptation to different probes and test material types can be developed, and (6) high specificity and sensitivity can (7) be achieved within a short time, usually in the range of two hours.

Many FISH applications require the cell geneticist to simply look through the eyepiece of a microscope or at the image on the monitor and determine whether a fluorescent label is present or absent. For less complex specimens, a simple count of one or two colored labels may be performed. The ability to manipulate digital images to extract numerical data from them,

adds a comprehensive new set of capabilities to the FISH techniques.

A suitable imaging technique can enhance very faint FISH images so that the labeled chromosomes and loci can be clearly identified. Under easily accessible experimental conditions, the number of labeled sites can be counted automatically. In addition, the intensity at each labeled site can be recorded and the amount of DNA calculated to show, for example, the number of copies of a particular gene present.

As discussed above, FISH provides information about the location of the labeled probe, the number of labeled sites on each chromosome, and the intensity of labeling (the amount of genetic material) at each site. Centromeric probes (repetitive DNA) and chromosome paints are used to label and count the number of copies present on each targeted chromosome. Locus-specific probes are used to map the location of small regions of genetic material. These types of probes can be used on an intact interphase nucleus as well as on metaphase chromosome spreads and can be counted visually or automatically by an appropriate algorithm. They are routinely used to identify genetic disorders characterized by having too many or too few copies of a particular chromosome, chromosome fragment, or gene.

In very early stages of some cancers, long before the cells are recognized as abnormal, an increase in the number of certain genes may occur, a phenomenon known in the art as gene amplification, which can be detected as homogenously stained regions (HSR) and/or double minute chromosomes using locus-specific probes. Using FISH to detect chromosomal abnormalities in cancerous cells can provide clues about the stage of disease development, allowing selection of the most appropriate treatment(s), with many of the treatments being stage-specific in their effectiveness.

It is possible to label the entire surface of a particular chromosome uniformly by isolating the chromosome (for example, using flow cytometry), physically (for example, by sonication), or enzymatically (for example, by endonucleases), breaking it up and generating a set of probes against all the fragments. Whole chromosome probes, also known as chromosome stains, label in a fluorescent manner

all copies of their target chromosome. An important application of chromosome staining is to detect translocation of genetic material between two chromosomes, which occurs characteristically in the earlier stages of certain cancers, but other chromosomal aberrations can also be detected.
5

For example, if chromosome A is specifically labeled with a green paint and chromosome B is specifically labeled with a red paint, a translocation of genetic material from A to B will appear as a green region adjacent to a red region (and vice versa).

Typically, chromosome stains generated from normal chromosomes are used to detect deletions or translocations on abnormal (patient) chromosomes. Reverse chromosome staining uses probes generated from an abnormal chromosome to identify DNA from several normal chromosomes that contributed material to the abnormal chromosome.

The method of the invention, as exemplified in the Examples section below, enables staining of the 24 different chromosomes comprising the human karyotype (i.e., genome), each in a different color, and simultaneously detecting, identifying and representing a human color karyotype using a single hybridization procedure followed by a single brief detection.

The introduction of combinatorial fluorescence strategies (e.g., combinatorial labeling and combinatorial hybridization) using various combinations of a few basic fluorescent dyes represents a marked improvement over the multicolor fluorescent dyes used to label chromosome stains. For further details on combinatorial labeling, see Ried et al., Simultaneous visualization of seven different DNA probes by in situ hybridization using combinatorial fluorescence and digital imaging spectroscopy, Proc. Natl. Acad. Sci. USA 89 (1992) 1388-1392; and Ried, Fluorescence in situ hybridization in genetic diagnostics, Faculty of Theoretical Medicine, Ruprecht-Karls-University Heidelberg (Jan. 1994), both of which are incorporated by reference as if fully disclosed herein. For further details on combinatorial hybridization, see du-Manoir et al., Detection of complete and partial chromosome gains and losses by comparative genomic in situ hybridization. Hum. Genet. 90 (1993) 590-610, which is incorporated by reference as if fully disclosed herein).

Numerous methods for labeling DNA probes are available for use in FISH assays, including indirect methods whereby a hapten such as biotin or digoxigenin is incorporated into the DNA using enzymatic reactions. Following hybridization to a metaphase chromosome spread or an interphase nucleus, a fluorescent label is attached to the hybrid using immunological methods. Recently, fluorescent dyes have been incorporated directly into probes and detected without the use of an intermediate step. Standard FISH dyes include fluorescein, rhodamine, Texas red and cascade blue, and a multiprobe FISH analysis method can be achieved by labeling different probes with different haptens or fluorescent dyes and combinations thereof, which is known in the art as combinatorial labeling [see Ried et al., Simultaneous visualization of seven different DNA probes by in situ hybridization using combinatorial fluorescence and digital imaging microscopy, Proc. Natl. Acad. Sci. USA 89, (1992) 1388-1392; and Ried, Fluorescence in situ hybridization in genetic diagnostics, Faculty of Theoretical Medicine, Ruprecht-Karls-University Heidelberg (Jan. 1994), both of which are incorporated by reference as if fully set forth herein.] Alternatively, a pool of a given probe can be divided into subpools, each labeled with a different fluorophore, and the subpools regrouped to otherwise achieve similar hybridization results, a process known in the art as combinatorial hybridization [see du-Manoir et al., Detection of complete and partial chromosome gains and losses by comparative genomic in situ hybridization, Hum. Genet. 90, (1993) 590-610, both of which are incorporated by reference as if fully set forth herein]. According to both labeling strategies, combinatorial probes are obtained. Consequently, when either of the terms "combination of fluorophores" or "combinatorial fluorescence strategy" is used in this document and in particular in the claims below, reference is made to both combinatorial labeling and combinatorial hybridization as described above.

The use of combinatorial fluorophores for chromosome staining and karyotyping, multicolor chromosome banding, and comparative genomic hybridization is described in detail in U.S. Patent Application No. 08/635,820 filed April 22, 1996, and in the journal Science [E. Schroeck et al., Multicolor spectral karyotyping of human chromosomes, Science 273 (1996) 494-497], both of which are incorporated by reference as if fully disclosed herein.

The main advance described in Science is that a complete

Genome scanning by spectral imaging enables instant visualization of defined emission spectra for each human chromosome after fluorescence m-siYw hybridization (FISH). By computer separation (classification) of spectra, spectrally overlapping chromosome-specific DNA probes are resolved and all human chromosomes are identified simultaneously.

The spectral imaging approach described therein combines Fourier spectroscopy, charge coupled device (CCD) imaging, and optical microscopy to simultaneously acquire visible and near-infrared emission spectra at all points in the sample. This allows the use of multiple spectrally overlapping samples. The approach is based on the acquisition of a discrete spectrum (identified by a sequence of intensities acquired at each pixel at many different wavelengths), which facilitates the discrimination of multiple fluorophores. In dramatic contrast to conventional epifluorescence microscopy, where fluorochrome discrimination is based on detection of fluorescence intensity through a small number of fluorochrome-specific narrow-band optical filters [see Speicher et al., Nature Genetics (1996) 12:358-375], the use of spectral karyotyping as described here allows all available information contained in the emitted photons within the spectrum of emitted light to be used for analysis.

The spectral-based method for distinguishing spectrally overlapping fluorophores (classification) can be easily extended to a large number of fluorochromes, provided that measurable and consistent differences in the emission spectrum of each fluorochrome exist.

Simultaneous identification of every human chromosome in metaphase preparations, an approach termed spectral karyotyping, is also reported. For this purpose, chromosome-specific composite libraries generated by polymerase chain reaction (PCR) from flow-sorted human chromosomes are immediately labeled with nucleotides conjugated to five different fluorophores or combinations thereof. A composite probe set containing all 24 chromosomes is then hybridized to metaphase chromosomes. Chromosome-specific labeling is achieved by suppression hybridization. Specifically, repetitive sequences in the composite libraries are blocked by the addition of an excess of unlabeled human Cot-1 DNA.

The hybridization is shown in both RGB display and classification colors. Display colors allow all human chromosomes to be easily visualized after spectral imaging and based on spectral detections at each pixel, a chromosome classification algorithm is applied to spectral karyotype all human chromosomes. One of the most important analysis algorithms is the spectral-based classification algorithm, which allows multiple or several different spectra to be identified in the image and highlighted in classification colors. This allows a specific classification color to be assigned to all human chromosomes based on their spectra. This algorithm assumes that the (reference) spectrum of each chromosome has been acquired and stored in a reference library in the computer. Each pixel in the image is assigned a distinguishable classification color according to the classification color assigned to the reference spectrum that is most similar to the spectrum at that given pixel. A minimum square error algorithm as shown in Equation 1

X2 XL

is calculated for each pixel, where I _x , _y (X) is the normalized spectrum at pixel coordinates x, y and Λ _η (λ) represents the normalized reference spectrum for each of the chromosomes η = 1,2,..., 23,24. After calculating the S _x ,y, _n value for all reference spectra, the smallest value is chosen and an artificial classification color is assigned to this pixel, in accordance with the classification color assigned to the most similar reference spectrum.

The potential of spectral karyotyping as a screening method for chromosomal aberrations was further explored by analyzing clinical specimens from several laboratories that had previously performed conventional banding or FISH with chromosome staining probes. In all cases, G-banding and spectral karyotyping showed consistent results. In some cases, Giemsa banding was not sufficient to fully interpret the chromosomal aberrations. In these cases, the diagnosis of chromosomal aberrations was confirmed via spectral karyotyping using conventional dual-color FISH analysis. The smallest distinguishable aberration analyzed for this report was a translocation t(l;ll)(q44;pl5.3), where the reciprocal translocation was not detected by conventional banding analysis.

were recognizable. The origin of the chromosomal material contributing to the reciprocal translocation was correctly classified. The translocated segments on chromosomes 1 and 11 were confirmed by subtelomere-specific cosmid probes for chromosomes Iq and 1 Ip. Based on the locus of the probes used, the size of the alteration was estimated to be > 1500 kb. In a second case, the banding analysis procedure indicated a translocation of a segment from chromosome 4 to chromosome 12. Spectral karyotyping clearly identified and classified the origin of the additional chromosomal material as being from chromosome 4. To determine the limit of sensitivity of spectral karyotyping, one case with a submicroscopic translocation (not recognizable on either metaphase or prometaphase chromosomes) involving chromosomes Io and 17 was examined. This t(16;17) was previously detected by FISH with cosmid probes and the reciprocal chromatin exchange was estimated to be approximately 500 kb. Spectral karyotyping using metaphase chromosomes from this patient failed to identify the known t(16;17), suggesting that the sensitivity of the metaphase chromosome analysis method with currently available staining probes is between 500-1500 kb.

Hybridization of previously G-banded chromosomes was performed to demonstrate that spectral karyotyping is an approach that can be used to complement conventional banding analysis methods. Due to the trypsin digestion required for G-banding, signal intensity was likely slightly reduced compared to metaphases not previously G-banded. The loss in signal intensity was approximately 10% and could therefore be easily compensated by increasing exposure times. Slightly increased noise at the edges of previously G-banded chromosomes compared to non-G-banded chromosomes was also observed. However, metaphase classification was readily achieved.

Thus, according to the approach described above, first, chromosomes of a single normal karyotype are identified by a conventional chromosome banding technique (e.g., G-banding or R-banding). Second, the same chromosomes are hybridized with chromosome dyes as described above. Third, the pixels characterizing the average spectrum, as previously determined by chromosome banding, are assigned to a chromosome type (e.g., 1-22, X, and Y in human males), and the resulting spectra (e.g., 24 in humans) form a library of reference spectra. Fourth, the reference spectra are then used to classify pixels of new karyotypes into their chromosomes.

&iacgr;&ogr;

However, as observed experimentally, this approach is not always successful. In other words, appropriate classification of new karyotypes fails in some cases. This is particularly the case when abnormal or aberrant karyotypes, such as karyotypes of cancer cells, are analyzed. This may occur due to one or more of the following reasons. First, different (differential) hybridization efficiency. Second, chromosomes from different sources consistently behave differently in terms of hybridization, light emission, etc. And third, the acquisition device is not calibrated identically for each acquisition.

Nevertheless, it was further observed that the best classification results were obtained when reference spectra were used to karyotype the chromosomes from which they were derived. In other words, the best results were obtained when the reference spectra used were internal spectra. However, as described above, this requires identification of the chromosomes via the conventional chromosome banding technique prior to classification. This in turn limits the scope of use of the classification method since (i) the method is not easily automated, (ii) a highly skilled cell geneticist is required to analyze the conventional banding patterns of the chromosomes, and (iii) in many cases of aberrant karyotypes, such as cancer cell karyotypes, the conventional banding pattern is not informative and therefore inconclusive.

The present invention is a method for identifying internal reference spectra for chromosome classification, wherein the reference spectra are calculated substantially automatically for each karyotype from the spectral information derived from each karyotype. In other words, the present invention is a method for identifying internal reference spectra for chromosome classification, wherein the reference spectra are calculated substantially automatically for each karyotype. The use of internal reference spectra for classification overcomes the limitations of the prior art described above.

SUMMARY OF THE INVENTION

According to the present invention, a method for classifying in situ stained chromosomes into staining (spectral) karyotypes and a method for identifying prominent internal reference spectra for making such classification are provided.

11
made available.

According to further features of preferred embodiments of the invention described below, there is provided a method for finding L internal reference vectors for classifying L chromosomes or chromosome regions from a cell, the L chromosomes or chromosome regions being stained with K different fluorophores or combinations thereof, wherein each of the K base chromosomes or chromosome regions of the L chromosomes or chromosome regions is stained with only one of the K different fluorophores, while each of the other L-K of the L chromosomes or chromosome regions is stained with a different combination of the K different fluorophores, the method comprising the steps of: (a) using a multi-band collection device to detect a first vector for each pixel of each of the L chromosomes or chromosome regions, (b) identifying pixels corresponding to each of the K basic chromosomes or chromosome regions, and defining the pixels as basic pixels so as to obtain K basic classes of the basic pixels, (c) using at least one basic pixel from each of the K basic classes to obtain K basis vectors, the K basis vectors being K internal reference vectors, (d) using the K basis vectors to identify pixels belonging to the other L-K chromosomes or chromosome regions, and (e) using the pixels belonging to the other L-K chromosomes or chromosome regions to calculate the other L-K internal reference vectors, thereby finding all internal reference vectors.

According to further features of preferred embodiments of the invention described below, there is provided a method for classifying L chromosomes or chromosome regions of a cell, the L chromosomes or chromosome regions being stained with K different fluorophores or combinations thereof, in which each of the K basic chromosomes or chromosome regions of the L chromosomes or chromosome regions is stained with only one of the K different fluorophores, while each of the other L-K of the L chromosomes or chromosome regions is stained with a different combination of the K different fluorophores, the method comprising the steps of: (a) finding L internal reference vectors by (i) using a multi-band collection device to detect a first vector for each pixel of each of the L chromosomes or chromosome regions, (ii) identifying pixels corresponding to each of the K basic chromosomes or chromosome regions, and defining the pixels as base pixels to obtain K base classes of the base pixels, (iii) using at least one base pixel from each of the

K basis classes to obtain K basis vectors, where the K basis vectors are K internal reference vectors, (iv) using the K basis vectors to identify pixels belonging to other L-K chromosomes or chromosome regions, and (v) using the pixels belonging to the other L-K chromosomes or chromosome regions to calculate the other L-K internal reference vectors, thereby finding all internal reference vectors, and (b) using the L reference vectors to classify each pixel into one of the L classification classes.

According to further features of preferred embodiments of the invention described below, there is provided a method for classifying L chromosomes or chromosome regions of a cell, the L chromosomes or chromosome regions being stained with K different fluorophores or combinations thereof, wherein each of the K basic chromosomes or chromosome regions of the L chromosomes or chromosome regions is stained with only one of the K different fluorophores, while each of the other L-K of the L chromosomes or chromosome regions is stained with a different combination of the K different fluorophores, the method comprising the steps of: (a) using a multi-band collection device to detect a first vector for each pixel of each of the L chromosomes or chromosome regions, (b) identifying pixels belonging to each of the K basic chromosomes or chromosome regions and defining the pixels as basis pixels to obtain K basis classes of the basis pixels, (c) using at least one basis pixel from each of the K basis classes to obtain K basis vectors, the K basis vectors being K internal reference vectors, and (d) using the K basis vectors to identify pixels belonging to other L-K chromosomes or chromosome regions.

According to still further features in the described preferred embodiments, the method further comprises the step of assigning a distinguishable artificial color to each of the classes of classified pixels.

According to further features in the described preferred embodiments, a method is provided for classifying L chromosomes or chromosome regions of a cell, wherein the L chromosomes or chromosome regions are stained with K different fluorophores or combinations thereof, in which each of the K basic chromosomes or chromosome regions of the L chromosomes or chromosome regions is stained with only one of the K different fluorophores, while each of the other L-K of the L chromosomes or chromosome regions is stained with a

different combination of the K different fluorophores, the method comprising the steps of: (a) using a multi-band collection device to detect a first vector for each pixel of each of the L chromosomes or chromosome regions, (b) identifying pixels belonging to each of the K basic chromosomes 5 or chromosome regions and defining the pixels as basic pixels so as to obtain K basic classes of the basic pixels, (c) using at least one basic pixel from each of the K basic classes to obtain K basic vectors, the K basic vectors being K internal reference vectors, and (d) using the K basic vectors to identify pixels belonging to the other L-K chromosomes or chromosome regions.

According to still further features in the described preferred embodiments, there is provided a method for classifying chromosomes or chromosome regions of a cell, wherein the chromosomes or chromosome regions are stained with different fluorophores or combinations thereof, such that each of the chromosomes or chromosome regions is stained with a different fluorophore or combination of fluorophores, the method comprising the steps of: (a) using a multi-band collection device to detect a first vector for each pixel of each of the chromosomes or chromosome regions, (b) identifying pixels belonging to base chromosomes or chromosome regions and defining the pixels as base pixels, so as to obtain base classes of the base pixels, (c) using at least one base pixel of each of the base classes to obtain base vectors, the base vectors being internal are basis reference vectors, and (d) using the internal basis reference vectors to identify pixels belonging to the other chromosomes or chromosome regions.

According to still further features in the described preferred embodiments, the classification of each of the pixels into one of the L classification classes using the L reference vectors is effected by a linear decomposition algorithm used to find binary vectors for each of the pixels belonging to the L chromosomes or chromosome regions.

According to still further features in the described preferred embodiments, a method is provided for detecting chromosomal aberrations using the above method for chromosome classification.
35

According to still further features in the described preferred embodiments, the multi-band collection device is selected from a group consisting of a spectral

Spectral imager combined with a filter cube and a device containing several additional filter cubes and additional emission and excitation filters.

According to still further features in the described preferred embodiments, each of the first vectors includes N items, where N is an integer selected from the range 3-150.

According to still further features in the described preferred embodiments, each of the first vectors represents a spectrum.

According to still further features in the described preferred embodiments, each of the first vectors is normalized.

According to still further features in the described preferred embodiments, the identification of the pixels belonging to each of the K basic chromosomes or chromosome regions is effected by a method selected from the group consisting of (a) identifying the pixels belonging to each of the K basic chromosomes or chromosome regions using a conventional banding pattern of the L chromosomes or chromosome regions, (b) identifying the pixels belonging to each of the K basic chromosomes or chromosome regions using an RGB algorithm, and (c) identifying the pixels belonging to each of the K basic chromosomes or chromosome regions using K external basis vectors from a library.

According to still further features in the described preferred embodiments, each of the K basis vectors is an average of several basis pixels belonging to one of the basis classes.

According to still further features in the described preferred embodiments, the identification of the pixels belonging to each of the K basic chromosomes or chromosome regions using the external basis vectors from the library is accomplished by (a) using a linear decomposition algorithm to define a decomposition K vector for each pixel of the L chromosomes or chromosome regions, (b) using a high cut off value to transform each of the decomposition K vectors into a transformed binary K vector, and (c) comparing each of the transformed binary K vectors to K defined binary K vectors defining each of the K different fluorophores to thereby identify the pixels belonging to each of the

K basic chromosomes or chromosome regions.

According to still further features in the described preferred embodiments, the cell is from a human.
5

According to still further features in the described preferred embodiments, the identification of the pixels belonging to the other L-K chromosomes or chromosome regions and the calculation of the other L-K internal reference vectors is effected by (a) using a linear decomposition algorithm to define a decomposition K vector for each pixel of the L chromosomes or chromosome regions, (b) using a low cut value or range to transform each of the decomposition K vectors into a transformed binary K vector, and (c) comparing each of the transformed binary K vectors to K-L defined binary K vectors, defining each of the combinations of the K different fluorophores to thereby identify the pixels belonging to each of the L-K chromosomes or chromosome regions.

According to still further features in the described preferred embodiments, there is provided a display comprising an image of all of the chromosomes or chromosome regions of a cell, each of the chromosomes or chromosome regions being stained with a different fluorophore or combination of fluorophores, the image representing the chromosomes or chromosome regions in different, distinctive colors, each of the chromosomes or chromosome regions being assigned one of the different, distinctive colors.

According to still further features in the described preferred embodiments, there is provided a display comprising a composite image of at least some chromosomes or chromosome regions of a cell, each of the chromosomes or chromosome regions being stained with a different fluorophore or combination of fluorophores and banded with a chromosome banding technique to obtain a distinctive banding pattern (shading) for each of the chromosomes or chromosome regions, the image showing the chromosomes or chromosome regions in distinguishable colors, each of the chromosomes or chromosome regions being assigned one of the distinguishable colors, and further wherein each of the distinguishable colors acquires an intensity pattern in accordance with the banding pattern of each of the chromosomes or chromosome regions, such that both the distinguishable

Colors as well as the banding patterns of each of the chromosomes or chromosome regions are superimposed.

According to still further features in the described preferred embodiments, there is provided a display comprising a composite image of at least some chromosomes or chromosome regions of a cell, wherein each of the chromosomes or chromosome regions is stained with a different fluorophore or combination of fluorophores and banded using a chromosome banding technique to achieve a distinctive banding pattern and shape for each of the chromosomes or chromosome regions, wherein the image shows the chromosomes or chromosome regions in distinguishable colors, wherein each of the chromosomes or chromosome regions is assigned one of the distinguishable colors, and further wherein each of the chromosomes or chromosome regions acquires the distinctive shape so that both the distinguishable colors and the shapes of each of the chromosomes or chromosome regions are superimposed.

According to still further features in the described preferred embodiments, there is provided a display comprising a composite image of a chromosome or a region of a chromosome, the chromosome or region of a chromosome being shown in a color, the color including a banding pattern of alternating lighter and darker bands, both the color and the banding pattern being indicative of the chromosome region of chromosome identification.

According to still further features in the described preferred embodiments, there is provided a display comprising an image of a chromosome or a region of a chromosome, the chromosome or region of a chromosome shown in a color, the color providing cues for identifying the chromosome or region of a chromosome, the chromosome or region of a chromosome having a shape similar to its shape after being banded using a chromosome banding technique.

According to still further features in the described preferred embodiments, there is provided a display comprising a composite image of a chromosome or a region of a chromosome, the composite image including a staining image and a banding image of the chromosome or region of a chromosome.

According to still further features in preferred embodiments of the invention described below, there is provided an apparatus for displaying chromosomes, the apparatus comprising (a) a multi-band collection device supplemented with algorithms to provide a visual presentation device with an image of all of the chromosomes or chromosome regions of a cell, each of the chromosomes or chromosome regions being stained with a different fluorophore or combination of fluorophores, the image showing the chromosomes or chromosome regions in different, distinguishable colors, each of the chromosomes or chromosome regions being associated with one of the different, distinguishable colors, and (b) a visual presentation device for displaying the chromosomes or chromosome regions.

According to still further features in the preferred embodiments of the invention described below, there is provided an apparatus for displaying chromosomes, the apparatus comprising (a) a multi-band collection device supplemented with algorithms to provide a visual presentation device with a composite image of at least some chromosomes or chromosome regions of a cell, each of the chromosomes or chromosome regions being stained with a different fluorophore or combination of fluorophores and banded using a chromosome banding technique to achieve a distinctive banding pattern for each of the chromosomes or chromosome regions, the image showing the chromosomes or chromosome regions in distinguishable colors, each of the chromosomes or chromosome regions being associated with one of the distinguishable colors, and further wherein each of the distinguishable colors acquires an intensity pattern consistent with the banding pattern of each of the chromosomes or chromosome regions so that both the distinguishable colors and the banding patterns of each of the chromosomes or chromosome regions are superimposed, and (b) a visual presentation device for displaying the chromosomes or chromosome regions.

According to still further features in preferred embodiments of the invention described above, there is provided an apparatus for displaying chromosomes, the apparatus comprising (a) a multi-band collection device supplemented with algorithms to provide a visual presentation device with an image of at least some chromosomes or chromosome regions of a cell, each of the chromosomes or chromosome regions being stained with a different fluorophore or combination of fluorophores and banded using a chromosome banding technique to provide a characteristic banding pattern and shape for each of the

chromosomes or chromosome regions, the image showing the chromosomes or chromosome regions in distinguishable colors, each of the chromosomes or chromosome regions being associated with one of the distinguishable colors, and further each of the chromosomes or chromosome regions acquiring the distinctive shape so that both the distinguishable colors and the shapes of each of the chromosomes or chromosome regions are superimposed, and (b) a visual presentation device for displaying the chromosomes or chromosome regions.

According to still further features in preferred embodiments of the invention described below, there is provided an apparatus for displaying chromosomes, the apparatus comprising (a) a multi-band collection device supplemented with algorithms to provide a visual presentation device with a composite image of a chromosome or region of a chromosome, the chromosome or region of a chromosome being shown in a color, the color including a banding pattern of alternating lighter and darker bands, both the color and the banding pattern indicative of the chromosome region of chromosome identification, and (b) a visual presentation device for displaying the chromosomes or chromosome regions.

According to still further features in preferred embodiments of the invention described below, there is provided an apparatus for displaying chromosomes, the apparatus comprising (a) a multi-band collection device supplemented with algorithms to provide a visual presentation device with a composite image of a chromosome or region of a chromosome, the chromosome or region of a chromosome being shown in a color, the color providing cues for identifying the chromosome or chromosome region, the chromosome or chromosome region having a shape similar to its shape after banding using a conventional chromosome banding technique, and (b) a visual presentation device for displaying the chromosomes or chromosome regions.

According to still further features in preferred embodiments of the invention described below, there is provided an apparatus for displaying chromosomes, the apparatus comprising (a) a multi-band collection device supplemented with algorithms to provide a visual presentation device with a composite image of a chromosome or region of a chromosome, the composite image comprising a color image and a banded image

of the chromosome or region of a chromosome, and (b) a visual presentation device for displaying the chromosomes or chromosome regions.

The present invention successfully overcomes the disadvantages of the currently known configurations by providing a method for chromosome classification which uses internal reference vectors for pixel classification, which internal reference vectors usually give better performance. The method is easily automated. Using this method, therefore, semi- or unskilled cell geneticists are enabled to obtain clear and informative karyotypes which may include superimposed spectral or spatial information.

BRIEF DESCRIPTION OF THE DRAWINGS

The invention is described herein by way of example only with reference to the accompanying drawings, in which:

FIG. 1 is a block diagram illustrating the major components of an imaging spectrometer constructed in accordance with U.S. Patent Application No. 08/392,019, now U.S. Pat. No. 5,539,517 (prior art);

FIG. 2 illustrates a Sagnac interferometer as used in an imaging spectrometer according to U.S. Patent Application No. 08/392,019, now U.S. Pat. No. 5,539,517 (prior art);

FIG. 3 a shows chromosomes classified according to the prior art classification method described in U.S. Patent Application No. 08/635,820, filed April 22, 1996, and in the journal Science [E. Schroeck et al., Multicolor spectral karyotyping of human chromosomes, Science, 273 (1996) 494-497] using artificial colors;

FIG. 3b shows the chromosomes from Figure 3a subjected to R-banding with DAPI;

FIG. 3c shows the arranged karyotype of the chromosomes of Figure 3a;

FIG. 4a-b shows the chromosomes from Figures 3 a-c, which have now been imaged using an RGB

Algorithm classified, according to the prior art described in US Patent Application No. 08/635,820, filed April 22, 1996 and in the journal Science [E. Schroeck et al., Multicolor spectral karyotyping of human chromosomes, Science, 273 (1996) 494-497], as a spread and as a karyotype;
5

FIG. 5 shows an example of the weighted functions w _T , w _g , W b for the RGB algorithm;

FIG. 6 shows the steps of integrating RGB maps for chromosome classification, where W _x , w _g , w\> are simple quadratic weighting functions;
10

FIG.7 shows the five external base spectra derived using the exemplary experimental procedure in which the fluorophores and base chromosomes are specified;

FIG. 8a-e shows drawings representing the number of pixels having a value (0-1.0) of the decomposition coefficients α, β, γ, 5, ε, respectively,

FIG. 9a-c shows chromosomes of a normal man classified according to the method of the invention; chromosomes of the same man are shown in RGB; and a combined karyotype, in each case,

FIG. 10a-b shows an RGB chromosome spread analyzed similarly to the spread in Figure 4a and whose banding pattern is the result of conventional R-banding (DAPI) in each case, and
25

FIGS. 11a-b show two images which are superimposed composites (composite image) of Figures 10a and 10b according to the invention.

DESCRIPTION OF THE PREFERRED EMBODIMENTS

The present invention relates to a method for classifying in situ stained chromosomes into a staining (spectral) karyotype that can be used to detect chromosomal aberrations. In particular, the present invention relates to a method for identifying prominent internal reference spectra for such classification, wherein the internal reference spectra are used to provide better classification results.

•«

The principles and operation of the chromosome classification method of the present invention will be better understood by reference to the drawings and accompanying descriptions.
5

Spectral imaging is the technology that enables the detection of the spectrum of light emitted by each point (pixel) of an object. A spectral imaging device (also referred to herein as an imaging spectrometer) is an instrument that detects the spectrum of light emitted by each point of the object placed within its field of view and stores it in memory for later retrieval. A spectral map is a collection of spectra of the object captured by a spectral imaging device. It is usually constructed as an intensity function defined in three-dimensional space where two dimensions are an image (x and y) and one is a spectral axis (λ). As such, a spectral map is usually referred to as a data "cube" or a "spectral cube."

The spectrum of each pixel is actually an N vector, where N is an integer that depends on the spectral range and spectral resolution of the imaging device.

The prior art teaches different methods for acquiring spectral images (i.e., spectral cubes). Devices designed according to these methods typically include light collection optics, a dispersion element (e.g., a grating or prism), a filter (or several filters) (e.g., AOTF or LCTF) or an interferometer, focusing optical elements, and a two-dimensional detector array (usually a CCD in the visible range and other types of detectors in the infrared range).

Each method has advantages and disadvantages, but a spectral imaging device based on a special type of triangular interferometer has advantages in terms of sensitivity, compactness and stability that more conventional spectral imaging devices do not possess, as can be seen in U.S. Patent Application No. 08/392,019 by Cabib et al., filed February 21, 1995, now U.S. Patent No. 5,539,517, issued July 23, 1996, and in Journal of Microscopy [182 (1996) 133-140], both of which are incorporated by reference as if fully disclosed herein. However, it should be noted that an N vector for each pixel can also be achieved using narrowband and/or broadband emission filters in combination with appropriately selected excitation filters, all of which can be used with the

22

ten fluorophores. Consequently, intensity values can be obtained for each pixel with each emission filter, the values forming an N vector for each pixel, where N is equal to the number of emission filters used. Unless a single excitation filter is used in combination with all emission filters, this vector does not represent what would be considered a low-resolution spectral vector. However, careful selection of filters and fluorophores makes it possible to obtain N vectors that allow distinguishing between the fluorophores. Therefore, the classification procedure described here can also be applied to these vectors.

A method for chromosome classification using mainly narrow band filters is disclosed in Speicher R. M., Ballard S. G. and Ward C. D., Karyotyping human chromosomes by combinatorial multi-flour FISH, Nature genetics (1996) 12:368-375, which is incorporated by reference as if fully disclosed herein. In this case the number of excitation filters is six and the N-vectors detected are therefore in fact 6-vectors.

A spectral imaging device according to the invention disclosed in U.S. Pat. No. 5,539,517 was developed by Applied Spectral Imaging Ltd., Industrial Park, Migdal Haemek, Israel and is hereinafter referred to as SpectraCube™. This instrument can provide N vectors, where in the visible range, N is equal to approximately 30-50, usually 40.

The importance of spectral imaging is that the spectrum of light carries information about the composition of the matter from which the object is formed and can therefore be used to map and visualize phenomena that are otherwise not apparent or distinguishable. Since color imaging is the next step after black and white imaging, spectral imaging is the next step after color imaging. Like the difference in the green tones of leaves of two different types of trees, or of a young and an old leaf, two fluorescent dyes, such as Texas red or rhodamine, appear to the human eye as the same color but can be easily distinguished by a spectrograph with a resolution of 10 nanometers. A complex biological system, such as a Giemsa-stained white blood cell, appears to the eye through the microscope when white light is transmitted as an object with structures composed of regions of purple, blue, and reddish colors of varying degrees of intensity. Since the colors as perceived by the human eye are composed of combinations of only three colors, red, green and blue (RGB), the number of different regions

in the cell that can be classified by color. For each point in the same cell, a spectral imaging device measures a spectrum or other vector depending on the chemical materials present at that point, as a function of wavelength, which involves on the order of fifty data (depending on spectral resolution), rather than just three as for a color image. As a result, small spectral differences or shifts between pixels can be distinguished by a spectral imaging device that the eye would recognize as belonging to the same color, and therefore considerably more classes of biological structures or components in the cell can be distinguished using a spectral imaging device compared to the human eye.

For example, in Fluorescence Imaging Spectroscopy and Microscopy, edited by X. F. Wang and B. Herman, Vol. 137, pages 87-124, 1996, John Wiley & Sons, it is shown that a nuclear wall is sharply distinguished from the rest of the nucleus in a spectral image, in contrast to a simple color image in which the wall is perceived as part of the nucleus (see ibidem Figure 4.1Od on page 115).

In E. Schroeck et al., Multicolor spectral karyotyping of human chromosomes, Science, 273 (1996) 494-497, it is shown how a spectral imaging device as disclosed in U.S. Pat. No. 5,539,517 can be used in combination with fluorescence/H-stain hybridization (FISH) techniques to analyze combinatorially stained chromosomes (human and animal) so that karyotyping and chromosomal aberrations can be detected and characterized.

According to this technique, which is further described in the Background section above and in the Examples section below, each chromosome is hybridized with complementary DNA material containing a different combination of fluorescent dyes from a larger set of dyes, such that each chromosome of a metaphase spread emits a different fluorescence spectrum substantially uniformly across its surface. Typically, each chromosome is labeled with a different combination of up to four dyes (e.g., one, two, three, or four dyes) selected from a set of five dyes, resulting in 24 different fluorescence spectra, one for each chromosome. This is done with human (24 different spectra or 24 dye combinations required), mouse and monkey chromosomes (the number of which differs from 24), and with healthy and diseased (e.g., cancerous) cells. The detection and identification of translocations using this method is immediate and reliable, since the different spectrum of a

translocation clearly stands out from the surrounding chromosome, whereas the information provided by the G-banding technique, which is widely used today for chromosome classification, is far less clear for this purpose.

The terms "spread" and "karyotype" are used interchangeably in this document to refer to the visualization of chromosomes. In most cases, the term "spread" is used to refer to chromosomes before they have been rearranged into chromosome pairs and the term "karyotype" is used to refer to chromosomes after such rearrangement. However, this is not intended to limit the identical meaning of these terms to refer to a set of chromosomes, each of which is identified.

However, as further described in the Background section above, the classification method used there is based on a universal reference spectrum derived from a normal karyotype, which is subsequently used to classify chromosomes in other karyotypes. As further mentioned above, this approach is not always successful because in some cases new karyotypes cannot be classified appropriately or can only be partially classified.

Since the best classification results are obtained using internal reference spectra, the present invention is directed to providing a method for calculating reference spectra whose calculation is independent of preclassification using standard chromosome classification techniques (e.g. G-banding or R-banding), as is the case in the prior art.

Accordingly, a method according to the invention is provided for finding L internal reference vectors for classifying L chromosomes or chromosome regions of a cell. The term "chromosomes or chromosome regions" as used herein and in the claims refers to chromosomes or chromosome regions derived from L types or kinds of chromosomes present in a cell, or any fraction or subset of these chromosomes. However, each of these types is usually present in two copies. In some pathological cases (e.g. cancer), the result of chromosomal aberrations is to have type(s) of one (or more) chromosome(s) present in more or less than two copies, with regions or entire chromosomes, for example, duplicated or missing. If the cell chosen is from a human and if all chromosomes are analyzed, L would in most cases be equal to 24 in males and equal to 23 in females.

MF

However, the invention also relates to analyzing a subset of the chromosomes of a cell, therefore "L" can be any integer smaller than the number of chromosomes characterizing the cell analyzed. According to the method, furthermore, the L chromosomes 5 or chromosome regions are stained with K, for example 5, different fluorophores or combinations thereof, wherein K basic chromosomes or chromosome regions of the L chromosomes or chromosome regions are each stained with only one of the K different fluorophores, while the other L-K, for example 19, for a male human having L chromosomes or chromosome regions are each stained with a different combination of K different fluorophores. Chromosome staining is carried out using a combined labeling approach, which can be, for example, combined labeling or combined hybridization, as further described herein. Preferably, combined hybridization is chosen.

The steps of the method are as follows. A multi-band collector is used to collect a first vector for each pixel of each of the L chromosomes or chromosome regions. Then the pixels belonging to each of the K basic chromosomes or chromosome regions are identified and the pixels are defined as basic pixels so that K basic classes of basic pixels are obtained. At least one basic pixel from each of the K basic classes is used to obtain the K basic vectors, the K basic vectors being K internal reference vectors. The K basic vectors are then used to identify the pixels belonging to the other L-K chromosomes or chromosome regions. Finally, the pixels belonging to the other L-K chromosomes or chromosome regions are used to calculate the other L-K internal reference vectors, thereby finding all of the L [K+ (L-K)] internal reference vectors.

In most applications, K may be equal to 5. However, other combinations of fluorophores are also applicable, depending on their type, spectral characteristics and the number L of chromosomes or chromosome regions analyzed. According to a preferred embodiment of the invention, the method described above is used to classify the L chromosomes or chromosome regions of a cell. For this purpose, the chromosomes are stained with K different fluorophores or combinations thereof as described above. For chromosome classification, L internal reference vectors are found as described above and are used to classify each pixel into one of the L classification classes.

In a preferred embodiment of the invention, each of the classes of classified

pixels are assigned a distinguishable artificial color. Preferably, the classification of each pixel into one of the L classification classes using the L reference vectors is effected by a linear decomposition algorithm used to find binary vectors for each of the pixels belonging to the L chromosomes or chromosome regions, all as described further hereinafter in the Examples section. The method for chromosome classification can be used to detect chromosomal aberrations.

According to further features of preferred embodiments of the invention described below, there is provided a method for classifying L chromosomes or chromosome regions of a cell, the L chromosomes or chromosome regions being stained with K different fluorophores or combinations thereof, wherein each of the K basic chromosomes or chromosome regions of the L chromosomes or chromosome regions is stained with only one of the K different fluorophores, while each of the other L-K of the L chromosomes or chromosome regions is stained with a different combination of the K different fluorophores, the method comprising the steps of: (a) using a multi-band collection device to detect a first vector for each pixel of each of the L chromosomes or chromosome regions, (b) identifying pixels belonging to each of the K basic chromosomes or chromosome regions and defining the pixels as basis pixels so as to obtain K basis classes of the basis pixels, (c) using at least one basis pixel from each of the K basis classes to obtain K basis vectors, where the K basis vectors are K internal reference vectors, and (d) using the K basis vectors to identify pixels belonging to other L-K chromosomes or chromosome regions.

According to still further features in the described preferred embodiments, the method further comprises the step of assigning a distinguishable artificial color to each of the classes of classified pixels.

The classification of chromosomes according to the invention can be effected by a classification based on base pixels, in which the step of classifying the identification of internal reference vectors is skipped. According to this embodiment, a method is provided for classifying L chromosomes or chromosome regions of a cell stained with K different fluorophores or combinations thereof as described above, the method comprising the steps of (a) using a multi-band collection device to detect a first vector for each

27 * ·*

pixels from each of the L chromosomes or chromosome regions, (b) identifying pixels belonging to each of the K basic chromosomes or chromosome regions and defining the pixels as basic pixels, thereby obtaining K basic classes of the basic pixels, (c) using at least one basic pixel from each of the K basic classes to obtain K basic vectors, the K basic vectors being K internal reference vectors, and (d) using the K basic vectors to identify pixels belonging to other L-K chromosomes or chromosome regions.

As described further below in Example 3, in some cases the base chromosomes or chromosome regions, and therefore the base pixels and reference vectors, are associated with more than a single fluorophore. According to the present invention, therefore, there is further provided a method for classifying chromosomes or chromosome regions of a cell, wherein the chromosomes or chromosome regions are stained with different fluorophores or combinations thereof such that each of the chromosomes or chromosome regions is stained with a different fluorophore, the method comprising the steps of: (a) using a multi-band collection device to detect a first vector for each pixel of each of the chromosomes or chromosome regions, (b) identifying pixels belonging to the base chromosomes or chromosome regions and defining the pixels as base pixels so as to obtain base classes of the base pixels, (c) using at least one base pixel of each of the base classes to obtain base vectors, the base vectors being internal base reference vectors, and (d) using the internal Base reference vectors for identifying pixels that belong to other chromosomes or chromosome regions.

Preferably, the multi-band collection device is either a spectral imaging device combined with a filter cube and a device including a plurality of additional filter cubes and additional emission and excitation filters as further described hereinabove.

Preferably, each of the above-mentioned first vectors further comprises N elements, where N is an integer selected from the range 3-150, which is preferably normalized as further described in detail in the Examples section below. Preferably, each of the first vectors further represents a spectrum.

In a preferred embodiment, the identification of the pixels belonging to each of the K basic chromosomes or chromosome regions is effected by a method

such as identifying the pixels belonging to each of the K basic chromosomes or chromosome regions using a conventional banding pattern of the L chromosomes or chromosome regions. Or such as identifying the pixels belonging to each of the K basic chromosomes or chromosome regions using an RGB algorithm. Alternatively, such as identifying the pixels belonging to each of the K basic chromosomes or chromosome regions using K external basis vectors from a library. In any case, preferably each of the K basis vectors is an average of a plurality of basis pixels belonging to one of the basis classes.

In a further preferred embodiment of the invention, the identification of the pixels belonging to each of the K basic chromosomes or chromosome regions using the external basis vectors from the library is accomplished by (a) using a linear decomposition algorithm to define a decomposition K vector for each pixel of the L chromosomes or chromosome regions, (b) using a high cut value to transform each of the decomposition K vectors into a transformed binary K vector, and (c) comparing each of the transformed binary K vectors to K defined binary K vectors defining each of the K different fluorophores, thereby identifying the pixels belonging to each of the K basic chromosomes or chromosome regions.

In yet another preferred embodiment of the invention, the identification of the pixels belonging to the other L-K chromosomes or chromosome regions and the calculation of the other L-K internal reference vectors is effected by (a) using a linear decomposition algorithm to define a decomposition K vector for each pixel of the L chromosomes or chromosome regions, using a low cut value or range to transform each of the decomposition K vectors into a transformed binary K vector, and (c) comparing each of the transformed binary K vectors to L-K defined binary K vectors, defining each of the combinations of the K different fluorophores to thereby identify the pixels belonging to each of the L-K chromosomes or chromosome regions.

According to the invention, colored chromosome karyotypes are further provided. Such a colored karyotype is a display that includes an image of all the chromosomes or chromosome regions of a cell. Each of the chromosomes or chromosome regions is colored with a different fluorophore or a combination of fluorophores. The image shows the chromosomes or chromosome regions in different distinguishable colors, with each of the chromosomes or chromosome regions

areas are assigned one of the different, distinguishable colors. The term "display" as used herein refers to any visual representation such as, but not limited to, a photograph, print, screen display or monitor display.
5

Another color karyotype according to the invention is a display that includes a composite image of at least some chromosomes or chromosome regions of a cell. Each of the chromosomes or chromosome regions is stained with a different fluorophore or combination of fluorophores and banded using a chromosome banding technique to obtain a characteristic banding pattern for each of the chromosomes or chromosome regions. The image shows the chromosomes or chromosome regions in distinguishable colors, wherein each of the chromosomes or chromosome regions is assigned one of the distinguishable colors, and wherein each of the distinguishable colors acquires an intensity pattern in accordance with the banding patterns of each of the chromosomes or region of chromosomes such that both the distinguishable colors and the banding patterns of each of the chromosomes or chromosome regions are superimposed.

Yet another color karyotype according to the invention is a display that includes a composite image of at least some chromosomes or chromosome regions of a cell. Each of the chromosomes or chromosome regions is stained with a different fluorophore or combination of fluorophores and banded with the chromosome banding technique to achieve a distinctive banding pattern and shape for each of the chromosomes or chromosome regions. The image shows the chromosomes or chromosome regions in distinguishable colors, wherein each of the chromosomes or chromosome regions has been assigned one of the distinguishable colors and wherein further each of the chromosomes or chromosome regions has acquired the distinctive shape such that both the distinguishable colors and the shapes of each of the chromosomes or chromosome regions are superimposed.

According to the invention there is further provided a display comprising a composite image of a chromosome or a region of a chromosome. The chromosome or region of the chromosome is shown in a color, the color including a banding pattern of alternating lighter and darker bands, both the color and the banding pattern providing indications of the chromosome region of chromosome identification (i.e., chromosome number and/or region).

According to the invention there is further provided a display which includes an image of a chromosome or a region of a chromosome. The chromosome or the region of a chromosome is shown in a color which provides clues for identifying the chromosome or the region of a chromosome. The chromosome or the region of a chromosome has a shape which is similar to its shape (outline) after banding using conventional banding techniques (e.g. G-banding, R-banding (DAPI), etc.).

According to the invention there is further provided a display comprising a composite image of a chromosome or a region of a chromosome, the composite image comprising the superposition of a staining image and a banded image of the chromosome or the region of a chromosome.

The method of the present invention has significant advantages over prior art methods due to the introduction of chromosome classification using internal reference vectors for pixel classification, where the internal reference vectors usually provide better performance. Another advantage of the present invention is that the method can be easily automated. Using this method, therefore, semi- or unskilled cell geneticists are enabled to obtain clear and informative karyotypes, which may include a superposition of spectral and spatial information.

Reference is now made to the examples which, together with the foregoing description, illustrate the invention in a non-limiting manner.

Example 1 Chromosome preparation for detection

The advent of multicolor FISH has expanded the applications of molecular cell genetics in basic research and genetic diagnosis. All existing multicolor FISH techniques require the use of fluorescent probes whose emission spectra can be separated with optical filters. [Ried et al., Simultaneous visualization of seven different DNA probes by in situ hybridization using combinatorial fluorescence and digital imaging spectroscopy, Proc. Natl. Acad. Sci. USA 89 (1992) 1388-1392; and Ried, Fluorescence/fluorescence hybridization in genetic diagnostics, Faculty of Theoretical Medicine, Ruprecht-Karls-University Heidelberg (Jan. 1994), both incorporated by reference.

as if they were fully disclosed herein]. This requirement limits the number of dyes that can be distinguished in a given sample. A new approach to FISH, using the SpectraCube™ system to detect and analyze multiple spectrally overlapping labeled probes (single and combinatorial), to classify chromosomes and therefore to detect chromosomal aberrations, has recently been introduced [E. Schroeck et al., Multicolor spectral karyotyping of human chromosomes, Science, 273 (1996) 494-497].

According to this new approach, bio-spectral imaging, which is a combination of Fourier spectroscopy, CCD imaging and optical microscopy, was used to acquire accurate spectral data simultaneously at all points of a biological sample to visualize a hybridization-based multicolor appearance of all (i.e., 24) types of human chromosomes and to generate a color map of the human karyotype (a color/spectral karyotype).
15

Below is a description of the dyes and their combinations which are presently preferred, which dyes and combinations have been used to carry out the chromosome classification method of the invention.

Thus, 24 chromosome stains (1 to 22, X and Y, Panel 1), each labeled with a different combination of four or fewer different fluorophores selected from a set of five fluorophores according to the combinatorial hybridization approach (a to e, Panel 1) (see Panel 1 for the different fluorophores and their spectral characteristics and Panel 2 for the assignment of the fluorophores listed in Panel 1 to yield the 24 chromosome stains) were simultaneously hybridized to human mitotic chromosome spreads from a few non-related male white blood cell cells prepared essentially as described in Ried et al. [Ried et al., Simultaneous visualization of seven different DNA probes by in situ hybridization using combinatorial fluorescence and digital imaging spectroscopy, Proc. Natl. Acad. Sci. USA 89 (1992) 1388-1392] were prepared for hybridization.

Hybridized chromosomes were viewed through an inverted fluorescence microscope connected to the SpectraCube™ system and were analyzed.
35

It is clear to one of ordinary skill in the art that other fluorophores, other combinations of fluorophores and different labeling approaches (e.g.

combinatorial labeling) can be used in a similar manner. Consequently, the information presented in Tables 1 and 2 below is merely illustrative in nature and is not intended to limit the scope of the invention to the fluorophores and/or labeling techniques listed.

TABLE:

Fluorophore symbol Suggestion (am) Human chromosome 1 1 b,c,d Human chromosome 6 6 b, c d, e Human chromosome 11 11 a, c, d Spectrum Orange a 540-580 Human chromosome 2 2 e Human chromosome 7 7 bc Human chromosome 12 12 be Texas Red b 540-570 Human chromosome 3 3 a, b, c, d, Human chromosome 8 8th d Human chromosome 13 13 a, e Cy5′ C 630-670 Human chromosome 4 4 CD Human chromosome 9 9 a, d, e Human chromosome 14 14 b Spectrum Green d 475-495 Human chromosome 5 5 a, b, d, e Human chromosome 10 10 c, e Human chromosome 15 15 a, b, c Cy5.5™ I e 630-670 from Amersham TABLE 2: chromosome Chromosome
Staining agents Fluorophores

TABLE 2 (continued):

chromosome Chromosome
Staining agents Fluorophores Human chromosome 16 16 b, d Human chromosome 21 21 d,e Human chromosome 17 17 C Human chromosome 22 22 b, c, e Human chromosome 18 18 a,b,d Human chromosome X X a, e Human chromosome 19 19 a, c Human chromosome Y Y c,d,e Human chromosome 20 20 a

EXAMPLE 2 The capture device

Figure 1 is a block diagram illustrating the major components of a prior art imaging spectrometer disclosed in U.S. Patent Application No. 08/392,019 by Cabib et al., filed February 21, 1995, now U.S. Patent No. 5,539,517, issued July 23, 1996, which is incorporated by reference as if fully set forth herein. This imaging spectrometer is very suitably constructed for carrying out the method of the invention because it has high spectral (ca. 4-14 nm depending on wavelength) and spatial (ca. 30/M μηλ where M is the effective microscope or fore optics magnification) resolutions.

Thus, the imaging spectrometer of Figure 1 includes: an optical collection system, generally designated 20; a one-dimensional scanner, as indicated by block 22; an optical path difference (OPD) generator or interferometer, as indicated by block 24; a one-dimensional or two-dimensional detector array, as indicated in block 26; and a signal processor and display, as indicated by block 28.

A critical element in system 20 is the OPD generator or interferometer 24, which outputs modulated light corresponding to a predetermined set of linear combinations of the spectral intensity of the light emitted by each pixel of the scene being analyzed. The output of the interferometer is focused on the detector array 26.

Consequently, all required optical phase differences are scanned simultaneously for all pixels of the field of view to obtain all the information required to reconstruct the spectrum. The spectra from all pixels in the scene are thus collected simultaneously with the image information, thereby enabling analysis of the image in a real-time manner.

The device according to U.S. Pat. No. 5,539,517 can be used in a wide variety of configurations. In particular, the interferometer used can be combined with other mirrors as described in the relevant figures of U.S. Pat. No. 5,539,517.

Thus, according to U.S. Pat. No. 5,539,517, alternative types of interferometers may be used. These include (1) a movable type of interferometer in which the OPD is varied to modulate the light, namely a Fabry-Perot scanned layer thickness interferometer, (2) a Michelson type interferometer which includes a beam splitter which receives the beam from an optical collection system and a scanner and splits the beam into two paths: (3) a Sagnac interferometer, optionally combined with other optical means, in which interferometer the OPD varies with the angle of incidence of the incoming radiation, such as the four-mirror combined beam splitter interferometer as further described in the cited U.S. patent (see Figure 14 therein).

Figure 2 illustrates an imaging spectrometer constructed in accordance with U.S. Pat. No. 5,539,517 using an interferometer in which the OPD varies with the angle of incidence of the incoming radiation. A beam entering the interferometer at a small angle to the optical axis experiences an OPD that varies substantially linearly with that angle.

In the interferometer of Figure 2, all radiation from source 30 in all pixels, after collimation by an optical collection system 31, is scanned by a mechanical scanner 32. The light then passes through a beam splitter 33, to a first reflector 34 and then to a second reflector 35, which reflects the light back through the beam splitter 33, and then through a focusing

Lens 36 to an array of detectors 37 (for example a CCD). This beam interferes with the beam reflected by 33, then by second reflector 35, and finally by first reflector 34.

At the end of a scan, every pixel of the scene has been covered by all OPDs and therefore the spectrum of each pixel can be reconstructed by Fourier transform. A ray parallel to the optical axis is compensated and a ray at an angle (θ) with respect to the optical axis experiences an OPD which is a function of the layer thickness of the beam splitter 33, its refractive index and the angle θ. The OPD is proportional to θ for small angles. By applying a suitable inversion and by careful bookkeeping, the spectrum of each pixel is calculated.

In the configuration of Figure 2, the beam incident on the beam splitter at an angle β (β = 45° in Figure 2) passes through the interferometer with an OPD = O, while a beam incident at a general angle β - θ experiences an OPD given by Equation 2:

(2) OPD(ß,0 ,*,«) = f[(« ² -sin ² (ß+e)) ⁰ · ⁵ - (« ² -sin ² (ß-0))°· ⁵ + 2sinßsine]

where θ is the angular distance of a beam from the optical axis or interferometer rotation angle with respect to the central position; t is the layer thickness of the beam splitter and η is the refractive index of the beam splitter.

From equation 2 it follows that by sampling both positive and negative angles with respect to the central position, one obtains a double-sided interferogram for each pixel, which helps to eliminate phase errors and gives more accurate results in the Fourier transform calculation. The sampling amplitude determines the maximum OPD achieved, which is related to the spectral resolution of the acquisition. The size of the angular steps determines the OPD step, which in turn is determined by the shortest wavelength to which the system is sensitive. In fact, according to the sampling theorem [see Chamberlain, The principles of interferometrie spectroscopy, John Wiley and Sons, (1979) 53-55], this OPD step must be less than half the shortest wavelength to which the system is sensitive.

Another parameter that should be considered is the limited size of a detector element in the matrix. Due to the focusing optical components, the element subtends a limited OPD in the interferometer, which reduces the effect of the

convolving the interferogram with a rectangular function. This leads to a reduction in the system sensitivity at short wavelengths, which drops to zero at wavelengths equal to or below the OPD limited by the element. For this reason, it must be ensured that the modulation transfer function (MTF) condition is met, i.e. that the OPD limited by a detector element in the interferometer must be smaller than the shortest wavelength at which the instrument is sensitive.

Thus, imaging spectrometers constructed in accordance with the invention disclosed in U.S. Pat. No. 5,539,517 not only measure the intensity of light emanating from each pixel in the field of view, but also capture the spectrum of each pixel in a predetermined wavelength range. They also make better use of all of the radiation emitted by each pixel in the field of view at a given time and therefore, as explained above, enable a significant decrease in frame time and/or a significant increase in the sensitivity of the spectrometer. Such imaging spectrometers may incorporate various types of interferometers and optical collecting and focusing systems and therefore may be used in a wide variety of applications including medical diagnostics and therapy and biological research applications, as well as remote sensing for geological and agricultural surveys and the like. As mentioned above, a spectral imaging device in accordance with the invention disclosed in U.S. Pat. No. 5,539,517 disclosed invention of Applied Spectral Imaging Ltd, Industrial Park, Migdal Haemek, Israel and is referred to herein as SpectraCube™.

The SpectraCube™ system optically connected to a microscope is used to carry out the chromosome classification method of the invention. The SpectraCube™ system has the following or better characteristics, which are listed below in Table 3.

TABLE 3

Parameters Performance

Spatial resolution: 30/M μηι (M= effective microscope or fore-optics magnification

(effective microscope or fore optics magnification))

Field of view: 15/M millimeters

Sensitivity: 20 milliLux (at 100 msec integration time, decreases with longer

Integration times linear with 4&tgr; to)

Spectral range: 400-1000 nm

Spectral resolution: 4 nm at 400 nm (16 ran at 800 nm)
Capture time.· 5-50 see, typically 25 see

FFT processing time: 20-180 see, typically 60 see

The prior art SpectraCube™ system is used here to acquire an N vector of spectral data (a spectrum in this case) from each metaphase pixel of in situ stained chromosomes as described above. However, as stated above, any spectral imaging device, i.e. an instrument that measures and stores for later retrieval and analysis the spectrum of light emitted from any point of an object placed in its field of view, including filter-based spectral imaging devices (e.g. acousto-optical tunable filters (AOTF) or liquid crystal tunable filters (LCTF) and dispersive element (e.g. grating or prism) or other spectral data or multi-band collection devices (e.g. a device according to the disclosure of Speicher RM, Ballard SG and Ward CD, Karyotyping human chromosomes by combinatorial multi-flour FISH, Nature genetics (1996) 12:368-375), may be used to acquire the required spectral data. It is therefore not intended to limit the scope of the present invention to the uses of any specific type of spectral data collection device or any specific type of spectral imaging devices.

EXAMPLE 3

Identification of internal reference vectors for chromosome classification 20

1. Definitions and preparatory treatment of spectral data

A spectrum can be represented as a vector of size N, where each component or element represents the intensity of light at a chosen wavelength. The number of components (N) that make up a spectrum depends on the spectral range (e.g. visible light range, .400-750 nm) and on the spectral resolution of the acquisition device. For example, a spectrum is represented by a 151 vector with a 2 nm spectral resolution homogeneously distributed over a spectral range of, say, 400 nm to 700 nm.

Therefore, when the term "spectrum" (or spectra in the plural) is used in the specification and the claims below, it refers to the N-vector representing that spectrum. In some cases, the equivalent term "spectral vector" is used.

Another term, a "normalized spectrum" (or spectra in the plural), is also used here. A normalized spectrum is calculated from the original spectrum by dividing the intensity at each value by the integrated intensity of the entire spectrum. This procedure ensures that a normalized spectrum has a total integrated intensity equal to exactly 1. This makes it possible to compare the spectral shape of two different spectra without considering their intensity. In many cases, the shape of the spectrum, rather than its intensity, is a better classifier or classification criterion. Normalized spectra are preferably used to exclude variable intensity effects, irregularities associated with the sample under investigation, the labeling procedure or the optical detection elements.

Furthermore, all mathematical procedures described herein are preferably performed using a background subtraction procedure. Background subtraction is a mathematical procedure in which a spectrum of a pixel or an average spectrum of some or all of the pixels located in the background region is subtracted from the spectrum of each pixel. As a result, background fluorescence signals common to all pixels are subtracted, thereby increasing the classification specificity of the signals. Furthermore, the background pixels are now characterized by spectra that are all close to zero intensity at all wavelengths, and the pixels are preferably eliminated for any further consideration. Therefore, only pixels associated with the chromosomes are included in further consideration.

2. Identification of external basis spectral vectors

The term "basic" as used herein in conjunction with other terms such as spectra, pixel, chromosome, cluster, class, etc., refers in most cases to a state involving a single fluorophore, as opposed to a combination of fluorophores. Since five fluorophores are used in the examples given herein, there are five classes of basics, each class being associated with a different fluorophore. However, it will be appreciated that in some applications a basic feature (e.g., chromosome, pixel, class, etc.) refers to a feature involving more than one fluorophore. In the present examples, the basic features refer to those associated with a single fluorophore, although this approach is not intended to limit the scope of the present invention to such cases. Thus, a

and linear decomposition, further described and exemplified below, is performed using basis pixels forming basis classes belonging to basis chromosomes stained by more than one fluorophore.

As shown above in Panels 1 and 2 of Example 1, the five basic chromosomes in the given example are chromosomes numbered 2, 8, 14, 17 and 20, which are labeled with Cy5.5 (e) or Spectrum Green (d) or Texas Red (b) or Cy5™ (c) or Spectrum Orange (a), respectively.

Therefore, basic pixels are those that display genetic material labeled with a single fluorophore, in this case genetic material derived from the chromosomes specified above. Therefore, there are five classes of basic pixels.

Basic spectral vectors, as referred to herein, are spectral vectors of pixels in which the genetic material is labelled (e.g., stained) by a single fluorophore. In the present example, there are therefore five classes of basic spectral vectors. In the present example, these are spectral vectors of pixels representing genetic material derived from the five chromosomes specified above.
20

External basis spectral vectors are basis spectral vectors derived from a preclassified reference color (spectral) karyotype, usually a normal male.

The external basis spectral vectors are used according to the method of the invention to find internal basis spectral vectors of a currently classified karyotype, which are subsequently used for an improved chromosome classification according to the method of the invention. According to another embodiment of the invention, as further described below, internal basis reference vectors

directly derived from the karyotype examined.
30

According to the present invention, there are few approaches to identifying the external basis spectral vectors.

The first approach, described herein with reference to Figures 3a-c, which show a normal color karyotype of a human male, was performed using the technique described in U.S. Patent Application No. 08/635,820, filed April 22, 1996, and in the journal Science [E. Schroeck et al., Multicolor spectral karyotyping of human chromosomes,

Science, 273 (1996) 494-497], both documents being incorporated by reference as if fully disclosed herein. The exact types and combinations of fluorophores used are listed in Tables 1 and 2 under Example 1 above.

The karyotype of Figures 3a-c was obtained as follows. First, the chromosomes were hybridized with chromosome stains described above in Panels 1 and 2 and labeled with DAPI for R-banding. Second, as shown in Figure 3b, chromosomes of the karyotype were identified by the conventional chromosome R-banding technique. Third, the averaged, normalized spectra characterizing pixels assigned to one of the 24 chromosome types (i.e., 1-22, X and Y ) as determined by chromosome R-banding were calculated and the resulting 24 averaged, normalized spectra were used to form a library of reference spectra. Fourth, the reference spectra were used to classify the pixels of that karyotype, the result of the classification being shown in Figure 3a. As shown in Figure 3c, the chromosomes are then arranged in sequence as is known in the art.

For classification, the normalized spectrum of each pixel in the image was compared to each of the 24 different reference spectra of the reference library and the similarity of it (e.g. minimum difference, as derived e.g. by a minimum square error method) to one of the reference spectra was determined. Each pixel was then assigned an artificial predetermined distinguishing color according to its classification. The result is a color karyotype in which each of the chromosomes is colored with a different, artificial color (24 different colors in total).

As mentioned, the classification uses different spectra as reference spectra and colors each pixel of a displayed image with a different, predetermined, artificial color, according to its classification as most similar with respect to one of the different reference spectra.

It is known that various algorithms are important (prominent) in classification, for example consider equations (3-6),

1 &igr;&eegr;&igr;&ugr;

(3)

g^ = —, ■ ——-^ &tgr; (5)

;-·

_l_ UHlX/ I

⁺ /40 )

where I _max is the maximum intensity of the image, G _x * is the brightness with which a pixel (with coordinates &khgr; and y) is imaged on the screen, I _x # (X) is its spectrum, < I _x , (X) > is the average of I^ (X) over the group of 3x3 neighboring pixels, S _max is the peak intensity of I _x , (X), T _max is the peak intensity of < I _x , (X) >, Rx is the reference spectrum with respect to which the similarity map is calculated, R _max is the peak intensity of the reference spectrum R _x and η is the number of wavelengths of the detection.

Alternatively, equation 7 below can be used to describe the degree of similarity

be used:

\

where I _x # (X) is the normalized spectrum of pixel x, y, I _R (X) is the normalized reference spectrum and N is the number of wavelengths. In this equation, S decreases if the similarity between the spectra increases and vice versa. In the limit, if both spectra are identical, S becomes zero.

By performing this calculation with k different reference spectra, k different values of S are obtained for each pixel, i.e. S ₁ , S ₂ , .., S _k . The least of these indicates which of the reference spectra is the most similar to the spectra of the pixel. Nevertheless, one of ordinary skill in the art will appreciate that other equations reflecting the similarity are also known and applicable. The classification can thus also be performed using a simple minimum square error distance algorithm.

(minimal square error distance algorithm) or a complicated principal component analysis.

Using this approach, the external basis spectral vectors can be identified, five in the present example. Note that any of the external basis spectral vectors can be of any specific basis pixel, showing the basis spectral vector or, preferably, the average of the spectral vectors of two or more basis pixels of a single basis pixel class.

Nevertheless, it should be noted that such identification is based on first identifying the chromosomes using a conventional banding technique (DAPI for R-banding in this case), which makes this approach less advantageous.

However, if a universal experimental protocol is maintained, the identification of the external reference spectral vectors must be performed once for all further chromosome classifications, as described herein with reference to some embodiments of the method of the invention.

The second approach is described herein with reference to Figures 4a-b, which depict a normal color karyotype of the same human male derived using another prior art method for chromosome classification described in U.S. Patent Application Serial No. 08/635,820 filed April 22, 1996, and in the journal Science [E. Schroeck et al., Multicolor spectral karyotyping of human chromosomes, Science, 273 (1996) 494-497]. The exact types and combinations of fluorophores used are listed above in Tables 1 and 2 under Example 1.

According to the second approach, an RGB algorithm was used that integrates the optical signal over the spectral range (e.g., from 400 nm to 760 nm) of the CCD array to obtain an RGB image of the chromosomes in which each pixel is assigned a combination of red, green, and blue intensities according to three weighting functions, {w _r (λ), w _e (λ), w^ (λ)}, which correspond to the tristimulus response functions for red (R), green (G), blue (B).

Figure 5 shows an example of the effectiveness of this simple algorithm. Consider that {w _r (λ), Wg (λ), Mh ₀ (λ)} are chosen in such a way that they are Gaussian or other functions distributed "inside" a spectrum under consideration, whereby the pseudo-color image thus obtained, which is displayed in this case, contains only data in the

weighting functions, thereby enabling the spectral differences in these three regions to be more clearly detected.

Figure 6 shows the effectiveness of the RGB algorithm for chromosome classification. In this case, w _r (λ), w _s (X), Wb (λ) are simple square weighting functions, where for w _T (red) λγ = 640 nm and %2 = 750 nm, for vv _g (green) λγ = 555 nm and %2 = 640 nm, and for Wb (blue) λγ = 450 nm and Xz - 555 nm. The simple weighting functions w _r , w _g , w\ are integrated to produce an RGB map of the chromosomes as shown on the right.

Given predetermined experimental conditions and predefined weighting functions for the RGB colors, the basic chromosomes and the basic pixels can be distinguished, as can most or all other chromosomes. In particular, the experimental conditions, i.e. the types and combinations of the fluorophores and the RGB weighting functions are chosen such that the basic pixels are assigned unique colors (i.e. basic colors) that clearly distinguish them from each other and from all other (19 in this case) "non-basic" chromosomes/pixels. Table 4 summarizes the association of fluorophores, basic chromosomes and basic colors chosen for the present example.

20

TABLE 4

symbol Fluorophore Basic chromosome Base color a Spectrum Orange 20 Green b Texas Red 14 Yellowish-green C Cy5&trade;' 17 Red d Spectrum Green 8th Blue e Cy5.5″ 2 Reddish brown

byAmersham

Figure 7 shows the five external base spectra derived using the example experimental procedure, with the fluorophores and base chromosomes indicated.

The use of the RGB approach to identify the external basis spectral vectors has an advantage over the classification approach described above, since no prior identification of the chromosomes by conventional banding techniques is required.

(e.g. R-banding) is required where it is necessary to generate an external reference library according to the classification approach described above. This is the case because the precise experimental procedure (e.g. the procedure of Tables 1 and 2 above) and the given weighting functions and the preselected three-colour output (e.g. RGB) can be chosen such that each of the pixels having a basis spectral vector will be highly distinguishably coloured. In fact, as described in more detail below, the use of the RGB approach can be used directly to find internal basis spectral vectors which are subsequently used to find all other reference spectra for classification.
10

3. Identification of internal basis spectral vectors

Using the external basis spectral vectors, a process for identifying the internal basis spectral vectors is initiated. According to the present invention, a linear decomposition algorithm is carried out for this purpose.

In the present example, the linear decomposition is performed using the five external basis spectral vectors, defined here as A, B, C, D, and E, to obtain a decomposition-K vector (a decomposition-5 vector in the present example) of the decomposition coefficients for each pixel in the image.

Therefore, during this mathematical process, for each pixel in the image, a set of five decomposition coefficients α, β, γ, δ, ε are calculated. The α-ε coefficients are chosen such that by multiplying each of the A-E external basis spectral vectors into its corresponding coefficients and summing the multiplication result spectral vectors, a spectrum (a spectral vector) results which is substantially identical to the spectrum of that pixel. Therefore, each of the pixels in the image can be represented as a decomposition K-vector, in the present example as a decomposition 5-vector composed of the α-ε decomposition coefficients. Of the decomposition 5-vectors thus calculated, the internal basis pixels are by definition represented by basis decomposition 5-vectors. These vectors are defined here as external basis decomposition K vectors because they are defined using K external basis spectral vectors.

This can be described mathematically for any number of decomposition coefficients required. If Q are the coefficient vectors, where J= 1 ... k (k equals 5 in the present example), then:

&phgr;&phgr;&phgr;&phgr; φ·φ· · · ·

&bull; * · ft · · * &igr;

&bull; * «ft · · &diams;&igr;

(8th)

where / (Xj) is the spectrum of a particular pixel, and Ij (XJ are the vectors.

Due to noise, it may be practically impossible to obtain a set of coefficients that exactly matches the above equations and therefore a small residual must be added. Due to the quality of the results and the high signal-to-noise ratio achievable with the SpectraCube™ system, this residual is relatively small.

According to the method of the invention, the linear decomposition coefficients used to identify internal basis pixels comprise α-ε, which, by definition, comprise internal basis spectral vectors defined here as A ¹ , B', C, D', and E ¹ , respectively.

Referring now to Figures 8a-e, drawings are shown which represent the number of pixels having a value (0-1.0) of the decomposition coefficients α, β, γ, 5, and ε, respectively.

As shown in Figures 8a-e, the plot for each of the α-ε decomposition coefficients is characterized by two peaks separated by a valley therebetween. The first peak represents pixels having low decomposition coefficient values (e.g., 0-0.3), while the second peak represents pixels having high decomposition coefficient values (e.g., 0.3-1.0). The pixels represented in the second peak of one of the decomposition coefficients (α-ε) have a high probability of containing the fluorophore associated with that coefficient.

As shown in Table 5 below, based on the fluorophore composition (ae), a basis binary K vector, in the present example a basis binary 5 vector, can be defined for each of the internal basis spectral vectors A'-E ¹ . To transform one of the five basis decomposition 5 vectors into a basis binary 5 vector using the defined basis binary 5 vectors and a high cut value, for example about 0.75, the basis pixels having the internal basis spectral vectors A'-E' can be selected.

46
Table 5

a b C d e ^A1 1 O O O O B' O 1 O O O C O O 1 O O ^D1 O O O 1 O ^E1 O O O O 1

Due to the chosen high cut values, the probability of correctly selecting internal basis pixels with internal basis spectral vectors is extremely high.

After finding the basis pixels having internal basis spectral vectors (A-F), internal reference basis spectral vectors can be calculated, for example by separately averaging the basis spectral vectors of the basis pixels belonging to each basis class. Alternatively, the internal basis spectral vectors of selected pixels (5 in this case, one for each basis class) can be used directly as internal reference basis spectral vectors.

In any case, according to the method of the invention, the internal reference basis spectral vectors thus defined are used to calculate the internal decomposition K-vectors (e.g., 5-vectors) for each of the pixels in the image, similarly as described with reference to the external basis spectral vectors hereinbefore.

The following is the mathematical description of the linear decomposition algorithm. In practice, the coefficients of the linear decomposition algorithms are calculated as follows.

Assuming that / (KJ is the spectrum of a particular pixel and that / is an index of the wavelength index i = 1..JV, where N is the total number of wavelengths in the spectrum and assuming that a set of K basis reference spectra is given, Ij (Ki)J = 1...K. We seek a set of K coefficients, Cj such that:

(9)

where ε ß-0 is the smallest possible error.

It is important to note that equation 9 should be valid for any of the N wavelengths λ,. This equation can also be written as:

or by summing over all N wavelengths.

Since the square of the remainder is always positive, one must now look for the set Cj , so JV

that JV(;L·) is minimized. This solution can be found by taking the partial derivative Ml

the right-hand side of equation 11 relative to each of the components Cj and set equal to zero. This results in a set of K equations having K unknowns, Cj. These equations can be written in a matrix format. The derivative y-. _n of the right-hand side of equation 11 is:

(12)

Equation 12 is set equal to zero and after a different transformation we get:

/V

/=1 Iy-I

/= I

Equation 13 can now be written in a matrix format (14):

·

&Sgr;'fa, .=1

&sgr;'.(&lgr;,

Xz ₂ (X ₇ )Z ₁ (X,) ...

J=I .V

&Sgr;&Lgr; (λ,·)· ζ, (λ,.)

/=1

N Î=&Igr;

&Sgr;&zgr;,(&khgr;.)·

Equation 14 can be written as Ä · C = F, where ^4 is a symmetric matrix and all its elements can be calculated directly. By finding the reciprocal matrix Ä ^A and multiplying from left to right, we get:

C = AV (15)
The components Cj are therefore found.

4. Identification of all internal Reference spectral vectors

As mentioned above, the K, say 5, internal base spectral vectors can serve as reference spectral vectors for classifying pixels showing genetic material labeled only by a single fluorophore a or b or c or d or e. Nevertheless, for a complete classification, additional reference spectral vectors are required representing the combinations of the a-e fluorophores used for hybridization (see Table 2).

According to the invention, the K internal basis spectral vectors are therefore used to identify pixels that have one of the additionally required (19 in this case) internal reference spectral vectors.

For this purpose, as shown in Table 6, each required

Reference spectral vector defined as a binary-5 vector, according to the associated

Fluorophores. As a reminder, the basic binary vectors have already been defined in Table 4 above.

TABLE 6

Binary vector

Chr.
No. Chr.
Coloring
ng agent Fluorophores a b C d e 1 1 b,c,d 0 1 1 1 0 2 2 e 0 0 0 0 1 3 3 a, b, c, d, 1 0 1 1 1 4 4 CD 0 0 1 1 0 5 5 a, b, d, e 1 1 0 1 1 6 6 b, c, d, e 0 1 1 1 1 7 7 bc 0 1 1 0 0 8th 8th d 0 0 0 1 0 9 9 a, d, e 1 0 0 1 1 10 10 a, b 0 0 1 0 1 11 11 a, c, d 1 0 1 1 0 12 12 be 0 1 0 0 1 13 13 a, e 1 0 0 0 1 14 14 b 0 1 0 0 0 15 15 a, b, c 1 1 1 0 0 16 16 b,d 0 1 0 1 0 17 17 C 0 0 1 0 0 18 18 a,b,d 1 1 0 1 0 19 19 a, c 1 0 1 0 0 20 20 a 1 0 0 0 0 21 21 d,e 0 0 0 1 1 22 22 b,c,e 0 1 1 0 1 X X a, e 1 0 0 0 1 Y Y CD, e 0 0 1 1 1

Then, using the abundance plots of Figures 8a-e and a second, lower threshold (e.g., about 0.3) or preferably a threshold range (e.g., about 0.25-0.35) preferably positioned substantially in the valley between the first and second peaks of the plot, each of the new decomposition K vectors from each of the pixels is transformed into a transformed binary K vector. In the present example, 24 types of transformed binary 5 vectors are expected for male karyotypes and 23 for female karyotypes. In cases where a threshold range is used, some are missing

The binary 5-vectors thus obtained have one or more binary components (0 or 1) for one or more of the decomposition coefficients. This is because values within the threshold range are not classified into a binary term (0 or 1).

According to a predetermined pattern associated with the defined binary vectors of Table 6 above, to represent a color karyotype, pixels having identical transformed binary K-vectors are assigned an identifying artificial color (24 different colors for males, 23 for females).

Alternatively, the spectral vectors of some or all pixels having identical transformed binary K-vectors are averaged to yield 24 (or 23 for females) internal reference spectral vectors, which are subsequently used for classification, for example as described with reference to Equations 3-7 above, or for another suitable classification method (e.g., minimum square error, principal component analysis, etc.), whereby a spectrum of a pixel is classified into a class according to its highest similarity to one of the reference spectral vectors used for classification.

Alternatively, the decomposition K-vectors of some or all pixels having identical transformed binary K-vectors are averaged to yield L (24 for male or 23 for female humans) internal reference vectors, which are subsequently used for classification, for example as described with reference to Equations 3-7 above, or for another suitable classification method (e.g., minimum square error), wherein a decomposition K-vector of a pixel is classified into a class according to its highest similarity to one of the L internal reference spectral vectors used for classification.

Referring now to Figure 9a, there is shown a normal male chromosome spread classified as described hereinabove. Figure 9b shows the same spread in RGB as described with reference to Figures 4a-b above. While Figure 9c shows a side-by-side combined color karyotype of the chromosomes from Figures 9a and 9b, with classified chromosomes shown on the left and RGB chromosomes on the right.

By carefully examining Figures 9a and 9c, one notices that some of the pixels associated with the chromosomes are wrongly classified, some others are not classified but are not visible against the background because they have been assigned a black color. The reason for this is that some of the pixels do not belong to one of the classes

can be classified, i.e. into one of the 24 types of binary-5 vectors (theoretically there are 8 additional such vectors), while other pixels are misclassified into the wrong class. In some cases, some of the 24 types of binary-5 vectors were associated with a very small number of pixels, leading to the conclusion that this is an artifact (not shown).

The following further illustrates this problem. A pixel - binary-5 vector association table has been computed. Each of the pixels (1,1 to x,y) is associated with one of the 24 binary-5 vectors (see Table 5). Then the total number of pixels associated with any of the 1 to 24 binary-5 vectors is counted. Some vectors will then have very few associated pixels. Some pixels will not be associated with any of the 1 to 24 binary-5 vectors. Both cases are artifacts due to the arbitrary selection of the threshold range as described above.

4. Spectral-spatial based autosegmentation

Up to this point, all the spectral information has been exploited for classification. However, the spatial information has not been used. It should be remembered that the distribution of pixels contains classification information, since adjacent or nearby pixels have a higher probability of belonging to a pixel cluster showing genetic material derived from a particular chromosome or region thereof and therefore should theoretically be classified in a similar way. Furthermore, each cell is characterized by a given number of chromosomes, each having a distinctive feature, for example in terms of size (e.g. length) and relative position of the centromere.

Typically, the number of clusters in a well-spread (no overlapping chromosomes present) normal human karyotype is 46 (one for each chromosome). These clusters are divided into 24 classes for males and 23 for females. If translocations occur, the total number of clusters increases. In cancerous cells, especially if they are cultured for several generations, the number of clusters caused by translocations can reach a hundred or more.

Recall that each of the pixels in the image is represented by a decomposition K-vector (obtained using either external or preferably internal basis spectral vectors and the second intersection value as described above). Therefore, a clustering procedure can be initiated.

For this purpose, variations among the decomposition K-vectors of adjacent pixels are detected. These variations are used for clustering. For example, if the K-vector variation (e.g. as predetermined, e.g. by minimum square error) between any two adjacent pixels is below a predetermined threshold, they are clustered into one cluster, while if the variation between two adjacent pixels exceeds a predetermined threshold, they are considered to belong to different clusters. The result of the clustering process is the generation of pixel clusters having similar decomposition K-vectors. Similarly, clustering can be accomplished using normalized spectral vectors from the pixels. In fact, clustering is a type of classification.

Using a third threshold, each of the identified clusters can be transformed into a transformed binary K-vector and thereby associated with one of the 24 defined binary K-vectors of Table 5. In other words, if the relative number of pixels of a given cluster associated with one of the 24 defined binary K-vectors is above the third threshold (e.g., above about 80%), then that cluster is associated with that vector. Optimally, a normal karyotype should have 46 clusters associated with 24 (male) or 23 (female) binary K-vector types. Thus, for a male, each of the 1 to 22 binary K-vectors is expected to be associated with two clusters, while the twenty-third and twenty-fourth binary K-vectors are expected to be associated with clusters of genetic material derived from chromosomes X and Y, respectively.

According to the method of the invention, to obtain internal reference vectors for classification, the spectra or the decomposition K-vectors of some or all pixels of clusters associated with each particular binary K-vector are further averaged.

These 23 or 24 averaged vectors are highly suitable for classifying the image currently being analyzed, since they are derived from pixels within the image itself, but this process does not require the use of conventional chromosome banding classification.

However, in some cases, a few clusters are not classified. By definition, an unclassified cluster is missing a group of pixels associated with one of the binary K vectors that is above the threshold of cluster classification. As a result, in

In some cases, some binary-K vectors have no associated clusters and some internal reference vectors for classification are therefore still missing.

The unclassified clusters typically include pixels that are not associated with any of the 24 defined binary K vectors because they contain at least one (usually one) unclassified binary component resulting from the threshold range used. By definition, the unclassified binary component can be either 1 or 0.

In a preferred embodiment of the invention, the two possibilities are examined against all defined binary K vectors that have not yet been associated with a cluster. In most cases, one of the two possibilities fits such that a vector and this pixel is thereby associated with this vector.

According to this analysis procedure, the clustering procedure as described above is repeated. In most cases, following this step, each individual defined binary K vector will be associated with at least one, usually two or more clusters, depending on, for example, sex, level of translocations, presence or absence of overlapping chromosomes in the analyzed chromosome spread.

When following the repeated clustering procedure, the internal reference vectors are recalculated for classification. For this purpose, the spectra or the decomposition K-vectors of some or all pixels of clusters associated with a given binary K-vector are preferably averaged. The reference vectors thus obtained are then used for classification as described above.

However, in rare cases, even after the reclustering step, one or more of the defined binary-5 vectors are still not associated with a cluster. A cluster associated with this vector can be identified using a conventional chromosome banding technique (e.g. DAPI for R-banding).

Now, in most cases, each binary-5 vector is associated with at least one cluster and therefore internal reference vectors can be calculated and classification performed.

As can be seen in Figure 9a, in some cases after classification, some

Chromosomes Clusters of pixels colored in a different color. These clusters can either be the result of a translocation, which is correct, or a classification artifact. Each of these clusters can be examined for this purpose. To this end, the degree of similarity (preferably in percentages) between the spectral vector or the decomposition K-vector of each or some of the pixels within the cluster under study with respect to the relevant reference vectors used is shown. The best five matches (for example, approximately five) are displayed and enable the operator to distinguish between artifacts and translocations.

EXAMPLE 4 Overlay composition of a spectral karyotype image and conventional Banding image

Referring now to Figures 10a-b and 11a-b, as is now well established, two sets of data can be obtained from a labeled metaphase chromosome spread, or in general from other multi-dye labeled materials. One of these is a monochrome image, such as a G-band transmission image or an R-band (DAPI) fluorescence image, the latter shown as a negative image in Figure 10a. The former is a multi-band image, such as a spectral image taken under fluorescence after labeling with chromosome-specific probes, as shown in Figure 10b.

It is common practice to display either the G-banded or DAPI monochrome images after some manual and/or automated preprocessing to normalize or enhance spatial features. These images are referred to herein as banded images. These images can be displayed either as a dark foreground on a light background or vice versa.

Whether the multi-band image is composed of many spectral wavelength bands or fewer bands obtained using image acquisition technology, such as described by Speicher et al. in Nature Genetics 12 (1996) 368-375, it is possible to create a color map to visualize these data. Such a color map can be constructed in several ways. In the simplest method, some of the bands are assigned colors and these layers are then superimposed to create a color image. Another method is to assign each color layer a color using a

applied function (for each pixel) to build up the complete band set and then overlay the color layers. This color map can then be enhanced manually and/or automatically using image enhancement techniques. Examples of color maps are the RGB maps and the classification maps described hereinabove. For the purposes of this description, a color map refers to any color map generated using the data acquired from the imaging measurement of the multi-dye labeled materials.

In the case of detecting a metaphase chromosome spread labeled with multiple fluorescent dyes using an interferometer-based spectral imaging device such as the SpectraCube™ system, it is found that there is a significant difference between the spatial characteristics of the post-banding and color imaging. In the color imaging, as seen in Figure 10b, the chromosomes appear thick and in many cases the centromere is not detectable. As seen in Figure 10a, for example, this is probably due to protruding probe molecules from the general outline of the chromosomes as seen during banding, which are attached to the chromosomes at one end and loose at the other end.

The idea described here is to derive a composite color/banding map using both the banding and color maps, showing features that were not visible in either map by itself. In this way, a fundamental advantage is achieved over displaying a banding or color map separately: the user gets an immediate and instantaneous impression of the chromosomes, in the way that the user has been used to seeing and analyzing them for many years, while at the same time getting the additional information not available in the banding map, provided by the color map based on spectral vector classification. The superposition can only be done after the maps have been fully acquired, analyzed and classified by software, since they cannot be acquired simultaneously due to the different labeling techniques required for the banding and the spectral or vector-based maps.

Thus, the approach described here uses the monochrome post-banding image and the color image to generate a composite color/post-banding image. The example shown below is from a chromosome metaphase,

56
although the idea can be applied to other dye-labeled materials.

Before the images can be used to create the composite image, they must be spatially registered so that the correspondence between the points in the two images is known. In this case, this was achieved by resizing, rotating and shifting the color image until it matches the monochrome image. In general, this correspondence is carried out automatically, manually or semi-automatically.

Figures 11a and 11b show composite images according to the present invention. In the method used to create the composite image of Figure 11a, the luminance or intensity value is derived directly from the (white on black) image after DAPI banding and the shade is derived from the corresponding point in the color image, in this example an RGB image. The difference in the method used to create Figure 11b is that the luminance value was derived from the inverse of the image after DAPI banding. It will be appreciated that any monochrome positive or negative image of the chromosomes after banding according to any banding protocol may alternatively be used. Currently, DAPI banding is preferred because the fluorophores used for chromosome staining and DAPI detections can be performed using the same detection device, e.g., the SpectraCube™ system.

The results provide the user with (i) chromosome shapes that are familiar to the user and (ii) overlaid information from spectral and conventional banding karyotyping.

Thus, this implementation uses a monochrome image and a color image and derives a composite. A variation of this idea is to skip the generation of intermediate images, using the initial acquisitions of the monochrome and multi-band data to directly construct a composite image. However, other superpositions are within the scope of the present invention, for example if the color chromosomes are given the shape (outline) of the post-banded chromosomes even though the composite chromosomes have not been banded, just as a recorded superposition of post-banded chromosomes and color chromosomes is within the scope of the invention.

The chromosome classification method of the present invention as described herein

can be used for various applications, some of which are described in detail below. The following is a brief summary of likely applications of the chromosome classification method of the present invention in the field of molecular cell genetics. Two major areas of cell genetics in which the chromosome classification method of the present invention will have a significant impact, with particular emphasis on diagnostic applications, are (i) clinical cell genetics and (ii) tumor cell genetics.

First, the chromosome classification method of the present invention can be used for diagnostic purposes to detect chromosomal aberrations, for example, in cancerous cells, fetal cells, etc. in a manner similar to that described in U.S. Patent Application No. 08/635,820. Using conventional chromosome banding analysis methods, approximately 400,000 cases in both clinical cell genetics and cancer cell genetics are analyzed each year in the United States alone. Chromosome staining using the classification method of the present invention could be performed in addition to the conventional banding analysis method and would be helpful in identifying marker chromosomes that cannot be characterized using conventional banding alone. Acquired chromosomal aberrations are primarily associated with malignant transformations. Broadly speaking, two malignant lesions can be distinguished: (i) hematologic malignant lesions and (ii) solid tumors. Since the cells of hematologic malignant lesions can be easily cultured in vitro , chromosome banding analysis of these tumors is one of the crowning success stories of cell genetics and genetic cancer research. Two well-known examples include the identification of the Philadelphia chromosome in chronic lymphocytic leukemia (CLL) and a specific chromosomal aberration in Burkitt's lymphoma. Many other tumor-specific chromosomal aberrations have been described in hematologic malignant lesions in the last decade and are used as a diagnostic and research tool. In many cases, delineation of a recurrent chromosomal aberration on a molecular basis has enabled the identification of the mechanism of the malignant lesion. Worryingly, little is known in the field of solid tumors (such as breast, colon, brain, lung, and other tumors). This discrepancy is all the more worrying because solid tumors play a higher role in human disease incidence than hematological malignancy. The difference is mainly due to technical difficulties common in solid tumor cell genetics. The often difficult cultivation of solid tumor cells, the poor quality of the chromosome

Preparations that prevent high-resolution cell genetic analysis and secondary chromosomal aberrations that are not necessarily associated with tumor initiation or progression are a common feature of these tumors. The availability of a hybridization-based chromosomal screening test (i.e., chromosome staining) closes a methodological gap and is urgently needed as described above. Comparative genomic hybridization helps in part in this regard. However, structural chromosomal aberration cannot be detected and always shows an average value of chromosomal aberration in a mixture of cells. It can be predicted that karyotyping-based hybridization could become a widely used method for detecting recurrent chromosomal aberrations in solid tumors, both in basic research and in the diagnostic laboratory.

Second, the chromosome classification method of the present invention can be used for comparative cellular genetics [see J. Weinberg and R. Stanyon, Current opinion in genetics and development 5 (1995) 792-797], in a manner similar to that described in U.S. Patent Application No. 08/635,820, to detect and visualize chromosomal aberrations that have resulted in changes in chromosome morphology during evolution. In the study of evolutionarily related species and in the study of model systems (e.g., mouse as a model system for humans), it is often necessary to obtain comparative genome maps in which chromosomes from two or more species are comparatively aligned according to their sequence similarities and, hence, according to the genetic information contained in the chromosomes. Use of the chromosome classification method of the present invention will facilitate the obtaining of such comparative maps.

For example, consider the preparation of a human-mouse comparative chromosome map. For this purpose, a complete set of chromosome stains from one of the species (e.g. human) must be simultaneously hybridized with chromosome spreads from the other species (mouse in the given example) and classified as described above. The result is a map of the mouse karyotype stained with the human chromosome stains. Consequently, a comparative alignment between the karyotypes of the two species can be made. Cross-species hybridization can even be used to generate a repetitive color banding pattern in the chromosomes analyzed, which is helpful in cell genetic analysis of chromosome aberrations. For this purpose, a smaller number of fluorophores can be used. For example, the use of two mouse stains, each containing approximately half of the mouse chromosomes and labeled with a

distinguishable fluorophores, produce a two-color banding pattern when hybridized with a human chromosome spread.

Third, the chromosome classification method of the invention can be applied to cells during interphase, mitosis or meiosis. For example, this classification approach can be used to detect three-dimensional arrangements of the interphase chromosomes. So far, little is known about chromosome organization during interphase, and yet it can be reasonably assumed that changes in chromosome organization occur during interphase in malignant cells. Consequently, the classification approach of the invention can be of great value for early detection of various malignant changes by defining the stage of the malignant disease and thus better tailoring treatment to patients under investigation, etc. It should be noted that use of the classification method of the invention in combination with three-dimensional reconstruction means (for example, a confocal microscope) can be used to obtain three-dimensional information about chromosome organization during interphase, mitosis, meiosis.

Fourth, many cancers and genetic disorders are characterized by chromosome deletions, translocations and other rearrangements and severe abnormalities (e.g., gene amplification). Use of the chromosome classification method of the present invention will increase the ability to detect such abnormalities.

Fifth, one of the common chromosomal aberrations is associated with Down syndrome. It has long been established that Down syndrome is associated with chromosome 21 trisomy. More careful investigation has shown that a particular region of chromosome 21 (21q22) is always associated with this common syndrome (i.e., occurs in trisomy). In some cases, however, the karyotype of individuals affected by Down syndrome is apparently normal when determined by conventional G- or R-banding karyotyping techniques. The widely accepted explanation for this phenomenon is that in these cases the trisomy arises from a fragment derived from the 21q22 chromosome region, which fragment is small and below the resolution of conventional banding techniques. However, the use of the classification method of the present invention will enable these previously undetectable chromosome 21 trisomies to be detected in embryonic cells, for example, obtained via chorionic villus sampling or from maternal peripheral blood, and will enable more informed genetic counseling of women at risk. It should be noted that

that chromosome 13 and chromosome 18 or their fragments have also been reported to occur in trisomies resulting in the birth of strikingly abnormal children and that the classification method of the invention can be similarly used for prenatal diagnosis of these devastating chromosome 13 or 18 trisomies.

Sixth, the classification method of the invention, combined with rapidly developing techniques for separating embryonic cells from the peripheral blood of a pregnant woman, will be of great value for prenatal low-risk karyotyping for the detection of chromosome 21 trisomies and other less common chromosomal abnormalities.

Seventh, the classification method of the invention can be used to generate a multicolor banding pattern of chromosomes (i.e., barcoding, multicolor banding karyotype). For details regarding chromosome barcoding, the reader is referred to C. Lengauer et al., Hum. Molec. Genet. 5 (1993) 505-512. The first goal of the Human Genome Project (HGP) is nearing completion. This goal is the generation of a physical map of the human genome. The term "physical map" refers to cloning the entire genome in large insert vectors such as YAC clones or BAC clones and mapping these clones using genetic, cellular genetic and physical mapping. Two major sources of human DNA have been used in this endeavor, radiation hybrid cell lines and YAC contigs containing overlapping clones for all human chromosomes. Completion of the map allows one to retrieve, for virtually any region in the genome, specific clones required to identify genes causally involved in inherited or acquired genetic diseases, including cancer. By combining FISH with multiple YAC or BAC clones or radiation hybrids and spectral imaging, it is possible to generate a multicolor banding pattern that will ultimately link the genetic and cell genetic maps. As an example, consider the use of a radiation hybrid panel (Stanford Panel) [see Barret J.H., Genetic mapping based on radiation hybrids, Genomics 13 (1992) 95-103]. Each individual panel of the Stanford panel contains a set of DNA fragments with an average fragment size of approximately 5000 Kbp. Each individual panel covers approximately 20% of the human genome. Cohybridization of fluorescent probes derived from five such panels should therefore provide coverage of most of the genome and thus label all human chromosomes. However, the fragments are randomly distributed in the individual panels. Therefore, the number of panels required for a

·

&iacgr; · &diams; <

complete coverage of the human genome is required (e.g. 6-10 panels). In the following description, it is assumed that five individual panels are used. The chromosome fragments from each panel are labelled with a different fluorophore (or with a different combination of fluorophores, e.g. combinatorial labelling or hybridisation strategies), for example by direct incorporation of dUTP-conjugated fluorophores using primers derived from interspersed repetitive sequences (IRS, e.g. Alu sequences) in a procedure known in the art as an IRS-PCR approach, such as AIu-PCR, which guarantees exclusive amplification and labelling of human sequences. If DNA from a species other than human is to be amplified and/or labeled in this way, a species-specific interspersed repetitive sequence (IRS) characteristic of the genome of that species must be used to derive suitable PCR primers. A single, separate hybridization of any of the individual panels would yield a banding pattern of a chromosome spread in one color with approximately 20% genome coverage and an average band size of 5000 Kbp. Due to the random overlap of single chromosome fragments in five hybrid panels, cohybridization of the five differentially labeled groups (each group represented by a single panel) of fragments would yield a banding pattern that includes bands that have pure colors, bands that each have a combination of two, three, four as well as five, for a total of 31 color combinations, which color combinations can be distinguished using spectral imaging. The generation of a high-resolution multicolor banding pattern of chromosomes has two distinct advantages over the use of chromosome staining probes (i.e., chromosome stains), as follows. Chromosome staining is a well-adapted tool for detecting interchromosomal aberrations, such as translocations or homogeneous staining regions, as well as extra chromosomal material, such as marker chromosomes or double-minute chromosomes. Intrachromosomal aberrations, such as deletions or duplications, would only be detected if the size of the aberrations affected the length of the chromosomes, while chromosomal inversions are not detectable at all by this method. However, using a multi-color banding pattern, both inter- and intrachromosomal aberrations could be diagnosed, as they affect the sequence of the chromosomal bands. A major advantage of high-resolution multi-color banding using pre-mapped DNA fragments (e.g. YAC clones and radiation hybrid cell lines) is the ability to integrate the genetic and cell genetic map. Each multicolor band is characterized by a specific set of sequence tagged sites. These are PCR

&bull; » &diams;

Products that occur only once in the genome. A description of the utility of an integrated genetic and cellular genetic map follows. For example, the lack of a specific color band on a chromosome derived from a tumor cell indicates a microdeletion, which often reflects the loss of a tumor suppressor gene. Knowledge of the sequence target sites (STS's) specific to this band would make it possible to screen a large insert clone collection and obtain a number of specific clones that are highly likely to contain the gene deleted in the example described. It should be noted that with large-scale sequencing efforts now underway and with the integration of expressed tagged sites (loci known to contain a gene), the value of a hybridization-based multicolor banding pattern would increase even further. It is also conceivable that such a multicolor banding pattern could be easily automated. Despite considerable efforts, the automation of cell genetic diagnostic procedures based on conventional chromosome banding has not yet been successful.

The approach described above will be applicable not only to generate a hybridization-based banding pattern of human chromosomes, but also for other mammalian (e.g. mouse) and non-mammalian species. This will be particularly useful in the analysis of human diseases, including cancer, in animal models. In analogy to the scenario described for the radiation hybrid panels, a multicolor banding pattern for all human chromosomes could be achieved by cohybridizing a set of individually large insert clones, such as YAC clones, Pl clones, BAC clones, or, depending on the desired resolution, by using contigs (overlapping clones) from these sources. In further analogy to the use of radiation hybrid panels, a multicolor banding pattern could be introduced by randomly labeling overlapping clones or contigs with different fluorophores. One of ordinary skill in the art will appreciate that obtaining clones involved in chromosome breakpoints or chromosomal aberration would be even simpler than using radiation hybrid panels. Another source of suitable chromosome fragments suitable for use in multicolor chromosome banding are fragments obtained via microdissection of chromosomes. Microdissected chromosome fragments are generated by manual or laser micromanipulation of chromosome spreads, as is also known in the art. The fragments thus prepared are usually amplified by polymerase chain reaction using, for example, degenerated oligonucleotide primers (DOP) in a method known in the art as DOP-PCR or using interspersed repetitive sequences (IRS, e.g. Alu sequences) in a method known in the art as IRS-PCR.

tt * · ■·■ &diams; · · · #■ *

known methods. Yet an additional source of chromosome fragments suitable for use in multicolor chromosome banding are fragments generated by DNA restriction approaches that generate large DNA fragments and electrophoresis approaches that size-separate large DNA fragments. With respect to generating large DNA fragments by DNA restriction, two approaches can be considered. According to the first approach, a partial digestion by an endonuclease (e.g., frequently or rarely cutting) is used, while according to the second approach, a complete digestion by a rarely cutting endonuclease (e.g., NotI) is used. The latter is presently preferred because a complete digestion can be repeated to achieve identical results in independent experiments, whereas a partial digestion is random in nature. Electrophoresis approaches that can size-separate large DNA fragments are well known in the art and include pulse field gel electrophoresis (PFGE). Thus, similar to that described above, for example, extracting DNA from successive regions along a PFGE strip, labeling the DNA extracted from each of the regions using a different fluorophore, and co-hybridizing probes so formed to chromosomes would result in a multi-color banding pattern of the chromosomes. Large DNA fragments can additionally be obtained via gradient centrifugation, such as sucrose or cesium chloride gradients, which is also known in the art. Nevertheless, it will be appreciated that using these approaches does not provide an opportunity to integrate the genetic and cellular genetic map as described above and is therefore less advantageous at present.

The generation of a multicolor banding pattern of chromosomes (i.e., a multicolor banding karyotype) based on fluorescent m-s#H hybridization and the chromosome classification method of the present invention can be used for various practical applications. These include, for example, (i) screening for chromosomal aberrations using, for example, specifically tailored cloning kits, (ii) screening for telomeric deletions that are otherwise difficult to detect, (iii) screening for chromosomal aneuploidies during prenatal diagnosis, (iv) screening for recurrent chromosomal breakpoints, (v) comparative multicolour genomic hybridization, (vii) combining multicolour FISH with other cytological and immunohistochemical staining for multiparameter analysis of tumor cells, (viii) combining multicolour banding with conventional R or G banding, (ix) analyzing genetic aberrations directly in interphase cells and (x) screening for

64
chromosomal aberrations in radiation- or mutagen-exposed populations.

The main advantage of the method of the present invention is the introduction of a method for chromosome classification that uses internal reference vectors for pixel classification, where usually internal reference vectors provide better performance. Another advantage of the present invention is that the method can be easily automated. Using this method, therefore, semi- or unskilled cell geneticists are enabled to obtain clear and informative karyotypes, which may include a superposition of spectral and spatial information.
10

While the invention has been described in terms of a limited number of embodiments, it will be apparent that many changes, modifications and other applications of the invention may be made.

Claims (6)

1. A device for displaying chromosomes, the device comprising: a) a multi-band collection device provided with algorithms to provide a visual presentation device with an image of all the chromosomes or chromosome regions of a cell, each of the chromosomes or chromosome regions being stained with a different fluorophore or combination of fluorophores, the image showing the chromosomes or chromosome regions in different, distinguishable colors, each of the chromosomes or chromosome regions being assigned one of the different, distinguishable colors, and (b) a visual presentation device for representing the chromosomes or chromosome regions. 2. A device for displaying chromosomes, the device comprising: a) a multi-band collection device provided with algorithms to provide a visual presentation device with a composite image of at least some chromosomes or chromosome regions of a cell, each of the chromosomes or chromosome regions being stained with a different fluorophore or combination of fluorophores and banded using a chromosome banding technique to obtain a characteristic banding pattern for each of the chromosomes or chromosome regions, the image showing the chromosomes or chromosome regions in distinguishable colors, each of the chromosomes or chromosome regions being associated with one of the distinguishable colors, and further each of the distinguishable colors acquiring an intensity pattern that is in accordance with the banding pattern of each of the chromosomes or chromosome regions, such that both the distinguishable colors and the banding patterns of each of the chromosomes or Chromosome regions are superimposed, and (b) a visual presentation device for representing the chromosomes or chromosome regions. 3. A device for displaying chromosomes, the device comprising: a) a multi-band collection device supplemented with algorithms to provide a visual presentation device with an image of at least some chromosomes or chromosome regions of a cell, each of the chromosomes or chromosome regions being stained with a different fluorophore or combination of fluorophores and banded using a chromosome banding technique to achieve a distinctive banding pattern and shape for each of the chromosomes or chromosome regions, the image showing the chromosomes or chromosome regions in distinguishable colors, each of the chromosomes or chromosome regions being assigned one of the distinguishable colors, and further each of the chromosomes or chromosome regions acquiring the distinctive shape so that both the distinguishable colors and the shapes of each of the chromosomes or chromosome regions are superimposed, and (b) a visual presentation device for representing the chromosomes or chromosome regions. 4. A display device for displaying chromosomes, the device comprising: a) a multi-band collection device supplemented with algorithms to provide a visual presentation device with a composite image of a chromosome or region of a chromosome, the chromosome or region of a chromosome being shown in a color, the color including a banding pattern of alternating lighter and darker bands, both the color and the banding pattern indicative of the chromosome region of chromosome identification, and b) a visual presentation device for representing the chromosomes or chromosome regions. 5. A device for displaying chromosomes, the device comprising: a) a multi-band collection device supplemented with algorithms to provide a visual presentation device with a composite image of a chromosome or region of a chromosome, the chromosome or region of a chromosome being shown in a color, the color providing cues for identifying the chromosome or chromosome region, the chromosome or chromosome region having a shape similar to its shape after banding using a conventional chromosome banding technique, and b) a visual presentation device for representing the chromosomes or chromosome regions. 6. A device for displaying chromosomes, the device comprising: (a) a multi-band collection device provided with algorithms to provide a visual presentation device with a composite image of a chromosome or region of a chromosome, the composite image including a color image and a banded image of the chromosome or region of a chromosome, and b) a visual presentation device for representing the chromosomes or chromosome regions.