AU2004201406B2 - A digital imaging system for assays in well plates, gels and blots - Google Patents

Description

Our Ref: 12237011 P/00/011 Regulation 3:2

AUSTRALIA

Patents Act 1990

ORIGINAL

COMPLETE SPECIFICATION STANDARD PATENT Applicant(s): Imaging Research, Inc.

Brock University 500 Glenridge Avenue, St. Catharines Ontario L2S 3A1 Canada Address for Service: Invention Title: DAVIES COLLISON CAVE Patent Trade Mark Attorneys Level 10, 10 Barrack Street Sydney, New South Wales, Australia, 2000 A digital imaging system for assays in well plates, gels and blots The following statement is a full description of this invention, including the best method of performing it known to me:- 5951 A DIGITAL IMAGING SYSTED FOR ASSAYS IN WELL PLATES, GELS AND BLOTS Field of The Invention The present invention relates generally to assay analyzing systems and, more particularly, concerns a system and method for creating digital images of randomly arranged specimens beads within gels, colonies within petri dishes) or specimens arranged in regular arrays wells in plastic plates, dots spotted onto membranes). The invention is capable of creating digital images and performing automated analyses of specimens which emit very low levels of fluorescence, chemiluminescence, or bioluminescence. More particularly, the invention is designed for the analysis of luminance arising from assays within well plates and gel media, and on membranes, glass, microfabricated devices, or other supports.

Background of The Invention Types of Assays Many chemical and molecular biological assays are designed so that changes in the absorbance, transmission, or emission of light reflect reactions within the specimen.

Therefore, instruments used to quantify these assays must detect alterations in luminance.

Wells. Some assays are conducted within discrete flasks or vials, while others are performed within plastic plates fabricated to contain a number of regularly spaced wells. "Well plate" assays are higher in throughput and lower in cost than similar assays in discrete containers. Standard well plates contain 96 wells in an area of 8 x 12 cm. The trend is to higher numbers of wells, within the s ame plate size. Today's highest commurercial density is 384 wells. Very high density arrays of small wells (microwells, e.g. thousands/plate with a fill volume of less than 1 ul/well) are under development, and will become commercially available as 'nicrowell filling and detection technologies mature.

Dot blots. Grids of small dots (reactive sites) are placed onto flat support membranes or slips of treated glass.

A higrh density grid can contain many thousands of discrete dots.

Grid assays usually involve hybridization with synthetic oligonucleotides, to look for genes containing specific sequences, or to determine the degree to which a particular gene is active. Applications include library screening, sequencing by hybridization, diagnosis by hybridization, and studies of gene expression. High density grids provide the potential for very high throughput at low cost, if analyzing the grids can be made simple and reliable. Therefore, considerable commercial attention is directed at companies developing technology for creating, detecting, and analyzing high density arrays of genomic sequences.

Combinatorial assays. Some assays involve small particles (typically beads coated with compounds) which act as the reactive sites. There might be many thousands of beads, each coated with a different compound molecular variants- of an enzy me) from a combinatorial library. These beads are exposed to a substance of interest a cloned receptor) in wells, or in a gel matrix. The beads which interact with the target substance are identified by fluorescence emission or absorption in the region around each bead. Beads which interact are surrounded by faint areas of altered luminance. Very sensitive detectors are required to identify the subtle alterations in luminance around the beads that interact with the target.

Electrophoretic separations. A solubilized sample is applied to a matrix, and an electrical potential is applied across the matrix. Because proteins or nucleic acids with different amino acid or nucleotide sequences each have a characteristic electrostatic charge and molecular size, components within the sample are separated by differences in the movement velocities with which they respond to the potential.

The separated components are visualized using isotopic, fluorescent, or luminescent labels. In many cases (e.g.

chemiluminescence), the luminance from the specimen is very dim.

Assays which occur within a regularly spaced array of active sites (wells, dot blots within a grid) can be referred to as fixed format assays. Assays which involve specimens that are irregularly distributed within a gel or blot matrix can be termed free format assays.

Fixed format assays are usually performed without imaging. In contrast, free format assays require the use of image analysis systems which can detect and quantify reactions at any position within an image.

Instruments designed for fixed format assays generally lack imaging capabilities, and have not been applied to free formats. Similarly, very few imaging instruments designed for free formats have been applied to wells, and other fixed format targets.

Nonimaging Counting Systems Nonimaging counting systems (liquid scintillation counters, luminometers, fluorescence polarization instruments, etc.) are essentially light meters. They use photomultipliers (PMTs) or light sensing diodes to detect alterations in the transmission or emission of light within wells. Like a light meter, these systems integrate the light output from each well into a single data point. They provide no information about spatial variations within the well, nor do they allow for variation in the packing density or positioning of active sites.

Each PMT reads one well at a time, and only a limited number of PMTs can be built into a counting system (12 is the maximum in existing counting systems). Though the limited number of PMTs means that a only few wells are read at a time, an array of wells can be analyzed by moving the PMT detector assembly many times.

The major advantages of nonimaging counting systems are that they are a "push-button" technology (easy to use), and that the technology is mature. Therefore, many such instruments are commercially available, and their performance is wellcharacterized.

The major disadvantages of counting systems are: a. Limited flexibility- few instruments can cope with 384 wells, and higher density arrays of fluorescent or luminescent specimens are out of the question.

b. Fixed format only- designed as well or vial readers, and cannot read specimens in free format.

C. Slow with dim assays- although scanning a few wells at a time can be very fast when light is plentiful, dim assays require longer counting times at each position within the scan. As there are many positions to be scanned, this can decrease throughput.

In summary, non- imaging counting systems are inf lexible and of fer limited throughput with some specimens.

Scanning Zmagere For flat specimens, an alternative to nonimaging counting is a scanning imager. Scanning imnagers, such as the Molecular Dynamics Storm, MD Fluormager, or Hitachi FMBIQ pass a laser or other light beam over the specimen, to excite fluorescence or reflectance in a point-by-point or line-by-line fashion. Confocal optics can be used to minimize out of focus fluorescence the Biomedical Photometrics MACROscope) at a sacrifice in speed and sensitivity. With all of these devices, an image is constructed over time by accumulating the points or lines in serial fashion.

Scanning imagers are usually applied to gels and blots, where they offer convenient operation. A specimen is inserted and, with minimal user interaction (there is no focusing, adjusting of illumination, etc.), the scan proceeds and an image is available. Like the nonimaging counting system, the scanning imiager is usually a push-button technology. This ease of use and reasonably good performance has lead to an increasing acceptance of scanning imagers in gel and blot analyses.

Scanning imagers have four major shortcomings: a. Slow scanning. The beam and detector assembly must be passed over the entire specimen, reading data at each point in the scan. Scanning a small specimen could easily take 5-10 minutes.- A large specimen might take X hour to scan. This slow scan limits throughput, and complicates the quantification of assays that change during the scan process.

b. Limited number of wavelengths. A limited number of fluorescence excitation wavelengths is provided by the optics. Therefore, only a limited number of assay methods can be used.

C. Low sensitivity. Most scanning imnagers exhibit lower sensitivity than a state of the art area imager.

d. Not appropriate for luminescence. Scanning imnagers a bright signal, resulting from the application of a beam of light to the specimen.

Therefore, specimens emitting dim endogenous luminescence reactions involving lucif erase or luminol) cannot be imaged.

e. Not appropriate for wells. Only flat specimens can be imaged. A limited number of confocal instruments can perform optical sectioning and then reconstruct the sections into a focused thick image.

Area xmging An area imaging system places the entire specimen onto a detector plane at one time. There is no need to move PDMTs or to scan a laser, because the camera images the entire specimen onto many small detector.elements (usually CCDs), in parallel.

The parallel acquisition phase is followed by a reading out of the entire image from the detector. Readout is a serial process, but is relatively fast, with rates ranging from thousands to millions of pixels/second.

Area imaging systems offer some very attractive potential advantages.

a. Because the entire specimen is imaged at once, the detection process can be very quick.

b. Given an appropriate illumination system, any excitation wavelength can be applied.

C. Luminescence reactions (emitting light without incident illumination) can be imaged, including both f lash and glow bioluminescence or chemiluminescence.

d. Free or fixed format specimens can be imaged.

Luminescence imaging is more easily implemented, in that illumination does not have to be applied. However, most luminescence reactions are quite dim, and this can make extreme demands upon existing area imaging technology. The standard strategy is to use sensitive, cooled scientific grade CCD cameras f or these types of specimens. However, in the absence of the present invention, integrating cameras will fail. to image many luminescent specimens. Therefore, the present invention can image specimens that other systems cannot.

Typical prior art systems apply area imaging to luminescent assays on flat membranes and luminescent assays in wells. Standard camera lenses are always used. The results of well imaging are flawed, in that there is no correction for parallax error.

There is more extensive prior art regarding use of area imaging in fluorescence. Fluorescence microscopy (see Brooker et al. US Patent No. 5,332,905) and routine gel/blot imaging are the most common applications. Prior art in microscopy has little relevance, as no provision is made for imaging large specimen areas.

The existing art relating to macro specimens is dominated by low cost commercial systems for routine gel/blot fluorescence. These systems can image large, bright areas using standard integrating CCD cameras. However, they have major disadvantages: a. Limited to the wavelengths emitted by gas discharge lamps. Typically some combination of UVA, UVB, tJVC, and/or white light lamps is provided. Other wavelengths cannot be obtained.

b. wavelengths cannot be altered during an assay. If the illumination must be changed during the assay as for calcium measurement with fura- 2) the devices cannot be adapted.

C. Insensitive to small alterations in fluorescence.

Transillumination comes from directly below the specimen into the detector optics. Therefore, even very good filters fail to remove all of the direct illumination, and this creates a high background of nonspecific illumination. Small alterations in fluorescence (typical of many assays) are lost within the nonspecific background.

d. Inefficient cameras and lenses. A very few systems use high-performance cameras. Even these few systems use standard CCTV or photographic lenses, which limit their application to bright specimens.

e. Parallax error precludes accurate well imaging. As fast, telecentric lenses have not been available, these, systems exhibit parallax error when imaging wells.

Novel features of the present invention minimize the disadvantages of known macro fluorescence systems. These novel features include: a. Illumination wavelengths may be selected without regard to the peak(s) of a gas discharge lamp or laser.

b. Using a computer-controlled. filter wheel or other device, illumination may be altered during an assay, C. Small alterations in fluorescence emission can be detected. Because fluorescence illumination comes via epi- illumination, or from a dorsal or lateral source, direct excitation illumination does not enter the optics. This renders the nonspecific background as low as possible.

d. Very ef ficient camera and lens system allow use with dim specimens.

e. Unique telecentric lens is both very fast, and removes parallax error so well plate assays are accurate.

A primary advantage of the present invention is its fast, telecentric lens, which can image an entire well plate at once, and which can provide efficient epi -illumination to transparent or opaque specimens. Fiber optic coupling to the specimen can be used instead of lens coupling. For example, a fiber optic lens has been used with an image intensified CCD camera run in photon counting mode for analyses of data in f ixed or free formats.- This approach yields good sensitivity, but has the following major disadvantages: a. Although it is suggested that the system could be used with fluorescent specimens, it would be limited to specimens that are trans illuminated, because there is no place to insert an epi- illumination mechanism.

Therefore, the fiber lens system would have degraded sensitivity, and could not be used with opaque specimens. Many specimens are opaque many well plates, nylon membranes).

b. Well plates are 8 x 12 cn. Image forming fiber optics of this size are very difficult and expensive to construct. Therefore, the specimen would have to be acquired as a number of small images, which would then be reassembled to show the entire specimen.

This multiple acquisition would preclude use of the device with assays which change over time.

An area imaging analysis system (LJANA) is disclosed by D. Neri et al. ("Multipurpose High Sensitivity Luminescence Analyzer", Biotechniques 20:708-713, 1996), which uses a cooled CCD, side-mounted fiber optic illuminator, and an excitation filter wheel to achieve some functions similar to the present invention (selection of wavelengths, area imaging). However, LUANA uses a side-mounted fiber optic, which is widely used in laboratory -built systems, and creates problems that are overcome by the present invention. Specifically, use of a side-mounted fiber optic provides very uneven illumination, particularly when used with wells. The epi- and transillumination systems of the present invention provide even illumination of both flat specimens and wells. Further, in LUANA, parallax would preclude imaging of assays in wells.

Another system (Fluorescence Imaging Plate Reader FIJIPR of NovelTech Inc., Ann Arbor MI) uses an area CCD to detect fluorescence within 96 well plates. This device is a nonimaging counting system, and uses the area CCD instead of multiple PMrs.

To achieve reasonable sensitivity, it runs in 96 well format and bins all pixels within each well into a single value. The device is not applicable to luminescence imaging, free format imaging, or higher density well formulations and is very costly.

There is extensive prior art in the use of imaging to detect assays incorporated within microfabricated devices (e.g.

"genosensors"). Some genosensors use scanning imagers, and detect emitted light with a scanning photomultiplier. Others use area CCDs to detect alterations at assay sites fabricated directly onto the CCD, or onto a coverelip that can be placed on the CCD. Genosensors have great potential when fixed targets are defined. For example, a chip is fabricated that looks for a specific sequence of genomic information, and this chip is used to screen large numbers of blood samples. While highly efficient for its designed sequence, the chip has to contain a great number of active sites if it is to be useful for screening a variety of sequences. Fabrication of chips with many thousands of sites is costly and difficult. Therefore,, the first generation of genosensors will be applied to screening for very specific sequences of nucleotides.

The inflexibility of the microfabricated device contrasts with the present invention, which does not require microfabrication of the assay substrate. Instead, the present invention permits assays to be conducted in wells, membranes, silicalized slides, or other environments. Almost any reaction may be quantified. Thus, the present invention could be used as an alternative technology to microfabrication. Because the present invention is flexible, and allows almost any chemistry to be assayed, it can be used for all phases of assay development. These include prototyping, and mass screening. The invention therefore provides an alternative to microfabrication, when microfabrication is not feasible or cost-effective.

Each of the prior art references discussed above treats some aspect of imaging assays. However, the prior art does not address all of the major problems in imaging large specimens at

I

P:\WPDOCS\AMD\spci\7587470.doc-27 Fb.ry 2001 low lights levels. The major problems in low light, macro imaging are: a. very high detector sensitivity required; b. flexible, monochromatic illumination of large areas is required; c. parallax error must be avoided; and d. more reliable procedures are need to find and quantify targets.

Broadly, the present invention seeks to provide an imaging system for assays which overcomes the shortcomings of prior art systems. It is specifically intended to provide a complete system for the area imaging of assays in wells and on membranes. It is specifically contemplated that the invention provide a complete system for the area imaging of chemiluminscent, fluorescent, chemifluorescent, bioluminescent, or other nonisotopic hybridisation assays, including high density dot blot arrays.

The present invention also seeks to image chemiluminscent, fluorescent, chemifluorescent, bioluminescent, or other nonisotopic assays, including combinatorial assays, in free format.

The present invention also seeks to provide software for digital deconvolution of the fluorescence image data. Application of the software decreases flare and out of focus information.

The present invention also seeks to provide a method and system for imaging assays which are flexible, reliable and efficient in use, particularly with low level emission.

The present invention also seeks to provide synergistic combination of detector, lens, imaging system, and illumination technologies which makes it able to image the types of specimens previously acquired with nonimaging counters and scanning imagers. In particular, it can be used with fixed or free formats, and with wells or flat specimens. It is able to detect fluorescence, luminesence, or transmission of light.

The features of the invention include that it detects and quantifies large arrays of regularly spaced targets, that it detects and quantifies targets that are not arranged in regular arrays, and that it performs automated analyses of any number of regularly spaced specimens, f rom small numbers of large wells to large numbers of very small wells or dot blots.

It is another f eature of the invention to provide an area illumination system that: can deliver homogenous monochromatic excitation to an entire well plate or similarly sized specimen, using standard and low cost interference filters to select the excitation wavelength; and can deliver varying wavelengths of homogenous monoqhromatic excitation to an entire well plate or similarly sized specimen, under computer control.

A system embodying the invention provides a lens designed specifically for assays in the well plate format. This lens is very efficient at transferring photons from the specimen to the CCD array (is fast), preferably contains an epiillumination system, and can be used with very dim specimens.

The lens is also telecentric.. A telecentric lens has -the property that it peers directly into all points within a well plate, and does not exhibit the parallax error that is characteristic of standard lenses.

A preferred system provides a telecentric and fast lens that generates an even field of epi- illumination, when required.

The lens is equipped with an internal fiber optic illumination system, that does not require a dichroic mirror. Preferably, the lens is constructed to accept an internal interference filter used as a barrier filter. Light rays passing through the lens are almost parallel when they strike the barrier f ilter, so that the filter operates at its specified wavelength and bandwidth tolerance.

It is a feature of the invention that it provides high light gathering efficiency, whether used with a fast telecenric lens, or standard photographic lenses.

A preferred system provides a CCD area array camera that has high quantum efficiency (approximately and high sensitivity (16 bit precision), so that most specimens can be detected by integration without intensification. Preferably, the system has an integrating, cooled CCD camera which has coupled thereto an optional image intensifier. In an embodiment intended for extremely low light levels, incident illumination from the specimen is amplified by the intensifier, and the amplified light is accumulated onto the integrating camera over an integration period. At the end of the integration period, the camera is read out to a dedicated controller or imaging apparatus to reproduce the light image. Multiple exposures may be used to increase the dynamic range of the camera. A light-tight specimen chamber is provided, to which all illumination and detection components may be mounted, and which contains the specimens.

A system in accordance with the invention may incorporate a translation stage (optional), that may be housed within the light-tight chamber and used to move large specimens 22 x 22 cm membranes) past the optical system. The invention controls the stage motion through software, and that creates a single composite image from the multiple "tiles" acquired with the translation stage.

Preferably, the invention provides software control that corrects the shading, geometric distortion, defocus, and noise errors inherent to the camera and lens system; and that removes as much nonspecific fluorescence as possible, using multiple images created with different excitation filters.

In particular, the invention provides software to deconvolve images from a single focal plane, using optical characteristics previously measured from the lens and detector system. It should be appreciated that data from multiple focal planes may also be deconvolved.

While the preferred embodiment of the invention uses a high-precision, cooled CCD camera, if cost is a major factor, the present invention could be constructed using lower cost integrating cameras. In this case, shorter integration periods can be achieved, with a reduction in image quality and ultimate sensitivity.

Brief Description of The Drawins Further objects, features and advantages of the invention will be understood more completely from the following detailed description of a presently preferred, but nonetheless illustrative embodiment, with reference being had to the accompanying drawings, in which: Figure 1 is a schematic illustration of a system in accordance with a first preferred embodiment (upright) of the invention; Figure 2 is a schematic illustration, in side view, of the fast, telecentric lens; Figure 3 is a detailed illustration of the optical and mechanical components of the lens and the emission filter holder; Figure 4 is a schematic diagram illustrating a second embodiment of a system in accordance with the invention useful for extreme low light applications, which has an intensifier mounted between the lens and the CCD camera; Figure 5 is a schematic illustration of the intensifier; Figure 6 is a schematic illustration of the diffuse illumination plate in side view, showing how discrete fiber bundles, from the main bundle are taken to locations within the .rectangular fiber holder; Figure 7 is a schematic illustration of the diffuse illumination plate in top view, showing how discrete fiber bundles from the main bundle are taken to an array of channels within the fiber holder; Figure 8 is schematic diagram of the CCD camera; Figure 9 is a flow chart illustrating the method utilized for image acquisition and analysis in accordance with the present invention; and Figure 10 is a flow chart illustrating the method utilized for locating targets in the process of Fig. 9.

Detailed Description of The Preferred Embodiments Turning now to the details of the drawings, Fig. 1 is a schematic diagram illustrating a preferred embodiment of an imaging system 1 in accordance with the present invention.

System 1 broadly comprises an illumination subsystem 10, an imaging subsystem 12 provided in an housing 14, and a control subsystem 16. The imaging subsystem 12 comprises a CCD camera subsystem 18 housed within a camera chamber 20 of housing .14 and a lens subassembly 22 extending between camera chamber 20 and a specimen chamber 24. In operation, illumination. subsystem io provides the necessary light energy to be applied to the specimen within chamber 24. Light energy emitted by the specimen is transmitted through lens subsystem 22 to camera 18, where an image is formed and transmitted to the control subsystem 16 for processing. Control subsystem 16 comprises a camera Control unit 26, which is a conventional unit matched to the particular camera 18 and a computer 28 which is programmed to control unit 26 and to receive data from camera 18, in order to achieve unique control and processing in accordance with the present invention.

The light source for the illumination subsystem 10 is preferably an arc lamp 30. Light from lamp 30 is Conducted via a liquid light guide 32 to the optical coupler or filter wheel 34. The liquid light guide 32 is advantageous in that it transmits in the UV range, and in that it acts to dif fuse the input illumination more than a fiber optic would do.

The optical coupler 34 contains a conventional f ilter holder (not shown) for standard, one inch diameter interference filters. In the preferred configuration, a computer controlled filter wheel is used instead of the optical coupler. The filter wheel can contain a number of filters, which can be rapidly changed under computer direction.

A fiber optic bundle 36 carries illumination from the optic coupler or filter wheel 34 to within the light-tight specimen chamber 24. The bundle 36 passes through a baffle 38, which allows it to move up and down during focusing of the specimen holder. Alternatively, the fiber optic bundle 40 from the epi-illumination ring light in lens 22 may be connected to the optical coupler 34.

Three forms of illumination system are described, each fed by a discrete fiber bundle. These are a transilluminating plate (42) a ring light external to the lens (not shown), and a ring light 44 internal to the lens (22) that performs epiillumination.

The transillumination plate is a rectangular chamber (see Fig-ures 6 and within which the discrete fibers 52 from

I

bundle 51 are separated and rotated by 90 degrees so that they point laterally, towards the specimen. The fibers 52 are distributed within the chamber in such a way that they minimize shading within the illumination pattern. To this end, a larger number of fibers lie in the peripherally outward portions of the chamber than lie at its center.

The rectangular chamber 50 contains a diffusing screen 54, and a quartz glass diffusing plate 56. These diffusing elements take as their input the discrete points of light from the fibers 52, and create a homogenous illumination over the surface of the plate 56. The chamber 50 may also contain a dark field stop, to allow light to enter the specimen from the side.

The external ring light consists of a ring of optical fibers aligned with the axis of the lens, with a hole in the center large enough to encircle the lens 22. The working distance of the ring light is matched to the focus distance of the lens 22.

The internal ring light 44 consists of a ring of optical fibers, mounted within and axially aligned with the. body of the telecentric lens 22, and behind its front lens element. A diffuser, polarizer, or other circular element may be placed at the front of the fiber ring 44.

The specimen well plate is carried within a holder 58 (Figure 6) that is mounted to the fiber optic chamber 50. The holder 58 grips the well plate at its edges. The bottom of the holder 58 is empty, so as not to impede viewing of the wells.

The holder 58 is mounted to a jack, which moves it in the vertical dimension. By adjusting the jack 60, the holder 58 moves relative to the lens 22 and the specimen is focused.

The lens 22 is a fast, telecentric lens. The lens contains an emission filter slot 62, which accepts three inch diameter interference filters for fluorescence imaging. it contains an internal fiber optic ring light 44, positioned behind the front lens element. The lens 22 is mounted to the camera chamber by a flange 64 (see Fig. 2) at its middle. The back of the lens projects into the camera chamber 20, providing ready access to the emission filter slot 62 without disturbing the specimen. The front of the lens projects into the specimen chamber 24.

The cooled CCD camera 18 is mounted directly to the lens. Because the camera has its own chamber 20, there is no need for concern regarding light leakage around the cooling, power and data cables that exit the chamber to the camera control unit.

All control, imaging, and analysis functions are resident within the computer 28.

Ill umin~ation Subsystemn The standard technology for monochromatic area illumination is to use gas discharge illuminators UJV light boxes), which can deliver about 5000 uW/cm2 of surface at the emission peaks (usually mercury). The lamps are coated with a filter that limits emission to a specific peak. Although fairly bright, gas discharge lamps are limited in wavelength to the peaks emitted by the excited gas within the lamp.

other than gas discharge lamps, very few descriptions of area illumination exist. The major problems are selection of wavelength, and that direct entrance of the illuminating beam into the collection optics degrades sensitivity. To avoid this, light can be delivered from above, from the side, or via dark field or refraction into the specimen. All of these techniques have severe limitations. Side-mounted fiber optic illuminators are uneven. They are also unsuited to wells or other non-flat specimens, because light enters the specimen at an angle and fails to penetrate deep targets. Refractive or dark field illuminators require special optical components at the well plate, and cannot be used with opaque specimens.

A more flexible area illumination system would use a broad-band illumination source, and would allow any wavelength of monochromatic illumination to be selected by precision f ilters (usually interference filters). Filters are preferred, because variable monochromators or low cost tunable lasers lack sufficient light output when diffused over large areas.

Mercury or xenon arc lamps are of ten selected f or filter-based monochromatic excitation. The advantage of an arc lamp is that its output can be made into a narrow beam that can be passed through a small and readily available interference filter, before being spread over the entire surface of the specimen. Either a lens or fiber optic may be used to transmit the monochromatic light from the filter to the specimen.

The present invention is much more flexible than any previous device. It applies diffuse transillumination (through the specimen), dorsal illumination (via ring light or other source), or epi-illumination (through the lens) to the entire surface of the specimen. Epi-illumination is preferred, because it usually results in lower backgrounds, broader dynamic range, and more linear fluorescence response under real-world conditions. The ability to deliver large area monochromatic epiillumination is one critical factor that sets the present invention apart from prior art.

The present invention addresses three main problems in illumination delivery.

a. Filter availability Close-tolerance filters a nm bandwidth filter), which are readily available ,in small sizes, are not available for large areas of illumination. This problem is overcome by use of standard interference filters.

b. Illumination delivery Application of even, monochromatic, and selectable illumination over an 8 x 12 cm area is a feature of the present invention.

An optical coupler or computer-controlled filter wheel accepts standard interference filters, and is used to select wavelengths. The optical coupler or wheel may be attached to a specially designed fiber optic plate for transillumination, to a fiber optic ring or panel light for dorsal illumination, or to a fiber optic illumination assembly within the lens, for epiillumination.

c. Intensity The excitation illumination is spread over a large area (typically 96 cm 2 As intensity decreases with the square of the illuminated area, the resulting excitation intensity is very low indeed. In many cases, emitted fluorescence will not be detected with standard, scientific-grade cooled CCD cameras.

The very sensitive detector of the present invention is capable of imaging the low levels of fluorescence emitted from large specimens. For the most extreme low light conditions, the present invention incorporates an optional light amplification system that may be inserted between the lens and the CCD camera (see.below).

Lens Subassembly Figure 2 shows the general arrangement of illumination and filter components within the telecentric lens 22. The lens has mounted within it a fiber optic ring light 44, which projects monochromatic illumination through the front lens element onto the specimen (leftward in Fig. The focus plane of the ring light is at B, while the focus plane of the entire lens is in front of that point, at A. Placing the focus of the ring light at a point beyond the specimen minimizes specular reflections from the specimen.

The emission filter slot 62 allows insertion of an interference filter that removes excitation illumination from the incoming rays, leaving only the fluorescence emitted by the specimen.

Figure 3 shows best the optical components of the telecentric, macro lens 22. The lens has 39 surfaces, and the following characteristics: Effective focal length 164.436 mm Numerical aperture .443 Magnification 0.25 Note that light rays are almost parallel at the emission filter slot 62. This allows the filter to operate at its specified wavelength and bandwidth.

Although the present invention may be used with any lens, the highest sensitivity is available from its specially designed lens. This lens is fast, telecentric, and incorporates the epi-illumination system appropriate to large specimen formats.

Epi-illumination is a standard technology in fluorescence microscopy, where small areas are illuminated. The most efficient way to illuminate a small area is to place dichroic beam splitter behind the objective. A dichroic beam splitter or mirror is a partially reflective surface that reflects one wavelength range, while allowing another wavelength range to pass through.

On a microscope, illumination enters the dichroic mirror from the side. The mirror is angled to reflect the excitation light down through the objective toward the specimen.

Fluorescence emitted by the specimen (shifted up in wavelength from excitation) is collected by the objective, which passes it upwards towards the dichroic mirror. The dichroic mirror is transparent to the emission wavelength, so that the light proceeds through the dichroic to the detector plane. A different dichroic is required for each excitation/emission wavelength.

There are major difficulties in applying the standard form of dichroic-based epi-illumination system to macro imaging.

a. The dichroic mirror must be at least as large as the objective it must fill. Camera lenses are much larger than microscope objectives, and would need correspondingly large dichroic mirrors. Dichroic mirrors this large are not readily available.

b. In a fast macro lens, it is critical that the back lens element be mounted as close as possible to the CCD. Any increase in the distance between the rearmost lens and the CCD markedly reduces the working f number and the light-gathering efficiency. Therefore, there is no roam for a dichroic to be mounted behind the lens.

c. In a normal epi-illumination system, the dichroic reflects excitation through the entire lens. For this reason, transmission of excitation illumination is highly subject to the optical characteristics of the glasses used in the lens. Very costly (and difficult to work) quartz glass optics are required for UV epiillumination. These UV-transparent optics can be constructed in the small sizes needed for a microscope objective, but would be astronomically expensive in the large sizes described for the present invention.

d. Dichroic beam splitters absorb light. Typically, they are 80-90% efficient.

A unique property of the present invention is that no dichroic is necessary. The telecentric lens is large, so there is room to install an illumination assembly within its body. The illuminator is mounted so that it shines directly at the front lens element, from behind. This illuminates the specimen, without any need of a reflective dichroic mirror. Any stray excitation illumination that is reflected back through the lens is removed by the emission barrier filter, located posterior to the illumination source.

Further, the lens is designed so that only one of the fifteen internal lens components resides in front of the internal illuminator. This has the advantage that internal flare and reflections are minimized. Of equal importance, only the front lens needs to be transparent to UV. A single UV-transparent lens is costly, but not prohibitively so.

The front element of the lens is calculated so as. to focus the illumination source beyond the plane of the specimen.

The defocus of the illumination source at the specimen plane minimizes reflections. As many well plates are constructed of polished plastic, and tend to generate specular reflections, this is an important feature.

The lens is highly efficient. The collection of the lens is 4.5. This implies a collection solid angle of 0.03891 sr, and a collection efficiency of 0.03891/4p .3096%. The expected transmission value is 0.85-0.90, giving an overall collection efficiency of 0.263-0.279%. In comparison to an F/1.2 photographic lens, the expected improvement with the present lens is about 340%.

The present lens is telecentric. A telecentric lens is free of parallax error. Images of deep, narrow targets, made with standard lenses, exhibit parallax error. Circular targets at the center of the image are seen as true circles. However, the lens peers into lateral targets at an angle. Therefore, these lateral targets are seen as semilunar shapes. In many cases, one cannot see the bottom of a well at all. A telecentric lens collects parallel rays, over the entire area of a well plate. Thus, it does not peer into any wells at an angle and is free of parallax error.

A critical advantage of the present lens is that the internal beam is collimated at a position appropriate to the insertion of a barrier filter. That is, the lens is calculated so that rays are nearly parallel, at a point about midway in the lens barrel. The lens accepts an interference filter at this point. The filter serves to remove excitation illumination, and other nonspecific light. The collimated beam at this point is critical, because interference filters must be mounted orthogonal to the incoming illumination. If the incoming illumination is at an angle, the filter exhibits alterations in the wavelengths that it passes. In the present invention, light rays are almost parallel when they strike the filter, yielding the best possible performance.

The telecentric lens has a fixed field of view (about 14.5 cm diameter, in this case) but, if larger specimens need to be imaged, a motorized translation table may be mounted within the light-tight chamber. The translation table moves the specimen relative to the lens, under computer control. After each motion, a single "tile" is acquired. When the entire specimen has been imaged, all the tiles are recomposed (by the software) into a single large image, retaining telecentricity, freedom from parallax error, and high resolution over its entire surface.

Extreme Low Light Modification Figure 4 shows a modification to system of Fig. 1, addition of an optional intensifier 70 to provide an alternate system useful for extreme low light imaging. In all other respects the system is essentially identical to that of Fig. 1.

The intensifier 70 is mounted between the telecentric lens 22 and the CCD camera 18.

Figure 5 shows best the intensifier 70 as being of the GEN 3 type, and including a photosensitive cathode 72, a microchannel plate (MCP) 74, a phosphor screen 76, and a vacuum sealed body or enclosure 78. The fast, telecentric lens 22 (Figures 2,3) is. placed in front of this assembly 70. At its output, the lens is focused on an input window of the cathode 72 so as to transfer the specimen image thereto. The photosensitive cathode 72 is selected to emit electrons in proportion to the intensity of light falling upon it. The MCP 74 is positioned within the vacuum sealed body 78, between the cathode 72, and the phosphor screen 76 and coupled to the cathode 72 at each end.

The MCP 74 is provided with an array of small diameter MCP channels, each of which is coated with gallium arsenide. The electrons emitted from the cathode 72 are accelerated along the MCP channels to the phosphor screen 76. As the electrons from the cathode are accelerated along the small diameter channels, they strike the coated channel walls to produce additional electrons. As the multiplied electrons leave the MCP channels, they strike the phosphor screen 76 and produce an intensified image of the specimen on an output window. This image is coupled to the CCD 84 element in the camera by a lens It has been found that the use of the Exctended Blue GEN 3 image -intensifier is advantageous, over other types of intensifiers in that the image provided on the output screen is sharper, has less shading error, and has less noise than those produced by GEN 1 and GEN 2 intensifiers. It is to be appreciated, however, that as better intensifier technologies are developed, they may be incorporated into the present system.

The integrating camera 18 is configured so that the highly amplified image generated on the output window 78 is focused by the intermediate lens 80 onto the CCD element 84. To image low light specimens, the CCD element 84 of camera 18 integrates for a period. During the integration period, photons from the output window incident to the CCD element 84 are stored as negative charges (the signal) in numerous discrete regions of the CCD element 84. The amount of charge in each discrete region of the CMD element 84 is accumulated as followt;.

signal =Incident light x Quantum efficiency x Integration time The greater the relative intensity of the incident light coming from the intensifier 70, the greater the signal stored in the corresponding region of the CCD element 84.

For the most extreme low light conditions, as with the scintillation proximity assay, the present invention allows a light amplifier to be inserted between the lens and the CCD camera. In the preferred configuration, this light amplifier is an image intensifier. Intensification, as for example, is disclosed in U.S. Patent No. 5,204,533 to Simonet, involves the coupling of an image intensifier to a CCD camera. The image intensifier typically includes a photocathode, a phosphor screen, and a microchannel plate (MCP) connected between the photocathode and phosphor screen. Light amplification factors of up to about 90,000 are possible with this type of device.

With the intensifier inserted into the optical chain, the present invention becomes an image intensified CCD (ICCD) camera. In an ICCD camera, the image is created at three or f our planes. At each of these planes, there is some loss of quantum efficiency. Therefore, the image intensifier is operated at high gain to overcome signal losses within the optical chain. At very -high gain factors, noise and ionic feedback through the MCP become so severe that further improvement of sensitivity is impossible. Even when run at maximum gain, conventional image intensified CCD cameras are not sensitive enough to image the dimmest specimens.

Faced with a typical very dim specimen, most IcCD cameras will fail to produce an image, or will produce a very poor image, in which the target will be difficult to discriminate from background, and the true range of target intensities will not be rendered. In the worst cases, the target will be indiscriminable from background.

Conventional image intensified CCD cameras use an integration period equal to a single television frame. The short integration period allows the intensifier to be used with standard, low-cost video cameras, as for example, are used in the television industry. In other cases, the inter.3if ier is gated, to use very short integration periods I msec). The use of gating allows the intensifier to be used in a photon counting mode.

The present invention offers two methods by which intensified light may be used. The preferred method involves continuous integration of the Output of the intensifier onto a cooled CCD camera. This method is fast and efficient, but has limited dynamic range. Cooling of the intensifier, Or multiple exposures for different times, may be used to improve the dynamic range. A second method involves looking at shorter periods of intensifier output, and photon counting. This method is much slower, but has broad dynamic range. The present invention allows either strategy to be selected, as warranted by the specimen.

Prior art exists for the use of intensified CCD cameras in well plate assay imaging. Martin and Bronstein (1994) and Roda et al. (1996) discuss use of an intensified CCD camera for the imaging of chemiluminescent specimens. ,Only brigh t specimens can be Been. No provisions are made for imaging deep wells without parallax error, or for applying monochromatic excitation, to the specimen.

U.S. Patent No. 4,922,092 (1990) to Rushbrooke et al.

discloses the use of an image intensified CCD camera which is coupled to a special fibre optic lens. The fibre optic lens consists of bundles which transmit light between an array of wells and the input of the intensifier. While the invention disclosed by Rushbrooke is free of parallax, and may be suitable for standard 96 or 384 well plates, it would be incapable of imaging the very high density well arrays addressed by the present invention. Further, the invention disclosed by Rushbrooke lacks illumination capabilities. It is also incapable of imaging specimens in free format, because there is space between the input bundles that is not addressed. By using lens input, as opposed to fiber optics, the present invention allows free format imaging.

In sum, the present embodiment of the invention allows the use of an optional intensifier placed behind the lens, to detect the most extreme low light specimens. When intensified, the device can be run in continuous integration or photon counting modes.

With the system shown in Figures 4 and 5, only the CCD sensor is cooled. This is sufficient for most purposes. It is to be appreciated however, that the intensifier photocathode 72 could also be cooled, thereby improving the signal to noise ratio of the intensifier. Similarly, the entire photosensitive apparatus (intensifier CCD) can be cooled. However, cooling the entire photosensitive apparatus has the disadvantage that the efficiency of the phosphor on the fibre optic output window is decreased.

Although a high quality, scientific grade CCD camera can detect about 50 photoelectrons incident to the CCD (depending on how we set reliability of detection), this is not an accurate indication of performance in imaging luminescent specimens.

Real-world performance is complicated by the emission and collection properties of the entire optical chain, as well as by the performance of the CCD camera. ThereforE, we need to go beyond the QE of the detector, and examine the transfer efficiency of the entire system.

Three factors domuinate the transfer efficiency (photoelectrons generated photons emitted) of the detector system. These are the light collection efficiency of the lens, the quantum efficiency of the CMD detector, and the lens transmittance. We can calculate the number of~ photoelectrons generated as follows: Npe r c. e. Nw where: 7 is lens transmittance, about 85-90% for our lens is quantum efficiency of the CCD detector, typically about 35-40t, up to S0% in our case, and c.e. is collection efficiency of lens, le~ss than .1k for fast photographic lenses, about 1.2% in our case.

In a typical scientific grade CCD camera system, using the fastest available photographic lens (f and with a high quality cooled detector, the ccD will generate 1 photoelectron for about 5,000-10,000 photons generated from a point source in the sample.

The lens of the present invention offers a collection efficiency of about 0.271W. The efficiency of the CCD detector is about double that of other CCDs. The result is that the present invention has the theoretical ability to generate one photoelectron for about 500-1000 photons generated from a point source within the sample. This very high transfer efficiency allows detection of specimens that cannot be imaged with prior art systems.

In the alternate embodiment of the invention shown in Figures 4 and 5, the system incorporates an extended blue type of GEN 3 image intensifier. Other types of intensifiers, although less preferre d, may also be used. -The three major types of intensif ier (GEN 1, GEN 2 and GEN 3) differ in the organization of their components and in the materials of which the components are constructed. In a GEN 1 intensif ier, illumination incident to a photocathode results in emissions at a rate proportional to the intensity of the incident signal. The electrons emitted f rom the ph otocathode are than accelerated through a high potential electric f ield, and f ocused onto a phosphor screen using electrostatic or proximity focusing. The phosphor screen can be the input window to a video camera (as in the silicon intensified target camera), or can be viewed directly. GEN 1 intensifiers suffer from bothersome geometric distortion, and have relatively low quantum efficiency (about The GEN 2 intensifiers, like the GEN 3, incorporate a MCP into an image tube, between the cathode and an anode. The GEN 2 intensifiers are smaller, lower in noise, and have higher gain than the GEN 1 intensif iers. However their quantum efficiency is fairly low (typically e20%), and they tend to suffer f romn poor contrast transfer characteristics. In contrast, the GEN 3 intensifier tube has a quantum efficiency of about or higher (needs less gain), and very high intrinsic contrast transfer. with recent versions of the GEN 3, gain levels are about equal to those of a GEN 2 (ultimate gain level available is about 90,000). Therefore, a GEN 3 intensifier will tend to yield better images than a GEN 2. Where necessary for reasons of cost or specific design features, other forms of intensifier could be used. Similarly devices with high intrinsic gain (such as electron bombarded back-illuminated CCD sensors) could be used in place of image intensifiers.

The CCD camera 18 of the present invention could use integration periods locked to a gated power supply in the image intensifier, with the result that the camera could be read out at very short intervals. Using the gating and fast readout feature, and with the intensifier run at highest gain or with a multistage intensifier, the present invention can thereby be operated as a conventional photon counting camnera. Thus, the present system can advantageously be used for both direct imaging of faint specimens, or as a photon counting camera by changing its mode of operation from integration to gating.

CCD Camera Syatem Figure 8 is a schematic representation of the CCD camera 18-. The camera 18 includes a CCD element 84 positioned behind a camera aperture. To reduce dark noise produced by electrons within the CCD, the CCD element 84 is mounted to a heat sink 88, which in turn is thermally coupled to a Peltier cooling element and liquid circulation system for: providing enhanced heat dissipation. The lens is positioned over the aperture to focus the image on the CCD element 84. The fast, telecentric lens 22 (Figures 2 and 3) is mounted directly to the camera body by screws, after removing the photographic lens mount. Similarly, the image intensifier 70 (when present) is mounted directly to the camera body.

Area imaging systems use CCD arrays to form images.

Factors which influence the ability of CCD arrays to detect small numbers of incoming photons include quantum efficiency, readout noise, dark noise, and the small size of most imaging arrays 2.25 cm2).

Quantum efficiency (QE) describes the ability of the photodetector to convert incident photons into electron hole pairs in the CCD. Consumer-grade CCDs typically exhibit QE of about 12-15%. Standard, scientific grade cooled CCD cameras exhibit QE of about 40%. A very limited number of thinned, backilluminated CCDs can achieve OE of as high as 80% at peak detection wavelengths.

Readout noise originates in the output preamplifier of the CCD, which measures the small changes in voltage produced each time the charge content of one or more CCD elements is transferred to it. Readout noise is directly related to the readout rate, and is decreased by use of slow readout.

Dark noise is produced by thermally generated charges in the CCD. By increasing the background level, dark noise decreases dynamic range. The constant dark noise level can be subtracted from the image, but dark noise, also has a random noise component which cannot be subtracted. This component adds to the noise level of the detector. Dark noise is decreased by cooling the CCD.

The size of the CMD element is related to its ability to store photoelectrons (known as the well capacity, and, hence, its dynamic range. The larger each CCD element in the array, the larger the full well capacity and dynamic range of that element.

A broad dynamic range allows the detector to be used for longer exposure times, without saturation, and this enhances the detection of very small signals. Further, the signal to noise.

performance of larger elements is inherently higher than that of smaller elements. Most area imaging "systems use relatively small CCDs. This results in limited resolution for devices in which the discrete CMD elements are large, and limited dynamic range for devices in which the discrete CCD elements are small.

Devices with limited dynamic range cannot achieve 16 bit precision, and mrust be used wit h relatively bright specimens fluorescence microscopy, UV gels, very bright chemiluminescence).

The present invention incorporates a CC., system which is designed to minimize all of the problems just described. The CCD array is unusually large (6.25 cm 2 and efficient (about quantum efficient) The result is very high detector sensitivity with broad dynamic range (true 16 bit). the preferred support electronics include a high-precision digitizer, with minimal readout noise. Preferably, the camera is cooled to minimize dark noise.

An electro -mechanical shutter mechanism is additionally provided within the camera, for limiting the exposure of the image on the CCD element. Preferably the camera is a thinned, back- illuminated 1024 x 1024 pixel black and white camera with asynchronous reset capability, and high quantum efficiency. The camera provides a 16-bit digital signal output via digitization circuitry mounted within the camera control unit, and an interface card mounted within the computer. Data from the CCD are digitized by the camera control unit at the rate of 200,000 pixels/second, and transferred directly to the computer memory.

Following the integration period, the CCD camera accepts a trigger pulse from the computer to initiate closure of the electromechanical shutter. With the shutter closed, the image is transferred -from the CMD to the internal frame buffer of the computer.

Although this camera could be used without cooling the CCD element, extended periods of integration are achieved by using a CCD camera with an integral cooling element. The.

effectiveness of integration is limited by the degree of cooling.

With a non-refrigerated liquid cooling device, sensor temperatures of about -50*C (below ambient) can be achieved. At this temperature, dark noise accumulates at a rate of about 7-10 electrons/second. This type of cooling has the advantage of low cost and easy implementation.

It is to be appreciated, however, that longer periods of integration are possible if refrigerated liquid or cryogenic cooling are employed.

Control Subsystem The control subsystem 16 comprises, control unit 26 and computer 28. Camera control unit is a computer controllable unit provided by the manufacturer of camera 18 to control the camera.

Computer 28 is preferably a conventional computer running in the WindowsO environment and is programmed to achieve image acquisition and analysis in accordance with the present invention.

Camera-based imaging systems lack the sort of pushbutton operation that is typical of counting or scanning systems.

Focusing the camera, adjusting exposure time, and so forth, can all be inconvenient.

In fact, imaging is inherently more complex than counting single targets within wells. Nonimaging counting systems have a relatively easy task. They only need to control the scanning process, control internal calibration, and create a small array of data points representing each well. The sequence of steps might be as follows.

a. Calibrate detector against internal standard.

b. Illuminate one well.

c. Position a PMT over the illuminated well.

d. Read well.

e. Transfer data to spreadsheet.

f. Illuminate next well and repeat.

An area imaging system has a much more difficult task.

Imaging a well plate might include the following requirements.

a. Provide adequate illumination over the entire plate.

b. Control a high performance camera.

c. Store geometric and density correction factors.

d. Image'specimen.

e. Correct geometric and density variation.

f. If necessary, calibrate image to standards within the specimen.

g. Locate each well and quantify intensity.

h. Transfer data to spreadsheet.

These tasks can only be performed if the imaging system is equipped with software that performs functions b-h, above.

The present invention incorporates such software.

In particular, one aspect of the present invention is software which corrects for nonspecific background fluorescence by using two images. The first image is made with an excitation filter that excites as little specific fluorescence as possible, while exciting nonspecific fluorescence. The second image is made with an excitation filter that excites specific fluorescence as much as possible, and as little nonspecific fluorescence as possible. An optimal specific fluorescence image is made by subtracting the nonspecific image from the specific image.

Figure 9 is a flow chart illustrating the primary process performed by computer 28 in controlling the system 1 and acquiring data therefrom. After initiation of the process, an image of the specimen is acquired at block 200 using camera 18.

Known processes exist for acquiring bias images of a specimen.

Such bias images take into account all significant distortions and errors introduced by the system itself when an image is taken. Utilizing one of the known methods, a bias image for the specimen is acquired at step 202.

At Step 204, a non-specific image is acquired. This image determines the contribution of non-specimen components, such as the support substrate, to the image. This step is indicated as optional, since it would only be performed in the event that the specimen had to be illuminated in order to acquire the specimen image, in which event some light would also be reflected from non-specimen elements. On the other hand, if the specimen were the source of the light for the image (as in chemiluminescence), the non-specific image would not be acquired.

Similarly, the step at block 206 is optional, since it involves obtaining a non-specific bias image.

At block 208, the specimen bias image is removed or subtracted from the specimen image, and at block 210 the nonspecific bias image is subtracted from the non-specific image.

This results in two images in which bias effects have been compensated. At step 212, the compensated non-specific image is removed from the compensated specimen image to produce a working image in which the effects of the specimen are isolated. Those skilled in the art will appreciate that if steps 204 and 206 were not performed, steps 210 and 212 would also not be performed.

Following bias removal, various other corrections are provided for geometric warping originating in the lens), using known processes.

At step 214, the operator inputs to the computer the nominal "grid" spacing and "probe template". The grid spacing is the nominal center-to-center spacing of specimen samples on the substrate. The "probe template" is the nominal definition of a single target in terms of shape and area) corresponding to one dot on a membrane, one well in a plate, or similar target. Typically the probe template is a circular area, and there is one probe template for each target in the specimen.

A grid is composed of a matrix containing one probe template for each of the targets.

Optionally, the operator can also define an array of "anchor points." The specimen may include an array of thousands of potential samples. In some instances, a large proportion of these will be populated, and in others relatively few will. In those instances in which relatively few sample points are populated, the specimen will include predefined "anchor" points to aid the system locating the probe template positions. In those instances in which a large proportion of the potential sample sites are populated, the samples themselves provide a sufficient population to position the probe templates, and anchor points may be unnecessary.

At block 216, probe templates of the defined size with the defined grid spacing are generated and superimposed over the working specimen image. At this point, the operator can optionally provide a manual adjustment to the superimposed grid of probe templates, in order to bring them into general alignment with the actual specimens. He could do so, for example, by utilizing a mouse to shift the entire array then "grab specific probe templates and center them over the appropriate targets on the specimen. The operator might, for example, perform a general alignment by centering the probe templates in the four corners of the grid over the appropriate targets of the specimen.

Although not essential, this manual adjustment will speed and simplify the processing done by computer 28.

At block 218, a process is performed, described in more detail below, in order to determine more precise locations for the probe templates relative to the actual location of potential targets. At the outset of this process, at block 218, a determination is made whether the targets or anchor points have been adequately identified or defined. If targets have been well-defined, control is transferred to block 222, where the array of probe templates is aligned to the defined targets; if not, but anchors have been well-defined, control is transferred to block 220, where the array of probe templates is aligned to the anchors; otherwise, control is transferred to block 224, where the predefined grid spacing and probe template for the array are utilized. It will be appreciated that, in some instances, it may be desirable to align the array on anchors and then on targets.

Once the probe templates and targets are aligned, the measurements within the individual probe templates are decoded to different conditions. For example, a probe may be capable of assuming any of n conditions, and the process of block 226 could decode the sample at each probe to one of those conditions. The actual process is performed on a statistical basis, and is best understood from a simple example relating to resolving a binary decision. However, those skilled in the art will appreciate that the process could actually be applied to resolving a multiple condition process. In the simplest case, the binary decision is a "yes" or "no" decision, which could be related to the presence or absence of a certain condition. In accordance with the process at block 226, the actual levels at every probe of the specimen are measured, a mean and standard deviation are determined for the set of samples, and this results in a working statistical distribution. The decoding of a "yes" or "no" could then be done to any level of confidence selected by the operator.

The operator's selection of a level of confidence results in the determination of a threshold level based upon that level being located a calculated number of standard deviations from the mean on the distribution curve), and any signal above the threshold level would be considered a "yes', while any signal below the threshold level would be considered a "no." At block 228, a process is performed to generate a report of the array data, based upon the process performed at block 226. It is contemplated that this may be any form of report writing software which provides the operator a substantial amount of flexibility in preparing reports of a desired format.

Once the reports are generated, the process ends.

Attached as Appendix A is a more detailed discussion of the process of Fig. 9.

Figure 10 is a flow chart illustrating the process performed in block 222 of Fig. 9.

After initiation of the process, image background and noise are estimated at block 300. At block 302, a determination is made whether a group alignment of the grid to the array of targets is necessary. This could be done either visually by an operator or by the system. The purpose of this test is to determine whether the grid is aligned to the targets overall.

If done by the system, it would be performed by a conventional procedure for testing alignment of two regular patterns of shapes. If it is determined that adequate alignment of the group exists, control is transferred to block 306.

At block 304, a group alignment is performed. The purpose of this operation is to align the probe template grid roughly with the respective targets. The alignment may be done on the basis of the whole grid or part of the grid selected by the operator. This alignment could be done by the process discussed below with respect to block 306 for maximizing ID, except that ID is maximized over the entire grid.

At block 306, a step-wise process is performed within the area of each individual probe template to locate that point which yields the maximum integrated density, ID, within the probe template, given by the formula ID (x0, yO) =f D y) W(x-xO, y-yO) dxdy (1) where: (x0,y0) is the center point of a probe template; S(xO,yO) is the probe template area at (xO,yO); D(x,y) is the density value brighteners) at and W(x,y) is a weighting function a twodimensional Gaussian function with its maximum value at P: WPDOCSAMD\Wecid757470.dcc-27 Febnwary 2001 This yields an "A location" for each probe template, which is that location that provides the maximum value in formula The probe template location prior to block 306 will be referred to as the "G location".

At block 308, a confidence weighting is performed between the A location and G location, in order to arrive at the final location of the center of each probe template, the confidence weighting factor for each A location is a form of signal-to-noise ratio. That is, the value of ID at each point is proportional to the ratio between the ID value at that point and the value determined at block 300 for that point. In effect the weighting factors are 10 utilised to determine the position of the probe center along, a straight line between the A and G locations, with weighting determining how close the point is to the A location.

Although the detailed description describes and illustrates preferred embodiment of the present apparatus, the invention is not so limited. Modifications and variations will now appear to persons skilled in this art. For a definition of the invention reference may be had to the appended claims.

Throughout this specification and the claims which follow, unless the context requires otherwise, the word "comprise", and variations such as "comprises" or "comprising", will be understood to imply the inclusion of a stated integer or step or group of integers or steps but not the exclusion of any other integer or step or group of integers or steps.

The reference to any prior art in this specification is not, and should not be taken as, an acknowledgement or any form of suggestion that that prior art forms part of the common general knowledge in Australia.

APPENDIX A I Contents Inlrodnclon Imaging and Librvy Srening 3 Studying G e Expression 4 Details of the Nigh Density Grid

S

Manual 6 Automated Seginftsim 6 Segzmmntatin With Fixed Sampliog 7 Consmtrtn and Aligning dho 7 Fine ASi 8 8a Detail and Screening 8 Detail mode Gene librar Screenmeg 8 Summary of detail and sceeing 9 Selecting inteeting Targetr. Statistical 9 9 Detail Mode and Gone Quick Date Cbmncaim Ile Bountai Effects of Background Summary: Features of The Eigh Density Grid Study 12 Referessoes 13 38 High Density Grids Introduction rapid pace of Imicyadoriin nsolWlsbiWloV, chemistry ad rubOtiCs is exe~rting I profound effect on biomedical Vessmch in geonra, aid an perroaccatical lead discovery in paticula. As sawe companies are quick to adopt innovative technoogies, presms. grow upon all compettr. Everyone wants to find ways to decrease oat and increase speed in do discovery snd evaknifuir of potenial therapajac targets. 1he result is ths rapid growth of a new are of phirnaosutical science biomolecular screening. Biemolecula screening can be definied as the rapid and efficient laboratosy testing of large numbers of *amp==&d for potential theraputic efficacy The growth of screening is an dnevitable consequence of innovations wan new undestgani n catnomWa clne-isty, biological diversity, molecular genedc, ad other areas. For exampe, ondrinaterial cemists can start with a single compound from which huindreds of thouesands of potentially interesting compounds are generated. All of these compounds, or a slpalllcant subset, must be teated for biological activit. Another example lies in the discovery, datection and c aacteioln of the Interaction between compound aud geugs.

'The demands of cosoLulatorial chemistry, molecular gesmeics, and other applicaon have lead to the development of a specialized aspect of screening technology, high throghput screening anTS). Mi is not bamentally differnt from the general definition of biomolecular screening. We night consider it as a shot trd form of screesning, in which considerable expense and effort are expended to incease the rale at which Compounds an tsted. Optimized technologies in assay C itry, de0tection systemauo tirbts, and bioinformatfics all have roleo to pa in HFS., Onw way in which assy chemistry can be optimized for nainasyn efficiency is by miniaturestin of the assay format. For example, moving ficem 96 to 384 wall plates increases the number of assay sites per unit arna. It can be assumed 0dugnshhow enity microwell formats will soo appear Mirowell assays will follow the le-d ofother screening Protocols in which microfabricated devices or spotting robots are used to provide very dense arrays of DNA clones on solid support media (eg. Beattie at aL, 199S; Eggers at al., 1994; Khrapko et al, 1999, 1991: Lmniume t al, 1994; lipwbnt at 199S; Maskos and Southern, 1 9M- Mason. Rampal and Coassin, 1994; Pearson and Tomiccl. 1995; Pease et 1994; Salk, wals, Levenson and Erlicha 1919; Schena, Sbman, Davis and Brown, 1995; Southern, Maskos and Elder, 199).

The nmber of targets in a typical specimen is increasing rapidly. Today's prototype, microfaicaed devices lixorporste about 1000 -20,00 grid elmt (fo abrief rview, see Southern, 1996). Spotted assays routinely achieve much higher densitims In our archives, we have images coutaining more than 50,000 isotopicallylabeled cDNA dot blots hybrdinan a mm of about 20 x 20 an. We hae" seen much higher densities schimved using oaisotapic methods, i which pacin densities mre nt limited by spread of emission from the isotope. As the microplates, membranes, or other media become more, densely popilated with assy sites, the reamb is a high denity of sites in a regularl spaced grid pater n.hi is usually referred to am a hingh-density grid.

The traditional mhdsfor analyzing wells (generally phtn- tpirae counters) ane usually inadequate for hWdnity grids. Counting devices, which must address each gr& point one after the other, ame not able to handle large arrays of small assy sites. In contrast, a camera -based Imaging systema is well suited to both high density gridis, and to specimns in fise formats (those niot in a regularly spaced pattern). Exampples of frepe format specimens include combinatorial beadi assys in dishes, cell based assays in tissu cultures, and colony assays.

The intersection bet reentimging and screening technologies is still in the formastive stages, and procedures. are only just becoming avalale. Innovations in microfabricalion, in noifabricated, arrays (eg. membranes, microwella), and in detection technologies will combine with image analysis to yield tools for new screening paradigm in sequencing, diagnosis, and binding.

Imaging Resesu& Inc.

High Density Grids IMMEand Lft= Scregfilg Molecular genet., at the Study Of gene at the level of tbe DNA molecuile, involves isolation and chmaicterratim of do DNA spnding specific, Iwoe. Themostbaicpurposeis to reblte egammn to diseases or pathological processes.. For examle MKt canes"' reswi from noiin-Aie genetic mutaions (somai mutations) which resultm bn ormal cel inictlon. f specifi gam is 111ua11ly altered in association with a Particular caner, ih Swed aitmtion nuigl be utable as a test for the cancer. Genetic magmneeing techniques could also be used to study'the function of the gem and its encoded protein, produce -RU qda1niltiom of sre proteins and create animal models of genetic disorders. 7be starting point for all of this is the screening of cDNA and gwnk;l libraries. We screen to simply identify clones that hybridize to our probe, or we screen to identif clones that she- gone expression, in response to some, independent variable two cell line.).

The most comaot application of DNA library screening is the idendtficaliosi and Isolation of clanes IoMuonin to a specifc gone of inteet In order to isolate clame for a speifi gene, a probe for tha gone must be available. The DNA library is then screened far done. containing sequences hybridizing with tho probe.

The first step in isolating cloned DNA for a gem of interest is to spread the individual clones of a library into a spatial poner, with a discrete clone at each position in the pattern. Tho clame placed at ach positiont com be a known entity as, for inmace, when we have an ordered array of cosmids, at YAC or DAC clones. This approach offers the advantage of sinmultaneously cloning a gem and mappig it to a chromnosomnal locatio.

Because much libraries ame often ittpated witi gometic maps, identifying a gene's location nay migges possibl associations with gee diseases which have beco, mapped to the same area of that chromosome. Alternatively, the clones may constitute a non-ordered sampling (woith redundancy and overlapping) of a particular library.

The library of discrete clones is duplicated onto a solid-supprt typically a membrane filter. With procise spotting robots, enie libraries can be spread out as high density grids on just a few muurum. Becatioe eah clone in alibray may contain coba pdici of the geno of intereat or a particular geone nay be underropreaezatod within the library, Ioor even hundreds of thousands of individual clnsneed to be screened in order to have good chance of finding positive (complementary) clones. The resul is that a labeled probe for a specific geme wigl interact with only a few clone. out of thousands of possible targets To identify positive clmwe hybridize the hIgh-density gri of dot blots with a labeled probe. Typically, the probe is a small DNA fragmniet or a synthetic oligaucleotide with a known seuawa. The hybriixation step amxplei the poweeful-p n equmue -st and highaffinity of conUplanrnas acleotide srads for ach other.

The goal of the procedlure Is to determine whether the library on our mhrilane contains a sequence oamupmnary to the known sequence In our probe. If it does, this hybridized sequence can be harvested and used to direct our attenion to unknwn parts of the genome flanking the smale hybridized portion.

Visualing the degree to which clones hybridize to the probe is am role of the imagin system. The high density grids as visualized by either by isotopic or ndnisotopic iagying. For isotopically labeled probes, phosphor inuaglug plate teebeology provides the snout convesdooi and accurate signal detection. Neniscopic; delecti may be performed using a sensitive digital comera, and -iesou or fluecent label.. We recommend "h use of our Tundra ultma-ow U&1 imaging system for this purpose. Whae m ur label the resul is an image of a high-density grid. In a typical screening study, this imiage contains manyr thosand of unlabeled points, and a few labeled ones. Analyzing the image to identify positive bits is a seco-nd major Fuinction of the imng system Image analysis involves localization of the dots, quantification of hybridization intensity, and parsing into negatives and positive hits. The most difficult aspect is dot localimtion. Because screening a hlary usually creates only a few visible spots b2the image, it would appear eas to localize the positive targets based on their relative intensity. In this case, we would 'eyeball' the image, and set a density level (a threshold) that selects only hits.

IrnaISag ReseArc InM.

High Density GIrids Selection of hits baued an density value. is mocre appropriate foe library screening than for gane expression stuies (see below), but remains a poor way to saladt bits for the foMowing reasons.

I Variations In background and, labe intensity usually lead to aceme uncertainty in delining a minge density vae for hits. Consdoth case of a probe hybddizirig to bmpoetant clone with week but gemno homology to the probe. This clone would aperas a faint spot.,O o nlsmewhat mere binens than most of the negative.. If the selection of positives was done with a Aipple denity cutoff" th spotiht be missed.

2. A density criterion eliminates I gs I Ware quantification and evaluation. Therefore, we are detecting bits an tbe basis of a subjective judgment (it look. darker or brWgWe), instead of aths basis of quantitative dat. 116s type of procedure is dffficuk to validte a&d subject to bins.

in contrast, our software exploits the regular spacing of the grid to locate and qu*nif all of the grid elements pro to hit identification, This approach (a fixed sampling pcobe strategy) use. specific 'macho spots on the fltmg, to proide quails pohms of reference. Using these spots. the software can iden*f th clone at each and every grid element. Tho software also corrects for loca bakground vulations, illuination differenes, and other error sources.

With the gri elements localized med correted for background, the systm reads dota from all the elemnts.

Every target in the grid Is quantified and eatered into a database for tha grid. Using the data, objective statistical methods can be appried co idenf6 positive hits. For example, we could coapare the intensities of all the grid elemnt, and select those element whose sigal is -re tdan 10 standard deviation units away frosii all mean signal. The wse of statstical methods asures that all positives, even those which may not appear visually distinot, are identified using objective critedia.

Fnally the software provide. a convenin graphic output that shows results in an eay intepreted way, and also provide. full numerical dats that can be exported to your data mnagemen software.

StdyngGone EXqROsin The goal of ges expression msuies is to find genes that are eP esed diferentially acros two or mr canditions. For ezampKe we might compare expression of a given gene in centrol cells from a diseased Eine, and in the same type of cells exposed to a pharmaceutical agn. Traditional mehds for evaluating gene expression are bad on assaying the RNA levels of individual gmnc sequentially or a few at a time. In contast.

the use of high-density grids a&How& the expression of thiomads of San=s can be skuded aionaataneously. The key difference between Noat screening and gae expressicas is tht we are no looking for a few hits thast identify hybridizatio to a specific Imow sequence. Rather, we are evaluating the degree to which a large number of gones are expressed.

Gene expiression is investigated by hybuiuffratiom of complex mixtues of probes to eDNA Ibraries. ThU lirary is spotted in replicate, to generate two or moe daioal high-density grds. Differnt probes are derived independoctly, using RNA isolated from each condition in the study. Eacb, RNA sample is actually a complex nixture of nmany different mItNA molecule., corresponding to the products of dfiesmt gin. The expression level of an individual gene is reflected in the number of RdNA molecle that Itcemributes to this complex mixture. The higher the expression, the more inRNA moolecules in the mixture.

When the comxplex probes are generated. (usig reverse transcriptae), radioactive or biot"inyted izuclootides are ~rpoated.Following hybridization, the activity of a particular gone is proportional to the signal detected at each spo on the high-density grid.

Unlike the library screening case, whore only a few spots hybridize, most spots will hybridize to the mixture of RNAa used in studying geme expression. Consequently identifying the position of all of the grid spots is not Wmaging Research Ine.

High Density Grids diffialL M gIng system uses Information from the ac--I spot matrix to align the sampling grid. The, real problem in otuyinS ga expression is the show volutme of the data. HEUh~ensity gride can contain teos of thousands of spots across multiple membrancs, and the expresion value of each spot mist be retained. It is important that the image analysis softwar be designed to hsandle such large data sets, and compare thesdt sets between conaditionis. Our software does include tbese capabilities.

The isuet of comparing eapression acros conditions is tion-trivial. Each grid will differ feem the others in the dkao'Ae innit of OP04l so we cno simply coqmr sgal strengh acos spec==ns -Our softwar, offer a number of methods by which irrelevant inter'.mcmne variation can be minimized. For example, we can defe taternal standards (a paiticular saw, or the mean expression level of all genesn) wthin each gVi. After normaliing value to an internal standard, inhensity Values can be compared between grids. Another strategy is to compare difference scores between conitions, using the disirbition of difference to localize hits. 7he difference score distribution is free of the influence of general intensity differences between membranes. and has some other benrficial statistical properties. Wsig thoe, and odher methods inchkde in our software, we can pedon varicts ty" of postboc data analyses to arrive at the best method for comparing gSae expression.

The results of a large gene expeasion study constitft, a massive daba set. 7heefre, it is important that data mianagemet funo siummarize and manipulate this data sad. Our elmnal display incationa create easily understood graphbi. dot amuinti doe resuls of complex expresion studies. Our data export funections creado matricea that can. be imported directly into your corporste data structure.

2. Detais of the High Density Grid Software Se gmen a tbn Three main tasks smut be accomplished before a high density grid can be analyzed.

I Identify each grid element (which we will refer to assa target) so dtats location in the maitrix.

is known. Ths identification must alDow for sme variabilit in creating the grid.

2. Quantify dainty (reflecting hkbrlization intensity) at each argeL This may require calibration to density standards, aetier some f(am of backgrmmnd correction.

3. Select'g tao s(~et. lc h re este have been quatified, we con report each endl evay target, to asy saw~ gex presson across spocimens. As an alternative, we can Just aelect a limited mumber targets of interest (oftan. called bids) from the large array.

The first step in the analysis is always target ideriffication. The proces of discriminating targets from bwakgrund is known as segmntation, and can be performed mamially or amwatically.

The simplest form of segmentation is to define each. target. ame at a time. This is done by moving the mouse to place a circle over acla target. With each click of the mouse, the intenity and position of the target are shown- In a typical lvary screwn there we large area of clear sobstrate with only a few, highly visible, dots indicating hits. Clicking on the obvious dota (to generate their location codes) might seem to be an acceptable detection mtd. However, mamml segmentation is niot practical, even with clearly visileo hits. The problem, lies in assigning hits to their proper locations In the SMi. Assigning any dot to Its correct position involves first detecting and aligning all the dots. Of course, you couldl click on all of the dots, mantally, but this would be limg Research Inc.

High Density Grid very teoosa iugin trying to kkaWil 10O0 targets, mnd consider that many labs ane hoping to reach 100,000 or more targets in the near future.

Autmtd Samta Automated segmnaion finds targets withou mmmIi guidance. The quality of your system' atmae segxnmaion flanctiona is critical to its succes. If automated segmentation is imppleented correctly, it can provide a rapid and convaiu way to idenify ad analyze many theusna of targets. If implgoned incorrectly, automated segmentation geneates many false positive. and false ngaptives, thai will requre extensive post-scan editing. At worst, Poor aUSomated Segmentation will yield sprious data.

The standard method for automated segmentation is to assign a specific density range to targets. This proeoms is referred tor uasthelding. Pizads that lie within the deovity rangean classified as targets. -Pixels which do not lie within this rage at claissiffied as backgrod.

A simpl d=4it thrmbe is dom best choice if btas ame of variable uim and shape. and/or do not occupy fixed locan wdble an iage. For exaqale amanocoatsd tissu section labeld with an isotopic In situ probe contaIns targets (dark grainus) thast lie anywhere within the image. We could psocess the image to make the xn-Iy visW de mas drsity tireshold, 50 gray levels), mud thenr detect any pixels darker donn this thbaold a grains.

With gids, in cotuit, targets are of a fixed aim an are regularly arranged (Figren We can take advantage of this regular arragement, to make segmentation more accurate and efficiet. In fact, a density threshold is rarely enough in grid iming. It is defeated by variable local badkgrouods (eg. different denaities at top left and bottom right), med by uncertainty in defining the level of itensity that discriminates a target. our experience is that density tluroslxldiag is a highly subjective and tricky task in most forms of grid analysis.

F~gure 1: SpadlaUy wlebfre trgets and VUi stau. The kwpu ina Me mange at Irfr v~ary in sue, wad awe qsaoed ingulary. They woedbe edcdbeeuag a daity £uhkoI The targets ka she image at right are organitd aagrid. They are deetdby pkwchg afixrdprobe aver e-di targ et oado m 0 0 en 0000 6660.0 000060 0006 @000 000000 *000 @900 00 000000 0000 e.g. @006gg 0000 Segmentation W46h Fixed Saannina Prahea Fortamatly, grids can be anayzesd quite objectvl, aid much more efficiently than irregular targets. We know the spacing, nter, ad diamete of the targets, so these is no need to use density differences as the only tool in target identification. Instead, AIS/MCU) use a fixed sampling probe strategy.

bmging Resarci Inw.

High Density Grids In using fixed probes, the S&i definition dinecs the sopew~ttion process. Fixed samplin probes am placed (inztnialically) over each locedn dite VAi. T1o difficuty with this is tint real specimnsn ane rarely pelfectly space. We Will find that spottin robots wander, zoanifolds vary, or some other factor has lead to variations in the grid. Therefore. segmntoation with the fixed probe strutep~ roquires two step. First, the gri is propagaed to its Hd locatiocs wsing dot locations firm the grid definition. 7hen, th system applins some rather sophisticated algorithm to perform sistoinsli alignmen of the grid with the actual targets (Figure 2).

Figur 2: A high deassky grid cauabuing 1536 dircre dog blots. As kftier ase Mhe mage wkhosa *amplng probes. The img at right JTIIOY she saiypfbg probes in place over ech of Mhe dot blots. We placd she probes, and aligned them %a thhe dots, ausoaicaly.

Mme an a nsdrb of advantages to using the fixed sampling probe mtdas opposed to scanning for targets thAt exceed a density thresold.

1. We detect and quantify every target, including these that contain dat at background ot arbakron lovea.

2. Detecting every target allows us to use objective statistical procedures (insteadl of sobjective density thresholds) to define bits.

3. We an follow a given target acres. any Umber of experimental conditions, incling those that do not activae the target in an obvious way. This is particularly important in gene expression studies.

Lmsimg Reseb I=n.

44 High Deal Grids Cnstueting and AIfirnft~ the rg Mie grid is constructed using spacing parameters (dlistancs between dots and between groups of daba) enterd &I the knyboard With the spacing defined, the origin of the grid is indicated end the eWWx grid then grows adt from the origin.

it like that soame of the grid camping probes will not lie directly aver their assigned targets. To overcom dim. problem, utaulic alignment can be performed. During this prooss, a fuzzy logic algorithm places Oach diacesa grid element over the location that best fits th a-t-a image data. Nf further ligomt b required, any grid element or group of ebeneas can be dragged halo position using the eae Anchars Not all specimens have visible data which can be used for fine alignmet In library screening, it is coman to hav a vary large grid with any a few data visible, In this cao*, that the fign allX~w algorithm lack data to work with valss Imown rofesunce an mped od th grid. These referencie points, known as wmso.*, contain an obvious target and ame used to guide the automated alignmet. For example, our spelling robot cretes a grid contan 61"4 discrete dots. We -a&l place easilly detected amut of label into, dots st the fbir corners, and at a few mer strategic locations. When we credo the grid AISIMCID will perform autonatic alignment, using the anchor points as referenoes.

Deaw adSogein adas DetaiI mode Gene expre~ion In detail mode, values are obtained for each and every tare wiffin the grid. Detail mae give, a comlt description of the targets, and allows direct cenIi Pi- between up to four grids. However, the data tables often cain my dxxxxod aflmit=V and this can slow yaour system down. In gonsral, there is no probnlemi Alg with a few way. of 5,000 datb each. However, muinaging very large ermobem of dots 30,000 or no-e) can lmi throul;%t Scanning the grid is not the major problem Rather, just recalculating a data table containing SO,000 numbers Ulme some time, If data mangement is proving to be too slow, consider using scenn mode.

Detail mode is most usefu in geme expression seies, and In samie case. of screening by hybridizain Both these applicaiam; involve move than smpl4e discriminations between obvious hits and background. Therefore.

it is useful to have knowledge of hew every point in the grid is reacting.

Library Screening Mnda Screening mode analyzes one grid at a time, and displays only the hits from that gid. Use screening mode to scrama unknowns against large libraries, when only a small proprtion of the grid elements contain label.

Im=2108 Rugarcb in@.

High Dewsily Grids Screening uses a similar logical structr to da i ode. That is, all of the data are scanned and quantified prior to statistical segmentation. The difference is that we want to identify a few labeled hits from a relatively clean q=6c' Data fiom the -nlabeled grW positions need no be reported, as they just alow down system response.

After screening segmntation, the misse are discarded and only the hits are retained. Because only the hits are displayed in the data tables, data management Is much faster.

S&nman ededlan md The difference between the daedi and screening modles lies in what happens after the grids have been quantified.

Detail mode reports each target value, so that you can perform exploratory data. analysis part hoc. In contrast, screening mode reports only valid its.. It smts a cutoff point in the distribution of targets S standard deviations sbove the mean) to select a limited number of hits for further attention. Only those targets that lie beyond the cutoff point am shown in theda" lists.

Im both the detail and screening modes, an initia setteentation a"e uses fied samplig probes and automted alignment to generate data from each and every target in the grid. Once dhe targets hav been segmnted and their densities or volumes quantified, various procedures are used to select targets of interest.

We use the term sladitirsep agentarimr to describe the prowes by which we select a limited mambe of hits firom the entire population (screening mode), or track the moat izqportant diffierences in gene expression fromn ane specimen to another (detail mode). We can define statistical segmentation as Mie qp~eadon oAuaid rwng to the -w-muad definidon of sSex.

Scrteg Consider an experimnti In which we screen a cDNA library with probes from thee cill lines. These library is spatted an three ideaticsl mnisa a~KprI Ise) and amo cell line is appied to each membrane.

Following th hybridization proceduores, we esie sulaadiograph. of the three nmnbwina and determine that all the cell linos show intense hybridization with two of the clones in the library. However, the third cell line &ain C) also appears to hybridize (somwhat less intesely) with an additional clone (Figure it is possible that line C is different than doe other two, but how am we to know? imaging skemar Inz.

High Density Grids Ron 3? 4=00ly varialie Arget, ad gSdIngegr. The &WC&e h the imag at 14t wy bt ate, md ar apaom d irrwsdarty. They would be detected by ceaag a dauby reahoki The "aget in the *map at rghtare organized as a :rid They are dcd by pchg a e probe owr each trge locoriLm Flgt 3 SarvemM a cDNA ibiy ufI pro be frfans m CaU l st Alldm Ca lbw hbrlize to Am qthe dlangi. Cel ueC we rto lyidbal to a ctrd de. Todetennie idwhr sakl aa redl effect.

w need go nopoar thie irauty of kiridftAaa at the third clie, acrors all dre membrue.

Several factors could affect our decision about the third clone in line C.

1. The third eoms may hybridize intensely to lne C, but also hybridizes less intensely to cel lines A and B.

2. The third clone hybridiz exclusively to line C, but the hybridiraion sigal is weak.

Imin Rmauft In.

High Density Grid 3. The preparation way contain -ny nmbe of ambiguous spats that complicate lierpretation of the third clone.

in these came, we might not have a simple, black-white discrimnato. Rather, we need to make a jugment call regarding the third done. bs the signal intense enough to be considered a bit, or not? Simply setting a density threshold that 'Looks good' is rather arbitrary.

To help you make an objectivo decision, tdo system finiis and calouastes intnities for all of the targets. The data Wnlai those targets that are at backgrumd intenity, those that ure somewhat sbove background but stio fint, and the more intense praoeclone interactions which are obvious to our eyes.

once we have the complete data met, we can apply variance-based statistical procedures to determine the probability that each obserevid intensity is a hiL A variance-based discrimination dafar from a simple threshold in that it doesn't Just detect hits on the basis of density. Rather, it compares the density of each target to the distrlbjtion characteristics of all the targets. Hits are those targets which am most unlik the other. The logic is that a target that is very different from the rest is probably one that should be investigated. This "yp of variancebsd threshold is both easily validated, (using statistics), and easily applied.

Detail Mode and Genn Erpression Statistical segmentation is usefu In trackng gene expression. In these studies, we are usually comparing the inensity of a given clone-probe irteraction across mualtiple rids. Ineresting interactions ae those which anu enhanced or inhibited by an experimental maniplation (eg. applying a potenial pharmaeutical compound).

The steps in analyzing this type of study by statistical segmeuntatioo are as follows.

I1. Quantif the level of gene expression by measuring densities in all the targets, across a nmber of specimens. Each specimen corresponds to a different condition in the experiment.

2.Calculate differuwve scome for every pair of targets. Now, we have a distribution of differences for each tullcDone, across all conditions of the experiment.

3. Calculate the mean and the SD of the distribution of iffereance scores.

4. The statistical segmentation step. Select targets showing difference scores lying some inmer of SD units above or below the nmn. 7these targets could be classified as hits. Being a variancebased discriuon, tdis 1 oia1 of hit selection Is free of inter-grid differences in overall dcnsity.

Display the targets in a graphic format, called an elemental display (see below).

SUMMay Statistical segmetation is the application of statistical reasoning to the analysis of grid data. Its advantages are that it provides objective criteria for discrimiwnatn targets, and that it allows us to compare data across specimens. This latter advantage is present because variance-based analyses nimize the influences of irrlevant influences on our analyses processing offees, exposure times).

In sm, statistical segmato is am objective procedure for extractig valid hits from large bodies of grid data.

Major advantages of statistical segmntation are that: a) we can docuiot the exact: criteria used to classify hits-, b) we can minimize the effects of extraneous processing vauisbles, to simplify comparison across gris.

vmging Resaebt Iuc.

48 High Density Grids Quick Date Communication:, The ElementlDsp~ Grid analysis can produce a laot a imenical dat points. These data can be presented in any way you see fit.

For example, export doe dat matrix to Excel, rank by hybridization inenity, and use the ranked data table to communicate your results.

Another option is to create an elmautal 4lrplay. This is a graphic image which uses various ways to flag a limitedf numiber of interest*n targets. In a screening study, for example, istruct the elemental display to flag any dote lying more *s six SD units above the mea dot density. The elemental display directs your attention to the most relevant data points from the thousands produced in a typcal run (Figure 4).

of course, with a clean memibrmne and a binaiy labe (on or off), The elasmetal display may not be required We can just look at the originalimuge toram which dots are hits. However, the elemntal display becomes very usefu when "th speciman Is not ciam, or if there.i a gruldia of I"bo Intesiies. In these case, our eyes tend to be uncertain and it is easier to let the computer show the hits in oor-eoded &a eaily appreciated foan.

The elemental display is particulal useu in detai mode, when comparing across two or more specimens. The uwan target inteesity, background level, and other factors will tend to viuy from grid to grid. Therefore, it is difficult to make visual Judgments shout the relative intensity of a particular grid point across specimens. In contrast, the elemental display clarly shows alterations; across gids. Any Outcome of statistical segmentation can be summarized by a single eletecal display (Figure Effetsofmucod Mradon Any farm of segmentation will be affected by background. Ideally, the level of backgrnound will be both low and constant. Low background improves sensitivity. Constant background simplifies target detection.

Unfortunatel, backgrounds are rarely low and cnstant. Thereforc, AMSMCIID have a variety of background corrections available.

None No background corrcton Selected dots Certain dots (eg. top left of each primary) are specified to contain beckgrond Surrounding pixels The system looks at the outer boundary of each dot, and calculates a background for that dot from than Surrounding User selections Yo efime am or mo aw of the apecane. which are used for bakgron correction.

image proc S Some of the image IPcesing functions am very effective in uunoving background dWn vaties ficmn pon to pon in the WNmage.

There are weaknesses with all of the orrection methods just described. The best option is to mi-nmz background variation In the original xyweim.L Summary: Femaf ur f The High DenRit GWi Sh* y pe

I

Larg pubd analyzed quikly and automaicaly Any umbter of targt ca lie witim a grid. Any meditum can be used as a matrix for to grid mebrnes.

gels, microtiter plates).

2. Automatic propagation, alignment, and background eorrection.

Th. grid is created (propatgated, automatically, using spacing@ entered by the user. Following propagation. the grid is automatically aligned to the best dat locations within the image and background is removed.

Imainfs Reseah In.

High Density Grids 3. Detai and Suning Moes On evay 2-sn- all the S grd el ame read. However, WO can choose to retain and display all of the data, or only those data which are defined as hits. Keeping all of the data allows very extensive, pox, hoc analyses.

Keeping only the hits can save timeo.

4. Stentrcil aegmemwoa Hits a- defined. ffitomaiCally, by locat them within the distribtion of all grid elements. Ths procedure is objective and easily doenmmied.

S. Cam tqI Specum se~ati gauj dmjjO die Wbm Of proesing factor, and allows comparison across specimens.

For exampl, we can look for hncrese or decreases in hybridization intensity, acros one conmrol and thre experimental membranes, each containing 5,000 dot blots.

6. Elementa disqiys These simplified graphi displys can show one or more amy. of dots reduced to a few color-coded hits.

[=5111 R"aerct Inc.

High Density Ga"d References Eggene, Hogan, R"da R.K, LAzinra, L, BnfichI Hollis, et al, A microchip for quanfitative detion of moleculus ilibing luniamewi and radioisotope TePotr group, Biotluziquer 17:316- 525(1994).

Khropko, Lysov, Miarlin, lvanov. LB., Yezuhov. Vasilooko, S.L, Flcrentev, V.L and Mizodbv, A.D. A mthd for DNA sequencing by hybridir~ion with oliganucleowie matrix, DNA Sequence Joumna qfDN4 Sequencing and Myppkg 1:375-38 (1991).

Lanture, Bealie, X.L, Burke, B. Egger. UbuidI DJ., Fowler, Hollia, Koslckj, Reich, R.L, Smith, S.L, Vurma, R.S. and Hogan, M. Direct detection of nucleic acid hidztion an d the sce of a charge coupled device, Nuclic Acds Rawn* 22:2121-2125 1994.

JUpchutz, LI., Morrs. Caoe. Hubbell, L, Kozal, MI., Shah, Shax, Yang, L and Fodor, S.p.A. Using oligonucleotide probe arrays to &ccess genetic diversity, Bioedu~qu, 19.442-447 (1995).

Maakos, U. and Southern E.M, Parale analysis of ollgodboxyribmoceotide (oligoaclaotide) interactions.

L Analysis of facor. knficain oligcmnzcleatidri &dpe formatio, Nudeic Acids Resrearc 20:167S- 1679 (199).

Mason, LBS., Ramnps, I.B. and Coassin, PJ. Biopolymur, synthesis on polypropylene spports. 1.

Ohigonucleotidea, AnhlBcjsmir"j217:306-310(1994).

Pearson, D.H. and Tomiccl, R.J. Nanochannel glass replica memubrines, Science 270:68.69 (1995) Pease, Belas, Sullivan. EJ., Cronim, Holmes, C.P. and Fodor, S.P.A. U&gh-generated oligonucleotie arrys for rapid DNA sequce analysis, Proceedings of d National Acudernyof Scence, USA, 91:5022-5026 (1994).

Saik, iLK., Walsh, Leeauaoa, C.H. and Erlich, H.A. Genetic analysis of amplified DNA with immobiliZed seqoew"peCiflc oligoinucleotic pobes, Pr'oceedings of lee Nationl Academy of Scinces USA, 96.623".234 (IM9.

Schena, Shsion. Davis, R.W. and Brown, P.O. Quantitative monitorig of sane expreasion paterns with a complemntary DNA microarray, Scince 270:467-470 (1995).

&xztben, E.M. DNA chp: analymig sequence by hybridization to oligommcleotides an a large scale, Trenk kn Genetdcs 12:110-115 (1996).

Scudxmcn, ELM., Maxims, U. and Elder, l.L Analying and comparing nucleic, acid seomes by hybridization to arrays of dolgzzwetides: Evaluation using eaperimetld model, ermica 13:10081017 (1992).

-msis Imureb Inc.

Claims (5)

1. In an area digital imaging system for assays, a method for extracting targets on a specimen containing an array of targets that may not be arranged in perfect regularity, including the steps of: defining a matrix of nominal target locations including a probe template of predefined, two-dimensional size and shape at each of a plurality of fixed, predefined grid points on the specimen; and determining the most probable location of the probe template corresponding to a target by: making use of an image of the specimen, deriving a most likely location for a selected target based upon pixel intensity in the image in the vicinity of a nominal target location corresponding to the selected target; and using a confidence value indicative of reliability of the most likely location as a weighting factor in shifting the location of the probe template from the nominal target location towards the most likely location for the selected target.

2. The method of claim 1, wherein said determining step is performed iteratively for each target.

3. The method of claim 1 or 2, wherein the image used in said deriving step is produced by generating a primary image of the specimen showing to best advantage the effect of interest, generating a secondary image which shows minimally the effect of interest and combining the secondary image with the primary image.

4. The method of claim 1 or 2, wherein the specimen is provided with predefined anchor points, the matrix being initially oriented relative to the actual target locations in the specimen by placing specific probe templates over one of: the anchor points; and those target points which are clearly definable. r P:\WPDOCS\AMfl\Dp d\I22370 l.do.- I Apil 2004 -52- The method of claim 1 wherein the confidence value for a target is determined by the detectability of the target.

6. In an imaging system for assays, a method for imaging assays as substantially hereinbefore described. DATED this 2nd day of April, 2004 IMAGING RESEARCH, INC. By Their Patent Attorneys DAVIES COLLISON CAVE