There are a number of known substrates for holding and analysing arrays of experimental samples. For example, microarrays of genomic DNA clones, synthetic oligonucleotides, or cDNA clones may be produced by piezoelectric printing, in situ synthesis using photolithographic methodologies or conventional ring and pin gridding onto flat planar substrates.

Such arrays, which are commonly known as microarrays, may be used for gene expression analysis, genotyping, gene discovery etc.

A well known example of an oligonucleotide-based array substrate is the GeneChip, produced by Affymetrix. The GeneChip is intended for the study of gene expression.

Presently, GeneChip arrays comprise the U95 series (five separate chips containing over 60,000 transcripts) and the U133 series (two separate chips comprising over 39,000 transcripts).

More specifically, GeneChip arrays contain oligonucleotide probes synthesised in situ on the surface of the supporting matrix (glass or wafer). Probes on a GeneChip array are referred to as features. The oligonucleotide features are designed to bind specific sequences within a labelled target either perfectly (i. e. the nucleotides within an oligonucleotide feature bear a 100% identity to target) or imperfectly (i. e. the central nucleotide within an oligonucleotide feature is mutated so that target does not bind). Perfect match and mis-match features are arrayed in adjacent pairs throughout the array. 16 feature pairs are used to represent different sections of each target in the U95

series. 11 feature pairs are used to represent each target within the U133 series. This redundancy ensures that a large proportion of any transcript is represented, and that target detection is specific. Algorithms (e. g. based on the proportion of the features pairs of a set reporting positive) are used to determine whether a transcript is present in a sample.

A DNA microarray experiment typically consists of hybridising a fluorescently labelled probe to the immobilised target DNA arrayed on the substrate, scanning the microarray to measure the fluorescence signal from each probe that was successfully hybridised to a target, and acquiring and analysing the fluorescence pattern on the substrate to extract information about the probe from the pattern. The acquisition and analysis steps may be performed e. g. by a suitably programmed computer controlling a dedicated array scanner that may comprise either a confocal imaging apparatus or a CCD camera imaging apparatus with appropriate narrow band pass fluorescent filters for discrimination of the emitted signal, as is well known in the art.

The number of registers in the microarray ultimately limits the amount of information that can be obtained from each experiment. For example, the Affymetrix GeneChip U133 has approximately 1 x 106 spots on two 1.64 cm2 slides, providing a spot density of 295,122 spots/cm2. However, this density comes at a price of operational flexibility; the oligonucleotide features of the U133 microarray are printed (using in situ synthesis and immobilisation techniques) on the substrate by the manufacturer, which prevents the user from customising the microarray for a particular application. Also the oligonucleotides are limited in length by the manufacturing

approach employed, meaning that more features have to be employed per transcript to maintain sensitivity.

More generally, if the spot density is increased the inter-spot distance is reduced, which can lead to problems of spot cross- contamination and signal interference between neighbouring spots. Decreasing the spot size can ameliorate these problems, but at the cost of a reduction in the measured signal intensity. Thus there are limitations on the amount of information that can be obtained from conventional spotted microarrays.

Similar considerations apply to substrates for holding arrays of e. g. protein and chemical samples.

An aim of the present invention is to provide an array substrate which provides a high sample density and hence a high information content. A further aim is to provide an array substrate which can support non-immobilised samples and/or which allows the user to select and locate the samples on the substrate.

Accordingly, in a first aspect the present invention provides a substrate for holding a microarray of experimental samples, the substrate comprising an array surface having a plurality of wells for holding respective experimental samples of the microarray, wherein the bottom of each well is at one of a plurality of levels, the bottoms of nearest-neighbour wells being at different levels.

Preferably, nearest-neighbour wells are determined by the distance, measured parallel to the array surface, between respective well bottoms, i. e. so that the nearest-neighbour

wells of a first well are those wells having well bottoms which are closest to the well bottom of the first well.

An advantage of the substrate is that the array surface can employ high well densities because signal overlap or interference between samples in nearest-neighbour wells can be reduced or eliminated, e. g. when the signals produced by the samples are measured using a confocal microscope (e. g. example as described in US patent no. 3,013, 467). Confocal microscopes image light from a single object plane, essentially eliminating out-of-plane contributions to the final image. Thus, for example, the samples held in the wells of an array surface having well bottoms at only two levels may be imaged by such a microscope in two imaging steps. In the first step, the object plane of the microscope is positioned at or near the first level, and all the samples held in the wells with bottoms at this level are imaged. In the second imaging step, samples held in wells with bottoms at the second level are imaged by shifting the object plane of the microscope to the second level. However, at each imaging step nearest-neighbour wells, being at different levels, can be located outside the object plane. Thus, higher well densities can be employed, as signal overlap from samples in nearest-neighbour wells can be substantially avoided. Effectively, the wells increase the spatial efficiency of the substrate compared to conventional flat substrates.

In an array surface having well bottoms at only two levels, the second wells are the nearest-neighbour wells for the first wells, and vice versa. However, the array surface may have wells with bottoms at three or more different levels. This allows the number of wells between each pair of wells having bottoms at the same level to be increased so that even higher

well densities can be employed, but at the cost of requiring more imaging steps to image all the samples of the array surface.

Yet another advantage is that samples contained in wells can, in general, be formed from bigger molecules than samples which are immobilised on flat surfaces using only in situ synthesis technology (e. g. of the type used in the production of the Affymetrix GeneChip U133). Thus, for example, an oligonucleotide microarray held on the substrate can be formed from longer chain oligonucleotide samples (e. g. 60mer oligonucleotides), with the result that the same amount of information can be provided by fewer wells so that the overall information content of the microarray is increased.

By holding each experimental sample in a well, the risk of sample spreading and cross contamination is reduced so that each sample can be better isolated from neighbouring samples, even when the samples are not immobilised in the wells. This facilitates selection and loading of samples onto the microarray by the user. Furthermore, as the wells form effective barriers to sample migration, compared to conventional spotted microarrays formed on flat substrates, the size of each sample can be maintained at a relatively high level as the sample density is increased. Thus, more information can be obtained from a substrate of a given size.

In one embodiment, the well bottom levels are spaced apart by at least the approximate minimum height of the in-focus object plane imagable by a confocal microscope. In this way signal overlap between nearest-neighbour wells can be substantially avoided. Preferably the well bottom levels are spaced apart by at least 0.5 pm (and more preferably at least 1,5, 10 or 20 or

50 um). Preferably the well bottom levels are spaced apart by at most 500 um (and more preferably at most 200 or 100 pm) to avoid excessive microscope focus changes.

Preferably, the array surface covers an area of no more than about 4.84 cm2 (i. e. the area of a standard 2.2 x 2.2 cm microscope cover slip). More preferably, the array surface covers an area of no more than about 1.64 cm2.

The substrate may have a well density of at least 40,000 wells per cm2 (and preferably at least 100,000 or 400,000 wells per cm2). With such densities, we believe it should be possible for all human cDNAs or other expression/splice variants to be provided on a single substrate having a 1.64 cm2 array surface.

This is advantageous because single substrate experiments are generally cheaper to perform (most experimental expense is associated with the preparation of each substrate) and the results can be more quickly processed. Also, single substrate experiments generally require smaller total amounts of samples, which is particularly advantageous when the samples are rare/difficult-to-obtain/expensive. Furthermore control of minor variations in conditions is much easier when only one substrate is involved.

In other embodiments, a lower well density may be appropriate, or even preferable. In such embodiments, the well density may be between 1,000 and 10,000 wells per cm2. The density may be chosen depending on the desired application of the array. The materials from which the array is formed may be chosen depending on the desired application and/or the desired well density.

The wells may be tapered in shape. For example the wells may be substantially conical, frustoconical, pyramidal or frustopyramidal. Alternatively the well may be substantially cylindrical with a flat or concave bottom. Such shapes should help to localise and increase the signal intensities obtained from samples held in the wells. Increased signal intensities allow the amount of sample loading to be reduced, thereby decreasing the likelihood of sample spreading.

In a further embodiment, the array surface further has a plurality of microterraces, a respective well being formed in each microterrace. By holding each experimental sample on a respective microterrace as well as in a well, the risk of sample spreading and cross contamination is further reduced.

Typically, each microterrace has a level that is higher than the level of the bottom of the well formed in the adjacent microterrace. In this way nearest-neighbour microterraces as well as the bottoms of nearest-neighbour wells can be at different levels, which facilitates sample imaging.

The microterraces may be triangular, square, rectangular, hexagonal etc. in shape. Such shapes can be tessellated allowing the microterraces to completely cover the array surface, thereby making most efficient use of available space.

In an array surface having a regular array of triangular, square or rectangular microterraces, each microterrace should be at one of two or more levels to avoid nearest-neighbour microterraces being at the same level. For the same reason, in an array surface having a regular array of hexagonal microterraces, each microterrace should be at one of three or more levels.

In use, each sample will have an upper surface which is spaced from the bottom of the well in which the sample is held.

Preferably, for each well having a bottom level which is lower than that of a nearest-neighbour, this upper surface is at least 0.5 um lower (and more preferably at least 1,5, 10,20 or 50 pm lower) than the well bottom of the nearest-neighbour.

Overfilling the well beyond this is undesirable because it can result in the upper portion of the sample held in the well being at the same level as the lower portion of the sample held in the nearest-neighbour well, whereby both samples could be simultaneously imaged by a confocal microscope leading to signal overlap/interference. In one embodiment, the level of this upper surface is 10 to 20 um lower than the level of the well bottom of the nearest-neighbour.

However, in a substrate in which the wells are formed in respective microterraces, each microterrace essentially defines the upper limit to which the respective well can be filled.

Thus, preferably, for each well having a bottom level which is lower than that of a nearest-neighbour, the level of the microterrace in which the well is formed is at least 0.5 um lower (and more preferably at least 1,5, 10,20 or 50 um lower) than the well bottom of the nearest-neighbour. In one embodiment, the level of the microterrace in which the well is formed is 10 to 20 um lower than the level of the well bottom of the nearest-neighbour.

In use, each sample held in a well may have a depth of no more than 15 um (and preferably no more than 10 or 5 um).

Furthermore, in use, each sample may have a volume of no more than 5 pl (and preferably no more than 4,3, 2 or 1 pl).

To facilitate this, in respect of a substrate in which the wells are formed in respective microterraces, preferably each well has a depth of no more than 15 um (and more preferably no more than 10 or 5 pm), and/or each well has a volume of no more than 5 pl (and more preferably no more than 4,3, 2 or 1 pl).

In other embodiments, in use, the samples and therefore the wells may have larger depths and/or volumes. For example, the wells may have depths of 20-40 um or greater, and/or volumes of 10-100 pl, particularly 50-100 pl. In some embodiments, the wells and/or samples may have volumes of up to 800 pl, with corresponding increases in well depth. These dimensions may be preferred in embodiments with lower well densities, these embodiments potentially having microterraces of greater area.

Preferably, the substrate is adapted for holding an array of biological or chemical samples. More preferably the substrate is adapted for holding an array of nucleic acid samples.

In one embodiment, each well holds a corresponding experimental sample.

In a further aspect, the present invention provides for the use of the substrate according to the first aspect of the invention for holding a microarray of experimental samples. The samples held by the substrate may then be used to perform e. g. expression analysis.

In a further aspect, the present invention provides an apparatus for analysing experimental samples, the system comprising: a substrate according to the first aspect of the invention, an imaging device (such as a confocal microscope, e. g. as described in US patent no. 3,013, 467, or a CCD camera,

e. g. as described in US patent no. 6,271, 042) arranged to image the array surface of the substrate, and a computer system which is configured to analyse the image received by the imaging device.

In a further aspect, the present invention provides a method of analysing a microarray of experimental samples, the method comprising: providing a substrate according to the first aspect of the invention, the substrate holding the experimental samples, providing a confocal microscope, imaging successive subsets of the experimental samples by changing the focal plane of the microscope, and analysing the images obtained by the microscope.

The present invention will now be described in relation to specific embodiments and with reference to the following figures in which: Fig. 1 shows schematically a perspective view of a portion of the array surface of a substrate, Fig. 2 shows schematically a plan view of a portion of the array surface of the substrate of Fig. 1, Fig. 3a shows a more detailed perspective view of a small number of the microterraces from the array surface of Figs. 1 and 2, and Fig. 3b shows a cross sectional view along plane I-I of Fig. 3a, Fig. 4a shows a plan view of a microterrace with a pyramid- shaped well, and Fig. 4b shows a cross section through neighbouring such microterraces,

Fig. 5 shows a cross-section through three adjacent wells of an alternative substrate, Figs. 6a-f show confocal images of a further substrate: Figs. 6a and d respectively showing reflection and fluorescence images at a focal depth of 13 um ; Figs. 6b and e respectively showing reflection and fluorescence images at a focal depth of 34 um ; and Figs. 6c and f respectively showing reflection and fluorescence images at a focal depth of 51 pm, Fig. 7 shows a scanning electron microscope picture of the array portion of a polycarbonate substrate, Fig. 8a shows a plan view of a microterrace with a cylindrical shaped well, and Fig. 8b shows a cross section through neighbouring microterraces, and Figs. 9a-f show confocal images of a polycarbonate substrate: Figs. 9a and d respectively showing reflection and fluorescence images at a focal depth of 64 pm ; Figs. 9b and e respectively showing reflection and fluorescence images at a focal depth of 118 um ; and Figs. 9c and f respectively showing reflection and fluorescence images at a focal depth of 164 um.

Figs. 1 and 2 respectively show perspective and plan schematic views of a portion of the array surface 2 of a substrate 1.

The array surface is made up of a plurality of square microterraces 3. The microterraces are arranged so that nearest-neighbour terraces share sides when viewed (as in Fig.

2) in the direction perpendicular to the array surface. A conical sample well is formed in each microterrace, although the wells are not shown in Figs. 1 and 2.

Each microterrace is at one of three (upper, middle and lower) levels. In Figs. 1 and 2 the levels are indicated by respective degrees of grey shading. Parallel diagonal lines of corner-sharing, same-level microterraces extend across the array surface. In this way the microterraces cover the entire array surface while nearest-neighbour microterraces having the same level are avoided.

Fig. 3a shows in more detail a perspective view of a small number of the microterraces from the array surface of Figs. 1 and 2, and Fig. 3b shows a cross sectional view along plane I-I of Fig. 3a. The conical sample well 4 formed in each microterrace is shown in Figs. 3a and b. The sample wells all have the same depth, so that the three microterrace levels provide three corresponding well bottom levels and the bottom levels of nearest-neighbour wells are at different levels.

Fig. 4a shows a plan view of a microterrace with an alternatively-shaped well; in this case an inverted square- based pyramid. The square base of the pyramid-shaped well has sides which are 15 um long and the depth of the well is also 15 pm, giving a well volume of 1125 pm3 or 1.13 pl. The well is set in a square microterrace with sides of 50 um. Fig. 4b shows a cross section through neighbouring microterraces of the type shown in Fig. 4a. This shows that the well spacing in the direction perpendicular to the array surface (i. e. the distance between the bottom of one well and top of a neighbouring lower well) is 10 um, which is significantly greater than the approximately 0.5 um practical minimum height of the in-focus object plane imagable by a conventional confocal microscope with a fluorescence objective. This helps to reduce or eliminate signal overlap between neighbouring wells.

With such dimensions it is possible to provide about 67, 000 wells on a 1. 64 cm2 array surface at a spot (i. e. well) density of 40,000 spots per cm2. When samples are loaded in the wells, the microterraces and wells help to reduce or eliminate sample spreading (even if the samples are not immobilised) and signal interference.

If a higher sample density is required, the size of the microterraces can be reduced until the base of each pyramid occupies an entire terrace. The square microterraces would then have only 15 pm sides. With this arrangement the wells and the height difference between neighbouring terraces still operate to reduce or prevent sample migration and cross- contamination, and a density of about 450,000 spots per cm2 can be achieved. At this density it should be possible to accommodate the entire known human transcriptome on a 1. 64 cm2 array surface at a sensitivity of 16 features (i. e. samples) per transcript, which is the same sensitivity level provided by the Affymetrix GeneChip U95. Indeed, if each sample comprised a 60mer oligonucleotide we could drop to 4 samples per transcript without compromising sensitivity and additionally sample the whole human genome on the same array surface at a resolution (i. e. sampling frequency along the genome) of 20 kilobases. Presently the best resolution that can be achieved with conventional microarray technology is 1-1.5 megabases.

Prototype substrates of the type described above in relation to Figs. 4a and b may be conveniently prepared from silicon wafers Si (100) using conventional photolithography and wet and dry etching of the wafer.

An example preparation protocol is as follows. A photoresist mask is applied to the cleaned wafer. The mask protects the areas of the wafer corresponding to the microterraces of the upper and middle levels. The wafer is then SF6/02 deep reactive ion etched (i. e. plasma dry etched) to form the microterraces of the lower level. The first mask is then stripped, a second mask is applied to the wafer to protect the areas of the wafer corresponding to the microterraces of the upper and lower levels, and dry etching is again carried out to form the microterraces of the middle level. The microterraces of the upper levels emerge as unetched surfaces when the middle and lower level microterraces are formed.

The second mask is then stripped and a third mask applied to the wafer, openings in the third mask corresponding to the positions of the wells. This time the wafer is KOH anisotropically etched (oxide wet etched). The KOH etching produces an inverted square-based pyramid shaped well or pit in the centre of each microterrace. Because the KOH anisotropic etch follows the Si <111> direction, the sides of the pit are bounded by Si {111) planes and hence form angles of 54. 47° with the wafer surface. Finally the third mask is stripped and the wafer cleaned.

Preferably, the third mask is a"hard mask"formed by SiN deposition and dry etching. For example, after the second mask is stripped, a SiN layer can be formed on the wafer by plasma enhanced chemical vapour deposition. A photoresist mask is applied to this layer, the photoresist mask having openings corresponding to the positions of the wells. Plasma dry etching is used to form openings in the SiN layer at the same positions as the openings in the photoresist mask. The photoresist mask is then stripped. Oxide wet etching can now

be performed as described above, the SiN layer forming a hard mask with openings corresponding to the positions of the wells.

Finally, the SiN hard mask is removed using buffered hydrofluoric acid.

A cross-section through three adjacent wells 14 of an alternative substrate is shown in Fig. 5. In this case the substrate is simpler, having a flat (i. e. non-terraced) array surface 12. However, the wells formed in the array surface have different depths so that each has a bottom at one of three levels. Furthermore, the wells are distributed over the array surface in a similar way to the wells of the substrate of Figs.

1 to 3 so that nearest-neighbour wells have well bottoms at different levels. In use, each well should not be filled beyond the level of the bottom of a shallower nearest-neighbour well.

The substrate of Fig. 5 may be prepared using the following protocol. A photoresist mask is applied to a cleaned Si (100) wafer. The mask has large, medium-sized and small openings respectively corresponding to the positions of the deep, medium depth and shallow wells. The wafer is KOH anisotropically etched to produce an inverted square-based pyramid shaped pit at each opening, the size of the pit depending on the size of the opening. The mask is then stripped and the wafer cleaned.

A further non-terraced substrate was prepared with the protocol used to form the substrate of Fig. 5. The substrate contained 50,000 wells with the wells being either small (A), medium (B) or large (C) in size. This produced three different levels for the well bottoms. The bottoms of A wells were at a depth of 13 pm, the bottoms of B wells were at a depth of 34 um, and the bottoms of C wells were at a depth of 51 um.

With this substrate we successfully demonstrated how confocal microscopy can be used to selectively view successive sets of samples held in the wells. 1 um, red fluorescent (580/605 nm), FluoSphere polystyrene microspheres from Molecular Probes were loaded into the wells by suspending the microspheres in ultra pure water at a concentration of 107 beads per ml and centrifuging at 1000 rpm for 4 minutes. Unloaded microspheres were removed from the substrate by gentle washing with ultra pure water. The substrate was then washed in successive ethanol-ultra pure water solutions (ending in absolute ethanol) and was left to dry by evaporation. Although some of the wells did not receive microspheres, sufficient numbers did to proceed with the demonstration.

The substrate was viewed using an LSM510 Zeiss confocal microscope. Figs. 6a-c show confocal reflection images of the substrate at depths of 13,34 and 51 pm. These reveal the substrate surfaces at the respective well bottoms. The positions of wells containing microspheres are indicated by dashed boxes. Fig. 6a also illustrates that in this substrate nearest-neighbour wells were determined by the distance, measured parallel to the array surface, between respective well bottoms (and not by the distance of closest approach of well edges).

Figs. 6d-f show corresponding confocal fluorescence images of the substrate at depths of 13,34 and 51 um. Clearly, at each focal depth only the microspheres located at that depth are imaged, demonstrating that samples held in the wells can be mutually exclusively imaged as the confocal microscope scans successive focal depths.

For mass-production it may be preferable to form substrates from plastics materials using an alternative manufacturing process, such as e. g. high precision hot embossing or spin coating. These processes are standard microfabrication techniques that require the use of a high precision former.

The former may be fabricated e. g. by an X-ray based LIGA (Lithographie, Galvanoformung, Abformung) process. For a discussion of LIGA processing and associated manufacturing process see e. g. Micro Structure Bulletin, Vol. 5, No. 1, Feb 1997; J. Hruby, MRS Bulletin, 1-4, April 2001; T. Otto et al., Fabrication of Micro Optical Components by High Precision Embossing, Micromachining and Microfabrication Symposium, Santa Clara (CA), USA, 18-20 September 2000, Vol. 4179, S. 96-106; and L. J. Lee et al., Biomedical Microdevices, 3: 4,339-351, 2001.

Alternatively the former may be fabricated by a laser LIGA process e. g. of the type described in E. C. Harvey and P. T.

Rumsby, Fabrication Techniques and their Application to produce Novel Micromachined Structures and Devices using Excimer Laser Mask Projection, Proc. SPIE, Vol. 3223, pp. 26-33,1997.

An example preparation protocol for substrates formed from plastics materials using laser fabrication is as follows. A laser projection mask is designed and fabricated by opening apertures in a thin metallic film that is opaque to the ultra- violet radiation from the excimer laser. The mask may be made using well established techniques to prepare a chrome-on-quartz mask, or may be cut, for example with a Nd: YAg laser, into thin sheets of stainless steel, or chemically etched into metal sheets, as is commonly done to make solder paste screens for printed circuit board manufacture. The choice of fabrication

methods for the mask will depend on the resolution of the pattern required.

The resultant mask is then placed at the object plane of an excimer laser image projection system. Typically the projection. optics will form a reduced image of the mask apertures, thereby increasing the laser energy density at the workpiece, and enabling laser ablation to machine the substrate while causing little or no damage to the mask. An additional benefit of image reduction is that the fabrication process for the mask does not need to have as detailed a resolution as the intended manufactured parts. Typically reduction factors of between 5 and 20 times are used.

A polymer substrate is then placed onto a holder supported by a set of precision translation stages. The pulsing of the excimer laser and the motion of the translation stages is co- ordinated by a computer numerical control (CNC) system. A feature of pulser excimer laser ablation is that the substrate is machined to a depth determined by the number of laser pulses applied. Typically one laser pulse will machine about 300 nm (0. 3pm) deep.

When the mask pattern is a series of squares, triangles or other polygons in a regular array, these may be tessellated onto the substrate in a step-and-repeat process. Here the workpiece is held stationary, a fixed number of laser pulses applied, then the workpiece moved to a new position and a second series of laser pulses applied. Where the first set of laser pulses is different from the second set, the resultant features will be machined to different depths. This will create a set of terraces. Any number of terrace levels may be

created depending upon the number of different pulsing regimes used.

Where the mask is held on a translation stage, a new mask or new part of the mask can be brought into the object plane of the laser projector. This new mask area has a series of circular apertures at the same relative pitch as the first set of polygons. Using the CNC system, the circular apertures can be used to laser machine wells into each of the previously created terraces using a similar step-and-repeat strategy. In this way large areas may be covered with repeating patterns of terraces and wells.

The resulting patterned surface may be cleaned of the laser ablated debris and used directly. Alternatively the part may be used as a mould and an inverse copy created by electrodeposition of nickel in a nickel suphamate bath. The resultant nickel part, when separated from the laser machined surface, can be used as a tool in a polymer replication process by either embossing, injection moulding, reaction casting or other high speed means to produce many copies of the original laser machined structure. Using this replication step the choice of materials used to produce final parts is greatly increased to allow any material that may be formed using the high speed well understood replication processes.

A general description of the laser technique can be found in e. g. E. C. Harvey and P. T. Rumsby"Micromachining and microfabrication process technology III", SPIE Vol. 3223,26- 33,25 June 1997.

Fig. 7 shows a scanning electron microscope image of an array formed by laser ablation of a polycarbonate substrate. The wells are substantially cylindrical on square microterraces.

Fig. 8a shows a plan view of a further microterrace with a different shaped well; in this case substantially cylindrical with a concave bottom, such as that shown in Fig. 7. The cylinder has a diameter of 40 um and a depth at its deepest point of 30 um, giving a well volume of 59,000 pm3 or 59 pl.

The well is set in a square microterrace with sides of 75 um, allowing 86,044 wells on a 2.2 x 2.2 cm microscope cover slip or a well density of approximately 18,000.

Fig. 8b shows a cross section through neighbouring microterraces of the type shown in Fig. 8a. This shows that the well spacing in the direction perpendicular to the array surface (i. e. the distance between the bottom of one well and the top of a neighbouring lower well) is 25 um, which is significantly larger than the 0.5 um which is approximately the practical minimum height of the in-focus object plane imagable by a conventional confocal microscope with a fluorescence objective. This helps to reduce or eliminate signal overlap between neighbouring wells.

With this polycarbonate substrate we again successfully demonstrated how confocal microscopy can be used to selectively view successive sets of samples held in the wells. 1 um, red fluorescent (580/605 nm), FluoSphere polystyrene microspheres from Molecular Probes were loaded into the wells by suspending the microspheres in ultra pure water at a concentration of 10' beads per ml and centrifuging at 1000 rpm for 5 minutes.

Unloaded microspheres were removed from the substrate by gentle washing with ultra pure water. The substrate was then left to

dry by evaporation. Although some of the wells did not receive microspheres, sufficient numbers did to proceed with the demonstration.

The substrate was viewed using an LSM510 Zeiss confocal microscope. Figs. 9a-c show confocal reflection images of the substrate at depths of 64,118 and 167 pm. These reveal the substrate surfaces at the respective well bottoms. The positions of wells in the terraces at that depth are indicated by boxes.

Figs. 9d-f show corresponding confocal fluorescence images of the substrate at depths of 64,118 and 167 pm. Clearly, at each focal depth only the microspheres located at that depth are imaged, demonstrating that samples held in the wells can be mutually exclusively imaged as the confocal microscope scans successive focal depths.

Samples may be loaded in the wells by techniques familiar to the skilled person such as gridding, in situ synthesis or by printing (e. g. piezoelectric thermal deposition). A useful review of contacting and non-contacting dispensing technologies is provided by D. Rose, Microarray Printing Technologies, in Microarray Biochip Technology, ed. M. Schena, Chapter 2, pp.

19-39, Biotechniques Books, Eaton Publishing (2001). Further discussion of loading technologies may be found in e. g. DNA Arrays Methods and Protocols, ed. J. B. Rampal, Humana Press Inc. , Totowa NJ, 2001; A. Watson et al. , M. B. Eisen and P. O.

Brown, Methods Enzymol., Vol. 303,179-205, 1999; Current Opinion in Biotechnology, Vol. 9, No. 6,609-614, 1998; P. S.

Linsley et al., Nature Biotechnology, Vol. 19, No. 4,342-347, April 2001.

A preferred use for the substrates is for the production, by in situ replication, of multiple array targets containing viable biological samples. Owing to the ability of each well to hold an individual viable biological sample, a master template array may be produced in which each well contains a unique non- immobilised viable sample (for example genomic DNA clones and bacterial strains). Whole array replication is achieved by short term culturing of the samples within the wells of the master array, and replicating these samples in to blank arrays that contain an appropriate growth media, through the use of a solid state replicator. Such a replicator may comprise pins that uniquely address each well of an array within a single application. The replicator is partially submerged into the wells of the template array, and subsequently used to inoculate the wells of several blank arrays. The inoculated arrays are short term cultured, prior to dehydration and fixation. The replicants may then be assayed as appropriate.

The substrate can then be used in biological assays. For example, to determine an expression profile of a tumour sample the following protocol may be employed: a. Obtain the total RNA from the sample b. Reverse transcribe and label the RNA to form cDNA targets c. Hybridise the targets to a microarray of oligonucleotide probes held on the array surface of the substrate d. Remove non-specifically bound targets by high stringency washing e. Fluorescently detect the bound and labelled targets using a detection reagent f. Image the microarray by confocal microscopy h. Analyse the image to determine significant changes in fluorescence relative to a standard sample and thereby obtain an expression profile.

However, the substrate may also be used in applications other than expression analysis, such as proteomics (see e. g. D. Stoll et al., Protein microarray technology, Front. Biosci., 2002, Vol. 7, pp. 13-32), metabolome screening (see e. g. D. J. Oliver et al., Functional genomics: high-throughput mRNA, protein, and metabolite analyses, Metab. Eng., 2002, Vol. 4., pp. 98-106), antigen testing (see e. g. N. Dhiman et al., Gene expression microarrays: a 21st century tool for directed vaccine design, Vaccine, 2001, Vol. 20. pp. 22-30), SNP analysis (see e. g. O. P.

Kallioniemi, Biochip technologies in cancer research, Ann.

Med.. Vol. 33, pp. 142-147), microELISA (see e. g. B. Schweitzer and S. F. Kingsmore, Measuring proteins on microarrays, Curr.

Opin. Biotechnol. , 2002, Vol. 13 pp. 14-19), toxicity testing (see e. g. H. K. Hamadeh et al., An overview of toxicogenopmics, Curr. Issues Mo. l Biol., 2002, Vol. 4. pp. 45-56) etc.

Another application is live cell array analysis (see e. g. I.

Biran and D. R. Walt, Optical imaging fiber-based single live cell arrays: a high density cell assay platform, Anal. Chem., 2002 Jul 74: 3046-54). In this cell lines are grown directly on an array spotted with a variety of metabolites. Cells that have the ability (either innate, through mutation, or due to a transfected construct) to metabolise the additive give rise to a detectable colour change.

The exact dimensions of the arrays with microterraces used will vary for each potential application, as will the nature of the loading and reading of samples. For example, a high density array (e. g. having 10,000-400, 000 wells/cm2) could be used for detection of genomic imbalances or changes in gene expression. Lower density arrays (e. g. having 1,000-10, 000

wells/cm2) could be used for determining the presence of bacterial glycoproteins in a serological assay.

Likewise the design of the array can be chosen depending on the end-user requirements. For example, it may be desirable to place array replicates on different terraces, thereby. enabling easy assessment of consistency across an assay. Alternatively sample groups could be arranged on different terraces, thereby effectively creating embedded sub-arrays that can be assayed within the same experiment. Since the array may be designed with any number of terraces at different heights, a multitude of such sub-arrays could be provided. Alternative tessellating patterns could also be provided for specific applications.

While the invention has been described in conjunction with the exemplary embodiments described above, many equivalent modifications and variations will be apparent to those skilled in the art when given this disclosure. Accordingly, the exemplary embodiments of the invention set forth above are considered to be illustrative and not limiting. Various changes to the described embodiments may be made without departing from the spirit and scope of the invention.

The references mentioned above are hereby incorporated by reference.