A novel sensitive microarray approach for differential screening using probes labelled with two different radioelements

Abstract

We have developed a novel microarray approach for differential screening using probes labelled with two different radioelements. The complementary DNAs from the reverse transcription of mRNAs from two different biological samples were labelled with radioelements of significantly different energies (³H and ³⁵S or ³³P). Radioactive images corresponding to the expressed genes were acquired with a MicroImager, a real time, high resolution digital autoradiography system. An algorithm was used to process the data such that the initially acquired radioactive image was filtered into two subimages, each representative of the hybridisation result specific for one probe. The simultaneous screening of gene expression in two different biological samples requires <100 ng mRNA without any amplification. In such conditions, the technique is sensitive enough to directly quantify the amount of mRNA even when present in small amounts: 10⁷ molecules in the probe as assessed with an added control sequence and 2 × 10⁵ molecules with an endogenous tyrosine hydroxylase mRNA. This novel technique of double radioactive labelling on a microarray is thus suitable for the comparison of gene expression in two different biological samples available in only small quantities. Consequently, it has great potential for various biological fields, such as neuroscience.

INTRODUCTION

DNA array technology is increasingly used for large-scale screening of gene expression. The availability of laser devices that can differentiate between several fluorescent dyes has led to most development efforts being concentrated on fluorescent labelling of probes to be hybridised onto DNA arrays (the immobilised nucleic acid is called the ‘target’ and the free nucleic acid is called the ‘probe’). The use of two different fluorescent dyes, one to label probes from a control tissue and one to label probes from a tissue of interest, allows normalised quantification of gene expression. For example, standard high density microarray protocols using fluorescence-labelled probes (1,2) have identified sets of genes for which the expression changes in different forms of diseases (3,4) or in response to experimental stimulation (5,6).

In some areas of biology, such as neuroscience, large-scale gene expression screening using high density arrays would be more valuable if it could, reproducibly: (i) detect small modulations of gene expression (down to 30%), because such modulations may be of major biological significance (7); (ii) analyse tissues or cell populations that are available only in small amounts, such as cells obtained by needle biopsy or particular rat brain structures (down to 1 mg tissue); (iii) detect rare mRNAs (a few copies per cell and fewer than one copy per cell in heterogeneous tissues). Indeed, many genes of major scientific and/or medical interest are expressed only weakly, unlike a large number of housekeeping genes, and rare mRNAs make up 80–90% of total mRNA. Furthermore, it is also desirable to detect and accurately quantify, in a single sample and during the same experiment, numerous mRNAs, the amounts of which may differ by 10⁴–10⁵ times (8,9). Any such developments involve the issue of signal detection sensitivity.

Protocols described in the literature using fluorescent labelling allow the analysis of small quantities of tissue or cells and even of a single cell if transcriptional amplification (10,11) or RT–PCR amplification is used. However, there is as yet no published proof that such amplification procedures do not modify the relative abundances of nucleic acid species, especially those which are not abundant. Thus, avoidance of any enzymatic amplification step is preferable to maintain the relative abundance of different mRNA species. To evaluate the sensitivity of a microarray method, two factors have to be considered: the quantity of starting material required and the threshold of molecule detection. If there is no amplification step, the detection threshold of high density array methods using fluorescence-labelled probes is of the order of 2 × 10⁷ molecules with a minimum quantity of starting material required of 200 ng to 10 µg mRNA (2,12). These technical considerations currently prevent fluorescent labelling from being fully suitable for the requirements of microarray-based large-scale gene expression screening as applied to various biological fields where only small samples are available.

Gene expression in small samples can be screened, however, with great sensitivity by using non-amplified radioactively labelled probes with DNA microarrays, as the minimum quantity of starting material required for radioactive labelling is only 2–400 ng mRNA to detect ∼2 × 10⁷ molecules (12). Previously, such analyses were possible only for one mRNA sample at a time. A technique comparing several mRNA samples on the same high density array but attaining the sensitivity discussed above would be of great value. For example, the results could be normalized, each RNA sample being used as a control for the other, on each target of the microarray, as is possible with double fluorescent labelling (2).

These considerations led us to develop a technique for simultaneous hybridisation of two differently labelled radioactive probes on the same glass support microarray and detection of the hybridisation result for each probe separately. The development of this procedure required a device for detection of radioactive emission that could discriminate between different radioactive emission spectra and also with a spatial discrimination appropriate for the microarray density. The MicroImager has these properties. We have previously shown the potential of this device in the discrimination of the radioactive emissions of two different radioelements for in situ hybridisation of two probes on a single tissue section (13,14). Here we describe methods of labelling and hybridisation allowing work with two radioactive probes simultaneously on a single glass support microarray. The sensitivity of this method was analysed and we demonstrate the potential of this novel approach in cases where only small samples are available.

MATERIALS AND METHODS

Gene array

PCR products 300–1500 bp long were purified using the concert nucleic acid purification system and then spotted with an arrayer (Genetix) onto polylysine-coated slides (15). The cDNA clones used were obtained from adult rat brains by RT–PCR, from a positive and exogenous control luciferase cDNA sequence (572 bp insert) in the pGEM-T easy vector (Promega, France) and from a negative and exogenous control neomycin phosphotransferase cDNA sequence (738 bp insert) in the pGEM-T easy vector (Promega). A total of 384 clones were spotted onto the microarray. The microarray plan was made up of four blocks of four rows and 24 columns (as shown in Fig. 2). This plan was in duplicate on every microarray.

Figure 2.

Visualisation (arbitrary colours) of the results of double radioactive labelling of probes on a microarray. Hybridisation images obtained with 100 ng [³⁵S]dATP-labelled probe and 100 ng [³H]dCTP-labelled probe from two different tissue samples. Targets were PCR products of 300–1500 bp spotted onto polylysine-coated slides. (A) Simultaneous visualisation of both ³H and ³⁵S labelling. The ³H labelling is represented in green, the ³⁵S labelling in red and overlapping of the two in shades of yellow. (B) Visualisation of only ³H labelling. (C) Visualisation of only ³⁵S labelling. Above the three microarray images, a spot of ³H, one of a mix of ³H and ³⁵S and another of ³⁵S were set down on the microarray as controls for filtering, allowing segregation of ³⁵S β from ³H β disintegrations.

Preparation of the luciferase RNA

The luciferase RNA was prepared from the luciferase cDNA described above using the riboprobe combination system T7 (Promega).

RNA extraction

mRNA was directly isolated from crude extracts of rat brain tissues on magnetic beads [oligo(dT)₂₅ Dynabeads; Dynal].

All experimental procedures were carried out in accordance with the European Communities Council Directive (24.xi.1986) and with the guidelines of the CNRS and the French Agricultural and Forestry Ministry (decree 87848, licence number A91429). All efforts were made to minimise animal suffering and to use only the number of animals necessary to produce reliable scientific data.

Sample preparation for hybridisation

Aliquots of 100 ng mRNA were mixed with 0.1 µg random hexamers from a Superscript First-Strand Synthesis System for RT–PCR (Life Technologies, France), heated to 70°C for 10 min and cooled on ice. Probe synthesis and labelling were then performed in the presence of 5 mM MgCl₂, 1× reverse transcription buffer (Life Technologies), 10 mM dithiothreitol, 100 U RNaseOUT RNase inhibitor (Life Technologies), 0.05 mM ddTTP, 0.5 mM dGTP and dTTP, 100 U Superscript II reverse transcriptase (Life Technologies) and 10 µCi [³⁵S]dATP (Amersham) and 0.5 mM dCTP or 20 µCi [³H]dCTP (Amersham) and 0.5 mM dATP for the phosphorylated and tritiated probes, respectively, by incubation of the mixtures at 42°C for 50 min. RNA was eliminated by heating at 70°C for 15 min and treatment with 2 U RNase H (Life Technologies) at 37°C for 20 min. Unincorporated nucleotides were removed by passage through a P10 column (Bio-Rad).

Hybridisation

The probes were added to the hybridisation buffer (3.5× SSC, 0.3% SDS), heated to 95°C for 2 min, cooled to room temperature and then put on the microarray under parafilm (Fuji). Hybridisation was performed in a cassette chamber (Telechem) submerged in a water bath at 60°C for 16–17 h. Following hybridisation, arrays were rinsed at room temperature in 2× SSC, 0.1% SDS, then 2× SSC, then 0.2× SSC, each washing step lasting 2 min.

Acquisition of radioactive images with a MicroImager (Biospace Mesures, Paris, France)

A thin foil of scintillating paper was placed in contact with the microarrays. β-Particles emitted by the hybridised probes were identified by acquisition of the light spot emissions in the scintillating foil by a CCD, coupled to an image intensifier. The acquired results were displayed live on a computer. The end of acquisition was chosen at a time such that the number of disintegrations was statistically satisfactory (i.e. 10⁶ counts detected in the whole field acquired by the MicroImager, which is sufficient for reliable measurement). In our experiments, acquisition times were ∼24 h. As the total number of counts was displayed, without knowing the number of counts due to ³H labelling and those due to ³⁵S labelling during acquisition of double radioactive images, quantities of ³⁵S and ³H radioelements required for the synthesis of probes were calculated so that the number of disintegrations would be similar for the two probes. Filter processing of the emission signals with dual labelling software (Biospace Mesures) allowed discrimination and quantification in each pixel of the respective contributions of the two radioelements, which have significantly different physical properties: the two types of radioactive emission were discriminated according to the sizes of the optical spots acquired by the CCD camera. The algorithm principle consists of a local analysis of the statistical distribution of spot size: H(x,y,n) is the distribution of β disintegrations, with x and y being the spatial coordinates of the spot centre and n being the spot size expressed in number of pixels. Assuming that H₁(n) and H₂(n) are the known size distributions corresponding to isotopes 1 and 2 (here ³H and ³⁵S), the algorithm solves the following equation for each (x,y): [For any n, H(x,y,n) = α₁(x,y)H₁(n) + α₂(x,y)H₂(n)], with α₁(x,y) and α₂(x,y) being the respective intensities of the radioactive emissions generated by isotopes 1 and 2. This software provides two files, each file containing the exact count of particles from one isotope specifically detected in each pixel. These files can be translated into images and the specific emission of each isotope precisely quantified for each microarray spot. The number of disintegrations of β-particles per spot was measured using β-Vision (Biospace) and GenePix software (Axon).

Calculation of the coefficient of variation

From four replicate experiments of hybridisation to microarrays, the coefficient of variation (CV) was calculated for each element of the microarray as follows: CV = 100 × (standard deviation/mean). CV is expressed as a percentage.

RESULTS

Our aim was to develop double radioactive labelling for gene expression screening on microarrays for small quantities of starting material. We investigated the possibility of using 100 ng mRNA as starting material for probe synthesis without any amplification. We used [³⁵S]dATP and [³H]dCTP to differently label two probes synthesized from 100 ng mRNA extracted from two different tissues, total brain of adult rat and cortex of 12-day-old rat. These probes were simultaneously hybridised to a single microarray. The principle of the differential screening is illustrated in Figure 1. The radioactive emission resulting from the two isotopes was simultaneously acquired in real time, providing a global signal. The hybridisation results were then analysed using a new signal filtering algorithm (dual labelling software from Biospace Mesures), discriminating and quantifying the radioactive emissions specific to each isotope. The initial image was filtered to segregate the image corresponding to ³H β disintegrations (Fig. 2B) from that corresponding to ³⁵S β disintegrations (Fig. 2C). The quantitative data for both ³H and ³⁵S labelling were incorporated into a single image (Fig. 2A). In this image, green corresponds to the cDNA clones only detected by the ³H-labelled probes, red to those only detected by the ³⁵S-labelled probes and shades of yellow to those that are detected by both. The technique showed specific differences in gene expression between the two tissues as revealed by the ratios of ³H signal intensities to ³⁵S signal intensities that are not equal to 1 (Figs 2 and 3A).

Figure 1.

mRNA was extracted from cells or tissues and reverse transcribed into single-strand cDNA (the probe). Probes were labelled by incorporation of radioactive nucleotides during their synthesis. The labelled probes were denatured and hybridised to the microarrays. Radioactive images were acquired with a MicroImager (Biospace Mesures, Paris, France), a real time, high resolution digital autoradiography system, with a 24 × 32 mm imaging area, a spatial resolution of 20 µm and a pixel size of 5 µm. After initial digital acquisition of the radioactive image with the MicroImager, including both ³H and ³⁵S/³³P labelling, the data were filtered to segregate the image corresponding to ³H β disintegrations (the green spots on the microarray) from that corresponding to ³⁵S β disintegrations (the red spots), each being representative of the hybridisation result for one probe.

Figure 3.

Reproducibility of the method of double radioactive labelling of probes on a microarray. (A) The results of two independent hybridisation experiments, each using 100 ng [³⁵S]dATP-labelled probe and 100 ng [³H]dCTP-labelled probe from two different tissue samples (total brain of adult rat versus cortex of 12-day-old rat) were compared. For each of the experiments, ratios of the ³H signal intensity to the ³⁵S signal intensity were calculated for each element of the microarray and normalised with respect to an external standard (luciferase cDNA sequence). The values from one experiment were plotted against those of the other. (B) Comparison of two independent hybridisations, each using 100 ng [³⁵S]dATP-labelled probe and 100 ng [³H]dCTP-labelled probe from two different tissue samples. For one experiment, 10⁷ molecules of transcribed luciferase RNA were added to 100 ng mRNA used for probe synthesis (experiment 1) and 10⁸ molecules of luciferase RNA were added for the other experiment (experiment 2). The ³H signal intensities (number of counts in 24 h) for each element of the microarray of one experiment were plotted against those of the other. The values of signal intensity are statistically reliable from 100 counts for 24 h. The clone shown by an arrow corresponds to the luciferase control. The ratio of the ³H signal intensity of the luciferase clone in experiment 2 to that in experiment 1 is ∼10, which is as expected.

As a control for filter segregation of the ³⁵S from the ³H β disintegrations, three control dots were spotted by hand on the slide as described previously (13). The dots contained the ³H-labelled probe (200 c.p.m.), a mix of the ³H-labelled (200 c.p.m.) and ³⁵S-labelled (200 c.p.m.) probes and the ³⁵S-labelled probe (200 c.p.m.). All three spots are observed in the total labelling image (Fig. 2A), but only two dots are observed after filtering, as expected (Fig. 2B and C). Quantification of the radioactivity emitted by each dot before and after filtering gave values in accordance with the amount of radioactivity spotted.

To assess the reproducibility of the technique of dual radioactive labelling on microarrays, four replicate experiments of hybridisation to microarrays were independently performed. For each of the four experiments, ratios of the ³H signal intensity to the ³⁵S signal intensity were calculated for each element of the microarray and normalised with respect to an external standard (the luciferase cDNA sequence). From four replicate experiments we calculated the mean and CV of the values for each element of the microarray. The CV, or relative standard deviation, provides a quantitative estimate of the precision of measurement of differential expression. The average CV was 13%. For differences of abundance of mRNA between two tissues of 1.3–2-fold, the average CV was 10%. The ratios in one of the four independent experiments were plotted against those from another experiment: the points are located along the diagonal, attesting the reproducibility of the method (Fig. 3A). We also tested the putative influence of the radioelements used on hybridisation. The same microarray was hybridised with ³H-labelled probes from adult rat brain and ³⁵S-labelled probes from the cortex of 12-day-old rats. The reverse experiment was also performed: hybridisation with ³⁵S-labelled adult brain probes and ³H-labelled 12-day-old cortex probes. The two experiments provided similar ratios after normalisation with respect to an external standard (luciferase cDNA sequence).

To assess the sensitivity of the method, two factors have to be considered: the quantity of starting material required and the threshold of molecule detection. This technique of double radioactive labelling on microarrays is satisfactory with 100 ng mRNA for probe synthesis without amplification (Fig. 2). A half dentate gyrus of rat that corresponds at most to 5 mg of starting tissue is sufficient starting material for probe synthesis (data not shown). To test the detection threshold, a luciferase cDNA sequence (572 bp insert), which has no homology with mammalian RNA, was cloned into the pGEM-T easy vector. The insert was spotted onto the microarray and RNA of this luciferase sequence was synthesized. Various quantities of luciferase RNA (10⁶, 10⁷, 10⁸ and 10⁹ molecules) were added to 100 ng rat mRNA before labelling. We observed that the signal intensities were proportional to the amount of luciferase mRNA in the probe, except for 10⁶ molecules, which gave a signal indistinguishable from the background. The signal intensities of the other elements of the microarray were unchanged, thus illustrating the reproducibility of results between independent hybridisations to the microarray (Fig. 3B). The point lying off the diagonal in Figure 3B corresponds to the luciferase gene: various amounts of the in vitro transcribed RNA were added to an aliquot of 100 ng rat mRNA (10⁷ and 10⁸ molecules for experiments 1 and 2, respectively; Fig. 3B). As expected, the ratio of the ³H signal intensity of luciferase RNA in experiment 2 to that in experiment 1 is close to 10 (Fig. 3B). These results are illustrated in Figure 3B, showing the ³H signal intensities. A similar pattern was obtained for the ³⁵S signal intensities (not shown). The limit of detection, i.e. the smallest quantity of luciferase mRNA detected, was 10⁷ molecules. The signal intensity corresponding to this limit quantity was twice that of the background. Thus we were able to use as little as 100 ng mRNA for probe synthesis and still detect 10⁷ molecules of RNA of an external gene, without any probe amplification. Under these experimental conditions, tyrosine hydroxylase mRNA, an internal control, was satisfactorily detected only in the total brain of adult rat. Tyrosine hydroxylase mRNA represents ∼0.0002% of total brain mRNA (16).

As a negative control, a neomycin phosphotransferase sequence, which has no homology with rat RNA, was spotted onto the microarray. No signal was detected for this negative exogenous control in any hybridisation experiment, as expected.

DISCUSSION

The use of radioactive detection for DNA microarray analysis had not previously been fully evaluated, despite radioactivity being a highly sensitive tool for molecular detection and its widespread use with macroarray membranes (12,17). One reason may be that no technique was previously available for the simultaneous detection of different isotopes on a single microarray, allowing direct comparison of two samples by simultaneously hybridising them to the same array.

The aim of our work here was to investigate the use of double radioactive labelling for microarray application under conditions allowing detection of weakly expressed genes in small amounts of biological materials, without any amplification step.

The limitless dynamic range of the MicroImager for each isotope allows, during a single acquisition on a single microarray, the comparative analysis of weak and strong signals. This allows the detection of differences of 10⁴–10⁵-fold in mRNA expression (18–20). Moreover, the spatial resolution of 20 µm and the 5 µm pixel size of the MicroImager are satisfactory for microarray analysis. These specifications make this device suitable for high density microarray analysis (>5000 spots/array).

Reproducibility of double radioactive labelling experiments

The average of the CV calculated for each element of the array of 13% is similar to that reported for fluorescent labelling by Yue et al. (2). Hybridisation experiments on the same microarray using mRNAs from one tissue labelled with ³H and a second tissue labelled with ³⁵S also provided similar results to the inverse experiment in which the same first tissue was labelled with ³⁵S and the second with ³H. This provides confidence in the reliability of the method.

The average CV for the differences of 1.3–2-fold in the expression of mRNAs between the two tissues was 10%. This indicates that small modulations of gene expression (down to 30%) could be reproducibly detected by our technique.

Sensitivity of double radioactive labelling experiments

The term sensitivity refers to the minimum detectable abundance level, i.e. the smallest fraction of total mRNA that can be detected. This includes two distinct factors that have to be considered to assess the detection sensitivity on microarrays: first, the amount of total mRNA (or total tissue) required for one hybridisation on one array; second, the smallest number of molecules of a given mRNA species that can be reproducibly detected on the array.

An amount of 100 ng mRNA can be used for probe synthesis by this technique of double radioactive labelling for hybridisation on a glass microarray, without any enzymatic amplification step. This is of the same order of magnitude as previously described for radioactive labelling with nylon-based microarrays: as expected, double radioactive labelling has a similar sensitivity to single radioactive labelling.

Double radioactive labelling with 100 ng mRNA can detect 10⁷ molecules of a given RNA, in this case control luciferase RNA. Many low abundance mRNA species are found at this level, i.e. <0.01% of the total mRNA sample [assuming that the average length of mRNAs is 2 kb, there are 9.1 × 10¹⁰ (≈10¹¹) mRNA molecules in 100 ng mRNA] and less than 30–100 mRNA molecules per cell (assuming that there are on average 3 × 10⁵ mRNA molecules in a cell; 12). This is in full accordance with previously published results with ³³P-labelled probes on nylon microarrays (12).

Based on calculations from published results, fluorescent labelling theoretically allows detection of ∼10⁷ molecules without an enzymatic amplification step (21), but the minimum quantity of starting material required is 2–100 times greater (200 ng to 10 µg mRNA) (2,12).

This difference between the two labelling techniques may be due to differences in the yields of the labelling procedures and hybridisation. Unlike radioactive nucleotides, fluorochromes are much larger than nucleotides and are very much more likely to have significant effects on enzymatic reactions and the hybridisation steps.

Sensitivity also depends on the ratio of signal to background. As the microarray has no intrinsic radioactivity, the background is extremely weak with radioactive labelling. The signal is spontaneously and directly emitted by the labels so that simply increasing the acquisition time leads to a proportional increase in sensitivity, assuming that the background is negligible. In the case of fluorescent labelling, the background is much higher and therefore impairs the sensitivity of the technique.

From these various arguments it would be expected that the smallest number of molecules detected would be higher with fluorescent labelling than with radioactive labelling. However, when working with small quantities of mRNA, increasing the concentration of total mRNA increases the yield of reverse transcription reactions. There is a similar effect of nucleic acid concentration on hybridisation reactions. Indeed, the two techniques require different quantities of biological sample (one to two orders of magnitude) to give a similar absolute detection threshold: this may be because the negative effect of fluorescent labelling on the yields of reverse transcription and hybridisation reactions is counterbalanced by the higher nucleic acid concentrations resulting from the use of greater amounts of starting material. Fluorescence techniques are being improved, especially to increase the incorporation of dyes using aminoallyl-dNTPs. More sensitive methods using tyramide signal amplification are also being developed. However, to our knowledge, no microarray applications have yet been published.

We used tyrosine hydroxylase as a model to validate our technique for small quantities of biological samples. Indeed, tyrosine hydroxylase mRNAs represent 0.0002% of total rat brain mRNA (16), which is two orders of magnitude less than the luciferase mRNA used as an external control. The technique was able to detect mRNAs present at as few as 1–10 copies per cell, using 100 ng total mRNA. mRNAs that differ in length or in sequence may be reverse transcribed and/or may hybridise to complementary sequences with different yields. These phenomena may account for the huge difference between the detection thresholds for the luciferase and tyrosine hydroxylase mRNAs. Double labelling procedures thus have a substantial advantage over single labelling techniques, because normalisation of the results (intra- or inter-microarray normalisations) may be very difficult for single labelling approaches. Different mRNAs present at different concentrations will be detected with different yields. When comparing two experimental conditions, performing both hybridisations simultaneously on the same microarray and reproducing the experiment several times would limit any quantitative bias, which can be difficult to discriminate and eliminate by comparing several microarrays and using normalisation techniques.

Perspectives and conclusion

Double radioactive labelling opens novel possibilities for large-scale gene expression screening on microarrays with small quantities of biological samples and requires no amplification step. Nevertheless, fluorescent labelling has some attractive characteristics not yet matched by radioactive labelling techniques. First, the availability of several fluorescent dyes that are differentiated by laser devices (22) makes it theoretically possible to compare four to six experimental conditions in the same screening experiment. Secondly, data acquisition is fast with fluorescent labelling (a few minutes for one acquisition), thus enabling high throughput analysis. While the acquisition time is only slightly longer for strong and moderate signals with radioactive labelling (15–30 min), it is much longer for very low signals (up to 24 h). However, such weak signals correspond to rare mRNAs that cannot be detected by fluorescence in the present state of the art. Nevertheless, reducing acquisition times for weak signals would make radioactivity more user friendly. In the future, as shown by preliminary experiments, it may even be possible to couple double radioactive labelling with fluorescent labelling (not shown). This would allow novel types of experiment which will lead to a better understanding of biological phenomena involving modulations of gene expression.