Four 285 E 6 male mice and four euploid males of identical age were used. RNA extraction was performed, following the experimental stages detailed in Fig. 2, using the RNeasy Lipid Tissue Mini Kit (QIAGEN).

The quantity (measured on NanoDrop ND-1000) and quality (Bioanalyser 2100) of the total RNA in the eight mice were measured as described in the last stage (Fig. 2) and reported in Table 1.

The aim was to hybridize the extracted RNA with the Whole Mouse Genome Microarray Kit, 4x44K that includes probes designed from 41,000 transcripts plus unique genes (Agilent Technologies). We used the Amp Labeling Kit, Two-Color Kit (Agilent Technologies) to perform:

The double-color marking method was used according to the manufacturer’s protocol, and the mouse cRNA samples were marked with either cyanine 3 (Cy3, green) or cyanine 5 (Cy5, red). Alternate marking of both 285 E 6 and euploid mice with Cy3 and Cy5 makes it possible to test the two chromophores for equal binding plus equal reaction to laser, thus consolidating the raw data. This alternate labeling, referred to as the “dye swap” experiment, produces an image at the end of the process which has (theoretically) a red/green color balance which is the exact opposite of the first one.

Post-synthesis and post-labeling controls are performed measuring

Basic information on cDNA microarray structure is a prerequisite for understanding the hybridizing experiment and measuring the amount of hybridized material. More detailed information on microarray manufacturing can be found in [8–10].

A cDNA microarray, also known as a DNA chip, is a set of probes, ranging from several thousand to one million, on a solid flat silicon or glass surface. A probe is a short set of oligonucleotides, the length (expressed in N-mers; N specifies the number of bases—“mer”) in a given molecule, and varies according to the manufacturer (from 25 to 60 mers, sometimes more). The probe is designed to hybridize to the tested cDNA (called the target). The hybridization of one probe with a cDNA strand does not mean that the whole gene has been transcribed as a gene and, on average, encompasses between 120,000 and 150,000 base pairs. Several probes per gene are therefore needed to cover the entire length of the exonic sequence. An average hybridizing score is computed for each gene. All the probes are constructed in parallel. The size of the RNA varies between 50 and 3,000 bases. It is therefore necessary to fragment the strands to optimize the hybridization process with the probes (60 mers with our provider). Shorter strands (50–200 bases) can be obtained by using a specific buffer. In the experiment cited as an example, 825 ng of each color-labeled cRNA was subjected to competitive hybridization for 17 h at 65 °C.

Several observations support the use of microarrays. The first questions concern the spatiotemporal pattern of gene expression. “Is the gene transcribed?” “Is the gene transcribed in this tissue?” “When does the transcription of the gene actually start?”

Spot reactivity to laser stimulation shows the number of cDNA strands that have hybridized against the probe. The higher the number of hybridizing cDNA strands, the greater the laser reactivity. A dark spot means non-transcription, but this almost never occurs. The transcription score of a spot (TSS) is estimated by comparing the average transcription score (ATS) of the spots present in the array. As there is intra-array variation, the comparison is done with the spots encompassed in one part of the array (one fourth or one eighth of the array). Perfect versus mismatch technology (Affymetrix) offers an alternative to array signal normalization. For this, a perfectly matching oligonucleotide (P) (ATCCCATGGGCACTATCGCATGGATCATTG) and a mismatching oligonucleotide (M) are designed. M differs from P by the introduction of a wrong base in the middle of the oligonucleotide to prevent matching with the cDNA strands. The ATCCCATGGGCACCATCGCATGGATCATTG M oligonucleotide differs from P by the underlined base (C instead of T). As the M spot does not hybridize by construction, it provides an absolute non-hybridizing reference. The method may induce bias [9]. Other companies have developed other strategies to define internal technical controls ([11, 12], for Agilent Technologies).

Second question: “Is the gene under- or overexpressed in sample A (patient or mutant mouse) compared to sample B (unaffected person or wild-type organism)?”

One color-labeled set of cDNA is all that is needed to answer the first question, but two different colors are needed to compare two samples on the same array: mutant and “normal” control DNAs are labeled, each with different dyes (usually, cyanines 3 (Cy3 green) and 5 (Cy5 red)). Values are measured for each spot, including controls, as previously indicated (TSS, ATS, or perfect match and mismatch). The statistical difference threshold (SDT) is the ratio TSS/ATS or perfect match/mismatch. Different transcription intensities often appear between the two colored samples requiring an adjustment of the two distributions. “In experiments where two fluorescent dyes (red and green) have been used, intensity-dependent variations in dye bias may introduce spurious variations in the collected data” [13]. “Lowess normalization” is in fact an adjustment to have the two histograms match. This must be done before any comparisons with the spot selected as control.

Various questions relevant to this technique should be incentive for readers to consult books covering the technical aspects of cDNA microarrays in greater detail. Filtering and spot elimination are discussed ([10] pp. 118–134). All processing involves additional correction for the background signal (see discussion on possible bias induced by the different noise subtraction methods in [14]). TSS/ATS perfect match and mismatch refer to one spot. One spot means 1 probe, and more than 20 probes are used for one gene (depending on the gene length). The probe values for each gene have to be averaged. Empty or unreadable spots are eliminated and a decision must be made to rule out a gene on the basis of a predefined number of missing spots. Similarly, a cutoff must be determined in advance, setting the number of missing samples; beyond the cutoff, the average number of transcripts in one gene cannot be computed.

The reply is straightforward when a single structure can be removed separately, e.g., the cerebellum, hippocampus, or olfactory bulbs. It is more difficult when the structures are mutually embedded, e.g., the motor cortex and sensorial cortex, or are difficult to distinguish, e.g., the amygdala. The target structure can be sampled according to published procedures. Niculescu and colleagues review different techniques and suggest a laser capture microdissection procedure to ensure a homogenized set of cells [15].

A factor analysis will confirm that the different samples come from the same structure. The principle is based on the similarity of the homothetic patterns of correlation within tissues. We took four brain samples in each of five C57BL/6 mice and used the coordinates given by Paxinos and Franklin [16]. The four samples were (1) the caudate putamen (Cpu) using a 1 mm biopsy punch (between bregma 0.22 mm and interaural 3.58 mm and bregma 0.94 mm and interaural 2.86 mm); (2) the primary visual cortex (VI), 1 mm biopsy punch (between bregma 3.40 mm and interaural 0.40 mm and bregma 4.88 mm and interaural 0.68 mm); (3) the whole hippocampus (Hip); and (4) the whole cerebellum (Cer). The samples were individually subjected to cDNA microarrays, producing a table with 20 columns (Cpu1, Cpu2…Cpu5; VI1, VI2…VI5; Hip1, Hip2…Hip5; and Cer1, Cer2…Cer5) and 20,437 rows, giving the transcript values of the genes explored. The correlations between samples were computed and subjected to a factor analysis (principal component method) with varimax rotations. The results are presented in Fig. 3 and show that Cer, Cpu, and Hip groupings each display within-sample homogeneity. The situation is not so simple for VI as VI4 has an outlying position compared to VI1, VI2, VI3, and VI5, indicating that the VI4 transcripts performed differently compared to the other four samples. This is the result of errors during sampling. VI is very close to two secondary visual cortices (the mediomedial area and mediolateral areas) and to the presubiculum. VI4 may contain some tissues from these structures, thus producing different transcript values. The VI4 sample should not be considered when averaging the transcript values in VI.

A general approach is given by the softwares. The most popular is the scatter diagram showing the correlation between transcript values. Figure 4 shows the correlation between 22,126 transcript values obtained in the left and right hippocampus (r = 0.95).