Introduction

The identification of genes mastering specific steps of brain development or function or involved in aging or pathological events is the focus of several laboratories. To this aim, gene expression analysis offers the most appropriate approach. Although numerous techniques are available to estimate messenger ribonucleic acid (mRNA) expression, the quantitative real-time reverse transcriptase polymerase chain reaction (RT-PCR) is the method of choice among neuroscientists, as it is a fast, sensitive, and reliable technique for the detection and the quantification of mRNA transcription levels of specific genes. However, a variety of problems affects the use of the real-time RT-PCR strategy. These are linked to the variability of the experimental procedure, including variations due to both technical and biological aspects. As the PCR step exponentially amplifies a small initial amount of complementary deoxyribonucleic acid (cDNA) to a large quantity of DNA product, even very small technical variations can produce large errors in the final product, leading to wrong interpretations of the gene expression analysis. To deal with the abovementioned difficulties, researchers frequently use to perform a relative quantification of their genes of interest, by normalizing their abundance to the expression level of a stably expressed gene, commonly termed reference or housekeeping gene, measured simultaneously in the same biological material. Reference genes are usually essential endogenous genes, often highly expressed in the cell, which are involved in various basic processes, such as cellular metabolism, cell structure, and protein synthesis. Therefore, they are supposed to be constitutively and similarly expressed in different cell types and not to be influenced by biological (e.g., gender, age, metabolic and hormonal conditions, disease stage) or experimental (e.g., temperature, treatments with soluble factors) conditions. An ideal reference gene should meet essentially one criterion: its expression must not undergo any regulation. Such an ideal reference gene does not exist. Thus, the most realistic approach is to choose the gene that shows the least variability under a specific study condition. If the expression of the housekeeping gene is largely altered by biological or experimental factors, the detection of small changes of the target gene becomes unfeasible. Many housekeeping genes commonly used in real-time RT-PCR were originally employed as internal controls for more qualitative assays such as Northern blot, RNase protection tests, and conventional RT-PCR, but these genes were often used without a precise knowledge of their individual variability in quantitative RT-PCR under defined experimental conditions.

Inspired by a recent work (Jung et al. 2007), we performed a PubMed search of articles using the terms “mouse brain,” “gene expression,” and “RT-PCR” combined with the Boolean operator “AND,” in order to obtain information about the most used reference genes. We carried out the same bibliographic survey also for mouse cerebellum, brain stem, hippocampus, septum, neocortex, and olfactory bulb. Glyceraldehyde-3-phosphate dehydrogenase (Gapdh), beta-actin (Actb), and the ribosomal subunit 18s were undoubtedly the most frequently used reference genes (supplementary Table 1). All other housekeeping genes were used in a minor part of the considered papers. This search clearly proves that a univocal housekeeping gene for mouse brain has not yet been identified. The use of some of the above-cited genes as internal controls for ribonucleic acid (RNA) quantification in the brain has been recently questioned, as their expression is modulated under different physiological and pathological circumstances, both in vivo and in vitro. For example, recent researches have demonstrated that sex differences and the secretion of hormones and neurotrophins could be sources of variability for Gapdh and Actb (Matsumoto et al. 1994; Barroso et al. 1999; Perrot-Sinal et al. 2001; Wong and Leung 2001; Zhang et al. 2001). In addition, it has been observed that Gapdh transcript and protein were upregulated in neuronal apoptosis (Sawa et al. 1997; Ishitani et al. 1998; Chen et al. 1999; Yang et al. 2007), which is an important mechanism for brain development, homeostasis, and disease. Furthermore, Gapdh and Actb levels decreased in an animal model of brain ischemia (Bond et al. 2002). An age-related hypermethylation of genomic DNA was found on the 18s and 28s ribosomal RNA genes in brain of aged mice (Swisshelm et al. 1990), suggesting that 18s expression could not be a stable housekeeping gene during brain aging. Regulation of the18s expression has been demonstrated also in Alzheimer’s disease pathology (Gutala and Reddy 2004). However, the use of such genes has been demonstrated useful under some experimental conditions. For example, no alterations in Gapdh mRNA in brain of aged rats or in postmortem brain samples of Alzheimer’s disease patients were found (Slagboom et al. 1990; Gutala and Reddy 2004; Tanic et al. 2007). In addition, the 18s ribosomal RNA (rRNA) showed to be a stable reference in the studies of addictive drug effects in rat brain (Proudnikov et al. 2003).

Identifying a good reference gene for brain is not trivial: The nervous tissue is intrinsically heterogeneous. First, we have to consider a spatial heterogeneity consisting in the functional and structural variety of brain subregions, which include further different cell types. In addition, as brain areas differ for the relative content of white matter, its contribution must not be ignored, especially for those genes whose transcripts are localized into axons, like Actb (Sotelo-Silveira et al. 2008). Researchers must take care also of the time: Brain development, maturation, and aging deeply modify the nervous system at structural, functional, and molecular levels. Moreover, different brain regions can be involved by age-related changes at different time points. For example, the development of cerebellum in mammals is mainly a postnatal process (Goldowitz and Hamre 1998), while in the cerebral cortex, it happens essentially in the embryonic life (Rakic 1974). Thus, no gene could be automatically considered the internal standard for different brain areas without taking into account its specific tissue- and age-dependent behavior.

No systematic real-time RT-PCR survey comparing the suitability of different candidate reference genes during mouse brain development and aging has been published to date. Therefore, the purpose of our work is essentially to identify a panel of fully characterized reference genes for distinct parts of the mouse brain. Thus, we compared the stability of a panel of genes, including three frequently used reference genes (such as Gapdh, 18s rRNA, and beta(2)-microglobulin [B2m]) and three relatively new housekeeping genes (such as phosphoglycerate kinase 1 [Pgk1], transferrin receptor [Tfrc] and beta-glucuronidase [Gusb]; Table 1), in mouse whole brain tissue and in a number of brain regions (cerebellum, brain stem, hippocampus, medial septum, frontal neocortex, and olfactory bulb). We analyzed two time points: the postnatal day 7 (P7) and 6 months of age (6M). The candidate reference genes were ranked according to their intragroup (the interindividual variability considered inside a single experimental group, e.g., P7 cerebella) and intergroup (the gene variability between the two ages and among different tissues) variations, first by simply comparing the coefficient of variation of the expression values and secondly by means of NormFinder and GeNorm softwares (Vandesompele et al. 2002; Andersen et al. 2004). As we found that the “new” housekeeping gene Pgk1 was one of the best control genes in the whole brain, cerebellum, hippocampus, and olfactory bulb, we studied in depth its expression, broadening our analysis to other time points (see “Materials and Methods”). Since Pgk1 was employed only a few times as reference gene for the nervous system (Pohjanvirta et al. 2006; Johansson et al. 2007), here we report the first complete study of Pgk1 expression throughout the development and the aging of mouse brain. The final goal of our study was to make available our data and our method for reference gene validation to the other neuroscientists.

Table 1 Panel of the candidate housekeeping genes evaluated in this study
Full size table

Materials and Methods

Tissue Collection and RNA Isolation

In order to compare the stability of the six candidate reference genes (Pgk1, 18s rRNA, Tfrc, Gapdh, Gusb, and B2m), a first set of experiments was performed on mouse hemibrain, cerebellum, brain stem, hippocampus, medial septum, frontal neocortex, and olfactory bulb at the P7 and 6M (n = 5 for each tissue at each age). A second set of experiments was carried out to accomplish an extended time course analysis of Pgk1 expression on mouse hemibrain at different ages from embryonic day 14 (E14) to 6M (including P0, P7, P14, and P60; six individuals per time point) and on the abovementioned brain regions at P7, P14, P60, 6M, and 18M (four to six individuals per time point).

All tissues were collected from CD-1 mice (Harlan, Corezzana, Italy). A pregnant mouse with E14 embryos was killed by a deep general anesthesia obtained by intraperitoneal administration of ketamine (100 mg/kg; Ketavet; Bayern; Leverkusen, Germany) supplemented by xylazine (5 mg/kg; Rompun; Bayer). The embryos were removed immediately by cesarean section and transferred in an ice-cold solution containing (in mM)125 NaCl, 2.5 KCl, 2 CaCl2, 1 MgCl2, 1.25 NaH2PO4, 26 NaHCO3, and 20 glucose and bubbled with 95% O2–5% CO2. Mouse pups (P0 and P7) were cryoanesthetized in melting ice. P14, P60, 6M, and 18M mice were anesthetized with isoflurane (Isoflurane-vet, Merial, Milan, Italy) and decapitated. Animal procedures were approved by the Animal Care and Use Committee of the University of Turin. Brains were quickly removed under semisterile conditions, placed in the previously described ice-cold solution, and then manually divided in two parts by a sagittal section. In the experiments performed on dissected brain regions, immediately after the whole-brain collection, cerebellum, brain stem, and olfactory bulbs were manually dissected. Then, coronal sections from the frontal neocortex, medial septum, and hippocampus were prepared using a vibratome, proceeding in rostro-caudal direction (Leica Microsystems GmbH, Wetzlar, Germany): starting from Bregma 1.10 mm, two 400-μm-thick slices were cut in order to collect pieces of neocortex containing the primary somatosensory cortex (S1), which included the S1 forelimb region (S1FL), S1 hindlimb region (S1HL), the S1 barrel field (S1BF), the S1 dysgranular zone (S1DZ), the primary and the secondary motor cortices (M1 and M2), and the cingulate cortex area 1 and area 2, and pieces of medial septum including the lateral septal nucleus dorsal part (LSD), intermediate part (LSI) and ventral part (LSV), the medial septal nucleus (MS), the lambdoid septal zone (Ld), the septohippocampal nucleus (SHi), the septofimbrial nucleus (SFi), and the dorsal peduncular cortex (DP; Fig. 1a). Two 400-μm-thick hippocampal slices were cut starting from Bregma −1.50 and included the oriens layer (Or), CA1, CA2 and CA3 fields, the dentate gyrus (DG), the stratum lucidum (SLu), the stratum radiatum (Rad), and the lacunosum moleculare layer (LMol; Fig. 1b). White matter parts (corpus callosum and fimbria) were manually separated from neocortical, septal, and hippocampal slices and not included in the gene expression analysis. Stereotaxic coordinates were obtained from the mouse brain atlas “The mouse brain in stereotaxic coordinates” (Paxinos and Franklin 2001).

Figure 1
figure 1

Schematic representation of the brain sections from which neocortical, septal and hippocampal regions were dissected. Neocortical and septal parts (gray parts in a) were separated from the same two 400-μm-thick slices, cut starting from Bregma 1.10 mm; hippocampus (gray parts in b) were obtained from two more caudal sections (400 μm thick), cut starting from Bregma −1.46 mm (b). White matter parts (corpus callosum and fimbria) were manually separated from neocortical, septal, and hippocampal slices and not included in the gene expression analysis

Full size image

All samples were rapidly frozen in −80°C 2-methylbutane and then stored at −80°C.

To ensure an RNase-free environment for tissue collection, the abovementioned solution was prepared with diethylpyrocarbonate-treated (DEPC; Sigma-Aldrich, St. Louis, MO, USA) water and RNAse-free salts. Moreover, RNase was removed from all solid materials used: Glass materials were heated at 180°C for at least 6 h, and plastic materials were rinsed with RNaseZap (Ambion, Austin, TX, USA) prior to the use.

Total RNA was isolated by extraction with the commercially available TRIzol Reagent (Invitrogen Life Technologies, Grand Island, NY, USA), in accordance with the manufacturer’s instructions. Briefly, thawed tissues were put in TRIzol Reagent (1 ml for hemibrains, cerebella, and brain stems; 500 μl for the other dissected brain regions) and homogenized on ice using a plastic homogenizer potter. After the addition of 200 μl of chloroform, the aqueous phase was separated from the organic one by centrifugation and then precipitated using a volume of isopropyl alcohol. RNA was subsequently washed with 70% ethanol and suspended in DEPC-treated water. Potential contaminating DNA was removed with DNase (Deoxiribonuclease I; Sigma-Aldrich) prior to the reverse transcription (RT) reaction. Concentration and purity of extracted RNA were evaluated prior and after the DNase treatment by spectrophotometry, by measuring the absorbance at 260 (A 260) and 280 nm (A 280). The precision of the measurements was controlled by repeating measurements ten times for each sample. The absence of RNA degradation was confirmed in a randomly chosen subset of samples by agarose gel electrophoresis with ethidium bromide staining. RNA samples were stored at −80°C until subsequent analysis.

Real Time RT-PCR

One microgram of total RNA was reverse-transcribed to single stranded cDNA using the commercially available High-Capacity cDNA Archive Kit (Applied Biosystems, Foster City, CA, USA), according to the manufacturer’s instructions. Each reaction consisted in RT buffer, deoxyribonucleotide triphosphate mixture, random primers, and Multiscribe RT enzyme. Samples were incubated at 25°C for 10 min and then at 37°C for 2 h in a PTC-100 Thermal Cycler (MJ Research, Waltham, MA, USA). Negative controls of the RT were performed by omitting the enzyme or by substituting the RNA for sterile RNAse-free water, to control for genomic DNA and RNA contamination, respectively. cDNA samples were stored at −20°C.

Quantitative real-time PCR was carried out using the ABI Prism 7000 Sequence Detection System instrumentation (Applied Biosystems) in combination with Taqman reagent-based chemistry: commercial Taqman gene expression assays, consisting in a specific set of primers and a fluorogenic internal probe, were purchased from Applied Biosystems to determine the amount of the six reference genes (codes are shown in Table 2) and of the target gene Kcnc3 (cod. Mm00434614_m1).

Table 2 Comparison of the six candidate reference genes in mouse hemibrain
Full size table

PCR amplifications were performed in Optical 96-well plates (Applied Biosystems) on cDNA samples corresponding to a final RNA concentration of 2 ng for the hemibrains and 200 pg for the dissected brain regions. PCR was performed in a total volume of 50 μl containing 25 μl of 2× Taqman Universal PCR Master mix (Applied Biosystems), 2.5 μl of 20× Taqman gene expression assays and 22.5 μl of the properly diluted cDNA. Reaction conditions were as follows: 50°C for 2 min, 95°C for 10 min, followed by 50 cycles 95°C for 15 s alternating with 60°C for 1 min. PCR amplifications were always run in duplicate. Blank controls consisting in no template (water) or RT negative reactions were performed for each run.

Data Analysis

For the quantitative comparison of the candidate reference genes, data extracted from each real-time RT-PCR run were analyzed by means of the 7000 v1.1 SDS instrument software (Applied Biosystems): The value of the noise fluorescence, usually indicated as the baseline of the run, was automatically determined. On the contrary, in order to guarantee the comparability between data obtained from different genes and from different runs, the threshold value was manually set to the value of 0.2. This threshold line was chosen to detect the point at which the reaction reaches a fluorescence intensity above the background. The cycle at which the sample exceeds this chosen threshold, usually called cycle threshold (CT), was automatically calculated and used to quantify the starting copy number of the target mRNA. As the raw CT values are log-based measurements inversely correlating to the cDNA concentration, all the data were transformed in the linear form 2−CT using a Microsoft Excel (Microsoft, Bellevue, WA, USA) spreadsheet.

A first descriptive statistical analysis was performed, investigating the average (Avg), the maximum (Max), the minimum (Min), and the standard deviation (Sd) of the collected 2−CT data, using Microsoft Excel software. The significance of the expression difference between P7 and 6M was assessed for each reference gene in every considered tissue, by means of unpaired two-tail Student’s t test (where a P < 0.05 was considered statistically significant).

Then, as the standard deviation and variance tended significantly to increase with the rise in expression values, we decided first to compare the variability of the candidate reference genes expressed by their coefficients of variation (CV, calculated as standard deviation/mean). In parallel, in order to deeply investigate the expression stability of the six genes, we used two Microsoft Excel add-in called NormFinder (kindly provided by Jensen JL, Andersen CL, and Gundesen C) and GeNorm (free available on the internet at the web site http://medgen.ugent.be/~jvdesomp/genorm/). These two softwares use two different approaches in order to identify the most stable reference genes. NormFinder calculates a stability value according to the combined estimate of intra- and intergroup expression variations of the gene studied. GeNorm calculates the stability value M as the average pairwise variation of a particular gene with all other candidate reference genes. Thus, we compared the CV, the NormFinder, and the GeNorm approaches and ranked the candidate genes in each tissue at P7 at 6M and during the tissue maturation (P7 versus 6M). By means of GeNorm, we assessed also the optimal number of reference genes which should be used for each analyzed tissue and age. The software measures the pairwise variation between two sequential normalization factors containing an increasing number of genes. It starts to calculate the pairwise variation V2/3 between the normalization factor NF2 (including only the two most stable reference genes) and the normalization factor NF3 (including the three most stable genes). Then, it performs a stepwise calculation of the V3/4 between the NF3 and the NF4 and then of the pairwise variation values V n /n + 1 between the subsequent normalization factors NF n and NF n  + 1. A large variation means that the added gene has a significant effect and should preferably be included for calculation of a reliable normalization factor. Based on GeNorm manual instruction, we considered 0.15 as a cutoff value, below which the inclusion of an additional reference gene is not required.

To assess the validity of the reference gene ranking, we selected a target gene, Kcnc3, that shows an age-dependent regulation in mouse cerebellum (Goldman-Wohl et al. 1994). Then, a correlation test (Pearson product moment correlation test) was used to verify if the expression of the candidate reference genes could show a correlation with the variation in Kcnc3 expression during cerebellar maturation.

The consequence of using an unstable reference gene for data normalization was evidenced by normalizing the expression of the target gene Kcnc3 to those of the six candidate reference genes in mouse cerebellum at P7 and 6M. We normalized Kcnc3 expression to each internal control by means of the 2−ddCT method (Livak and Schmittgen 2001). The calibrator was the P7 time point. Furthermore, we carried out a normalization to the normalization factor NF3, calculated by geometrically averaging the expression values obtained by the three most stable reference genes identified by the GeNorm software (Pgk1, Tfrc, and B2m). This geometric average was considered a unique reference entity, as proposed by Vandesompele and colleagues (2002).

Pgk1 expression differences throughout mouse lifespan were statistically evaluated for each tissue by means of a two-way analysis of variance test (where a P < 0.05 was considered statistically significant). Then, in order to verify the expression stability of the Pgk1 gene, we performed a linear regression analysis to verify if the experimental data could be fitted by a horizontal straight line. At the end, a correlation test (Pearson product moment correlation test) was used to verify if Pgk1 expression could vary depending on mouse aging.

Results

Assessment of RNA Yield and Quality

Tissue collection and storage were optimized to guarantee the maximum yield of RNA. To prevent RNA degradation, tissue extraction was performed using RNase-free materials and solutions. Then, the integrity of RNA was confirmed by visual analysis on agarose gels stained with ethidium bromide: All samples analyzed showed clear 28s and 18s rRNA bands, with the 28s band approximately twice intense as the 18s band, indicating that RNA was almost intact (data not shown). As protein contamination could severely affect enzymatic reaction efficiency, the purity (the ratio A 260/A 280) of the 191 RNA samples used for the real-time RT-PCR analysis was evaluated by spectrophotometry. The average purity was 1.92 ± 0.23, indicating high-quality RNA. It was notable that similar values of purity were obtained by both tissues of great mass (like hemibrain, cerebellum, and brain stem) and dissected regions of small size (as hippocampus, neocortex, and septum slices), indicating a reliable and standardized method of RNA extraction.

Expression Levels of Candidate Housekeeping Genes

In order to investigate the abundance and the variability of the six candidate housekeeping genes in each used experimental condition, descriptive statistical analysis was performed on their linearized expression data (2−CT): the supplementary Tables 2, 3, 4, 5, 6, 7, and 8 include the average (Avg), the maximum (Max), the minimum (Min), and the standard deviation (Sd) of the collected data in mouse hemibrain, cerebellum, brain stem, hippocampus, medial septum, frontal neocortex, and olfactory bulb, respectively, at the P7 and 6M. The expression differences between P7 and 6M were statistically evaluated for each reference gene in every tissue by means of Student’s t test (Fig. 2). It is remarkable that all considered reference genes showed a significantly lower level of expression in the adult hemibrain in comparison to the young one (Pgk1, 18s, Tfrc, Gapdh, Gusb, and B2m obtained P values of 0.0007, 0.0007, 0.0003, 0.0369, 0.0002, and 0.01, respectively). Since the six candidate reference genes resulted all significantly downregulated in the adult hemibrain, we first checked if this decrease could be related to an unreliable RNA extraction or to RNA degradation prior the retrotranscription. RNA yield was approximately the same for P7 and 6M hemibrains (2,472.496 ± 1,120.922 ng/μl for P7 samples and 2,949.746 ± 820.649 ng/μl for 6M samples), without any significant difference (P = 0.5). Each reaction was performed with the same reagents and technical equipments on the two age group samples. Finally, the quantification of other genes (as Kcnc3) in the same two sample groups produced predictable results (data not shown). Thus, we excluded that differences in reference gene expression in mouse hemibrain could be linked to inaccurate experimental procedures.

Figure 2
figure 2

Variations of candidate reference gene expression at 6M relative to P7 in mouse brain regions. Symbols represent mean ± SE. Points are connected by solid lines to better illustrate the tendency to increase or decrease during development. Note the significant decrease of Gusb expression in the adult cerebellum and the significant decrease of all the candidate reference genes in the adult hemibrain (P < 0.05). The symbol legend applies to all panels

Full size image

With the exception of Gusb, which showed a significant downregulation in the adult cerebellum, all the candidate genes remained constant in every selected brain region. To facilitate the comparison of gene abundance in different tissues, a logarithmic histogram graph was designed to represent the mean level of expression of the candidate genes in each brain region (where P7 and 6M data were considered together; Fig. 3). The overall expression of the six candidate reference genes in different brain areas was quite constant, even if a tissue-specific behavior was evident for most of them. For example, every candidate gene showed the lowest expression level in medial septum. 18s rRNA and Gapdh transcripts were the most abundant reference genes in any brain region. They reached, respectively, a 3,500-fold and a 17-fold difference in expression in comparison to the other four housekeeping genes. Gusb was the least expressed gene in all mouse brain regions.

Figure 3
figure 3

Candidate reference gene expression in mouse brain regions. Logarithmic histogram graph representing the mean level of expression of the candidate genes in each brain region (where P7 and 6M data were considered as a unique group). Values are given as 2−CT

Full size image

Ranking of Candidate Housekeeping Genes

The stability of the transcriptional level was evaluated for each candidate gene first by simply comparing the CV of its expression level. A lower CV value (defined as the ratio of the standard deviation to the average 2−CT) corresponds to a higher intragroup stability. In parallel, the housekeeping genes were ranked by means of the NormFinder and GeNorm softwares. Tables 2, 3, 4, 5, 6, 7, and 8 include the results of the reference gene comparison in mouse hemibrain, cerebellum, brain stem, hippocampus, medial septum, frontal neocortex, and olfactory bulb, respectively. Genes were listed first according to their stability in the single time points (P7 and 6M, considered as two distinct and independent groups) and secondly according to their stability between the two ages. Thus, from the above-cited tables, it could be possible to identify the best reference gene in a single brain region at a single time point or during the maturation of the corresponding brain region. Finally, Table 9 contains the output of the global analysis on all P7, 6M, and P7 versus 6M samples: Genes were ranked according to their stability among different tissues in a single time point (P7 and 6M, considered as two distinct and independent groups) and then to their general stability between P7 and 6M in all the considered tissues.

Table 3 Comparison of the six candidate reference genes in mouse cerebellum
Full size table
Table 4 Comparison of the six candidate reference genes in mouse brain stem
Full size table
Table 5 Comparison of the six candidate reference genes in mouse hippocampus
Full size table
Table 6 Comparison of the six candidate reference genes in mouse medial septum
Full size table
Table 7 Comparison of the six candidate reference genes in mouse frontal neocortex
Full size table
Table 8 Comparison of the six candidate reference genes in mouse olfactory bulb
Full size table
Table 9 Candidate reference genes in mouse brain (global analysis of the six brain regions)
Full size table

Determination of the Optimal Reference Gene Number

The optimal number of reference genes which should be used for each experimental group (each analyzed tissue and age) was assessed by means of GeNorm, based on the measurement of the pairwise variation value. With the exception of the frontal neocortex at P7, the best normalization factor for all the analyzed brain regions included only the two most stable reference genes (supplementary Fig. 1 and Tables 2, 3, 4, 5, 6, and 7) both at P7 and at 6M.When P7 samples were considered versus 6M, the best normalization factor for the cerebellum, hippocampus, frontal neocortex, and medial septum included the three most stable reference genes (supplementary Fig. 1). In contrast, for the olfactory bulb the best normalization factor for P7 versus 6M included only the two most stable reference genes (supplementary Fig. 1). No normalization factor reached a pairwise variation value lower than the cutoff 0.15 for brain stem and hemibrain samples in the analysis P7 versus 6M (supplementary Fig. 1). Globally, the best normalization factor for the analysis of all the analyzed brain regions (considered as a unique experimental group) included four genes at P7 and three genes at 6M. When all the experimental groups (the different brain region samples at the two time points) were considered together, the optimal number of reference genes to be used was 4 (supplementary Fig. 1).

Expression Level of a Target Gene Influenced by the Choice of an Appropriate Reference Gene

In order to validate the method of gene stability assessment, we measured the correlation between the expression of the target gene Kcnc3 and those of the candidate reference genes in mouse cerebellum at P7 and 6M. Only Gusb variations were significantly correlated with Kcnc3 changes in expression during cerebellar maturation (P = 0.0278 and correlation coefficient = −0.808). In contrast, no correlation was observed between Kcnc3 levels and the expression data of the other candidate genes (P > 0.05). To exemplify the consequence of using an unstable internal control, we determined Kcnc3 mRNA expression in mouse cerebellum, where Gusb showed a significant age-dependent variation. We normalized the expression of the target gene Kcnc3 to those of the six candidate genes and to the normalization factor NF3 (the geometric average of the expression data of Pgk1, Tfrc, and B2m, the three most stable reference genes identified by GeNorm software). When normalized to stable reference genes (Pgk1, 18s, Tfrc, Gapdh, and B2m) or to the normalization factor NF3, the Kcnc3 transcript showed a sevenfold increase in expression during cerebellar maturation (Fig. 4). On the contrary, since the expression of Gusb significantly decreased during cerebellar maturation, the normalization of Kcnc3 expression data to Gusb caused an erroneous overestimation of Kcnc3 increasing trend.

Figure 4
figure 4

Expression of Kcnc3 in cerebellum at 6M relative to P7: comparison of normalization with different reference genes. Histograms represent Kcnc3 expression differences (±SE) in mouse cerebellum between P7 and 6M, depending on different normalization procedures. Gene expression was calculated as multiple of the expression in P7 tissue, that was set 1.0 (dashed line). When normalized to stable reference genes (Pgk1, 18s, Tfrc, Gapdh, and B2m) or to the normalization factor NF3 (the geometric average of the expression data of Pgk1, Tfrc, and B2m, the three most stable reference genes identified by GeNorm software), the Kcnc3 transcript showed a sevenfold increase in expression during cerebellar maturation. It is notable that, since the Gusb expression significantly decreased during cerebellar maturation, the normalization of Kcnc3 expression data to Gusb (hatched column) caused an overestimation of Kcnc3 increasing trend

Full size image

Pgk1 Expression in Mouse Brain

Real-time RT-PCR experiments were performed to carry out an extended time course analysis of Pgk1 expression in mouse hemibrain at different ages from E14 to 6M (including P0, P7, P14, and P60) and in the different abovementioned brain regions at P7, P14, P60, 6M, and 18M (Figs. 5 and 6). The supplementary Table 17 includes the average and the standard error (SE) of the collected expression data at the different considered time points. As shown in Fig. 5, Pgk1 level increased significantly (P < 0.05) in the brain stem, frontal neocortex, and hippocampus in the second week of life, while in every other considered tissue, its expression did not significantly vary during the mouse lifespan. Results obtained from the linear regression analysis and the Pearson product moment correlation test are shown in supplementary Table 17; it is notable that Pgk1 increase in expression correlated with the age only in cerebellum and in olfactory bulb (P < 0.05).

Figure 5
figure 5

Pgk1 expression levels in single mouse brain regions. Time course of the Pgk1 expression levels, shown as 2−CT values, obtained by real-time RT-PCR on five age group samples, including P7, P14, P60, 6M, and 18M in mouse cerebellum (a), brain stem (b), hippocampus (c), medial septum (d), frontal neocortex (e), and olfactory bulb (f). Each point is from a determination in a single animal. Solid, dotted, and dashed lines represent, respectively, the regression line, the 95% confidence intervals, and the mean

Full size image
Figure 6
figure 6

Pgk1 expression in mouse brain. Global analysis of Pgk1 expression in all the analyzed mouse brain regions (a) and in the whole hemibrain (b). Each point is from a determination in a single animal. Solid, dotted, and dashed lines represent, respectively, the regression line, the 95% confidence intervals, and the mean

Full size image

Discussion

Settings of the Experimental Work

Since a growing number of real-time RT-PCR-based studies continue to be published in order to highlight genes specifically involved in brain development, maturation, and function, the identification of stable reference genes suitable for this kind of experiments is now an urgent need for the neuroscientist community. On the base of the published data, we decided to test the suitability of some common reference genes (such as Gapdh, 18s, and B2m) and of some relatively new control genes (such as Pgk1, Tfrc, and Gusb) during the development, maturation, and aging of mouse brain. These candidate reference genes were chosen on the base of their different physiological roles (Table 1), to reduce the risk of being coregulated during mouse brain development and aging. First, we compared the stability of the six candidate housekeeping genes between the P7 and the adult stage (6M). Secondly, we deepened the study of Pgk1 expression throughout the whole mouse lifespan, from the E14 to the mature age (6M) for the experiments on the whole mouse brain and from the first week of life (P7) to the elderly time point 18M for the experiments on the single mouse brain parts (cerebellum, brain stem, hippocampus, medial septum, frontal neocortex, and olfactory bulb).

Data Analysis

For the comparison of the candidate reference genes, data extracted from real-time RT-PCR runs were analyzed first by means of a classical statistical test, such as the unpaired two-tail Student’s t test. A more detailed discussion about the subsequent methods (the CV, NormFinder, and GeNorm approaches) used to assess gene variability should be done. The CV of gene expression data was calculated as the ratio standard deviation/mean, so that it gives an estimation of variability not affected by the absolute mean level of expression. Therefore, it was chosen as the simplest parameter to compare the gene variation. In parallel, we used the two most used and reliable programs for the identification of stable reference genes: NormFinder (Andersen et al. 2004) and GeNorm (Vandesompele et al. 2002). NormFinder applies a mathematical model in order to estimate not only the overall variation of the candidate genes (as the CV method does) but also the intra- (the variation inside any subgroups of the sample set) and the intergroup (the variation among sample subgroups of the sample set) expression variations. GeNorm applies a pairwise comparison approach that ranks the candidate reference genes according to the similarity of their expression profiles. Since the three approaches used are based on different mathematical models, the results obtained by means of the three methods were in some cases different. Moreover, since the possible coregulation of the candidate reference genes could affect the reliability of the GeNorm method, the parallel use of NormFinder and GeNorm softwares avoided the risk to obtain inconsistent results. It is notable that in almost all brain regions, the gene ranking obtained by the two programs was identical or very similar, demonstrating that the selected reference genes were not coregulated in our experimental model.

Although significant results have been obtained from our work, since our study did not start from a more extended study, such as microarray analysis, we cannot exclude the existence of genes which could be better references than those we used in our analysis.

Ranking of Candidate Housekeeping Genes in Mouse Brain

Surprisingly, all the candidate genes showed a significant downregulation in the adult mouse hemibrain in comparison to the young one. Therefore, in this first analysis, no one of the candidate genes showed to be an optimal reference gene for the whole mouse brain tissue. Anyway, in order to have an idea about the relative stability of our genes in this tissue, we applied NormFinder and GeNorm approaches on our data: Pgk1 gene was the top listed at P7, while Gusb was the best reference at 6M. Moreover, Pgk1 was the most stable gene between P7 and 6M. Our results are substantially in contrast with the past literature, where Gapdh was undoubtedly the most used gene and Pgk1 was never used in mouse whole brain tissue (supplementary Table 1).

Concerning the analysis on the specific brain regions, no one of the tested reference genes showed to be a suitable control for all the considered mouse brain areas. This fact was quite predictable, since the brain parts selected for our study differed for function, cell populations, embryonic derivation, and developmental periods. With the exception of Gusb, all the considered control genes showed no significant differences during the maturation of single brain regions. Unexpectedly, although Gapdh was one of the most regularly used genes in literature, NormFinder and GeNorm ranked it at the top of the list only in the frontal neocortex. Our results are in accordance with a recent work (Tanic et al. 2007), which demonstrated the suitability of Gapdh for the normalization of gene expression data in cortical samples of aged rats. In most of the other brain regions, Gapdh showed a rather high variability. For example, interestingly, Gapdh was one of the less reliable reference genes during the cerebellar growth. A number of studies (Sawa et al. 1997; Ishitani et al. 1998; Chen et al. 1999;Yang et al. 2007) demonstrated that Gapdh transcript and protein levels were altered in neuronal apoptosis. Since the apoptosis of the granule cells in mouse cerebellum peaks just between P5 and P15 (Wood et al. 1993; Lossi et al. 2004), an involvement of Gapdh in this kind of developmental process could be responsible for its low stability. The development of mouse cerebellum is in large part a postnatal series of events; thus, any gene involved in neuronal migration, apoptosis, growth cone extension, and synaptogenesis could show considerable changes during the first 3 weeks of life. A similar consideration could be done for those brain regions in which neurogenesis persists in the adult, such as the olfactory bulb and the hippocampus (Bagley et al. 2007; Enhinger and Kempermann 2008): Since specific neuronal populations (periglomerular and granular interneurons in the olfactory bulb, granule cells in the dentate gyrus of hippocampus) continue to be generated and the circuitries are continually renewed, any statement about a role of a gene in the maturation of these two regions should be done with caution. Perhaps, the relative unstability of Tfrc in the hippocampus and, more significantly, in the olfactory bulb could be due to its essential role in cell proliferation (Mescher and Munaim 1988; Jones et al. 2006).

Ribosomal RNAs, such as 18s, 28s, 5s and 16s, are very common control genes in gene expression studies (supplementary Table 1), as they are abundant and necessary in any cell type. Therefore, in order to have a representative of the rRNA category, we included the 18s rRNA in our set of candidate genes. The 18s rRNA was absolutely the most expressed reference gene in our panel, but its stability values differed largely depending on the brain region. Since, in principle, the 18s rRNA and the other members of the rRNA class are not mRNA, some authors did not consider them as real internal controls for gene expression studies. However, since in our study, the 18s rRNA showed to be one of the best references during the maturation of frontal neocortex and olfactory bulb, we could recommend its use, at least in association with other reference genes.

Determination of the Optimal Reference Gene Number

By means of GeNorm, the optimal number of reference genes which should be used for each analyzed tissue and age was assessed based on the pairwise variation between sequential normalization factors containing an increasing number of genes. A large pairwise variation value indicates that the inclusion of an additional reference gene in the normalization factor produces a significant effect and thus should be included for calculation of a reliable normalization factor. According to the GeNorm manual, a pairwise variation value of 0.15 was considered as the cutoff value, below which the inclusion of a further reference gene is not necessary. Interestingly, in most of the analyzed experimental groups, the best normalization factor included only the two or the three most stable reference genes identified by GeNorm. Our results are in agreement to Vandensompele’s considerations (Vandensompele et al. 2002), which recommend, in most of the cases, the minimal use of the three most stable genes for the calculation of the normalization factor. It is notable that, in the analysis P7 versus 6M, no normalization factor reached a pairwise variation value lower than the cutoff 0.15 for brain stem and hemibrain samples.

Expression Level of a Target Gene Influenced by the Choice of an Appropriate Reference Gene

In order to validate the method of reference gene selection, it is important to show the appropriateness of the control genes by comparing their expression to that of a known target gene. In our study, the expression level of the target gene Kcnc3 in mouse cerebellum was used for this purpose. The Kcnc3 voltage-gated potassium channel subunit is expressed and functional since P7 in mouse cerebellar neurons (Goldman-Wohl et al. 1994; Sacco et al. 2006) and is indispensable for firing action potentials at high frequency (Rudy and McBain 2001). The developmental regulation of Kcnc3 transcript level is known: It was demonstrated to increase in parallel with the maturation of Purkinje cells during the postnatal development and to be expressed in all terminally differentiated Purkinje cells and in the adult deep cerebellar nuclei (Goldman-Wohl et al. 1994).

We assessed the correlation between the expression of the target gene Kcnc3 and those of the candidate reference genes in mouse cerebellum at P7 and 6M. Only Gusb variations were significantly correlated with Kcnc3 changes during cerebellar maturation, while no correlation was observed between Kcnc3 and the other candidate genes. Similarly, the normalization of Kcnc3 expression data to Gusb, which is significantly downregulated during cerebellar maturation, caused an erroneous overestimation of Kcnc3 increase, although not affecting the qualitative interpretation of gene expression data. Since the estimation of the optimal number of housekeeping genes performed by GeNorm identified a normalization factor including the three best reference genes as the best option (supplementary Fig. 1), we normalized the expression of the target gene Kcnc3 also to the normalization factor NF3 (the geometric average of the expression data of Pgk1, Tfrc, and B2m, the three most stable reference genes identified by GeNorm software). Although the normalization to the NF3 reduced significantly the variability (measured as SE) of Kcnc3 expression data, in our experimental model, the use of this normalization factor produced similar results to the normalization to the single stable reference genes (Pgk1, 18s, Tfrc, Gapdh, and B2m).

Pgk1 expression in mouse brain

Globally, in our study, the less used reference genes (Pgk1, Tfrc, and Gusb) showed to be most stable than the more common ones, like Gapdh and 18s rRNA. Therefore, we could recommend their use in gene expression studies. Above all, Pgk1 was identified as the most stable control gene during the maturation of mouse whole brain tissue, cerebellum, and hippocampus. The Pgk1 gene codes for a kinase involved in carbohydrate metabolism, whose deficiency or mutation causes a human inherited disease characterized by hemolytic anemia, muscle stiffness, and variable defects in the central nervous system (Flanagan et al. 2006; Svaasand et al. 2007). To our knowledge, it was used only twice as reference for the nervous tissue in literature (Pohjanrvirta et al. 2006; Johansson et al. 2007).

Thus, our study is the first survey concerning its expression during the maturation and the aging of mouse brain regions.

Pgk1 expression showed a significant change only in brain stem, frontal neocortex, and hippocampus between P7 and P14, while in every other considered tissue, its level did not significantly vary during the mouse lifetime. A further analysis of Pgk1 stability was carried out by performing linear regression analysis of the expression data, in order to quantify the dispersion of our experimental points and to appreciate a trend in data. Moreover, we performed the Pearson product moment correlation test in order to highlight a possible link between the age and Pgk1 expression; it is notable that Pgk1 expression showed a slight correlation with the age in cerebellum and a more significant correlation in the olfactory bulb. However, our experimental data demonstrated that, in all other conditions and regions, Pgk1 displayed a high stability. Although no one of the tested reference genes showed to be a suitable control for all the considered mouse brain regions, Pgk1 could be considered a good internal control for mouse nervous tissue, in particular for those studies which concern adult and aging brain. In this study, for the first time in literature, we brought to light the specific time points and brain regions in which its expression changed significantly, in order to provide scientists with information necessary to guide their choice for quantitative experiments of gene expression in mouse brain.