In the search for a stable tool for normalization, we focused on EARs because they are the most abundant repeats in the human genome. More than 1,500 human genes contain one or more Alu repeat in the UTRs regions. The EARs are about 280 nucleotides long and are divided into 30 well-conserved subfamilies (see 7 for a review). These features make EAR amplification a reliable tool for normalization in RT-qPCR based on the principle that the differential expression of a small number of genes bearing the EAR will not influence EAR abundance in the transcriptome. Figure 2a shows a BLAST representation of EAR element sequence conservation in the human genome. The small blue dots represent the position of EAR in the intergenic and intragenic regions and show that EARs are particularly located in the UTR and intronic regions. The Alu interspersed repetitive sequence and primers selected for the study are represented in Figure 2b while Additional file 1 shows the protocol for EAR amplification in the human transcriptome. As indicated, we first performed a retrotranscription (RT) reaction with total RNA at a final concentration of 12.5 ng/μl (see Materials and methods for additional details). The obtained cDNA was then diluted 1:200 and 5 μl of this solution tested using RT-qPCR (see Materials and methods). Notably, another advantage of this assay is in the low amount of cDNA used for the analyses, allowing many experimental replicates when the starting material (RNA) is limiting. As Figure 2c shows, we obtained a good RT-qPCR outcome, as indicated by the PCR efficiency (102%) and correct slope value (-3.27). In addition, another important outcome arises from the analysis of the amplified products with the melting curve. As Figure 2d shows, multiple melting peaks emerge from these analyses, and the expected peaks by amplified products (88,69 T) can be recognized. This result is anticipated given that several peaks are generated by the polymorphisms present in the EAR sequences and by different Alu subfamilies 7. We conclude that when working with EARs, different melting curves in different samples are to be expected and that for each sample the replicates have to share their melting curve.

As indicated above, the use of NF obtained from the geometric mean of three or more stable reference genes is suitable for normalization 1. We therefore decided to compare the EAR normalization tool to the normalization performed based on three more stable control genes expressed in blood and selected by the GeNorm algorithm. We started by choosing six different reference genes with expression in human blood that has been evaluated by RT-qPCR. The GeNorm algorithm was applied to calculate the M (see Table 1 for full gene name, accession number, chromosomal localization, indication that primers span an intron, and primer sequences). Genes with higher M values have greater variation in expression, and the threshold proposed for eliminating a gene as unstable was M ≥ 0.5.

On the basis of our analyses, the ranking of the expression stability in these genes was (from most to least stable): GNB2L1 (guanine nucleotide binding protein (G protein), beta polypeptide 2-like 1), HPRT1 (hypoxanthine phosphoribosyltransferase 1), YWHAZ (tyrosine 3-monooxygenase/tryptophan 5-monooxygenase activation protein, zeta polypeptide), β2M, GAPDH, and βACT (β-actin) (Figure 3a). As shown, the M values of GNB2L1, HPRT1, and YWHAZ were lower than 0.5; therefore, these genes were concluded to be stably expressed reference genes in human blood. The NF value, used for normalization of RT-qPCR, was calculated based on the geometric mean of the expression levels of these three most stable control genes.

To assess the validity of EARs as a normalization tool versus the GeNorm strategy, we evaluated BDNF mRNA levels normalized to EARs and to NF in blood from ten healthy volunteers. Figure 3b shows that the quantification of BDNF mRNA in human blood is comparable after normalization with the two different strategies. The median value shown in Figure 3c confirms this result: the median values of blood BDNF mRNA levels normalized versus EARs and versus NF are similar to each other (0.95 versus 1.1), while the median value of BDNF mRNA normalized versus GAPDH mRNA is much lower (0.65). These data indicate that EARs represent a good tool for normalization and that normalization over a single reference gene provides incorrect results.

Subsequently, we applied the EAR normalization tool for the quantification of BDNF mRNA in the eight control participants analyzed in Figure 1, in which the messenger level of the neurotrophin is represented relative to GAPDH or β2M mRNA content. Figure 3d shows the median value of BDNF mRNA normalized over GAPDH (1), β2M (1.6), or EARs (0.55). As shown, the distribution of the BDNF mRNA values normalized to EARs is tighter with respect to the distribution of BDNF mRNA normalized to GAPDH or β2M mRNA, indicating that the EAR normalization strategy removes most of the non-specific variations arising from the use of an inappropriate internal reference gene.

The experiments described above prove the reliability of the EAR normalization strategy with respect to GeNorm as well as the efficiency of this new method with respect to normalization to a single housekeeping gene to measure BDNF mRNA levels in human blood.

To evaluate the reliability of the EAR normalization strategy with other, more blood-specific genes, we analyzed IL8 (interleukin-8) mRNA levels in blood from control participants. Validation of EAR normalization for cytokine quantification in blood might be particularly useful for measuring the balance between pro- and anti-inflammatory cytokines in chronic inflammatory disorders or autoimmune disease. Cytokine mRNA quantification is therefore of clinical relevance to immuno-monitoring and for the evaluation of the disease status in patients.

In our measurements of IL8 mRNA levels in ten control participants, normalized versus EARs and versus NF, the IL8 mRNA levels with the two different approaches were similar in single human samples; also, again, the similar median value obtained (Figure 4b) confirmed the result.

We concluded that EARs are a suitable normalization tool for gene expression study of cytokines in human blood.

Fast, sensitive, and accurate quantification in RT-qPCR analysis is a crucial tool not only for assessing mRNA in human blood but also for evaluating the levels of given molecules in tissues such as post-mortem human brain. In a previous study, we analyzed BDNF mRNA levels in 38 post-mortem human cortices from control participants and HD patients using GAPDH as the reference gene 8. The GAPDH gene is, in fact, one of the more commonly used reference genes in such analyses 9.

Here we tested whether GAPDH is a stable control gene for brain tissues versus EAR and whether normalization over EAR still confirms the finding of a robust reduction in BDNF mRNA in HD cortical samples over controls 8. To evaluate the efficiency of GAPDH, we used the same RNA as in Zuccato et al. 8 and performed a GeNorm analysis on six different reference genes. As shown in Figure 5a, the ranking of the expression stability in these genes was (from most to least stable): GAPDH, YWHAZ, βACT, β2M, HPRT1, and GNB2L1. The M values of GAPDH were lower than 0.5, confirming that this gene is a stably expressed reference gene in human cortex and also validating the previously performed quantification of the BDNF mRNA in human HD samples versus controls 8. Using the same RNA, we next repeated the analysis of BDNF mRNA levels and used normalization to EARs or to GAPDH. Figure 5b shows that the BDNF mRNA level was lower in HD patients with respect to control participants (Mann-Whitney, P < 0.05, two-tailed test). These data demonstrate the efficiency of EAR normalization for quantifying a transcript of interest in other human tissues, such as cerebral cortex, that are normally subjected to a long post-mortem delay.

Moreover, we extended the application of EAR normalization strategy to samples from skeletal muscle, an accessible tissue that responds to hormonal, metabolic and neural inputs. The muscle, therefore, is a reasonable tissue to examine for gene expression changes that might be turned into biomarkers for neurodegenerative diseases 10. In particular, we have analyzed the expression levels of TNN1C (troponin C type 1) mRNA, a stable gene expressed in the skeletal muscle. The GeNorm algorithm was applied to evaluate three more stable control genes expressed in human muscle. On the basis of our analyses, the ranking of the expression stability in these genes was (from most to least stable): GNB2L1, HPRT1, YWHAZ, β2M, GAPDH, and βACT (Figure 6a). As shown, the M values of GNB2L1, HPRT1, and YWHAZ were lower than 0.5; therefore, these genes were concluded to be stably expressed reference genes in human skeletal muscle. The NF value, used for normalization of RT-qPCR, was calculated based on the geometric mean of the expression levels of these three most stable control genes. Importantly, in our measurements of TNN1C mRNA levels in four control participants, normalized versus EARs and versus NF, the TNN1C mRNA levels with the two different approaches were overlapping in single human samples. Figure 6b also shows a similar median value between the two measurements.

The data described above show that, in addition to peripheral blood, the EAR normalization strategy can be applied to samples from human brain and skeletal muscle, implying that this method could be used for assessing different human samples.

Individuals were recruited for the study from the National Hospital for Neurology and Neurosurgery, London (UK) and the National Neurological Institute-IRCCS "Carlo Besta", Milan (Italy). Full ethical review and consent was obtained for the research use of blood and skeletal muscle from healthy volunteers and all subjects gave informed written consent. Table 2 gives participant demographic characteristics.

RNA was extracted from whole blood using the PAXgene™ Blood RNA System Kit (PreAnalytiX, QIAGEN) following the manufacturer's guidelines. Total RNA concentration and purity were measured using a NanoDrop 1000 Spectrophotometer (Thermo Scientific, Waltham, MA, USA), and quality was verified by means of agarose gel electrophoresis of 1 μg of each sample (Additional data file 2). The total RNA was stored in aliquots at -80°C.

The muscle samples were homogenized in TRIZOL (Invitrogen, Carlsbad, CA, USA) using a Ultraturex homogeniser. Total RNA was isolated according to the manufacturer's protocol and stored in aliquots at -80°C.

Using Superscript III RNaseH reverse transcriptase (Invitrogen) and random primers in a volume of 20 μl, 250 ng total RNA was reverse-transcribed to single-stranded cDNA, according to the manufacturer's instructions.

Three independent PCR analyses were performed of each of two independent RT reactions, for a total of six independent measurements for each of the analyzed mRNA samples.

Because BDNF values did not have a normal distribution, the non-parametric Mann-Whitney two-tailed U-test was used for the statistical analyses, with significance set at P < 0.05. Median values were chosen because of the skewed distribution of the data. In each figure, the boundary of the box closest to zero indicates the 25th percentile, the line within the box marks the median or 50th percentile, and the boundary of the box farthest from zero indicates the 75th percentile. When ten or more samples were analyzed, whiskers above and below the box indicate the 90th and 10th percentiles.

The validation of the EAR normalization strategy with respect to the GeNorm algorithm was measured using Pearson correlations (r).

The materials and methods section and nomenclature follows the MIQE reporting guidelines 18.