Evaluating the applicability of mouse SINEs as an alternative normalization approach for RT-qPCR in brain tissue of the APP23 model for Alzheimer’s disease

BACKGROUND
The choice of appropriate reference genes (RGs) for use in reverse transcription quantitative polymerase chain reaction (RT-qPCR) has been thoroughly investigated, since the inclusion of unstable RGs might cause inaccurate gene expression results.


NEW METHOD
Short interspersed nuclear elements (SINEs) such as B elements, might represent an alternative solution given the high occurrence of these repetitive elements in the rodent genome and transcriptome. We performed RT-qPCR to investigate the stability of nine commonly used RGs and two B elements, B1 and B2, across different age- and genotype-related experimental conditions in the hippocampus and cortex of the APP23 amyloidosis mouse model for Alzheimer's disease. Gene stability was assessed using geNorm, NormFinder and BestKeeper. Human amyloid precursor protein (APP) levels in transgenic versus wild-type animals were also determined to validate the use of B elements as an alternative normalization approach.


RESULTS
Whereas B elements were stably expressed in the hippocampus, they were ranked as least stable in the cortex. The optimal normalization factor (NF) in hippocampus was a combination of Gapdh and Rpl13a, whereas in cortex, Actb and Tbp constituted the ideal NF.


COMPARISON WITH EXISTING METHOD
When comparing B1 and B2 as NFs for APP with the optimal panel of RGs in hippocampus, we found that B1 and B2 performed similarly to the optimal NF, while these SINEs performed less well in cortex.


CONCLUSIONS
Although B elements are suitable as an alternative normalization strategy in the hippocampus, they do not represent a universal normalization approach in the APP23 model.


Introduction
The reverse transcription quantitative polymerase chain reaction (RT-qPCR) is one of the most commonly used techniques in the field of molecular biology (de Jonge et al., 2007), and has become the gold standard for quantification of mRNA in biological samples (Bustin, 2000;Bustin et al., 2005) given its flexibility, sensitivity, specificity and the possibility for high-throughput analyses. Quantitative determination of gene expression is either relative or absolute, however, both methods are associated with a number of pitfalls. Since absolute quantification requires the inclusion of a standard curve (Bustin, 2000;Fronhoffs et al., 2002;Pfaffl and Hageleit, 2001), a major downside of this method is the need for such a serial dilution on each plate, thus hampering large RT-qPCR studies incorporating many samples and/or genes. In contrast, the delta-delta Cq method (Livak and Schmittgen, 2001) for relative quantification normalizes expression of the gene of interest to one or more endogenous reference genes (RGs), serving as internal controls. The internal control genes should ideally be constitutively and stably expressed across cell/tissue types, developmental stages, age and treatment conditions.
The selection and validation of appropriate RGs received a lot of attention in the previous decade (De Spiegelaere et al., 2015;Guénin et al., 2009;Rydbirk et al., 2016;Svingen et al., 2015), as the inclusion of unstably expressed RGs might give rise to an important source of bias. As such, commonly used RGs including glyceraldehyde-3-phosphate dehydrogenase (GAPDH), beta-actin (ACTB), and 18S rRNA have repeatedly been shown to be variably expressed across experimental parameters and tissues (de Jonge et al., 2007;Eissa et al., 2016;Svingen et al., 2015). Another topic of discussion was the ideal number of RGs to be included in the experimental design. Following the publication of the MIQE (Minimum Information for Publication of Quantitative Real-Time PCR Experiments) guidelines for studies reporting RT-qPCR results, which recommended the inclusion of at least two RGs (Bustin et al., 2009), a restricted literature review investigated the number of RGs used in gene expression studies in the period from 2010 until 2015, showing that merely 13% of all gene expression studies used more than one RG (Chapman and Waldenström, 2015). Another survey investigating papers published from 2009 until 2011 in journals with a wide range of impact factors (IF), noted that almost 30% of journals with an IF < 5 and 73% of high-impact journals failed to publish a single paper including validated RGs (Bustin et al., 2013). The notion that (the number of) appropriate RGs should be validated for each experimental design, is currently becoming increasingly accepted in the field. However, the repeated validation of RGs for each new study entails important disadvantages, as it renders genetic research more expensive and labor-intensive. Limited RNA sample availability also hinders an extensive investigation of appropriate RGs. These drawbacks, in addition to the unstable expression of historical RGs (e.g. GAPDH, ACTB and 18S rRNA), motivate the search for alternative normalization approaches. Short interspersed nuclear elements (SINEs) are a class of highly occurring retrotransposons, which are generally 100-500 base pairs (bp) in length (Ichiyanagi, 2013;Singer, 1982). Since SINEs make up approximately 10% of the total mammalian genome (Bovine Genome Sequencing Analysis Consortium et al., 2009;Mouse Genome Sequencing Consortium et al., 2002;Lander et al., 2001;Lindblad-Toh et al., 2005;Mikkelsen et al., 2007) and since they are mainly located in intronic and untranslated regions of many genes (Tsirigos and Rigoutsos, 2009), it is hypothesized that temporary or event-related changes in a certain number of genes will not have a large impact on the overall SINE content in the transcriptome (Renard et al., 2018;Crans et al., 2019). Therefore, SINEs might represent a valuable alternative to the regularly used RGs. As an example, expressed Alu repeats were previously found to be an accurate normalization tool for RT-qPCR studies in human blood (Marullo et al., 2010), human embryonic stem cells (Vossaert et al., 2013), and cancer cells (Rihani et al., 2013). This family of retrotransposons, derived from 7SL RNA (Ullu and Tschudi, 1984;Weiner, 1980), is ubiquitously present in the primate genome (Deininger, 2011), and contains a common restriction site for AluI (Houck et al., 1979). Due to the ability to integrate in various regions of the genome, Alu repeats can regulate gene expression by acting as transcriptional enhancers (Sasaki et al., 2008), by influencing pre-mRNA splicing (Lev-Maor et al., 2003;Sorek et al., 2002) or other documented mechanisms (for review, see Brosius, 1999;Elbarbary et al., 2016;Tomilin, 1999).
The identification of a second family of SINEs, present in the rodent genome (Kramerov et al., 1979;Krayev et al., 1980Krayev et al., , 1982, arose from the analysis of the structural features of pre-mRNA (Ryskov et al., 1972). Its nomenclature is based on the homology to double-stranded (ds) regions in pre-mRNA, termed as dsRNA-B molecules (Kramerov et al., 1979;Ryskov et al., 1972Ryskov et al., , 1973. Further analysis of the DNA sequences complementary to dsRNA-B showed two abundant categories, B1 and B2 (Kramerov et al., 1979), each with a distinct degree of sequence variation (Kramerov et al., 1979), and derived from distinct RNA ancestor molecules. The first category, B1, consists of sequences with a total length of 130 bp (Krayev et al., 1980). Similar to Alu repeats, B1 elements are derived from 7SL RNA and were termed quasidimers because of an internal duplication of 29 bp (Labuda et al., 1991;Veniaminova et al., 2007). In contrast, B2 sequences, which are generally 190 bp in length, originate from tRNA (Daniels and Deininger, 1985;Sakamoto and Okada, 1985). Analogous to the Alu repeats, B elements are hypothesized to be involved in splicing, processing of RNA polymerase II transcripts, and the regulation of gene expression (Krayev et al., 1982). Thus, since B1 and B2 elements are well characterized and because of their ubiquitous presence in the rodent genome and transcriptome, these SINEs might represent an interesting alternative to classical RGs in RT-qPCR experiments. In this paper, we investigated the applicability of nine commonly used mouse RGs: Gapdh, Actb, beta-2-microglobulin (B2m), phosphoglycerate kinase 1 (Pgk1), hypoxanthine guanine phosphoribosyl transferase (Hprt), tyrosine 3-monooxygenase/tryptophan 5-monooxygenase activation protein zeta (Ywhaz), beta-glucuronidase (Gusb), TATA-box binding protein (Tbp) and ribosomal protein L13a (Rpl13a), as well as B1 and B2 elements as a normalization strategy for gene expression analysis in hippocampal and cortical tissue of the APP23 amyloidosis mouse model for Alzheimer's disease (AD). Although various work reported about methylation of B elements (Tommasi et al., 2013), posttranscriptional gene regulation (Lucas et al., 2018), and their transcriptional response to viral cell infection (Williams et al., 2004), few previous studies used these SINEs as a normalization approach for RT-qPCR experiments. Only one recent study performed a similar experimental analysis concerning the applicability of B elements for normalization of gene expression in hippocampus and cortex of a rat model of temporal lobe epilepsy (Crans et al., 2019), while, to the best of our knowledge, these SINEs have not previously been used for analogous purposes in mice.

Animal model and tissue collection
The APP23 model is a transgenic mouse model, containing human cDNA of amyloid precursor protein gene (APP) with the Swedish double mutation (K670N/M671L), resulting in overexpression of APP and causing a familial form of early-onset AD in humans. From the age of 6 months onwards, APP23 mice display neuropathological lesions, i.e. Aβ plaques and tau-related pathology (Sturchler-Pierrat et al., 1997;Sturchler-Pierrat and Staufenbiel, 2000;Van Dam and De Deyn, 2011). Further characterization with respect to cognitive and behavioral abnormalities, and validation of this model, have been described previously (Van Dam and De Deyn, 2011;Van Dam et al., 2005). Male mice were group-housed in standard mouse cages in the animal facility of the University of Antwerp, in a 12:12 h light-dark cycle (lights on at 8:00 AM, lights off at 8:00 PM). Room temperature was maintained constant at 22 ± 2°C, while the humidity was 55 ± 5%. Food and water were supplied ad libitum. Genotyping was performed on tissue derived from ear punches using custom primers (Biolegio, Nijmegen, The Netherlands). Heterozygous (HET) (n = 5) and wild-type (WT) (n = 5) mice belonging to each age group (6-8 weeks, 6, and 24 months) were euthanized by cervical dislocation, followed by brain collection and regional dissection of each hemisphere using binoculars. First, both bulbi were dissected, followed by the cerebellum and pons/brain stem. Cortex tissue was sliced off using a scalpel, and as such removed from the underlying white matter. In addition, cortical tissue was visually checked for remaining pieces of white matter. A small sagittal midline incision was made to detach the left and right hippocampus from the fornix, after which a forceps was used to remove both hippocampi in the caudolateral direction. Tissue samples were immediately snapfrozen in liquid nitrogen and stored at −80°C until further use. All procedures involving animals were approved by the local ethics committee for laboratory animal experiments (file number 2016-06) and complied with the European Communities Council Directive (2010/63/ EU) and ARRIVE guidelines.

RNA extraction and cDNA synthesis
All analyses were performed on left hemispheric cortices (n = 30) and hippocampi (n = 29), derived from HET and WT animals aged 6-8 weeks, 6 and 24 months, except for one right hemispheric hippocampus implemented in the 24-month-old group. RNA extraction was performed using the RNeasy ® Plus Universal Mini kit (Qiagen, Hilden, Germany) following the manufacturer's instructions. In addition, oncolumn DNase treatment was performed using the RNase-free DNase set (Qiagen, Hilden, Germany) according to the manufacturer's instructions. RNA eluate volume varied between 22 and 25 μl. RNA concentration and purity were determined by UV-VIS spectroscopy measurements at 230, 260 and 280 nm with the Nanodrop 1000 instrument (ThermoFisher Scientific, Waltham, MA, USA). Individual tissue weights, RNA yield and purity are comprised in Table A.1 (Appendix A). Reverse transcription was performed on 800 ng RNA, using the iScript cDNA Synthesis kit (Bio-Rad Laboratories, Hercules, CA, USA) in 20 μL reactions. To analyze expression stability of B elements and Hprt in hippocampus, cDNA samples were diluted 1:1000, while a dilution of 1:10 was adopted for Actb, B2m, Gapdh, Gusb, Pgk1, Tbp, Rpl13a and Ywhaz. In addition, cDNA was diluted 1:10 for Actb, B2m, Gusb, Hprt, Pgk1, Tbp, Rpl13a and Ywhaz, and 1:1000 for Gapdh and B elements in the case of cortical samples. Expression levels of the human APP construct as well as mouse App were also analyzed, and normalized to the best combination of RGs according to gene stability software, or to B elements. For this analysis, cDNA was diluted 1:10 and 1:1000 for hippocampus and cortex, respectively. All cDNA samples were stored at −80°C until further use.

Primer design
Primers for all RGs were developed and checked for specificity using the NCBI Primer-BLAST software (Table 1) and RefSeq mRNA sequences of Mus musculus, while sequences for B1 and B2 elements were designed using Primer3 software, and checked for specificity using NCBI Primer-BLAST (J.V., UGent). All primers were synthesized by Biolegio (Nijmegen, The Netherlands).

Reverse transcription quantitative PCR
Reverse transcription quantitative PCR analyses were performed on the Applied Biosystems StepOnePlus instrument (Foster City, CA, USA), with automatic threshold settings. All analyses were carried out in 10 μL reactions containing 5 μL of the Power SYBR Green Master Mix (Applied Biosystems, Foster City, CA, USA), 1.5 μL of each primer (250 nM final concentration) and 2 μL cDNA (8 ng and 0.08 ng cDNA in case of a 1:10 dilution and 1:1000 dilution, respectively). Cycling conditions for all analyses consisted of initial denaturation at 95°C during 10 min, followed by 40 cycles of denaturation at 95°C for 15 s and annealing at 60°C for 60 s. Finally, melting curve analysis was performed by heating from 60°C to 95°C in increments of 0.3°C/s. cDNA samples and corresponding negative control reactions to which no reverse transcriptase was added during cDNA synthesis (-RT reactions), were analyzed in triplicate using StepOne Software v2.3 (Applied Biosystems, Foster City, CA, USA). All corresponding -RT reactions were negative (defined as reactions with an undetermined Cq or if the difference between the -RT control and the corresponding cDNA sample was ≥ 10 Cq values). PCR efficiencies for each assay were determined using technical triplicates of 1:2 or 1:10 serial dilutions, consisting of up to six dilution points, of a representative cDNA sample and are displayed in Table 1. Normalized relative quantities (NRQs) of B elements and APP were calculated using the qbase+ software (version 3.1, Biogazelle), implementing an efficiency-corrected delta-delta Cq method (Hellemans et al., 2007). Melt curves of each RT-qPCR product are shown in Appendix A.

Software and statistical analysis
Gene stability analysis was carried out on previously mentioned nine RGs, in conjunction with the two B elements separately and combined. We analyzed all RGs using qbase+ software, version 3.1, incorporating the geNorm algorithm (Vandesompele et al., 2002). This algorithm ranks distinct RGs based on the geNorm M value, which represents the average pairwise variation of remaining RGs after stepwise exclusion of the least stable genes. In addition, the geNorm V value indicates the pairwise variation of two sequential NFs upon inclusion of additional genes. Respective cutoff values for geNorm M and V are 0.50 and 0.15, so that all RGs with a geNorm M value < 0.50 are generally classified as stable RGs. A geNorm V value < 0.15 indicates that there is no benefit in including additional RGs for normalization purposes. Moreover, the NormFinder algorithm, v0.953, was used as an add-on in Microsoft Excel and ranks RGs taking into account a systematic error measure based on the inter-and intragroup variation. Similar to geNorm, the RG with the lowest stability value, is ranked as most stable (Andersen et al., 2004). Raw Cq values were transformed into efficiency-adjusted relative quantities before performing the NormFinder analysis, as described previously (De Spiegelaere et al., 2015). Lastly, the BestKeeper algorithm, version 1, was implemented in Microsoft Excel, calculating the geometric mean of all RGs included in the study as the BestKeeper Index, as well as the correlation coefficient of each individual gene with aforementioned index. Consequently, the gene with the highest correlation factor, is characterized by the most stable expression.
Since these three algorithms use distinct approaches to rank RGs, we compared the stability scores from geNorm, NormFinder and BestKeeper, and finally, compiled them into one final rank using the RankAggreg package (Pihur et al., 2009) in R, version 3.4.0 for Windows, using the cross-entropy Monte Carlo algorithm with Spearman's footrule distance (File A.1 in Appendix A).
Relative quantities of B elements and APP, either normalized to the optimal NF as determined by rank aggregation analysis, or to a NF containing B elements in the case of APP, were log-transformed for all statistical analyses. Test statistics were considered statistically significant if P < 0.05. To detect the effects of age and genotype on NRQs of APP and B elements, two-way ANOVA was applied, with Tukey's post-hoc tests for age, if applicable. In case the assumption of equality of variances was violated, separate one-way Welch's ANOVA and independent samples t-tests were performed. All statistical analyses were performed in IBM SPSS Statistics for Windows, version 24.0.

Sample size
In total, 30 mice were included in the RT-qPCR experiments. Age groups of 6-8 weeks, 6, and 24 months, consisted of 10 mice each, with 5 HET APP23 animals and 5 WT littermates in all age groups.

Expression profiles based on raw Cq values
Descriptive statistics of Cq values for each B element and RG in hippocampus and cortex are depicted in Figs. A.1 and A.2 (Appendix A). In hippocampus, almost all genes and B elements show a similar expression pattern across age and genotype groups. In animals aged 24 months, however, higher levels of B2m can be discerned relative to animals aged 6-8 weeks and 6 months. Overall, the variation in Cq values represented by the interquartile range is larger in cortical tissue compared to hippocampal samples for all tested genes. Surprisingly, the expression profiles of the B elements in the cortex showed the highest variability.

Stability ranking of reference genes according to geNorm, NormFinder and BestKeeper
For both tissue types, the ranking of RGs was largely similar between the three algorithms. The ranking of all RGs and separate B elements, as well as their gene stability scores, are displayed in Table 2. In hippocampal tissue, B1 and B2 performed relatively well, with a top 5 ranking in the gene list generated by each algorithm (Fig. 1A-C). Furthermore, almost all genes included in the analysis proved to be acceptable RGs, indicated by the finding that all RGs had a geNorm M value < 0.50. Nevertheless, caution should be exercised when choosing B2m as RG in hippocampus, since this RG shows considerably higher stability indices compared to the other RGs in both NormFinder and BestKeeper algorithms. Furthermore, B elements and B2m were amongst the least favorably ranked genes in cortical tissue, crossing the geNorm M threshold value. NormFinder and BestKeeper algorithms both appear to confirm this finding, with distinctly higher stability indices for B elements as well as B2m. In addition, when both B1 and B2 were included, B1 had a higher ranking compared to B2 in hippocampus, while both SINEs were ranked as the two least stably expressed transcripts in cortex (Fig. 1D-F). As the BestKeeper algorithm is limited to 10 genes, B2m was excluded from the analysis since it was the least stably expressed transcript according to both geNorm and NormFinder.

Rank aggregation analysis
Although the ranking of all analyzed genes was fairly similar within tissue type, some genes performed differently when ranks were compared across algorithms. A final overall ranking including either B1 or B2, indicated that in hippocampus of the APP23 model, Rpl13a showed the most stable expression, followed by Gapdh and Actb. When a combination of both B elements was included, B1 was ranked as the fourth most stably expressed gene transcript, while B2 was ranked sixth. In cortex, Actb, Tbp and Ywhaz, showed the most stable transcript levels overall, whereas B1 and B2 were ranked last.
Since B1 and B2 had a very distinct stability in hippocampus compared to cortex, they were excluded from analysis when an overall ranking across tissues and across algorithms was performed, to obtain an indication of the most stable gene transcripts in experiments investigating gene expression levels in cortex and hippocampus. This rank aggregation returned Actb as the most stably expressed transcript, followed by Gapdh and Ywhaz (Table 3).

Number of reference genes to be included
The pairwise variation between normalization factors (NFs) incorporating n and n+1 RGs, was lower than the proposed cut-off value of 0.15 for all n (Fig. 2). For hippocampal as well as cortical tissue, the pairwise variation between NFs containing two and three RGs was smaller than 0.15, indicating that the addition of a third gene was not required. However, the geNorm V values were consistently higher in cortex (Fig. 2B, Fig. A.3C, D in Appendix A) compared to hippocampus ( Fig. 2A, Fig. A.3A, B in Appendix A), regardless of the number of candidate RGs.

Normalized expression of B elements and APP
Since the results of the geNorm V analysis indicated that two RGs should be included in the normalization panel, and as rank aggregation indicated the best RGs in either tissue, we normalized expression levels of B1 and B2 to the optimal NF in hippocampus (Gapdh and Rpl13a) and cortex (Actb and Tbp). Both SINEs are characterized by a stable Table 1 Primer sequences, amplicon lengths and PCR efficiencies of every assay. Abbreviations: Actb: beta-actin; B2m: beta-2-microglobulin; bp: base pairs; Gapdh: glyceraldehyde-3-phopshate dehydrogenase; Gusb: beta-glucuronidase; Hprt: hypoxanthine guanine phosphoribosyl transferase; Pgk1: phosphoglycerate kinase 1; Rpl13a: ribosomal protein L13 A; Tbp: TATA-box binding protein; Ywhaz: tyrosine 3-monooxygenase/tryptophan 5-monooxygenase activation protein zeta.

Strengths and limitations
Our study comprised three equally sized age groups, from juvenile to old age, comprising WT and HET animals of the well characterized APP23 amyloidosis mouse model (Sturchler-Pierrat et al., 1997; Table 3 Final stability ranking of RGs after rank aggregation analysis implementing the cross-entropy Monte-Carlo algorithm with Spearman's footrule distance across algorithms and tissue types. Rank 1 corresponds to the most stably expressed RG, while rank 9 represents the least stable RG. Abbreviations: Actb: betaactin; B2m: beta-2-microglobulin; Gapdh: glyceraldehyde-3-phopshate dehydrogenase; Gusb: beta-glucuronidase; Hprt: hypoxanthine guanine phosphoribosyl transferase; Pgk1: phosphoglycerate kinase 1; RG: reference gene; Rpl13a: ribosomal protein L13 A; Tbp: TATA-box binding protein; Ywhaz: tyrosine 3-monooxygenase/tryptophan 5monooxygenase activation protein zeta.  Sturchler-Pierrat and Staufenbiel, 2000), thus introducing a considerable amount of biological variation and facilitating statistical analysis. In addition, nine commonly used RGs were analyzed in conjunction with two SINEs, using three independent gene ranking algorithms.
Since those three algorithms indicated slightly different rankings, rank aggregation was performed to retrieve a final ranking of all RGs and B elements combined, adding further reliability to our study. Apart from gene stability analysis, we also conducted a normalization approach implementing B elements to investigate the expression of APP, which is known to be overexpressed in HET relative to WT mice. Despite the fact that no protein-level analyses were conducted, our study adds relevant information to the characterization of the APP23 model, unambiguously reporting APP transgene mRNA levels in transgenic versus WT mice. A study aspect conceived as both a strength and a limitation, is that regional brain dissection was applied to collect brain samples. Although we have ample expertise in doing so and although this technique was performed with great precision, it gives rise to a larger amount of unwanted variability compared to other techniques such as laser capture microdissection. Moreover, the BestKeeper algorithm does not produce a gene ranking by itself. Other rankings, e.g. based on coefficient of variation and standard deviation, might give rise to alternative results. However, we believe that a gene ranking analysis should preferentially be based on the relation between genes or a measure taking into account distinct sources of variation, and, therefore, we chose to implement the correlation coefficient between each candidate RG and the BestKeeper Index.

Stability of B elements and commonly used reference genes in hippocampus and cortex
In both hippocampus and cortex, the results of the gene stability analysis were comparable between the three distinct algorithms, with B elements having stability values lower than the geNorm threshold value of 0.5 in hippocampus, while in cortex, their expression levels were less stable compared to the RGs, with stability values near or above the threshold. Two of the most implemented RGs in literature, Actb and Gapdh (Kozera and Rapacz, 2013;Suzuki et al., 2000), were in the top 5 of most stably expressed transcripts, with geNorm M values far below the threshold value, regardless of tissue type. Moreover, in the final rank merging all analysis methods and tissues, and including only commonly applied RGs, Actb and Gapdh were classified as the most stable candidate genes. These results do not entirely comply with previous reports on RG stability in the literature. A study investigating the   4. APP overexpression levels in hippocampus and cortex. For hippocampus, NRQs of WT (n = 15) versus HET (n = 15) animals were obtained using four NFs, including either B1, B2, a combination of B1 and B2, and the optimal combination of Rpl13a and Gapdh. In case of cortical tissue, 13 WT and 15 HET animals were included for the analysis of APP levels normalized to B1, B2, a combination of these B elements, and the optimal NF comprising Actb and Tbp. Data are represented as mean ± SD. Abbreviations: APP: amyloid precursor protein; Gapdh: glyceraldehyde-3-phosphate dehydrogenase; HET: heterozygous; NFs: normalization factors; NRQ: normalized relative quantities; Rpl13a: ribosomal protein L13 A; SD: standard deviation; WT: wild-type. expression stability of six commonly reported RGs in brain regions of mice at postnatal day 7 and at 6 months of age, found that Gapdh was ranked sixth of all analyzed genes in the latter age group, with a geNorm M value of 13.396 (Boda et al., 2009). However, it should be noted that 6-month-old CD1 mice were first anesthetized by inhalation of isoflurane, followed by decapitation and brain collection, while our results were obtained in transgenic mice with a C57BL/6 J background and in the absence of potentially interfering anesthetics. Indeed, inhalation anesthesia with isoflurane has been reported to induce changes in mRNA, even after a short time span (Staib-Lasarzik et al., 2014). In contrast, another report, focusing on the stability of RGs in the developing mouse brain, indicated that Actb was the second most stably expressed RG in the male developing C57BL/6 mouse brain across time points ranging from embryonic day 11.5 until 15.5, with a GeoMean score of 2.73 (Cheung et al., 2017). As our study pinpointed aging and disease progression rather than development, and since we used a distinct method of euthanasia along with a different mouse model, it is difficult to compare our results with the findings of these previous studies.

Tissue-specific differences in stability of B elements
When B1 and B2 were normalized to the ideal normalization panel in hippocampus and cortex, both SINEs showed stable expression across hippocampal samples. Although still moderately stable, these SINEs were characterized by a lesser stability in cortex compared to hippocampus. Surprisingly, no significant effects of age or genotype were found in normalized B1 nor B2 levels in the cortex, while their mean values vary considerably (Fig. 3). Possibly, the variation in normalized levels of the B elements was too high to detect a statistically significant difference. Although the macroscopic dissection method used for this work is not as accurate as laser capture microdissection, it is unlikely this influenced our findings. Despite the fact that all samples were visually checked for the presence of pieces of white matter, we cannot exclude the possibility that some traces were still present in the collected samples, thereby minimally altering the relative amount of cortex tissue. However, applying laser capture microdissection, would have made the tissue collection process more labor-intensive and timeconsuming. Secondly, epigenetic changes such as DNA methylation or histone modification are hallmarks of genomic instability (Li and Zhang, 2014) and might underlie the disparity with regard to the stability of B elements in hippocampus versus cortex. As such, B1 elements were found to be methylated in somatic (Ichiyanagi, 2013) and embryonic cells (Papin et al., 2017), with a higher degree of methylation and subsequent transcriptional repression, compared to B2 (Papin et al., 2017).
Besides B elements, human Alu sequences are also known methylation targets (Jang et al., 2017;Kochanek et al., 1993;Luo et al., 2014). In addition, these SINEs were previously shown to exhibit ageand tissue-specific differential methylation (Ianov et al., 2017;Maegawa et al., 2010;Singhal et al., 1987;Thompson et al., 2010). One study even showed that methylation differences between brain areas were more apparent than differences in methylation profile with respect to age, gender, postmortem delay, race, diagnosis, or cause of death (Ladd-Acosta et al., 2007).
Since we did not generate proof of the nature, nor the direction of epigenetic changes, we can only speculate that a tissue-specific differential degree of epigenetic signaling across age groups and, possibly, genotypes, underlies the differences in the stability of B elements in cortex versus hippocampus. However, to the best of our knowledge, no reports exist of distinct age-or genotype-related epigenetic regulation of SINE expression in brain areas of the APP23 model. Since the first amyloid beta (Aβ)-plaques appear in the frontal cortex of HET APP23 mice aged 6 months (Sturchler-Pierrat and Staufenbiel, 2000), preceded by an increase in soluble Aβ oligomers from the age of 6-8 weeks onwards (Janssen et al., 2016), one could hypothesize that this brain region displays early epigenetic changes as a consequence of initial Aβ exposure, as was previously suggested in female mice of the 5xFAD model (Griñán-Ferré et al., 2016;Oakley et al., 2006) and in murine cerebral endothelial cells exposed to synthetic Aβ 1-40 peptides (Chen et al., 2009). However, one would also expect similar alterations in hippocampal samples of older age groups of the APP23 model, as amyloid plaques are present in virtually all brain regions in HET APP23 mice aged 24 months (Sturchler-Pierrat and Staufenbiel, 2000). Thus, it remains unclear why B elements are less stably expressed in cortex compared to hippocampus.

APP overexpression to validate the use of B elements as a novel normalization approach
Previously, the transgene in APP23 HET animals was reported to be sevenfold overexpressed relative to endogenous App mRNA, as determined by semi-quantitative PCR and confirmed by western blot experiments (Sturchler-Pierrat et al., 1997;Sturchler-Pierrat and Staufenbiel, 2000). However, it is unclear if these expression levels were determined across several ages or in a single age group. The use of a semi-quantitative technique, along with the number of tissue samples included in these analyses, further hamper a straightforward comparison of our APP overexpression data with these initially reported results. However, regardless of age, APP was at least twofold overexpressed in HET relative to WT animals in the cortex of the youngest age group, while an increasing trend was observed in the older groups ( Fig. A.4, Appendix A). Conversely, tissue samples derived from hippocampus showed a relatively constant expression of APP in all age groups. Thus, as suggested in the original paper describing the APP23 model, crossing a certain threshold expression of APP bearing human disease-associated mutations, might represent an initial incentive for neuropathology (Sturchler-Pierrat et al., 1997). In addition, a specific threshold of two-to threefold APP overexpression was previously proposed for successful reproduction of pathological AD-like features (Guénette and Tanzi, 1999). In the youngest age group of this study, APP is also overexpressed in HET versus WT animals, although no plaque-related neuropathology is present at this age. However, as stated previously, high levels of soluble Aβ oligomers were found at this early stage (Janssen et al., 2016) as a possible result of initial APP overexpression. Lastly, when hippocampal APP levels were not only normalized to the ideal panel of RGs, but also to B1, B2, or a combination of both (Fig. 4), the overexpression of human APP was almost identical across all normalization strategies, indicating that these SINEs can indeed be adopted as a novel normalization approach in hippocampus. In cortex, the results of Fig. 4 appear to indicate that B1, B2 and a combination of these SINEs are superior to the previously determined optimal NF consisting of Actb and Tbp. However, one should bear in mind that this figure depicts normalized APP levels and that the large standard deviation of APP NRQs, when the optimal NF was used for normalization purposes, might just reflect a large biological variation of APP levels in the cortex of HET mice. In this aspect, B elements as NF in cortex might fail to indicate the actual results, leading researchers to draw incorrect conclusions. Conversely, a recent paper showed that 11 expressed repeat elements in mice, other than B1 and B2 (amongst others lower in copy number), were superior to commonly used RGs in the literature with respect to normalization purposes in a wide range of experimental setups (Renard et al., 2018). Thus, as contemporary research is providing growing evidence concerning the applicability of repeat elements with respect to normalization of mRNA expression levels in various conditions (Marullo et al., 2010;Renard et al., 2018;Rihani et al., 2013;Vanhauwaert et al., 2014), they might become increasingly popular as reference targets.
In conclusion, this study emphasizes the importance of validating RGs in distinct tissue types for optimal normalization results. The expression levels of B elements and APP varied to a much larger extent in cortex compared to hippocampus, which might be due to epigenetic alterations. In addition, we provide first evidence for the use of mouse B elements as an alternative normalization approach in hippocampus, but not cortex, of the APP23 amyloidosis mouse model.

Declarations of interest
None.