miRNAs constitute a class of noncoding regulatory RNA that functions by binding to target RNAs 17 . We have conducted a bioinformatics search for miRNA binding sites in BACE1 mRNA and predicted the presence of a binding site for miR-485-5p in the sixth exon of BACE1 mRNA. Previously we showed that the same region of BACE1 mRNA may interact with a natural antisense transcript, BACE1-AS, and that there is potential for sense-antisense RNA duplex formation (Figure 1a, b). Considering RNA duplex formation between BACE1 and BACE1-AS, we postulated that an additional regulatory function of BACE1-AS may be 'masking' the miR-485-5p binding site and thereby blocking the inhibitory effects of this miRNA on BACE1 translation and mRNA decay (Figure 1a, b). Some other miRNA target sites were found in the overlapping region of the BACE1 mRNA; however, the assigned score and binding energy were not sufficient to be considered as strong candidates. We selected a number of these miRNAs, including miR-17-3p, miR-652, miR-593 and miR-183, and over-expressed these in our cellular model, with a beta-galactosidase reporter assay corresponding to BACE1 protein as a read-out (see below). We found that, unlike miR-485-5p, these miRNAs were not able to alter BACE1 protein concentrations (Figure 1c). Although these miRNAs did not pass our validation studies, we cannot completely exclude potential interactions between these miRNAs and BACE1 or BACE1-AS transcripts.

To validate the predicted miR-485-5p target site in BACE1 mRNA, we engineered a miR-485-5p target site downstream of a luciferase reporter gene. This sequence corresponds to the predicted target site of miR-485-5p on BACE1 mRNA. We found that the presence of this target site is sufficient for luciferase reporter protein reduction upon miR-485-5p over-expression (Figure 2a). We also constructed a luciferase reporter with either full complementary or mismatch target sites for miR-485-5p as positive and negative controls, respectively. Each one of these three luciferase constructs was transfected into HEK293T cells in the presence of miR-485-5p over-expressing or empty control vectors. We observed significant down-regulation of luciferase in the construct with miR-485-5p target sites in the presence of a miR-485-5p over-expressing vector. Our results indicate that the presence of our predicted miR-485-5p target site is sufficient for miRNA binding, suggesting the possibility of in vivo interactions between miR-485-5p and its cognate binding site in the sixth exonic region of the BACE1 mRNA.

Although luciferase reporters are extensively utilized as a validation tool for miRNA targets, these constructs have limitations for evaluating binding sites located in the coding region. To create a construct that resembles an in vivo setting, we cloned full-length BACE1 cDNA, excluding the 3' UTR, into a ProLabel C3 expression vector (DiscoveRx). The ProLabel C3 vector, upon transfection into mammalian cells, expresses BACE1 mRNA and protein with a small fusion tag. The protein tag is then used for the detection of protein synthesis utilizing the enzyme fragment complementation (EFC) system. Next, we examined the effect of miR-485-5p over-expression on BACE1 protein in HEK293T C3 cells using enzyme complementation protein quantification technology (DiscoveRx). We found that miR-485-5p over-expression causes a reduction in BACE1 protein concentrations (Figure 2b). Our results indicate that miRNA-binding sites in the coding parts of mRNAs may still be functional and further suggest the possibility of in vivo interactions between miR-485-5p and mature BACE1 mRNA.

To test the specificity of the reduction of BACE1 and to further validate the miR-485-5p target site in the coding region of BACE1, we applied a synthetic locked nucleic acid (LNA)-antimiR molecule to block the miRNA binding. Such antimiRs are synthetic, LNA-modified RNA molecules with sequence complementary to the mature miRNA. As expected, over-expression of miR-485-5p reduced the BACE1 protein levels and this reduction was blocked by application of LNA-antimiR-485-5p (Figure 3a). We observed that the LNA-antimiR increased BACE1 protein levels in our EFC reporter assay, which may possibly be explained by inhibition of endogenous miR-485-5p. The reversal of mir-485-5p-mediated BACE1 protein reduction by LNA-antimiR indicates the specificity of the miRNA effect and further validates the miR-485-5p binding site in the coding region of BACE1 mRNA.

Considering that BACE1-AS and miR-485-5p share potential binding sites in the BACE1 mRNA, we aimed to check the possible counteraction of these two ncRNA transcripts. If these two ncRNAs can compete for binding sites, then simultaneous over-expression should block the effect of miR-485-5p. Indeed, we noted that over-expression of BACE1-AS eliminated the effects of miR-485-5p and returned the BACE1 protein to basal levels (Figure 3b). We previously showed that over-expression of BACE1-AS caused an increase in BACE1 protein level and we were able to reproduce these data, using our in vitro EFC assay. On the other hand, we observed that miR-485-5p over-expression reduced the BACE1 protein levels. Simultaneous over-expression of both BACE1-AS and miR-485-5p returned the BACE1 protein level to the basal level. These data imply that miR-485-5p and BACE1-AS may compete for binding to BACE1 mRNA and support the novel regulatory role of masking a miRNA-binding site in BACE1 by the non-coding BACE1-AS transcript. This proposed miRNA masking effect is in concert with the concordant antisense regulatory action of BACE1-AS.

To confirm the expression of BACE1, BACE1-AS and miR-485-5p in brain and other human tissues, we performed real-time PCR (RT-PCR) on RNA samples from human and mouse. We observed that miR-485-5p is present and significantly higher in brain compared to other regions (Figure 4a). BACE1 and BACE1-AS transcripts showed ubiquitous expression patterns with minimal variation among different tissues. Similar results were observed in mouse tissues and various regions of mouse brain showed high expression of miR-485-5p, BACE1 and BACE1-AS transcripts (Figure 4b). The high concentration of miR-485-5p and BACE1-AS in the brain regions suggests the likelihood of their functional interaction with the BACE1 mRNA target site, and involvement in BACE1 regulation.

Next, we examined the expression of BACE1, BACE1-AS and miR-485-5p in four brain regions from human control subjects (Figure 4c). These RNA samples originated from post-mortem brains of 35 elderly individuals with an average age of 72.3 years (range 53 to 91 years) who had passed away from causes other than Alzheimer's disease. Although not all regions were available from all cases, we examined RNAs from cerebellum (18 subjects), entorhinal cortex (8 subjects), hippocampus (12 subjects) and superior frontal gyrus (18 subjects). Unlike BACE1-AS, miR-485-5p was two- to four-fold higher in entorhinal cortex, hippocampus and superior frontal gyrus compared to cerebellum. BACE1-AS transcript was expressed two- to three-fold lower in similar regions compared to cerebellum. It is worth noting that the transcript expression data represent the relative quantity of each RNA transcript to that of reference tissue (brain in Figure 4a, and cerebellum in Figure 4b, c). Therefore, these data do not directly support the conclusion that expression of miR-485-5p represses BACE1 transcript levels and that BACE1-AS expression enhances BACE1 transcript levels. Nevertheless, the relatively high expression of miR-485-5p and BACE1-AS in brain regions that are affected by Alzheimer's disease pathology may suggest a role for these ncRNAs in Alzheimer's disease-related pathogenesis.

We also examined the abundance of miR-485-5p in various human tissues by next generation sequencing of the small RNA fraction, using the Illumina Genome analyzer. Two individuals were used for sequence profiling of the small RNA fraction and identification of known miRNAs from a set of eight tissues. We found that the number of normalized reads for miR-485-5p was significantly higher in the orbital gyrus from brain compared to skeletal muscle, pancreas, lung, heart, liver, spleen and kidney (Figure 4d). The raw read count for miR-485-5p differed between the two individuals as follows: pancreas, 1.5%; lung, 1.6%; skeletal muscle, 1.5%; heart, 1.25%; brain, 1.4%; liver, 3.5%; spleen, 8.6%; and kidney (only one sample). The normalized reads from deep sequencing experiments provide absolute quantities, in contrast to relative quantity values obtained from RT-PCR methods. Therefore, the high expression of miR-485-5p in orbital gyrus of brain revealed by deep sequencing data confirms our RT-PCR findings. Moreover, the substantial abundance of miR-485-5p in the brain, compared to other tissues, suggests a neuronal-related function, and by reason of co-expression, increases the likelihood of involvement in BACE1 regulation.

We previously showed that the BACE1-AS transcript is significantly up-regulated in several brain regions of subjects with Alzheimer's disease. We measured the expression of BACE1, BACE1-AS and miR-485-5p in two different sets of RNA samples from control subjects and individuals with Alzheimer's disease. Initially, we examined the parietal lobe and cerebellum of 5 subjects with Alzheimer's disease and 5 normal elderly individuals (20 samples total). BACE1-AS, and to a lesser degree BACE1, transcripts were up-regulated in Alzheimer's disease patients compared to control individuals and miR-485-5p was down-regulated by 30% in parietal lobe and close to 60% in cerebellum of Alzheimer's disease patients (Figure 5a).

We have also examined a second set of RNA samples from 35 Alzheimer's disease and 35 control individuals (Figure 5b). Although not all regions were available from all cases, we examined RNA from cerebellum (18 control and 23 Alzheimer's disease subjects), entorhinal cortex (8 control and 11 Alzheimer's disease subjects), hippocampus (12 control and 12 Alzheimer's disease subjects) and superior frontal gyrus (18 control and 18 Alzheimer's disease subjects). Consistent with our previous work, BACE1-AS transcript concentrations were up-regulated in all four regions tested. To a lesser degree, BACE1 transcripts were up-regulated in entorhinal cortex as well as in superior frontal gyrus. On the other hand, miR-485-5p was significantly reduced in entorhinal cortex and hippocampus. However, miR-485-5p was not significantly altered in cerebellum and superior frontal gyrus (Figure 5b). The difference between miR-485-5p expressions in cerebellum of the two sets of RNA samples can conceivably be explained by the relatively high variability among human samples. Considering the increased level of BACE1-AS and reduction of miR-485-5p in the brain of Alzheimer's disease subjects, we postulated that dysregulation of these two ncRNAs might cause increases in BACE1 mRNA as well as the removal of the miRNA brake on BACE1 mRNA and protein expression.

Our data suggest an interaction between two classes of ncRNAs in the regulation of BACE1 gene expression. We sought to determine the extent of this computational regulatory mechanism as a general theme in the human genome. We selected a set of evolutionarily conserved sense-antisense pairs, previously published as complex loci in human and mouse genomes 9 . Predicted miRNA binding sites within pairing (sense-antisense overlapping) regions and non-pairing (non-overlapping) regions were counted and are listed in Additional file 1. In summary, among 894 sense-antisense pairs included in this study, 391 (43.7%) contain a sense-antisense overlapping region equal to or more than 25 nucleotides, which were selected for further miRNA binding-site scanning. A total of 18,704 predicted miRNA binding sites were identified in the sense-antisense pairing regions, spanning 358,663 nucleotides. In non-pairing regions, 111,192 miRNA binding sites were predicted across 1,570,606 nucleotides. After normalization of the predicted miRNA targets over sequence lengths, the miRNA binding sites within sense-antisense pairing regions ranged from 0 to 0.18790 miRNAs per nucleotide, with an interquartile range of 0.01398 to 0.08620, and a median value of 0.04780 miRNAs per nucleotide. The miRNA binding sites within non-pairing regions range from 0 to 0.2173, with an interquartile range of 0.03350 to 0.10020, and median value of 0.06490 miRNAs per nucleotide. There are more predicted miRNA binding sites in non-overlapping regions of the sense-antisense pairs compared to overlapping regions (P-value < 0.0001, Wilcoxon test). On average (mean), each 100 nucleotides of overlapping region have 5.21 miRNA binding sites, while each 100 nucleotides of non-overlapping region have 7.07 miRNA binding sites (Figure 6). This result was corroborated by our randomization test (P-value < 0.0001), in which 1,000 Monte Carlo randomizations with shuffled sequences were carried out. Our result indicates an evolutionary selection against miRNA binding to the pairing region. These findings suggest that overlapping regions between sense and antisense RNA transcripts are functional regulatory elements per se, and that sense-antisense RNA duplex formation may prevent miRNA binding. Therefore, there might be a selection to avoid a clash of two regulatory elements in one particular region.

RT-PCR was carried out with the GeneAmp 7900 machine (Applied Biosystems, Foster City, CA, USA). The primers and probe for miR-485-5p were bought from Applied Biosystems. The primers and probe for BACE1 and BACE1-AS were previously reported 10 . The PCR conditions were as follows: 50°C for 2 minutes then 95°C for 10 minutes then 40 cycles of 95°C for 15 s and 60°C for 1 minute. The results are based on cycle threshold (Ct) values. Differences between the Ct values for experimental and reference genes (Human beta-actin or U6 small RNA) were calculated as ΔΔCt.

Sequencing was carried out on RNA from two individuals for eight tissues, pancreas, lung, heart, skeletal muscle, brain, liver, spleen and kidney (kidney had only one sample). Sequencing was carried out using the Illumina genome analyzer. Small RNA libraries were prepared and 36 cycle sequencing carried out according to the manufacturer's instructions. Briefly, total RNA was fractionated and the 18-35 nucleotide fraction isolated. RNA adapters were ligated to the 3' and 5' ends of the samples and used for cDNA synthesis. Libraries were sequenced on the genome analyzer (Illumina, San Diego, CA, USA) and the sequences analyzed using miRanalyzer 52 . The number of unique reads for miR-485-5p were counted for each tissue and normalized to the total number of reads. The short read sequence data were submitted to the Sequence Read Archive at the National Center for Biotechnology Information; the submission number is SRA012516.1 and is freely available.

HEK293T cells were cultured in DMEM plus 10% fetal bovine serum and transfected with BACE1-AS, miR-485-5p, other miRNAs or empty vectors.

All experiments were performed with at least three to six biological and technical repeats. Beta-galactosidase and luciferase studies were performed in 96-well plates with at least 24 replicates for each treatment. The data are presented in graphs as a comparison with controls, after post hoc test of treatment factor using main effect in two-way analysis of variance (ANOVA). Alternatively, an unpaired t-test with Welch correction was used to calculate significance of the differences. The significance of each treatment was calculated as a P-value and is reported in the legend of each figure; P < 0.05 was considered significant.

ECA is a technology developed by DiscoveRx (Fremont, CA, USA) that allows the measurement of changes in protein levels. The cDNA of BACE1 was cloned into pCMV-ProLabel vector upstream of the ProLabel and transfected into HEK293T cells to produce a fusion protein (BACE1 and the enzyme donor fragment) expressed in a stable cell line we call C3. When the two fragments (enzyme donor and enzyme acceptor) of the β-galactosidase combine in solution, the enzyme becomes active and hydrolyzes a substrate that produces a chemiluminescent signal. The strength of this signal is proportional to the protein being produced (in this case BACE1). The stable cell line C3 over-expressing BACE1 was transfected with BACE1-AS over-expression vector, miRNA over-expression vector or control vectors and protein expression measured 48 hours later with the DiscoveRx method; data are plotted as a percentile of control vector.

The first set of human brain samples was prepared at the USC Alzheimer's Disease Research Center, which obtained informed consent from all subjects; the USC Institutional Review Board then approved the use of the human tissues. RNA was extracted from parietal lobes and cerebellum of postmortem brains from five subjects with Alzheimer's disease and five matched controls. The average age of subjects with Alzheimer's disease was 85 years (range 75 to 92 years) and 91.8 years (90 to 95 years) for controls. The postmortem interval ranged from 3.75 to 10.1 h with a mean of 5.87 h. We treated RNA samples with DNase and purified them with RNeasy mini columns (QIAGEN Valencia, CA, USA). We prepared cDNA from 400 ng of RNA samples and used RT-PCR for relative quantification of different transcripts.

The second set of human brain samples and the tissues used for deep sequencing were prepared from rapid autopsy brain tissue and were collected by J Rogers and T Beach (Sun Health Research Institute); all enrolled subjects or legal representatives had signed a Sun Health Research Institutional Review Board-approved informed consent form allowing both clinical assessments during life and several options for brain and bodily organ donation after death. These cases included 35 autopsy confirmed cases of Alzheimer's disease with an average age of 81.8 years (range 64 to 92 years) and 35 controls with an average age of 72.3 years (range 53 to 91 years). The postmortem interval ranged from 1.25 to 5 h with a mean of 2.5 h. The average duration of disease in the subjects with Alzheimer's disease was 9.2 years. Total RNA was isolated via CsCl purification from tissue dissected from specific regions of brain. Although not all regions were available from all cases, we examined a total of 120 RNA samples from superior frontal gyrus, entorhinal cortex, hippocampus and cerebellum for BACE1, BACE1-AS and miR-485-5p expression by RT-PCR.

A panel of total RNA from human tissues was bought from Applied Biosystems and used for expression profiling of BACE1, BACE1-AS and miR-485-5p.

Published human miRNA sequences were retrieved from the miRBase database 53 . A list of 993 evolutionarily conserved natural antisense transcripts was retrieved from the previously published FANTOM-3 dataset 9 . Out of this list, 99 pairs were deleted from any analyses described in this study, as we could not verify that they indeed qualify as sense-antisense pairs. miRNAs were screened in the sense-antisense RNA overlapping and non-overlapping regions, respectively. The BLAST program 54 was used to obtain the overlapping and non-overlapping regions of each sense-antisense pair. Then, the miRanda algorithm 55 was used for miRNA binding-site predictions. The numbers of predicted miRNAs within sense-antisense pairing regions and non-pairing regions were normalized to sequence length. Wilcoxon two-sided test was computed in R to compare the difference in predicted miRNA binding sites within sense-antisense pairing and non-pairing regions. In the randomization test, all sequences derived from the pairing and non-pairing regions were shuffled and sequences were randomly selected to create two data sets to represent the simulated pairing and non-pairing region data, respectively. The numbers of sequences included in the simulated data sets were equal to those in the real data sets. Wilcoxon two-sided test was used to compare the predicted miRNAs between the simulated pairing and non-paring data sets. The randomization procedure was repeated 1,000 times. We calculated the number of times (x) that the Wilcoxon P-value/W values in the randomizations are smaller than those from the real data sets. The P-value of the randomization is x/1,000. The statistically significant difference cut-off was 0.05.