Increased processing of SINE B2 non coding RNAs unveils a novel type of transcriptome de-regulation underlying amyloid beta neuro-pathology

More than 97% of the mammalian genome is non-protein coding, and repetitive elements account for more than 50% of noncoding space. However, the functional importance of many non-coding RNAs generated by these elements and their connection with pathologic processes remains elusive. We have previously shown that B2 RNAs, a class of non-coding RNAs that belong to the B2 family of SINE repeats, mediate the transcriptional activation of stress response genes (SRGs) upon application of a stimulus. Notably, B2 RNAs bind RNA Polymerase II (RNA Pol II) and suppress SRG transcription during pro-stimulation state. Upon application of a stimulus, B2 RNAs are processed into fragments and degraded, which in turn releases RNA Pol II from suppression and upregulates SRGs. Here, we demonstrate a novel role for B2 RNAs in transcriptome response to amyloid beta toxicity and pathology in the mouse hippocampus. In healthy hippocampi, activation of SRGs is followed by a transient upregulation of pro-apoptotic factors, such as p53 and miRNA-34c, which target SRGs creating a negative feedback loop that facilitates transition to the pro-stimulation state. Using an integrative RNA genomics approach, we show that in mouse hippocampi of an amyloid precursor protein knock-in mouse model and in an in vitro cell culture model of amyloid beta toxicity, this regulatory loop is dysfunctional due to increased levels of B2 RNA processing, constitutively elevated SRG expression and high p53 levels. Evidence indicates that Hsf1, a master regulator of stress response, mediates B2 RNA processing in cells, and is upregulated during amyloid toxicity accelerating the processing of SINE RNAs and SRG hyper-activation. Our study reveals that in mouse, SINE RNAs constitute a novel pathway deregulated in amyloid beta pathology, with potential implications for similar cases in the human brain, such as Alzheimer’s disease (AD). This data attributes a role to SINE RNA processing in a pathological process as well as a new function to Hsf1 that is independent of its transcription factor activity.


INTRODUCTION
The number of patients with Alzheimer's disease (AD) is expected to skyrocket in upcoming years (Cornutiu, 2015). The exact molecular mechanisms underlying AD are not fully understood, a fact that is underlined by the inauspicious results of recent clinical trials for potential therapeutic agents (Cummings, Lee, Ritter, Sabbagh, & Zhong, 2019;Hane et al., 2017). Amyloid pathology, and particularly, amyloid beta peptides and their aggregated forms have been connected with AD pathogenesis (Bloom, 2014) as well as with neurotoxicity in mouse models of amyloid pathology (Ittner et al., 2010). Nevertheless, the transcriptome changes involved in cell stress response to amyloid toxicity in brains with extensive amyloid beta pathology are still not entirely clear. The hippocampus is a primary target of amyloid pathology in humans. In healthy hippocampi, among the genes that have been implicated in transcriptome-environment interactions are stress response genes (SRGs) (Gallo, Katche, Morici, Medina, & Weisstaub, 2018). These genes have been initially described in other biological contexts, such as thermal and oxidative stress, as prosurvival genes activated early after the application of a stress stimulus, such as heat shock, that help the cell overcome the stress condition (Mahat, Salamanca, Duarte, Danko, & Lis, 2016).
However, in addition to the cellular response to stress, many SRGs were shown to have a central role in the function of the mouse hippocampus, by mediating cell signalling and genomeenvironment interactions. In particular, we and others have shown that in the healthy hippocampus during neural response to environmental stimuli, SRGs, such as those of the MAPK pathway, are transiently activated (Peleg et al., 2010). Activation of SRGs is followed by a transient upregulation of the pro-apoptotic factor p53 and, subsequently, of a pro-apoptotic miRNA, miR-34c, which are transiently induced by and as a response to the activation of prosurvival SRGs (Zovoilis et al., 2011). p53 activates the expression of genes engaged in promoting cell death in response to multiple forms of cellular stress including miRNA-34c (Yamakuchi & Lowenstein, 2009). This miRNA acts transiently as a guard and fine tuner of the expression of many SRGs by targeting them, thus creating a negative feedback regulatory loop that keeps SRG expression in healthy cells under strict control. This facilitates the return to the pro-stimulation state and normalization of p53 levels in approximately 3 hours after the application of the stimulus (Zovoilis et al., 2011). In contrast, hippocampi of mouse models of amyloid pathology and postmortem brains of human patients of AD are characterized by abnormally high miR-34c levels that subsequently can lead to prolonged high p53 levels and neural death (Zovoilis et al., 2011). Given that many SRGs are upstream regulators of p53-miR-34c activation (Gao et al., 2010), high p53 and miR-34c levels in amyloid pathology implied a possible transcriptome deregulation of the pathways that involve SRGs but whether such a deregulation exists, and which is the mechanism underlying this, it remained unknown. Interestingly, in a recent publication, we showed that expression of a number of SRGs is regulated by a class of non-protein coding (non-coding) RNAs called B2 SINE RNAs (Zovoilis, Cifuentes-Rojas, Chu, Hernandez, & Lee, 2016), raising the possibility that these non-coding RNAs may be a missing link in the pathways connecting amyloid beta toxicity with transcriptome changes in mouse hippocampus during amyloid pathology. SINE non-coding RNAs are transcribed by repetitive Small Interspersed Nuclear Elements (SINEs), with the subclass B2 repetitive elements being one of the most frequent in mouse (Kramerov & Vassetzky, 2011). SINE B2 elements, which are retrotransposons present in millions of copies, are part of the non-protein coding genome, and over the long haul they have been regarded as genomic parasites and "junk DNA" with no function (Karijolich, Zhao, Alla, & Glaunsinger, 2017). However, SINE B2 elements can be transcribed by RNA Polymerase III into SINE B2 RNAs (Kramerov & Vassetzky, 2011) and a number of recent studies have revealed a key role for SINE B2 RNAs in cellular response to stress. In particular, studies from the J.Kugel and J. Goodrich labs have shown that during response to cellular stress, levels of SINE B2 RNAs increase and suppress the transcription of a number of housekeeping genes through binding of RNA Polymerase II (RNA Pol II), potentially facilitating the redirection of cell resources to prosurvival pathways (Yakovchuk, Goodrich, & Kugel, 2009) (Yakovchuk et al., 2009). In addition, we have recently shown that SINE B2 RNAs mediate cellular response to stress through the regulation of pro-survival stress response genes by acting as transcriptional switches. In particular, in the pro-cellular stress state SINE B2 RNAs bind RNA Pol II at several SRGs and suppress their transcription. In this way, stalled or delayed RNA Pol II remains poised for a fast activation and ramp up of transcription when needed. Upon application of a stress stimulus, SINE B2 RNAs, which have a self-cleavage activity that is accelerated by their interaction with some proteins (Hernandez et al., 2020), are processed into unstable fragments that lack the ability to bind and suppress RNA Pol II. This event releases the delayed or stalled RNA Pol II and enables fast transcriptional activation of stress response genes (Zovoilis et al., 2016).
The above findings have revealed a novel role in cellular function for processing of SINE B2 RNAs (hereafter referred simply as B2 RNAs) through the regulation of SRGs. However, an association of the processing and destabilization of B2 RNAs with any pathological cellular process remains unknown. Here, given the importance of SRGs in hippocampal neuronal function, we examine whether this newly described B2 RNA-SRG regulatory mechanism is linked with pathological processes, focusing on transcriptome response to amyloid beta toxicity and pathology. To this end, we investigate whether B2 RNA regulated SRGs are indeed deregulated during amyloid pathology, which would imply a role for B2 RNAs in this condition and we examine whether amyloid toxicity is connected with changes in B2 RNA processing. Subsequently, we investigate the further upstream molecular mechanisms underlying any potential deregulation of B2 RNA processing in response to amyloid toxicity in hippocampal neural cells.

B2 RNA regulated SRGs are enriched in neural functions
In a previous study, we have identified genomic locations that are subject to regulation by SINE B2 non-coding RNAs through binding and suppression of transcription by the RNA Pol II at the pre-stimulus state. Upon induction of cellular stress through the application of a stimulus, SRGs in these locations become activated through B2 processing and release of RNA Pol II suppression ( Fig.1A) (Zovoilis et al., 2016). A list of B2 RNA regulated SRGs in these locations is available in Suppl. Table 1.
We questioned whether there are known cellular functions connected with B2 RNA regulated SRGs. As shown in Fig. 1B, after performing a tissue enrichment analysis to identify tissue terms that are over-represented in the list of our SRGs, we found a significant enrichment of neural tissue terms compared to other tissues in our list. Similarly, during Gene Ontology term enrichment analysis, cellular compartments closely related to neural functions top the list of enriched terms in these genes, including terms such as synapse, postsynaptic density and membrane, dendrites, neural projections and pre-synaptic membrane (Fig.1C). Most importantly, after performing the same analysis for Biological Processes GO terms, B2 RNA regulated SRGs were found to be enriched considerably in neural function related terms. GO terms, such as learning, nervous system development, synaptic transmission, synapse receptor localization and neurotransmitter transport were among the first 10 entries with the highest adjusted P value scores enriched in B2 RNA regulated SRGs (Fig.1D).
Among the biological processes potentially affected by B2 RNAs that were identified above was learning. We and others have already shown that learning processes in the mouse hippocampus are connected with the transient activation of a number of learning associated genes and pathways, including many known SRGs, (Peleg et al., 2010). For this reason, we examined whether learning associated genes are among the binding targets of B2 RNAs. To this end, we compared the distribution of B2 RNA binding sites in the genome between learning associated genes and all genes. Indeed, as shown in Fig.1E, B2 RNA binding sites were found to be enriched in learning associated genes compared to other genes. This data suggests that B2 RNA regulated SRGs could affect a wide spectrum of neural functions including learning. Learning is among the biological processes heavily impaired in AD.
Given the role of hippocampus as a primary target of amyloid pathology in AD, we decided to focus further on the potential role of B2 RNA-SRGs regulation in this pathological process.

B2 RNA regulated SRGs get hyper-activated during progression of amyloid pathology
We have previously shown that in hippocampus the pro-apoptotic miRNA miR-34c, which is induced among others by p53, targets in a negative feedback loop manner many SRGs and helps to restore SRG and p53 expression levels after application of a stimulus ( Fig.1F) (Zovoilis et al., 2011). In both mouse models of amyloid beta pathology and AD patient brains, we found persistently high levels of miR-34c. High miR-34c levels in hippocampi with amyloid pathology are indicative of a transcriptome-wide deregulation of p53 and SRG levels, as many of these genes are either direct or indirect upstream regulators of miR-34c ( Fig.1F) (Rokavec, Li, Jiang, & Hermeking, 2014). Thus, we questioned whether transcriptome changes in response to amyloid pathology involve SRGs regulated by B2 RNAs.
To test whether B2 RNA regulated SRGs are indeed de-regulated in amyloid pathology, we employed a transgenic mouse model of amyloid pathology, APP NL-G-F , (Saito et al., 2014) and the respective wild type control (C57BL/6J). The same animal cohorts that were previously characterized through a battery of immunohistochemistry (IHC) and behavioral tests (Mehla et al., 2019) were used to isolate whole hippocampi for the transcriptome analysis conducted in the current study. Fig We have focused on three different mouse ages that correspond to different phases of amyloid beta pathology and represent the: i) pre-symptomatic stage with undetectable (very low) amyloid plaque load, (3 months -3m, Fig.2B left panels), ii) stage of symptom manifestation (6 months -6m) (Suppl. Fig.1) that coincides with the active neurodegeneration phase and appearance of amyloid plaques (Fig.2B, middle-panels), and iii) terminal stage of the pathology (12 months -12m, Fig. 2B right panels) when mice have already acquired the extensive brain atrophy due to neural cell death. Whole hippocampi from mice of these three groups were isolated and the extracted RNA was subjected to next-generation sequencing. We performed directional RNA sequencing for these samples and subsequently quantified gene expression levels.
In accordance with increased cell death during the active neurodegeneration phase at 6 months, levels of p53 were elevated in APP mice of this age compared to controls of the same age (t-test, p<0.05, n=3/group) (Fig.2C). Moreover, consistent with our hypothesis of a transcriptome-wide deregulation, levels of B2 RNA regulated SRGs were found to be strongly upregulated during the active phase of neurodegeneration at 6 months ( Fig.2D-F). In particular, in healthy hippocampi (WT control animals), levels of these genes are normally downregulated in 6m and 12m old mice compared to 3m old mice (p=0.02 and p=0.04, respectively) ( Fig.1A and   1F). This is in accordance with the higher level of neural synaptic activity and plasticity that younger mice have (Lilja et al., 2013). In contrast, levels of B2 RNA regulated SRGs in APP mice remain abnormally high in 6m old APP mice compared to both the 3m old APP mice and the 6m old control mice (p<0.05 in both cases) ( Fig.2D and 2F). This pattern in B2 RNA regulated SRG expression was not observed when a random set of genes was tested (Fig.2E).
These findings show a transcriptome wide hyper-activation of B2 RNA regulated SRGs during the active phase of amyloid pathology.

B2 RNA processing rate increases upon progression of in vivo amyloid pathology
Given the role of B2 RNAs in the regulation of SRGs, we hypothesized that the observed hyper-activation of B2 RNA regulated SRGs in the hippocampus of 6m old APP mice may reflect a similar upstream deregulation at the level of the B2 RNAs.
To test this, we employed a customized version of RNA sequencing and analysis used in our previous study (Zovoilis et al., 2016) that allows for enrichment and sequencing of the short SINE RNA fragments (<100nt) produced by B2 RNA processing (short-RNA-seq). In contrast to standard long-RNA-seq protocols that include RNA fragmentation and, thus, may introduce bias, this approach circumvents this problem. Moreover, the long-RNA-seq protocols exclude short RNA fragments of <100nt making the identification of B2 short fragments challenging. After short-RNA-seq (Fig.3A), mapping of the 5'ends of the sequenced fragments across the B2 loci enables the determination of processing points at B2 RNA (depicted as "X" in Fig.3A), including those at the critical RNA Pol II binding region (depicted as a rectangle in Fig.3A). Fig.3B depicts an example of these fragments from one of the samples. As shown in Fig.3A, B2 RNA is extensively processed in hippocampi with amyloid pathology. In particular, this data revealed an increased number of B2 RNA fragments in 6m old mice compared to controls (Fig.3A), a difference not observed when comparing full length B2 RNA levels between these groups of mice using long-RNA-seq (Fig.3C).
This data suggests that the processing rate of B2 RNAs may be higher in APP mice of this age. To estimate this rate, we used the short RNA-seq data in combination with standard long-RNA-seq (Fig.3D). As in our previous study, for normalization of short RNA 5'end values we used a class of short RNAs that is not affected by B2 RNAs, the RNA Pol III-transcribed tRNAs.
Moreover, the absolute numbers of fragmented B2 RNA may vary according to the underlying expression of full length B2 RNA transcripts. Therefore, in order to factor in any differences in the amount of fragments due to basal expression levels of the full length B2 RNA, B2 RNA fragment values from short RNA-seq were normalized to the levels of the full length B2 RNAs calculated by the directional long-RNA-seq. The long-RNA-seq approach excludes short fragments.
Consistent with our hypothesis and Fig.3A findings, 6m old APP mice were found to have substantially increased rate of B2 RNA processing compared to control mice of the same age ( Fig.3E) (p<0.05, n=3/group).
Thus, APP mice at the active neuro-degeneration phase are characterized by higher destabilization and processing rates of B2 RNAs, consistent with the observed increase in B2 RNA regulated SRG levels in these same animals.

Hsf1 accelerates B2 RNA processing
We then focused on the molecular mechanism underlying the increased B2 RNA processing during response to amyloid toxicity. When we had previously examined the mechanism of B2 RNA processing in non-neural cells, a member of the PRC2 protein complex, Ezh2, was reported as being responsible for the B2 RNA accelerated destabilization and processing during response to stress. However, as shown in Fig.4A-B, scant expression levels of Ezh2 levels in neural cells indicate that Ezh2 is not a key factor in B2 RNA processing in brain.
Thus, the factors that may mediate destabilization for B2 RNAs during stress in neural cells remain elusive.
In our earlier study that described the induction of B2 RNA processing by Ezh2, it remained unclear how Ezh2 exerted its impact on B2 RNAs, since Ezh2 lacked any known RNAse activity (Zovoilis et al., 2016). However, in subsequent experiments (Hernandez et al., 2020) we showed that instability is in fact inherent to the B2 RNA molecule and interaction with Ezh2 only accelerates this destabilization. This finding suggests that other proteins may have a similar effect on B2 RNA stability. Therefore, we started searching for stress related candidate proteins that could affect the B2 RNA processing in hippocampus.
We showed before that B2 RNA binding during response to stress is enriched near stalled RNA polymerase genomic sites. These areas are known to be highly enriched in binding sites of various stress related proteins, among which it is a stress related protein called Hsf1, which is a master regulator of stress response for various types of cellular stress (Pandey, Mandal, Jha, & Mukerji, 2011). Hsf1 has been previously connected with activation of SRGs through both transcriptional factor (TF) activities as well as other yet unknown TF-independent processes (Inouye et al., 2007). As shown in Fig.4A and 4B, in contrast to Ezh2, Hsf1 is expressed in neural tissues and especially in hippocampus. We then examined the proximity of Hsf1 binding sites, identified by the Lis lab (Mahat et al., 2016), to genes with B2 RNA binding sites (Suppl. Table 1) (Zovoilis et al., 2016). Increased Hsf1 binding was found near B2 RNA regulated SRGs (Fig.4C) and was further enriched after application of a heat-shock stimulus (KS-test < 0.05). Moreover, Hsf1 levels were found to be upregulated in APP 6m old mice compared to control group of the same age (p<0.05, n=3/group) (Fig.4D). As mentioned above, at the same time, 6m old APP mice have increased B2 RNA processing rates. Thus, this data suggests that Hsf1 may be a good candidate for accelerating B2 RNA processing in the context of amyloid pathology.
To check this, we investigated whether the interaction between B2 RNA and Hsf1 can accelerate B2 RNA destabilization. In particular, we incubated full length B2 RNA with the Hsf1 protein in vitro. As shown in Fig.5A, in the presence of Hsf1, the destabilization of full length B2 RNA was accelerated in contrast to the control protein (PNK) or no protein. No processing was observed in an RNA fragment co-incubated with B2 RNA and the protein (marked with an asterisk across Fig.5). B2 RNA destabilization in the presence of Hsf1 was accelerated compared to no protein ( Fig.5B and 5C). Consistent with the above finding, the rate of processing of B2 RNA was dependent on Hsf1 concentration in the reaction ( Fig.5D and 5E) and increased upon increase of Hsf1 concentration. This data shows that Hsf1 has the potential in vitro to accelerate B2 RNA processing.

Hsf1 mediates increased B2 RNA processing in response to amyloid beta toxicity
The increased levels of Hsf1/B2 RNA processing in APP 6m old mice raised the question whether response to amyloid toxicity in hippocampal cells is connected with an Hsf1-mediated increase in B2 RNA processing. In order to test this, we employed the a hippocampal cell culture model using the a HT-22 cell line, which has been used extensively in the past as hippocampal cell stress model (Davis & Maher, 1994;Liu, Li, & Suo, 2009). We incubated these cells with amyloid beta peptides (1-42 aa) and compared their transcriptome to the one of cells incubated with an inverted sequence control peptide (R, reverse 42-1) (Fig.5F). Incubation of these cells with 1-42 amyloid beta peptides results in upregulation of a number of genes ( Fig. 5G), that are also found upregulated in the 6m old APP mice (Suppl. Table 2). The increase in expression levels of genes associated with amyloid pathology suggests that this cellular model simulates to a certain extend the amyloid toxicity effect on the transcriptome of hippocampal cells.
In order to test the impact of amyloid toxicity to Hsf1-mediated B2 RNA processing we treated HT22 cells with either an LNA against Hsf1 or a scramble LNA (control) followed by incubation with the 1-42 peptides, that subject the cells to amyloid toxicity stress, or the respective control peptide (Fig. 6A). As shown in Fig.6B, treatment with the anti-Hsf1 LNA has suppressed the increase of Hsf1 levels observed in the control cells upon application of the amyloid beta peptides. As observed in the APP mice, amyloid beta peptides resulted in increased B2 RNA processing in control cells (with no Hsf1 knock down) suggesting that it is the toxicity of these peptides that induces the SINE B2 RNA transcriptome changes (Fig.6C). In contrast, under anti-Hsf1 LNA, application of the 1-42 peptides was unable to increase B2 RNA processing in the cells compared to the control cells (Fig. 6C). A similar effect was observed in B2 RNA regulated SRGs compared to other random genes ( Fig.6D-6F).
These results suggest that amyloid beta toxicity induces B2 RNA processing also in vitro and Hsf1 comprises a necessary component in the upstream activation pathway of B2 RNA processing and, thus, of the genes regulated by B2 RNA.

DISCUSSION
Fewer than five ribozymes have been identified in mammals, including our previous work on RNAs made from two retrotransposon families, murine SINE B2 RNAs and human SINE ALU RNAs, which are self-cleaving RNAs (Hernandez et al., 2020). We previously showed that cleavage in SINE B2 RNAs controls response to cellular stress through activation of stress response genes. However, no connection between this novel molecular mechanism and pathologic processes was until now known. Moreover, since B2 RNAs are intrinsically reactive, and contact with Ezh2 only accelerates cleavage, it remained plausible that other stress-related proteins may also have a similar effect on accelerating B2 RNA processing, which would link this ribozyme-like property to stress response through pathways other than Ezh2. This is especially relevant in mouse tissues, such as the brain, where Ezh2 expression is limited, an expression pattern observed also in human (Suppl. Fig. 2).
Here, we unveil increased processing of SINE B2 RNAs as a novel type of transcriptome de-regulation underlying amyloid beta neuro-pathology. Our data provides a new link in the murine hippocampal pathways connecting amyloid beta toxicity with transcriptome changes in SRGs through processing of B2 RNAs. In particular, the B2 RNA processing rate increases upon progression of amyloid pathology both in mouse hippocampus and a hippocampal cell culture model, and B2 RNA regulated SRGs become hyper-activated. Consistent with the spatial proximity between B2 RNA regulated SRGs and Hsf1 binding sites, Hsf1 proved to be key for mediating B2 RNA processing in response to amyloid toxicity. Our work assigns to Hsf1 a new function that is independent of its long-established transcription factor function and includes the interaction with and processing of SINE B2 RNAs. The high levels of Hsf1 trigger a downstream cascade of events which are orchestrated into a cell-wide, SRG-mediated response to stress conditions. This axis is mediated by the ability of B2 RNA to get processed and, thus, act as a molecular switch. While healthy cells and animals are able to restore the expression levels of Hsf1, SRGs and their regulating B2 RNAs upon removal of the stressgenerating stimulus, the amyloid beta load in our biological models acts as a continuous stimulus that causes the Hsf1 -B2 RNA -SRG axis to "lock" into an activated mode.
Upregulation of SRGs results in increased p53 levels that induce neuronal cell death (Fig.7).
In this study, we employed a mouse model of amyloid pathology in order to test the impact of increased amyloid beta load on B2 RNA processing in vivo. This mouse model has the NL-G-F mutations in the amyloid precursor protein knocked-in to a C57BL genomic background, with each mutation contributing to an increased severity and speed manifestation of the disease . The combined effect of the APPNL−G−F mutations results in mice that experience rapid onset of AD-like symptoms at approximately 6 months (Mehla et al., 2019). This is exactly the time point that we observed the massive hyper-activation of B2 RNA SRGs, p53 and B2 RNA processing, suggesting that these changes constitute a molecular signature for the active neurodegenerative phase months, before the end state of 12 months that the mice develop terminal AD-like pathology and symptoms. Interestingly, these molecular changes were not observed in younger mice suggesting the existence of a yet unknown protective mechanism against increased neuronal activity of SRGs in 3m old APP mice. This may be attributed to the increased neuronal plasticity observed in younger brains (Lilja et al., 2013), which suggest that mechanisms in the younger brain may exist for counter-acting this excessive activity. In contrast, during aging, SRG activity in WT mice ramps down, which is not the case in APP mice, explaining at the molecular level the increased p53 levels and subsequent activation of cell death (Fig.1F).
Our study leaves a number of open questions. Particularly, our study raises the question whether a similar mode of regulation of SRGs by SINE RNAs may exist also in human and which SINE RNAs could play such a role. Similarly to B2 RNA, such SINE RNAs would be able to bind and inhibit RNA Pol II, and would be subject to a similar RNA processing mechanism enabling the release of RNA Pol II. A number of studies have described that in human, repetitive SINE RNAs of the Alu class are also upregulated during cellular stress and can bind RNA Pol II inhibiting the transcription of target genes (Yakovchuk et al., 2009). Alu RNAs are widely regarded as the equivalent in human of B2 RNA. Most importantly, as we showed before, human Alu RNAs, alike B2 RNAs, are self-cleaving RNAs and can become destabilized in vitro (Hernandez et al., 2020). It remains unknown whether SINE RNAs and Hsf1 play a similar role in amyloid pathology in the case of humans and whether we can extrapolate the generated conclusions in murine models to deduce that SINE RNAs are key components of the pathophysiological mechanisms underlying debilitating diseases such as AD. One major limitation compared to human pathophysiology is that the phenotype of amyloid pathology is not observed in mice even during aging. Nonetheless, a stress-central role of Alu RNAs, the human counterpart of B2 RNAs is plausible and, thus, future studies need to elucidate whether Alu RNA processing is also hyperactivated in the brain of patients with amyloid pathology in the context of AD.
Since B2 RNA regulated SRGs are highly interconnected and interweaved into various pathways, the impact of SRG hyperactivation by B2 RNA processing may extend to pathways that lie downstream of SRGs and affect various gene programs without binding directly B2 RNA.
However, here it should be noted that high levels of Hsf1 are certainly expected to affect transcription also through Hsf1's conventional transcriptional factor function while there are still many SRGs that are not directly regulated by B2 RNA. Thus, B2 RNA processing described here does not constitute the only one but just one of the parameters in the equation of SRG regulation.
It remains unclear which is the interplay between Hsf1 traditional transcription factor activities and its ability to affect B2 RNA processing. Moreover, given how easily B2 RNA is processed in the presence of certain proteins, Hsf1 may be only one of the factors accelerating B2 processing in mouse hippocampus as we are just beginning to understand the implications of this form of SINE RNA regulation in cells. A broader role of SINE RNA processing in brain physiology and pathophysiology constitutes, thus, a significant possibility that could further revise our understanding of these RNAs as something more than just transcriptional noise and "junk DNA" products.

Amyloid beta peptide preparations
The amyloid beta 1-42 peptides and the respective control peptides (having the reverse aa sequence compared to 1-42 peptides) were synthesized by Sigma Aldrich's custom synthesis service using the following sequences:

Animals and behavioural measurement
Immunohistochemistry and behavioral studies of mice used are described in our previous study . In brief, mouse pairs of APP-KI mice carrying Arctic, Swedish, and Beyreuther/Iberian mutations (APPNL-G-F/NL-G-F) were gifted by RIKEN Center for Brain Science, Japan, and the colony of these mice was maintained at Canadian Center for Behavioral Neuroscience vivarium. C57BL/6J mice were used as a WT control and all animals were housed in groups of 4 mice in each cage in a controlled environment (22 °C-25 °C, 50% humidity and a 12-hour light:dark cycle). All experimental procedures were approved by the institutional animal care committee and performed in accordance with the standards set out by the Canadian Council for Animal Care. We used 10 mice for behavioral tests and 2-4 mice for immunohistochemistry at each of the testing age points. Mice were individually handled each day for 1 week before starting the acquisition training and were given 8 days of acquisition trials. On each acquisition day, mice received 4 training trials from each quadrant randomly distributed each day. The latency to reach the platform was analyzed to confirm the learning behavior of the mice. A single probe trial was conducted on day 9 to assess the integrity and strength of spatial memory 24 hours after the completion of the last trial of the acquisition phase. We analyzed the data in the probe trial by measuring the time spent by mice in the target quadrant and average proximity to the escape annulus.

Cell culture and transfections
HT22 cells, from an immortalized mouse hippocampal cell line (Davis & Maher, 1994) (Sigma Aldrich, SCC129), were cultured in DMEM (Sigma) and 1% Penicillin/Streptomycin (Gibco), Cells were thawed and passaged at least three days before the transfection date in order to allow sufficient time for cells to recover from the stress of cryopreservation and not interfere with the assessment of cellular response to stress in subsequent experiments. For knocking down of Hsf1 mRNA levels, we used an LNA long RNA GapmeR against Hsf1 (Exiqon/Qiagen) with the following sequence: 5'-CGAAGGATGGAGTCAA-3' and an LNA long RNA GapmeR with the following scramble sequence: 5'-CCTCAATTTTATCAC-3'. The day of LNA transfections, following 5min incubation with TrypLE TM Express Enzyme (Gibco) (1x), cells were passaged and transferred to a 6-well plate at a 100,000 cells/well density and LNA transfections were performed simultaneously, using the HiPerfect reagent (Qiagen). Transfection was performed as follows: Firstly, LNAs were reconstituted in nuclease free water at a 50μM. Subsequently, 3μL 50uM LNA were mixed and incubated with 4uL Hiperfect reagent and 30ul nuclease free water, at room temperature for 20 min, and then added drop-wise to cells that had just been plated in 1ml of medium/well (still not attached) to a final LNA concentration of 150nM. Subsequently, incubation with amyloid beta (1-42) and control peptides (Reverse) was performed 24h after transfection with LNAs. Peptides were initially dissolved in DMSO and then added to cells to a final concentration of 30uM for incubation at 37C for 1 hour before treating with 0.5mL TrypLE TM (Gibco), pelleting cells at 1000rpm for 5min and resuspending the pellet in 1mL Trizol reagent (Thermofisher) for RNA extraction based on Manufacturer's instructions.

RNA in vitro transcription and RNA-protein incubations
B2 template for in vitro RNA transcription was ordered as IDT g-block TM : 5′-GGGGCTGGTGAGATGGCTCAGTGGGTAAGAGCACCCGACTGCTCTTCCGAAGGTCCGGA GTTCAAATCCCAGCAACCACATGGTGGCTCACAACCATCCGTAACGAGATCTGACTCCCTC TTCTGGAGTGTCTGAAGACAGCTACAGTGTACTTACATATAATAAATAAATAAATCTTTAAAA AAAAA -3′. The template was amplified by PCR using a T7 promoter sequence as the forward primer: 5′-TAATACGACTCACTATAG and the following sequence as reverse primer: 5′-TTTTTTTTTAAAGATTTATTTATTTATTATATGTAAGTACA. Primers were diluted to 20mM and PCR was performed using the NEB Q5 polymerase, Q5 reaction buffer (10x), Q5 high GC enhancer (10x). The reaction proceeded at hot start 98C -30s, (98C -5s, 58C -10s, 72C -10s) X35 cycles, 72C -10min. The samples were then analyzed by agarose gel electrophoresis (Bio-Rad, 1613100EDU) and the bands were gel extracted at the desired size and purified using the BioBasic EZ-10 gel extraction kit (BS353). A subsequent PCR was then repeated and 1ug of the B2 RNA re-folding and B2 RNA -Hsf1 incubations were performed as described previously (Zovoilis et al., 2016). In brief, in-vitro-transcribed B2 RNA was folded with 300mM NaCl through incubation for 1 min at 50°C and cooling at a rate of 1°C/10sec until 4°C. Fragmentation of B2 RNA was analyzed on 8M urea 10% PAGE gels stained by SYBR II (Invitrogen, S7564). Gel analysis occurred on Amersham Typhoon instruments. Band absorbance was analyzed using ImageJ area under the curve software and normalized by the ratio of experimental over initial as described previously (Zovoilis et al., 2016).

Short-RNA-seq and long-RNA-seq
1.5μg total RNA was size separated into short and long fractions using the miRvana TM miRNA size selection kit (Thermo Fisher) as described before (Zovoilis et al., 2016). In brief, following addition of the lysis/binding buffer and the homogenate additive solution to the RNA, 1/3 of the volume 100% EtOH was added and the mix was passed through the column for binding long RNAs. 100% EtOH at 2/3 of the flow through volume was subsequently added to the flow through and passed through a second column for binding short RNAs. Eluted RNAs were tested for size and quality using the Agilent Bioanalyzer RNA pico-kit. For long-RNA-seq, the long RNA fractions were cleaned and concentrated using the RNeasy Minelute kit (Qiagen) and ribodepleted using the rRNA depletion kit (NEB). The library was then prepared using the NEBNext Ultra II direction Library preparation kit (NEB, E7760), and sample cleanups were performed using the Omega NGS Total Pure Mag Beads (Omega, SKU: M1378-01) 0.5X and 1.2X before library amplification and 0.9X following amplification. 9 cycles were used during amplification. For short-RNA-seq, the short RNA fractions were subjected to 3'-phosphoryl removal for 1 hr at 37C, treated with the T4 PNK enzyme (NEB, M0201), using exclusively the 10X PNK buffer (NEB). The short fractions were cleaned and concentrated using the RNeasy Minelute kit and the library was prepared using the NEBNext small RNA library prep set (E7330) as described before (Zovoilis et al., 2016). Sample cleanups were performed using again the  (Li & Durbin, 2009). Long-RNA-seq reads for each sample were mapped to reference genome ensembl GRCm38 (November 2018) primary assembly using hisat2-2.1.0, in single end mode, with the following parameters: Report > alignments tailored for transcript assemblers including StringTie, searches > for at most 1 distinct, primary alignments for each read (Kim, Paggi, Park, Bennett, & Salzberg, 2019). SAM format files generated from mapping were converted to BAM format files using samtools-1.6 , and to files in BED format with bamToBed utility from BEDTools-2.26.0 (Quinlan & Hall, 2010 In long-RNA-seq, FPKM (Fragments Per Kilobase of transcript per Million) and TPM (Transcripts Per Million) for genes were generated using StringTie-1.3.4d (Pertea et al., 2015) with the following annotation: ensembl GRCm38 patch 94 gff3 file, and parameters limiting the processing of read alignments to only estimate and output the assembled transcripts matching the reference transcripts given in annotation and excluding non regular chromosomes. For data visualization, statistics and differential expression analysis we employed R (version 3.4.3) (https://www.R-project.org/) and the package DESeq2 (Love, Huber, & Anders, 2014). Differential expression analysis was implemented on gene count data to perform differential gene expression analysis for 6 m old mice between APP and wild type. Boxplots central line represents median and t-test was applied on the group numbers mentioned in the text.
For Hsf1 metagene analysis, we used peak.txt files of Hsf1 peaks for ChIP-seq from (Mahat et al., 2016). Peaks were analyzed with Seqmonk around Transcription start sites of genes (TSS) based on the Eponine annotation (Down & Hubbard, 2002), and filtered based on their overlap with B2 RNA regulated genes (Zovoilis et al., 2016). Also, B2 RNA binding (CHART-seq) peaks were analyzed with Seqmonk around TSS (Eponine), then filtered by overlapping with learning associated SRGs or all genes (Peleg et al., 2010). Relative density metagene plots of the distribution of the above peaks were generated using Seqmonk.

Data access
Short and long-RNA-seq raw data have been deposited to GEO with access number GSE149243.
Resource Allocation Grant to AZ. AZ is supported by the Canada Research Chairs Program and the Canada Foundation for Innovation and is a former EMBO and DFG long-term fellow. YC is supported by an Alberta Innovates (AITF) fellowship. We are grateful to Dr. Angeliki Pantazi for extensively reviewing, editing and commenting on the manuscript.

AUTHOR CONTRIBUTIONS
YC: Bioinformatics analysis, statistical analysis, establishment and testing of the analysis pipelines, data visualization and writing of the manuscript; BG: establishment and testing of the next-generation-sequencing pipelines, library construction; LS: testing of the next-generationsequencing and in vitro assay pipelines, library construction, in vitro assays, data visualization and writing of the manuscript; CI: establishment and testing of the in vitro assay pipelines, in vitro assays; JM: acquisitions of brain tissue, hippocampal extractions; MM: design, data interpretation, writing of the manuscript; AZ: conception and design, establishment and testing of data generation and analysis pipelines, data interpretation, data visualization and writing of the manuscript, overall supervision

CONFLICT OF INTEREST
The authors declare no conflict of interest.  (B) Immunohistochemistry for identifying amyloid-beta load in the brain sections of mice from the same cohort as in (A) (image credit: Mehla and colleagues, adapted from previously published figure ). Higher fluorescence intensity corresponds to higher amyloid load in APP mice from 6m old onwards.
(C) Expression levels of the pro-apoptotic gene p53 (official symbol TrP53) as defined by long-RNA-seq. p53 transcripts per million reads (TPM) from the APP mice compared to WT, grouped by age. Boxplot depicts distribution of expression levels of p53 gene among different age groups of mice (WT: wild type, APP: mice with amyloid pathology). Black line denotes median. p < 0.05 between 6m old APP and 6m old WT (depicted as asterisk, t-test, n=3/group).  Espinosa and colleagues (Espinoza, Goodrich, & Kugel, 2007). As in our previous study (Zovoilis et al., 2016), we depict the B2 RNA processing points based on short RNA-seq data and mapping of the 5'ends of B2 RNA fragments. X marks which cleavage sites of B2 RNA in the upper panel correspond to enriched processing points (the peaks of 5'end fragments distribution) at the lower panel. The green rectangle depicts the critical region that binds and suppresses RNA Pol II (Espinoza et al., 2007;Ponicsan et al., 2010;Ponicsan, Kugel, & Goodrich, 2015;Yakovchuk et al., 2009)  (E) Boxplot depicts distribution of SINE B2 RNA processing rate among different age groups of mice between wild type and APP. p < 0.05 between 6m old APP and 6m old WT (depicted as asterisk, n=3/group, t-test).  (C) Metagene analysis of increase in Hsf1 binding (ChIP-seq data) across the start of B2 RNA regulated genes (Suppl. Table 1) between pre-stress and 12 minutes post-stress (KS<0.05).
ChIP-deq data are from Mahat and colleagues (Mahat et al., 2016). TSS represents the Transcription Start Site of these genes.
(D) Boxplot depicts distribution of expression levels of Hsf1 gene among different age groups of mice between wild type and APP. Values are based on TPM counts of long-RNA-seq data.

Hsf1
Amyloid load Amyloid load