Natural SINEUP RNAs in Autism Spectrum Disorders: RAB11B-AS1 Dysregulation in a Neuronal CHD8 Suppression Model Leads to RAB11B Protein Increase

CHD8 represents one of the highest confidence genetic risk factors implied in Autism Spectrum Disorders, with most mutations leading to CHD8 haploinsufficiency and the insurgence of specific phenotypes, such as macrocephaly, facial dysmorphisms, intellectual disability, and gastrointestinal complaints. While extensive studies have been conducted on the possible consequences of CHD8 suppression and protein coding RNAs dysregulation during neuronal development, the effects of transcriptional changes of long non-coding RNAs (lncRNAs) remain unclear. In this study, we focused on a peculiar class of natural antisense lncRNAs, SINEUPs, that enhance translation of a target mRNA through the activity of two RNA domains, an embedded transposable element sequence and an antisense region. By looking at dysregulated transcripts following CHD8 knock down (KD), we first identified RAB11B-AS1 as a potential SINEUP RNA for its domain configuration. Then we demonstrated that such lncRNA is able to increase endogenous RAB11B protein amounts without affecting its transcriptional levels. RAB11B has a pivotal role in vesicular trafficking, and mutations on this gene correlate with intellectual disability and microcephaly. Thus, our study discloses an additional layer of molecular regulation which is altered by CHD8 suppression. This represents the first experimental confirmation that naturally occurring SINEUP could be involved in ASD pathogenesis and underscores the importance of dysregulation of functional lncRNAs in neurodevelopment.


INTRODUCTION
Autism Spectrum Disorders (ASD) are a heterogeneous group of complex neurodevelopmental conditions characterized by socialcommunicative deficits as well as repetitive sensory-motor behaviors, appearing during early childhood (American Psychiatric Association, 2013). ASD prevalence is steadily increasing, such that the estimated global prevalence is currently 1 in 68 (Elsabbagh et al., 2012). Adding to complexity, prevalence in males is 4 to 5-fold higher than in females (Baio et al., 2018). Albeit affecting such a significant portion of the world population, the underlying mechanisms of the disease have not yet been elucidated. However, several etiological hypotheses have been proposed, with risk factors ranging from environmental, to epigenetic (Shulha et al., 2012;Ladd-Acosta et al., 2014), to genetic. Consistently with clinical heterogeneity, the genetic architecture of ASD includes variable inheritance patterns, including rare de novo variants, chromosomal alterations, and common inherited variation (De Rubeis and Buxbaum, 2015;de la Torre-Ubieta et al., 2016). To date, more than 1,000 genes have been ranked as potential risk factors for ASD [SFARI Gene (Abrahams et al., 2013)], and it is challenging to determine whether they converge on shared molecular mechanisms.
Therefore, we decided to investigate the potential presence of functional lncRNAs among the dysregulated genes in CHD8 suppression Human induced Neural Progenitor Cells (hiNPCs) model, hypothesizing that they may constitute a further layer of molecular regulation in ASD. However, in silico prediction of lncRNAs functionality is intrinsically challenging, as non-coding transcripts only rarely have a modular structure, therefore structure-to-function relationships are not always straightforward (Mattick, 2018). SINEUP is a novel class of functional antisense lncRNA, which can up-regulate protein translation of their target sense mRNAs, without altering their transcription (Zucchelli et al., 2015a;Zucchelli et al., 2015b). First discovered in mouse (Carrieri et al., 2012), where Uchl1-AS was found to up-regulate protein translation of Uchl1 mRNA, SINEUP translational increase is mediated by two functional domains, namely 1) a region overlapping the Translational Initiation Site (TIS), head-to-head antisense to the 5′ end of the target sense mRNA, which confers specificity to the protein coding transcript (binding domain, BD) and 2) a SINEB2 repeat, Alu, MIR transposable element (TE) (Carrieri et al., 2012;Patrucco et al., 2015;Zucchelli et al., 2015b;Schein et al., 2016) on the 3′ end, that mediates the effect on the target mRNA translation (effector domain, ED). Several transcripts with this structure were computationally identified, and their function as SINEUPs successfully confirmed (Carrieri et al., 2012;Schein et al., 2016). Thus, it is possible to confidently hypothesize the function of such lncRNAs based merely on their structure.
In this work, we sought to identify SINEUP-like molecules among the dysregulated transcripts in an ASD-relevant cellular model system, human neural progenitors where CHD8 expression was suppressed by approximately 50% using short hairpin RNAs (shRNAs), mimicking the haploinsufficiency condition (Sugathan et al., 2014). Among the identified candidates, we prioritized the RAB11B-AS1 lncRNA, and provided experimental evidence of its regulatory role on its sense counterpart RAB11B mRNA by means of its SINEUPspecific domains. Our results suggest that ASD transcriptional dysregulation might affect previously unrecognized lncRNAsmediated networks and underline SINEUP molecules as unacknowledged players in ASD molecular phenotypes.

Identification of Candidate SINEUP
SINEUP-like transcripts were identified among the Differentially Expressed Genes (DEGs) comparing WT and CHD8 KD hiNPCs (GSE61491, GEO, NCBI) (Sugathan et al., 2014) according to a series of filtering steps ( Figure 1A): 1) LncRNAs were selected relying on GENCODE v37 lncRNA gene annotation (Frankish et al., 2019); 2) Screening for SINE/Alu or SINE/MIR TEs was performed using Dfam Tool Repeat Masker v3.0 (Smit et al., 1996); 3) Antisense overlapping transcripts to sense coding mRNAs were identified using the BioConductor package GenomicRanges (Lawrence et al., 2013) (minoverlap 1 L); 4) Transcripts overlapping the respective sense protein coding gene on the first ATG were chosen by comparing the "start codon" position of the sense mRNA to exons "start" and "end"positions of the lncRNAs from GENCODE v37 comprehensive annotation, and confirmed by using Ensembl (Yates et al., 2020) and UCSC Genome Browser (Kent et al., 2002) annotations.

RNA Extraction and Retrotranscription
Total RNA was extracted using TRIzol (#15596018 Ambion, Life Technologies) following the manufacturer's instructions. Genomic DNA was removed using DNase I (#AM2222  with Oligo-dT/random hexamers primers according to manufacturer's instructions. cDNA diluted 1:10 was used for qPCR. Transcripts relative expression levels in human tissues were determined using Human Total RNA Master Panel (#636643, LOT1409502A, ClonTech). 1 µg was retrotranscribed for each tissue/cell type as previously described, and cDNA was diluted 1:20 for application in qPCR.

Quantitative PCR
Primers for qPCR were designed spanning an exon-exon junction by using the Universal Probe Library Assay Design Center (Roche Life Science, 2019). iTaq ™ Universal SYBR ® Green Supermix (#1725121, Biorad) was used following manufacturer's instructions. NONO reference gene (Eisenberg and Levanon, 2013) was used for normalization, and relative expression values were calculated using the 2 −ΔΔCq method (Segundo-Val and Sanz-Lozano, 2016). The co-expression pattern of S/AS pairs across Human Total RNA Master Panel was evaluated by plotting the normalized and relativized expression values (2 −ΔΔCq ) matrix into a heatmap. In this case, for each gene, the ΔΔCq ratio was calculated with respect to the highest expression value across tissues. Amplicons size and specificity were verified through gel electrophoresis and Sanger sequencing.

Statistical Analysis
Statistical analysis tests were performed using R as described in figure captions. Significance level was set to 0.05. Data were plotted using R (ggplot2) and represented as Mean ± Standard Error of the Mean (SEM), as specified in figure legends with sample sizes. The significance level was reported as NS p > 0.05, *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001.

RESULTS
Identification of RAB11B-AS1 as a SINEUP-like Transcript Dysregulated Upon CHD8 Suppression To select functional lncRNAs altered following CHD8 suppression, we applied the selection pipeline schematized in Figure 1A and detailed in Methods. We first resorted to the complete list of dysregulated genes from Sugathan et al., 2014, where 1756 DEGs were reported following CHD8 KD compared to control hiNPCs (shGFP). Selection of natural SINEUP molecules was performed based on their specific structural criteria 1) annotation as lncRNAs, 2) presence of a SINE/Alu and/or SINE/MIR TE, 3) antisense to protein coding genes, 4) overlapping the start codon of coding gene). These sequential Frontiers in Genetics | www.frontiersin.org November 2021 | Volume 12 | Article 745229 filtering steps ( Figure 1A and Methods) led to the isolation of five candidate lncRNA genes ( Table 1) containing at least one inverted SINE/Alu and/or SINE/MIR repeats and overlapping in antisense orientation to the TIS of their respective sense protein coding mRNA. RAB11B-AS1 (ENSG00000269386) was identified as the most promising candidate, since the structure of the transcript precisely mirrored the one of a canonical SINEUP molecule ( Figure 1B). RAB11B-AS1 transcript is the antisense counterpart of a sense, protein coding gene, RAB11B (ENSG00000185236), a small GTPase belonging to the Ras superfamily responsible for vesicle formation, transport, and fusion (Stenmark and Olkkonen, 2001). RAB11B is enriched in the brain (Lai et al., 1994), and it is involved in membrane and vesicle trafficking and apical proteins recycling (Kelly et al., 2012), processes of relevance for brain development and synaptic plasticity (Villarroel-Campos et al., 2014). The RAB11B S/AS pairs were both significantly up-regulated upon CHD8 suppression in hiNPCs (Table 1). Notably, another previously generated and independently characterized sh-CHD8 suppression model (Cotney et al., 2015) displayed similar RAB11B-AS1 upregulation, although not nominally statistically significant after multiple test correction. RAB11B-AS1 overlaps in opposite orientation with RAB11B, specifically with 96 nucleotides encompassing the TIS, representing the putative BD. As for the ED, RAB11B-AS1 contains two classes of partially overlapping inverted embedded TE, a FRAM repeat (free right arm monomer) and a SINE/Alu repeat. Because SINE/Alu might arise from dimerization of two different REs (Mighell et al., 1997), the 2 TEs were jointly considered as the potential RAB11B-AS1 ED, a 222 nucleotides long region near the 3′ end of the transcript ( Figure 1B). Significantly, an ortholog transcript in M. musculus, Gm17251 (ENSMUSG00000090952), was identified, displaying a high sequence similarity [83% identity score, BLAST (Nucleotide BLAST, 2019)] to the human counterpart ( Figure 1C). Equivalently to the human transcript, it possesses the SINEUP-like putative BD, overlapping with the sense Rab11b (ENSMUSG00000077450) on the TIS, and a putative ED consisting of a SINEB2 repeat in inverted configuration ( Figure 1C). Because co-expression of the transcripts pair is essential to SINEUP protein translation function, the spatio-temporal co-expression of RAB11B and RAB11B-AS1 S/AS pair was examined. RAB11B and RAB11B-AS1 transcripts levels were quantified across an RNA panel from various human body districts. RAB11B-AS1 showed a fairly ubiquitous distribution, with detectable levels in skeletal muscle, testis and heart and highest expression in the central nervous system (spinal cord, whole brain) ( Figure 1D). Importantly, RAB11B-AS1 and RAB11B displayed a concordant expression pattern primarily in whole brain, heart, and skeletal muscle ( Figure 1D), thus supporting a possible S/AS functional regulatory mechanism.

RAB11B-AS1 and CHD8 Display Inversely Correlated Expression
In order to validate the transcriptional upregulation initially observed following CHD8 suppression (Sugathan et al., 2014) ( Table 1), RAB11B/RAB11B-AS1 S/AS pair was quantified by qPCR in independent biological replicates of CHD8 KD hiNPCs. Conforming to initial RNA-seq results (Table 1), RAB11B exhibited a mild upregulation (p-value 0.06), while RAB11B-AS1 dysregulation was more robust and significant in CHD8-suppressed lines (Figure 2A). We further calculated a linear regression analysis to appreciate possible correlation between the levels of CHD8 KD and the expression of the S/AS pair. By resorting to the initial logCPM from the hiNPCs models with CHD8 suppression (Sugathan et al., 2014), we uncovered a significant anti-correlation between RAB11B-AS1 and CHD8, while RAB11B correspondence was milder ( Figure 2B).
Next, to further dissect the expression crosstalk between CHD8 and RAB11B/RAB11B-AS1 S/AS pair we resorted to blood transcriptomic data of the Italian Autism Network (ITAN) (Muglia et al., 2018). RNA-seq data derived from peripheral blood samples of ASD and unaffected siblings (Filosi et al., 2020) were tested. While a modest (R ASD 0.036; R NO ASD −0.18) anti-correlation between RAB11B and CHD8 was observed ( Figure 2C, top), a significant inverse correlation between RAB11B-AS1 and CHD8 expression levels was found in both ASD and control siblings ( Figure 2C, bottom). Altogether, these results suggest a possible functional suppression mechanism by CHD8 on the RAB11B/RAB11B-AS1 locus, which might be impaired in CHD8 haploinsufficiency conditions.

RAB11B-AS1 Over-Expression is Able to Enhance RAB11B Translation With No Transcriptional Alteration
Because a measurable effect of a functional SINEUP molecule is the increase in translation of its sense counterpart and considering the over-expression of RAB11B-AS in the CHD8  (Sugathan et al., 2014). Linear regression analyses unveiled a significant inverse correlation for RAB11B (R 2 0.39, p 0.01697) and particularly for RAB11B-AS1 (R 2 0.8, p 1.741e-05) compared to CHD8. (C) Dot plots display the abundance (CPM) of RAB11B (top) and RAB11B-AS1 (bottom) compared to CHD8 from RNAseq analyses of peripheral blood samples from ASD patients (ASD diagnosis) and healthy siblings (NO ASD) in the ITAN family cohort. RAB11B-AS1 transcript abundance results in significative inverse correlation relationship with CHD8 both in the ASD patients group (R −0.38, p 0.0007) and in the healthy family members, i.e., the control group (R -0.59, p 1.63e-08).
Frontiers in Genetics | www.frontiersin.org November 2021 | Volume 12 | Article 745229 suppression lines, we predicted increased levels for RAB11B protein. After Western Blot quantification, densitometric analysis of the 24 kDa bands corresponding to RAB11B highlighted a significant increase in protein levels upon CHD8 suppression with respect to the control condition ( Figures 3A,B). While RAB11B upregulation was solid and reproducible in CHD8-Sh2 and CHD8-Sh1, displaying roughly 50% of CHD8 KD, the data on the third KD line (CHD8-Sh4) seem to be more variable. Finally, to functionally characterize RAB11B-AS1 as a SINEUP molecule, the full-length (WT) sequence of the transcript ( Figure 3C, top) was cloned and over-expressed in GM8330-8 hiNPC parental line. As a result of RAB11B-AS1 over-expression, RAB11B protein level was increased by approximately two times compared to control ( Figures  3C,D). RAB11B protein up-regulation was significant and reproducible, as confirmed by statistical analysis of replicate experiments ( Figure 3D). Furthermore, qPCR experiments confirmed that, despite the RAB11B protein increase, RAB11B transcriptional levels were substantially stable ( Figure 3E) while RAB11B-AS1 was abundantly overexpressed ( Figure 3F). These results strongly support our initial hypothesis, as they are coherent with the functional mechanism of a SINEUP molecule.

RAB11B Translational Increase is Dependent on the Presence of RAB11B-AS1 SINEUP Functional Domains
In order to fully prove the SINEUP nature of RAB11B-AS1 lncRNA, we wanted to test whether the absence of one of the putative functional domains might impair the SINEUP-like mechanism. To this purpose, two domain-specific deletion mutants were generated by site-specific mutagenesis (ΔBD or ΔED RAB11B-AS1). RAB11B-AS1 WT, ΔBD or ΔED were then delivered in parental GM8330-8 hiNPCs, and subsequently Western Blot experiments were performed to quantify RAB11B protein level. While the over-expression of WT, full-length RAB11B-AS1 elicited the expected increase in RAB11B protein, ΔBD, and ΔED mutants failed to evoke RAB11B protein upregulation, in line with the anticipated SINEUP activity ( Figure 4A). Such observation was confirmed by densitometric analysis on replicated experiments (n 4-6) ( Figure 4B). Importantly, qPCR revealed that RAB11B transcriptional levels were stable ( Figure 4C) while RAB11B-AS1 WT and deletion mutants were significantly and strongly over-expressed ( Figure 4D). These results suggest that RAB11B protein translation increase is mediated by its antisense transcript RAB11B-AS1 functional domains. Taken together, these results are reinforcing the hypothesis that RAB11B-AS1 lncRNA is a CHD8-suppression-sensitive SINEUP molecule, able to upregulate protein translation of its target mRNA RAB11B and potentially relevant for CHD8 haploinsufficiency defined ASD.

DISCUSSION
NcRNAs constitute the major product of mammalian transcription (The FANTOM Consortium, 2005), however their functions are still largely unexplored. Hinting at their possible role in higher cognition, lncRNAs are primarily expressed in the brain (Mercer et al., 2009), with definite patterns across cerebral areas, and several of them exclusively described in primates (Mattick, 2018). Increasing evidence underscores their role in neuronal physiology and pathology. In fact, lncRNAs have been implicated in neural development and FIGURE 4 | Deletion of functional SINEUP domains (BD, ED) impairs RAB11B-AS1 effect on RAB11B protein translation. (A) The RAB11B-AS1 (WT, ΔBD, and ΔED) sequence schematics report the location of the putative SINEUP domains (up). Representative image describes a Western Blot of RAB11B after RAB11B-AS1 WT, ΔBD, and ΔED over-expression. β-TUBULIN was used as loading control. (B) Barplot of RAB11B densitometric analysis, confirming that protein increase was statistically significant upon over-expression of RAB11B-AS1 (WT), while deletion mutants over-expression caused no protein expression difference from the control empty vector. (C) The barplot shows RAB11B transcriptional levels, tested by qPCR and expressed as 2 −ΔΔCq . mRNA levels were stable upon plasmid constructs overexpression (WT black, ΔBD light grey, ΔED dark grey) with respect to the control empty vector (white). (D) RAB11B-AS1 was abundantly and significantly overexpressed upon plasmid delivery (WT black, ΔBD light grey, ΔED dark grey) with respect to the control (white). Data are plotted as Mean ± SEM, N 4 for deletion mutants, N 6 for WT over-expression. For statistical analysis, un-paired one-tailed Student's t-test was performed. NS p > 0.05, *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001.
In this study, we sought to characterize the effects of transcriptional dysregulation of lncRNAs in a model system of neuronal development and relevant to ASD. Specifically, we resorted to hiNPCs, where CHD8 expression was reduced by short-hairpins administration to roughly 50%. CHD8 protein haploinsufficiency represents one of the highest confidence risk factors for ASD with profound consequences for the whole transcriptome. In this work, we identified SINEUP-like antisense lncRNAs among the pool of dysregulated genes following CHD8 suppression (Sugathan et al., 2014). SINEUP ncRNAs are a class of regulatory, antisense modular transcripts, which increase protein translation of their sense mRNA by means of their characteristic functional domains (Zucchelli et al., 2015a;Zucchelli et al., 2015b). Thus, we filtered, relying merely on structural features, the 1756 DEGs upon CHD8 suppression (Sugathan et al., 2014). Only an exiguous list of candidates met our stringent criteria. However, we cannot exclude that the number of dysregulated SINEUP molecules might be underestimated in our study. In fact, due to the poly-A mRNA enrichment protocol used for library preparation (Sugathan et al., 2014), a large portion of poly-A-minus lncRNAs (Mattick, 2018), and possibly also nonpolyadenylated SINEUP, might have been missed. A total of 5 SINEUP-like molecules have been identified (Table 1), however, we prioritized for further functional validation RAB11B-AS1, which displayed the structural organization more typically associated with natural SINEUP. Importantly, a murine ortholog of RAB11B-AS1 was identified, with an inverted SINEB2 TE. LncRNAs containing embedded TEs are more conserved across species with respect to non-TE-derived sequences, and display significantly less variance (Kapusta et al., 2013), sustaining the hypothesis that TEs in lncRNAs are subject to functional and/or structural constraints during evolution. Previous reports in osteosarcoma, lung, and breast cancer development described different, discrepant modes of RAB11B-AS1 regulation on RAB11B mRNA and protein levels: downregulation , upregulation (Li et al., 2020) or no effect (Niu et al., 2020) of the sense transcript was observed, generating an inconclusive scenario. Notably, RAB11B-the headto-head protein coding transcript, overlapping RAB11B-AS1-has critical roles in apical recycling of cargo proteins (Delisle et al., 2009;Silvis et al., 2009;Best et al., 2011;Butterworth et al., 2012). Moreover, it was reported to inhibit Ca 2+ -triggered exocytosis in neuronal and neuroendocrine cells, and to be enriched in purified synaptic vesicles (Khvotchev et al., 2003). Importantly, RAB11B de novo mutations were correlated with Intellectual Disability and microcephaly (Lamers et al., 2017). Thus, these observations globally support a role for RAB11B-and possibly its overlapping lncRNA-in vesicular trafficking and synaptic activity, of relevance for ASD and other neurological conditions. Firstly, we validated the upregulation of RAB11B/ RAB11B-AS1 transcripts pair following CHD8 suppression by qPCR. Secondly, we confirmed a comparable expression pattern between RAB11B-AS1/RAB11B across human body districts and CAGE data, coherently with previous observations reporting a similar spatio-temporal distribution of S/AS pairs (Chen et al., 2005). Furthermore, we uncovered an anti-correlation between CHD8 and RAB11B-AS1 in ASD-affected and healthy siblings of the ITAN cohort. Thus, aberrantly reduced expression of CHD8 seems to correlate with RAB11B-AS1 upregulation. However, in our hiNPCs model transcriptomic data, both RAB11B-AS1 and RAB11B appear to be upregulated, although with different strength and significance. While further studies will be needed to fully dissect this interplay, the observed upregulation of the sense RAB11B transcript might be directly mediated by CHD8 protein, since CHD8 binding sites were identified on RAB11B, but not on RAB11B-AS1 promoter (Sugathan et al., 2014).
Finally, we moved to the functional characterization of RAB11B-AS1 as a potential new SINEUP molecule. To this task, we cloned and overexpressed the full-length human lncRNA transcript. Over-expression of RAB11B-AS1 did not affect RAB11B transcriptional levels but led to a reproducible increase in the production of RAB11B protein. This posttranscriptional effect is consistent with a SINEUP role, as translation is typically expected to increase in the range of 1.5-3 fold (Zucchelli et al., 2015a). To further strengthen our results, we created deletion mutants of RAB11B-AS1, removing the SINEUP functional domains (BD and ED). Consistently with our hypothesis, the mutant forms of the transcript failed to exert a regulatory effect on RAB11B mRNA translation. Thus, here we propose that RAB11B-AS1 SINEUP molecule potentially represents a further indirect layer of protein translation regulation, independent of RAB11B transcriptional control. This finding seems to be discordant with previous studies Li et al., 2020;Niu et al., 2020), however, AS-lncRNAs have been previously reported to have dual functions, and this could depend on the cellular context and availability of specific co-factors. To this point, Uxt-AS1, initially found to act as a SINEUP by upregulating protein translation of its sense counterpart Uxt (Carrieri et al., 2012), in a later study was, instead, found to regulate alternative splicing of UXT in human colonic carcinoma cell lines (Yin et al., 2017). Thus, alternative roles for some lncRNAs could be described when using different cell lines or other cellular contexts or tissues. This could suggest that expression of different mediators could drive different functional effects of specific AS-lncRNA on their sense counterparts.
In conclusion, we provided evidence that naturally occurring SINEUP could be involved in ASD pathogenesis, highlighting the importance of dysregulation of functional lncRNAs during brain development.

ETHICS STATEMENT
The studies involving human participants were reviewed and approved by the Verona Hospital Ethical Review board (study protocol AUT-SFK001, CE1419) which approved the study protocol in first instance, followed by the Ethical Review Committees of each recruiting site for the ITAN collection.
Written informed consent to participate in this study was provided by the participants' legal guardian/next of kin.