Abstract
The smallest genomic region causing Prader-Willi Syndrome (PWS) deletes the non-coding RNA SNORD116 cluster; however, the function of SNORD116 remains a mystery. Previous work in the field revealed the tantalizing possibility that expression of NHLH2, a gene previously implicated in both obesity and hypogonadism, was downregulated in PWS patients and differentiated stem cells. In silico RNA: RNA modeling identified several potential interaction domains between SNORD116 and NHLH2 mRNA. One of these interaction domains was highly conserved in most vertebrate NHLH2 mRNAs examined. A construct containing the Nhlh2 mRNA, including its 3’-UTR, linked to a c-myc tag was transfected into a hypothalamic neuron cell line in the presence and absence of exogenously-expressed Snord116. Nhlh2 mRNA expression was upregulated in the presence of Snord116 dependent on the length and type of 3’UTR used on the construct. Furthermore, use of actinomycin D to stop new transcription in N29/2 cells demonstrated that the upregulation occurred through increased stability of the Nhlh2 mRNA in the 45 minutes immediately following transcription. In silico modeling also revealed that a single nucleotide variant (SNV) in the NHLH2 mRNA could reduce the predicted interaction strength of the NHLH2:SNORD116 diad. Indeed, use of an Nhlh2 mRNA construct containing this SNV significantly reduces the ability of Snord116 to increase Nhlh2 mRNA levels. For the first time, these data identify a motif and mechanism for SNORD116-mediated regulation of NHLH2, clarifying the mechanism by which deletion of the SNORD116 snoRNAs locus leads to PWS phenotypes.
Introduction
Prader-Willi Syndrome (PWS) is a devastating human genetic condition, affecting up to 1 in 10 000 live births. Affected infants present with hypotonia, as well as weak suckling and failure to thrive (1). Adolescents are hypogonadal and males show cryptorchidism, with hypogonadotropic hypogonadism at puberty (1). Hyperphagia typically begins around age 2 years (1), and results in morbid obesity in adulthood unless drastic calorie limitation is initiated.
PWS is most often due to a de novo paternal deletion of the chromosome 15q11-q13 region (at least 60% of cases) (1). The paternally-inherited 15q deletion minimally contains the SNORD116 locus, (a group of 30 small nucleolar RNAs ‘snoRNAs’), and the IPW gene (2). The SNORD116 snoRNAs are ‘orphan’ snoRNAs meaning that they have no predicted RNA targets.
A significant gap in knowledge exists in our understanding of how deletion of this region of 15q leads to the complex phenotypes of PWS. A hint about this came in late 2016 when Burnett and colleagues showed two mRNAs—NHLH2 and PCSK1 were downregulated in human PWS induced pluripotent stem cell (iPSC)-derived neurons, and in the Snord116m+/p- mouse—a model animal for PWS (3). As noted in their paper, targeted deletion of Nhlh2 (N2KO) in mice results in phenotypes that overlap that of human PWS, and the Snord116m+/p- mouse (3).
N2KO mice show an initial delay in total body weight until about 7 weeks of age in males, and then a later onset obesity, significant in both sexes by around 13 weeks of age (4). Total body fat in the females becomes significantly higher than age-matched wild-type females in the 13–19 week ages, while in males, body fat levels are significantly higher in animals older than 20 weeks old (4). Both of these phenotypes are similar in developmental stages to PWS patients. Likewise, N2KO mice demonstrate male cryptorchidism, with hypogonadotropic hypogonadism in males and females at puberty (5,6). Low physical activity levels are characteristic of both N2KO mice (4) and PWS patients—the latter showing low infantile movement (1) as well as overall physical activity levels and response to exercise (7). Both N2KO mice (KJ/hour, in preparation) and PWS patients (Kcal/day) (8) demonstrate low energy expenditure levels, contributing to overall lower metabolism levels for both the model organism and the human counterparts.
Nhlh2 is a neuronal basic helix–loop–helix transcription factor that controls genes within the leptin-melanocortin pathway (9), including prohormone convertase 1/3 (PCSK1/3) (10), melanocortin-4-receptor (MC4R) (11), and even a locus within the imprinted region of chromosome 15q, necdin (NDN) (12). Reduced expression of NHLH2 in PWS patients could explain both the obesity and hypogonadal phenotypes of these patients. However, a significant gap in knowledge exists about how deletion of the paternal 15q region would lead to lower NHLH2 levels.
Our expertise in NHLH2 regulation and function, combined with the findings by Burnett and colleagues led us to ask if SNORD116 might regulate NHLH2 levels at a molecular level. We specifically focused on SNORD116 due to the finding that the Snord116m+/p- mouse as well as iPSC derived neurons from patients with a minimal deletion that included SNORD116 showed the lower NHLH2/Nhlh2 protein and mRNA expression levels (3). In this paper, using in silico tools and a mouse hypothalamic neuron cell line, we demonstrate the mechanism by which Snord116 controls levels of Nhlh2. Together, these results effectively change SNORD116 from an orphan snoRNA, to one with a verified target, identify a SNORD116 interaction motif that can be used to search for more targets, and provide a molecular genetic mechanism for how deletion of the 15q11-q13 region in humans leads to PWS.
Results
In silico determination of an NHLH2: SNORD116–3 interaction motif
As the SNORD116 locus codes for orphan snoRNAs, meaning that there were no known interaction targets for any of its 30 snoRNAs, in silico tools were used to determine if any region of the NHLH2 mRNA showed strong interactions with any of the 30 SNORD116 snRNAs. Both the mouse and human NHLH2 genes contain three exons, with only a small coding region at the beginning of exon 3, and a long 3′ untranslated region (UTR) (Fig. 1A). Within the 3’-UTR, several regions of high homology exist between mouse and human NHLH2. An in silico RNA:RNA interaction analysis using all of the Human SNORD116 snoRNAs, and the human NHLH2 mRNA sequence indicated that at least one of the 30 SNORD116 transcripts had the potential to form a highly stable RNA:RNA interaction. As shown in Fig. 1A, the location of the putative interaction region is within a larger region showing 100% conservation between mouse and human NHLH2 (as indicated by the gray shading). In silico RNA:RNA analysis identified other SNORD116:NHLH2 dyads with lower interaction energies (shown as arrow heads, Fig. 1A), but these interactions were not pursued further, as they did not interact with homologous regions in the mouse and human NHLH2 mRNA, were not located in the snoRNA antisense region, and had much lower interaction energies compared to that for SNORD116–3: NHLH2. The putative SNORD116–3: NHLH2 interaction motif is highly conserved in vertebrates (Fig. 1B-C). For NHLH2, this region is 100% conserved in the vertebrates shown (Fig. 1B), and in 70/77 vertebrates tested phylogenetically (data not shown). SNORD116 consensus sequences for all paralogs within a cluster show high phylogenetic conservation in the predicted interaction region (Fig. 1C). Critically, the SNORD116–3 predicted interaction is in the canonical antisense region just 5′ of the D′ box, conserved in all species shown (Fig. 1C). Of note, the SNORD116–3 interaction motif on the NHLH2 3’UTR is found in 2 of 2 of the human NHLH2 splice variants, and 2 of 2 of the murine splice variants that are registered in the RefSeq database (Supplementary Material, Fig. S1).
Strongest predicted RNA:RNA interaction between SNORD116 and NHLH2 is found within highly conserved regions. (A) Nucleotide alignment for human and mouse NHLH2 mRNA sequences with predicted SNORD116 interaction sites. Predicted interaction sites are displayed in small antisense arrows above or below the given sequence. Light gray regions in the aligned sequences show identity agreement, black regions show disagreements, and horizontal lines show gaps in alignment. One region with high phylogenetic agreement and predicted SNORD116 interaction sites found in both human and mouse are labelled. (B) Phylogenetic alignment of NHLH2 3’UTR predicted interaction site among vertebrates. Nucleotide numbers above the alignment are based on human NHLH2, NM_005599. NCBI reference sequence numbers used for the analysis are shown. *indicates perfect non-ambiguous homology with the human sequence for all nucleotides in the site above. (C) Phylogenetic Alignment of consensus SNORD116. Underlines indicate C/D′ box motifs. Boxed region indicates predicted interaction site with NHLH2, which includes the canonical snoRNA antisense region. *indicates perfect non-ambiguous homology with the human sequence for all nucleotides in the site above. Human n = 30, chimp n = 28, rhesus n = 29, rabbit n = 29, rat n = 26, and mouse n = 17.
Numerous single nucleotide variants (SNV) exist within the NHLH2 and SNORD116–3 genomic loci, although none have been characterized as having clinical significance. However, an analysis of the effect of each naturally-occurring SNVs on the NHLH2 3’UTR in the region of this motif revealed five of these that disrupted the energy of interaction, as predicted using in silico RNA:RNA interaction analyses. Each of these disrupting SNVs were found towards the 3′ end of the motif (Fig. 2A) and reduce the predicted strength of interaction from −13.4 kcal/mol for the wildtype, to −9.4 for both rs1218168750 and rs1051613841 (Fig. 2B). As shown in Fig. 2C, compared to Fig. 2D rs1051613841 was predicted to disrupt the stem area of the NHLH2:SNORD116–3 dyad. The predicted mouse secondary structure for the wildtype (Fig. 2E) and mutagenized form (Fig. 2F) are also shown.
Single nucleotide variants have the potential to reduce the predicted NHLH2: SNORD116 motif interaction energy. (A) RNA alignment of the predicted SNORD116 interaction region for mouse and human NHLH2 with five human SNVs displayed. (B) Reference SNV data provided from NCBI SNP database, including accession number, and minor allele frequency (MAF). The predicted effect on the SNORD116–3 interaction energy (kcal/mol) with NHLH2 is shown. (C) INTA-RNA predicted structure of the wild-type NHLH2:SNORD116–3 interaction. (D) INTA-RNA predicted structure of the NHLH2:SNORD116–3 interaction containing NHLH2 SNV rs1051613841 (E) INTA-RNA predicted structure of the wild-type mouse Nhlh2:Snord116 interaction used in the current study. (F) INTA-RNA predicted structure of the mutagenized mouse sequence U2449C. SNV site of interest is labelled in bold, underlined, and/or with arrows.
The intact NHLH2:SNORD116–3 interaction motif mediates high Nhlh2 expression levels
N29/2 hypothalamic cells have been previously used to measure Nhlh2 expression at the transcriptional and posttranscriptional levels (13,14). A set of constructs, using a myc-tag motif to differentiate endogenous Nhlh2 from the transfected Nhlh2 with different 3′ tails were used to measure Nhlh2 response to Snord116 in hypothalamic cells. These constructs all contained the myc-tagged Nhlh2 (mouse) coding region linked to either high-expression SV40 3’UTR, a partial Nhlh2 3’ UTR (13), a full-length Nhlh2 3’-UTR, or a full-length Nhlh2 3’-UTR containing the rs1051613841 SNV (Fig. 3A). Each were transfected into N29/2 hypothalamic neurons (15) which express low levels of Snord116, and Nhlh2 during regular growth phase. Addition of a mouse Snord116 consensus expression plasmid (which is most phylogenetically homologous to group 1 SNORD116 snoRNAs (16)) to cells expressing the full-length Nhlh2-myc mRNA construct led to a ~ 15-fold increase of Nhlh2-myc mRNA levels (Fig. 3B). The construct containing the engineered SNV in the consensus motif, led to a smaller, yet still significant increase of ~ 6-fold in Nhlh2-myc mRNA levels, while the partial tail without the predicted interaction site showed no increase in response to Snord116 overexpression. Interestingly, the construct with the SV40 3’UTR showed a similar increase of ~ 5-fold, although the strongest predicted Snord116 interaction site for this tail was a relatively weak (−5.54 kcal/mol) in the 3’UTR. Figure 3C shows that the different 3’UTR constructs lead to different steady-state Nhlh2 protein levels, consistent with the mRNA levels. Like the protein levels, SV40 3’UTR shows greater baseline RNA expression when normalized across 3’UTR (Supplementary Material, Fig. S2). Snord116 overexpression ranged from about 100–1000-fold higher (Supplementary Material, Fig. S3). Of note, all constructs were under the control of the CMV promoter and had the identical Nhlh2-myc coding region through the Nhlh2 stop codon. Thus, differences in expression level could only be attributed to differences in the 3’-UTR of the constructs, and the presence/absence of the putative motif. These data suggest that Snord116 controls expression of Nhlh2, post-transcriptionally, through the 3’UTR motif identified by in silico binding analysis.
Post-transcriptional Regulation of Nhlh2 by Snord116 is reduced by a SNV in the predicted interaction region within the Nhlh2 3’UTR. (A) Reporter plasmids expressing mouse Nhlh2 mRNA. Mouse Nhlh2 mRNA (NM_178777) with annotated exons, coding sequence (CDS), and predicted Snord116 interaction sites. Four different expression vectors for tagged Nhlh2 mRNA with varying 3’UTRs are shown to scale in comparison to Nhlh2 mRNA. The mouse vector was engineered to contain a T > C SNV within the predicted interaction region which is shown by the white bar on that vector. (B) Relative expression of tagged Nhlh2 RNA with different 3’UTRs during Snord116 overexpression in N29/2 cells. Nhlh2 SNV at predicted Snord116 interaction site decreases Snord116-induced fold change. For the +Snord116 condition, one-way ANOVA F (3,29) = 30.992, p < 0.001 with Bonferroni correction shows the only 3’UTR types to not be significantly different to each other are SV40 and SNV. All other comparisons are significantly different (p < 0.05); N = 6 for partial 3’UTR conditions; N = 9 for SV40, Full, and SNV conditions. Error bars indicate ±SD. **p < 0.01, ***p < 0.001. Statistical analysis performed on respective ddCT values from which Relative Quantification (RQ) values are derived. (C) Representative Western blot shows higher protein translation of the +Snord116 conditions and the SV40 3’UTR construct. Both blots are from the same samples but the bottom blot is with anti-myc tag antibody alone to allow for longer signal development. Note the western blot is normalized across 3’UTR type, unlike Figure 3B. (D) Optical density of representative Western blot.
Snord116 snoRNA stabilizes Nhlh2 mRNA
While post-transcriptional regulation of mRNA can be due to changes in mRNA translation efficiency, the 3’-UTR of an mRNA often directs regulation due to changes in mRNA stability. To test whether Snord116 levels altered mRNA stability of Nhlh2, the same constructs as above and the N29/2 cell line was used in an actinomycin D assay. Actinomycin D (also known as dactinomycin) is an antibiotic and chemotherapeutic that inhibits DNA-dependent RNA synthesis through DNA intercalation and inhibition of RNA polymerase I, II, and III (17,18). As transcription of new mRNA is blocked when actinomycin D is added to a cell culture, the level of mRNA at times following addition of the chemical can be used to determine mRNA stability. As shown in Fig. 4, Snord116 shows general overall stabilization of the SV40, Full, and SNV tails, throughout the 3-hour time course. This is consistent with Snord116’s effect on steady-state levels (Fig. 3B). The stabilization effect size is greater for the Full and SNV tails than SV40 (Fig. 4E). Additionally, this difference in stabilization effect size seems to be mediated by an early rapid decay of the Full/SNV tails without Snord116. About 50% of the RNA from the SNV/Full tails are degraded by 45 minutes, while both the partial/SV40 tails’ levels remain higher (p < 0.05) and degrade to about 50% between 90–180 minutes (Fig. 4A-D). This indicates a relative baseline instability for the Full/SNV tails. This high initial decay rate within the first 45 minutes also slows down by the 90 min and 180 min intervals for Full/SNV tails (Supplementary Material, Fig. S4). Interestingly, with the addition of Snord116, this initially high rate of decay is blunted, indicating that Snord116 has a protective effect against RNA decay within the first 45 minutes for the Full/SNV tails (Fig. 4G). Critically, the Snord116 protective effect at 45 minutes trends to being less effective for the SNV tail when compared to the full tail but does not reach a significance difference between the two (p = 0.10). Of note, Snord116 RNA does not decay over time (Supplementary Material, Fig. S5).
Snord116 stabilizes target RNAs and protects against early rapid decay dependent on 3’UTR. (A-D) Relative expression of reporter mRNAs with different 3’UTRs during Snord116 overexpression and actinomycin D treatment. Snord116 overexpression indicated by solid lines, and negative control indicated by dotted lines. Significance indicators for main effect across all time points between +/-Snord116 are shown just below the title of each figure. Significance indicators for individual time points between +/-Snord116 are shown just above the X-axis. (E) Fold change of +Snord116 relative to -Snord116 condition for each 3’UTR and time point. (F) Relative expression of Full and SNV 3’UTRs modified from Fig. 4C-D. Boxed area emphasizing 45-minute time point for the context of Fig. 4G. (G) Decay rate shown as percent decay for the 45-minute time point for each 3’UTR and Snord116 condition. Significance indicators correspond to differences between +/-Snord116 conditions within a 3’UTR condition. N = 6 for partial 3’UTR condition; N = 9 for SV40, Full, and SNV conditions. Error bars indicate ±SD. *p < 0.05, **p < 0.01, ***p < 0.001, NSp > 0.05.
Discussion
Snord116 has been previously implicated in changes of gene expression, with proposed molecular mechanisms including alternative splicing and RNA-stability (3,19–21). The current study is the first to demonstrate that RNA stability indeed is a molecular mechanism for Snord116-dependent changes in gene expression. As this study was done using cell lines and molecular constructs, there are minimal confounds from the PWS locus and the Snord116 host gene (116HG). Rather, these results can be directly attributed to a Snord116 snoRNA and its target gene.
The current study is the first to show evidence for an interaction of Snord116 RNA with a predicted target. Although RNA interaction prediction for targets of SNORD116 have been performed previously, there was no experimental validation of these interactions (19,22). Using the snoTARGET software (22), previous predictions failed to include NHLH2 as a potential target. Furthermore, the predicted strength of interaction with NHLH2 is relatively low compared to other predicted target RNAs. The current study suggests that while interaction prediction is a useful tool, as implicated by the predicted effect of the SNV on the 3’UTR:Snord116 interaction, other factors, such as protein or other interactions should be considered.
Using both the data in this paper, and previous work, we predict that high NHLH2 levels may be present in cells expressing SNORD116 and/or treated with leptin. Snord116 expression from wild type mouse hypothalamus may increase with refeeding after overnight fasting (3). As both leptin and Snord116 levels increase during a refeeding period, these two paths appear to converge synergistically on Nhlh2 upregulation: transcriptional upregulation of Nhlh2 through leptin signaling (14), and post-transcriptional upregulation of Nhlh2 through Snord116. While some previous studies have not detected changes in the levels of hypothalamic NHLH2 mRNA in PWS patient post-mortem tissue or PWS model mice (19–21), our data suggest that changes in Nhlh2 levels may occur with a short temporal window that the other studies failed to capture. Conversely, another study has demonstrated that growing lymphoblastoid cells derived from PWS patients show a − 1.5 fold expression of NHLH2 mRNA compared to normal lymphoblastoid cells (23).
It may be questioned if the current study is relevant to the biological levels of the genes in question. In our approximation, when compared to mouse hypothalamus tissue by RT-QPCR, the current study’s Snord116 levels are greatly under-expressed, and Nhlh2 levels range from equal expression to slight under-expression (data not shown). Part of the rationale for using the N29/2 cell line is that Snord116 levels are low, replicating a PWS-like condition, and allowing specific overexpression of exogenous Snord116. Precise quantification of these data between a cell line and whole mouse hypothalamus presents issues, due to a lack of suitable validated reference genes for RT-QPCR appropriate for use. Considering raw CT values and equal total RNA per RT-QPCR reaction, endogenous Snord116 in mouse hypothalamus is 8000—100 000-fold higher than our overexpression conditions in N29/2 cells. This general pattern of expression is a large enough difference that the authors are comfortable coming to this general estimate and conclusion for Snord116 levels between cell-line and tissue. However, it must be addressed that RT-QPCR of Snord116 can sense both the non-spliced Snord116 host gene, and the mature Snord116, so this too may partially explain the large difference between N29/2 cells and mouse hypothalamus levels.
The current study shows that high expression of Snord116 leads to higher stability of Nhlh2 mRNA and subsequently, higher levels of translated protein. At steady-state levels, the SNV 3’UTR seems to disrupt Snord116-dependent upregulation when compared to the full wildtype 3’UTR, implying that the predicted interaction site is playing a role in Snord116 interaction (Fig. 3B). However, this disruption from the SNV is not seen in the RNA decay experiment (Fig. 4). The Snord116 protective effect at 45 minutes only trends to being less effective for the SNV tail when compared to the full tail (p = 0.10). This may be due to some differences between steady-state and RNA decay experiments. The steady-state experiment may be sensitive to the accrual of RNA over a 24-hour period post-transfection and any RNA stability differences may lead to a cumulative effect. The RNA decay experiment is sensitive to a smaller timeframe, and highlights that the initial 45 minutes may be critical for stabilization effects between the Full and SNV 3’UTRs. Therefore, small differences in stability that do not show statistical significance at 45 minutes may be amplified when these small differences accumulate over time and show differences at steady-state levels. Additionally, it is possible that a more disruptive SNV may show stronger effects than the one used in the current study. Furthermore, the steady-state experiment did not include a media refresh prior to cell lysis, while the RNA decay experiment refreshed the media with the addition of Actinomycin D at the 0-minute timepoint, which could add a serum or media factor that affects expression levels.
While the current study implies Snord116 stabilizes target RNAs, it is possible that results from steady state experiments are influenced by differences in transcription (Fig. 3B). However, the consistent pattern of effect size between steady-state and RNA decay studies implies that RNA stability is playing a major role. Future work will have to examine possible effects on transcription.
The use of the Snord116 expression construct minimizes confounds associated with the PWS genomic locus, as there are many species of ncRNAs and host genes expressed from the locus (24–26). Use of the N29/2 neuronal cell line, which has endogenously low levels of Snord116 lends confidence that the results shown are due to an overexpression of Snord116 rather than changes in the many host genes and ncRNAs found at the PWS genomic locus that are often associated with a genomic deletion. It is worth noting that the construct used contains Snord116 within its natural intron, as this is necessary for proper snoRNA maturation and processing (20). The natural genomic processing of the Snord116 host gene (116HG) results in a much longer ncRNA with many exons that are not within the construct used (24,27). There is no way to completely exclude a requirement for the endogenous host gene or other genes within the PWS locus which are present in N29/2 cells. Expression levels of 116HG were not tested in the present study, while Snord116 snoRNA levels were tested by RT-QPCR to validate overexpression (Supplementary Material, Figs S3 and S5).
Use of the SNV-containing construct appears to confirm that the interaction motif is within the predicted regions on both Nhlh2 and Snord116. Based on this, a model was created that predicts the secondary structure and interaction between NHLH2 and SNORD116–3 (Fig. 5). As shown in the prediction, interaction with SNORD116–3 occurs within a stem-loop structure on the NHLH2 3’UTR. In examining the structure, the stem-loop may be either stabilized or disrupted by SNORD116–3 interactions, which likely depends on RNA-binding protein interactions within that region. Of interest, hnRNP-U has previously been shown to stabilize NHLH2 mRNA, although the position of that interaction is not known (28). Thus, while the most well-established mechanism of non-methylating SNORDs is through pre-mRNA splicing (20), enhancement of mRNA stability by SNORDs remain a tantalizing mechanism. It remains to be seen whether the mRNA stability observed in the current study is through the canonical SNORD mechanism of 2’-O methylation or non-canonical mechanisms.
Predicted model of SNORD116:NHLH2 interaction. The ViennaRNA package software was used to predict the folding structure for NHLH2 3’UTR, and then the interaction with SNORD116. The miR-9-5p interaction site was identified using TargetScan. Both interactions were added to the overall structure for viewing of their relative positions on the NHLH2 3’UTR only, not to indicate whether one or both simultaneously interaction with the stem-loop structure. *indicates the site of the NHLH2 SNV used in the current study.
While the predicted RNA interaction in Nhlh2 3’UTR and the SNV results indicate a possible direct RNA:RNA interaction region, the increased mRNA stability seen with the SV40 3’UTR indicate that Snord116’s mRNA stability effect may be non-specific. The predicted interaction between Snord116 and the SV40 3’UTR is relatively weak and does not correspond to the Snord116 antisense target region (Supplementary Material, Fig. S6). This differs from the Nhlh2 3’UTR that is predicted to be complementary to the Snord116 antisense target region. These results suggest that Snord116 may have general effects on RNA stability independent of sequence complementarity. However, the partial 3’UTR is unaffected by Snord116 indicating that Snord116’s effect may be dependent on RNA sequence and potential co-factors.
In examining the NHLH2 RNA sequence around the putative SNORD116 interaction motif using RegRNA 2.0 (29), there are no overlaps of known motifs within the putative SNORD116 interaction motif. However, a poly(A) signal is about 100 bp downstream of the predicted interaction site for both mouse and human. The mouse transcript has a splice variant that ends at this poly(A) signal (XM_006501112), while the one used in the current study is the splice variant with the longest 3’UTR (NM_178777) (Supplementary Material, Fig. S1). Snord116 may be mediating it’s RNA stability effect through poly(A) signal pathways, as the partial 3’UTR RNA used in the current study does not contain a strong poly(A) signal, while the SV40 3’UTR does. Additionally, in mouse Nhlh2, but not human, there is a 60-72 nt (depending on strain) trinucleotide repeat region of GAA about 20 nt upstream of the predicted Snord116 interaction region. This region is predicted to be an exon splicing enhancer, although there is little evidence of any splice variants using this exon splice site. Furthermore, just five base pairs upstream from the interaction site is a putative microRNA interaction site, predicted using TargetScanHuman (30). Both the SNORD116 and miR-9-5p sites lie within the predicted stem-loop structure on the NHLH2 mRNA 3’UTR (Fig. 5). MiR-9-5p is a neuronally-expressed microRNA whose dysregulation has been implicated in a number of neurodegenerative diseases (31). Interestingly, interaction with miR-9-5p has been shown to decrease target mRNA stability (32,33). Experiments to examine if SNORD116 disrupts the NHLH2 stem-loop structure are out of the scope of this study, but would allow one to determine if miR-9-5p directly binds to and increases the stability of NHLH2 mRNA. In either scenario, the possible interaction with miR-9-5p or other secondary factors, including proteins may explain why the SNV had a weaker effect overall on Nhlh2 stability levels than we had predicted using in silico tools, and suggest that a second region mediating additional mRNA stability control for NHLH2 may occur outside of the predicted NHLH2:SNORD116 interaction motif. The entire region shown in Fig. 5 has high homology in vertebrates (data not shown). Phylogenetic conservation of small motifs in untranslated regions suggest a functional conservation (34,35) that can be used to predict and test functional interactions, as was done herein. This longer region within the NHLH2 3’UTR can be used to identify other possible SNORD116–3 targets that may also contain this motif in their 3’-UTR. Likewise, as all of the SNORD116 snoRNAs within group 1 are highly conserved but still have some nucleotide-based substitutions between them (16), we can use this information to determine if all group 1 snoRNAs can interact with similar targets or whether each has a specific set of targets with highly conserved/high energy interaction. Indeed, as shown by our work, a single SNV on the target mRNA can reduce the ability of overexpressed Snord116 RNA to stabilize the target mRNA. It is possible that the reverse is true as well—with SNVs within the SNORD116 locus possibly leading to PWS-like phenotypes.
While we do not find an exact match to the NHLH2:SNORD116 motif in the PCSK1 mRNA or other RNA-seq identified RNAs to date (3), it is possible that a similar motif, and different SNORD116 cluster snoRNAs contribute to these downregulations in PWS models. It is also possible that downregulation of PCSK1 and other mRNAs in neuronally-induced PWS stem cells simply results from the loss of direct, leptin-induced transcriptional regulation by NHLH2, as we have previously shown for Pcsk1 (10). Additionally, there is a convincing mechanism for PCSK1 downregulation through MAGEL2 loss in PWS patients with a large deletion of the genomic locus (36). However, these findings do not explain the low levels of PCSK1 mRNA and protein found in a PWS microdeletion model of neuronally differentiated iPSCs containing a genomic deletion of only SNORD109A, SNORD116 cluster, and IPW (3). Additionally, PWS mouse models with intact Magel2 and loss of the paternal Snord116 cluster have lower levels of Pcsk1 mRNA and protein in islet cells and stomach (3,36). These data suggest that loss of Snord116 may lead to downregulation of Nhlh2-dependent Pcsk1 in addition to the MAGEL2 mechanism. This may add to the ongoing explanation of why PWS patients with large genomic deletions show stronger phenotypes.
NHLH2 is a basic-helix–loop–helix transcription factor, with multiple known and putative targets (9–11,37–39). Many of these targets (i.e. PCSK1, MC4R) are involved in neuronal control of body weight, and may provide an explanation for some of the phenotypes of PWS. SNORD116-mediated post-transcriptional regulation of NHLH2 could result in hundreds of downstream regulatory changes. For PWS patients, these data now suggest why PWS patients, and the Nhlh2 knockout mouse share many of the same phenotypes (3,9), and may open further analyses into therapeutic interventions that can increase levels of NHLH2 or one of its transcriptional targets, even in individuals with impaired SNORD116.
Materials and Methods
Nucleotide Alignments
Nucleotide alignment and annotation performed using Geneious Prime 2020.1.2 (https://www.geneious.com, Biomatters, New Zealand). The ‘Geneious alignment’ algorithm on default settings was used for pairwise nucleotide alignment of human and mouse sequences with annotations. Phylogenetic nucleotide alignment of SNORD116 was based on original alignments from Kocher and Good (16).
RNA–RNA Interaction Prediction
IntaRNA version 2.4.0 (40–43) was used with default parameters to predict the interaction structures between SNORD116 and its targets. Specifically, the entire mRNA sequence of NHLH2 (Gene ID: 4808) was input for analysis. IntaRNA is an algorithm that computes an interaction structure with minimum free energy from two input sequences using dynamic programming routine. All secondary structures of single stranded RNA are predicted by ViennaRNA 2.0 (44).
Generation of Constructs
The Nhlh2-myc tagged construct with SV40 p(A) tail was a generous gift from Dr Thomas Braun, Max Planck Institute, Bad Nauheim, Germany. This construct contained in the pCS2-MT backbone was used to generate the Nhlh2-myc tag with a partial 3’UTR used in a previous study (13). The partial 3’UTR construct was then used for generating the full 3’UTR and the full 3’UTR containing the rs1051613841 SNV. Specifically, a separate vector containing the 1145 bp of extra 3’UTR cDNA was cloned using PCR amplification with PstI sites on the 5′ and 3′ ends of the PCR product, and subsequently cloned into a PstI site at the end of the partial 3’UTR. This vector containing the full 3’UTR was then used to create the SNV construct by site-directed mutagenesis using the Phusion Site-Directed Mutagenesis Kit (ThermoFisher # F541), and mutagenesis primers (Supplementary Material, Table S1). The myc-tags on these constructs allow for QPCR analysis of tagged RNA separate from endogenous RNA and allow for primary antibody detection for Western blotting.
The mouse Snord116 expression vector was a generous gift from Dr Stefan Stamm, University of Kentucky, Lexington, USA (20). This construct was used to generate a negative control vector for Snord116 expression by excising the Snord116 insert using MssI digestion (2722 bp), and ligation of the blunt ended fragment of interest (5035 bp). This left a backbone nearly identical to the pCDNA5/FRT/TO vector, but missing 102 bp of the multiple cloning site in between the CMV promoter and bGH p(A) terminator.
All constructs were sequenced and subjected to restriction enzyme digests followed by agarose gel electrophoresis for validation of cloning procedures.
Cell Culture and Transfections
The mouse hypothalamus neuron cell line, N29/2 (15), was maintained in T25 flasks in DMEM (4.5 g/L glucose, with 110 mg/L sodium pyruvate) (ThermoFisher # 11995065) and 10% fetal bovine serum (GE Healthcare #SH30396.03HI) with penicillin (50 units/mL)/streptomycin (50ug/mL) (Thermo # 15070063) at 4–6% CO2 and 37°C. Cells were detached from flasks using trypsin–EDTA (ThermoFisher # 25300054) and 6-well plates were seeded with 2 x 104—105 cells per well. Transfections were done 2–4 days post seeding using Opti-MEM® media (ThermoFisher # 31985070) and Lipofectamine® 3000 (ThermoFisher #L3000008) according to manufacturer’s instructions at 60–90% cell confluence. All transfections were performed between 3 p.m. and 6 p.m. 200 ng of each plasmid DNA was transfected for Leptin receptor, Stat3, +/-Snord116, Nhlh2-myctag for a total of 800 ng DNA per well of a 6-well plate. Leptin receptor and Stat3 expression vectors were included to ensure the expression of these key regulatory components of Nhlh2 and for consistency with previous studies of Nhlh2 (10,11,13,14).
RNA Purification
24 hours post transfection, cell culture media was removed, and cells were lysed with 1 mL of TRIzol® Reagent (ThermoFisher #15596018) directly in 6-well culture plates. TRIzol samples were frozen at −20°C in microfuge tubes until purified (1–14 days) using the TRIzol+Purelink RNA minikit (ThermoFisher #12183025) following manufacturer’s instructions for the TRIzol® Plus Total Transcriptome Isolation protocol. Purified RNA was then DNAse treated using TURBO DNA-free™ Kit (ThermoFisher #AM1907) according to manufacturer’s instructions, diluted to 60 ng/μL in nuclease-free water, and stored at −80°C.
Reverse-Transcriptase Quantitative PCR
For RT-QPCR, Power SYBR® Green RNA-to-CT™ 1-Step Kit (ThermoFisher #4389986) was used according to manufacturer’s instructions. 10uL reactions were performed using 150 nM final primer concentration. Primers were assessed for efficiency using a dilution series and fell within 90%–110% efficiency. 90 ng RNA was used per 10uL reaction. Two technical replicates were performed. Control reactions for each sample (No reverse-transcriptase and no-template controls) were used for quality control. 384-well plates were run on the ViiA 7 Real-Time PCR System (ThermoFisher) according to RT-QPCR mix instructions and thermocycling conditions were not modified from suggested protocol (1-step annealing/extension at 60°C). Quality control measures including melt-curve analysis, technical replicate analysis, etc. were analyzed by thermocycler software and by operator; any major errors were excluded from analysis when appropriate, and/or new samples and plates were run when appropriate. Candidate reference genes for ddCT analysis were analyzed for appropriate reference controls. Mouse beta-actin was used as reference gene control for steady-state experiments. Potential reference genes were evaluated for the Actinomycin D experiments, but none were satisfactory and thus no reference gene was used for RNA decay experiments. The CT values used reflect molarity of the target RNA, as total RNA remained the same for each reaction (90 ng). Relative change in molarity over the RNA decay time course is thereby relative to the average decay of total RNA (e.g. If a target gene’s relative quantification stays at 1.0 throughout the 180 minutes of Actinomycin D treatment, it decays at the same rate as the total RNA average decay rate).
Western Blot
Constructs were transfected into N29/2 cells, and 24 hours following transfection, cells were washed, lysed in RIPA buffer, scraped from the tissue culture plates, and processed for Western analysis using standard methods. Equal amounts of protein (7 μg/lane), as determined using Bradford Reagent (AMRESCO #E530-1 L) were separated on a 12% SDS polyacrylamide gel and transferred to PVDF membrane. Western blotting was performed using rabbit anti-myctag polyclonal primary antibody (Proteintech #16286–1-AP) with goat anti-rabbit horseradish peroxidase-linked antibody as a secondary antibody. Chemiluminescent signal was detected using the SuperSignal™ West Femto Maximum Sensitivity Substrate (ThermoFisher #34095).
RNA stability assay
24 hours post-transfection, cell culture media was refreshed with 5 μg/ml Actinomycin D (Sigma #SBR00013-1ML) for 45, 90, and 180 minutes before lysis with Trizol directly in 6-well plates. The 0-minute timepoints were refreshed with media containing no Actinomycin D for 50 minutes before lysis. Plates with Actinomycin D were concealed from light during the duration of Actinomycin D incubation. The full time-course experiment was replicated 3 independent times with 3 separate vials of frozen N29/2 cells.
Statistical Analysis
All RT-QPCR data were analyzed using Microsoft Excel 16 for Microsoft 365, IBM SPSS Statistics 26 for Windows, and GraphPad Prism 9.0.0. N = 6 for partial 3’UTR conditions; N = 9 for SV40, Full, and Full SNV conditions. Error bars indicate ±SD. The 2ddCT method of relative quantification was used. Statistical significance tests performed on respective ddCT values from which Relative Quantification values are derived. Significance is expressed at *p < 0.05, **p < 0.01, ***p < 0.001, NSp > 0.05.
For steady-state experiments, a two-way ANOVA with Bonferroni correction was used for relative expression of tagged RNA normalized within 3’UTR type. CT and dCT values used are provided in Supplementary Material, File S1.
For RNA decay analysis, CT values were normalized to the average CT of the 0-minute time point within a transfection, within 3’UTR condition, and within Snord116 condition [CT(0.minute.average)-CT(individual.value)]. 0-minute-normalized CT values (0dCT) were then pooled across transfection trial and assessed for normality. For Fig. 4A-D, 2-Way ANOVAs with Šídák's multiple comparisons test within timepoints were run on each 3’UTR excluding the 0-min timepoints. 20dCT are displayed in Fig. 4A-D. For Fig. 4E, +Snord116 CTs were normalized to average -Snord116 CTs within transfection trial, 3’UTR, and timepoints. [CT(-Snord116.average(TransfectionTrial’x’.Timepoint’y’.3’UTR’z’))-CT(individual.value(TransfectionTrial’x’.Timepoint’y’.3’UTR’z’))] These -Snord116 normalized CT values were analyzed using 2-Way ANOVA with Šídák's multiple comparisons test excluding 0-minute timepoints. 2(−snord116.normalized.CTs) are graphed. For decay rate, 0-minute-normalized CT values were used to calculate CT differences between one timepoint and the next successive timepoint and adjusted for 45-minute intervals. Individual CT values were normalized to the average CT value of the previous timepoint within a condition. These values were used to perform 2-Way ANOVAs with Šídák's multiple comparisons test within 3’UTR, shown in Fig. 4G and within 3’UTR/Snord116 for Supplementary Material, Fig. S4. Normalized CT values were adjusted to percentage values for graphs [(1-(2(decay.rate.normalized.CT)))*100]. CT values used are provided in Supplementary Material, File S2.
Acknowledgements
We would like to thank Joe Grieco (Virginia Tech) for his assistance in Western Blots and Dr Stefan Stamm (University of Kentucky), for the gift of the mouse Snord116 expression vector. The work was funded by a grant (#532939) from the Foundation for Prader-Willi Research (FPWR) to D.J.G., an internal grant from the Adaptive Brain and Behavior Destination Area at Virginia Tech to D.J.G, and a student research award from Sigma Xi to M.A.K.
Conflict of Interest Statement. D.J.G. receives funding from the BASF Chemical Company, Berkeley, CA for a PWS-related project, but none of these funds were used for the project reported herein.
