Premature polyadenylation of MAGI3 is associated with diminished N⁶-methyladenosine in its large internal exon

Abstract

In cancer, tumor suppressor genes (TSGs) are frequently truncated, causing their encoded products to be non-functional or dominant-negative. We previously showed that premature polyadenylation (pPA) of MAGI3 truncates the gene, switching its functional role from a TSG to a dominant-negative oncogene. Here we report that MAGI3 undergoes pPA at the intron immediately downstream of its large internal exon, which is normally highly modified by N⁶-methyladenosine (m⁶A). In breast cancer cells that upregulate MAGI3^pPA, m⁶A levels in the large internal exon of MAGI3 are significantly reduced compared to cells that do not express MAGI3^pPA. We further find that MAGI3^pPA transcripts are significantly depleted of m⁶A modifications, in contrast to highly m⁶A-modified full-length MAGI3 mRNA. Finally, we analyze public expression data and find that other TSGs, including LATS1 and BRCA1, also undergo intronic pPA following large internal exons, and that m⁶A levels in these exons are reduced in pPA-activated breast cancer cells relative to untransformed mammary cells. Our study suggests that m⁶A may play a role in regulating intronic pPA of MAGI3 and possibly other TSGs, warranting further investigation.

Introduction

Polyadenylation is an essential process controlling gene expression, yet how cancer cells deregulate this process to drive malignancy is only beginning to be appreciated. Polyadenylation requires cis-acting RNA sequence elements, most notably the AAUAAA sequence motif known as the poly(A) signal (PAS), which is recognized by trans-acting cleavage and polyadenylation proteins¹. The AAUAAA motif is fairly ubiquitous and, besides its presence in terminal exons, can frequently be found in introns. Typically, intronic PAS are prevented from triggering cleavage and polyadenylation by ribonucleoprotein complexes that bind to suppressive RNA sequence elements, such as U1 snRNA-binding sites^2,3. Despite these molecular safeguards, we previously showed that instances of intronic PAS activation do occur in cancer⁴. For example, in the MDA-MB-231 human breast cancer cell line and in primary human breast tumors, we found that oncogenic truncations of MAGI3 (MAGI3^pPA) are caused by premature polyadenylation (pPA) triggered by intronic PAS activation⁴. We also previously characterized MAGI3^pPA and found that this truncation interfered with the ability of full-length MAGI3 to bind and inactivate YAP, thereby promoting malignant transformation in breast cancer cells by functioning in a dominant-negative manner. However, the molecular mechanism that activates pPA of MAGI3 remains unknown since no cis-acting mutations were found in the gene⁴. In addition, it is unclear how and why pPA of MAGI3 occurs specifically in intron 10 but not in any of the other nineteen introns of the gene, most of which also harbor cryptic PAS.

In principle, imbalances in trans-acting factors could give rise to MAGI3^pPA. In practice however, because such factors participate widely in PAS recognition, changes in their activity impart widespread consequences on the polyadenylation of most multi-exon genes. The result is the production of many pPA-truncated products per gene; yet this is not observed for the 21-exon MAGI3⁴. Indeed, depletion of U1 snRNP, which protects pre-mRNAs from pPA, results in activation of multiple intronic PAS in the 5′ regions of most genes³, with a strong bias for PAS in intron 1⁵. These results cannot account for the focal pPA event occurring in intron 10 of MAGI3⁴, yet not in upstream introns that are more likely to be affected by trans-acting factors. Intrigued by the specific occurrence of pPA following exon 10 of MAGI3 but not following other exons of the gene, we hypothesized that novel cis-acting elements may mark and render this gene region, and possibly others like it, susceptible to focal pPA events.

Results

Intronic pPA of MAGI3 occurs following the gene’s large internal exon

To understand the focal nature of pPA in the MAGI3 gene, we first examined the structure of the entire gene. MAGI3 is a large gene comprised of 21 exons (Fig. 1A). As reported more extensively in our previous work⁴, breast cancer-associated pPA of MAGI3 occurs in intron 10, following exon 10 (Fig. 1A). This event leads to the expression of a truncated, dominantly-acting oncogene (Fig. 1B), which can be detected by both 3′ rapid amplification of cDNA ends (RACE) and immunoblotting in MDA-MB-231 human breast cancer cells but not in MCF10A non-transformed human mammary cells (Fig. 1C and D)⁴. Upon examining the gene structure, we noticed that the occurence of pPA in intron 10 might be particularly significant since the preceding exon is by far the largest in the gene at 606 nt. The size of exon 10 is additionally noteworthy because large internal exons are a rare class of exons in the genome, likely because efficient splicing favors exon sizes less than 200 nt⁶. Thus, we hypothesized that a molecular mechanism which normally limits the usage of the intronic PAS downstream of the large internal exon of MAGI3 may be deregulated in cancer.

Figure 1

Intronic pPA of MAGI3 occurs following the gene’s large internal exon. (A) Diagrams showing the exon/intron arrangement of the full length MAGI3 gene and its truncated variant, MAGI3^pPA. The large internal exon is colored blue. (B) Domains of the encoded gene products are shown for full-length MAGI3 and MAGI3^pPA. (C) Full-length MAGI3 and truncated MAGI3^pPA mRNA are detected in the MDA-MB-231 human breast cancer cell line but not the non-transformed MCF10A human mammary cell line by 3′ RACE. Amplification of GAPDH is included to show loading for 3′ RACE and approximate molecular mass markers are indicated in kb. (D) Full length MAGI3 and truncated MAGI3^pPA proteins are detected by immunoblotting. Immunoblot of β-actin is included to show loading, approximate molecular mass markers are indicated in kDa, and the relative levels of full-length and pPA-truncated MAGI3 proteins were normalized to β-actin levels.

N⁶-methyladenosine (m⁶A) is normally enriched in the large internal exon of MAGI3, but its levels are reduced in pPA-activated MDA-MB-231 cells

To begin to test our hypothesis, we asked whether molecular marks enriched in large internal exons might correlate with the expression of MAGI3^pPA. Interestingly, studies examining methylation of mRNA at N⁶-adenosine (N⁶-methyladenosine or m⁶A) on a transcriptome-wide scale have previously reported consistent enrichment of m⁶A in large internal exons as well as terminal exons across several human cell lines^7,8. While the functional significance of these modifications in large internal exons has remained unclear, m⁶A density in terminal exons has been found to correlate inversely with proximal PAS usage in 3′ UTR alternative polyadenylation⁹. These data raise the possibility that m⁶A may influence the usage of proximal downstream PAS.

Interrogating two transcriptome-wide m⁶A sequencing (m⁶A-Seq) datasets generated in the human hepatocellular carcinoma HepG2 and non-malignant human embryonic kidney HEK293T cell lines^7,8, we found strong enrichment of m⁶A peaks in the large internal exon of MAGI3 (Fig. 2A). Notably, the concordance between the m⁶A peaks found in HepG2 and HEK293T cells was very strong. By normalizing the number of m⁶A reads to exon length, we observed that the vast majority of m⁶A marks in the MAGI3 mRNA are contained in the large internal exon (Fig. 2A).

Figure 2

The large internal exon of MAGI3 is highly modified by m⁶A in HEK293T, HepG2 and MCF10A cells but shows diminished m⁶A levels in pPA-activated MDA-MB-231 cells. (A) Distribution of m⁶A-Seq peaks across the MAGI3 gene locus, based on analysis of previously published m⁶A-Seq data in HepG2 cells⁷. Peak number and positions in HepG2 cells were found to be highly concordant with those found in HEK293T cells by an independent m⁶A-Seq study⁸. Below, the normalized number of m⁶A-Seq reads mapping to each exon of MAGI3 is plotted. (B) Distribution of m⁶A-Seq peaks across the large internal exon of MAGI3, exon 10. The locations and sequences of putative m⁶A sites within the large internal exon are indicated. (C) m⁶A levels at the indicated m⁶A consensus sites of MAGI3, relative to a distal MAGI3 exonic segment (exons 1–2), as determined by m⁶A RIP-qPCR in MCF10A cells (n = 3 m⁶A RIP replicates). (D,E) Relative m⁶A levels at the indicated m⁶A consensus sites of MAGI3 large internal exons, as determined by m⁶A RIP-qPCR in MCF10A and MDA-MB-231 cells (n = 3 m⁶A RIP replicates). Data in (C–E) are presented as mean ± SEM. ***p ≤ 0.001 (two-tailed Student’s t-tests).

Previous work has identified the m⁶A consensus sequence RRACU, where R is either G or A^7,8. In the 606-nt large internal exon of MAGI3, we found only two RRACU sequences, each positioned at the center of the two observed m⁶A-Seq peaks (Fig. 2B). To validate m⁶A presence in the MAGI3 large internal exon, we used a m⁶A-specific antibody to perform RNA immunoprecipitation (RIP) on ~100-nt chemically fragmented, poly(A)-purified RNA from MCF10A mammary epithelial cells. Relative methylation levels of fragments containing m⁶A consensus sites in the large internal exon of MAGI3 were determined by real-time PCR (qPCR) using flanking primers. To confirm the specificity of m⁶A RIP-qPCR, we included as negative controls primers flanking exonic regions (exons 1–2) of MAGI3 located far from m⁶A consensus sites (distal mRNA segments). Indeed, after we performed m⁶A RIP-qPCR, immunoprecipitated mRNA fragments containing the m⁶A consensus sites of the MAGI3 large internal exon were detected at high levels, whereas distal mRNA fragments were hardly detected at all (Fig. 2C).

We next focused on validating that m⁶A modifications at the two identified sites in the large internal exon of MAGI3 functionally promotes interaction with known m⁶A-binding proteins. Thus, we synthesized two biotinylated RNA moieties spanning each site, one m⁶A-modified within the RRACU motif and the other unmodified. Following incubation with MCF10A nuclear lysates, we immunoprecipitated the synthesized RNA by streptavidin-bound beads and performed mass spectrometry analysis (RIP-MS) on the bound samples. This analysis yielded three proteins enriched in the m⁶A-modified RIP samples of each site, including the m⁶A-binding proteins YTHDF1 and YTHDF3 (Table 1), thereby demonstrating that m⁶A modification at either site of MAGI3 exon 10 functionally promotes interaction with experimentally validated m⁶A readers^10,11,12. Following confirmation that m⁶A modification of MAGI3 exon 10 is functionally significant, we asked whether m⁶A modification in this exon differed between MDA-MB-231 and MCF10A cells by performing additional m⁶A RIP-qPCR experiments. We found that the relative abundance of m⁶A at both sites in the large internal exon of MAGI3 was significantly reduced in MDA-MB-231 compared to MCF10A cells (Fig. 2D and E).

Table 1 Proteins interacting with m⁶A-modified MAGI3 exon 10 sites as identified by RIP-mass spectrometry.

pPA-truncated MAGI3 transcripts are largely depleted of m⁶A modifications

Having shown an overall reduction in large internal exon m⁶A modification for MAGI3 in pPA-activated cancer cells, we next endeavored to determine whether this overall depletion of m⁶A marks in the large internal exon is specific to pPA-truncated transcripts or whether it occurs indiscriminately between full-length and truncated isoforms. We hypothesized that if m⁶A levels do not contribute to the activation of pPA, then full-length and pPA-truncated MAGI3 transcripts will not differ significantly in methylation status. We modified the m⁶A RIP protocol used previously in order to test this null hypothesis by eliminating the chemical fragmentation step such that we could immunoprecipitate intact, poly(A)-purified RNA from MDA-MB-231 cells. In addition to the immunoprecipitated RNA, we also extracted mRNA from the unbound fraction. We subsequently performed 3′ RACE using MAGI3-specific forward primers and an oligo-d(T) reverse primer for each extracted fraction.

Strikingly, these experiments using m⁶A RIP-RACE revealed that pPA-truncated transcripts of MAGI3 were significantly enriched in the unmethylated fraction and depleted from the methylated fraction (Fig. 3A and B). In contrast, full-length MAGI3 transcripts were highly enriched in the methylated fraction, and only a minority was observed in the m⁶A-unbound fraction (Fig. 3A and B). As a control, we performed m⁶A RIP-RACE for GAPDH, which has no large internal exons and is not modified by m⁶A⁷. GAPDH transcripts were detected only in the unmethylated fraction, thus confirming the specificity of the m⁶A RIP-RACE (Fig. 3C). These data demonstrate that hypomethylation of N⁶-adenosine in the large internal exon of MAGI3 is significantly associated with pPA-truncated, oncogenic MAGI3 transcripts. Taken together, our data to this point suggest that depletion of m⁶A modifications from the large internal exon of MAGI3 may somehow bias the favorability of using the downstream cryptic PAS in intron 10 (Fig. 3D). However, the generality of this proposed model remains uncertain and requires an understanding of whether other tumor suppressor genes (TSGs) also show evidence of pPA events following m⁶A-depleted large internal exons like MAGI3.

Figure 3

The pPA-truncated MAGI3 isoform is predominantly unmodified by m⁶A. (A) Full-length and pPA-truncated MAGI3 mRNA isoforms from MDA-MB-231 cells, fractionated into m⁶A-bound and m⁶A-unbound pools and detected by 3′ RACE (m⁶A RIP-RACE). Products from nested 3′ RACE reactions performed on MDA-MB-231 input, m⁶A-bound and m⁶A-unbound samples were separated by agarose gel electrophoresis. pPA-truncated and full-length MAGI3 transcripts are indicated. (B) Ratios of full-length to pPA-truncated (FL:pPA) MAGI3 mRNA isoforms from MDA-MB-231 input, m⁶A-bound and m⁶A-unbound fractions as detected by m⁶A RIP-RACE and quantified by densitometry using ImageJ (n = 3 technical replicates of 3′ RACE per m⁶A RIP, 2 biological replicates of the m⁶A RIP procedure). Data are presented as mean ± SEM. **p ≤ 0.01, ***p ≤ 0.001 (two-tailed Student’s t-tests). (C) Full-length GAPDH mRNA from MDA-MB-231 cells, fractionated into m⁶A-bound and m⁶A-unbound pools and detected by m⁶A RIP-RACE. Products from nested 3′ RACE reactions performed on MDA-MB-231 input, m⁶A-bound and m⁶A-unbound samples were separated by agarose gel electrophoresis. (D) Model for m⁶A-mediated repression of the MAGI3 intronic PAS downstream of large internal exons. Methylation of m⁶A sites (green tick marks) in large internal exons represses cryptic intronic PAS usage in the downstream intron, favoring the generation of full-length transcripts (upper panel). Hypomethylation of m⁶A sites in large internal exons reduces the bias against downstream cryptic intronic PAS usage, leading to increased production of pPA-truncated transcripts (lower panel).

Evidence of pPA events following the large internal exons of additional tumor suppressor genes

To begin addressing these questions, we investigated whether other tumor suppressor genes (TSGs) also show evidence of pPA events following large internal exons like MAGI3. Using public mRNA isoform expression databases to survey fifty TSGs from the Cancer Gene Census¹³, we found that twenty of them harbor at least one large internal exon (defined as >500 nt) (Supplementary Table S1). Of these, seven TSGs (ATRX, BCOR, BRCA1, BRCA2, LATS1, MSH6 and RNF43) have previously annotated mRNA isoforms terminating in introns immediately following large internal exons (Table 2). As a caveat, we note that having identified truncations arising from pPA in these seven TSGs does not preclude the possibility that the other thirteen TSGs in the list might also undergo intronic pPA following large internal exons. These data suggest that pPA may act as a more common mechanism for truncating TSGs than previous appreciated.

Table 2 TSGs with large internal exons truncated by pPA in immediate downstream introns.

Among the seven TSGs showing evidence of pPA, the truncated LATS1 isoform is particularly similar to MAGI3^pPA since previous studies have suggested that truncation products of LATS1 act to functionally oppose its tumor suppressive function. In the LATS1 gene, pPA occurs at a cryptic PAS in intron 4, following the 1.5-kb exon 4 (Fig. 4A)¹⁴. This pPA-truncated transcript of LATS1 was identified in a candidate full-ORF cDNA library generated from a variety of cellular sources¹⁴, and has not been extensively studied since its initial annotation. Thus we performed 3′ RACE to validate its expression specifically in the “pPA-activated” MDA-MB-231 breast cancer cell line and non-transformed MCF10A cell line. Indeed by 3′ RACE, we observed an upregulation of the truncated LATS1 mRNA in MDA-MB-231 compared to MCF10A cells, apparently at the expense of full-length LATS1 levels (Fig. 4B). By immunoblotting with an antibody raised against the N-terminal region of LATS1, we also found upregulation of LATS1^pPA in MDA-MB-231 compared to MCF10A cells (Fig. 4C). Interestingly, the truncated LATS1 isoform (hereafter LATS1^pPA) lacks the kinase domain necessary for suppressing oncogenic YAP activity but retains the YAP-interacting domain (Fig. 4D). Overexpression of experimentally truncated LATS1 products of similar length to LATS1^pPA has been reported to dominantly interfere with LATS1-mediated regulation of the centrosome during mitosis, thus promoting mitotic delay and tetraploidy^15,16, and additionally bind to full-length LATS1 proteins in an inhibitory manner^17,18. Taken together, these data suggest that MDA-MB-231 breast cancer cells may have positively selected for the pPA-truncated product of LATS1 as a potentially oncogenic protein variant.

Figure 4

Intronic pPA events occur following the large internal exons of additional TSGs and correlate with reduced large internal exon m⁶A levels. (A) Diagrams showing the exon/intron arrangement of the full-length LATS1 gene and a truncated variant. The large internal exon is colored blue. (B) Detection of full-length LATS1 mRNA isoforms (lengths vary depending on 3′ UTR PAS selection) as well as a truncated LATS1^pPA mRNA isoform corresponding to intronic pPA downstream of exon 4 in the MDA-MB-231 and MCF10A cell lines by 3′ RACE. Approximate molecular mass markers are indicated in kb. (C) Immunoblot of LATS1 full-length and pPA-truncated products in the indicated cell lines. The membrane from Fig. 1D was stripped and re-probed with an anti-LATS1 antibody. Immunoblot of β-actin is included to show loading, approximate molecular mass markers are indicated in kDa, and the relative levels of full-length and pPA-truncated LATS1 proteins were normalized to β-actin levels. (D) Domains and functional regions of the encoded LATS1 full-length and pPA-truncated proteins. (E) Diagrams showing the exon/intron arrangement of the full-length BRCA1 gene and a truncated variant. The large internal exon is colored blue. (F) Immunoblots of BRCA1-p220 and BRCA1-IRIS proteins in the indicated cell lines. Immunoblot of β-actin is included to show loading, approximate molecular mass markers are indicated in kDa, and the relative levels of full-length and pPA-truncated BRCA1 proteins were normalized to β-actin levels. (G) Domains and functional regions of the encoded gene products, BRCA1-p220 and BRCA1-IRIS. (H) m⁶A levels at the m⁶A consensus sites of LATS1, relative to a distal LATS1 exonic segment (exons 2–3), as determined by m⁶A RIP-qPCR in MCF10A cells (n = 3 m⁶A RIP replicates). (I) m⁶A levels at the m⁶A consensus sites of BRCA1, relative to a distal BRCA1 exonic segment (exons 2–3), as determined by m⁶A RIP-qPCR in MCF10A cells (n = 3 m⁶A RIP replicates). (J–M) Relative m⁶A levels at the indicated m⁶A consensus sites of LATS1 (J,L) and BRCA1 (K,M) large internal exons, as determined by m⁶A RIP-qPCR in the indicated cell lines (n = 3 m⁶A RIP replicates). Data in (H–J) are presented as mean ± SEM. **p ≤ 0.01, ***p ≤ 0.001 (two-tailed Student’s t-tests).

We also looked at BRCA1 as another example of the seven TSGs showing evidence of pPA. Following the 3.4-kb exon 10, the BRCA1-IRIS isoform is prematurely polyadenylated downstream of a close variant of the canonical PAS (AGUAAA) in intron 10 (Fig. 4E)¹⁹. The expression of this truncated mRNA isoform has previously been extensively characterized by 3′ RACE, sequencing, RT-PCR and Northern blot analysis¹⁹. We immunoblotted MDA-MB-231 and MCF10A cell lysates with an antibody recognizing the N-terminal region of BRCA1 and observed that BRCA1-IRIS was present at higher levels in MDA-MB-231 versus MCF10A cells (Fig. 4F). BRCA1-IRIS lacks key functional regions, such as the BRCT domains and protein-interacting regions (Fig. 4G), and its expression has been previously reported to promote growth factor-independent cell proliferation, anchorage-independent colony formation, and subcutaneous tumor xenograft growth^19,20,21,22.

Reduced m⁶A modification of LATS1 and BRCA1 large internal exons in MDA-MB-231 cells

To investigate whether the large internal exons of TSGs are also typically enriched in m⁶A modifications, we again examined transcriptome-wide m⁶A sequencing (m⁶A-Seq) datasets^7,8. Consistent with the pattern of m⁶A modification for the large internal exon of MAGI3, we found enrichment of m⁶A peaks in the large internal exons of LATS1 and BRCA1, as well as other TSGs (Supplementary Fig. S1). It is worth noting that the complexity of m⁶A modification patterns increased with greater internal exon lengths, and the largest internal exons frequently exhibited multiple, strong m⁶A peaks with additional, weaker m⁶A peaks throughout. After identifying putative large internal exon m⁶A sites by finding the consensus sequence RRACU in the strongest m⁶A peak regions of LATS1 and BRCA1, we performed m⁶A RIP-qPCR in MCF10A cells to validate the presence of m⁶A modifications. We validated m⁶A modifications in the two strongest peaks of LATS1 exon 4, with the downstream site exhibiting the highest modification level (Fig. 4H). Meanwhile, for BRCA1 exon 10, we validated high levels of m⁶A modification in the most downstream site, but the upstream site showed much weaker enrichment by m⁶A RIP (Fig. 4I).

We subsequently asked whether MDA-MB-231 cells differ in the levels of m⁶A modification in TSG large internal exons compared to MCF10A cells. We found that the relative abundance of m⁶A at the strongest, most downstream sites in the large internal exons of LATS1 and BRCA1 was significantly reduced in MDA-MB-231 cells (Fig. 4J and K), accompanied by less dramatic reductions at weaker upstream m⁶A sites (Fig. 4L and M). These data suggest that like MAGI3, reduced m⁶A levels in the large internal exons of LATS1 and BRCA1 also correlate with intronic pPA following large internal exons.

Overall m⁶A levels and expression levels of m⁶A-modifying enzymes are comparable between MDA-MB-231 and MCF10A cells

Because we observed pPA-associated m⁶A hypomethylation in the large internal exon of MAGI3, as well as a general reductions in large internal exon m⁶A levels for BRCA1 and LATS1, we asked whether this phenomenon might be caused by an overall reduction in m⁶A levels transcriptome-wide in pPA-activated MDA-MB-231 cells compared to pPA-protected MCF10A cells. Thus we performed dot blot assays on purified poly(A) RNA from each cell line. These experiments showed that overall levels of m⁶A modification in the two cell lines are comparable (Supplementary Figure S2A and B). We further examined whether the expression levels of genes encoding known m⁶A methyltransferase components (writers) or demethylase proteins (erasers) differ dramatically between MDA-MB-231 and MCF10A cells. We therefore assessed the expression levels of m⁶A writers METTL3, METTL14 and WTAP^23,24,25,26, as well as the expression levels of the m⁶A erasers FTO and ALKBH5^27,28, in MDA-MB-231 and MCF10A cells by qPCR (Supplementary Figures S2C–G). Overall, we found that the expression levels of m⁶A-modifying enzymes were comparable between the two cell lines (Supplementary Figures S2C–G), with only slight differences observed. Moreover, when m⁶A writers or erasers were considered together as functional groups, we did not observe collective trends in one cell line versus the other. For instance, while METTL3 levels were slightly higher in MDA-MB-231 cells, the other two m⁶A methyltransferase components, METTL14 and WTAP, were expressed at slightly lower levels compared to MCF10A cells (Supplementary Figures S2C–E). Similarly, of the two m⁶A demethylases, FTO was expressed slightly more highly in MDA-MB-231 cells while ALKBH5 was expressed slightly more highly in MCF10A cells (Supplementary Figures S2F and G). Taken together with the results from dot blot assays, the overall levels of m⁶A modification and the expression levels of m⁶A-modifying enyzmes do not necessarily distinguish the pPA-activated cell line, MDA-MB-231 from the non-transformed MCF10A cell line.

Discussion

The molecular mechanism underlying cancer-associated, intronic premature polyadenylation of MAGI3 has remained unknown because no cis-acting genetic mutations were found in the gene, making it unclear how pPA of MAGI3 can specifically be activated in one intron but not in other introns that also harbor cryptic PAS⁴. In this study, we have identified N⁶-methyladenosine as a cis-acting epitranscriptomic mark associated with MAGI3 mRNA shortening. We have found that MAGI3 is affected by pPA at the intron immediately downstream of its single, large internal exon. The large internal exon of MAGI3 is by far the most highly m⁶A-modified exon in the gene, and we have shown by RIP-MS that the lack of m6A modification at the two m⁶A consensus sites in the exon diminishes the frequency of physical interactions between the mRNA and m⁶A-reading proteins. Furthermore, we have discovered that MAGI3^pPA transcripts are largely depleted of m⁶A modifications while full-length MAGI3 mRNA remains highly m⁶A-modified.

Since its discovery, the functional impact of high m⁶A levels in the large internal exons of genes has remained unclear⁷. By identifying m⁶A as a cis-acting epitranscriptomic mark associated with MAGI3 mRNA shortening, we have drawn an unexpected connection between large internal exon m⁶A modifications in MAGI3 and the expression of cancer-associated, pPA-truncated MAGI3 transcripts. How cancer cells modulate m⁶A levels in the MAGI3 large internal exon to trigger pPA, and how this modulation of levels impacts pPA of MAGI3 from a mechanistic standpoint, are new questions that require further investigation. Regarding the former, several m⁶A-modifying enzymes have been recently identified, and alterations in some of these components, especially the m⁶A demethylase FTO, have been observed to correlate with human cancer risk^29,30. For the latter, a bias against pPA of MAGI3 rendered by m⁶A modification could be achieved via changes to the secondary structure of large internal exonic regions of the mRNA thus preventing downstream PAS recognition, or by binding of a m⁶A-binding protein that acts in concert with other protein factors to prevent intronic PAS usage, or a combination of both mechanisms. Indeed, similar mechanistic concepts regarding the structural aspects of genes and m⁶A-mediated post-transcriptional gene regulation have recently been put forth for consideration as a new paradigm for the coordination of gene expression^31,32.

We have additionally analyzed publicly available mRNA expression data to report that intronic pPA-generated isoforms of other TSGs such as LATS1 and BRCA1 have been previously identified^14,19. These findings suggest that pPA may act as a more pervasive oncogenic mechanism for truncating TSGs with large internal exons than previously appreciated. Interestingly, we have also found that m⁶A levels in the large internal exons of LATS1 and BRCA1 are significantly lower in pPA-activated breast cancer cells relative to untransformed mammary cells. Taken together with the experiments showing that reduced m⁶A modification is associated with pPA-shortening of MAGI3, these data are conceptually consistent with those of a previous study showing that m⁶A density is inversely correlated with proximal PAS usage in terminal exons⁹. Thus, it is intriguing to speculate that m⁶A modification of large internal exons may play a role in regulating intronic pPA of TSGs beyond MAGI3, and additional studies of broader scope investigating the relationship between m⁶A levels in large internal exons and intronic pPA-mediated mRNA truncation for other TSGs are warranted.

Materials and Methods

Cell Lines and Tissue Culture

The cell lines used in this study were purchased from ATCC and grown as described previously⁴.

Immunoblotting and Dot Blot Assays

Cell lysis, SDS-PAGE and immunoblotting were performed as described previously⁴. For dot blot assays, poly(A) RNA was purified from total RNA using DynaBeads mRNA Purification Kit (ThermoFisher). Poly(A) RNA was serially diluted to 180 ng/µl, 45 ng/µl, 11.25 ng/µl. Each dilution was dotted (2.5 µl) on a BrightStar-Plus positively charged nylon membrane (Invitrogen) in duplicate. The poly(A) RNA was crosslinked to the membrane in a Stratalinker 2400 Crosslinker twice (1,200 µl joules) and the membrane was washed for 5 minutes in wash buffer (Phosphate Buffered Saline, 0.02% Tween-20) before blocking for 1 hr (Phosphate Buffered Saline, 5% Milk, 0.02% Tween-20). The membrane was incubated overnight at 4 °C in polyclonal rabbit anti-m⁶A antibody (2 µg/ml) diluted in blocking buffer. Treatment with secondary antibody was performed according to standard immunoblotting procedures and m⁶A detection was visualized using enhanced chemiluminescence. Levels of m⁶A were quantified by measuring density of dots using Fiji ImageJ. Antibodies used are: β-actin (Abcam ab6276); BRCA1 (ThermoFisher MA1-23160); BRCA1(Santa Cruz Biotechnology sc-642); LATS1 Goat Santa Cruz Biotechnology sc-9388; m6A for RIP (New England Biolabs E1610); m⁶A (Synaptic Systems 202–003); MAGI3 (Novus Biologicals NBP2-17210).

RNA Preparation, m⁶A RIP-qPCR and m⁶A RIP-RACE

Total RNA was extracted from MCF10A and MDA-MB-231 cell pellets using the RNeasy Maxi Kit (Qiagen). Poly(A) RNA was purified from total RNA using the Oligotex Midi Kit (Qiagen). For m⁶A RIP-qPCR, RNA samples were chemically fragmented into ~100-nt length fragments by a 5 min incubation at 95 °C in NEBNext RNA fragmentation buffer from New England Biolabs (40 mM Tris-OAc, 100 mM KOAc, 30 mM Mg(OAc)₂, pH 8.3). The fragmentation reaction was stopped with 50 mM EDTA, and one round of ethanol precipitation was performed to purify the fragmented poly(A) RNA. 3 µg fragmented poly(A) RNA was incubated for 1 hr at 4 °C with 1 µl EpiMark anti-m⁶A antibody (New England Biolabs) pre-bound to pre-washed Protein G magnetic beads in reaction buffer (150 mM NaCl, 10 mM Tris-HCl, pH 7.5, 0.1% NP-40). m⁶A-bound complexes were then washed twice in reaction buffer, followed by two washes in low salt reaction buffer (50 mM NaCl, 10 mM Tris-HCl, pH 7.5, 0.1% NP-40) and two washes in high salt reaction buffer (500 mM NaCl, 10 mM Tris-HCl, pH 7.5, 0.1% NP-40). Immunoprecipitated RNA was eluted in 30 µl Buffer RLT (Qiagen), then cleaned and concentrated using Dynabeads MyOne Silane (ThermoFisher) followed by ethanol washes. Bound RNA was eluted in 20 µl nuclease-free water and used for first-strand cDNA synthesis as described previously⁴. cDNA was also synthesized from total RNA, representing the input for m⁶A RIP. Samples were prepared for qPCR using isoform-specific or exon-specific primers. qPCR was performed in triplicate for each sample-target combination as described previously⁴. For determining gene expression, mRNA abundance was normalized to GAPDH. For m⁶A RIP samples, m⁶A levels of each target were normalized to overall expression levels of the target as determined by the same primer pair from total RNA. Targets spanning exons of the same gene but located far from the m⁶A sites within the large internal exons (distal mRNA segments) were also assayed. Primer sequences used for qPCR are: Forward Primer (F: 5′ to 3′), Reverse Primer (R: 5′ to 3′): BRCA1 exon 10 site 1F: TGAGTGGTTTTCCAGAAGTGA R: TCCCCATCATGTGAGTCATC; BRCA1 exon 10 site 2F: TCTCAGTTCAGAGGCAACGAR: TGGGTTTTGTAAAAGTCCATGTT; BRCA1-distal (exons 2–3) F: CGCGTTGAAGAAGTACAAAATG R:CAGGTTCCTTGATCAACTCCA; LATS1 exon 4 site 1F: GACCTGGAATGCAGAATGGT R: GCAGGGACAACATTTTGGTG; LATS1 exon 4 site 2F: GCCTGTGAAAAGTATGCGTGT R: GGCTGTGGTATCCAAGAAGG; LATS1-distal (exons 2–3) F: ACTTGCAAGCTGCTGGATTT R: TGTTGTTAGTTTTCTGAAGAGCTTG; MAGI3 exon 10 site 1F: TGGACAGTCATTAACCAAGGGA R: GCTCCAGAACCATTGCTCCT; MAGI3 exon 10 site2 F: CATCGTCAGGCAGCTCCC R: TGCAAACCCAAACCCTTTAGG; MAGI3-distal (exons 1–2) F: CGTCTCAAGACTGTGAAACCA R: GACTTAGGTAATGCCGCAATC; GAPDH F: CCATGGGGAAGGTGAAGGTC R: TAAAAGCAGCCCTGGTGACC; ALKBH5 F: TTCAAGCCTATTCGGGTGTC R: GGCCGTATGCAGTGAGTGAT; FTO F: AATCTGGTGGACAGGTCAGC R: TGCCTTCGAGATGAGAGTCA; METTL3 F: CCCACTGATGCTGTGTCCAT R: CTGCAGGAGGCTTTCTACCC; METTL14 F: TCCAAAGGCTGTCTTTCAGAGA R: GAAGTCCCCGTCTGTGCTAC; WTAP F: ACAAGCTTTGGAGGGCAAGT R: GATGTTTTCCCTGCGTGCAG.

For m⁶A RIP-RACE, RNA samples were not subjected to the fragmentation step and used directly for m⁶A-RIP. 3 µg unfragmented poly(A) RNA was incubated for 1 hr at 4 °C with 1 µl EpiMark anti-m⁶A antibody (New England Biolabs) pre-bound to pre-washed Protein G magnetic beads in reaction buffer (150 mM NaCl, 10 mM Tris-HCl, pH 7.5, 0.1% NP-40). Following this binding step, m⁶A-bound RNA (beads) and m⁶A-unbound RNA (supernatant) were reserved. The m⁶A-bound fraction was washed twice in reaction buffer, twice in low salt reaction buffer and twice more in high salt reaction buffer. Immunoprecipitated RNA was eluted in 30 µl Buffer RLT (Qiagen). The eluted m⁶A-bound RNA and the reserved m⁶A-unbound RNA were cleaned and concentrated using Dynabeads MyOne Silane (ThermoFisher) followed by ethanol washes. The bound and unbound RNA fractions were then eluted in 20 µl nuclease-free water, and 3′ RACE was performed as described previously⁴. MAGI3 and GAPDH gene-specific forward primer sequences used for 3′ RACE are: GAPDH-primary CCATGGGGAAGGTGAAGGTC;GAPDH-nested GATTTGGTCGTATTGGGCGC; MAGI3-primary CTGTGTCCTCGGTCACACTC; MAGI3-nestedGTTGCTGCTACCCCTGTCAT.

RIP-MS Analysis

Nuclear MCF10A lysates were obtained using the NE-PER kit (ThermoFisher) supplemented with protease inhibitors (Roche) and phosphatase inhibitors (Sigma), then precleared by incubating with streptavidin-conjugated magnetic beads (New England Biolabs) for 1 hr at 4 °C. 5′-biotin-labeled RNA oligonucleotides (42-nt in length with the RRACU m⁶A consensus motif in the center) were synthesized (Dharmacon). Two RNA oligonucleotide versions were synthesized for each MAGI3 exon 10 m⁶A site, differing only in their m⁶A modification status. Precleared MCF10A nuclear lysates were incubated with 2 µg of the RNA oligonucleotides supplemented with 0.4 units/µl RNasin (Promega) for 1 hr at 4 °C. The RNA-nuclear lysate mixture was subsequently added to streptavidin-conjugated magnetic beads pre-blocked with 1% BSA and 50 µg/ml yeast tRNA (ThermoFisher) for 1 hr at 4 °C. Immunoprecipitated complexes were washed in Tris-HCl buffer (20 mM Tris-HCl, pH 7.5), and bound proteins were eluted by boiling in SDS loading buffer for 5 min. Protein samples were separated by SDS-PAGE according to standard procedures, fixed in the gel, stained with a 0.3% Coomassie Blue R250 solution, then destained overnight. Gel slices were digested with trypsin and analyzed by liquid chromatography-tandem mass spectrometry (Taplin Mass Spectrometry Facility, Harvard Medical School). The accepted list of interacting proteins was obtained by filtering out common cytoplasmic protein contaminants and setting stringency thresholds of six or greater peptides identified in m⁶A-modified RIP samples and three or fewer peptides identified in m⁶A-unmodified RIP samples. The modified RNA oligonucleotide sequences used for RIP-mass spectrometry are: MAGI3 site 1 m⁶A-modified Bi-gacagucauuaaccaagggagag(m6A)cuugcaugaauccucagg; MAGI3 site 1 m⁶A-unmodified Bi-gacagucauuaaccaagggagagacuugcaugaauccucagg; MAGI3 site 2 m⁶A-modified Bi-ucgucaggcagcucccagccuga(m6A)cuagugacuaucccuuug; MAGI3 site 2 m⁶A-unmodified Bi-ucgucaggcagcucccagccugaacuagugacuaucccuuug.

Bioinformatic Analysis of m⁶A-Seq Data and Identification of Putative m⁶A Sites

Sequence data were downloaded from the Gene Expression Omnibus (GEO). The identifier for the GEO dataset is GSE37005⁷. Alignment data was obtained by following a previously published protocol for m⁶A-Seq analysis³³, converted to bigWig format normalized per total filtered reads and loaded to the UCSC genome browser for downstream analyses. To identify putative m⁶A sites, the locations of RRACU motifs, where R is either G or A, were cross referenced with peak locations along each exon. For the 606-nt MAGI3 exon 10, only two sequences matching the RRACU motif were found, and their locations corresponded to the approximate center of the m⁶A peaks from m⁶A-Seq. For LATS 1 exon 4 and BRCA1 exon 10, the pattern of m⁶A peak signals was considerably more complex. This was due to the exon lengths and increased frequency of RRACU sequences. LATS1 exon 4 had nine RRACU sequences across 1.5-kb, and BRCA1 exon 10 had 24 RRACU sequences across 3.4-kb. m⁶A-Seq data showed that each exon had two highly modified sites (strong peaks). Besides the two strongest peaks, LATS1 exon 4 had three moderate-to-high signal peaks and two weak signal peaks, while BRCA1 exon 10 had four moderate signal peaks and nine weak signal peaks. The two strongest peaks within each exon were chosen for validation as weaker peaks were likely to represent low stoichiometry m⁶A modifications that would be difficult to distinguish from background noise in m⁶A RIP-qPCR.

Statistical Analysis

Data were analyzed and compared between groups using two-tailed Student’s t-tests. A p < 0.05 was considered statistically significant.