Evidence for an Integrated Gene Repression Mechanism Based on mRNA Isoform Toggling in Human Cells

We recently described an unconventional mode of gene regulation in budding yeast by which transcriptional and translational interference collaborate to down-regulate protein expression. Developmentally timed transcriptional interference inhibited production of a well translated mRNA isoform and resulted in the production of an mRNA isoform containing inhibitory upstream open reading frames (uORFs) that prevented translation of the main ORF. Transcriptional interference and uORF-based translational repression are established mechanisms outside of yeast, but whether this type of integrated regulation was conserved was unknown. Here we find that, indeed, a similar type of regulation occurs at the locus for the human oncogene MDM2. We observe evidence of transcriptional interference between the two MDM2 promoters, which produce a poorly translated distal promoter-derived uORF-containing mRNA isoform and a well-translated proximal promoter-derived transcript. Down-regulation of distal promoter activity markedly up-regulates proximal promoter-driven expression and results in local reduction of histone H3K36 trimethylation. Moreover, we observe that this transcript toggling between the two MDM2 isoforms naturally occurs during human embryonic stem cell differentiation programs.

translational repression is viewed as a switch-like mechanism, where the uORFs prevent translation of the downstream ORF under certain conditions, but this repression can be bypassed under other conditions.
Recently, we described a form of gene regulation that relies on the obligate coupling of transcriptional interference and uORFmediated translational repression to downregulate protein expression ( Figure 1A; Chen et al. 2017;Chia et al. 2017). During budding yeast meiosis, the amount of protein for the conserved kinetochore protein Ndc80 is determined by toggling between two functionally distinct NDC80 mRNA isoforms. The mRNA isoform produced from a distal promoter cannot be efficiently translated due to uORF translation that prevents ribosomes from reaching and translating the NDC80 ORF, and the transcription of this isoform interferes with the proximal NDC80 promoter activity in cis. In this manner, production of a 59-extended mRNA isoform inhibits Ndc80 protein production Chia et al. 2017). In the case of NDC80, the uORF-mediated repression appears to be constitutive, conditional only on the existence of the 59-extended transcript, rather than condition-specific like the best studied regulatory uORF cases [ (Mueller and Hinnebusch 1986;Vattem and Wek 2004), for example]. Thus, translational down-regulation by uORFs is simply a consequence of the choice of an upstream transcription start site. Re-expression of Ndc80 protein is achieved by a developmentally induced switch in promoter usage, from distal to proximal, during meiotic progression.
Improved methodology for genome-wide gene expression measurements have resulted in a more complete characterization of the set of transcripts expressed and regions translated than was previously possible. These studies have provided evidence for translation of thousands of uORFs and widespread existence of alternate transcript isoforms [(Ingolia et al. 2009;Ingolia et al. 2011;Brar et al. 2012;Sterne-Weiler et al. 2013;Dieudonné et al. 2015;Floor and Doudna 2016;Wang et al. 2016), for example], including during the yeast meiotic program. Despite their prevalence, the biological impact of both alternative transcript production and uORF translation to gene expression output in most of these newly identified cases has been unclear. Using analyses of parallel global mRNA, translation, and protein datasets, we found that many of the uORFs and alternate transcripts seen during the yeast meiotic program were indicative of the mode of coordinate regulation seen for NDC80, with at least 379 other genes showing protein levels that are driven by this type of integrated transcriptional and translational control over time through meiotic development (Cheng et al. 2018). It was also recently found that this mode of regulation functions to mediate down-regulation of proteins involved in aerobic respiration as a core part of the unfolded protein response [UPR; (Van Dalfsen et al., 2018)].
We noted that some of the hallmarks of the integrated mode of gene repression that were seen for NDC80 regulation, and thus used to annotate new cases in yeast, are also known to be common in mammals. For example, almost half of human genes show evidence of alternative promoter usage, resulting in transcript isoforms that differ in their 59 leader (Wang et al. 2016). Additionally, transcripts with extended 59 leaders that contain uORFs result, in some cases, in a poorly translated transcript compared to isoforms with shorter 59 leaders (Law et al. 2005;Floor and Doudna 2016). Alternative uORF-containing transcripts were also previously defined for several individual mammalian genes, including Mouse double-minute 2 homolog (MDM2), an oncogene and repressor of the tumor suppressor p53 (Arrick et al. 1994;Barak et al. 1994;Brown et al. 1999;Hughes and Brady 2005). The MDM2 transcript isoform produced from the distal P1 promoter contains a longer 59 leader than the one produced from the proximal P2 promoter ( Figure 1B) and this P1-derived MDM2 isoform specifically is poorly translated due to the presence of two uORFs in its extended 59 leader, as established by polysome analyses and reporter assays (Landers et al. 1997;Brown et al. 1999;Jin et al. 2003). These assays had established differential translation of the two individual MDM2 isoforms but had not investigated whether a relationship existed between them.
The integrated mode of gene repression seen for NDC80 in yeast relies on three key features [ Figure 1A, Chia et al. 2017)]. First, a developmentally regulated switch between alternative promoters for the same gene leads to the usage of different transcription start sites (TSSs). Second, due to upstream open reading frame (uORF)-mediated translational repression, the distal promoter-generated transcript is inefficiently translated. Third, transcription from the distal promoter represses the expression of the canonical mRNA isoform through transcriptional interference associated with co-transcriptional histone modifications. When all three features exist together, and only then, the activation of LUTI mRNA production causes an increase in the co-transcriptional H3K36me3 marks at the proximal gene promoter and transcriptional repression of the canonical mRNA isoform. Because LUTI mRNA is not well translated due to uORFs in its extended 59 leader and because the well-translated canonical mRNA is repressed, the net effect of LUTI mRNA production is the downregulation of translation from the LUTI target gene locus. Bottom panel: In the absence of LUTI expression, transcription from the canonical gene promoter occurs, leading to translation. B. Illustration of the MDM2 gene structure. MDM2 is transcribed from two different transcription start sites (TSS1 and TSS2) regulated by two different promoters (P1 and P2). Transcription from the distal TSS1 produces a 59-extended, uORF-containing transcript, which is poorly translated. Hereafter, the P1 promoter-driven transcript isoform is referred to as MDM2 LUTI , while the P2-driven isoform, transcribed from the proximal TSS2 is referred to as MDM2 PROX . p53 RE refers to the location of the p53 response element within the P2 promoter.
NDC80 transcription can result in a decrease in translation from this locus. We termed the distal promoter-generated transcript "NDC80 LUTI " for long undecoded transcript isoform, because, despite containing the entire NDC80 ORF, NDC80 LUTI is not efficiently translated into protein Chia et al. 2017). We noted that the regulation at the MDM2 locus shared key features with this LUTI-based mechanism, with the important exception that it was unknown whether there was transcriptional interference between the two MDM2 promoters.
It is well established that MDM2 P2 can be activated by p53 (Wu et al. 1993;Barak et al. 1994;Honda et al. 1997), but little is known about the factors that activate P1 and whether transcription from the P1 promoter affects P2 activity had not been explored. To our knowledge, this was true of all other individually and globally defined examples of alternative transcripts of differential translatability, as transcriptional interference and uORF-based translational control have been topics studied independently, and typically by different labs. Transcriptional interference is an essential feature of the LUTI mechanism, because it causes a toggle between the two transcript isoforms, which is necessary for effective gene expression repression. Transcriptional interference also enables efficient developmental regulation in the case of NDC80. Evidence of transcriptional interference from nearby transcription has previously been established in human cells (reviewed in (Shearwin et al. 2005;Palmer et al. 2011)), but these cases have not involved the production of ORF-encoding mRNAs. Nevertheless, we hypothesized that the type of integrated regulation that is common in yeast might occur at the MDM2 locus. Here, we report evidence that this is indeed the case, based on observed transcriptional interference between the two promoters at this locus. We also observe developmental regulation of the two MDM2 transcript isoforms and conclude that the type of integrated transcriptional and translational regulation that we described as a developmental gene regulatory strategy in yeast is also seen in human cells. These findings suggest value in considering translational and transcriptional regulation not only as independent steps, but rather, as potential collaborators in gene expression outcomes.

MATERIALS AND METHODS
Cell lines MCF-7-dCas9 and -dCas9-KRAB cells were cultivated at 37°with 5% CO 2 in high glucose Dulbecco's modified Eagle media (Gluta-Max, Gibco) supplemented with 10% FBS and 1% penicillinstreptomycin. K562-dCas9 cell lines were cultivated at 37°with 5% CO 2 in RPMI1640 media (Gibco) supplemented with 10% FBS and 10mM HEPES. hESCs (WIBR3 NIH#0079) were maintained in culture as described in (Lengner et al. 2010). The differentiation into definitive endoderm was performed using the STEMdiff Definitive Endoderm Kit (Stem Cell Technologies) following the manufacturer's instructions. MCF-7-dCas9 and -dCas9-KRAB cells were kindly provided by Howard Y. Chang (Stanford University). K562-dCas9 cells were kindly provided by Jonathan Weissman (University of California, San Francisco). Cell lines were authenticated by STR profiling and were tested to be negative for mycoplasma (MycoAlert Mycoplasma Detection Kit, Lonza).
RNA isolation, cDNA synthesis and quantitative polymerase chain reaction Total RNA from hESCs differentiating into definitive endoderm and hESCs differentiating into neurons, MCF-7-dCas9, MCF-7-dCas9-KRAB and K562-dCas9 cells was isolated using Trizol (Life Technologies) according to the manufacturer's instructions. Equal amounts of RNA were primed with random hexamers and reverse transcribed using SuperScript II Reverse Transcriptase (ThermoFisher) according to the manufacturer's instructions. RNA levels were quantified using SYBR Green/Rox (ThermoFisher) and the StepOnePlus Real-time PCR system (ThermoFisher). Primers used for RT-qPCR are in Table 1.
Samples for total RNA isolation from hESCs differentiating into neurons (hESC, NPC, neurons D14, neurons D50) were a gift from Helen Bateup [University of California, Berkeley].
The crosslinking reaction was stopped by adding 0.125M PBSglycine and cells were harvested by centrifugation. Cells were subsequently resuspended in ice-cold PBS containing 0.25mM PMSF and 10ug/ml aprotinin (Millipore) and pelleted by centrifugation. Chromatin immunoprecipitation of these pellets was performed as previously described (Testa et al. 2005) with minor modifications. Chromatin was sonicated 50 · 30 sec ON/30 sec OFF with a Bioruptor Pico (Diagenode) to obtain fragment sizes of 200 bp. The sheared samples were incubated in RIPA buffer II (10 mM Tris-Cl, pH 8.0, 1 mM EDTA, pH 8.0, 0.5 mM EGTA, 1% Triton X-100, 0.1% SDS, 0.1% Na-deoxycholate, 140 mM NaCl) containing protease inhibitors and PMSF, with Dynabeads Protein A (Invitrogen) for 2 h at 4°on rotation. After removal of Dynabeads Protein A, precleared lysates were immunoprecipitated overnight with 4 ug of rabbit anti-mouse IgG (Ab46540, Abcam) or anti-Histone H3 tri methyl lysine 36 (Ab9050, Abcam). Immunoprecipitates were recovered by incubation for 2 h at 4°with previously blocked Protein A Dynabeads in RIPA buffer II (1 mg/ml bovine serum albumin, protease inhibitors, and PMSF). Reverse crosslinked input DNA and immunoprecipitated DNA fragments were amplified with SYBR Green/Rox (ThermoFisher) and quantified with the StepOnePlus Real-time PCR system (ThermoFisher).

Data availability
All the reagents generated in this study are available upon request. Supplemental material available at Figshare: https://doi.org/10.25387/ g3.7670699.

RESULTS
A key prediction, if MDM2 were regulated by a LUTI-based mechanism, would be an inverse relationship between the two MDM2 transcript isoforms, such that reduction in transcription from P1 should lead to increased transcription from P2. We were able to reliably assay the levels of the P1-and P2-derived transcripts by reverse transcription followed by quantitative PCR (RT-qPCR) using isoform-specific primers, based on the differential splicing of the two transcripts that results in unique 59 sequences ( Figure S1). To directly test if P1-dependent transcription interferes with P2-dependent transcription, we inhibited transcription from P1 by using CRISPRi Qi et al. 2013). We first examined MCF-7 breast cancer cells stably encoding the catalytically dead Cas9 (dCas9), which is thought to interfere with PolII elongation when targeted near transcription start sites Qi et al. 2013). Expressing each of four different single guide RNAs (sgRNAs) targeting the P1 promoter region led to modest but significant increases, of up to twofold, in the P2-derived MDM2 PROX transcript levels, which was associated with the expected reduction of transcription from P1 (Figure 2A and Figure S2A). This result was notable, given that the maximal knockdown of P1 activity was only 40% relative to control cells in these lines (Figure 2A). We tried to enhance the P1 transcriptional knockdown by using CRISPRi in MCF-7 cells that carry a version of dCas9 fused to the Krüppel-associated box (KRAB) transcriptional repression domain . However, targeting of dCas9-KRAB to the P1 promoter led to repression of both the P1 and P2 promoters ( Figure S3). This finding is consistent with the long-range effect of the KRAB domain up to 1Kb (Gilbert et al. 2014), beyond the 845 bp distance between the P1 and P2 regulated transcription start sites. Therefore, we performed all subsequent experiments using cell lines that stably expressed dCas9 without the KRAB domain, as this firstgeneration version of CRISPRi allowed us to achieve promoter-specific repression.
We further probed the relationship between P1 and P2 by knockdown of the gene encoding p53 (TP53) in MCF-7 cells using CRISPRi.
Given that p53 is a well-characterized transcriptional activator for P2, it was not surprising that TP53 knockdown resulted in a significant, 43% reduction of the P2-derived MDM2 PROX transcript ( Figure 2B, left panel). However, additional CRISPRi knockdown of the P1-derived transcript, hereon referred to as MDM2 LUTI , still resulted in the transcriptional activation of P2, as evidenced by the 2-to threefold increase in MDM2 PROX levels in this background compared to the TP53 knockdown alone ( Figure 2B, right panel; Figure S2B; Figure S4). The observation that MDM2 LUTI repression leads to an increase in MDM2 PROX expression, even in cells with reduced p53 levels, suggests that transcription from P1 actively represses P2 activity and that relief of this repression alone can lead to increased expression of MDM2 PROX independent of p53, consistent with transcriptional interference at this locus. To test whether transcriptional interference based on MDM2 LUTI occurs in a different cell type, we performed similar experiments in K562, a TP53 2/2 myeloid leukemia cell line that routinely shows robust CRISPRi-based repression (Gilbert et al. 2014). Inhibition of MDM2 LUTI transcription in these cells resulted in a dramatic increase (up to 10-fold) in MDM2 PROX expression ( Figure 3A and Figure S2C). A range of MDM2 LUTI knockdown efficiencies were achieved in this cell line. Consequently, the degree of P1 down-regulation generally correlated with the degree of P2 activation ( Figure 3A), suggesting tunability of the transcriptional interference at this locus.
H3K36me3 is a co-transcriptionally established modification that occurs in regions downstream of active promoters (Xiao et al. 2003;Bannister et al. 2005;Mikkelsen et al. 2007), and in budding yeast is associated with a decrease in spurious transcription initiation from within transcribed genes (Li et al. 2003;Xiao et al. 2003;Carrozza et al. 2005;Keogh et al. 2005;Kim et al. 2016) and in noncoding RNA transcription-dependent repression of gene promoters (Kim et al. 2012;Van Werven et al. 2012). H3K36me3 is enriched at the proximal NDC80 promoter as a result of NDC80 LUTI transcription and is involved in the transcriptional interference seen at the NDC80 locus . In mammalian cells, H3K36me3 has been n implicated in silencing, including the repression of spurious intragenic transcription (Dhayalan et al. 2010;Xie et al. 2011;Wagner and Carpenter 2012;Baubec et al. 2015;Suzuki et al. 2017), but its involvement in promoter repression has been less clear. We found that down-regulation of MDM2 LUTI expression resulted in a greater than threefold decrease in the H3K36me3 signal over the P2 promoter ( Figure 3B, Figure S5). In contrast, H3K36me3 signal remained high within the MDM2 gene body, likely due to increased MDM2 PROX transcription under these conditions. These data are consistent with a  mechanism whereby MDM2 LUTI expression represses transcription from the P2 promoter through co-transcriptional histone modifications, and provide further support for a model in which the proximal MDM2 promoter is controlled by a similar mechanism to that defined in yeast.
In budding yeast, developmentally controlled switching between the LUTI and canonical mRNA isoforms occurs pervasively during meiotic differentiation (Cheng et al. 2018). To begin to test whether such transcript toggling naturally occurs in human cells, we used two different human Embryonic Stem Cell (hESC) differentiation models. In human hESCs, both MDM2 transcript isoforms were expressed ( Figure 4A). When these cells were induced to undergo neuronal differentiation, a transient switch in transcript isoform expression from MDM2 PROX to MDM2 LUTI was evident between hESCs and neuronal precursors (NPCs) ( Figure 4B). We also observed an anti-correlation between MDM2 PROX and MDM2 LUTI expression as hESCs differentiated into an endodermal fate, as determined by endoderm-specific markers ( Figure 4C, Figure S6). This inverse pattern of proximal and distal promoter usage seen during hESC differentiation suggests that the LUTIbased mechanism regulates MDM2 expression during normal cellular differentiation. Data were normalized relative to MDM2 LUTI or MDM2 PROX transcript abundance in hESCs. Error bars refer to the range measured for two biological replicates. C. RT-qPCR data showing the changes in the expression of MDM2 PROX and MDM2 LUTI in hESCs differentiating into endoderm. D1-D4 refers to days after transfer of the hESCs to endoderm differentiation medium. Data were normalized relative to MDM2 LUTI or MDM2 PROX transcript abundance in hESCs.

Figure 5
Model of the LUTI mRNA based mechanism of the MDM2 gene. The MDM2 gene has two promoters, P1 and P2. MDM2 PROX is regulated by P2 whereas MDM2 LUTI is regulated by P1. In comparison to MDM2 PROX , MDM2 LUTI is poorly translated because of the existence of two upstream open reading frames (uORFs) in its extended 59-leader. Top panel: When P1 promoter is active ("ON"), MDM2 LUTI transcription establishes H3K36 trimethylation at the downstream P2 promoter and causes repression of P2 ("OFF"). As a result, MDM2 LUTI becomes the predominant transcript product from the MDM2 locus. Bottom panel: When P1 promoter is "OFF", transcriptional repression of the downstream P2 promoter is relieved, culminating in the expression of MDM2 PROX . MDM2 PROX is efficiently translated, resulting in higher MDM2 translation.

DISCUSSION
We report here that an integrated transcriptional and translational regulatory strategy that relies on obligate coupling of transcriptional interference and uORF-based translational repression occurs in human cells. Based on the ubiquitous use of alternative promoters and uORF translation in humans (Ingolia et al. 2011;Floor and Doudna 2016;Wang et al. 2016;Tresenrider and Unal 2018), this type of regulation may control the expression of many mammalian genes. Both the type of transcriptional control described here and the uORF-based translational regulation already established for the distal promoter-derived MDM2 transcript are known modes of gene regulatory control. However, our study shows that transcriptional interference can be combined with the production of alternate transcripts of differential translatability in mammals, in a manner that resembles the LUTI-based regulation seen in yeast. These findings suggest value in revisiting the sequential model of gene expression control that dominates the interpretation of past and current data in favor of a more holistic view of different levels of gene expression control.
Canonical models to explain the prevalence of mammalian alternative promoter usage suggest that one promoter might serve as a "backup" or that the use of two promoters could simply allow activation by different transcription factors that are present in different cell types (Davuluri et al. 2008). However, in the case of MDM2, we argue that its two promoters are fundamentally different in function. The P1 promoter produces a poorly translated MDM2 LUTI transcript and the production of MDM2 LUTI from this promoter interferes with P2 activity in cis, reducing the transcription of the well-translated MDM2 PROX isoform. Therefore, P1-driven MDM2 LUTI mRNA production serves to downregulate MDM2 expression. Further, this repression occurs in a tunable manner ( Figure 5). A similar trend was also observed for the Adh gene in fruit fly, which is expressed from two distinct promoters (Jorgensen et al. 2018). Both findings are consistent with the notion that transcriptional interference can be used to tune gene expression, rather than acting as an on-off switch .
Contrary to traditional gene regulatory models, mRNA and protein abundances generally show poor correlations over developmental programs in genome-scale yeast and vertebrate studies (Peshkin et al. 2015;Cheng et al. 2018). In yeast, hundreds of such cases can be explained by LUTI-based regulation, whereby developmentally regulated transcript distal and proximal promoter usage drives final protein output levels (Cheng et al. 2018). In these conditions, we found that a single developmentally regulated transcription factor could drive distinct sets of canonical and LUTI targets, which resulted in coordinate up-and down-regulation of protein levels for the two target sets (Cheng et al. 2018). We propose that this mechanism may be advantageous for developmental programs, which are typically characterized by multiple transcription-factor driven switches in cell stage, because of the ability to temporally coordinate up-and down-regulation of gene sets at each transition. In principle, whether a gene is in one set or another could be defined simply by the position of the binding site for a given transcription factor relative to the ORF and uORF sequences Cheng et al. 2018;Otto and Brar 2018;Tresenrider and Unal 2018). The natural toggling between MDM2 isoforms during differentiation shown here suggests that the broad use of this type of regulation for developmental modulation of gene expression may be conserved.
MDM2 levels are elevated in a variety of cancers (reviewed in (Rayburn et al. 2005)) and this elevation has been attributed in some cases to an increase in translation of the pool of MDM2 transcripts, based on increased transcription from the P2 (Landers et al. 1997;Capoulade et al. 1998;Brown et al. 1999). Much research has focused on identifying alternate transcription factors that can activate P2-as it is clear that transcription can occur from this promoter in the absence of p53-and several have been found (Phelps et al. 2003;Zhang et al. 2012), but relatively little is known about P1 regulation. Our study argues that MDM2 expression levels could be modulated by changes in activation of P1 alone, suggesting a promising new general area for the development of gene regulatory tools that modulate P1 activity, and the activity of other yet-to-be-identified LUTI mRNA promoters, as a means to fine tune gene expression.