 Hi, so I'm going to talk about a potentially heretical topic to this audience, which is post-transcriptional regulation at the level of translational control in my post-doctoral work with Jennifer Doudna at UC Berkeley. So control of protein translation broadly affects expression of genes in circumstances where cells need acute spatial temporal control over regulation, such as this example from Maria Barna's group, where deletion of a small RNA element causes mice to fail to develop their bottom rib. In neurons, there's extensive translational control, where RNA granules are shipped out to dendritic processes to locally activate translation. And in cancer, translation is broadly dysregulated, such as this leukemia model, where introduction of either of two translation factors causes a more severe leukemic phenotype than knockdown of the tumor suppressor P10. The focus of my post-doc has been trying to discriminate between two models of translational control driven by RNA binding proteins and translational control driven by alternative transcript isoforms, and that's actually what I'm going to talk about today. So as you all know, the human transcriptome is very complicated, where according to the ensemble annotations, the median human gene has about five annotated transcript isoforms. And as you probably know, a common method to measure translation across the genome is ribosome profiling, where you digest messenger RNA down to a small fragment that's protected by the ribosome. However, when you do that, you lose the transcript context that the ribosome protected fragment was derived from. So I wanted to come up with an alternative approach to actually measure transcript-specific translation across the genome. I'm not going to go into this in detail because this talk is very short, and I have a poster on it over there. But in some, you freeze ribosomes in place with cyclohexamide, you then take the cytoplasmic fraction that you separate by size on a sucrose gradient, which is a technique that's been around for 60 years to resolve ribosomal protein complexes and their associated mRNAs. You can then sequence the mRNAs that are bound to different numbers of ribosomes and extract transcript isoforms that are present in each of those fractions, and therefore know how many ribosomes were on each individual transcript isoform. And I thought this would be particularly interesting for this audience because it gives you a plot that looks like this, where in HAC 293 cells, you have about 62,000 transcript isoforms, which for each one, you now know how many ribosomes are actually bound to. This means that for a given RNA sequencing dataset, you can actually use this as a filter to apply translational information onto that RNA-seq data. I'm not going to go through this at all, but suffice it to say, you can look at a large number of features that are present in different transcript isoforms. The one that jumped out from HAC 293 cells was actually the length and the content of the 3-prime UTR, strongly negatively controls translation. And again, this is on my poster if you'd like to see it in more detail. What I wanted to focus on today, these pioneering studies that I published earlier this year were in HAC 293 cells, and I wanted to move on to a more relevant cell type, which is human ES cells, and isogenic differentiated neuropigender cells and neurons in vitro. So I've now performed, at this point, in collaboration with Dirk Hockemire and Helen Beta at Berkeley, so-called TRIP-seq, which is measuring transcript specific translation, as well as ribosome profiling and RNA-seq in these three different cell types. And I'm just going to show you a couple of highlights of that today. So first of all, you can extract immediately very interesting things. So for example, in human ES cells, there are two abundant isoforms of the nanogene, which in the cytoplasm are roughly equally abundant. But if you look in polyzones, so this is the monosome. This is aggregated low polyzone fractions, so something that's poorly translated. This is aggregated high polyzones, so something that's well translated. You see, this black isoform is pretty flat across polyzones, while this blue isoform is strongly enriched in the polyzonal fractions, suggesting that despite the fact that these two isoforms have relatively equal cytoplasmic abundance, the protein that's produced from these two is radically different. And I should say that for these two particular isoforms, the coding sequence is almost identical. Another thing that I thought that would be interesting for this audience is if you cluster all of these samples, we're on the y-axis here, you have 30 different samples from the cytoplasm and nucleus, as well as the polyzonal fractions. What jumps out at the bottom here is actually total RNA. So this is all of the ES cells, neuropigenerous cells and neurons, all of your total cytoplasmic or nuclear RNA, all clusters down at the bottom. What you see on the top is actually separation of translated RNA. So this is the human ES cell, polyzonal RNA, and this is the neuron in neuropigenerous cell, polyzonal RNA, which suggests that translated or ribosome-associated RNA might actually be a better way to stratify cell types than looking at total RNA. And lastly, I mentioned in 293 cells I saw, and there might be one more thing after this. In 293 cells I saw a strong effect from 3-prime UTRs, and now I can do the same analysis where in ES cells, neuropigenerous cells, and neurons, I actually cluster transcripts, compare two transcript isoforms that are derived from the same gene and see what features are different between those isoforms. And in 293 cells, there was a strong effect from 3-prime UTRs, but in these cell types, I actually see that there's a cell type specific effect of 3-prime UTRs, where the length of the 3-prime UTR in human embryonic stem cells has essentially no effect on translation, while neuropigenerous cells in neurons, as you've differentiated along this neuronal trajectory, 3-prime UTRs are increasingly more powerful at regulating translation. Okay, and the last thing that I wanted to come to this meeting, because of what I was very excited about, is the new RNA binding protein encode data that's coming out, for example, what Eric talked about earlier today. And so what you can do now is you can say, for two transcript isoforms of the same gene, what RNA binding proteins are different between these isoforms? And so I did this for all roughly 80 RNA binding proteins that were in K562 cells, and data that I'm not showing you, translational control by isoforms is often similar between at least cancer cell types. So I picked a couple of interesting examples to show you here. This one is the FMRP, or Fragilex mental retardation protein, where it's known to stall translation of elongating ribosomes. And you can actually see, if you look in a region-specific manner for transcript isoforms that are bound by more FMRP, if it binds inside of the orph, those transcripts are associated with high polyzones. So this is probably something that's elongation-stalled, okay, whereas it binds in the 5-prime UTR, it has a smaller effect, and there are actually elongating ribosomes in annotated 5-prime UTRs in upstream open reading frames, whereas if it binds in the 3-prime UTR, it has essentially no effect. So this means two things. First of all, you can identify which transcripts are controlled by FMRP in cells, and second, you can filter those that are probably elongation-stalled by using this data. The second one that I wanted to highlight was the spicing factor SF3B4, where if you have binding sites for SF3B4, more binding sites in an open reading frame that's associated with better translation. If you have more binding sites for SF3B4 in the 3-prime UTR, that's associated with worse translation or ribosome occupancy. And what I think is going on in this case, or one possibility, is that exons inside of an open reading frame are known to promote translation both through nuclear export and actively recruiting translation factors, whereas exons that are in 3-prime UTRs are usually downstream of the stop codon, which would actually activate nonsense media to decay. So I'm very excited by the possibility of using ENCODE RNA binding protein data to add, again, another layer of interpretation on top of, in this case, transcript-specific translation. And if you look at the metagen plot of SF3B4, it does look like a splicing factor, so it binds mostly at intronaxon junctions. So I don't think that it's binding to mature transcripts. And I'd like to thank many people at Berkeley, including, in particular, for this project, Dirk Hockmeier and Helen Bade up, as well as members of the down the lab and funding sources. And thank you. We have time for one question. Thanks for the time management. One question? So there will be a poster session tomorrow as well, so all the Latin talk speakers will be there for the whole session. So we can ask them more questions.