Scientist Stories: Peter Schultz, Playing with the Molecules of Life

It is my great pleasure to introduce our first speaker, Pete Schultz. Pete's been a CEO, I think, and a professor at Scripps. Before that, he was at the Genomics Institute of Novartis Research Foundation, and oh my, many things before that as well, including at Berkeley, etc. And of course trained at Caltech, very much a California product. When I think of Pete Schultz, I can't help but always think of unusual amino acids, amino acids that are backwards in shape. More recently, he's actually gone on to do such things with DNA bases, showing us that, uh, what should I say, unphysical science could be even more exciting than the natural sciences. So Pete, please talk to us, and it's yours.

Great. Thanks a lot, Dick. Let me see if I can do this. Can everybody see that?

Yes.

Great. Great. Thanks, Dick. And I want to start out by congratulating Bob on a really well-deserved award and many more to come, Bob. And what I thought I'd do is tell you about our work in synthesis. We really do synthesis at the interface of chemistry and biology, using both the tools of chemistry and biology to create macromolecules and even complex molecular systems with novel properties that you might not be able to achieve by either chemical or biological routes alone. And today I thought I'd give you two perspectives. One, a forward, using synthesis to take a forward look into evolution, and that's our work with an expanded genetic code, or in other words, what would we look like if God had worked on the seventh day? And the second is a backward look into evolution and using synthetic biology to look at the putative transition from the RNA to the DNA world. With a chimeric DNA-RNA genome. And on another related topic, the evolution from prokaryotes to eukaryotes and where did mitochondria come from. So let's start out with a forward look and expanding the genetic code. With the exception of just a handful of proteins, all forms of life on this planet share the same 20 amino acid code. And as a chemist, you ask why these 20, you know, where's the ketone group? It's the most important group in organic synthesis. It's lacking. And why 20, and why not 21 or 25? And if you find an organism with a 21 amino— 25 amino acid code, will it have an evolutionary advantage? So that's the question we wanted to ask. How do we do that? Well, nature's invested, you know, a billion years in and the genetic code and the translational machinery. So we, and we know the structures of basically every molecule involved in this, you know, 2 to 3 angstrom resolution. So we thought we could simply reprogram the code. And so what do you need to do that? You need a blank codon. You can use one of the nonsense codons. There are 3 stops. You stop translation once, so you have 2 degenerate codons there. You can go to a 4-base code. You can use a codon-deleted genome. You then have to build the tRNA, the delivery vehicle that recognizes your unique codon, and it's orthogonal to all the other tRNAs in, in the host, i.e., it's not recognized by the aminoacyl tRNA synthetases. And then you have to evolve the aminoacyl tRNA synthetases to uniquely recognize your new tRNA and none of the other tRNAs in the host cell. Then you have to convince that synthetase to uniquely charge your tRNA with the 21st amino acid and no other. And finally, get the amino acid into the host or biosynthesize it in the host. And we managed to do this using a combination of structure-based design and in vitro evolution tools. That's been reviewed, but, you know, to date, I think roughly 200 unnatural amino acids have been added to the code, and high yields, up to 8 grams per liter on industrial scale, in a variety of cell types and organisms. So what can you add? Well, if you're interested in protein structure function, fluorophores, NMR probes, if you're interested in building new sites, catalytic sites, metal-binding amino acids, cell biological tools, amino acids you can turn on with light, photocrosslinkers to map the interactomes, and also amino acids with unique chemical reactivity. And there's a lot of interest in the latter to site-specifically attach biophysical probes, drugs, and oligonucleotides even to proteins. And one example is in the formation of antibody drug conjugates. This is the old idea of Ehrlich's magic bullet, that you put a drug onto an antibody that's tumor-specific and you make it increase its efficacy and safety. The problem is, is the way we modify proteins in general is with electrophilic chemistry that targets cysteines and lysines, and there are a lot of cysteines and lysines in antibodies, so you can't really do selective chemistry. What we did was genetically encode a ketoamino acid. A keto group will react— it's bio-orthogonal— it reacts with alkoxy amines to form an oxime linkage, which is really stable. So it's a very simple orthogonal chemistry to modify proteins. And by genetically encoding that amino acid, you can actually put it in at specific sites and specific stoichiometries, and in essence do protein medicinal chemistry and do an SAR study to optimize the pharmacology and safety. And when we did that, we made an ADC that was selectively modified at alanine 121 on the heavy chain. And it retained wild-like, a wild-type-like half-life, and had really excellent efficacy. And it just completed a Phase 2 clinical trial at Ambrix, and going in with pretty remarkable results, suggesting that the ability to really optimize geometry and stoichiometry is important. If you're interested in cell biology, there's huge interest in post-translational modifications of histones, of signaling proteins, and the like. It's hard to study specific post-translational modifications because in general we don't have the selective tools to make them, except in smaller proteins where you can do semi-synthesis. So the answer to that is simply, for instance, if you want to look at phosphotyrosine genetically encoded, phosphotyrosine or phosphonotyrosine, a stable analog. It's hard to get phosphonotyrosine into the cell, but you can do it with a simple dipeptide, so we did that, and we evolved a synthase that incorporated that fairly well in bacteria, and we solved the crystal structure of the aminoacyl-tRNA synthase, and I show it here just to point out that there are very significant active site conformational changes that occurred in evolving from a tyrosyl site to a phosphono— phosphono-tyrosyl site. And it just points out that the remarkable plasticity of this class of enzymes that allows you to genetically encode so many diverse structures. Now another interesting use of an expanded genetic code is to genetically encode photocrosslinkers to probe protein interactomes. A lot of times when you probe protein-protein interactions or nucleic acid interactions, it's with an immunoprecipitation pulldown followed by mass spec. And unfortunately, a lot of the interactions you're looking at are either transient or they're not stable to those pulldown conditions. So what we did to overcome that is genetically encode a whole series of photocrosslinkers, for instance, the area with it shown here. And to exemplify this, let me show you what we did with SEPs. SEPs are short orphan coded peptides. These are short open reading frame peptides that really weren't annotated in the first annotation of the genome because of their small size and in some cases unique start sites. So how do we figure out what these SEPs do? Well, one way is to cross-link the SEPs to their interacting partners, and maybe you'll get a clue. Blue. So what we did was flag tag the CEPs at either the N or C terminus and then put a photocrosslinker throughout the sequence in independent experiments, transect that into cells, photocrosslink, and then pull down with a covalently fixed protein-protein interaction. And so with Alan Saglotelli and his self, we applied this to 24 CEPs that conserved across mammalian mammals. And as a first example, we looked at a known CEP called MRI2, which is associated with DNA repair. So when we did this labeling experiment in cells, we pulled down the known interacting partners, but we also pulled down unknown interacting partners that may suggest this CEP also has a role in DNA import and export. From the nucleus. But then we looked at these 24 SEPs in parallel, and you can see in 8 cases we identified the interactome. And so if you look at SEP10, it's pretty clear it does something with transcription. It binds to histone 2B. So we confirmed that by Western blot, and then we actually measured the binding affinity. It's 25 nanomolar. We showed SEP10 localizes to chromatin. We would ChIP-seq profiling, we showed it occupies almost 1,800 transcriptional start sites, and it modifies the expression of genes in a Sept10-dependent manner. So it's clear this micropeptide is a transcriptional regulator. Its exact function we're still working on. And there's another interest now in creating organisms that conditionally depend on a small molecule for growth, for example, a non-canonical amino acid, and that's to make live vaccines in biocontainment. The idea is, can you make a protein replication dependent on the presence of an unnatural amino acid? You grow it in the presence of that amino— unnatural amino acid, but when you put it into a host, either as a vaccine or a different environment where that amino acid is missing, After rounds of replication, you deplete the amino acid and the organism fails to grow. So, so people have approached this with a number of clever strategies. One that I think is really useful is, is to create a bio-orthogonal protein-protein interface in an essential protein that depends on the presence of that non-canonical amino acid. So to demonstrate that, let me show you what we did with chorismate mutase. It's an essential enzyme in E. coli. For the biosynthesis of aromatic amino acids. It's a homodimer. So we went into the, the homodimer interface and replaced tyrosine 72 with a large bulky amino acid, benzophenone. And what that does is it destroys the protein-protein interface. The protein is no longer active. So then what we did is simply make a library of the residues surrounding position 72 and repack those residues to bind the benzophenone. And And that you can do with a simple growth-based selection. So we did that. We isolated a mutant that basically had the same catalytic parameters as wild-type protein, grew almost as well as wild-type E. coli in the presence of the non-canonical amino acid, and its absence didn't grow at all. And the escape frequency was really low. It's 10^-11. So we're now actually applying this to to live vaccine development with DNA N mutants. Now there's the other question, does an expanded genetic code confer a selective evolutionary advantage? And so we've looked at that and others have looked at that in a number of systems. Let me just tell you about one selected system. We put E. coli under thermal stress, so we grew it at elevated temperature. And the growth of E. coli at elevated temperatures depends on the stability of a protein called MetA. And so what we did was we randomly mutagenized roughly half the sites in MetA and grew those mutant MetAs in 10 different strains that encoded 10 different 21st amino acids, simply grew it at elevated temperature and looked at what survived. And what we found was this protein, the that actually has a ketoamino acid forms a covalent cross-link across the MetA homodimer interface, and with one mutation stabilizes the protein roughly 25 degrees, which is unheard of with a natural amino acid. So it really does show in certain settings, 21 amino acids provides an advantage relative to 20. So can you genetically encode 22, 23, 25? And one obvious way to do that is to go to a 4-base code, so 256 codons. So we used in vitro methods to evolve very efficient 4-base frameshift suppressor tRNAs, which work as well as the tRNAs that suppress the TAG, the nonsense codon. And then going from there, you can make a whole series of mutually orthogonal pairs to put in even more amino acids. But the problem with a 4-base codon is you're competing with a 3-base codon, and even if it's a stop, you're competing with termination of the protein. So the way to get around that is to go to George Church, and George was kind enough to recombineer the E. coli genome and removed all of the TAG codons and replaced them all with TAA. So now we weren't competing with anything. There were no TAGs, there was no RF. Release factor. So you can see if you go now TAGx or UAGx, you get very efficient amino acid incorporation. So going to higher organisms, people have gone and genetically encoded a 21st amino acid in flies and worms, in mammalian cells. A lot of this is done with baculovirus and other viral delivery systems. So in this case, we made a baculovirus system., and we can efficiently introduce a 21st genetically encoded 21st amino acid in cell lines and primary cells and stem cells and even in neurons. And just to show the utility of that, we genetically encoded a small fluorophore in mammalian cells. You know, we all use GFP. It's large. You can really only in general label the N or C terminus of a protein without affecting function. But when we genetically encoded the small amino acid protein, you can basically put it any site in the proteome you like. So to show its utility, we put it— substituted it for asparagine 160 in a glutamine-binding protein. And then you can just look at the fluorescence, a LAMBDA Max shift, and titrate the binding of glutamine to this protein. And in essence, this mutant protein binds glutamine with almost wild-type affinity. And then, can you go all the way to a mouse? Can you make a 21-amino acid mouse, which I could argue would be a milestone in synthesis? So, how do you do that? Well, the obvious way is to genetically encode 21 amino acids in embryonic stem cells. And that's a little challenging, so to warm up, we asked whether we could genetically encode 21 amino acids in hematopoietic stem cells, the stem cells that make up all the cells in the blood. And so to do that, we used an EVV-based episomal vector system, and we could add to hematopoietic stem cells, okay, a 21st amino acid in vitro, and then we could differentiate those cells into all the cells of the blood lineage— T cells, B cells, macrophages, and the like, and they all had 21 amino acid codes. So then we got a little daring and introduced and grafted those engineered hematopoietic stem cells in mice, and now we got about 20% of all the blood cells in the mouse have a 21 amino acid genetic code, including their progeny, okay, the hematopoietic stem cells. So then anything now is to challenge these mice with, with an immunogen and see if the B cells during clonal selection actually put in a 21st amino acid into the variable region. So going forward in evolution, you know, for many years, billions of years, life on the planet's been largely constrained by a 20 amino acid genetic code. And I think using chemistry and biology together, we've shown you can lift that constraint to a large degree. To make proteins and even whole organisms with interesting new properties. So while we were doing this, another obvious way to generate more codons is to go to, instead of a 2-base pair genetic code, a 3-base pair genetic code. And so we and others have tried to do this. With Floyd Romesberg, we used the principle of hydrophobic orthogonality. That is, if the third base pair is hydrophobic, it won't hydrogen bond with A, G, C, or T, but in water it'll want to form a stable self, stable self-pair. And so we generated a whole series of hydrophobic third base pairs that were stable selective, and in vitro we showed we could replicate DNA with those base pairs and Wade went on in a really terrific series of experiments to actually show you could do it in vivo. So then we said, well, can we actually now substitute every base in the E. coli genome with another base? And so we chose hydroxymethyl C. It's a really interesting epigenetic modification. The question is, can you replace every C with hydroxymethyl C, which you can then also glycosylate? And we thought we could do this because T4 phage, the T4 phage genome has all Cs replaced with 5-hydroxymethyl-C. So to do this, we did some metabolic engineering and reduced the levels of C and increased the levels of hydroxymethyl-C. And when we looked at the E. coli, it turns out about 65% of the C in the E. coli genome was replaced with hydroxymethyl-C. So we thought that was pretty good. We built a clever selection and did random mutagenesis to see if we could get it up to 95%. And what happened was a little bit surprising. When we isolated the genomic DNA and actually analyzed it, we found that roughly half of the G and half of the A was ribose G and ribose A. So all of a sudden we're making a chimeric genome And which, you know, a third to half of the entire genome is RNA and not DNA. And how could that be? Well, first of all, there's no U. We didn't see U, which means it's not a message ribosomal RNA, tRNA contaminant. We did spiking experiments to rule that out. And those levels are way too high to be the result of an RNA H, A, or B repair mutant. So what is going on? Is it simply an RNA-DNA hybrid we're isolating? And it's not. So we, we isolated genomic DNA and broke it down into tri- and dinucleotides and analyzed the structures of those. And what you have is a ribose covalently linked to a deoxyribose. Okay, so it's not hybridized, it's actually covalently linked in a single strand on both strands. So that was a big surprise. And so if that's true, well, then the genomes ought to be sensitive to RNases. And so you can see here, Th10b coli is resistant to RNase I and RNase H, but these new strains we're generating with these chimeric genomes are degraded, completely degraded by RNases as expected. And we can get down to like 20-nucleotide fragments. So maybe there are multiple genomes, and the genome responsible for replication is a DNA genome, and we have these other RNA-DNA chimeric genomes. We ruled that out by specifically labeling the genomic DNA at very high levels and looking at correlating radioactivity with colony-forming units. And really there's only one genome. Then we did a metabolome analysis and as expected, we saw a very large drop in the deoxy levels, no drop in T levels, and that's why we don't see U, no drop in ribose levels, that's why we get ribose incorporation. And then in vitro we showed if you really increased ribose levels over deoxy levels, you actually start with clonidine, you start to get RNA incorporated into a template. So the problem with this is it's really hard to study that genome intact because it's fragile. Okay, so we've done candidate PCR sequencing and haven't found the mutants yet that are responsible for this. So what we're now building is an orthogonal T7M15 phage system in E. coli where we can actually replicate a plasmid that has a chimeric plasmid but leave the genome intact, so we have a stable host strain and we can isolate plasmid DNA and analyze that. We're well on our way to doing that, I think. We're also looking at the mitochondrial yeast polymerase as an orthogonal system, and also Cheng Lu's orthogonal yeast system. But interestingly here, you know, our efforts in synthesis to make new base pairs and replace bases in DNA actually went backwards and potentially is shedding light into a major, you know, transition during evolution. And that's from storing information in the form of RNA to storing information in the form of DNA. How did that happen? Well, maybe this kind of chimeric genome that we serendipitously generated right represents a transition between those two worlds. So after we did this, we said, well, what other major events in evolution can we study from a synthetic point of view? So the other major transition that occurred was the evolution from prokaryotes to eukaryotes. And eukaryotes have a nucleus, which would be hard to synthesize, and mitochondria, which provide energy. And where did mitochondria come from? Well, there's this endosymbiont theory of mitochondrial evolution that early proto— pro-eukaryotes engulfed— early pro-eukaryotes engulfed prokaryotes, and the prokaryotes provided an energy source to the eukaryotes and allowed them to colonize different niches. And also because the prokaryotes now living in the eukaryote minimize the genome. So to recapitulate that, we used two, you know, well-characterized systems. Yeast is the eukaryote and E. coli is a prokaryote. We knocked out the mitochondrial function of the yeast by making COX-2 mutants so the mitochondria don't make energy. And then what we did was engineer the E. coli with an ATP translocase so they spit ATP out. Into, in theory, the yeast cytoplasm. We made the E. coli depend on the yeast to provide a cofactor, so the E. coli are auxotrophic in a cofactor. And then we coated the E. coli with SNARE-like proteins to prevent their degradation in the cytosol of yeast. And then finally, we figured out how to transform live E. coli into the yeast, and what we found then When we isolated the yeast and PCR'd up the DNA, we found yeast genes and E. coli genes as expected. We have a GFP on the E. coli. And the yeast are now sensitive to antibiotics. If you kill the E. coli, you kill the yeast. And, you know, rough PCR analysis suggested that there are about 4 E. coli providing the energy to the yeast acting as the mitochondria. And when we do fluorescence imaging, that's exactly what we see here. For instance, on the upper left, you see 4 E. coli swimming around in the yeast providing now energy. So we have kind of an artificial laboratory system now. So the first thing is, is, well, the E. coli are in yeast, they don't need to make amino acids or cofactors anymore. And that's true, you can basically knock out the biosynthesis of each amino acid and cofactor. And now we're actually trying to remove the whole cell wall to get to a mitochondria, and you can do that by making mutants in cell wall biosynthesis and using what's called L-form E. coli. But an even more interesting question is, is you have a rapidly replicating bacteria and a slowly replicating yeast, how can that be stable, okay. And so we looked at making more stable systems. The first systems we looked at underwent around 40 rounds of replication. We now have a system that's stable for a year. And we got that by simply selecting for stable mutants of the E. coli. And what we did was we got transposon insertions into 3 genes: RCSC, CPXA, and IDNK. And we've shown that those trosphodoson mutations actually change the function. They don't delete the gene, they change the function of the gene. And these genes are involved in things like biofilm formation and cellular stress and carbohydrate metabolism, so they all make sense. We haven't entirely deconvoluted exactly how these genes are contributing to the stability of this endosymbiont, but I think it's a really interesting opportunity to explore, you know, how did mitochondria actually evolve. So that's— that was the last slide. It remains to thank me for just the terrific group of co-workers I have and collaborators who carried out the work, and take any questions in the few minutes we have left. Thanks a lot.

Thank you very much, Pete. I'm impressed by how you are actually peeking into evolution, not only what may have happened in the past, but what could be happening in the future that you could drive. It's awesome to me.

Yeah, I know. It's a different world of synthesis, Dick.

Um, you, you, you, uh, kept stressing 20 amino acids, and I wanted to know what about the possibilities of yet other amino acids like those that contain selenium? Are you sort of ruling those out, or they don't count?

Well, you know, there's selenocysteine and there's pyrrolysine and those type of things. I mean, yeah, they're found in, in a handful of organisms, you know, in a handful of proteins. It's interesting to make selenocysteine, you got to go through, you know, you amino acylate serine and convert the serine to a selenium. So it's really complicated. The kind of neat thing about pyrolysine is after we kind of figured out the simplest way to do it and people showed pyrolysine, it's genetically encoded, it's pretty much the same thing. You build an orthogonal tRNA, you know, it's the same drill. So The question is, is why hasn't that been more widely used? I mean, it's not obvious that pyrolysing brings a whole lot of interesting functions, and selenocysteine is really oxidatively, you know, problematic. So, but the question is, why not a keto amino acid? You got to go through all this work to make cofactors like pyridoxal and thiamine to get, you know, carbon-centered anions. There's a lot easier way to do it. So It is an interesting question. Why didn't Mother Nature do that?

Okay, much to ponder here.

Thanks very much. Okay, well, thanks a lot, and again, Bob, congratulations. Thank you.

Bye-bye.

Great work. Thank you.