Scientist Stories: Jill Banfield, Pioneer in Metagenomics, CRISPR, and Gene Editing

Good morning. I have to confess that what you just heard is completely true, and despite the fact that probably everybody expected me to come here and talk about acid mine drainage, which we've worked on for a very long time, I'm not going to do that. In fact, in contrast to most of the talks we've heard and probably will hear, I'm not going to talk about integrating a body of research that represents a decade. I'm actually going to talk about stuff that's all unpublished and none of it's older than about a year. So I come to the topic of sediments sort of in a backwards and roundabout way, but I have to tell you from the beginning that actually I'm an Earth scientist, a geologist. And I came to start to study microbes because they're fundamentally important in shaping the Earth around us, particularly the surface. So I start out with this photograph taken in a quarry with a bunch of sediment to give you a little bit of context. And I'm going to talk about why we should study sediment and what we can learn and how in fact this will address some of the big questions that we've been hearing about at this meeting, particularly the topic of the dark matter of the biosphere. Carbon cycling, and novel enzymes and processes. So why should we study sediment? First of all, in sediment we find a lot of organic carbon. In fact, that's the carbon that's transitioning from the active cycling biosphere into the deep earth or into the geosphere. Processes that take place during that transition over hundreds, thousands, millions of years, of course, determine the nature and abundance of carbon that does end up in the fossil biosphere. In fact, if we look back over time, time we know that processes in sediment have a very big impact on something as fundamental as atmospheric composition. In fact, in the Proterozoic, there were times of active tectonics where sedimentation rates increased, bearing more organic carbon, leaving more oxygen in the atmosphere. So this kind of dynamic is actually really fundamentally important, but I would say actually sedimentary systems are relatively unstudied compared to a lot of other systems. In fact, even soil, which is a topic of current investigation by many teams. Sediment's different. It's the pile of detrital material that lies underneath. It's often deep, and I would say typically much, much, much greater in abundance by volume than soil. There are some various statistics we find around that tell us that a large fraction of the biosphere is in sediment. Surprising, actually. I'm not sure which of these numbers to believe, and you can see that they range quite considerably. But there's a lot of stuff down there. The other reason I think that the sedimentary system is an interesting place to study is because it's anoxic for the most part. In fact, I think about Earth history a lot, and I know that early in Earth history, the atmospheric composition Earth was very different, and you've all heard about the Great Oxidation Event around 2.4 billion years ago. And just to put that in context, here's time coming from the ancient Earth, rise of the biosphere potentially around here, through to the modern time. We can see that there's been a tremendous change in oxygen composition, particularly around this point about 2.4 billion years ago. So what happened to life on Earth at that time? All of a sudden there's oxygen in the atmosphere, so if you're an anaerobe, you have a problem. So probably what happened is that the anaerobes retreated to the deep subsurface where they're hanging out today. Because you will be diversifying, because you will be representing the kinds of metabolisms and organisms that we otherwise might not encounter. So it sort of seems logical that this would be a pretty interesting place to study if we want to understand about the nature of biology today. Let me tell you the system in which we're working. I'm going to talk about 3 little parts of this story. All of this research is being conducted at the Department of Energy's IFRC research site at Rifle, Colorado. It's an aquifer. It's an aquifer that's controlled by the Department of Energy, so you can do in situ manipulations, which I think is really exciting and going to be very critical in terms of the future, in terms of understanding what organisms are doing in natural systems, the ability to do experiments. I'm going to start off with the— actually backwards, I'm going to start off with the background sediment and talk about data we've had for exactly 2 weeks. So very, very, very recent information, 2 or 3 weeks, that starts to describe what's out there in the background before we start to mess with the system. And then I'm going to talk about 2 different experiments in which we sample the biosphere in the sediment in the subsurface or below the water table, which is here, either in the planktonic fraction, meaning the cells that are swimming around in solution, or directly in the sediment after it's stimulated. In the case of the stimulation, I'm going to talk about acetate amendment. So we're going to manipulate the system and see what happens. Okay, so these are the kinds of questions we're going to be asking. What kinds of organisms are in the subsurface? How diverse are these communities? What kinds of metabolisms do they have? And how do they individually and in aggregate affect biogeochemical cycles? We're very interested, of course, in methane that's been released through degradation processes under anaerobic conditions, complex organic carbon sedimentation, possibly some of the 6,000-year-old carbon that makes it into the sediments, nitrate, sulfate, and the list goes on and on. So here's our approach. We start off with metagenomics, something we've been doing for a long time, methods developed mostly through study of this acid-mind lineage system. And we're going to talk just at the beginning about what you can get from assembly of one lane of Illumina metagenomic sequencing. One lane by the Joint Genome Institute two weeks ago was 52 gigabases of sequence, 150 base pair reads. I'm going to talk about the grand challenge of binning, which to me is actually the most significant thing we need to deal with when we're dealing with metagenomics, the systems where we can get assembly. And I think that's something we should all be looking for, systems where we can get assembly, because assembled data is so much more powerful, as I hope you you will see the unassembled data. The methods we use involve all sorts of different metrics, tetranucleotide composition, that's basically how the genome uses ATCG, coverage, how deep we sampled. Most important we realize now is time series abundance patterns. If we can get a time series, the power of our binning methods go up orders of magnitude. And of course there's the phylogenetic methods that can be added onto that. Critical to our approach is these things called emergent self-organizing maps, which are basically just a tool to analyze analyze these signatures. I'm going to show you a bunch of sort of pretty pictures. I'm not going to go into the details of how these work, but this is published actually and tested in the acid mine drainage system some years ago now. This picture on the right-hand side is an emergent self-organizing map. It looks actually at first glance a lot like the sediment cobbles, but actually this is a representation of the data. Every point in that map is a 5-kb fragment, and basically its placement on the map in this particular map is determined only by sequence composition. We've not added any assumptions, anything else. We've just basically created the map. Any dots, and actually I'm sorry you can't see the dots because they're tiny, but any dots that fall together within an area are presumed to belong to the same genome. They share the same signatures, and these mountain ranges that separate them, the dark lines, are the boundaries of the cluster. So it's just a clustering method. In itself you can see by looking at this 50 gigabases worth of data, actually co-assembly of data from 3 samples, we can obviously see that we can right away pull out bins of genome sequence. To give you a bit of context of how much information we can get out from a relatively straightforward assembly with that scale of sequencing, this is what we're dealing with. A lot of assembled data coming out of this. We're using an assembler that takes into consideration the fact that different organisms in the community have different abundance levels, which has been a traditional challenge with assemblies, metagenomic data that don't have isolates within them. But using these approaches, We can see here that just for example on 5-kb fragments we have almost 400,000 genes. And I would say to you that each one of those fragments is potentially binable. This is just a single lane of sequencing. In fact, we will probably triple the amount of sequencing, so you'll see, I hope in a few moments, the extent to which this can be scaled to genome recovery. Onto the emergent self-organizing map. This time I'm showing one with a kind of landscape coloring to give you the topology, but again Again, the dots represent 5-kb fragments. Actually, I think this one was 10-kb fragments, I have to correct myself here. Never mind, it doesn't matter, it's a detail. Superimposed upon this map is another amount of information, of the type of information, and that information is how the abundance of the organism that contributed that dot changed across the time series. So the dot is black if it's very abundant in the deep sediment, and it's blue if it's abundant in the near-surface sediment. So by superimposing on that, we see we can resolve the structure of the come up a great deal better. This is by considering another ratio of abundance. We've got 3 samples: A to B, A to C, B to C. And in each case, you can see that addition of the time series abundance information helps us to resolve these bins. So the bottom line is that relatively quickly and easily we can pull out the fragments that represent a genome. A genome that in the case of this particular sedimentary system is almost certainly from a highly novel organism. It's hard to communicate this with trees and I just have one, I think. This is the archaea in the sediment. We're very interested in the archaea in the background sediment. Onto this tree you can see we have branches marked. The take-home point is that here we have the Crinarchaeota, and the purple are the Euryarchaeota, and coming out of the sediment are whole branches on the archaeal tree that we've not sampled previously. These are not branches represented by 16S ribosomal RNA genes. These are branches represented by now genomes, genomes for things that were in fragments large enough that they can be bent. So as I said, this is very novel data. I'm not going to present any of the detailed analysis, but let me just show you what the system really looks like that we're dealing with here. This is not Acid Mine drainage where the dominant organism is 40% of the sample when we started in 2001. No. This is the rank abundance curve from one sample, 21 gigabases worth of data, half a lane. The most abundant organism in this sample is 0.2%. So we are not talking about genome recovery, but genome assembly of abundant things. This is that's a tiny fraction of 1%. And we go on out here to 45 different organisms sampled at depths sufficient, even in this small amount of data, to recover genomes. Almost every single genome is novel. Many of the genomes are on branches we don't even know what they are. They're coming off in different parts of the tree, almost certainly represent members of previously unstudied phyla. And just imagine what we'll get when we throw in another 50 gigabases onto this samples. So one of the motivations of trying to study the subsurface is in fact to learn how to manipulate the subsurface to do something, to remediate uranium, to change the porosity and permeability, to change the rate of methane flux from the sediment, many different objectives. In this regard, there's been a lot of work at the Rifle Site with acetate amendment. And the next thing I'm going to show you is what happens, at least in one case, In this case, we studied the microbial communities as they transitioned through this geochemical period from iron reduction into sulfate reduction. So here we have time on the horizontal axis. This is a conceptual diagram. I'll show you the real data in a moment. Here we have concentration. You can see basically this is the loss of uranium, which is of course the motivation for this particular study. We see we go through a period of iron reduction into sulfide reduction. When sulfide becomes abundant, the iron is titrated out iron sulfide. So in this context, we collect samples and we ask, how does the microbial community change as we transition along that gradient? We would expect to see communities with different compositions, different members, some overlapping members, some distinctive. In this particular experiment, the aquifer had actually been amended one year previously, so this is what we call secondary stimulation. And this is probably why this result I'm going to present now differs from what's previously been reported for subsurface acetate stimulation by Derek Lovely and others, where you expect to get Geobacter bloom. This is what we've been studying years and years and years and years at the rifle site. This is different. Here we've already removed a lot of the readily available iron for anaerobic respiration, so we're doing some different sorts of things. And I also should tell you that these samples were filtered, so we're focusing on sort of a fraction of the community that's available. A little bit smaller in size. On the left, the emergent self-organizing map, the dots representing the data points. On the right, superimposed, we have the time series abundance with just 3 samples, and then we have the bins. And what we'll see from analysis of this figure, if you look carefully, is that there are 87 genome bins. This is 25 gigabases of data, equivalent to half a modern lane of sequencing. Not very much data, 87 genome bins. But what are they? To me, this is where it gets interesting. Whoops. Candida division Bd1-5, Candida division Od1. Has anybody ever heard of these? No. Candida division Op11. These are branches of the tree of life. I see Jonathan. These are branches of the tree of life for which we had no previous significant amounts of genomic sampling, with the exception being Op11. There was one cell for which there was 270 kb of sequence recovered by single-cell genomics. And we didn't just get one genome. We didn't just get two genomes. In that 87 genomes, there were 21 genomic samplings of OD1, OP1119, BD155, multiple genomes, and then some new lineages we didn't even know existed. So we can go into the background sediment and pull out hosts of genomes, lineages of genomes, not all complete. We wouldn't expect them to be with this level of sequencing, but many of them quite highly significantly complete, and many of them, for example, in the case of OD1, greater than 90% complete. Comprehensive samplings of novel lineages. So just to put this into perspective, for those of you who aren't so familiar with the candidate divisions, this is the bacterial tree of life. And in black are the leaves, the branches of the tree for which we have isolate sequences. Those are the phyla, like animal phyla, the phyla. The gray branches are those for which we have no isolates, and you can see that's less than half have isolates. And almost all those for which we have no isolates, we have obviously no genome sequence. So just Just in this one study alone, we can add lineages of genomes from 3 of these candidate divisions. So this is the starting point. That's why I'm talking about the future. We can go across this tree, I think, and very quickly build an understanding of the lineages of the tree of life, the protein families, and discover new functionalities. I should point out that these are tiny little leaves from the tree OP11, and both OD1 and OP11 in fact are suspected to be bigger than the proteobacteria based on what we know from 16S sequences. Surveys. Okay, so now we've got 87 genomes from one sample, potentially the next sample hundreds, thousands, tens of thousands of genomes. How are we going to analyze them? This is an enormous challenge, and I'd just like to share a little bit about what we've been thinking in terms of this intermediate scale. Basically, we're using a knowledge-based structure in which we can simultaneously interrogate the genomes no matter how many genomes there are. We can make lists of the proteins proteins across the data set, and then look at how those proteins are distributed at the organism level. So let me just give you a quick snapshot of the kinds of things we might ask. We might ask how accurate, how complete I should say first of all, how complete are your genomes? Well, let's look at the marker genes, just a subset of them shown here, that we expect to find in the genome. We can ask, do all of them have RecA, DNA gyrase A, DNA gyrase B, and so on and so forth. And by examination very quickly, by making this kind of an approach, we can see that these genomes are pretty darn accurate. Complete. We can also look across the organisms in the dataset in terms of their metabolism. So I'm just going to give you this one example. This is BD1-5, for which we have the genomes shown here. We can see that they have, at a glance, through this kind of knowledge-based approach, they all have ATP synthase. They have very few redox proteins, basically no electron transport chain, a few blue copper proteins and one cytochrome, no NADHD hydrogenase. No Lipid A, essentially no lipopolysaccharides, so what are they, gram-negative, gram-positive? Probably not gram-negative at a glance, but we can recognize genes for various structural components, peptidoglycan is well represented, and we can see in fact a bunch of other metabolic components, transporters for example. So what do we take away from this? We take away by looking at this lineage, never before seen, never isolated, that these are obligate fermenters. That's our current hypothesis. No TCA cycle genes, genes not shown here, but they do have essentially no TCA cycle genes. So from this kind of approach, we can build a representation of the metabolism of these organisms. This is an example put together by Kelly Wrighton, who's the first author of this publication describing this work. This is a sub-branch of OD1, which we call OD1-1. There's 10 genomes in this branch. Based on analysis of the genome, because we've got the whole genome essentially, we can We can get a strong sense of what they can do and what they can't do. We can build a reconstruction. Notice they have very, very interesting and novel hydrogenases. In fact, these are archaeal-type hydrogenases not found previously in bacteria. They're coupling hydrogen, they're coupling sulfur. They have all sorts of interesting pathways for degradation of organic compounds, complex organic compounds. And on top of that, this is the only mention I'm going to make pretty much of proteomics, we can make proteomic measurements and see whether or not at the time these samples were collected these pathways were active. So we can superimpose the omic data onto these kinds of studies as well. This is a new potential candidate division, same kind of thing. I show this one because it's neat, because I don't know how many of you are interested in RuBisCO, but actually it turns out that this lineage has a novel type of RuBisCO. It's a 2-3 hybrid. And it's basically found in Archaea. It's referred to as the RuBisCO-AMP salvage pathway. And I actually have to tell you, just because nobody's going to believe me about this, that I have had Frank Larimer look at this I've talked to him very closely, and he agrees that this is actually correct. So it is a new branch. So there's lots of new metabolism to be discovered out there, and this approach has the ability to uncover it. The last thing I'm going to show you is what happens when we stimulate background sediment, because although this is one of the things we did first, I think it's actually a very nice pointer to where we want to go in the future. We want to make hypotheses based on these kinds of genomic analyses, and we want to test them by adding things to see who responds. So we ended up doing this kind of backwards by accident, acetate amendment that's been going on for a long time. And what you see is, I think, exactly what we should be doing in the future with regard to exploring this physiology. This is a 16S ribosomal RNA representation of diversity, and you can see here lots of diversity in the background sediment. When we add acetate to the sediment, a lot of simplification. Enormous amount of different organisms in there, a few of them respond very, very, very strongly. This tells us something. Provides a system where we can move in and selectively recover genomes, new genomes, and it also provides a system that's a lot like our initial acid-mind drainage system insofar as we have relatively high abundance of organisms, so we can deploy proteomics and other kinds of omic approaches to really understand how these organisms are responding, not just who responds, but in what manner. So again, this is a pretty— this is an emergent self-organizing map showing the various bins that were recovered and genomes recovered through this process. Process. These are more similar to known things, and so we can compare them to known genomes, and we know very high confidence that these are essentially complete genomes for at least the dominant organisms, and accurate. And so from this kind of an approach, with this new understanding of who's out there in a cultivation-independent manner, no reference genomes, no mapping of reads to genomes, this is de novo assembly, we can build a sense of what these individual organisms are capable of doing, and then onto that, we can look for connections between them, and we can map information from protein abundance data that are obtained from the same samples at the same time. From the overview we see here, we can see that there's a complex interaction among these organisms involving nitrogen cycling, hydrogen, and of course oxidation of iron and also sulfur actually. So a very highly interlinked system in the sediment. Okay, so I'm going to make some concluding comments. A lot of times for many, many years people have told me you can do it in acid mud energy, you can't do it anywhere else. Never worked does work in complex systems, and I'm here to tell you today it does work in complex systems. I think 0.2% of a microbial community, everything else in lower abundance, is a pretty complex system. That in such systems with a trivial amount of sequencing by today's standards, one lane of sequencing, we can recover not one, not two, not a dozen, but potentially hundreds or even thousands of genomes. They can be near complete, and they can provide tremendous hypotheses about the metabolism of the organisms that exist in the subsurface. And by the way, they can give us that sampling that we heard about yesterday, so desired, a sampling of the tree of life in terms of, for example, the protein families. I really believe that this is the important next step to test these hypotheses in the field, to look at the genomes and instead of using those insights to try and cultivate the organisms as has been done previously, to bring them into cultivation, very important, but nonetheless not here. I'm not talking about that. I'm talking about cultivation of the whole community, stimulation of the whole community so we can study organisms in context to provide insight about what pathways they are or are not in fact capable of. And also I think this will provide a route to understand new genes and metabolisms, hypothetical proteins for which we have no function at this present time. The rifle site has a very strong emphasis on modeling, and I'm not talking about it today because I certainly don't have time, but there are complex transport geochemical models that incorporate biological components into which we need desperately biological constraints, and so this kind of approach I think will provide those. In the long term, I think it's really important we know how to manipulate the environment. We're doing it all around us all the time, but maybe if we're going to do that, doing it with intelligence and based on science would be a good idea. So how do we manipulate the environment to achieve a goal? Not just by remediation, but I think more management of carbon repositories in the terrestrial environment, for example, could come into the same kind of category. Kind of vision. I didn't talk about this, but I did want to just make one last comment that basically when we study natural samples, we recover the DNA in its entirety, and every single read comes from an individual organism, not just a bacterium or archaeon. In fact, we do this with fungi too. Not just a microbe, but viruses and plasmids as well. And we study variation within natural populations, and so we're not looking at clones, we're looking at the nature of how individuals in one sample at one time differ from each other, and of course this really ultimately represents the raw material upon which selection acts, and I think provides a great deal of insight into dynamics on short timescales. So my time is up. I have a lot of people to thank. I think the rightful IFRC team has been critical in this, and I think it's absolutely essential that we continue to be able to conduct experiments in natural systems to really understand them. A lot of people have contributed to the data analysis. Various points. Kelly Whiten is a particularly important contributor to the center, the middle of the three studies. Brian Thomas is our major bioinformatics support, and these other people have been involved in various other ways. And I want to specifically acknowledge the funding from the Department of Energy in this particular SBIR program. It's under threat at the moment. I think that's really very unfortunate because priorities in the government change, but I really hope that we can convince them that this kind of subsurface research is important and sediments are something that's incredibly important to study. And finally, I want to thank the Joint Genome Institute who are our partners in this current sequencing effort. So thank you.