AUSTRALIAN FRONTIERS OF SCIENCE, 2003
Canberra, 31 July to 1 August 2003
Monitoring, molecules
and models – integrating genetics and demography in the analysis
of plant population viability
by Dr Andrew Young
 |
Andrew Young is a Principal Research Scientist at the CSIRO Division of Plant Industry where he leads the conservation biology subprogram. He received his BSc and MSc (Hons) in plant ecology from Auckland University and his PhD in genetics from Carleton University, Ottawa. He moved to the CSIRO in 1993 to take up a postdoctoral fellowship and has continued his ecological and genetic research there for the past 10 years. His work integrates the use of molecular genetic markers, demographic analysis and simulation modelling to investigate the spatial and temporal dynamics of ecological processes such as mating, seed dispersal and recruitment in plant populations. He is the recipient of the Australian Academy of Science's 2003 Fenner Medal for his work on the importance of self-incompatibility genes in regulating plant population viability. He has published more than 40 scientific papers and edited two books. He is on the editorial boards of Conservation Biology, Conservation Genetics and Biological Conservation. |
I am going to talk today about some of the work that we do at CSIRO Plant Industry in the area of conservation biology. Perhaps we are better known more generally for our work in plant production, but along with that, over the past 10 or 15 years, we have developed a fairly large research group in the area of plant conservation biology. That has come out of a recognition that, along with needing to have production systems in Australia, we also need – in parallel – to manage our native biodiversity.
Australia has a rich array of plant biodiversity: tropical grasslands and shrublands in the north, exciting and interesting mallee woodlands in the south, open gum forests in the west and acacia-dominated shrublands in the arid zone – these are just examples of the incredible diversity we have.
If we look at what we think vegetation used to be like, looking at the greens and browns about 100 to 150 years ago, that is the kind of cover we had. But with the expansion of our agricultural base over the past 100 to 150 years we have seen the loss of about 100 million hectares of native vegetation. That is continuing in some areas apace, at the rate of about 60,000 hectares a year.
Click on image for a larger version of figure 1
Given that environmental context, it is fairly obvious that a large part of managing our native plants and animals for long-term conservation and persistence of species and communities is going to involve our managing plants and animals in these kinds of landscapes, where we have potentially relatively small patches of vegetation – a few hectares, maybe, up to a few hundred hectares – within our agricultural matrix. Figure 1 was taken in Western Australia – I think that is a Eucalyptus wandoo – and in the Dongolocking region some shrubland, a couple of intrepid researchers, and a large area of cultivation.
These kinds of remnant patches are important for a whole range of reasons, and I might just pick three. One is that they often represent all that remains of unique ecosystems. A good local example is the white box woodlands. We just don't have big patches. We can't say we are going to manage these species in national parks – we don't have national parks that contain these vegetation types. So if we want to keep them at all, then we have to manage them in the agricultural landscape.
The second reason is that, as you all would probably be aware, the federal and state governments are putting a lot of money into revegetation, to try and deal with a range of environmental issues. And one of the big question marks over that is: where are we going to get the seed from, for doing this? We don't have large-scale native seed orchards. Most of that seed is currently being harvested from natural populations, and because we have the desire to try and source seed locally, to try and maintain local genetic provenance and local adaptation, these remnant sites are really important sources of seed for those projects.
Finally, it is becoming more and more obvious that we don't really need to manage the remnant vegetation just for the native species. Vegetation remnants in agricultural systems can provide at least a few ecosystem services. Sometimes they can provide off-season habitat for the pollinators of crops, and we are also finding that larger patches play an important role in stabilising the hydrological cycle and preventing some of the dramatic effects of dryland salinity. So it is a useful thing to keep these remnant vegetation patches in our agricultural landscapes, in terms of sustainable production.
So it is pretty clear to us that if we are going to move forward in terms of conservation of our native biodiversity, we need to move forward in terms of integrated land management, managing for both production goals and conservation goals in the same landscapes. To do that we need to understand something, I think, about the population biology of how our native plants and animals function in these kinds of fragmented landscapes. The problem is we don't know much about that currently. Most of our classical population biology, our understanding of how plants and animals work, get around the landscape, mate, have kids, where those kids go to school, all those kinds of thing, comes from reasonably intact systems. This has often been by choice – people have tried to look at large systems that they think of as stable-state systems, to understand the 'true dynamics'. But we are mostly now going to be managing highly disturbed systems. So I think we really need three pieces of information.
We need to quantify the spatial and temporal dynamics of key ecological and genetic processes in these fragmented landscapes, and that is a difficult thing to do. We also need to try and get those data on a very broad range of species. And we need to try and replicate those experiments and gather data across a broad range of different fragmented and disturbed landscapes. The landscape I showed you in WA is very different from the landscape here in south-eastern Australia, say on the southern temperate grasslands. So those three things we need to do, and it is a difficult thing.
There are a couple of areas I would like to talk about today: the use of molecular markers to do ecology, so molecular ecology, and simulation modelling. Hopefully, we can get some data out of molecular marker techniques and we can use those data to make some predictions using simulation modelling techniques. So I am just going to briefly talk about those two things.
We have heard a lot today about the power of molecules, from a lot of people who probably know more about molecules than I do, in many respects. But from our perspective there are two things that make molecular genetics a useful thing. One of those has been the development of high-resolution genetic markers, by which we can uniquely identify individuals and keep track of them, and know them again when we see them again by doing multi-locus genotyping, using a range of techniques. The second is the development of powerful statistical tools for inferring paternity and parentage, so we can use those data to do something useful. Taken together, those two things give us the promise, at least, of doing two really useful things.
One is, essentially, molecular ecology. Let's use these genotypes and statistical techniques to do paternity and parentage analysis so we can actually use molecular tools to tackle ecological questions that have previously been really difficult to get at. It is hard to look at patterns of pollination over a long period of time and across a range of landscapes, using traditional techniques like chasing pollinators.
The second is to actually say, 'Well, there are genetic processes that could be crucial to understanding the viability of these populations' – things like inbreeding, and genetic mate limitation because of incompatibility genes, which I will get on to later. So we can start to get some direct information on the genetic processes involved as well in determining population viability.
The second thing I will talk about is modelling. There have been a few advances. One is the development of computing power, so we can have individually-based spatially explicit models which allow us to actually integrate demographic and genetic processes in the same models. We don't have to run demographic models or genetic models and try and pull the data together; we now have the power that allows us to have these processes in the same models and let them interact, and that is certainly more useful. We can also now cope with multiple populations in our models, and because we have, hopefully, some great data from our molecular analyses, we have got better input data for the models so we can have a better level of prediction.
There are really three things models are very good at, in my opinion, in driving the research forward. One is maybe the most obvious one, and in some ways the most pedestrian: they give us the ability to predict, hopefully in detail, the fate of individual populations if we are concerned about particular species. That is a useful thing to do, even if we can only get relative viabilities out of those, when we are trying to make conservation choices about where to put our limited conservation dollar.
The second, perhaps more importantly, is that models allow us to play a bunch of management scenarios, which we cannot do empirically in the real world, and try and determine then what is the best management strategy, based on the outcome. We can impose all sorts of management regimes that we physically cannot do.
And, very critically, because there are so many species we need to deal with and the data are so hard to get, we can use models to try and extend the range of species that we understand, by using the data from the few we have and then tweaking the models by changing the life history characteristics of the species, and get some output on the fate of species that we have not actually studied directly.
I am going to go quickly through two case studies from work done in our lab. One is very much a molecular ecology study, the other a simulation and modelling study.
The first one, which I think is pretty interesting, is work that was done by a graduate student of mine Susan Hoebee. She is now doing a postdoc in Zurich, having left last year – she finished up the last parts of the work and that is now handed in. It is a really nice piece of work looking at Grevillea iaspicula, which is a proteaceous shrub from around the Lake Burrinjuck area.
This grevillea is bird-pollinated; it has bird- and gravity-dispersed seed, and it generally occurs in a bunch of populations with reproductive sizes of about 20 to 100 flowering plants. These are scattered on limestone outcrops about half a kilometre up to 20 km apart. They are in a highly fragmented landscape, with sheep grazing being the main land use in between them. And this plant has very poor seed set, which is what we were really interested in.
The suggestion by National Parks is that there is a problem with the pollinators. They don't think the pollinators were getting around the landscape and that is why we had low seed set. But in reality we didn't know much about how pollinators get round these fragmented landscapes, and that is what we wanted to know.
There are a bunch of ways Sue could have gone. She could have gone out and started looking at the pollinators – bird-banding them, following the birds – but the problem when you do that is that you get the movements of the birds but even though you are pretty sure they are the ones doing the pollination, were they really carrying pollen? And you can't look at all the birds all the time, so your datasets are quite limited. Even if they were carrying pollen, did the pollen actually fertilise the flowers and were they the ones that produced the seed?
Click on image for a larger version of figure 2
So she decided to go the paternity analysis route – I will spend a little bit of time on this – using microsatellite markers, which are really good codominant markers for doing this kind of thing.
This is pretty typical of what she did (figure 2). Here you have got a mum. So she went into the field and collected seed from a bunch of plants in a population. She then genotyped the mother, say for a single gene locus such as we have got here. These are the different possible alleles, and we know that mum's genotype is 148/152. She has got two alleles, it is a diploid species.
Then she genotyped a whole bunch of seed from that mum. So she could go through and pretty easily say, 'Okay, these must have been the alleles that the mum contributed to all those different seed.' Some we are not sure; here is a heterozygous progeny, and it is of the same heterozygote as the mother, so the mum could have contributed either allele. Some are homozygotes, so we definitely know what she contributed there. For others, the maternal contribution is unique.
She could then go through and say, 'Okay, for the first seed I should be looking for a father who has allele no. 154; for the second seed, allele no. 156; for the third seed, 150.' She could go through doing this, and start to build up a profile of what the fathers must have looked like and who the fathers must be, based on their genotype.
Click on image for a larger version of figure 3
That is quite powerful, but it becomes very powerful when you start using a whole bunch of markers, and she was using eight or nine. Here are three of them, and you can see again we have got the maternal genotypes, down the side; we can infer the paternal contributions; and then again if we look at columns 1 and 2 we now know that the father must have contributed all three of those alleles (figure 3). So we know his profile must be 154/129/210.
As Sue went out and genotyped large numbers of individuals in most of the populations she could find, she could actually go looking for that tree and find out where it was. And the same sort of thing for the second one.
Click on image for a larger version of figure 4
Here is one of her study populations (figure 4). It is about a hectare in size. I think it has got something like 43 trees in it, each marked by a square. A typical question she asked is, 'Who is pollinating the flowers on plant no. 32?' The answer is a whole bunch of plants, and these are the percentages of seed that were sired by these different plants. You can see that here. There is a whole bunch of dads who are responsible. The blue segment is for where we could not pin it down to a single father but we knew it was somewhere in the population. This white segment is the interesting one: 4 per cent of the pollen seems to be coming from outside the population.
When you do that for a whole bunch of individuals in the population, you start to get some really interesting results. First of all, again multiple paternity seems to be the rule. But the really interesting thing is that a fairly large proportion of the seed in a population is actually sired by trees that are in other populations, often as far as 16 or 17 km away. So we are seeing really long-distance pollinator movements. And these are populations for which, up to now, most of the management was based on the idea that they were all individual populations and totally unconnected.
So what do we actually know? We certainly know that there is no pollinator limitation in this species, or these populations. The low seed set is pretty normal; that is just the way the species works. Multiple pollination and paternity of seed is the rule. But the big thing is that this inter-population pollination is a really critical part of the system. If we knock that out, then we may see significant reductions in seed set. So what we need to do is to be sure that we manage multiple populations together, and the kind of landscape scale that we need to think about, in terms of our management, is around 10-20 km. This a real paradigm shift in how we would manage these populations. What it means is that if you knock over a population 5 km away on another farm, it could well affect the population you have on your land.
We are also getting some inferential data about the scale of movements of these honeyeaters, which is not something we had really thought about to start with. And Sue also did some really nice analyses – which I won't talk about – where she looked at seedlings in these populations and found that there were quite a few seedlings for which there were no extant possible genetic parents, suggesting a really long-term seed bank, which is a sort of long-term genetic reservoir.
The second thing I will talk about is a case study on a personal favourite of mine that I have worked on, on and off, for a few years, Rutidosis leptorrhynchoides. It is a grassland daisy. I called it Rutidosis leptorrhynchoides because its common name is worse: the button wrinklewort. So we stick with the scientific name.
It is a herbaceous perennial. It is widespread in the grasslands in the south-east. I was interested in the talk about the virtual system for the astronomical data, because we have got the same thing going with our herbarium records. It lets you pull up these powerful data from all the different herbaria in the country and see where species used to be and where they are now. So this plant is widespread, it has been subject to severe habitat loss because it is prime grazing land, and what we see is that small populations have poor seed set.
Click on image for a larger version of figure 5
Now, this thing has a genetic self-incompatibility system, and this really the crux of the matter. These systems work very generally like this (figure 5). You have a bunch of different genotypes in the population – again we are dealing with a primarily diploid species here – represented by the different flower colours: this individual has the 'orange' genotype. The idea of these systems is that they prevent self-pollination, because there is a genetic recognition mechanism on the stigma, but if you get pollen from an individual of another genotype you are fine. As long as the genotypes don't match, the pollen germinates and fertilises. If you get pollen from another individual of the same genotype in your population, it is just like self-pollen: no success. But all those other individuals in nice big populations are great and there are no problems with mate limitation.
This is tremendous in big populations. You avoid inbreeding and you still get to have lots of mates. But remember we are dealing with fragmented systems, where often populations become very small and the numbers of genotypes also become small. So what happens is that you run into the situation where there aren't many individuals you can mate with.
Click on image for a larger version of figure 6
We have done a lot of crossing and genetic marker work to show that this is indeed exactly what goes on. As population size goes down in these populations, the number of S alleles goes down – these are the genes that control the incompatibility recognition system – and as that number of S alleles goes down we see a dramatic reduction in seed set (figure 6).
This is a nice empirical result, and probably one of the first really clear demonstrations of a clear link between genetic diversity per se and the demographic performance of populations. We now wanted to use our simulation models to do two things.
The first is to determine if there are genetic diversity thresholds. We know the seed set goes down, but what does that do the long-term viability? So we want to know if there are genetic diversity thresholds for population persistence, (a) so we can assess which populations we should put our effort into, and (b) so we can assess which ones we might be able to recover, and how many S alleles we might want to try and put into a population.
Secondly, we have done a lot of work to get to this point on this single species. There are a whole lot of species out there that are self-incompatible. Over half the angiosperm families have self-incompatible species in them; this is not an odd characteristic for plants. We would like to know how other plants with different kinds of incompatibility systems – because there are different forms of genetic control – might respond.
We built a few iterations of different simulation models, but they have a few things in common. They are individually-based and spatially explicit, so plants are generated, they grow, they produce ovules and pollen, they mate, based on a set of rules about incompatibility, they produce and disperse seed, and the whole thing starts again. The allocation of pollen and seed dispersal is all based on the cool kind of molecular data we have been able to get, using a range of markers, so we think it is well based in reality. We track the individual's demographic and genetic status, and we are using data to parameterise those from multiple years of monitoring of multiple populations and multiple individuals.
Click on image for a larger version of figure 7
The results are quite interesting. This is for Rutidosis itself (figure 7). What we have here is population size. We started our populations at 0.1, which is the size of 100 individuals at t0, and we monitored mate availability and the inbreeding coefficient, which we were interested in. The inbreeding coefficient gets more negative, usually, as mate availability goes down, because of the disassortative mating at the incompatibility locus.
We see that if we start them with 10 S alleles, mate limitation gets worse and, uniformly, populations go to extinction. If we start them with 25 S alleles, the situation is not quite as bad but things do go extinct. By the time we get to 50 S alleles, you see initially there is a reduction in mate availability but that recovers under frequency-dependent selection and the populations do okay. And with infinite S alleles, basically where you are self-compatible, things are fine.
So we now know that, in the field, populations with 50 or more S alleles should do okay. They are our best bets for conservation. But it also means we can start shifting seed around to increase the number of S alleles in the small, marginalised populations, and improve seed set.
Click on image for a larger version of figure 8
The second thing about the modelling we want to do is to try and extend this, as I said, to understand generally how limiting self-incompatibility is. This is pretty complex, but what we have got is three different incompatibility systems, the gametophytic system, which is less limiting, we would think, than the sporophytic one we are dealing with, with codominance, and the sporophytic one with dominance is somewhere in between. In figure 8 we are looking at population persistence on this axis of the graph, and here we have got changes in death rates. So this is what our daisy is like, a death rate of about 0.2; this [in one of three columns of charts] might be a medium-term shrub. So we are changing life history characteristics along here, in this case the death rate, and we are changing the incompatibility system here, and along the individual X axes we are changing the fecundity of the plants – how many ovules they produce.
The take-home message is that under some sets of circumstances they all pretty much operate the same. So when death rates are low, all the incompatibility systems have very little effect, but as death rates get higher you start to see that the different incompatibility systems place different limits on population persistence. The overall message out of this is that it is the shorter-lived species, with medium fecundities, that are ones that are really going to run into trouble from loss of S alleles.
To finish up: there are really three things that are holding us back at the moment. One is developing a lot of these multi-locus marker systems for many, many species. If you are dealing with wheat, or cotton, you are working on the same species for a long time; you can get a lot of great markers. But when you are dealing with different species every two or three years, the R&D involved in marker development is a hassle. The second thing is incorporating rare events into our simulation models. And the third, which we have not even touched on, is dealing with symbiotic interactions, things like the plant/pollinator interactions.