Evolution 101: Host-Parasite Interactions: Of Mice and Cheese and Men and Zombies

Posted on September 24, 2012 by Danielle Whittaker

This week’s Evolution 101 Post is by MSU graduate student Alita Burmeister.

Hollywood loves a good parasite story—from zombies and vampires to Alien and Star Trek II — nothing creeps audiences out like a parasitic infection that controls its host’s actions.  Though notorious in sci-fi films, the parasite control strategy is not unknown in the natural world. For example, the parasite Toxoplasma gondii infects both rats and cats, but it can only reproduce within cats. While we know that normal rats avoid cats, T. gondii-infected rats seem drawn to the scent of cat urine. The parasite manipulates its rodent host by reversing the normal aversion to cat scent, thereby increasing the chances of its host being eaten, and consequently, its own reproductive success in the belly of Felix.

Mind-control is not a defining characteristic of a parasite, which is simply an organism that lives in or on a host, deriving benefit at the host’s expense. But like in the movies, the most abundant earthling parasites do indeed replicate by controlling their hosts. These parasites are the bacteriophage. Bacteriophage (“phage” for short) are predators-and-parasites in one that live off of their single celled bacterial hosts. Like other viruses, phages are simple critters—a small genome surrounded by a bit of protein—and reproduce only within a host cell.

Phage parasitize bacteria by binding to the outside of a host cell and injecting their genomes. Once inside the cell, the phage DNA goes to work, co-opting the host machinery and reprogramming the cell into a phage assembly line. If it is unable to stop the infection, the bacterial host cell does the work to replicate the phage—the cell provides building blocks, energy, and a protected environment. The cell replicates the phage genome, turning one copy into a hundred copies. The cell replicates the body of the phage, again producing hundreds of copies. And in a last, self-defeating step, the cell’s own machinery stamps out a protein that eats the cell from the inside-out. The cell dies in a burst of a hundred new infectious phage. Take that, Shark Week. (For more, Radiolab’s lively telling of marine phage cycling is worth a listen.)

The cutthroat simplicity of lytic phage infection makes it a great way to study how interacting hosts and parasites evolve. When bacteria and phage are grown together in the lab, the bacterial population evolves resistance to phage infection, and the phage population evolves new infection mechanisms. This process is an example of coevolution— an interaction in which one organism’s evolution responds to another organism’s evolution. The evolutionary interactions of hosts and parasites are part of a general class of interactions called exploiter-victim interactions, which also includes interactions between predators and prey.

Outside of the lab, we see the implications of hosts and parasites in health, agriculture, and industry. A quirky example comes from Wisconsin, where the (un)official State Microbe is Lactococcus lactis—the bacterium that makes cheese. L. lactis produces an acid that turns white gold into curds and whey and produces cheddar’s distinctive flavor. However, when bacteriophage hang out in cheese factories, reproducing as they kill L. lactis, the cheesemaking process is disrupted. The dairy industry has developed a small arsenal of approaches to deal with phage, saving cheesemakers money and strife. 

Many parasites, however, are not so easy to defeat. The World Health Organization estimates that the parasitic disease malaria caused 665,000 deaths in 2010. The number of people killed by malaria each year would fill Spartan Stadium eight times over.

Malaria presents a double evolutionary challenge. Plasmodium falciparum—the malaria-causing parasite—is evolving resistance to anti-malarial drugs including artemisinin. Artemisinin-based combination therapies (ACTs) are the first-line of defense against malaria, but resistance to artemisinin has already been observed in four countries—all since 2009—making drug resistance an emergent threat to eradication efforts. Another major evolutionary challenge is the emergence of insecticide resistance in mosquito populations. Mosquitoes are the primary way malaria spreads from person to person, and with forty-five countries reporting insecticide resistance, this problem is even more widespread than artemisinin resistance.

Malaria demonstrates a general problem encountered in fighting diseases. The effectiveness of vaccines and drugs can change as their targets evolve. However, an understanding of the evolutionary processes that shape a disease will help us devise prevention and treatment strategies.  That’s good news for us, bad news for the zombies.

Posted in Evolution 101 | Tagged , , | Leave a comment

Zachary Blount shows us evolution in action in E. coli

Posted on September 20, 2012 by Danielle Whittaker

BEACON member Zachary Blount, formerly a graduate student and now a postdoc in Richard Lenski’s lab, is the lead author on a new paper in Nature describing the step-by-step process by which E. coli evolved the ability to consume citrate in Long Term Evolution Experiment. Blount and Lenski first described the appearance of this new trait in a 2008 PNAS paper which made headlines as an example of evolution in action. Carl Zimmer explains the new study on his blog The Loom.

To hear the whole story in Zachary Blount’s own words, watch his doctoral dissertation defense from last year!

Posted in BEACON in the News, Videos | Tagged , , | Leave a comment

BEACON Distinguished Postdoctoral Fellows Program

Posted on September 18, 2012 by Danielle Whittaker

BEACON is once again accepting applications for the Distinguished Postdoctoral Fellows Program.

BEACON is an NSF Science and Technology Center headquartered at Michigan State University with partners at North Carolina A&T State University, University of Idaho, University of Texas at Austin, and University of Washington. BEACON brings together biologists, computer scientists, and engineers to study evolutionary dynamics using biological and computational techniques and to apply evolutionary principles to engineering problems. We seek outstanding post-doctoral scholars to pursue interdisciplinary research on evolution in action with BEACON faculty members, in the fields of biology, computer science, and/or engineering.

Applicants will propose a research project within the scope of BEACON’s mission and must have two BEACON faculty sponsors who will serve as research mentors should the fellowship be awarded. At least one sponsor must be from the MSU faculty; the other sponsor may be from any of the five BEACON institutions. Preference is given for interdisciplinary research. The postdoc fellow will be based at Michigan State University in East Lansing. Please see our website (http://www.beacon-center.org) for information about BEACON’s mission, participants and ongoing research projects.

Applicants must submit the following, in a single PDF, to BEACON Managing Director Danielle Whittaker via email (djwhitta@msu.edu):

  1. CV
  2. A two-page description of their research plan
  3. A one-page summary of their doctoral research

Also required (may be sent separately):

  • Letters of support from two BEACON sponsors (at least one must be from MSU)
  • Two additional letters of recommendation

Fellowships are for two years and include a salary of $50,000/year and modest funds to support research and travel. The successful applicant will help foster collaborations among faculty and disciplines and serve as a professional model for pre-doctoral trainees.

A Ph.D. in biology, computer science, engineering or related fields is required. Current MSU graduate students or postdocs are not eligible for this fellowship. Minority applicants are especially encouraged to apply. MSU is an Equal Opportunity/Affirmative Action Employer.

The deadline for applications is December 15 of each year. Finalists will be invited to give research seminars in January/February, and the award will be announced in late February.

Posted in Job Openings | Tagged | Leave a comment

Evolution 101: Epistasis

Posted on September 17, 2012 by Danielle Whittaker

BjørnThis week’s Evolution 101 post is by MSU postdoc Bjørn Østman. Bjørn also blogs at Pleiotropy.

What is epistasis?

Epistasis is a measure of the strength of epistatic interactions. Epistatic interactions are non-additive interactions between alleles, loci, or mutations. That is, if the combined effect of a pair of mutations is not what we expect from their individual effects, we then say there is epistasis between those two mutations.

Two mutations that are both detrimental on their own can be beneficial when they occur together. An example of this is from Joe Thornton’s lab: the present function of reduced sensitivity to hormone in vertebrate glucocorticoid receptor is an example of this. Two mutations both reduced sensitivity and destabilized the newly duplicated gene shortly after its birth 450 million years ago. A third mutation – neutral without the first two mutations – buffered the destabilization, and allowed to gene to go fixation (Carroll et al., 2010).

Epistasis is mostly measured in terms of fitness, as the deviation from additivity, but in principle any trait-value can be used*. If mutation A increases fitness by 5% and B increases fitness by 10%, then we might expect that an organism with both mutations get a fitness increase of 1.05×1.10=1.155 or 15.5%. This would be the case if the two mutations do not interact, so that their effects on fitness are independent of each other. The deviation can be measured in various ways, but the proper way of doing it would be like this:

ε  = log10[WAB × W0/ (WA × WB)],

where W0 is the fitness of the organisms with neither mutation. This is the best definition (!), because we assumed above that the effects of the mutations are to increase fitness by a fraction of the current fitness, rather than by adding a number. If mutations did increase fitness by an absolute number, we might measure epistasis as

ε  = WAB + W0 – (WA + WB).

Both of these measures are then zero when there is no epistasis, and both can be extended to deal with more than two mutations interacting. When ε>0 we call it positive epistasis, and negative epistasis when ε<0 (Fig. 1).

So, if an organism with both mutations have a fitness of 1.20, then the amount of epistasis is ε  = log10[1.20 / (1.05 × 1.10)] = 0.01660. If two deleterious mutations together have a beneficial effect, the sign of the joint effect is reversed, and this is called reciprocal sign epistasis (e.g., WA = 0.95, WB = 0.90, WAB = 1.20, giving ε = 0.1472). A trivial case of negative epistasis is when both mutations are independently neutral, but their joint effect is deleterious (e.g., WA = 1.0, WB = 1.0, WAB = 0.90, ε = -0.04576). I say this is a trivial case, because this type of interaction could be one where two genes carry out the same function, thereby exhibiting robustness by being redundant; the organisms then only suffers a fitness decrease when both genes are not working properly.

Fig. 1: Schematic illustration of epistasis. Two mutations A and B can interact epistatically in different ways with varying effects on fitness. The fitness of the wild-type is represented by the black baselines, and the heights of arrows represent the fitness after one mutation (WA or WB) and after both mutations (WAB). Green, positive epistasis, red, negative epistasis, black, no epistasis. In (a), two independently beneficial mutations may have their joint effect increased or diminished (WAB larger or smaller), while in (b) the independent effect of the two mutations is deleterious and beneficial, respectively, and the combined expected effect on fitness is deleterious. In (c), each mutation by itself is deleterious, but when they interact, the result can be reciprocal sign epistasis (green arrow). These sketches illustrate an additive model, where the sum of WA and WB is equal to WAB without epistasis. In our model, using the geometric mean this corresponds to taking the logarithms of the fitness. From Østman et al. (2012).

Epistasis is a feature of the genotype-phenotype map, and of genetic architecture. The genes that together are responsible for a trait (e.g., eyes, lungs, blood-clotting) are likely to interact and have non-zero epistasis. Many genes are also pleiotropic, i.e. part of gene-networks of more than one trait (Fig. 2), as they are expressed in different contexts (tissues, cell-types, in response to different environmental cues, etc.).

Fig 2: Epistatic modules. (A) Hypothetical genotype-phenotype map with three modules of groups of genes affecting three traits: eyes, lungs, and blood-clotting. The genes within each module interact epistatically, while some genes exhibit pleiotropy (black arrows). Not all pairs of genes affecting the same trait necessarily have a non-zero epistasis. (B) Human liver coexpression network and corresponding gene modules. The gene coexpression network consists of the top 12.5% most differentially expressed genes (5,012 expression traits). The colors of the nodes represent their module assignments. Each of the colors correspond to a trait, and most genes are only expressed in that trait, though some are expressed in more than one (pleiotropy), as indicated by lines signifying coexpression. From Friend (2010).

Why is epistasis important in evolution?

One reason why epistasis is so important in evolutionary biology is that it affects the fitness landscape. The structure of the fitness landscape in large part determines many important things in evolution, such as evolvability, robustness, repeatability, contingency, and speciation. If the environment dictates that on set of genes/loci has a particular combination of alleles that optimizes fitness, then without epistasis each gene can be optimized individually until the optimal combination is reached (i.e., there is one peak in the local fitness landscape, aka smooth landscape). Deterministically, the population will end up on the peak. However, if the genes/loci interact, then fitness values are modified, and the fitness landscape will no longer be smooth, but contain multiple local peaks with valleys in between. Evolution in such a rugged fitness landscape will not be predictable, and multiple outcomes are now possible. Because there are multiple peaks the population might get stuck on a local peak with lower fitness than the highest peak in the landscape. Another possibility is that more than one peak is climbed at the same time, and if such a situation can be sustained, it can lead to evolutionary branching and even speciation.

Another reason why epistasis is so important is that interactions between genes means that much more complex traits can be made. If genes did not interact

, then no trait would be affected by more than one gene (is this necessarily always true?). It is of course not possible to make a complex structure with only one kind of protein. Conversely, the more genes interact within a module, the more complex the trait can be, which in turn translates into higher fitness. With only a handful of genes available, only a simple eye can develop, while many genes together can make a more complex structure, which can increase the organism’s fitness. The fact that genes interact epistatically is why complex multicellular organisms with abundant cellular differentiation are possible at all.

How prevalent is epistasis?

Very. Basically, when people measure it, pretty much all pairs of mutations are epistatic. That’s hard to believe is true, and it probably isn’t. Measuring fitness is generally difficult; you have to measure the fitness of four organisms, and just a little bit of error will give ε different from zero. Therefore it is reasonable to attribute lots of non-zero measures below some limit to no epistasis. And then still, it turns out lots of pairs of mutations have significant epistasis between them. For example, Costanzo et al. (2010), using data from a genome-wide, quantitative analysis of genetic interactions in yeast, showed that even when including only high values of epistasis (|ε|>0.08), then a large fraction of gene pairs are epistatic (Fig. 3A). Or in Drosophila melanogaster, where 15 insertions in the genes involved in startle-induced locomotion show extensive genetic interactions (Fig. 3B)

Fig. 3: Prevalance of epistasis. (A) The distribution of genetic interaction network degree for negative (red) and positive (green) interactions involving query genes. From Costanzo et al. (2010). (B) Epistatic interactions for startle-induced locomotion among 15 P[GT1] insertion lines in double heterozygous genotypes. From Yamamoto et al. (2008).

What is the current research focus?

Two major areas of research in evolution are adaptation and speciation. This has been so for a long time, and while we do know a lot about both, there is little doubt that this will not change in the foreseeable future. Adaptation is particularly affected by epistasis and pleiotropy, and it is an outstanding question to what extent adaptation is enhanced or mitigated by epistasis. Empirical data suggest that epistasis causes diminishing returns (e.g., Kahn et al, 2010), but this probably just means that the shape of fitness peaks are shallower the closer you get to the apex, which would just mean that the biggest returns on fitness comes with the first beneficial mutations (which are more likely to go to fixation in the first place). How much does epistasis affect evolvability? Fitness landscape ruggedness can limit a population’s ability to evolve, and ruggedness depends on the amount of epistasis among and within genes. But are these epistatic interactions set in stone, or are they malleable? In other words, how easy is it to create epistatic interactions, and once formed, can they be broken and allow for new advances in adaptation?

Speciation is also a much studied area of evolutionary biology, but the impact of genetic architecture is only recently coming into focus. Epistasis can cause Dobzhansky-Muller incompatibilities, which can lead to reproductive isolation (which is cool if your gold standard of speciation is the Biological Species Concept). But more generally, the epistastic nature of the genetic architecture causing multiple fitness peaks implies that evolutionary branching can occur. It remains an open question how much this is governed by epistasis, and particularly whether epistasis is a prerequisite for speciation of microbes.

* Not that I am thereby saying that fitness is just another trait. I hold the view that fitness – reproductive success – is a function of other traits, such that a network would point from genes to traits, and traits to fitness.

References
Carroll SM, Ortlund EA, and Thornton JW (2011). Mechanisms for the evolution of a derived function in the ancestral glucocorticoid receptor. PLoS Genetics, 7 (6) PMID: 21698144

Costanzo M, et al. (2010). The Genetic Landscape of a Cell Science, 327 DOI: 10.1126/science.1180823

Friend SH (2010). The need for precompetitive integrative bionetwork disease model building. Clinical pharmacology and therapeutics, 87 (5), 536-9 PMID: 20407459

Khan AI, Dinh DM, Schneider D, Lenski RE, and Cooper TF (2011). Negative epistasis between beneficial mutations in an evolving bacterial population. Science (New York, N.Y.), 332 (6034), 1193-6 PMID: 21636772

Yamamoto A, Zwarts L, Callaerts P, Norga K, Mackay TF, and Anholt RR (2008). Neurogenetic networks for startle-induced locomotion in Drosophila melanogaster. Proceedings of the National Academy of Sciences of the United States of America, 105 (34), 12393-8 PMID: 18713854

Østman B, Hintze A, and Adami C (2012). Impact of epistasis and pleiotropy on evolutionary adaptation. Proceedings of The Royal Society Biological sciences, 279 (1727), 247-56 PMID: 21697174

 You can leave comments for Bjørn on his blog, where this is cross-posted: http://pleiotropy.fieldofscience.com/2012/09/epistasis-in-evolution.html

Posted in Evolution 101 | Tagged , , , | Leave a comment

Watch a talk by BEACON's Titus Brown

Posted on September 12, 2012 by Danielle Whittaker

C. Titus Brown, “Streaming lossy compression of biological sequence data using probabilistic data structures.” A talk given at the Michigan State University Computer Science department on September 7, 2012.

Posted in Videos | Tagged , , | Leave a comment

Eco-Evolutionary Dynamics

Posted on September 10, 2012 by Danielle Whittaker

This week’s Evolution 101 blog post is by MSU graduate student Byron Smith.

Evolution is often thought of as a constant, gradual change in the characteristics of a species based on how well those traits are adapted to a static environment. A more nuanced understanding of evolution, however, discards this linear and rigid concept in favor of a much more dynamic (and interesting) process. Environments are generally recognized to be anything but constant, resulting in an ever changing fitness landscape, the relationship between the characteristics of organisms and their reproductive success.

Darwin's FinchesEven the most commonly used examples of evolution demonstrate the impact of ecology: think of the adaptation of Darwin’s finches to the fluctuating availability of seeds, or the peppered moth, which increased its pigmentation, maintaining its camouflage after soot from the industrial revolution changed the color of the moth’s habitat. Ecology clearly influences evolution.

What is less obvious, and has only recently become apparent, is the reverse: the potential for evolution to impact ecology. The bi-directional interaction of the two, called eco-evolutionary dynamics, was long considered unlikely because of the assumption that evolutionary changes occur on much larger time scales than ecological ones.  Among others, a landmark study by Takehito Yoshida et al. (2003) demonstrated that microscopic communities of algae and rotifers (a predator of algae) show different predator-prey dynamics when natural selection is given genetic variation to act on.

It is legitimate to point out that this system benefits from an increased rate of evolution due to the short generation time and large population sizes of the microorganisms studied. These characteristics suggest that eco-evolutionary dynamics play a larger role in microbial systems than slower evolving animal/plant dominated ecosystems. Understanding these phenomena will likely be an important part of the current revolution in microbial ecology.

Many simple, real world evolutionary games, especially those in microbial systems, are great examples of a feedback between the environment and evolution. A common example of an evolutionary game is the group production of iron scavenging machines called siderophores. Since iron, a vital nutrient, is found outside of the cell, siderophores are released by an entire population of needy microbes. After a siderophore grabs a molecule of iron it is collected by a bacterial cell to be used in important cellular processes. Since everyone is doing their part and producing the costly collecting machines, everyone benefits from the increased iron availability.

What happens, however, when a few of these bacteria mutate to not produce siderophores, but continue to collect them from the environment? Since the freeloaders don’t pay the cost of production, its fitness is larger, and its population grows. At some point, however, the number producing bacteria drops too low and iron collection becomes nearly impossible. The ecology of iron scavenging in bacteria is at the whim of evolution.

GuppiesThat is not to say, however, that eco-evolutionary feedbacks are limited to microbes.  Among others, Eric Palkovacs et al. (2009) demonstrated that evolution of native guppies in Trinidadian streams influenced a variety of ecosystem properties, including algal and invertebrate biomass.

The field of ecology is finally realizing the deep truth said best by Dobzhansky (1964):  “nothing in biology makes sense except in the light of evolution.”

Posted in Evolution 101 | Tagged , | Leave a comment

Evolution 101: Maternal Effects

Posted on September 3, 2012 by Danielle Whittaker

This week’s Evolution 101 blog post is by MSU graduate student Emily Weigel.

Pregnant woman's bellyThis is a moment to thank your mom.

Mothers have more of an effect on their offspring than one might first think. In addition to the DNA, both mitochondrial and nuclear, that a mother provides to her children, mothers also influence the development of their children through “maternal effects.” Maternal effects act on the expressed characteristics of the organism (the organism’s “phenotype”) so that the organism is influenced not just by its own environment and genes, but also by the environment and phenotype of its mother.

What are maternal effects, and how do they work?

Maternal effects can be seen in the way a mother provisions her eggs with mRNA, proteins, hormones, or antibodies, which can control the size, sex, growth, or behavior of her offspring. Mothers can also directly influence their offspring through their own behavioral traits; behaviors like nursing, grooming, predator defense, and “decisions” on when and where to lay eggs can all affect the offspring and its survival.

For example, in many species, including frogs, fish, and mites, traits delaying a mother’s reproduction can result in offspring that hatch later, develop more slowly, and mature at larger body sizes, all of which can promote offspring survival and reproductive success. In this same way, maternal effects may not always be adaptive, as is the case with mayfly mothers who mistake the sheen of an asphalt road for a good, wet place to deposit eggs. Those offspring are not likely to survive not because of their inherent characteristics, but because of their mother’s egg-laying behavior.

Although offspring-and maternal traits obviously interact (for example, feeding behavior by a mother and begging behavior by her offspring), maternal effects are generally thought to be a direct result of the mother’s actions. Considering this definition, indirect effects resulting from the mates a mother chooses, for example, are not maternal effects, but simply yet another way mothers impact their offspring.

Why do maternal effects matter?

Maternal effects can produce meaningful variation in a population upon which selection can act. Selection acting on these traits will naturally affect the fitness of such organisms and therefore the evolutionary dynamics of the population, particularly with respect to the evolution of offspring traits subject to maternal effects and the maternal traits themselves. Additionally, maternal effects may be important for the evolution of adaptive responses (“phenotypic plasticity”) to rapidly changing environments, as they allow for immediate adjustments in phenotype based on the environment. Because those changes are often in both the offspring and mother, multi-generational effects under certain conditions may spread through the population much quicker than favored genetic mutations alone. Conversely, under perhaps less-rapidly changing conditions, time lags in the response to selection could be shown in traits subject to maternal effects because selection in one generation depends on selection of both the current generation and that of the previous. Understanding how and the speed at which populations may respond to environmental change informs our predictions of how populations may change in the future and to understand the interplay between evolutionary and ecological dynamics.

Because mothers impact their offspring through countless other ways (maternal cytoplasmic inheritance, genomic imprinting, and mitochondrial DNA, to name a few), perhaps today is a day to thank your mom for making the majority of you, you. 

Posted in Evolution 101 | Tagged , , , , | Leave a comment

Evolution 101: Adaptive Dynamics Models and Game Theory

Posted on August 27, 2012 by Danielle Whittaker

This week’s Evolution 101 blog post is by MSU undergrad Faisal Tameesh and MSU grad student Emily Weigel. 

Mathematics developed from game theory has been used to study phenotypic evolution, or evolution in the way something “appears” or “behaves.” Particular models, known as adaptive dynamics models, use game-theoretic concepts of frequency-dependence — that the success, or fitness,  of an individual depends on the fitness and abundance of other phenotypes in the population —  to give more ecologically-realistic description for traits that continuously vary. These models link population and evolutionary dynamics to describe the spread (“invasion”) of very small mutations through a population.

To understand how adaptive dynamics models work, imagine a population in which all members share the same phenotypic trait. Let’s call this our “resident” population.  What we will then try to do is to describe mathematically how mutants (“invaders”), whose phenotype is just slightly different from the residents, could invade and spread within the population. What is the outcome of competition, then, between the residents and the invaders? Can they coexist, or will one population exclude the other? From here, scientists try to understand when a population is vulnerable to an invasion, or when an unbeatable strategy is reached — a so-called Evolutionarily Stable Strategy (ESS). These strategies are simply traits which, when the vast majority of individuals express them, no rare mutant with a different trait can invade and increase in numbers. If a successful invasion occurs, the invading population becomes the new resident population, and because evolutionary timescales are much longer compared to the ecological timescales under which the two phenotypes compete, this change is more or less instantaneous. The graphical tool called a Pairwise-Invasibility Plot (PIP) helps us visualize under what conditions a resident wins, an invader wins, or coexistence occurs. The following is an example of what this can look like.

Example Pairwise Invasibility Plot

These dynamics models are also related to the concept of game theory, which is the study of the interaction of intelligent, rational entities, such as humans or intricate computer programs, in an environment that is either natural or designed. The organisms must have some form of memory and a way to make their actions visible to their opponents. The memory is required to keep a log of the opponent’s actions to possibly detect patterns in behavior that can be responded to. Actions must be visible to one another, because, by default, an organism must be able to observe its opponent’s activities in order to come up with actions that will ultimately benefit it.

There are two main aspects of game theory: cooperation and defection. Evolutionary simulations have demonstrated that organisms may evolve to cooperate, so that both organisms benefit; however, these simulations have also shown that, under certain conditions, defection becomes a more feasible approach for an individual to survive in an environment. Whether an organism uses cooperation, defection, or both strategies depends on the environment and the task to accomplish.

Mathematical models can be created, depending on the behaviors of these organisms in terms of cooperation and defection; the organisms’ behaviors can be represented and possible predicted. These models consequently enable humans to gain a better understanding of processes in fields like psychology, economics, and biology. Computer scientists are also able to apply the models to create more challenging opponents in video games in the form of artificial intelligence. Findings like these can also pave the way of enhancing the interactions between intelligent machines and humans in the future. 

Posted in Evolution 101 | Tagged , , | Leave a comment

Evolution 101: Kin Selection

Posted on August 20, 2012 by Danielle Whittaker

This week’s Evolution 101 blog post is by MSU graduate student Sara Garnett.

As soon as spadefoot toad tadpoles are born, the pressure’s on. Adults lay their eggs in temporary ponds, created from sudden rainstorms in the southwestern desert. Because these ponds don’t last long, anything that helps tadpoles to develop and turn into toads before the pond dries can be a huge advantage. Some tadpoles even develop the ability to be carnivorous, since fairy shrimp – and other tadpoles – can be a good source of nutrition when rapid growth is necessary. Despite this environmental pressure, these cannibal tadpoles have often been observed releasing some of their potential tadpole prey after “tasting” them. With so much riding on getting out of the pond, why would tadpoles expend the energy to catch another tadpole, only to give it up?

Researchers noticed this happening in cases where the prey tadpole was a sibling of the cannibal (individuals from the same clutch of eggs can develop both ways); the cannibal released the other tadpole if it tasted a sibling, but it consumed its prey if the other tadpole were unrelated. Such kin discrimination behavior – behaving differently depending on whether the other individual is related or not – seems at odds with a simplified understanding of evolution as an intense struggle to maximize personal success without regard for other members of one’s own species, but it’s more common than you might think. Many species with cannibalistic tendencies have some way of reducing the risk of harming relatives, even if the individual might benefit from a meal. Nestling birds adjust how intensely they beg for food depending on the frequency of half-siblings relative to full-siblings in the nest. Some plants change how they distribute resources to different structures depending on whether they’re surrounded by strangers or siblings. In addition to these examples of restraint shown toward kin, there are also examples of individuals seeming to actively put themselves in harm’s way; certain types of alarm calls that make an animal more likely to be caught by a predator are more likely to be given in the presence of kin.

To understand why these behaviors exist, we need to go back to what it means to succeed in an evolutionary sense. Survival is pretty important, but what really matters is how successful an individual is at passing on copies of its genes. This success is what is described by the term “fitness.” Although there is some debate within the field about how best to define and measure fitness, one determination of it can be based on the number of offspring an individual produces. While it might seem like this would capture how many copies of one organism’s genes show up in the next generation, it ignores the fact that many of those genes are not unique to one individual. If the gene isn’t a new mutation, at least one parent has a copy, meaning that siblings may also have received a copy. Depending on how far back the gene goes, it may also be found in half-siblings, cousins, grandparents, or other relatives as a result of common descent. What this means is that, if we broaden our concept of fitness, helping relatives to survive and reproduce may also increase the chances of passing on genes, an idea known as kin selection.

Kin selection provides a mechanism for how helping relatives may increase an individual’s fitness, in spite of personal costs. Evolutionary biologist W. D. Hamilton laid out a set of conditions, now known as Hamilton’s rule, under which genes governing such behavior should spread in a population. This rule takes into account the genetic relatedness between an actor and the recipient of the behavior (that is, how likely that a given gene is shared between the two individuals due to common descent), the additional fitness benefit the recipient gets as a result of the behavior, and the fitness cost to the actor. If the fitness benefit to the recipient multiplied by the relatedness is greater than the cost to the actor, it will actually ultimately benefit the actor to perform the behavior. This ultimate benefit comes from thinking about an individual’s inclusive fitness, which takes into account not just the direct fitness gained by producing offspring, but also the adjusted fitness benefits gained by relatives who share genes.

Thinking about Hamilton’s rule illustrates how changing each of these parameters might be expected to alter kin-directed behavior. We might expect to see more instances of behavior that helps relatives as well as the individual when the cost to the actor is not very low, or when the benefit to the relative is very great. Individuals should also be willing to incur greater costs if relatedness to the recipient of the action is greater. The chance that a given gene will be identical due to common descent is greater between siblings than half-siblings (1/2 vs. 1/4 on average), which is greater than the average relatedness between cousins (1/8), affecting the inclusive fitness that results from helping different relatives. This relationship is captured well by a quote attributed to evolutionary biologist J.B.S. Haldane, who claimed he “would lay down [his] life for two brothers or eight cousins.”

Giving up a meal to avoid consuming a sibling makes more sense for a spadefoot toad tadpole when inclusive fitness is taken into account. As long as an individual is not in immediate danger of starvation, it receives a fitness benefit of its own by not taking a relative out of the gene pool. Kin selection and inclusive fitness provide some ideas for understanding behaviors that don’t initially seem to make sense from an evolutionary perspective, even if we still don’t always understand how we tolerate our own siblings.

Posted in Evolution 101 | Tagged , , , , | Leave a comment

Evolution 101: Neuroevolution

Posted on August 13, 2012 by Danielle Whittaker

This week we introduce a new occasional blog series: Evolution 101. Enjoy!

If you were to ask a random person what the best example of Artificial Intelligence is out there, what do you think it would be?

Most likely, it would be IBM’s Watson.


In a stunning display of knowledge and accuracy, Watson blew away the world Jeopardy champions Ken Jennings and Brad Rutter without blowing a fuse, and ended with Jennings proclaiming, “I for one welcome our new computer overlords.”

IBM’s Watson represents the current popular approach to AI: that is, spending hundreds of hours hand-coding and fine-tuning a program to perform exceedingly well on a single task. Most people in the field of AI call machines like Watson an expert system because they are designed to be experts at a single task. This approach has been wildly successful lately, producing machines that drive cars and fly UAVs by themselves, beat world chess and Jeopardy champions, and even fool some people into thinking they’re human.

However, imagine how hard it would be to hand-code a system that could do everything the human brain is capable of. Do you think that sounds impossible? That’s the reason why the field of neuroevolution was born: scientists wanted to harness the creative power of evolution to design the programs that could achieve human-level intelligence.

What is Neuroevolution?

Neuroevolution, or neuro-evolution, is a form of machine learning that uses evolutionary algorithms to train artificial neural networks. It is useful for applications such as games and robot motor control, where it is easy to measure a network’s performance at a task but difficult or impossible to create a syllabus of correct input-output pairs for use with a supervised learning algorithm.

-Wikipedia

What does all that mean?

Broadly speaking, the goal of neuroevolution is to evolve an artificial brain with a genetic algorithm to solve a specific task. The artificial brain, oftentimes called the artificial neural network, is designed based on our understanding of how biological brains work. This video does a great job of explaining artificial neural networks:

As the video mentioned, oftentimes the genetic algorithm starts out with a bunch of random artificial brains. The genetic algorithm then emulates the process of evolution:

  1. Fitness evaluation: each of the artificial brains are tested on how well they perform at a task.
  2. Selection: the brains that perform better are chosen to reproduce into the next generation of artificial brains.
  3. Descent with modification: the offspring of those artificial brains are created as copies of their parent brains with slight modifications.

This process repeats over and over until the artificial brains master the task. Here’s an example of an artificial brain being evolved to walk in a two-legged robot. Notice how the artificial brain does a really bad job of walking at first, but eventually learns walk without falling at all.

Why is that useful?

Genetic algorithms have been proven to be a creative and powerful designer.

For example, researchers once used a genetic algorithm to design an antenna for one of NASA’s satellites. The original antenna took months for engineers to design; cost thousands of dollars per antenna; and didn’t even perform as well as NASA had hoped. An entrepreneurial group of researchers at UCSC decided to make an attempt at designing their own version of the antenna with a genetic algorithm, and evolved an antenna that used a single piece of wire that cost next to nothing and performed better than the antenna designed by the engineers.

The same concept applies for evolving artificial brains.

Researchers at UT Austin have evolved artificial brains to control a rocket into space without fins, which is an otherwise extremely difficult problem to engineer. [videos]

Meanwhile, researchers at UCF have evolved artificial brain controllers for two-legged robots that walk and balance all by themselves. [video]

Evolved artificial brains are even being used in video games, such as UT Austin’s NERO video game. [video]

There are plenty more examples of “neuroevolution in action” out there; these are just a few choice examples. Neuroevolution has a promising future of designing intelligent algorithms for robot control, vehicle navigation, and many, many, many more applications.

Neuroevolution and Artificial Intelligence

The real advantage of neuroevolution is what it brings to the development of Artificial Intelligence. In the past, computer scientists working on AI would design an algorithm that would exhibit intelligent behavior, then tweak that algorithm’s parameters until it exhibited “optimal” intelligent behavior. The AI they designed either worked or it didn’t, and oftentimes their results didn’t teach us much about how human brains work.

On the other hand, in neuroevolution, scientists can begin to ask questions about the evolution of human-level intelligence:

  • “What challenges (or set of challenges) were ancient organisms faced with that required them to evolve intelligence to succeed?”
  • “What were the ‘building blocks’ to human-level intelligence?”
  • etc.

Indeed, neuroevolution promises to be an insightful field of study, since scientists can not only attempt to create an artificial intelligence, but also hypothesize about how intelligence was created in the first place. (Which is why neuroscientists and biologists are also interested and involved in this field!)

Written by Randy Olson. Randy is a PhD student in Michigan State University’s Computer Science program. Along with his lab mates in Dr. Chris Adami’s lab, he studies biologically-inspired artificial brains and algorithms with the goal of evolving intelligent behaviors inside of the computer.

Posted in Evolution 101 | Tagged , , , , | Leave a comment