in Artificial Life VIII, Standish, Abbass, Bedau (eds)(MIT Press) 2002. pp 227{232 1 Evolution of Stable Ecosystems in Populations of Digital Organisms Tim F. Cooper and Charles Ofria Center for Microbial Ecology Michigan State University, East Lansing, MI 48824 Abstract continued use will diminish and organisms able to use un- derutilized resources will be favored. Over time density- Competition for resources has long been believed to be dependent selection created by this process can lead to fundamental to the evolution of diversity. However, the difficulty of working with natural ecosystems has meant adaptive radiation and even speciation as organisms di- that this theory has rarely been tested. Here, we use the verge to into distinct lineages (Schluter 2001). Avida experimental platform to demonstrate the evolu- In this study, we use digital organisms (self-replicating tion of ecosystems composed of digital organisms. We and evolving computer programs) to examine the di- show that stable ecosystems are formed during evolu- versity arising in populations during adaptation to an tion in an environment where organisms must compete for limiting resources. Stable coexistence was not ob- environment containing nine distinct resources. (We served in environments where resource levels were not use the term population to refer collectively to all or- limiting, suggesting that competition for resources was ganisms in the environment.) We compare the result responsible for coexistence. To test this, we restarted of adaptation when resources are in infinite supply to populations evolved in the resource-limited environment when resources can be depleted. Because there are but increased resource levels to be non-limiting. In this environment, ecosystems previously supporting multiple no limiting resources in the former environment, only genotypes could maintain only a single genotype. These one niche is present; the organism most efficiently us- results demonstrate the utility of the Avida platform for ing the most valuable resource combination will always addressing ecological questions and demonstrate its po- have a selective advantage and be able to outcompete tential in addressing questions involving ecosystem-level all other organisms (Kassen, et. al. 2000; Hardin 1960; processes. Tilman 1982). In contrast, in the depletable resource environment, all resources can be limiting, potentially Introduction creating as many niches as resources and thus allow- One of the most striking features of life is its diver- ing as many species as resources to coexist (Bell 1997; sity. However, the selective pressures which have led to Huisman 1999). this diversity have not yet been comprehensively identi- We show that stable ecosystems evolved in popula- fied (Morin 2000; Schluter 2000; Tilman 2000). In part, tions grown in environments where organisms competed the lack of progress has been due to the inherent dif- for depletable resources. Diversity was maintained for ficulty of performing precise, replicated and controlled many thousands of generations when the mutation rate experiments on whole ecosystems (Morin 2000). In an was set to zero but rapidly disappeared when we changed effort to ameliorate these limitations, some researchers the environment to make resource levels infinite. have turned to laboratory microcosms (reviewed in Trav- isano & Rainey 2000). However, even in these model sys- The Experimental System tems, ecosystems can evolve to be much more complex The Avida platform is a computer software system in than is experimentally tractable and identification of the which digital organisms evolve by means of random mu- causes of diversity can still be difficult (Notley-McRobb tation and natural selection (as opposed to artificial & Ferenci 1999; 1999; Rainey & Travisano 1998). selection, as in genetic algorithms) (Ofria, Brown, & Competition between organisms for limited resources Adami 1998)1. The genetic code of the digital organ- has long been suspected to play a key role in the struc- isms in Avida is a Turing complete programming lan- turing of communities (Tilman 2000; Schluter 1996; guage. To replicate, a digital organism performs a self- Tilman 1982). Though proof of this idea has been de- analysis that determines information about its genome, ceptively difficult to come by (Schluter 1996; 2000), the 1 reason for its enduring appeal is obvious: if organisms The Avida software is available from deplete resources as they use them, the benefit for their http://nemus.dllab.caltech.edu/avida/ 2 in Artificial Life VIII, Standish, Abbass, Bedau (eds) (MIT Press) 2002. pp 227{232 it then allocates enough extra memory to hold its off- The genome of the ancestral organism in each run was spring, and copies its genome line by line into that extra 100 instructions in length. This organism could self- space before finally dividing off the offspring as an in- replicate but could not perform any logical computa- dependent organism. This replication process is subject tions. The descendants of this organism rapidly filled to two types of errors: copy mutations, where a random up each population (to its maximum capacity of 2500). instruction is written in place of the ancestral one and in- During the first 100,000 updates2 of evolution the copy sertion/deletion mutations, where a random instruction mutation rate was set to 0.0075 per instruction and the is added or removed from the genome. insertion/deletion rate was set to 0.05 per genome. After In Avida, the success of an organism depends upon its the first 100,000 updates we turned off all mutations and growth rate. This is determined by an organism's repli- let the population replicate for an additional 50,000 up- cation efficiency and by its interactions with the environ- dates. This latter phase allowed us to identify genotypes ment. An Avida environment determines the mutation able to stably coexist as a result of ecological interactions rates the organisms are subjected to, as well as the col- without the complications of new genotypes being intro- lection of resources that they can use to increase the rate duced through mutation. Hereafter, we call the 100,000 at which they execute the instructions contained in their updates of evolution with mutation turned on the \evo- genomes. To \metabolize" a resource and garner a ben- lution phase" and the subsequent 50,000 updates with efit from it, an organism must perform a mathematical mutation turned off the \ecology phase". computation specified by the experimenter. Resources We evolved a total of 69 populations; 30 in a lim- are globally available to all organisms. In previous ver- ited resource chemostat environment, and 39 in an en- sions of Avida, resource levels were infinite, such that the vironment where all resources were infinitely abundant. number of organisms using that resource did not affect In both environments 9 resources were present. These its level or the benefit gained for its use. In this study, could be \metabolized" by performing logical computa- we add the capability of feedback between the use of a tions. The full set of computations rewarded were: Not, resource and its level, such that the use of a resource Nand, And, OrNot, Or, AndNot, Nor, Xor, and Equals. by one organism lowers the availability of that resource To complete one of these computations, the organisms to all others. A low concentration of a resource will re- must input one or two 32-bit numbers, perform the logic duce the benefit that can be gained by performing its operation in a bitwise fashion on them, and output the associated computation. This model of resource compe- correct 32-bit result. The ability to perform one of these tition is analogous to that occurring during competition computations (and therefore use the associated resource) in bacterial chemostats. increases the speed at which an organism executes its in- In Avida the experimenter determines the environ- structions, allowing it to replicate more often. The out- ment that populations will evolve in, but does not spec- put of a computation must be exact{a single bit in error ify which organisms are favored. Instead, the outcome disallows the organism to use the resource in question. of the evolutionary process is determined by popula- To determine the genotypic diversity present in popu- tion genetic processes, that is, natural selection and lations we used the Shannon-Weaver index, calculated as th genetic drift. For this reason, Avida populations are − P pi ln(pi), where pi is the frequency of the i geno- genuinely evolving systems, rather than being \simula- type in the population. The same formula was used to tions" of evolution. Avida allows perfect knowledge of measure phenotypic diversity except substituting the fre- the genetic, phenotypic and ecological state of the sys- quency of phenotypes (based on the logic tasks each or- tem at all times and ensures that no measurement in- ganism could perform) in a population for the frequency fluences the course of evolution. These attributes have of genotypes. Mann-Whitney rank sum tests were used enabled Avida to be used to address fundamental evolu- to compare measures of diversity between the two treat- tionary questions in ways not possible with traditional ments because diversity values were non-normally dis- systems (Lenski, et. al. 1999; Adami, et. al. 2000; tributed. Wilke, et. al. 2001). Results: Ecosystem Evolution Materials and Methods We measured the effect of competition for resources The experiments reported here were performed using on genotypic and phenotypic diversity present in the version 2.0beta of the Avida software. Unless otherwise evolving populations. If competition for limited re- noted, each replicate evolving population was started sources creates multiple niches, we expect diversity to with identical initial conditions but with a different ran- be higher in resource-limited populations. We found dom number seed. This seed causes runs to differ at 2 An update is a unit of \real time" for the organisms points where a random choice is made, such as the oc- during which 30 × population size instructions are executed currence of mutations or the location of an offspring in globally.