Elsevier

Biosystems

Volume 94, Issues 1–2, October–November 2008, Pages 68-74
Biosystems

Do motifs reflect evolved function?—No convergent evolution of genetic regulatory network subgraph topologies

https://doi.org/10.1016/j.biosystems.2008.05.012Get rights and content

Abstract

Methods that analyse the topological structure of networks have recently become quite popular. Whether motifs (subgraph patterns that occur more often than in randomized networks) have specific functions as elementary computational circuits has been cause for debate. As the question is difficult to resolve with currently available biological data, we approach the issue using networks that abstractly model natural genetic regulatory networks (GRNs) which are evolved to show dynamical behaviors.

Specifically one group of networks was evolved to be capable of exhibiting two different behaviors (“differentiation”) in contrast to a group with a single target behavior. In both groups we find motif distribution differences within the groups to be larger than differences between them, indicating that evolutionary niches (target functions) do not necessarily mold network structure uniquely. These results show that variability operators can have a stronger influence on network topologies than selection pressures, especially when many topologies can create similar dynamics. Moreover, analysis of motif functional relevance by lesioning did not suggest that motifs were of greater importance to the functioning of the network than arbitrary subgraph patterns.

Only when drastically restricting network size, so that one motif corresponds to a whole functionally evolved network, was preference for particular connection patterns found. This suggests that in non-restricted, bigger networks, entanglement with the rest of the network hinders topological subgraph analysis.

Introduction

In biological genetic regulatory networks (GRNs), genes encode proteins and proteins may in turn regulate the expression level of genes. The dynamics of these interactions not only play a key role in development (Davidson, 2001) but also in the ongoing metabolism of all cells during their lifetime (Alberts et al., 2002). These interactions can conveniently be depicted as networks, with the nodes representing genes and the directed edges the regulating proteins. Understanding GRN dynamics is still a hard task and so methods for breaking their complexity down have been proposed. Very influential has been the static structural analysis method of searching for subgraph patterns—so called network motifs “are those patterns for which the probability P of appearing in a randomized network an equal or greater number of times than in the real network is lower than a cutoff value” (Milo et al., 2002).

The main research questions motivating our analysis were: are there significant structural patterns that arise in the course of evolution in GRNs necessary for controlling the realization of particular functionality? Are some motifs more prevalent than others in evolution for particular functions? How unique are the networks that realize particular functionalities and how robust are they?

These questions cannot be adequately answered on the basis of the current data available for natural GRNs; for this type of statistical analysis, many networks with a particular functionality must be compared and currently data are possibly incomplete and only available for networks in, at most, a few organisms. In addition, observing GRN interactions and analysing metabolism in vivo is very difficult and techniques are still in their infancy.

The model, employing a computational evolutionary simulation, has been shown to exhibit some characteristics of natural GRNs Knabe et al., 2006a, Knabe et al., 2006b, Knabe et al., 2008 and has the substantial advantage that all variables can be controlled; also the GRNs are open to “metabolic” inspection. Note that this study refines earlier work (Knabe et al., 2007a) as possible sampling biases are removed through always seeding populations with the exact same random network and also extends it by adding analysis of activatory/inhibitory internal sub-structures of motifs.

Mazurie et al. (2005) compare available data from S. cerevisiae to four species that belong to the same class of hemiascomycetes, finding that “network motifs undergo no special evolutionary pressure as compared to a generic interaction pattern”. Investigating the function of motifs they conclude that in the limited number of cases where enough information was available no specific functional role could be found. As a likely reason they suggest that entanglement with the rest of the network and the higher order context such as post-transcriptional regulation are more important in realizing information processing capabilities.

Artzy-Randrup et al. (2004) warn against drawing too far-reaching conclusions from motif analysis as one might be using an incomplete null model, i.e. randomization of network data could introduce a bias in the analysis as it does not conserve spatial or other constraints to network topology. They show this by the example of a “Gaussian toy network” modeling the neural-connectivity map of C. elegans, where neurons are more likely to be connected the closer they are to each other. This model, without any action of selection, produces networks showing motif distributions similar to the ones found by Milo et al. (2002) in C. elegans.

Kuo et al. (2006) also employ a GRN model and analyse static network topology. However the networks included in their systematic analysis were randomly created (focusing on whole genome duplication) without being exposed to evolution under selective pressure—and therefore were not comparable to that case and could not be lesioned to check for impact on function. The authors stress that methods of network construction, especially duplication and divergence events, could well explain motif distributions observed in natural regulatory networks. Similarly, Ho (2007), van Noort et al. (2004) as well as Cordero and Hogeweg (2006) all devise mechanisms for the creation of networks structures. The generated topologies are very similar to those of real GRNs, without any selection for function or other sophisticated mechanisms needed.

Very recently Isalan et al. (2008) where the first to systematically explore the effect of adding new links to the Escheria coli’s GRN in vivo. They find that network entanglement makes it often difficult to find functional modules: “Also, our results indicate that partition of a network into small modules (negative feedback, feedforward, and so on) could in some cases be misleading, as the behavior of these modules is affected to a large extent by the rest of the network in which they are embedded.”

Section snippets

Methods

First we present a short introduction to our GRN model followed by a description of the evolutionary algorithm and the evolutionary conditions compared. A description of the topological measure used on networks completes this section.

Results

In the following we compare the population subgraph patterns for the original, differentiation and random evolution conditions. Every population consists of 80 individuals, each of them the best-performing individual from generation 1000 (or 500 where mentioned) of one independent run, i.e. there were 80 independent runs per condition.4

Discussion

One might expect to find a switch motif to control differentiation when the requirement to differentiate was the only difference between the two evolutionary niches (target functions). That is, one might expect that motifs reflect evolved function. However our results show this view may be too naive—there was no convergence on the same single motif or a small set of switching motifs, and uniqueness of motifs was not observed. Instead a wide variety of network patterns and topologies was found.

References (19)

  • B. Alberts et al.

    Molecular Biology of the Cell

    (2002)
  • U. Alon

    An Introduction to Systems Biology—Design Principles of Biologial Circuits

    (2006)
  • Y. Artzy-Randrup et al.

    Comment on network motifs: simple building blocks of complex networks and superfamilies of evolved and designed networks

    Science

    (2004)
  • O. Cordero et al.

    Feed forward loop circuits as a side effect of genome evolution

    Mol. Biol. Evol.

    (2006)
  • E.H. Davidson

    Genomic Regulatory Systems: Development and Evolution

    (2001)
  • D.E. Goldberg

    Genetic Algorithms in Search, Optimization, and Machine Learning

    (1989)
  • S.J. Gould et al.

    The spandrels of San Marco and the Panglossian paradigm: a critique of the adaptationist programme

    Proc. R. Soc. Lond. B: Biol. Sci.

    (1979)
  • R.T. Gregory

    The Evolution of the Genome

    (2004)
  • Ho, J.W.K., 2007. Modeling the Evolution of Gene Regulatory Networks. Presentation at The 8th International Conference...
There are more references available in the full text version of this article.

Cited by (36)

  • Self-selection of evolutionary strategies: adaptive versus non-adaptive forces

    2021, Heliyon
    Citation Excerpt :

    Research into functional network models, focused on the performance of a network rather than the non-adaptive forces that form it, are meant to understand how function influences network structure, and have used a variety of functions models which include electronic logic circuits [8], neural networks [29], and networks that exhibit oscillatory behavior [6, 30]. The algorithms used to identify potential functional network structures are not always evolutionary, which would ignore the influence of non-adaptive processes altogether, and those which are suffer shortfalls by evolving networks using fixed mutation rates [29, 31, 32] or fixed network size [6]. In addition to the non-adaptive network models and functional network models, the third set of relevant models look at the influence of selective pressure on mutation rate.

  • Drivers of structural features in gene regulatory networks: From biophysical constraints to biological function

    2015, Physics of Life Reviews
    Citation Excerpt :

    Note that degeneracy is very common in genotype to phenotype maps and arises in particular in the maps associated with GRNs [186]. More generally, the fact that different motifs can lead to the same dynamical behavior has been demonstrated in other in silico contexts [187–190], with recent example in evolutionary genetics [191,192]. Furthermore, the inverse is also true: one motif may allow for a number of different dynamical behaviors and thus phenotypes.

  • When one model is not enough: Combining epistemic tools in systems biology

    2013, Studies in History and Philosophy of Science Part C :Studies in History and Philosophy of Biological and Biomedical Sciences
    Citation Excerpt :

    This part of the research can be described as highly hypothesis-driven, and the fact that the network motifs were found in a variety of different organisms and in different networks (see Alon, 2006) further supported the experimental realization of the new epistemic entity. Due to the recent critique, network motifs can, however, still be considered as question-generating machines—as partly stable epistemic entities that produce new questions for further experimental analysis (Knabe et al., 2008; Mazurie et al., 2005). To conclude I want to emphasize a strong parallel between Rheinberger’s account and existing accounts of modeling within philosophy of science.

  • Computational Intelligence in the Design of Synthetic Microbial Genetic Systems

    2013, Methods in Microbiology
    Citation Excerpt :

    While the motif concept is appealing, the underlying assumption that these biological modules can be combined in arbitrary ways has not yet been tested in vivo. Consequently, there is some debate about whether genetic motifs actually represent units of selection and therefore whether they act in vivo as they are designed to do in silico (Artzy-Randrup, Fleishman, Ben-Tal, & Stone, 2004; Hallinan & Jackway, 2005; Ingram et al., 2006; Knabe, Nehaniv, & Schilstra, 2008). Despite these drawbacks, small functional modules have been the subject of considerable research in synthetic biology (Paladugu et al., 2006; Hallinan & Wipat, 2007; Purnick & Weiss, 2009).

View all citing articles on Scopus

This paper substantialy extends and supersedes (Knabe et al., 2007b).

View full text