Thread by @yoginho

More on networks and their influence on robustness & evolutionary dynamics in today’s “Beyond Networks” lecture (ufind.univie.ac.at/en/course.html…). Slides available here: drive.google.com/open?id=1kfA7R…. @univienna

Evolution requires evolving systems to be robust and evolvable. Non-robust phenotypes cannot be found by evolutionary search, non-evolvable phenotypes cannot produce adaptive variation. This poses an apparent paradox as robustness seems to be the opposite of evolvability. /1

We can start to resolve the robustness/evolvability paradox with Maynard Smith’s wonderful little 1970 Nature paper on “protein spaces.” nature.com/articles/22556…. /2

Maynard Smith notes that in order for proteins to be evolvable “...functional proteins must form a continuous network which can be traversed by unit mutational steps without passing through nonfunctional intermediates,” anticipating notions of neutral or genotype networks. /3

Maynard Smith ends his classic little paper by laying out a research programme that keeps us busy to this day. /4

In the 1990s, Schuster et al. @univienna demonstrate the existence of neutral networks for secondary structures of RNA, and show how these networks govern the probability of evolutionary transitions to new structures. This work is reviewed here: onlinelibrary.wiley.com/doi/full/10.10…. /5

Schuster et al.’s insights are not confined to RNAs. Andreas Wagner shows in his “Origins of Evolutionary Innovations” (global.oup.com/academic/produ…) that they also occur in metabolic and gene regulatory networks. /6

In the gene regulatory case: many network configurations (genotypes) produce the same patterning output (phenotype). Few phenotypes are produced by a large number of genotypes. These frequent phenotypes cluster tightly in genotype space (see figure): journals.plos.org/ploscompbiol/a…. /7

Two wonderful papers by Ciliberti, Martin & Wagner in 2007 establish a number of important facts about robustness and evolvability.

journals.plos.org/ploscompbiol/a…
pnas.org/content/104/34…

/8

Networks producing the same phenotype are connected into large components of a meta-graph (a graph of graphs) by single mutational steps (adding or removing an interaction, for example). This is called a genotype network (equivalent to neutral networks from the RNA example). /9

Robust networks tend to evolve from non-robust ones, simply because more mutations cause individuals in a population to “fall off” their genotype network if that network is not large & well connected. Robustness becomes a by-product of neutral network evolution. /10

Genotype networks span a vast area of genotype space. They allow a population of individuals to traverse and explore this space by neutral network drift. /11

The identity of neighbouring genotype networks depends on where in genotype space you are. Therefore, the neighbours at far ends of the genotype network will have very little in common. In this way, large genotype networks allow a population to “discover” novel phenotypes. /12

Taken together, this indicates that the resolution of the robustness/evolvability paradox requires us to distinguish these concepts at the genotype and phenotype levels. rspb.royalsocietypublishing.org/content/275/16… /13

Genotypic robustness anti-correlates with genotypic evolvability (the more robust, the less likely a mutation is to alter the current phenotype). /14

In contrast, phenotypic robustness creates genotype networks which allow a population to explore vast regions of genotype space, thereby discovering novel phenotypes. Phenotypic robustness and evolvability are positively correlated! /15

Phenotypic robustness is a consequence of canalisation or structural stability in the underlying developmental processes that generate the phenotype. It is the structure of the underlying configuration space that explains the existence of genotype networks. /16

Understanding this underlying dynamic structure is crucial for us to understand the probability of evolutionary transitions—and therefore the evolvability—of a regulatory system. /17

One particular manifestation of network drift is developmental system drift. onlinelibrary.wiley.com/doi/abs/10.104… /18

Developmental system drift is predicted to be a huge pain in the ass for evo-devo. It implies that even closely related homologous traits may be produced by non-homologous processes that have drifted apart, causing us to misclassify numerous potential cases of true homology. /19

At least 2 potential mechanisms exist to explain network drift: (1) (Pseudo-)compensatory evolution via “molecular stepping stones” (much like Maynard Smith’s “protein spaces; Haag, 2007), (2) or compensatory evolution in response to selection on a (mildly) pleiotropic trait. /20

Selection, pleiotropy, compensation (SPC, Pavlicev & G. Wagner) postulates that beneficial mutations at mildly pleiotropic loci must outweigh the negative pleiotropic effects, which then induce selective sweeps for compensatory mutations. ncbi.nlm.nih.gov/pubmed/22385978 /21

Evidence of a (pseudo-)compensatory mechanism for network drift comes from our own study on the gap gene system in dipteran insects. Changes in maternal inputs are compensated by mutations altering the dynamics of downstream gene interactions. @KoralWotton @antoncrombach /22

However, both neutral and selective (SPC) compensatory mechanisms could apply in different evolutionary contexts. /23

The take-home message is that understanding evolvability requires understanding the properties of the underlying configuration space of the evolving developmental system. /24

Since many mechanisms produce the same phenotype (albeit maybe in different ways), we need to move away from ever-more detailed molecular explanations towards a systems level understanding of evolving networks at the level of their configuration space. /25

Next lecture: the role of the environment. See you again next week Monday! /26