Variational properties of organisms

Variation among individuals arises from genetic, environmental and intrinsic causes. It is shaped by the size and structure of populations, and by ontogeny. Studying all these things at once is, for most organisms, impractical. I mostly use small, abundant bactivorous invertebrates – nematode worms of the Caenorhabditis genus – which combine many of the experimental advantages of microbes with the complex development of an animal.

I draw information and materials (genetically diverse individuals) from ecology and population genetics, approaches and ideas from statistical and quantitative genetics, and trade the simplification of experimental evolution in the lab for the relative statistical certainty that it can provide. I also am interested in Caenorhabditis ecology and population structure, both in their own right, and as for the greater realism they will bring to future long-term evolution experiments through, for example, laboratory microcosms incorporating migration and environmental heterogeneity.

Read on for details of current and future research. Published work is listed here.

Trait evolution in C. elegans

What determines the capacity for traits to respond to selection?  Beyond the simple fact that variation must exist for adaptation to occur, things are less certain, and the scope of our explanations may be restricted. Do large populations harbouring greater genetic diversity always adapt faster than smaller populations? Do population structure and migration influence adaptation in predictable ways? How predictive are quantitative genetic descriptions of genetic (co)variances of short and long-term selection responses, or, restated, how important are the details of development, genetic architectures and environmental interactions? I am addressing these and other questions using experimental evolution of C. elegans.

Some of this work uses a new genetic resource derived from a long-term evolution experiment. Populations were founded by 16 diverse, fully-inbred wild isolates and evolved under standard lab conditions at moderate population size (Ne = 1000) for many non-overlapping generations (effectively around 250 total). Under these conditions, the fitness of hermaphrodites is precisely the number of viable embryos present at a fixed stage of life (4 days). From a base outbred population, replicate populations were then evolved that varied in (1) the frequency of outcrossing (lots, some, or none) and (2) environment (continued adaptation to standard lab conditions, or adaptation to a moving optimum). In the diagram below, each circle represents an outbred population (each of which is replicated at least three times) and the conditions of experimental evolution (mating system and environment) from the base A140 outbred population are shown, where red arrows represents adaptation to a moving optimum (increasing [NaCl]). See Noble et al., Teotonio et al., Poullet et al. for more details.

The results of all this evolution are replicated, highly recombined outbred populations. Hermaphrodites were sampled from these populations and inbred by selfing to create homozygous lines for genotyping – at present around 800 lines have been genotyped by sequencing. All populations and inbred lines are frozen, and so are a permanent, stable and extensible resource on which to build.

To date, traits have been measured for these recombinant inbred lines (RILs), a major convenience that has been the default approach for genetic mapping resources in many species. Although selfing is the dominant mode of reproduction for C. elegans in the wild, and outbreeding depression is seen among wild isolates, the extent to which the long, shared phase of outcrossing has potentiated inbreeding depression in RILs is unknown. Theory suggests dominance variance for fitness should be preserved under directional selection while additive variance is eroded (empirical results are scarce, and mixed, e.g.), and this is being tested by phenotyping the progeny of intercrosses among RILs.

Genetic architectures

A trait is any thing that is measurable and biologists are naturally fond of measuring anything they possibly can. From an evolutionary perspective, however, there is just one trait that matters: fitness (its mean and variance). An advantage of the experimental evolution scheme is that fitness is clearly defined and, importantly, measurable under conditions identical to those in which evolution has taken place. An advantage of the experimental system is that many organismal and population demographic traits can be measured at scale, and their varied alignments with fitness (phenotypic and genetic correlations) quantified. Traits we have measured for RILs to date include fitness, morphology (worm size and shape), and locomotory behavior.

Size and fertility, as in many animals, are correlated traits. In worms, the connection is in part an unusually direct one; embryos can occupy over half the volume of a gravid mother. Repeatabilities for these traits are relatively high (>0.6), indicating significant heritability (in the vicinity of human height, incidentally). Yet for even a relatively lax false discovery rate of 10%, a total of 0 significant QTL are detected using single marker additive tests. This is not shocking; after hundreds of generations of consistent directional selection alleles with strong additive effects (accounting for more than a few per cent of the variance), if they exist, are likely to have been pushed to extreme frequencies, to the extent that background selection can be overcome. Indeed, variance decomposition and LD score regression support the expected relationship of lower additive variance and higher polygenicity for fertility relative to size and behavioural traits.

Additive genetic variance for these traits is still far from zero, however, which could be due to any of a number of factors, including epistasis. In the presence of random epistasis (unbiased on average with respect to directional effect on a trait), selection response may be prolonged. And in the presence of sign epistasis, where the directional effect of alleles changes over time as the genetic background changes, the response may also be unpredictable. If a genotype-phenotype map features multiple fitness peaks, the optimum found by any one population may be local, subject to chance (meiosis and order effects) and initial conditions (allele frequencies). However this depends critically on the statistical properties of populations and environmentally-conditional genetic architectures. In the case of high polygenicity and finite population size, the fate of most individual causal alleles (Nes << 1) may still be dominated by drift (e.g.). This implies similar trait values can be attained by many different genotypes, and, under a constant environment, initial levels of genetic variance may be predictive of the response to selection. Conversely, in less idealized cases, details of the genetic architecture – polygenicity, interaction structure and life-history effects – may be more important than quantitative genetic variance components.

To know the genetic architecture of a trait, in a population, is to possess a model that predicts the trait value of an unknown individual from genetic data. Such models are of course limited by the quality, quantity and representativity of the data that go into them, and how well modeling assumptions correspond to reality.