Nature © Macmillan Publishers Ltd 1997

Genome sequencing projects have generated a flood of informationabout the molecular basis of life. As

the catalogue of genes has grown, so too hasour understanding of gene function. Forexample, of the 4,288 protein-coding genes inthe bacterium Escherichia coli, 62 per centhave been assigned to a functional class1. Yetwe remain quite ignorant about how thesegenes work together to create a whole organism. Understanding how genes interactis important in its own right but also has profound implications for evolutionary biology2.

On page 395 of this issue3, Elena andLenski provide the clearest picture to date ofthe ways in which genes, randomly chosenfrom throughout the genome, interact toaffect fitness. They inserted a known numberof transposable elements into random posi-tions of the E. coli genome, and calculatedgrowth rate over a six-day period for bacteriawith different numbers of mutations. If thefunction of a gene relies on a complex web ofinteractions, the effect of a mutation on fit-ness will depend on which other mutationsare present. In this case, evolutionists say thatthere is ‘epistasis’ between the genes, becausethe effects of mutations are not independentof one another. In these experiments, how-ever, the authors found that, on average, eachadditional mutation slowed the rate of cell division by roughly the same amount,suggesting that epistasis was not present.

Just as one settles into the comfortableinterpretation that random mutations affectsuch different aspects of cell performancethat they are effectively independent, Elenaand Lenski demonstrate that there are inter-actions between many pairs of mutations. Inan additional set of experiments, they creat-ed recombinant E. coli with different pairs ofmutations and compared them to parentalstrains carrying each of the mutants sepa-rately. In 14 out of 27 cases, the fitness of thedouble mutant was significantly different

from that expected from the product of thesingle-mutant fitnesses (the probability ofobserving so many significant differences is<10–11). No epistasis was observed on aver-age because the pairs of mutations amelio-rated each others’ effects about as often asthey exacerbated them. This result is remark-able and has a number of implications.

With interactions between fully half of thepairs of mutations, we must re-evaluate theimportance of gene interactions. Why woulda pair of genes randomly chosen from agenome be sensitive to each other’s action?This result cannot be explained by the existence of protective mechanisms thatbreak down as more mutations accumulate(mutations would then only exacerbate oneanother’s effects). Perhaps enzymatic path-ways and regulatory networks are longer,more interconnected and more sensitive tostructural changes in the cell than is oftenappreciated.

Indeed, this explanation is consistent withan experiment in Drosophila melanogaster,where characteristics such as protein, glyco-gen and fat content, and body weight andenzyme activities, were examined in linescontaining transposable-element insertions4.Epistatic interactions were rampant, withinsertions often affecting several characteris-tics. It is also consistent with the observation3

of great variability in the form of epistasis be-tween pairs of genes, because different epista-tic relationships are expected to occur withindifferent enzymatic pathways5. The availabil-ity of the entire genome of E. coli1 shouldmake it possible to identify the genes affectedin such experiments and, for many of them,their putative function. We will then be ableto determine more precisely the nature ofepistatic interactions.

The observation that genetic interactionsare common but highly variable will have aneven greater impact on the evolutionary theory of sex and recombination. One of themore popular theories for why sexual repro-

duction is ubiquitous, despite being a riskyand costly mode of reproduction, is that sexallows deleterious mutations to be eliminat-ed more efficiently from a population (themutational deterministic hypothesis6,7). Sexonly facilitates the elimination of deleteriousmutations, however, when fitness decreasesfaster than predicted with each additionaldeleterious mutation, a phenomenon calledsynergistic epistasis (see Fig.1; refs 8–10). Ifeach additional mutation has less and less ofan impact on fitness (antagonistic epistasis),then sex actually hinders the elimination ofdeleterious mutations. The fact that epistasiswas not predominantly synergistic in thenew experiments3 therefore contradicts themain requirement of the mutational deter-ministic theory for sex. Even if synergisticepistasis were present, sex and recombina-tion need not evolve. Variance in the extentof gene interactions, as found by Elena andLenski3, reduces selection for sex and recom-bination (see Fig. 1; ref. 9). So the absence ofsynergistic epistasis and the great variance inepistatic interactions observed are doubleblows to the mutation-elimination hypoth-esis for sex and recombination.

One might object that, after all, E. colibacteria rarely exchange genetic material11

and may not be subject to the type of geneticinteractions typical of sexual organisms.There is, however, no obvious reason whygenetic interactions in sexual eukaryotesshould necessarily differ from those in E. coli— all cells must perform housekeeping tasksand use similar, often homologous, genes to do so. Perhaps sexual eukaryotes haveevolved elaborate protective mechanismsthat minimize the effects of a few mutationsbut fail under the burden of many mutations(resulting in synergistic epistasis). Or per-haps sexual eukaryotes face different meta-bolic demands compared with bacteria, such that the balance between enzyme path-ways is altered and synergistic epistasis ismade more common5. Studies similar toElena and Lenski’s, but in a number of differ-ent species, are needed to determine thenature and variety of genetic interactions —information that is of fundamental impor-tance to both molecular and evolutionarybiology.Sarah P. Otto is in the Department of Zoology,University of British Columbia, Vancouver, British Columbia, Canada V6T 1Z4.e-mail: otto@zoology.ubc.ca

1. Blattner, F. R. et al. Science 277, 1453–1474 (1997).

2. Whitlock, M. C., Phillips, P. C., Moore, F. B.-G. & Tonsor, S. J.

Annu. Rev. Ecol. Syst. 26, 601–629 (1995).

3. Elena, S. F. & Lenski, R. E. Nature 390, 395–398 (1997).

4. Clark, A. G. & Wang, L. Genetics 147, 157–163 (1997).

5. Szathmary, E. Genetics 133, 127–132 (1993).

6. Kondrashov, A. S. Nature 336, 435–441 (1988).

7. Lyons, E. J. Nature 390, 19–21 (1997).

8. Barton, N. H. Genet. Res. 65, 123–144 (1995).

9. Otto, S. P. & Feldman, M. W. Theor. Pop. Biol. 51, 134–147 (1996).

10.Redfield, R. J., Schrag, M. R. & Dean, A. M. Genetics 146, 27–38

(1997).

11.Selander, R. K. & Levin, B. R. Science 210, 545–547 (1980).

news and views

NATURE | VOL 390 | 27 NOVEMBER 1997 343

Evolutionary genetics

Unravelling gene interactionsSarah P. Otto

Figure 1 Fitness versus the number ofdeleterious mutations. The main figureshows fitness as a function of the number ofdeleterious mutations under conditions ofsynergistic, multiplicative and antagonisticepistasis. The inset shows the strength ofselection favouring an increased amount ofsex as a function of the variance in epistasis(calculated from ref. 9), assuming that, onaverage, there is synergistic epistasis(necessary for sex to be advantageous). Sexand recombination are able to evolve onlywhen there is little variability in the amountof epistasis (shaded area).

1.00

Fitn

ess

0.95

0.90

0.85

0.8043210

106

1 06

0

Sel

ectio

n fo

r se

x

0.0002 0.0004 0.0006

Variance in epistasis

Number of mutations

AntagonisticMultiplicativeSynergistic