Experimental illumination of a fitness landscape

Abstract

The genes of all organisms have been shaped by selective pressures. The relationship between gene sequence and fitness has tremendous implications for understanding both evolutionary processes and functional constraints on the encoded proteins. Here, we have exploited deep sequencing technology to experimentally determine the fitness of all possible individual point mutants under controlled conditions for a nine-amino acid region of Hsp90. Over the past five decades, limited glimpses into the relationship between gene sequence and function have sparked a long debate regarding the distribution, relative proportion, and evolutionary significance of deleterious, neutral, and advantageous mutations. Our systematic experimental measurement of fitness effects of Hsp90 mutants in yeast, evaluated in the light of existing population genetic theory, are remarkably consistent with a nearly neutral model of molecular evolution.

Fig. 1.

EMPIRIC approach to experimentally determine fitness landscapes. Randomized individual codon libraries are introduced into a host cell whose only other copy of the gene is regulatable. The fitness of each individual codon mutation is determined by measuring its abundance in the mixed culture as a function of time under selective conditions.

Fig. 2.

Hsp90 region analyzed and application of selection pressure to point mutants of Hsp90 in yeast. (A) Positions 592–600 are highlighted in yellow in the dimeric structure of S. cerevisiae Hsp90. (B) Growth of an Hsp90 temperature-sensitive yeast strain at 36 °C is rescued with a wild-type Hsp90 plasmid. (C and D) Deep sequencing analysis of a library of single-codon mutants of Hsp90 from amino acids 582–590 grown in mixed culture. (C) Relative abundance of wild-type sequence as a function of time in selective conditions where the only other copy of Hsp90 is inactivated. (D) The ratio of TAG stop codons to wild-type codons decreases steeply over time in selective conditions. (E) Observed fitness of leucine synonyms at positions 582, 583, and 584.

Fig. 3.

Amino acid profile in phylogenetic alignment poorly predicts EMPIRIC fitness profile. (A) Heat map representation of the EMPIRIC fitness profile with the wild-type amino acids outlined in red. Information content logos generated from amino acids with WT-like EMPIRIC fitness (B) and a phylogenetic alignment of 448 Hsp90 protein sequences (C). (D) The dominant genetic code is optimized for single-base substitutions between codons with WT-like fitness compared with randomly simulated codes (+2.4σ). (E) Distribution of tolerated and phylogenetically observed amino acids expressed as an entropy where zero corresponds to a frozen position and 3 corresponds to unrestrained positions. (F) Relationship between tolerated amino acid profile from EMPIRIC fitness measurements and phylogenetic alignment. Linear regression indicates a very weak correlation with R² of 0.15. (G) EMPIRIC fitness analyzed as a function of amino acid prevalence in the phylogenetic alignment. Most amino acids observed in the phylogenetic alignment are well-tolerated when made in the yeast homolog.

Fig. 4.

Distribution of fitness effects of mutations from population genetic models and EMPIRIC measurement. For each model, the relative abundance of each type of fitness effect is illustrated by the length of the bar segment, and whereas the expected proportions of mutations in each class are unknown, the EMPIRIC approach provides an experimental measurement of class occupancy. The experimentally measured selection coefficients were binned in 0.05 increments.