An experimental phylogeny to benchmark ancestral sequence reconstruction

Abstract

Ancestral sequence reconstruction (ASR) is a still-burgeoning method that has revealed many key mechanisms of molecular evolution. One criticism of the approach is an inability to validate its algorithms within a biological context as opposed to a computer simulation. Here we build an experimental phylogeny using the gene of a single red fluorescent protein to address this criticism. The evolved phylogeny consists of 19 operational taxonomic units (leaves) and 17 ancestral bifurcations (nodes) that display a wide variety of fluorescent phenotypes. The 19 leaves then serve as ‘modern' sequences that we subject to ASR analyses using various algorithms and to benchmark against the known ancestral genotypes and ancestral phenotypes. We confirm computer simulations that show all algorithms infer ancient sequences with high accuracy, yet we also reveal wide variation in the phenotypes encoded by incorrectly inferred sequences. Specifically, Bayesian methods incorporating rate variation significantly outperform the maximum parsimony criterion in phenotypic accuracy. Subsampling of extant sequences had minor effect on the inference of ancestral sequences.

Non-conservation of folding rates in the thioredoxin family reveals degradation of ancestral unassisted-folding

“…Overall, it is clear that in vitro thioredoxin folding rates are not evolutionary conserved. As we have previously noted (Candel et al, 2017), recent claims to the contrary (Tzul et al, 2017a, b) are probably related to the use of strongly destabilizing solvent conditions which buffer the effect of landscape ruggedness on in vitro folding experiments, as it has been known for many years (Randall et al, 2016) and it is visually apparent in our data of Figure 6a. Note that the folding rates are roughly similar for most thioredoxins at 4M urea, while they diverge as the denaturant concentration becomes lower and the solvent becomes less destabilizing.…”

“…Furthermore, the Bayesian posterior probabilities for the inferred residues in both ancestors at position 74 is 100% (Figure 1b). We have previously shown that such sites rarely, if ever, are incorrectly inferred with such a high posterior probability (Randall et al, 2016).…”

Evolution of chalcone isomerase from a non-catalytic ancestor

“…Therefore, ancCC appears to be correctly folded and stable, suggesting that the lack of enzymatic activity is not due to inference errors. However, we only reconstructed the statistically most probable sequence while in effect, ancestral inference yields a “cloud” of sequences that relate to the historical ancestor 11–13 . Results described below indicate that this “cloud” includes sequences that possess basal CHI activity, but the probability that ancCC’s isomerase activity exceeded low, promiscuous levels is negligible.…”

“…The crude product was further purified by silica gel column chromatography using a hexane:ethylacetate gradient (100% hexane, then 7:1, 3:1, 2:1). 1 H NMR (400 MHz, DMSO- d 6 ): δ 12.56 (s, OH-2’/6’), 10.46 (s, OH-4’), 10.12 (s, OH-4), 7.95 (d, J =15.57 Hz, H-β), 7.64 (d, J =15.57 Hz, H-a), 7.51 (d, J =8.32 Hz, H-2/6), 6.81 (d, J =8.32 Hz, H- 3/5), 5.83(s, H-3’/5’); 13 C NMR (100 MHz, DMSO- d 6 ): δ 191.88, 164.86, 164.54, 159.96, 142.49, 130.50, 126.26, 123.96, 116.14, 104.39, 95.03.…”

Structural, thermodynamic and catalytic characterization of an ancestral triosephosphate isomerase reveal early evolutionary coupling between monomer association and function

“…The sequences obtained by ASR are phylogenetic inferences, as such, rather than producing a single reconstruction of the ancestral sequence this tool provides representative pictures in the statistical ensemble of ancestral sequences . The extent to which the inferred functional, structural or thermodynamic properties of resurrected proteins are robust to uncertainty is an important issue of ASR that was recognized long ago and is currently a foregoing topic of research . In this respect, a recent survey of several protein families showed that although the precise values of quantitative parameters may vary, the qualitative functions of reconstructed proteins seem to be robust to the inherent uncertainty of the process .…”