Testing for adaptive signatures of amino acid alphabet evolution using chemistry space

In recent years, chemistry space has revolutionized the search for pharmaceutically relevant molecules (e.g. [1–4]). The utility of chemistry space as a concept, however extends far beyond the pharmaceutical industry and is here used to investigate adaptive properties of the genetically encoded amino acids. The fundamental point of chemistry space is to assign molecules with numeric values that define specific aspects of their physical and/or chemical attributes. This simple step transforms a collection of unique molecules into a set of points in multi-dimensional space, which are therefore amenable to powerful visualization and quantitative analysis. Key to this transformation is replacing conceptual properties (such as “hydrophobicity”), with precisely defined, measurable molecular descriptors that quantify them (e.g. LogP).

Testing for ‘internal’ adaptive properties of the genetically encoded amino acids involves two steps. First, any such test must define an appropriate chemistry space for the amino acids. This requires careful selection of molecular descriptors that accurately depict amino acids in terms of their relationships with one another and their roles within proteins. Given an appropriately defined chemistry space, it becomes possible to perform quantitative tests for the adaptive logic of the genetically encoded amino acids. For this second step, we test the three hypotheses outlined in the introduction. The first test probes the concept of early amino acids, asking whether this specific subset of 10 amino acids distinguishes itself relative to other possible subsets of the standard amino acid alphabet. The second test complements the first by turning to assess the idea of late amino acids, asking whether this subset amounts to a literal expansion of chemistry space. The third and final test places these two investigations into a broader, unified context by asking whether the early and late amino acids make adaptive sense when considered alongside a third component: dimers constructed from the early amino acids.

Test 1: are the early amino acids an adaptive subset?

Our hypothesis predicts that if the 10 amino acids designated as “early” were indeed used by some earlier genetic code as a protein-building alphabet, they should exhibit unusual qualities as a subset that are analogous to the unusual properties detected for the entire set of 20 genetically encoded amino acids. In particular, the concept of amino acid coverage was used previously to measure the adaptive value of an amino acid alphabet in terms of its protein building potential [5]. Coverage considers both the range of values covered within a particular descriptor and how evenly these values are distributed within that range (Figure 1). An adaptive set of amino acids comprises members evenly distributed across a broad range for key physicochemical properties of size (molar volume), charge (Isoelectric Point) and hydrophobicity (LogP). Such a set of amino acids can combine within an evolving protein to approximate any suite of properties required by shifting environmental conditions.

A simple way to test the adaptive value of the early amino acids is therefore to take the 20 amino acids and ask: if we were to select 10 amino acids at random and measure their coverage in size (Molar Volume), hydrophobicity (ACD LogP) and charge (Isoelectric point), then how often would these random subsets exhibit a better coverage of amino acid chemistry space than that of the “earlies”? We therefore wrote a script to measure the coverage of the 10 early amino acids and of 1 million random samples of 10 amino acids chosen randomly from within the twenty encoded amino acids. This script was run separately for each dimension of amino acid chemistry space and then for each possible combination of 2 and 3 dimensions simultaneously.

Test 2: do the late amino acids expand the chemistry space of the early amino acids?

If the “late” amino acids were added to expand the universe of genetically encoded proteins, then we might expect them to be associated with some measurable expansion of amino acid chemistry space. In other words, we may predict that the late amino acids lie further from the earlies than would be expected for an arbitrary division of the amino acids.

To test this idea, we wrote a further script that first calculated the mean of the cluster of early amino acids for a given molecular descriptor and then measured the distance between this mean and each of the lates as a summed total distance (Figure 2). It then replicated this measurement 1 million times, each time randomly designating 10 of the 20 amino acids as early and 10 as late in order to record how often the true late amino acids show greater expansion from the earlies than occurs by chance. As with Test 1 above, this second test was performed for each individual dimension of amino acid chemistry space, and for combinations of 2 and 3 dimensions simultaneously. As an additional test of the robustness of our results, we repeated all calculations having first removed Glycine and Alanine, which, owing to their unique structural simplicity, can arguably be considered outliers rather than meaningful degrees of freedom of amino acid possibility space.

Test 3: do the late amino acids fill an adaptive gap?

Our general adaptive hypothesis is that the late amino acids were added to the genetic code by natural selection because they expanded the protein-building chemistry space available to some simpler precursor of the standard genetic code. Test 2 therefore considers whether the late amino acids populate new points in chemistry space that were unavailable to the early amino acids. However, amino acids do not act in isolation: they are polymerized to form proteins. This implies that the biologically relevant chemistry space of the early amino acids also includes the chemistry space of their dimers, trimers, etc. For the late amino acids to be adaptively advantageous, they must not only expand the chemistry space of the early amino acids, but also do so in such a way that they are performing a novel role; that is, one not already fulfilled by existing amino acids or their oligomers. Within one dimension, early amino acid dimers and late monomers overlap considerably in range for size and hydrophobocity (Figure 3). Any adaptive separation must therefore be found in combinations of these properties.

In order to test this, we first defined an area around all points in chemistry space (late amino acid monomers and early amino acid dimers) to represent the region of chemistry space populated by each molecule. This area was calculated as a circle centered on each point with a radius equal to the average distance between all points (see Figure 4). Using the designation of late amino acids and early-dimers, we measured the frequency with which dimers were found to overlap with the chemistry space of late amino acids. We then randomized the designation of late amino acids and early dimers and measured equivalently how often dimers were found to share chemistry space with late amino acids. This allowed us to measure the overlap in chemistry space between these two sets of molecules compared to what could be expected by chance and to therefore quantify the novelty contributed by the late amino acids.

Our aim here was to apply the concept of chemistry space in order to corroborate or refute previous claims for adaptive properties of the set of 20 genetically encoded amino acids. Whereas previous claims tested the genetically encoded amino acids against a background of plausible alternatives that were never, as far as we can tell, incorporated into genetic coding, our focus here was on the internal logic of the 20 genetically encoded amino acids, building from the premise of early versus late amino acids.

Our first test asked whether the 10 amino acids that are thought to have formed a simpler, earlier genetic code exhibit, as a set, similar adaptive properties as have been recorded for the full amino acid alphabet. Our results show a weak adaptive signal for any individual dimension of chemistry space. However, if we accept that amino chemistry space becomes more meaningful for protein structure and function as it is measured in two or three dimensions, then our results provide strong support for this notion of an internal adaptive logic to the early amino acids. In terms of their coverage of chemistry space, the early amino acids prove to be a highly unusual subset of the twenty genetically encoded amino acids. This is consistent with the inference that, at some point before the emergence of the standard genetic code, they could have constituted a cohesive and functional alphabet. In the same way as the standard 20 were shown to be adaptively advantageous compared to a broader pool of alternatives [5], the “early” amino acids also appear to cover chemistry space exceptionally well.

Our findings complement a body of recent research spanning multiple approaches that suggests the early amino acids contain enough chemical information to form a coherent functional set. This includes a study that suggests the set of early amino acids are sufficient to enable protein folding by reducing the amino acid composition of a protein while maintaining its foldability [10]. We draw attention to the unusual terminology of this paper, which is consistent with similar efforts by others but misleading to those outside the field. The bold claim that a functional protein has been made entirely from early amino acids actually refers to a protein sequence that comprises 80% early amino acids, with the remaining 20% of the sequence drawing from the full alphabet of 20 amino acids. More straightforwardly, our results also agree with a study that used phylogenetic analysis of amino acid compositional bias in ribosomal proteins to conclude that “at a more primitive state, the code would still contain a similar diversity of physiochemical properties” [21]. A third study that examined modern protein sequences deficient in late amino acids of functional significance (the basic amino acids Lysine and Arginine) also supported the idea of “small proteins without basic amino acids performed important functions in the prebiotic chemistry of early Earth” [22].

Test 2 asked whether the 10 amino acids that are thought to have been later additions to the genetic code represent a quantifiable expansion of amino acid chemistry space. Here we see a weak signal that the late amino acids lie further in chemistry space from the earlies than expected by chance. It appears, however, that most of this effect is coming from the dimension of size. In other words, the later additions to the standard amino acid alphabet differ from the early amino acids primarily in that they were larger. Considering that the early amino acids include the smallest L-alpha amino acids that are chemically possible, this is perhaps not surprising. Nevertheless, so long as we accept that size is an important component of an adaptive chemistry space, then these results support an adaptive explanation: the late amino acids show a literal expansion of two and three-dimensional chemistry space. This assumption is bolstered by the results of the same test performed with Glycine and Alanine removed, where we still find that, in almost 90% of trials, the true late amino acids show a greater expansion in chemistry space than would be expected by chance. Our findings again are in general agreement with previous inference drawn from a different approach, where the late amino acids are believed to have been advantageous precisely because of their “unique specialized” structure [21].

Our third test offers clarification for the otherwise somewhat ambiguous results of Test 2. If the concept of early versus late amino acids is correct, then it implies that all dimers comprising two early amino acids were present before any late amino acid was incorporated into the genetic code. We therefore tested whether the expansion of amino acid chemistry space brought about by the addition of late amino acids was steered by an additional consideration: avoiding overlap with regions already occupied by dimers made from the earlies. In accordance with this adaptive hypothesis, we find a strong signal that the lates indeed filled a gap in chemistry space that was not already occupied by the early amino acids or their dimers.

This third test represents a qualitative expansion of thinking about adaptive amino acid chemistry space in that it is the first time molecules larger than monomers have been considered. Indeed, it is largely thanks to the availability of a free database, Chemspider, which includes relevant molecules and their key descriptors that these measurements were made possible. In this context, it is unfortunate that the molecular descriptor isoelectric point is not readily available for dimers, but this implies a simple, logical future step to verify or undermine our current interpretation.

Another clear question raised by our analysis is whether other molecular structures might have complemented early amino acids to form the functional chemistry space in which the genetic code emerged. Many variations of the RNA-world model predict that metabolism during the time of genetic code evolution would have involved a heavy presence by enzymatic cofactors featuring nucleobases and their derivatives [23]. Indeed, it has been specifically proposed that “aromatic amino acids could have been selected for as RNA “replacements” in a dying RNA world, mimicking nucleotides in important structural roles (e.g., stacking interactions)” [21]. We therefore considered visually a simple 2-dimensional plot of the chemistry space analyzed in test 3 (that is the chemistry space of the earlies, lates, and early dimers) in order to add some obvious examples of molecules that suggest themselves to this way of thinking (Figure 7). To our best inspection, nothing here suggests an obvious complementary role for these cofactors, which may well indicate the need of more sophisticated, higher-dimensional analysis.

Here we present three, logically independent tests that each generates quantitative evidence to corroborate or challenge previous claims for adaptive properties of the standard amino acid alphabet. Each test operates in terms of a simple, 3-dimensional chemistry space built from amino acid charge (pI), size (ACD Molar volume), and hydrophobicity (ACD LogP). Whereas previous claims focus on the external chemistry space of amino acids that are not part of the genetic code, we turn to consider the internal logic of the 20 genetically encoded amino acids. In particular, we consider the surprisingly strong, multidisciplinary consensus that has emerged in recent years to suggest that the genetic code may have begun with only half the 20 amino acids currently found in the standard genetic code. This division of the standard amino acid alphabet into “early” and “late” amino acids allows us to make three predictions based upon an adaptive hypothesis: (i) the “early” amino acids should form a cohesive sub-set with similar properties to the final, full-sized amino acid alphabet; (ii) the “late” amino acids should demonstrate quantifiable expansion of amino acid chemistry space in terms of dimensions that define protein-building potential, and (iii) the expansion of “late” amino acids should populate regions of chemistry space that were not already available to a genetic code that build dimers from the early amino acids. Taken together, our results provide strong support for these predictions.

In order to implement our methods, we wrote our source code in Java version 6. We then ran it on Mac OS X Version 10.6.8. Source code is available upon request to the corresponding author.