Comprehensive experimental fitness landscape and evolutionary network for small RNA

Abstract

The origin of life is believed to have progressed through an RNA world, in which RNA acted as both genetic material and functional molecules. The structure of the evolutionary fitness landscape of RNA would determine natural selection for the first functional sequences. Fitness landscapes are the subject of much speculation, but their structure is essentially unknown. Here we describe a comprehensive map of a fitness landscape, exploring nearly all of sequence space, for short RNAs surviving selection in vitro. With the exception of a small evolutionary network, we find that fitness peaks are largely isolated from one another, highlighting the importance of historical contingency and indicating that natural selection would be constrained to local exploration in the RNA world.T he nucleotide sequence of an organism's genome determines its fitness in a given selective environment. All possible sequences of length L constitute a discrete sequence space containing 4 L points. Including fitness as another variable creates a "landscape" in sequence space, in which highly fit sequences occupy peaks (1, 2). Evolution can be thought of as a random walk on this landscape with a bias toward climbing peaks (3). Knowledge of the fitness landscape is a fundamental prerequisite for a quantitative understanding of evolution. Although several models of theoretical landscapes have been proposed (reviewed in refs. 4 and 5), there is a lack of empirical data. Landscapes based on RNA secondary structure have been explored computationally, but the relationship to function is unknown (6-10). Experimental efforts at determining a comprehensive fitness landscape are generally stymied by the astronomical size of sequence space, but synthesis of nearly every variant is feasible for relatively short sequences (i.e., RNAs with length L <30 nucleotides). Therefore, complete landscapes for short RNAs could be mapped in principle. Such landscapes are of particular interest for understanding evolution in the RNA world of early life (11)(12)(13)(14).Limited fitness landscapes localized around known functional sequences have been explored for proteins (15-17), viruses (18, 19), and functional nucleic acids, including ribozymes and ribosomal RNA (20-24). The local fitness landscape around a known RNA ligase ribozyme (L = 54) was mapped using highthroughput sequencing (25). However, random sampling of RNA and DNA sequence space can be done by in vitro selection, or SELEX, for de novo discovery of functional molecules (26)(27)(28)(29). Such studies generally take very sparse samples of sequence space, owing to their focus on obtaining functional, and therefore longer, sequences. Thorough sampling techniques have been used to explore all possible DNA targets for a known DNAbinding protein (30). Such studies illuminate the biology of extant organisms but do not address the initial evolution of macromolecular activity. The entirety of a macromolecular fitness landscape has not yet been explored. Therefore, fundamental questions about the shape of fitn...

Computational analysis of fitness landscapes and evolutionary networks from in vitro evolution experiments

“…If the map possesses sufficient (e.g., base-by-base) resolution in sequence space, potential evolutionary pathways can be discovered by identifying networks of sequences that are connected by small mutational steps. The importance of understanding the relationship between functional activity and sequence, together with the potential to study evolutionary issues, provides motivation for constructing and analyzing maps of molecular fitness landscapes [13][14][15]. However, until the advent of high-throughput sequencing techniques, maps of the fitness landscape were difficult, if not impossible, to obtain, owing to the small amount of sequencing information available via low-throughput techniques.…”

“…We have discussed these issues in detail elsewhere [16]. Here, we briefly review these issues and describe the computational methods we have used to reconstruct fitness landscapes from pre-and post-selection frequencies [15] as a guide to potential practitioners. This includes expanded capabilities compared to our previously described platform.…”

“…Thus, the number of molecules that we can experimentally count is much too small to directly measure the true distribution of molecules in a pool of this diversity. Although the pre-selection abundance of an individual 24-mer sequence cannot be directly measured by sequencing, one may calculate a good estimate for its abundance through the creation of a quantitative, semi-empirical model that describes how an initial pool of sequences is synthesized [15,16]. This is necessary since oligonucleotide synthesis is subject to chemical biases (e.g., in coupling efficiency), which prevent true randomness.…”

“…We can thus estimate the pre-selection abundances of sequences in the initial pool. Using model-selection criteria [17], we have found [15,16] that a reasonable model is one that assigns different reactivities to a growing chain depending on the identity of the last two nucleotides and of the incoming nucleotide. This model requires a relatively small number of parameters that are estimated from the sequences' statistics of the pre-selection pool.…”

“…These segments are needed as constant regions, designed as binding sites for PCR primers, or to anneal the sequences to a support for sequencing. We had previously chosen an approach of ligating single-stranded adapter sequences to the randomized N24 sequences, in order to avoid effects from built-in PCR primer regions during the selection [15]. However, this introduces bias from the process of adapter ligation, which is catalyzed by T4 RNA ligase.…”