Array-based evolution of DNA aptamers allows modelling of an explicit sequence-fitness landscape

Abstract

Mapping the landscape of possible macromolecular polymer sequences to their fitness in performing biological functions is a challenge across the biosciences. A paradigm is the case of aptamers, nucleic acids that can be selected to bind particular target molecules. We have characterized the sequence-fitness landscape for aptamers binding allophycocyanin (APC) protein via a novel Closed Loop Aptameric Directed Evolution (CLADE) approach. In contrast to the conventional SELEX methodology, selection and mutation of aptamer sequences was carried out in silico, with explicit fitness assays for 44 131 aptamers of known sequence using DNA microarrays in vitro. We capture the landscape using a predictive machine learning model linking sequence features and function and validate this model using 5500 entirely separate test sequences, which give a very high observed versus predicted correlation of 0.87. This approach reveals a complex sequence-fitness mapping, and hypotheses for the physical basis of aptameric binding; it also enables rapid design of novel aptamers with desired binding properties. We demonstrate an extension to the approach by incorporating prior knowledge into CLADE, resulting in some of the tightest binding sequences.

Analysis of a complete DNA–protein affinity landscape

“…Detailed descriptions of chemicals and chip synthesis can be found elsewhere (Knight et al 2009). Briefly, each microarray chip possessed 93 311 individual spots at each of which a known DNA sequence was synthesized in situ.…”

“…Each replicate chip consisted of the same sequences randomized spatially. The chips were hybridized with the target protein APC, in 1Âphosphate-buffered saline ( pH 5.5) for 1 h at 378C, imaged and analysed as described previously (Knight et al 2009). The mean scores across both chips were taken as the overall binding scores.…”

“…In contrast to this, a recent paper by Knight et al developed aptamers to the fluorescent protein allophycocyanin (APC) on high-density DNA microarrays, by optimizing affinity using an evolutionary algorithm (Knight et al 2009), the so-called CLADE (closedloop aptamer-directed evolution) method. As a consequence of using the 'on-chip' procedure, the sequence-fitness relationships of over 40 000 aptamers were determined during the optimization process.…”

“…This ethos was exhibited in a study by Warren et al where the interaction profiles of a transcription factor and a small molecule were determined with every permutation of a duplex DNA sequence 8 bp in length (Warren et al 2006). In the CLADE system described above, the binding affinity of the APC aptamer is strongly dictated by the bases furthest away from the microarray chip surface (Knight et al 2009). If these bases are determinant of protein binding, it is feasible that an attenuated aptamer will be as effective as the longer aptamers.…”

Theory and Practice of Optimal Mutation Rate Control in Hamming Spaces of DNA Sequences

“…However, hitherto, these attempts have maintained a continuous view of the landscape space, in contrast to the reality of its discrete domain. Discrete views have typically been restricted to abstracted biological systems, such as aptamer (Knight et al, 2009) or RNA structure evolution, where landscape analogies can be dropped in favour of networks of sequences (Cowperthwaite and Meyers, 2007) which do not lend themselves to consideration of variable mutation sizes.…”