Are there two distinct research strategies for developing biologically active molecules: rational design and empirical selection?

Exploiting models of molecular evolution to efficiently direct protein engineering

Directed evolution and protein engineering approaches used to generate novel or enhanced biomolecular function often use the evolutionary sequence diversity of protein homologs to rationally guide library design. To fully capture this sequence diversity, however, libraries containing millions of variants are often necessary. Screening libraries of this size is often undesirable due to inaccuracies of high-throughput assays, costs, and time constraints. The ability to effectively cull sequence diversity while still generating the functional diversity within a library thus holds considerable value. This is particularly relevant when high-throughput assays are not amenable to select/screen for certain biomolecular properties. Here, we summarize our recent attempts to develop an evolution-guided approach, Reconstructing Evolutionary Adaptive Paths (REAP), for directed evolution and protein engineering that exploits phylogenetic and sequence analyses to identify amino acid substitutions that are likely to alter or enhance function of a protein. To demonstrate the utility of this technique, we highlight our previous work with DNA polymerases in which a REAP-designed small library was used to identify a DNA polymerase capable of accepting non-standard nucleosides. We anticipate that the REAP approach will be used in the future to facilitate the engineering of biopolymers with expanded functions and will thus have a significant impact on the developing field of 'evolutionary synthetic biology'.

QSAR of multiple mutated antibodies

The aim of this study was to develop predictive quantitative structure-activity relationship (QSAR) modeling for antibody-peptide interactions. A small single chain antibody library was designed and manufactured around the murine anti-p24 (HIV-1) monoclonal antibody CB4-1 by use of statistical molecular design (SMD) principles and site directed mutagenesis, and its affinity for a p24 derived antigen was determined by fluorescence polarization. A satisfactory QSAR model (Q(2) = 0.74, R(2) = 0.88) was derived by correlating the affinity data to physicochemical property scales of the amino acids varied in the library. The model explains most of the antibody-antigen interactions of the studied set, and provides insights into the molecular mechanism involved in antigen binding.

Engineering proteinase K using machine learning and synthetic genes

Background

Altering a protein's function by changing its sequence allows natural proteins to be converted into useful molecular tools. Current protein engineering methods are limited by a lack of high throughput physical or computational tests that can accurately predict protein activity under conditions relevant to its final application. Here we describe a new synthetic biology approach to protein engineering that avoids these limitations by combining high throughput gene synthesis with machine learning-based design algorithms.

Results

We selected 24 amino acid substitutions to make in proteinase K from alignments of homologous sequences. We then designed and synthesized 59 specific proteinase K variants containing different combinations of the selected substitutions. The 59 variants were tested for their ability to hydrolyze a tetrapeptide substrate after the enzyme was first heated to 68°C for 5 minutes. Sequence and activity data was analyzed using machine learning algorithms. This analysis was used to design a new set of variants predicted to have increased activity over the training set, that were then synthesized and tested. By performing two cycles of machine learning analysis and variant design we obtained 20-fold improved proteinase K variants while only testing a total of 95 variant enzymes.

Conclusion

The number of protein variants that must be tested to obtain significant functional improvements determines the type of tests that can be performed. Protein engineers wishing to modify the property of a protein to shrink tumours or catalyze chemical reactions under industrial conditions have until now been forced to accept high throughput surrogate screens to measure protein properties that they hope will correlate with the functionalities that they intend to modify. By reducing the number of variants that must be tested to fewer than 100, machine learning algorithms make it possible to use more complex and expensive tests so that only protein properties that are directly relevant to the desired application need to be measured. Protein design algorithms that only require the testing of a small number of variants represent a significant step towards a generic, resource-optimized protein engineering process.

The rational design of biological complexity: A deceptive metaphor

Biologists often claim that they follow a rational design strategy when their research is based on molecular knowledge of biological systems. This claim implies that their knowledge of the innumerable causal connections present in biological systems is sufficient to allow them to deduce and predict the outcome of their experimental interventions. The design metaphor is shown to originate in human intentionality and in the anthropomorphic fallacy of interpreting objects, events, and the behavior of all living organisms in terms of goals and purposes. Instead of presenting rational design as an effective research strategy, it would be preferable to acknowledge that advances in biomedicine are nearly always derived from empirical observations based on trial and error experimentation. The claim that rational design is an effective research strategy was tested in the case of current attempts to develop synthetic vaccines, in particular against human immunodeficiency virus. It was concluded that in this field of biomedicine, trial and error experimentation is more likely to succeed than a rational design approach. Current developments in systems biology may give us eventually a better understanding of the immune system and this may enable us in the future to develop improved vaccines.

Osmosensing by bacteria

Osmosensors are proteins that sense environmental osmotic pressure. They mediate or direct osmoregulatory responses that allow cells to survive osmotic changes and extremes. Bacterial osmosensing transporters sense high external osmotic pressure and respond by mediating organic osmolyte uptake, hence cellular rehydration. Detailed studies of osmosensing transporters OpuA, BetP, and ProP suggest that they sense and respond to different osmotic pressure-dependent cellular properties. These studies also suggest that each protein has a cytoplasmic osmosensory or osmoregulatory domain, but that these domains differ in structure and function. It is not yet clear whether each transporter represents a distinct osmosensory mechanism or whether different research groups are approaching the same mechanism by way of different paths. Principles emerging from this research will apply to other osmosensors, including those that initiate signal transduction cascades in prokaryotes and eukaryotes.

VeloceGenomics: An Accelerated in Vivo Drug Discovery Approach to Rapidly Predict the Biologic, Drug-Like Activity of Compounds, Proteins, or Genes

PURPOSE

The aim of this study is to test the predictive power of in vivo multiorgan RNA expression profiling in identifying the biologic activity of molecules.

METHODS

Animals were treated with compound A or B. At the end of the treatment period, in vivo multiorgan microarray-based gene expression data were collected. Investigators masked to the identity of the compounds analyzed the transcriptome signatures to define the molecular pathways affected by treatment and to hypothesize the biologic activity and potential therapeutic indications of the blinded compounds.

RESULTS

For compound A, G-protein-coupled receptors and factors associated with cell growth were affected-growth hormone/insulin-like growth factor-1, glucagon/insulin axes, and general somatomedin-like activity. Deblinding showed the compound to be a somatostatin analog, SOM230, confirming the accuracy of the predicted biologic activity. For compound B, components of the inflammatory cascade potentially mediated by lipopolysaccharide, tumor necrosis factor, or proinflammatory cytokines were affected. The gene expression signatures were most consistent with an interleukin-6 family activity. Deblinding revealed that compound B was leukemia inhibitory factor.

CONCLUSIONS

VeloceGenomics is a strategy of coupling in vivo compound testing with genomic technologies. The process enables prediction of the mechanism of action and, coupled with other relevant data, prediction of the suitability of compounds for advancement in the drug development process.

Directed molecular evolution by machine learning and the influence of nonlinear interactions

Alternative search strategies for the directed evolution of proteins are presented and compared with each other. In particular, two different machine learning strategies based on partial least-squares regression are developed: the first contains only linear terms that represent a given residue's independent contribution to fitness, the second contains additional nonlinear terms to account for potential epistatic coupling between residues. The nonlinear modeling strategy is further divided into two types, one that contains all possible nonlinear terms and another that makes use of a genetic algorithm to select a subset of important interaction terms. The performance of each modeling type as a function of training set size is analysed. Simulated molecular evolution on a synthetic protein landscape shows the use of machine learning techniques to guide library design can be a powerful addition to library generation methods such as DNA shuffling.

Putting engineering back into protein engineering: bioinformatic approaches to catalyst design

Complex multivariate engineering problems are commonplace and not unique to protein engineering. Mathematical and data-mining tools developed in other fields of engineering have now been applied to analyze sequence-activity relationships of peptides and proteins and to assist in the design of proteins and peptides with specified properties. Decreasing costs of DNA sequencing in conjunction with methods to quickly synthesize statistically representative sets of proteins allow modern heuristic statistics to be applied to protein engineering. This provides an alternative approach to expensive assays or unreliable high-throughput surrogate screens.

Reductionism and the search for structure-function relationships in antibody molecules

One of the claims of reductionism is that it can explain all the features of living systems by an analysis of their physico-chemical constituents. Such a claim disregards the existence in biological systems of emergent properties that do not exist in their isolated components but which allow autonomous organisms to be directively organized in a self-regulated and integrated manner. It is not possible to describe biological systems adequately without using functional language that is meaningless in the physical sciences. The description of biological functions is also an essential part of immunology and functional explanations are more useful than causal explanations also in this discipline. Since causality is not a relation between a material object and an event, the structure of an antibody cannot be the cause of its binding activity. When structure-function relationships are analysed, the search should be for correlations rather than for causal relations. Methods used to find correlations between the atomic structure of antibody binding sites and their binding activity are mostly based on mutagenesis studies. Since the effect of any mutation depends on the molecular context, it is usually very difficult to predict the effects of multiple mutations on antibody function. Our knowledge of the molecular basis of antigen-antibody recognition has led to the expectation that it may be possible to develop new vaccines using molecular design principles. Such unwarranted hopes arise because of a confusion between antigenicity and immunogenicity. Although knowledge of antibody structure is of little use in vaccine design, it may help to develop therapeutic inhibitors and antibodies effective in the passive immunotherapy of viral infection.

Polysaccharides, mimotopes and vaccines for fungal and encapsulated pathogens

Vaccination is a rational alternative to treatment for Cryptococcus neoformans infections, as these infections are currently intractable in immunocompromised (including HIV-infected) individuals. Vaccines composed of the cryptococcal capsular polysaccharide glucuronoxylomannan (GXM), the key C. neoformans virulence factor, elicit protective antibodies in mice, although deleterious antibodies can also be induced. By contrast, polysaccharides are poor immunogens in HIV-infected humans and others with B-cell defects. Peptide mimotopes of GXM can induce protective immunity to C. neoformans in mice, however, our knowledge of the mechanisms of mimotope-induced protection is incomplete and further work is needed if polysaccharide- or mimotope-based vaccines are to be used to manage C. neoformans infection.

Exploration of sequence space for protein engineering

The process of protein engineering is currently evolving towards a heuristic understanding of the sequence-function relationship. Improved DNA sequencing capacity, efficient protein function characterization and improved quality of data points in conjunction with well-established statistical tools from other industries are changing the protein engineering field. Algorithms capturing the heuristic sequence-function relationships will have a drastic impact on the field of protein engineering. In this review, several alternative approaches to quantitatively assess sequence space are discussed and the relatively few examples of wet-lab validation of statistical sequence-function characterization/correlation are described.

Predicting the kinetics of peptide-antibody interactions using a multivariate experimental design of sequence and chemical space

A multivariate approach involving modifications in peptide sequence and chemical buffer medium was used as an attempt to predict the kinetics of peptide-antibody interactions. Using a BIACORE system the kinetic parameters of the interaction of Fab 57P with 18 peptide analogues of an epitope of tobacco mosaic virus protein were characterized in 20 buffers of various pH values and containing different chemical additives (NaCl, urea, EDTA, KSCN and DMSO). For multivariate peptide design, three amino acid positions were selected because their modification was known to moderately affect binding, without abolishing it entirely. Predictive mathematical models were developed which related kinetic parameters (k(a) or k(d)) measured in standard buffer to the amino acid sequence of the antigen. ZZ-scales and a helix-forming-tendency (HFT) scale were used as descriptors of the physico-chemical properties of amino acids in the peptide antigen. These mathematical models had good predictive power (Q(2) = 0.49 for k(a), Q(2) = 0.73 for k(d)). For the non-essential residues under study, HFT and charge were found to be the most important factors that influenced the activity. Experiments in 19 buffers were performed to assess the sensitivity of the interactions to buffer composition. The presence of urea, DMSO and NaCl in the buffer influenced binding properties, while change in pH and the presence of EDTA and KSCN had no effect. The chemical sensitivity fingerprints were different for the various peptides. The results indicate that multivariate experimental design and mathematical modeling can be applied to the prediction of interaction kinetics.

Is there a role for potassium channel openers in neuronal ion channel disorders?

Malfunction in ion channels, due to mutations in genes encoding channel proteins or the presence of autoantibodies, are increasing being implicated in causing disease conditions, termed channelopathies. Dysfunction of potassium (K(+)) channels has been associated with the pathophysiology of a number of neurological, as well as peripheral, disorders (e.g., episodic ataxia, epilepsy, neuromyotonia, Parkinson's disease, congenital deafness, long QT syndrome). K(+) channels, which demonstrate a high degree of diversity and ubiquity, are fundamental in the control of membrane depolarisation and cell excitability. A common feature of K(+) channelopathies is a reduction or loss of membrane potential repolarisation. The identification of K(+) channel subtype specific openers will allow the recovery of the mechanism(s) responsible for counteraction of uncontrolled cellular depolarisation. Synthetic agents that demonstrate K(+) channel opening properties are available for a variety of K(+) channel subtypes (e.g., K(ATP), BK(Ca), GIRK and M-channel). This study reviews the realistic therapeutic potential that may be gained in a broad spectrum of clinical conditions by K(+) channel openers. K(+) channel openers would therefore identify dysfunctional K(+) channel as therapeutic targets for clinical benefit, in addition being able to modulate normally functioning K(+) channels to gain clinical management of pathophysiological events irrespective of the cause.