Combined use of NMR and computational tools for fragment based drug discovery targeting protein-protein interactions VEGF protein surface recognition as a case study

Abstract

[spa] En el contexto de la presente tesis hemos abordado los siguientes objetivos: 1. El uso de métodos de RMN, basados tanto en la observación del ligando como en la observación de proteína, para estudiar la unión de los compuestos de una quimioteca a la zona de VEGF involucrada en la unión a sus receptores. La interacción VEGF/VEGFR puede ser considerada como un caso de estudio para la evaluación de las interfaces proteína-proteína mediante cribado de fragmentos. 2. Desarrollar herramientas basadas en la combinación de RMN y métodos computacionales para abordar: i) un sistema automático de diseño de mezclas de fragmentos; ii) el análisis automático de datos procedentes de cribados basados en RMN; iii) la evolución de fragmentos con muy baja afinidad. 3. Explorar el uso de técnicas de “mRNA display” para el descubrimiento de nuevos ligandos peptídicos para VEGF.

[eng] The capacity of proteins to interact with each other rests at the core of biology. Given the ubiquitous nature of these interactions they have attracted the attention of scientists for the development of inhibitors or biochemical tools.The use of biologics to target protein-protein interfaces is relatively advanced; suffer although from some intrinsic drawbacks as the danger of immunogenicity, the inability to cross biological barriers efficiently and high production costs. Small molecule inhibitors do not necessarily share these drawbacks. Unfortunately the druggability of protein-protein interfaces and strategies to target them with small molecules is under open debate. In this work we explore the druggability and methods to target the protein-protein interface of VEGF, a model system with therapeutic relevance in the fields of tumor biology and macular degeneration. We focus mainly on a fragment based approach for the following reasons: i.) proved to be successful at least for some particular protein-protein interfaces, ii.) offers a good coverage of the chemical space with small libraries, iii.) may lead to compounds with improved physicochemical properties compared to HTS. NMR, which is an omnipresent method in the field of fragment based drug discovery since the pioneering work of Fesik, was our tool of choice with a strong focus on combination with novel computational approaches. In the first part of our work we express the required amounts of recombinant VEGF. Then we design and prepared a fragment library after the “SAR by catalog” principle. We developed a new methodology that allowed the preparation of fragment mixtures for NMR based screening with minimized signal overlap. This allowed the direct assessment of nearly all fragment mixtures without the need of mixture deconvolution. Further we developed a program that allowed the automatic evaluation of NMR derived fragment screening data. While being faster than tedious manual interpretation of NMR data it offered a degree of quantitative analysis that would otherwise not be possible in a reasonable amount of time. Our library consistent of over 500 fragments was screened using STD- and CPMG filtered NMR experiments. The analysis of the NMR data resulted in high hit rates but apparent very weak affinity of identified ligands. We successfully developed a competitive 19F NMR based screening assay to identify ligands that bind to the protein-protein interface of VEGF, however none clear competitors could be identified. A second library of over 350 19F containing molecules was screened which led to low hit rates and identification of ligands with apparent very weak affinity. Finally a computational analysis of VEGF surface predicted a low druggability of the protein-protein interface which was in accordance to our experimental observations. The characterization of weak binding fragments and their structural evolution was elaboration was achieved by a combined approach based on NMR and computational experiments. Ligand binding was assessed by the NMR chemical shift perturbation methodology using as probes both amide backbone N-H groups of the protein and its side chain methionine methyl groups. Binding poses were predicted by induced fit docking with the PELE algorithm under strong guidance by NMR derived restrains. Predicted binding modes were used to select fragment analogs with improved binding parameters. This was performed for three cycles and led finally to the discovery of several scaffold families that bind to or in proximity to the protein-protein interface of VEGF. Finally, we present a preliminary exploration of mRNA display for the selection of novel peptide based VEGF ligands.