Predicting second harmonic generation from crystals

Levi M Haupert, Purdue University

Abstract

Determining protein structure is important for identifying drug targets in order to treat human disease. Crystallized proteins are required for the predominant method of structure determination, x-ray crystallography. Because there is no a priori method for crystallizing proteins, a large series of conditions must be tested. Optical second harmonic generation (SHG) microscopy, which is sensitive to chiral crystals, may be employed to speed screening of protein crystallization trials. However, the physical origin of differences in SHG between individual protein crystals remains largely unexplained. The purpose of this research is to explain differences in SHG between protein crystals and to explore applications of polarization dependent analysis of SHG from protein crystals. We approximated the SHG response of proteins by adding together the theoretical response of the protein's amide linkages. Applying this technique to a random sample of proteins from the RCSB protein data bank, we analyzed influences from protein secondary structural motifs, protein size and crystal symmetry. According to our model, crystal symmetry proved to be largest predictor of SHG activity. However, similar sized proteins within crystals of the same symmetry were still expected to exhibit an order of magnitude of diversity in their response. We suspect this diversity is due to both the general structure of the protein and its orientation within the crystal unit cell, but further investigation is still needed. The importance of symmetry for the SHG response from chiral crystals suggests crystal symmetry screening as one possible application for SHG microscopy. This application could be useful for quality control of crystalline pharmaceutical products. In this application, large quantities of semi-randomly oriented crystals of a given molecule must be screened. In order to overcome the complication of random sample orientation, we employed an artificial neural network developed using PyBrain. Training sets were composed of the complete polarization dependent SHG responses from simulated collections of randomly oriented crystals of a particular molecule. Neural networks trained this way were generally 75% accurate at determining the crystal symmetry of simulated, randomly oriented crystals of the molecule of interest. Though the initial results are promising, complete polarization dependent SHG response with phase information is difficult to obtain at this time. However, research to develop an instrument capable of obtaining full phase information is underway.

Degree

Ph.D.

Advisors

Simpson, Purdue University.

Subject Area

Biochemistry|Physical chemistry|Biophysics

COinS