An isomorphous replacement method for efficient de novo phasing for serial femtosecond crystallography

Abstract

Serial femtosecond crystallography (SFX) with X-ray free electron lasers (XFELs) holds great potential for structure determination of challenging proteins that are not amenable to producing large well diffracting crystals. Efficient de novo phasing methods are highly demanding and as such most SFX structures have been determined by molecular replacement methods. Here we employed single isomorphous replacement with anomalous scattering (SIRAS) for phasing and demonstrate successful application to SFX de novo phasing. Only about 20,000 patterns in total were needed for SIRAS phasing while single wavelength anomalous dispersion (SAD) phasing was unsuccessful with more than 80,000 patterns of derivative crystals. We employed high energy X-rays from SACLA (12.6 keV) to take advantage of the large anomalous enhancement near the LIII absorption edge of Hg, which is one of the most widely used heavy atoms for phasing in conventional protein crystallography. Hard XFEL is of benefit for de novo phasing in the use of routinely used heavy atoms and high resolution data collection.

Figure 1. Phase quality (map CC) as functions of the numbers of the native and the derivative indexed patterns.

(a) SIRAS case. (b) SIR case. The success and failure of phasing are represented as circular and triangular symbols, respectively. The CC of electron density maps (defined in the main text) is indicated by colors. Some data points are missing because SHELXE failed to trace map. The figure was prepared using R with ggplot2 package.

Figure 2. Isomorphous difference Patterson map.

The v = 1/2 Harker section of the isomorphous difference Patterson map calculated using indexed patterns from 10,792 native and 10,000 Hg derivative LRE crystals. The Patterson map was calculated using CCTBX functionality. This figure was prepared using R with ggplot2 package.

Figure 3. The SIRAS electron density map produced by SHELXE with a refined model of LRE.

10,000 and 10,792 patterns of derivative and native crystals were used for the calculation. The electron density map is contoured at 1.0σ. The figure was prepared using PyMOL.

Figure 4. The atomic structure of LRE determined by SIRAS method.

The model was refined against the SFX native data. The figure was prepared using PyMOL.

Figure 5. Effect of the high resolution cutoff on SIRAS/SIR phasing.

The indexed patterns of 10,792 native and 10,000 Hg derivative LRE crystals were used for this calculation. The color and symbol scheme are the same as in Fig. 1. It should be noted that the phasing was successful with lower resolution data when more patterns were used. The figure was prepared using R with ggplot2 package.

Figure 6. CC_ano and CC_anoref as a function of pattern numbers used for Monte-Carlo integration.

CC_ano was calculated using two data sets grouped randomly. CC_anoref is defined as in the main text. The figure was prepared using R with ggplot2 package.

Figure 7. Anomalous difference Patterson maps.

The v = 1/2 Harker section is calculated using different numbers of indexed Hg derivative patterns. When 10,000 patterns are used, a peak of 3.2σ is observed at the Hg position. There are, however, other peaks higher than this in the Harker section. When 20,000 and 30,000 patterns are used, Hg peaks with heights of 5.0σ and 6.9σ, respectively, are observed and they are the highest peaks in the sections. The Patterson map was calculated using the CCTBX functionality. The figure was prepared using R with ggplot2 package.