Innovative Structural Biology


How does cross-linking and mass-spectrometry (XL-MS) work?
Cross-linking and mass-spectrometry is an emerging experimental approach that provides structural information on protein complexes that are currently inaccessible by X-ray crystallography. The idea is to cross-link the intact complex under native conditions with a short bi-functional cross-linking reagent (a). After the covalent cross-linking occurred, the complex is denatured and digested with a protease (b). The digest is analyzed by mass-spectrometry (c), and pairs of residues that underwent cross-linking are identified (d). Such analysis can identify hundreds of cross-linked sites on a large complex. The cross-links are converted to distance constraints and drive computational modeling (e).


Studying the architectures of very large protein complexes with XL-MS
We recently used XL-MS to solve the architectures of two central eukaryotic complexes. The CCT chaperonin (16 subunits; 1 Mega-Dalton) is essential for the correct and efficient folding of certain cytosolic proteins. The transcription pre-initiation complex (32 subunits; 1.5 Mega-Dalton) assembles before every round of transcription to correctly position the RNA polymerase II enzyme on the DNA. The lab is employing XL-MS to expand these architectures by identifying new subunits that interact with the complexes, and by obtaining more cross-links to refine the models. We have also started to probe the molecular structure of the neuronal synapse with XL-MS.

Structural modeling with XL-MS data
Our lab is developing new computational ways to effectively convert XL-MS data into useful 3D models. The current focus is on coarse-grained modeling, where each subunit is represented by a small set of points. We are working to merge this level of modeling with other structural modalities such as SAX and cryo-Electron Microscopy.

Modeling with XL-MS

Getting more out of low-resolution X-ray crystallography data
In collaboration with the lab of Michael Levitt at Stanford University, we are trying to find new computational ways to use low-resolution crystallographic data more effectively. We have recently showed that crystallographic data sets with resolutions as low as 5.5Å can single out the correct structural model out of millions of alternative models. This finding, which directly relies on the mathematical fit between the model and the data, opens the way to new uses of low-resolution crystallography.

Combinatorial Homology Modeling