Structural biology
Visualizing heterogeneous structural ensembles
We are actively working on new computational frameworks to extract ensembles of heterogeneous structures from single particle cryo-EM datasets. Our first application of these ideas produced cryo-DRGN (Deep Reconstructing Generative Networks), which has garnered great interested from the cryo-EM community, and was recently published in Nature Methods. CryoDRGN employs a series of neural networks arranged as a variational autoencoder to 1) embed single particle images in a low-dimensional latent space, and 2) reconstruct 3D density maps from data-occupied regions of this latent space. Together, these networks efficiently classify the particles in a continuous space, which allows us to visualize continuous motions as well as discrete compositional changes. Critically, in contrast to many approaches, one need not specify the number of classes expected nor the types of motion expected.
We can use the maps cryoDRGN generates to identify new structural states that help us understand how molecular machines work. Moreover, we can generate movies of the molecules transitioning from one state to another by traversing along the latent space data manifold, generating structures along the way. Such movies help us understand how structural dynamics facilitate assembly or function.
In related work, we are developing analysis tools to interrogate and interpret the overwhelming number of distinct 3D maps that tools like cryoDRGN can generate.
More importantly, cryoDRGN has it’s own emoji - ❄🐉!!
You can download and try the software yourself here.
in-situ structural biology
We have extended our machine-learning based methods to complexes imaged directly in cells, and this work was recently published in Nature Methods. Using this technique, we have now begun analyzing ribosome assembly directly in bacterial cells and, in collaboration with the Mosalaganti Group, have developed a rapid processing workflow to push the throughput of microbial cryo-electron tomography. Combined, these techniques allows us to wide ensembles of conformational changes ribosomes undergo in their native cellular environment. I describe our work in this area during a recent SBGrid talk, and you can see an exemplar analysis below. This work was funded with generate support from the Sloan foundation - learn more here. See an example of our application of the tomoDRGN in a rapid processing workflow below.
model COMPLEXES FOR STRUCTURAL STUDIES
To guide our development of the computational tools described above, we are also building software to simulate highly heterogeneous single particle cryoEM datasets. This tool, which we call cryoSRPNT is implemented in Python, is based a seminal description of how one could model noise in single particle cryo-EM datasets by Baxter et al.
Our tool allows users to 1) convert an atomic model to a 3D density map; 2) generate noiseless 2D projections of this 3D density map from desired projection angles; 3) corrupt the particle image with white Gaussian noise and the contrast transfer function using desired noise and CTF parameters. The resulting images can then be used to test the efficacy of 3D reconstruction algorithms.
In related work, we are also building physical complexes with defined structural heterogeneity, which we can use to benchmark our 3D reconstruction tools. More coming soon!
AFFINITY-EM and grid support systems
In collaboration with Dave Thompson’s group at Purdue University, we are working on new methods to affinity purify protein complexes of interest directly on cryo-EM grids. These reagents would allow users to directly isolate an appropriately tagged protein complex from a cell lysate and and to determine the ensemble of conformations it adopts. In preliminary experiments, we have successfully isolated an intact complex from a E. coli.
In related work, we have adapted methods from others to deposit graphene and graphene oxide monolayers on cryoEM grids to aid in concentrating and supporting our protein complexes during the vitrification process. This work is published as a very nice JoVE protocol you can find here and here.
Mass spectrometry
Complex compositional analysis by mass spectrometry
Using quantitative mass spectrometry, we routinely determine the composition of macromolecular complexes we’ve isolated or assembled in vitro.
This work relies on application of a standard complex of known composition that was purified from media bearing mass-distinguishable isotopes. We typically collect “DIA/SWATH” data, and use a series of in-house scripts to carefully (and quantitatively) determine the relative abundance of each subunit in the complex.
When looking at many (10s-100s) of subunits, it sure beats a western blot!
Pulse labeling mass spectrometry
We have continued to develop and apply methods to couple pulse-labeling to quantitative mass spectrometry, with the goal of carefully monitoring synthesis and degradation rates of individual proteins across the proteome.
In recent studies, we have applied these methods to bacterial, yeast, and human cells cultured in vitro to understand the relative contributions of various proteostasis pathways, and to identify novel targets for this protestasis machinery. Some of this work is now on bioRxiv here and here, and the underlysing datasets we generated can be found here.
