Public Policy in the Age of Big Data

Public policy needs to find a way of protecting peoples’ health records while still allowing the biomedical research community access to medical records for the potential identification of new diagnostic and therapeutic products. What Dr. Erlich’s work has shown is that restricting the information available on public genomic data does not ensure anonymity of an individual’s genome. Further restrictions would only lower the value of the data provided to the biomedical research community. The solution to this problem is to shift the burden of protecting the anonymity of the data from the repository to the user.

• The identity and intent of the users of publicly available genomic data needs to be recorded with the data repository.
• The users of the data need to be identified using protocols that are the best practices in the security industry, beyond the standard email confirmation. Out-of-network identity confirmation or the use of devices that remain in the control of the user should be employed.
• The user authentication should be done every time a user requests information from a public medical repository.

Potential Benefits to the Individual Supplying Their Genomic Sequences

One of the guiding principles of research on humans is that there must be a potential benefit to the individuals participating in the research. Researchers using public genomic data should try to identify potential benefits to the individuals whose information they are using. This information could be returned to the individual through the genomic repository.

Marketing based on an Individual’s Genome

Prescription drug marketing is already regulated by the Food and Drug Administration (FDA). What if a company wanted to market to people that have a particular genetic signature? Should they be able to use public or private genomic data to identify these individuals, and market to them directly, or to their physicians? I would think that some individuals would want this and others would find it invasive. There needs to be a way of allowing individuals to opt-in to contact by various organizations through the repository. Once again we see that the repository is acting as an intermediary between the individual and organizations analyzing their genome.

Expansion Genomics Repository Responsibilities

What I’m suggesting is that genomic repositories become hubs that manage more than the depositing individual’s data. Instead of allowing anyone access to the data, repositories should ensure that they know who is requesting data and for what propose. Finally, the repositories should distribute information about the potential benefits of new technologies to the individual whose information they have been entrusted with.

Identifying Individuals Methodology

Dr. Ehlich’s group started from public sequence data of an individual. From the supporting meta data they were able to obtain:

1. The individual’s year of birth
2. sex
3. state of residence

They developed a short tandem repeat (STR) profiler² that allowed them to identify the Y-chromosome STR of male individuals, and fed this information into a surname search on the genetic genealogy sites.

Having the most probable surname, year of birth and the state of residence they used publicly available web services, such as USearch.com and PeopleFinders.com to locate the best potential matches, and did subsequent followup studies using public resources to confirm their results.

Of the 1000 Y-chromosome STR data sets Dr. Ehlich’s group examined, they were able to “completely” identify almost 5% of the individuals.

Reference

1. Identifying Personal Genomes by Surname Inference. Gymrek M., McGuire A.L., Golan D., Halperin E., Erlich Y. Science. 339, 321-4.
2. lobSTR: A Short Tandem Repeat Profiler for Personal Genomes. Gymrek M., Golan D., Saharon S., Erlich Y. Genome Research 22, 1154-1162.

Colabrativ’s Role in Small-Angle X-ray Scattering Experiments

Colabrativ’s role in this work started at the conceptual level during discussions with the Biogen Idec scientists about structural approaches to compare the FVIII and rFVIIIFc proteins. Crystallization was an obvious choice, but due to the flexible linker between the FVIII and the Fc in rFVIIIFc, crystallization of the rFVIIIFc was considered to have a low probability of success. SAXS had it own challenges, i.e. obtaining aggregate free samples of both proteins at reasonable concentrations. Early on while evaluating SAXS as an analysis option, we contacted Dr. John Tainer, who put us in touch with Dr. Susan Tsutakawa at SIBYLS beamline 12.3.1 at the Advanced Light Source (ALS). We made the contractual arrangements with the ALS and worked with the Biogen Idec staff to prepare and submit a successful SAXS proposal. We worked with both the Biogen Idec and ALS staffs on the preparation and shipping of the protein to the ALS. Working with Dr. Tsutakawa, we purified both proteins at the ALS by size-exclusion chromatography, prepared in a range of concentrations, and collected the SAXS data on the samples and their corresponding filtrates all in the same day.

Analysis of Factor VIII and Factor VIII-Fc SAXS Data

The analysis of the SAXS data and model refinement was performed by Dr. Tsutakawa beginning with initial models supplied by Biogen Idec.

The Guinier plots of the SAXS data for FVIII and rFVIIIFc showed both samples to be monodisperse; i.e. they have linear Guinier plot slopes. The radii of gyration (Rg) for FVIII and rFVIIIFc from the Guinier plots are 38Å and 51Å, respectively. The Electron Pair Distributions have a Dmax of 125Å and 175Å for FVIII and rFVIIIFc, respectively.

Models of FVIII and rFVIIIFc used the crystallographic structures of Factor VIII (3CDZ.pdb) and Fc as starting models. The program BILBOMD was used to identify a small set of conformations for these starting coordinates that best match the experimental SAXS data. Missing loops and carbohydrate chains were added to improve the fit to the SAXS data. For FVIII, the Chi value improved from 3.5 to 1.8 after modeling the missing loops and optimizing the conformation with BILBOMD, and to 1.5 after modeling the missing carbohydrate structures; see Figure 1.

Figure 1
Modeling of the FVIII Structure in Solution

The molecular dynamics simulation program BILBOMD was used to model conformers of BDD FVIII and compare their theoretical X-ray scattering curves to experimental scattering data. Three models are shown: The crystal structure of BDD FVIII (3CDZ.pdb – Chi=3.5), BDD FVIII with loops (Chi=1.8), and BDD FVIII with addition of N-linked carbohydrate (Chi=1.5). The fit of the three FVIII models to the experimental SAXS data is shown in the left panel.

The rFVIIIFc BILBOMD model has two conformations contributing to the best fit of SAXS data when only the missing loops were added. Unlike the FVIII modeling, the fit to the SAXS data did not improve when the carbohydrate structures were modeled and optimized with BILBOMD; see Figure 2.

Figure 2
Modeling of the rFVIIIFc Structure in Solution

BILBOMD was used to generate a minimal ensemble of structures for which theoretical X-ray scattering curves optimally fit experimental scattering data. Two models are shown: The rFVIIIFc model based on the BDD and Fc crystal structures with surface loops but lacking N glycans has a (Chi=1.5), and the same model with of N-linked carbohydrate present (Chi=1.6). The fit of the two rFVIIIFc models to the experimental SAXS data is shown in the upper-left panel.

Conclusion

The Fc domain in rFVIIIFc extends away from the regions in FVIII that are essential for interaction with elements of the Xase complex (anionic phospholipids, Factor IXa, and Factor X) and von Willebrand factor. This result is consistent with the conclusions from the H/DX study that shows that the fusion of Fc to FVIII does not perturb the structure of either the FVIII or Fc elements of rFVIIIFc.

Structural Studies of the 26S Proteasome

Stefan Bohn, Max-Planck Institute of Biochemistry, Martinsried, Germany

 
Image from the “Molecular Architecture of the 26S Proteaseome Holocomplex” produced at Max Planck Institute of Biocehmistry, Department of Molecular Structural Biology, showing the arrangement of the non-ATPase regulatory proteins.

Dr. Bohn presented their work on the the 26S Proteasome. The proteasome degrades unfolded proteins that have been tagged with multiple ubiquitins. The 45 nm long by 20 nm wide rod shaped proteasome is built up of approximately 35 protein subunits. The poly-ubiquinated proteins are recognized by the Regulatory protein non-ATPase 10 (Rpn-10) near one end of the proteasome and peptides come out the other end. Crystal structures of many of the proteasome proteins have been solved, including that of the proteolytic core particle (CP) which has 7-fold rotational symmetry. The overall structure of the S. pombe at approximately 8.4 Å resolution was obtained using cryoelectron microscopy and single particle analysis. A variety of techniques to apply constraints between the subunits of the proteasome, including different labeling techniques of different proteins during the Cryo-EM studies, residue-specific chemical crosslinking, and proteomics methods. They created a beautiful video showing the showing the structure and how it is assembled, is coming soon to their Protease website.

Molecular Simulation Methods for Modeling Macromolecular Behavior in Vivo

Adrian Elcock, University of Iowa

 
Snapshot of a cytoplasm in a Brownian dynamics simulation courtesy Adrian Elcock, University of Iowa.

Dr. Elcock’s group has been working on methods to simulate large multi-component systems, up to and including whole cells. Given the computational limitations, his group has reduced the number of parameters in these large systems by grouping residues into a single pseudo-atom or treating proteins as rigid bodies. This has allowed them to take larger time steps, as large as 10-20 ns, a million times longer than what is normally used in a full-atom dynamic simulation. Dr. Elcock has posted several of these simulations on YouTube including his presentation on Biological Diffusion and Brownian Dynamics Brainstorm 2, and Cytoplasm Full Energy Model.

Visualizing and Interacting with the Molecular Cell

Art Olson, The Scripps Research Institute

 
ePMV PyAutodock screen shot courtesy Arthur Olson, The Scripps Research Institute.

Dr. Olson showed the state-of-the-art molecular and cellular graphics that his group has been developing, and demonstrated some of the tools he uses as a researcher and and a educator. Embedded Python Molecular Viewer (ePMV) is an open-source Python plug-in that runs molecular model tools inside of professional 3D animation applications. He demonstrated many interactive interfaces, including the docking of an inhibitor into the active site of a protein using PyAutodock/cAudodock. As he pushed the inhibitor into the site, residues that were interacting with the inhibitor would move out of the way in real time. You can view an ePMV-created video of the Fantastic Voyage Part 1: AutoFill on a synaptic vesicle with placeholder proteins and mock bilayer on YouTube.com.

Current State of High-Speed Atomic Force Microscopy: Its Advantages and Limitations

Toshio Ando, Kanazawa University, Kanazawa, Japan
High-speed Atomic Force Microscopy (AFM) is one of the few methods that allows us to see the dynamic motion of these supramolecular assemblies. This technique is able to visualize the motions on the subsecond to sub-100 millisecond time scales. Dr. Ando showed a high-speed AFM movie of myosin V “walking” on an actin filament, the stochastic nature of the myosin V step duration and the consistent forward progress of myosin V along the actin filament. The myosin V/actin system is ideal for these studies because myosin orients itself in such a way that it can be easily seen using the top-down view of the high-speed AFM technique. Several of these movies can be obtained from Kodera, N. et al. Video Imaging of walking myosin V by high-speed atomic force microscopy. Nature 486, 72-76 [2010]) supplementary material.