Protein-protein interactions play critical roles inside and outside the cell, maintaining structure, enabling motility, regulating signal transduction and assembling molecular machines. An outstanding goal in biology is to understand how protein interactions are determined by sequence and structure. Work in the Keating laboratory is focused on how interaction specificity allows some proteins to select one or a few interaction partners out of a large number of alternatives. This is especially important for paralogous gene families and for other protein interactions mediated by structurally conserved motifs. 

Studies in the lab are focused on alpha-helical coiled coils and Bcl-2 family proteins because of their biological and medical importance, and because questions of specificity can be addressed most powerfully in model systems that are well characterized. We combine computational and experimental approaches to understand, predict and design protein interaction specificity. We also develop new computational and experimental methods to address problems in this area.


Coiled coils BCL-2        



Computational Modeling and Design

Coiled coils

The alpha-helical coiled coil is an extremely common interaction motif found in proteins with many different biological functions. Coiled-coil structures consist of two or more helices that wrap around one another with a superhelical twist. Their sequences are characterized by a seven amino-acid repeat, (abcdefg)n. The a- and d-position residues are predominantly hydrophobic and the e- and g-position residues are often polar or charged. Their simple sequence patterns and symmetrical structures make coiled coils amenable to computational modeling. Short lengths and autonomous, reversible folding make some examples very experimentally tractable as well. A rich body of literature dating to the 1950s reports many sequence/structure/function relationships for coiled coils, as well as an abundance of important roles in biology. Areas of particular interest to the lab are highlighted below.

Interaction specificity

An important, unanswered question about coiled coils is how interaction specificity is encoded in sequence. Coiled-coil helices can form a variety of complexes of differing topology, including parallel and antiparallel dimers, trimers, tetramers, pentamers and even a reported heptamer. Also, coiled-coil-forming sequences are confronted with an enormous variety of potential partner helices in the cell. We are studying how the correct complexes are specified by amino-acid sequence, using both computational and experimental approaches. In recent years, we have focused on the interactions of bZIP transcription factor leucine zippers, which form parallel coiled-coil dimers.

The structure of the heterodimeric Fos/Jun bZIP transcription factor bound to DNA is shown at left. Approximately 53 bZIPs are encoded in the human genome, and these can form a variety of homo- and hetero-dimers. (Structure by Glover & Harrison, 1995)

Assays for coiled-coil interactions. We use peptide microarrays and other biophysical assays to measure bZIP leucine-zipper interactions on a large scale. To generate microarrays, peptides are printed directly onto chemically derivatized glass substrates and probed with fluorescently labeled peptides in solution. We continue to develop protein-interaction assays at many scales, seeking methods that can provide large amounts of high-quality data to complement computational modeling. We have measured human bZIP coiled-coil interactions, viral-human bZIP interactions, and the interactions of many designed peptides. Newman Science 2002, Grigoryan Nature 2009, Reinke Biochemistry 2010, Reinke JACS 2010

In collaboration with the Singh lab (Princeton University) we have applied machine learning to the problem of predicting bZIP coiled-coil dimerization preferences. Performance is impressive – good enough to be useful for efficiently prioritizing interactions for experimental testing. Fong Genome Biology 2004

bZIP interaction specificity can be modeled by predicting atomic-resolution structures and evaluating these using a physics-based energy function. The availability of large datasets, such as our bZIP microarray data, makes it possible to critically assess the strengths and limitations of such methods and identify good models. Grigoryan J. Mol. Biol. 2006

Coiled-coil dimers can adopt either a parallel or an anti-parallel helix orientation. We have developed and tested structure-based computational methods to predict this preference, given the sequences of two helices. The best methods are ~80% accurate on examples drawn from the PDB. Apgar Proteins 2008

Two-component histidine kinases dimerize through a coiled-coil-like four-helix bundle, as shown at left. Prokaryotes encode large numbers of such kinases in their genomes. We are studying the determinants of kinase dimerization specificity in collaboration with Prof. Michael Laub’s laboratory at MIT. We have developed assays for kinase homo vs. heterodimerization and identified individual residues important for preventing crosstalk of specific homodimers.

Coiled-coil design

The large number of coiled coils encoded in the proteome makes these attractive targets for the design of synthetic interaction partners. We are pursuing the design of peptides that interact with native coiled coils, which could be used as probes or dominant-negative inhibitors. An important challenge is how to design peptides that are specific for their intended targets over a large number of related coiled-coil “decoys” that may be present in the cell.

The human bZIP transcription factor leucine zippers are attractive but challenging design targets. Using novel computational methods that address the tradeoff between optimizing affinity and optimizing specificity, we have designed and characterized >50 synthetic peptides that bind to native human bZIP coiled coils. In the figure, thick red donuts show successful examples of targeting the intended bZIP selectively, as assessed using peptide microarrays. Grigoryan Nature 2009

Several viruses encode bZIP proteins with only distant similarity to mammalian genes. We have designed peptides that bind two of these viral targets: Meq and BZLF1. Reinke Biochemistry 2010, Chen submitted

Coiled coils in Saccharomyces cerevisiae

The yeast proteome encodes hundreds of putative coiled coils. We have measured interactions for many of these, particularly for coiled coils that are components of the spindle-pole body or are involved in vesicular trafficking.

The spindle-pole body is an ~400 MDa complex that organizes microtubules in yeast. The ~20 proteins that comprise the structure are nearly all predicted to contain regions that form coiled coils. We have characterized interactions among these regions using a variety of biophysical assays and used the resulting information to put constraints on a model for the architecture of the spindle pole body. Zizlsperger Biochemistry 2008, Zizlsperger J. Structural Biology 2010

Coiled coils are predicted to be common in yeast proteins involved in vesicular trafficking. In collaboration with the Jiang lab (Chinese Academy of Sciences), we developed a computational/experimental screening strategy for efficiently uncovering coiled-coil mediated interactions that may participate in this process. Zhang J. Mol. Biol. 2009

Coiled coils for molecular engineering

Coiled coils are widely used in protein engineering to induce oligomerization or to assemble nanomaterials. We have designed and/or screened numerous helical peptides for their interaction behaviors, providing a “parts list” of reagents that can be applied in synthetic biology or materials applications. We continue to optimize these reagents for a variety of uses.

In collaboration with the Imperiali lab (MIT), we have used computational methods to design a short peptide that specifically adopts a heterotetrameric structure. The structure of the complex consists of a helical bundle (somewhat like a coiled coil) that is capped by anti-parallel beta hairpins. Ali Structure 2005

In a 52 x 52 microarray screen of binary coiled-coil peptide interactions, we have identified numerous interaction sub-networks that will be useful for synthetic biology and protein engineering. We have also tested many of these for their interaction properties in solution and in yeast cells. Reinke JACS 2009

Bcl-2 family proteins

The Bcl-2 family consists of ~25 proteins important for controlling apoptosis. Five mammalian anti-apoptotic family members have a conserved globular structure, and all known family members share a weakly conserved short BH3 (Bcl-2 homology 3) sequence. Peptides corresponding to the BH3 region can adopt an alpha-helical structure and bind a hydrophobic groove on the surface of anti-apoptotic proteins. Critical junctures that govern cellular life-vs.-death decisions are regulated by specific interactions among pro- and anti-apoptotic members, making inhibition of Bcl-2 family interactions a promising therapeutic strategy for treating cancers. From a molecular recognition perspective, the docking of a single BH3 helix into a globular receptor is simple enough to allow comprehensive experimental analysis and extensive computational sampling, yet complex enough to stretch our understanding of specificity in molecular recognition. The lab is using a variety of computational and experimental methods to understand and manipulate Bcl-2 family interaction specificity.

Structural studies of Bcl-2 family complexes

The overall structure of complexes between anti-apoptotic Bcl-2 family members and pro-apoptotic peptides is highly conserved. Thus, fairly subtle differences in structure must account for large variations in affinity observed for different pairs. A recent explosion of structures in the PDB provides an opportunity to address how this occurs. We are contributing to this body of structural data.

The structure of a Bim point mutant bound to anti-apoptotic protein Mcl-1, compared to the wild-type peptide in the same complex, illustrates how structural plasticity may play a role in interaction specificity. We have solved several structures of mutated Mcl-1/Bim complexes. Fire Protein Science 2010

In collaboration with Prof. Loren Walensky’s group (Dana Farber/Harvard), we solved the structure of a covalently cross-linked helical peptide derived from Mcl-1, bound to Mcl-1. This molecule is a selective inhibitor of Mcl-1—peptide interactions. Stewart Nat. Chemical Biol. 2010

Design and selection of novel BH3 peptides

The known BH3 regions that bind to Bcl-2 proteins are diverse in sequence and show a range of interaction specificities with respect to anti-apoptotic family members. By using both computational design and experimental screening, we are adding to what is known about the requirements for BH3 binding, and exploring new regions of sequence and interaction space. Novel BH3 ligands will provide valuable reagents for studying the role of Bcl-2 interaction specificity in apoptosis.

We have used computational protein design to identify new peptides that bind to anti-apoptotic protein Bcl-xL yet have properties distinct from known native ligands. Fu J. Mol. Biol. 2007

Novel peptides with specific binding properties can be identified via experimental screening of combinatorial libraries. We applied yeast-surface display to identify peptides that bind to Bcl-2 family protein Mcl-1 in preference to Bcl-xL, or vice versa. We also explored determinants of interaction specificity in our identified peptides to derive a simple model that describes binding preferences.Dutta J. Mol. Biol. 2010

Systems biology of apoptosis

The Keating lab is part of the Cell Decision Processes Center (http://www.cdpcenter.org), an interdisciplinary group seeking to understand cell death in a mechanistic and predictive way at a systems level. Bcl-2 family interactions sit at a critical juncture in the extrinsic apoptosis pathway, and we are working with systems-level modelers and cell biologists to explore how interaction specificity within the family may be important for cellular life/death outcomes. We are also interested in how designed peptide reagents can be used to profile cell state and probe molecular mechanism.

We have collaborated with the Sorger lab (Harvard Systems Biology) to expand the Albeck & Burke model of mitochondrial permeabilization to account for differential interactions among Bcl-2 family members.

Computational Methods

Computational modeling of structure can provide insights into determinants of protein interaction specificity, and structure-based design is becoming a powerful tool for generating new molecules. We are developing tools to facilitate the use of structure-based methods for designing interaction specificity.

Backbone flexibility is important for protein design. We have used normal-model analysis to introduce simple degrees of freedom into alpha-helical backbones when designing peptides to bind to Bcl-xL. Fu J. Mol. Biol. 2007

We have adapted the method of cluster expansion, frequently used in materials science, to the problem of evaluating different sequences on a protein backbone scaffold. This converts a structure-based procedure for determining energies to a much simpler sequence-based expression, resulting in dramatic computational speed-ups. We have implemented this method in the software package CLEVER 1.0. Zhou Phys. Rev. Lett. 2005, Grigoryan PLoS Comp. Biol. 2006, Apgar J. Comp. Chem. 2009, Hahn J. Comp. Chem. 2010

Using CLEVER to generate efficient energy expressions, we can pose specificity design as a constrained optimization problem. We accomplish this in our method CLASSY. CLASSY makes it possible to systematically explore tradeoffs between interaction affinity and specificity, and we have applied it to the design of specific coiled-coil peptides. Grigoryan Nature 2009