Drugging the undruggable with small, structured proteins

From the Olson Lab, Clinical Research Division

Most pharmaceutical drugs are small molecules that bind to the active site of a target enzyme. They are often similar in structure to the enzyme’s natural substrate, but carry key differences that allow them to out-compete other molecules or permanently turn off enzyme activity. For many diseases, however, it will be necessary to disrupt interactions between two different proteins to achieve a therapeutic result. Designing drugs to target protein:protein interactions has proven difficult because small molecules are generally unable to tightly bind protein surfaces, while larger molecules like antibodies cannot penetrate deep into tissues or enter cells. Thus, finding a “Goldilocks” molecule that is big enough to disrupt protein binding but small enough to access the right locations is highly desirable.

The Olson Laboratory in the Clinical Research Division may be onto a solution. “Our group works with a class of proteins, called cystine-dense peptides (CDPs), which are inspired by protein-based drugs used widely in nature for things like muscle activity suppression in scorpion venoms,” explains Dr. Zachary Crook, a post-doctoral fellow in the Olson lab. CDPs are smaller than the average protein and tend to be highly structured due to the unique ability of cysteine residues to form disulfide bonds with each other. These properties make CDPs ideal starting points for developing drugs that target protein:protein interactions.

Finding the right drug candidate for a given job is like finding a needle in a haystack, so the researchers needed a way to screen thousands of CDPs to find those with desirable properties. In work recently published in Nature Communications, Dr. Crook and his colleagues developed such a system using a technique known as mammalian surface display, in which CDPs are tethered to the cell surface via a transmembrane domain. The authors constructed a library of 10,000 CDPs by combing genomic sequences of plants, animals, bacteria and fungi. Most of the CDPs used in the study were cysteine-rich fragments of proteins that have not been structurally characterized.

After cloning the CDP library into a lentiviral vector and transducing it into 293F cells, the researchers used protease digestion, cell sorting and DNA sequencing to evaluate each library member’s expression and degree of structure. Overall, they found that about 80% of CDPs did not fold well, indicating that these peptides might not be able to bind targets tightly and consistently, which are essential features of drug-like molecules. Poor-performing CDPs were mainly derived from large proteins, which suggested a role for surrounding protein context in CDP stability. In addition, CDPs containing glycosylation sites tended to have a less consistent structure. 

Figure 1: Illustration of a designed peptide (blue) sitting on the surface of the TEAD protein (pink) where the YAP protein (green) would otherwise bind. Because the YAP:TEAD interaction is involved in a number of different cancers, the engineered peptide could be the starting point for a new protein-based therapeutic. Image provided by Dr. Zachary Crook

To test whether CDPs that fold well when tethered to the cell surface also fold well as soluble proteins, the authors expressed 600 library members as secreted peptides. Of these proteins, 45% were classified as well-folded in solution, suggesting that many could be promising as drug-like candidates. In the future, the Olson lab will build on these findings to generate libraries with more consistently structured members and plans to integrate mutagenesis methods to test how slight variations in a CDP’s sequence affect its physical properties.

Now that the authors had a validated system for screening CDPs, they sought to identify a candidate capable of disrupting a specific protein:protein interaction. They chose to target binding of the transcription factor TEA domain family member 1 (TEAD) to its co-activator Yes-associated protein (YAP). This interaction activates wound healing in normal cells but is often co-opted to promote uncontrolled cell growth in cancer. Because the YAP:TEAD interface has been structurally well characterized, the authors did not need to start with a library of random CDPs. Instead, Dr. Crook and Dr. Phil Bradley (Public Health Sciences Division) used the Rosetta software suite, developed in the lab of Dr. David Baker at the University of Washington, to design a custom library of CDPs that were predicted computationally to bind TEAD and block YAP (Figure 1).

Figure 2: (A) Flow cytometry data showing enrichment for cells bound to streptavidin-Alexa Flour 647, which indicates surface display of CDPs that bind TEAD-biotin. (B) The candidate CDP TB1G1 reduces the ability of FLAG-YAP and Myc-TEAD to bind each other in a co-immunoprecipitation assay. Image modified from the original publication, reproduced with permission under the terms of the Creative Commons license: https://creativecommons.org/licenses/by/4.0/legalcode

The resulting CDP library was screened by flow cytometry for candidates capable of binding biotinylated TEAD (Figure 2A). Following four rounds of enriching for cells with a strong fluorescent signal (due to binding of fluorescently labeled streptavidin to biotin-TEAD-bound CDPs), the Olson lab collaborated with the Strong Laboratory in the Basic Sciences Division to characterize the stability and TEAD binding affinity of each CDP using biochemical assays and high-performance liquid chromatography (HPLC). The top candidate, denoted TB1G1, bound TEAD with a nanomolar dissociation constant and caused a dose-dependent reduction of the YAP:TEAD interaction (Figure 2B).

Next, the authors sought to improve the affinity of TB1G1 for TEAD by making a library of hundreds of similar variants and testing whether any worked better. This process yielded a variant termed TB1G2 containing five point mutations that strengthened the CDP’s interaction with TEAD. The improved candidate was resistant to numerous insults, including heat, proteolysis and reducing conditions, all of which could challenge a protein-based drug in the body.

TB1G2 was able to disrupt YAP:TEAD binding in the surface display assay and in cells transfected with TB1G2-expressing DNA. However, when testing whether pre-made TB1G2 could inhibit intracellular YAP:TEAD when mixed with cells, the researchers ran into a problem: TB1G2 couldn’t get in to the cells on its own. However, when administered with a cell-penetrating peptide, TB1G2 was able to cross the cell membrane and reduce co-localization of YAP and TEAD. Follow-up work is currently underway in the Olson lab to produce a variant of TB1G2 capable of passing into cells and accessing the nucleus, where YAP and TEAD reside. Dr. Crook is optimistic: “if this can be done, this will be an early example of a new class of anti-cancer drugs that are specialized to inhibit targets that are otherwise difficult to address.”


Crook ZR, Sevilla GP, Friend D, Brusniak M-Y, Bandaranayake AD, Clarke M, Gewe M, Mhyre AJ, Baker D, Strong RK, Bradley P and Olson JM. 2017. Mammalian display screening of diverse cystine-dense peptides for difficult to drug targets. Nature Communications. 8(1):2244.

This research was supported by the National Institutes of Health, the Washington Research Foundation and Project Violet.

Fred Hutch/UW Cancer Consortium members James Olson (Fred Hutch), Roland Strong (Fred Hutch), Philip Bradley (Fred Hutch), and David Baker (UW) contributed to this work.