Wetlab Summary

During the course of our experimental wet lab phase, we designed a genetic construct to be assembled and produced in E. coli. To create this, we utilized Combinatorial Golden Gate assembly to produce four unique genetic cassettes, each encoding a specific portion of our biosensor complex. Each cassette existed as a plasmid, with Cassettes A and C each encoding one half of the binder/linker portion of our construct. Cassette B encoded our split complementation system, which “sandwiched” our RiboJ sequence. Cassette D encoded the backbone of our final plasmid that would be transfected into cells and translated into protein. This cassette included our selected resistances as well as a D-Promoter.

---

Future Steps

After initial experimentation, we discovered that our RiboJ sequence was not functioning correctly in the Golden Gate assembly, so we needed to fix one end of the riboJ sequence and rerun the Golden Gate assembly. Furthermore, a wise step would be to increase the number of split reporter proteins we are experimenting with, as we gathered many more during the research phase that we never got to test. This pursuit would also adjust split sites and substrate concentrations to maximize signaling.

---

Computational Summary

There are many biomarkers associated with concussions and mTBI, but their binding interactions with particular proteins are not often studied so some forms of binding (particularly the strength of the biomarker/binder complex) are difficult to characterize using only the literature. After narrowing the field of likely biomarkers and binders from our literature search, we obtained sequences and then used AlphaFold¹ to model complexes between binders and biomarkers. Based on the results from alphafold, we also determined binding effectiveness with two different docking suites, HDOCK and ZDOCK. Based on the results of all these simulations, we decided to use UCHL1 as our biomarker and split ubiquitin as our binder. GFAP had some promising binders as well, but we ultimately decided to focus purely on UCHL1 since we were unable to express GFAP.

---

Folding with AlphaFold

The four biomarkers we modeled binder complexes for were UCHL1, ENO2, GFAP, and S100β. For UCHL1, wild-type ubiquitin binds to UCHL1, and a particular modified form ubiquitin Ub-VME is used to study UCHL1². We didn’t know if reconstituted split ubiquitin would bind similarly, so we used AlphaFold to predict the structure of the complex. As expected from literature research, we found the highest number of hydrogen bonds and binding residues at the biomarker-binder interface and the binding was quite strong. There were over 20 hydrogen bonds and 67 residues at the interface. We calculated binding residues using a python script, this procedure can be found in our experiments page under “Dry Lab Protocols”. These generated structures were also fairly consistent based off of the executive RMSD’s (see below) calculated by aligning all 5 generated structures to the best structure in PyMOL³.

The dimer ENO2 had only 4 binding residues and no hydrogen bonds with bZIP75, and the binding location of bZIP75 blocks the dimerization of ENO2.

For GFAP, the binders modeled in alphafold were A10, VHHB7, G11, p53, VC1 RAGE, and VC1 w61. A10 and VHHB7 have the least viable modeling as there were no binding residues or hydrogen bonds. We decided to move forward with VHHB7 instead of A10. G11 has one binding residue with GFAP. It is worth noting that, compared to the strength of UCHL1-split ubiquitin complex, all of the GFAP-binder interactions are quite poor.

For S100β, binding with the VC1 (RAGE domain) and VC1 (w61) showed 17 and 23 interface residues respectively and no hydrogen bonds. However, using S100β in the experimental system was abandoned since S100β is expressed in the body during exercise and is expressed later in the body than UCHL1 after an mBTI event⁴.

AlphaFold RMSD Values

Biomarker	Binder	Binding Residues	Hydrogen Bonds	Rank 2 RMSD	Rank 3 RMSD	Rank 4 RMSD	Rank 5 RMSD
UCHL1	Split Ubiquitin	67	10+	0.210	0.285	0.311	0.247
ENO2	bZIP75	4	0	14.452	17.952	0.246	0.328
GFAP	A10	0	0	17.408	13.697	12.079	20.369
GFAP	VHHB7	0	0	19.242	12.461	18.651	18.757
GFAP	G11	2	0	16.186	14.927	12.463	21.550
S100β	VC1 RAGE	17	0	0.431	12.097	25.098	0.322
S100β	VC1 W61	23	0	0.251	0.161	0.240	0.208

---

Protein-Protein Docking with HDOCK and ZDOCK

Through protein modeling and docking, we decided to further investigate UCHL1 using split ubiquitin as a binder, as well as GFAP using S100β, G11, and VHHB7 as binders. Following the protocol listed on our experiments page, we used ZDOCK and HDOCK to dock our biomarkers and binders. Specifically, we modeled UCHL1 with split ubiquitin as a binder, as well as GFAP with S100β, G11, and VHHB7 as binders.

Regardless of docking software, the location of binding between the split ubiquitin and UCHL1 was consistent. Taken together with the results of the AlphaFold folding, which also predict binding in the same site with approximately the same strength we can be confident in the binding location and more confident in the strength of our models in silico.

Going forward it would be useful to have a model of the entire biosensor with the biomarker, binder, and linker together, but we were unable to model this due to technical limitations of the computational tools at our disposal.