Our Process began with finding our Missing Sequences and PDB Files in Our Systems

Binding Peptides Required for Tau

And after a literature Search, we reached many Binding Peptides and filtering was according to parameters: Clinical trials, IC50, Bind to tau, AB or both, Inhibition of PHF6 or PHF6* or Both,Inhibition of whole tau filaments, Cell penetration and nucleus penetration .

After Modeling, Quality assessment and Ranking, eight out of nineteen peptides ranked first.

M4W39, Mink, MMD3rev, MMDP6, TD28, TED28REV, Wink, WWW (Sequence of amino acid Tau Inhibitors generated based on the interface of two structure VQIINK[2], Sequence segment derive formation of Tau aggregations relying upon the beginning of repeat 2 )

Linkers Required for Trim System

Based on literature search for the needed linker, we reached (GGGGS)3 Protac linker from NUDT_China Project and GSGS from the literature [3] And finally we reached (GGGGS)3 to assemble the TBP with either trim21, Coh2 or Docs

GST and His modeling with our Dockerin and Cohesin Importance:

Our two complementary modules, Dockerin and Cohesin [4]. According to previous reports, Dockerien proved to produce a high yield when it is fused with will folded-protein [5-8] due to its proven degradation capability in E. coli [9] and its instability in solution when its expressed alone.

So we decided to express it once with GST and with His tag to see how they will affect our protein folding and stability by our models Quality assessment and by docking results. To best perform in our pull down assay we will also express cohesin with GST and His to estimate the previously mentioned parameters.

Table 1 : describes our used parts and its reference either from iGEM Registery or literature

Modelling jobs and Ranking

In Our Modeling Approach, we first Modeled our tau binding Peptides to complete our missing PDBs in our Protein Pipeline of our system

Tau binding Peptide modeling

According to our QA code, After Ranking to choose between 35 Models had 5 and 6 value out of 6 among 239 model. 35 Models has 0 clash score and 0 Cbeta deviations, very close molprpbity value, Ramachandran favored = 100 and Ramachandran outlier = 0.

1-Model” peptide (WWW) modeled by itasser model2.pdb” ranked the best one among 35 Model for WWW according to its score 6 , Q Mean_4 value = -1.602 , Q Mean_6 value = -1.34 , and molprobitey = 1.49 .

Fig. 3: 3D structure Prediction of peptide (WWW) modeled by itasser

2- Model “ peptide (TD28rev) modeled by apptest fold_peptide_82.pdb ” ranked the best between our 35 Models ranked the best one among 35 Model for WWW according to its score 6 , Q Mean_4 value = -0.41 , Q Mean_6 value = -1.47 , and molprobitey = 1.2.

Fig. 4: 3D structure Prediction of peptide (TD28rev) modeled by apptest

TRIM_Linker_DOC

According to a literature search, Trim system was split in our assembly jobs for 2 major parts to be assembled depending on the experimental proved Interaction Between DOC2 and COH2. we found there was no difference in fusing either Docs and Coh2 from the N terminal or C terminal. We made all assembly probabilities for Trim fused with Docs or Coh2 and TBP with each binding module with two different linkers we have gone through the same filtration process till we reached best model combination. After our Quality assessment process 2 Trim -the linker DoC model ranked first out of 15 models that were:

1- “peptide (trim-(GGGGS)3-docs) modeled by trrosetta model1.pdb” that had had score 5 with c beta deviation = 0 , clash score = 141.09 , molproberty = 2.61 , ramachandran favored = 98.57 , ramachandran outlier = 0, Q mean_4= 3.73 , Q mean_6 = 2.3

Fig. 5: 3D structure Prediction of (trim-(GGGGS)3-docs) modeled by trrosetta

2- peptide (trim-(GGGGS)3-docs) modeled by trrosetta model3.pdb” with score 5 out of 6 , c beta deviation = 0, clash score= 140.33 , ramachandran favored = 99.43 , molproberty = 2.61, ramachandran outlier = 0 , Q mean_4 = 3.98 , Q mean_6 = 2.66

Fig. 6: 3D structure Prediction of (trim-(GGGGS)3-docs) modeled by trrosetta

COH_Linker_WWW

Out of the eight prevuisoly mentioned pepetide, two were sucessful in modeling with the Coh2 with GS linker. From 3 models out of 16 Models modeled in 4 software, 1 Model ranked best according to its Q_Mean values 4 and 6. The best Model has values of : score 5 out of 6 , C-Beta deviation = 1 , clash score = 3.08 , molprobitey = 1.35 , Ramachandran favored = 96.31, Ramachandran outlier = 0.79 , Q Mean_4 = -0.01 , Q_Mean_6 = -0.29.

Fig. 7: 3D structure Prediction of GST- COH_Linker_WWW Modeled by tRRosetta Model2

GST-CoH-linker-TD28rev

For two Models with scores five out of six and among eleven Models, we choose our best-ranked model according to its score 5 out of 6 Q Mean_4 value. Our model values : C Beta deviation = 1 , clash score = 2.74 , molprobity = 1.22 , Ramachandran favored = 97.11 , Ramachandran Outlier = 0.52 , Q Mean_4 = 1.96 , Q Mean_6 = 1.88

Fig. 8: 3D structure Prediction of GST- COH_Linker_TD28rev Modeled by TRRosetta Model1

GST and His Tag modeling with our parts CoH and DoCs

Our four top ranked models were according to Q Mean score filtrations neglecting high clash score results:

GST_Coh2

Fig. 9: 3D structure Prediction of GST- COH modeled by Alphafold

His-CoH

Model “peptide (His_Coh2) modeled by TRrosetta model1” Ranked second model from our 4 ones with score 5 out of 6 and values of C Beta deviasion = 0 , clash score = 3.23 , molpro0bitey = 1.59 , Ramachandran favored = 92.31 , Ramachandran outlier = 1.4 , Q Mean_4 = 0.835 , Q Mean_6 = 0.95

Fig. 10: 3D structure Prediction of His- COH modeled by TRRosetta

His-DoCs

Model “peptide (His_DocS) modeled by TRrosetta model1” Ranked Third model from our four ones with score 5 out of 6 and values of C Beta deviasion = 0 , clash score = 0 , molprobitey = 0.77 , Ramachandran favored = 96 , Ramachandran outlier = 0 , Q Mean_4 = -0.29 , Q Mean_6 = -0.27

Fig. 11: 3D structure Prediction of His- DOC modeled byTRRosetta

GST-DOCs

Model “peptide (GST_DocS) modeled by Rosettafold full_model_407305_3” Ranked fourth model from our 4 ones with score 4 out of 6 and values of C Beta deviation = 0 , clash score = 182.78 , molprobitey = 2.72 , Ramachandran favored = 98.25 , Ramachandran outlier = 0.7 , Q Mean_4 = 0.75 , Q Mean_6 = 1.02

Fig. 12: 3D structure Prediction of GST- DOC modeled by Rosettafold

The beginning of our Journey was by searching for our Missing PDB Files that was:

I. HTRA1 Inhibitors: that were found on Uniprot, bind with the catalytic domain. our filtration was according to human inhibitors, Sequence length to be (1-200), and the Reviewed models. after the filtration process, reached 21 Inhibitors to go through Modeling and Quality assessment steps to choose between them[1].

II. HTRA1 Binding Peptides: Our Reason behind choosing HTRA1 Peptide to bind to the PDZ domain was according to literature results that revealed many PDZ ligands [2] can increase protease activity, ensuring correct trimerization for HTRA1 and strong binding between it and the rest monomer system (Inhibitor-Clamp-linker ) that will increase HTRA1 activity. our filtration was based on Binding affinity and IC50 results from literature and after the filtration process, we reached the Final HTRA1 binding peptides. The was [DARIWWV], we entitled it H1A, with a binding affinity (ΔG) proved to be -8.1 (kcal/mol) and experimental IC50 to be 0.9 ± 0.1 μM, Its IC50 was ranked first among 10 experimentally tested peptide [3] since its lower tested toxicity between 10 HTRA1 peptides and the most tried experimental one with promising results.

III. Amyloid beta binding peptides: was filtered and compared, after that we choose the seed, GGVVIA, for many experimental trials, proved binding to amyloid beta, its PDB already found or RCSB(PDB ID:2ONV) [4]

VI. A flexible linker was chosen to bind the clamp to each other and test their binding to Tau and amyloid beta, whereas the linker choice also was critical upon fusing the clamp, H1A Binding peptide, and the inhibitors that was from our choice for G and S amino acids [5] and a specific length of the linkers for the clamp and clamp-peptide will be detected in detail on the Docking and Modeling page.

Table 1 : describes our used parts and its reference either from iGEM Registery or literature

HTRA1 Criteria

Our aim behind this system is to degrade both tau and amyloid beta with our Clamps which consist of TD28rev and WWW and two beta amyloid seeds.

Our reason behind using tau-binding peptides and seed in the same clamp was for their proven ability of them to degrade Tau and amyloid-beta with high affinity, in addition to our criteria implemented in our trim21 system [10].

And To ensure correct HTRA1 protease activity, we put specific criteria based on the total switch modelling, their affininty to bind their targets (Tau, Amyloid beta and HTRA1), which will be valid through our Docking steps and Prodigy raking [11]. Our criteria will be explained below while the results of both found in the docking page

Clamp binding affinity > HTRA1 Peptide - PDZ binding affinity > Inhibitor- HTRA1 Catalytic domain Binding peptide affininty for Tau or Beta-amyloid should have the highest affinity then the H1A bind with higher affininty to PDZ domain than HTRA1 inhibitor to the catalytic domain to make it unbind from the catalytic domain and perform its protease activity.

According to docking and prodigy results, SPINK8 inhibitor proved to be approximate to achieve our criteria that it has lower binding affinity than HTRA1(H1A) binding peptide and (H1A) has much lower binding affinity than Tau and amyloid-beta affinity to our clamp. Accordingly, Amyloid beta and tau must bind strong enough to unbind our inhibitor from HTRA1 so protease activity can be achieved and degradation occurs.

In the case of our inhibitor WAP-four, our clamp and amyloid beta has higher binding affinity than the H1A binding peptide which has the lowest value resulting in a probability of inhibitor to unbind H1A from PDZ affecting HTRA1 protease activity. For more details about the binding energies check the docking page

HTRA1 Monomer Modeling

According to HTRA1 References, we needed to model our HTRA1 2 domains that are responsible for HTRA1 activity [1] that were Catalytic domain and PDZ domain. Two domain sequences were acquired from Uniprot [ Q92743 ID]. PDZ domain binds our HTRA1- binding Peptide[DARIWWV] (We entitled it H1A)[1,3] and catalytic domain [3] that bind with HTRA1 inhibitors (SPINK8(C-N) [6] and WAP-four (N-C)[7].

HTRA1 Clustering

HTRA1 existence in a homomultimeric state as a trimer in our brain to be active made us go through the clustering phase before going through Modeling. After HTRA1 Monomer modeling, Modeled Monomer went through clustering through 2 Servers that were Clouspro 2.0[12] and GalaxyTongDock server. In Cluspro, we used the Multimer docking option that supports dimer and trimer formation till now. Multimer docking is considered a subclass of protein Interactions that allow the formation of homomultimers by the interaction between two or more identical Proteins [5]. For trimer formation, Cluspro rotates the coordinate system generated by our monomer using C3 symmetry [120 angles] in our case. And also trimerization is done on GalaxyTongDock server[13], Ab initio protein-protein docking that does symmetric and asymmetric rigid-body docking. In our case, we performed clustering on symmetric body docking of our homoligomeric protein monomer HTRA1 depending on Cn and Dn symmetries using GalaxyTongA Algorithm. the best model was chosen based on Cβ deviation and RMSD after alignment with the experimental model (RCSB: 3NZI) was from cluspro with 1.47 Å value on RCSB Pairwise Structure Alignment.

Fig. 1. 3D Structure prediction of HTRA1 Clustered protease

Linker length detection and filtration

we used a flexible linker because of our system mechanism of binding and unbinding between inhibitor, clamp, and HTRA1 binding peptides, also to check the best linker length for making clamp- tau or amyloid beta complex unbind inhibitor from HTRA1 catalytic domain according to our criteria. Our method was inspired by Heildberg team; we used Pymol to measure the linker distance between N and C Terminal for our Inhibitor and peptide, then subtracted this distance from clamp distance [NN, NC, CN, CC]. Afterward, we reached the amino acid number and length by calculating the Mean of lengths /3.5 length[15], the average length for every amino acid (1). Also, we did statistical analysis for every length we measured by calculating Mode, standard deviation, and Median. Approximation length was between (4-7 ) for Q8UIB5 inhibitor and 6 for SPINK8. According to these lengths, modeling jobs were performed with the system assembly for detecting the best length. For detailed methods check the docking page

HTRA1 Modeling Jobs

Modeling Jobs was to choose the best linker length, the best inhibitors out of 21 Inhibitor, and the best binding peptide model as will be detailed described below:

Inhibitor Modeling

After Modeling and Quality assessment for 21 Inhibitors, we reached 3 final Inhibitors that got a score of 6 out of 6, 0 value for Ramachandran outlier and C beta deviation value = 0. he top three inhibitors were: SPINK8, WAP-four, and Q9H1F0 that proceed to Docking to test their binding affinity to the HTRA1 Catalytic domain.

Inhibitor SPINK8

“peptide (SPINK8) modeled by trRosetta model2 “ with clash score = 1.36 , molprobitey = 0.87 , Ramachandran favored = 98.95 , Q Mean_4 = -0.415 , Q Mean_6 = -1.55.

Fig. 2. 3D Structure Prediction of Inhibitor (P0C7L1) modeled by trRosetta

Inhibitor WAP-four

““peptide (WAP-four) modeled by TRrosetta model3 “ with clash score = 0 , molprobitey = 0.5 , Ramachandran favored = 98.9 , Q Mean_4 = 0.86 , Q Mean_6 = -0.182.

Fig. 3. 3D Structure Prediction of Inhibitor (WAP-four) modeled by trRosetta

H1A Modeling

After modeling and QA, we filtered between four models of H1A that have scores of 5 out of 6 according to Q Mean_4 and Q Mean_6 values. The best model was” H1A_76358_relaxed_rank_3_model_2.pdb” that has Ramachandran score = 100, 0 value for C Beta deviation and clash score, molprobitey = 1.43 , Q Mean_4 = -2.36 , Q Mean_6 = -1.082.

Fig. 4. 3D Structure Prediction of H1A HTRA1 Peptide modeled by Alphafold

Our 4 clamp Probabilities

Three binding peptide was used: seed to bind Beta-amyloid and our tau binding Peptides TD28rev and WWW , we then had four probabilities according to this that were:

Seed-linker length - Seed
Seed-linker length - Td28rev
Seed-linker length - WWW
TD28rev - Linker length - WWW

We modeled them to find the best models for every probability and the results were:

After Modeling and QA of the clamp probabilities , we were filtering between more than 1500 model reaching two best final models that were :

1- “Probability 18” with score 5 out of 6 consists of TD28rev_GGSGGGG_WWW with values of C_Beta deviation = 0 , clash_Score = 6.86 , mol_Proberty = 1.38 , Ramachandran_Favored = 100 , Ramachandran_outlier = 0 , Q mean_4 = 0.256 , Q Mean_6 = -0.38

Fig. 5. 3D Structure Prediction of Clamp probability 18 modeled by TRRosetta

2- “Probability 4 Model 14 Modeled by Alphafold“ with score 6 out of 6 C_Beta deviation = 0 , clash_Score = 0 , mol_Proberty = 0.5 , Ramachandran_Favored = 100 , Ramachandran_outlier = 0 , Q mean_4 = -1.2 , Q Mean_6 = -0.7

Fig. 6. 3D Structure Prediction of Clamp probability 4 modeled by AppTest

Linker part assembly Modeling

After testing the binding of four linker lengths with Two Inhibitors, 1 binding peptide, with four clamp probabilities. We did 40 Modeling jobs; each was modeled in three software. Reaching 681 models, the swiss-Model quality assessment was performed, followed by filtration. We reached the top 40 models getting a score of 6 and 5 out of 6.

To choose between them, we did Structure alignment in Pymol for estimating RMSD against H1A Models and HTRA1-H1A docking jobs, getting from them four Top models.

Every part consists of an Inhibitor - one of 4 Clamp Probabilities - one of 4 linker lengths - H1A Peptide and best models will be detailed below:

Switch 12

Our first switch “ peptide (12) modeled by trrosetta model3 ”

Include 1- Inhibitor SPINK8 (C-N) - GGGGSG linker length - TD28revGGSGGGGWWW clamp probability - GGGGSG - H1A (C-N) 2- that has score 6 out of 6 , C Beta deviasion = 0 , molprobitey = 0.98 , clash score = 2.12 , ramachgandran favored = 98.51 , Ramachandran outlier = 0 , Q Mean_4=0.149 , Q Mean_6 = -1.63.

Fig. 7. 3D Structure Prediction of Switch 12 modeled by TRRosetta

Switch 10

Our second switch “ “peptide (10) modeled by trrosetta model1” has a quality score 5 out of 6 Include :
1-Inhibitor SPINK8(C-N) - GGGGSG linker length - SeedGGSGGGGGSeed - GGGGSG - H1A (C-N)
2- Peptide values were : C Beta deviation = 0 , molprobitey = 1.24 , clash score = 4.74 , Ramachandran favored = 98.6 , Ramachandran outlier = 0.7 , Q Mean_4=0.212 , Q Mean_6 = -1.44.

Fig. 8. 3D Structure Prediction of Switch 10 modeled by TRRosetta

Switch 15

Our third switch “ peptide (15) modeled by trrosetta model3 ” has score 5 out of 6 Including:
1- WAP-four (N-C) - GSGSGS linker length - TD28rev GGSGGGG WWW clamp probability - GSGSGS linker - H1A (N-C)
2- Peptide values are: C Beta deviation = 0 , molprobitey = 1.24 , clash score = 4.74 , Ramachandran favored = 98.6 , Ramachandran outlier = 0.7 , Q Mean_4=0.212 , Q Mean_6 = -1.44.

Fig. 9:3D Structure Prediction of Switch 15 modeled by TRRosetta

Switch 18

Our fourth switch “ peptide (18) modeled by trrosetta model1.pdb” has a score 5 out of 6 including:
WAP-four (N-C) - GSGSGS linker - SeedGGSGGGGGSeed - GSGSGS-H1A (N-C)
Peptide Values are : C Beta deviation = 0 , molprobitey = 1.17 , clash score = 3.77 , Ramachandran favored = 98.46 , Ramachandran outlier = 0 , Q Mean_4=1.27 , Q Mean_6 = -0.1.

Fig. 10: 3D Structure Prediction of Switch 18 modeled by TRRosetta

In conclusion, testing switches 10, and 12 could approve our theory and criteria in further wet lab results as it contains the inhibtor of lower binding affinity than H1A to HTRA1. wheres as switch 10 target Beta-Amyloid while switch 12 targets tay degradation.
In contrast the inhibtor with the highee affinitny to HTRA1 than H1A peptide was in the Peptide sequences 15,18 each target Tau and Beta-amyloid, respectively.