MACRO DESIGN: EVALUATION METRICS + PROTEIN MODELING
Evaluation Matrix: Design
To ensure that our computationally generated mutations of RdhANp would indeed
dehalogenate selected PCB congeners, we began by developing an evaluation
protocol using Rosetta, a powerful command-line program suite for computational
macromolecular modeling. The initial evaluation protocol was generated based on
a basic Rosetta GALigandDock to recapitulate the reported activities of RdhANp
against a set of bromo/chloro organic halides.
Evaluation Matrix: Build
We downloaded the set of bromo/chloro organic halides from PubChem and docked
them against the crystal structure of RdhANp (PDB ID: 6ZY1 [1]). We then relaxed
the structure using the following Rosetta script:
$ROSETTA/main/source/bin/relax.static.macosclangrelease -s complex_relaxed.pdb
-relax:constrain_relax_to_start_coords -relax:coord_constrain_sidechains
-relax:ramp_constraints false -missing_density_to_jump -extra_res_fa params/B12.params
-beta -extra_res_fa params/LIGAND.params -default_repeats 2
-optimization:default_max_cycles 200
Evaluation Matrix: Test
The relaxed structure was docked using a basic GALigandDock protocol on “eval” mode,
which calculates the Rosetta energy scores of the entire structure without significant
repacking (allowing significant rotamer changes or sampling ligand binding space).
dH and dG obtained from the score files were our primary measures of interest. Unfortunately,
there was only some correlation between the dH/dG values with the reported activities of RdhANp
against the set of bromo/chloro halides we tested with this initial protocol, and
the ligand positions varied.
Evaluation Matrix: Learn & Redesign
With further research, we found that the dehalogenase facilitates catalysis
primarily via proton transfer at Y426 and the ideal distance between O of
Y426A and the halogenated carbon is 3.5 Å [2]. A 5.0 Å distance between
cobalamin and halogenated carbon was also an important determinant for effective
coordination [3]. We thus added these parameters into our relax protocol using
constraint files.
We also experimented with different relax parameters and constraint weights,
repeating the relaxation and docking steps to better recapitulate the reported
activities. We tried pairfitting the tested substrates with the ligand of another
RdhANp crystal structure (PDB ID: 6ZXX [1]). These repeated experiments led us
to a set of parameters that more accurately and reliably recapitulated the reported
activities with output dH values. Our final the relaxation command used was the following:
$ROSETTA/main/source/bin/relax.static.macosclangrelease -s $prefix.pdb
-relax:constrain_relax_to_start_coords -relax:coord_constrain_sidechains
-relax:ramp_constraints true -missing_density_to_jump -extra_res_fa B12.params
-extra_res_fa LIGAND.params -default_repeats 2
-optimization:default_max_cycles 200 -beta
-constraints:cst_fa_file LIGAND.txt -constraints:cst_fa_weight 0.1
And our final constraint file was the following:
AtomPair OH 426A C9 1X SCALARWEIGHTEDFUNC 1000 HARMONIC 3.5 0.1
AtomPair CO 703A C9 1X SCALARWEIGHTEDFUNC 50 HARMONIC 5.0 0.1
Our working protocol suggests that a dH of -17 to -20 would represent a RdhANp
activity of 65-80%. This would hence be the threshold dH value we use for
determining computationally whether the RdhANp mutation sequence confers
sufficient activity against the selected PCB congeners.
Protein Design: Design
Once we have established a working protocol for evaluating RdhANp activity
against various substrates, we constructed models of the PCB congeners the
wet lab would be using (PCB 11, 132 and 209) with RdhANp. PCB 132 has two
docking conformations – para and ortho – which were treated as two separate
docked structures. We evaluated the initial activities of the structures with
our protocol. PCB 11 seemed to have the potential for highest activity with
the wild-type enzyme based on the dH values resulting from the docking evaluation.
All results obtained may be found in the results page of our wiki.
To redesign the protein-ligand interface such that RdhANp may possess greater
affinity for the PCB congeners, we looked into different softwares to use.
This is as although RosettaDesign is a plausible platform for making redesigns,
it is less effective for native protein structures which we are using. Our
graduate advisor, Preetham, strongly recommended that we use an updated version of
ProteinMPNN [4], a MPNN protocol for redesigning proteins to better interact with
ligands of choice which is yet to be published.
Protein Design: Build
We sent our PCB-docked enzyme structures to Preetham to generate the mutated
sequences. Mutations were constrained to be within 8 angstroms of the ligand
and 5 angstroms away from the cobalamin as ligand interaction with the cofactor
is critical to the dehalogenation activity.
As a preliminary test, the mutational sequences for RdhANp docked to
2,6-dichlorophenol were generated to verify whether the mutations generated
changed RdhANp energy according to our evaluation model. Then, Preetham helped us
generate 16 sequences for each of the PCB-RdhANp constructs. However, the first
set of mutations did not fix the catalytic tyrosine Y426 and it was mutated to a
phenylalanine. As phenylalanine is structurally similar to tyrosine, we reverted
the mutation and moved forward with our evaluations as well. At the same time, a
second set of mutations were generated with ligandMPNN, with Y426 fixed.
Protein Design: Test
These mutated sequences (both the Y426F and fixed Y426) were uploaded to Benchling
and aligned to identify the full set of mutations. Of the mutation sequences
generated, 70-80% were unique sequences. We then ran our evaluation protocol
against the PDB files for each sequence. While sequences resulted in worse dH,
most sequences improved the dH energy of our PCB-RdhANp structure. However,
only a select few crossed the dH threshold of -17 to -20. We selected the
sequences that resulted in the lowest dH to be sent for experimental validation.
Protein Design: Learn
The mutations which appeared most frequently were across all 4 docked PCB models
were K295L, E298L, K310A, P314Y, M315L, S423T, S425V, M426L, H457D, V556I and
A563G. Hence it is likely that the same mutation sequence may have good activity
against all the PCB congeners.
MACRO LEARN: PROTEIN REDESIGN
Protein Redesign: Design
While waiting on feedback from Wetlab, we decided to further
evaluate the mutation sequences generated by the MPNN design. This was to
assess whether the mutations made were indeed beneficial to improving RdhANp
activity against the target PCB congener.
Protein Redesign: Build
Each mutation of the sequences sent to the wet lab was reverted to the wildtype
sequence using Pymol and saved as a new structure. The structures were then
relaxed and scored using our evaluation metrics previously established.
Protein Redesign: Test
Using the original dH as a baseline for comparison, we found that indeed certain
mutations improved the dH energy while others have made it worse. Additionally,
reverting all mutations that made the dH score worse did not result in a
significant improvement in overall dH. Instead, only a unique combination of
reversions effectively improved the dH energy of the structures. We have thus
far managed to improve the dH of PCB 132 ortho and para structures with a unique
set of reversions
Protein Redesign: Learn
We learnt that not all mutations introduced by MPNN led to better scores and
that energy scores are not additive. Thus moving forward, an additional assessment
of each mutation of the best mutational sequence generated by MPNN should be
included to ensure that all mutations introduced are effectively improving the
activity of the protein against the target ligand.
