RESULTS

Overview

  After the design of our functional binder + transcytosis sequence (process mentioned in Engineering page) and the synthesis of our plasmid (described in Experimental) we engaged in both computational and wet lab procedures for the assessment of various properties. Through these methods we can divulge information like if the engineered pockets are predicted to bind to PFOA the same way as albumin, estimate the Δ G of binding for use in the model we developed (described in Modeling) which can tell us vital details regarding the thermodynamics of our binder, Albumin (PFAS-binding competitor), and PFAS itself within a transcytosis model.

Ligand Docking and Δ G estimation

 We used RosettaLigand (1) to generate fifty ligand-docked protein predicted structures for each engineered pocket, and performed this for both PFOA and Myristic Acid. We refolded the sequences in Alpha Fold 2 (2) since structures generated with Alpha Fold 3 (3) are not permitted for use in docking; Alpha fold 2 structures had very high structural similarity to Alpha Fold 3(RMSD > 1.0 Å) and were used in docking.

  We used the FA4 pocket with PFOA from 7AAI (4) as the comparison structure for docking runs with PFOA, and used the Albumin FA4 pocket with Myristic Acid from 1BJ5 (5) as the comparison structure for runs with docking runs with Myristic Acid. We chose myristic acid as the comparison ligand since myristic acid was found to be co-crystallized with PFOA in Albumin (4).

  After generating the ligand-docked structures, we used PRODIGY-Lig (6) to estimate the ΔG values of binding for the top 3 structures from each ground which had the lowest ligand RMSD. The structures with the lowest ligand RMSD represent the structures generated by RosettaLigand with the most similar binding pose to the comparison structure. We reasoned these structures would provide the best estimates for ΔG since they are closest to the established binding mode from x-ray crystallography experiments.

  At first glance, the data suggests that our designed pockets bind myristic acid more strongly than PFOA, as indicated by the lower (more negative) ΔG values for MYR. However, the error margins represented by the RMSE are substantial, suggesting that the observed differences in ΔG between the two ligands may not be statistically significant. Specifically, the signal-to-noise ratio of 0.87, calculated by dividing the effect size (the difference between the predicted ΔG values of the two ligands) by the RMSE, indicates that the differences are within the noise of the model. This leads to the conclusion that while our models appear to bind myristic acid slightly better, this difference may not be meaningful due to the high uncertainty in the model’s predictions. Thus, we can say more confidently from the PRODIGY-Lig predictions that our engineered pockets likely have similar ΔG of binding values for both PFOA and Myr, even if we cannot pinpoint the exact value. Additionally, from these ΔG predictions, all four pockets would be predicted to bind to PFOA and Myristic Acid at similar strengths—no significant differences were predicted across designs.

Immunogenicity

We predicted the immunogenicity of our proteins using AntigenPRO (7) which predicts the likelihood of a protein being recognized as a foreign antigen by the human body.

Interestingly, the native FA4 sequence had the highest probability of being an antigen despite the fact it is a native protein that circulates the blood. However, this might be attributed to the fact that a considerable portion of the FA4 pocket does not face the solvent and instead makes contacts with the rest of the Albumin protein. Additionally, since LigandMPNN is trained on the Protein Database, which is biased towards globular proteins (8), the algorithm likely excels at designing globular proteins which are less likely to trigger an immune response than proteins with exposed hydrophobic regions. Indeed, recognition of solvent accessible hydrophobic areas of proteins plays a key role in how antigens are recognized and presented in mammalian immunity (9).

However, we need to be cognizant of limitations of AntigenPRO; although it was trained on a balanced dataset of antigens and non-antigens, it was primarily trained on bacteria toxin data.

We were also interested in understanding the basis for design 119’s lower estimated probability of being an antigen since each of the designs have relatively high sequence similarity (70-80%, not including the linker or PASR sequence.) We identified 6 instances of a unique amino acid in design 119 compared to the other 3 designs:

• K137 - others are N
• E1 - others are S, A
• R23 - others are E
• T50 - others are I
• R72 - others are K, E
• H74 - others are E, D

The substitutions that are the least synonymous are T50I and R23. As expected the residues project towards the solvent, and none of these residues face the ligand-binding pocket.

Results and Analysis

Our main results from the model is that we can determine the steady state concentration of PFOA in the bloodstream with parameters. Using literature-determined values for the dissociation constant of PFOA and albumin, along with their initial concentrations, we developed a plot to look at such PFOA concentrations. We note that since we do not have an exact value for our binder, we instead use a ratio to characterize our binder, where the ratio is defined as the dissociation constant of the binder-PFOA complex divided by that of albumin-PFOA complex. This is equivalent to the scaler S defined earlier and for the following example, this ratio is set is 2.0 and the transcytosis rate is kept at 0.1 uM/time:

Indeed, as the initial binder concentration increases, the final PFOA concentration decreases in a nonlinear trend. However, additional analysis can provide insight into how effective the binder needs to be relative to albumin in terms of binding affinity. To provide a more complete picture of this aspect, we show how the variation of the ratio between the dissociation constants of the two complexes affects the steady state concentration of PFOA as shown in figure 3:

Initial binder concentrations of 100 uM along with having a binding efficiency of approximately two times that of albumin with PFOA yields an approximately two-fold decrease in the amount of PFOA in the bloodstream. We also note that even with an unfavorable ratio of dissociation constants (greater than 1), there is still a decrease in the final PFOA concentration. Thus, for these conditions, we believe that our engineered binder has potential to effectively remove PFOA from the bloodstream, even if it is marginally better at complexing than albumin.

Using the same parameters, we were also able to determine the steady state concentrations of the engineered protein in the bloodstream. Having high levels of the binder may cause toxic effects, so being able to estimate the remaining amount of binder provides insight into how safe our engineered protein is:

Indeed, we see practically none of the binder remains in the bloodstream. Furthermore, even with the large initial binder concentrations, there is less than 10% that readily remains in the bloodstream, suggesting that the binder will not have a major toxic effect due to being in the bloodstream. However, since we can not determine the duration that it is in the bloodstream, we are unable to assess if it stays at high concentration for long periods of time. Rather, we can only conclude that steady state in the long-term is likely safe in terms of binder concentration.

Finally, since an experimentally determined value for transcytosis rate was not determined and given the lack of time scale, we test four discreet magnitudes of transcytosis rate and look at the effects on steady state PFOA concentration:

Indeed, increasing the transcytosis rate increases the efficiency of the binder as for similar initial binder concentrations and ratios, we see that final PFOA steady state concentrations are smaller as the k4 value increases by magnitudes of 10 (emphasis on color map scale). Thus, we conclude the engineered protein has potential for efficient removal of PFOA from the bloodstream and modification to increase the transcytosis rate can make the binder more efficient in removal.

REFERENCES