Discussion

Conclusions

In this project, AlphaFold2 was used to predict the structures and binding interactions of two proteins, Ttc29 and Lrrc56. Eno2 and Zmynd12 are predicted with a high degree of confidence to bind the N-terminus of Ttc29. In a clinical setting, C-terminus truncations have been observed to cause ciliopathies; if AlphaFold2's predictions are accurate, then disrupted binding of Eno2 and Zmynd12 may not be responsible for the patient phenotype. Likewise, Ttc29 dimerization is predicted to occur in both wild-type and truncated isoforms, so loss of some C-terminal binding to other proteins may be occuring.

Lrrc56 was predicted to have a high degree of intrinsic disorder, which is fitting given its localization to the DynAP, a liquid organelle. deePhase v1, PSPredictor, and FuzDrop all returned a high likelihood that this protein will phase separate with consistent control results. catGRANULE and PSPer gave inconsistent results for both proteins of interest and controls; this could be due to these programs being trained on datasets that do not fully reflect the phase separation capabilities of human proteins (catGRANULE was trained on prion-like proteins, and PSPer was trained solely on yeast proteins). The clinical allele Lrrc56 p.Leu140Pro is not predicted to disrupt liquid-liquid phase separation, nor does the mutation disrupt the highly-ordered leucine rich repeats in AlphaFold2's predictions. However, this could be due to a limitation of AlphaFold being trained largely on wild-type LRR structures.

Future Directions

To investigate C-terminal interactions with Ttc29, AlphaFold2 should be performed with more dedicated memory to handle heterodimers with larger proteins Dnah1 and Myh10. Subsequently, a full complex including all 5 proteins of interest should be run using AlphaFold2 on TACC.

Lrrc56's intrinsic disorder is problematic for predicting binding partners. While RNA structures are not as commonly reported as protein structures, as databases of them continue to grow, new programs may be able to predict protein-RNA interactions with a high degree of confidence. Such added data may lead currently predicted IDRs to resolve into secondary structures.