Concluding Remarks
Key Achievements
To conclude, the biochemical design, production and investigation of PROTAC ligands for telomerase knockdown was enhanced by the mathematical exploration of the yeast protein interaction network. Our team has successfully designed a set of experiments and methods that will result in the identification of ligands that can be integrated into PROTAC therapeutics, which will undergo rigorous in vitro and in vivo analysis. This methodology has been designed for two proof of concept proteins: ATM and POP1, however they can be applied to novel targets identified through the protein protein interaction network of telomerase.
The key achievements in the mathematics part of the project include the partitioning of the protein network into communities and the subsequent identification of proteins which we deemed to be related to the telomerase proteins. Using the Walktrap community detection algorithm, we managed to partition the network in a way that roughly corresponds to the functional communities, and subsequently identified the communities most relevant to the telomerase proteins. By performing analysis on these communities using diffusion state distance, we obtained potential proteins that were then subjected to both mathematical and biological validations.
Future Directions - Therapeutic Development and Testing
Nonetheless, there are undoubtedly limitations with our methodology that we could approach differently in the future. Namely, the proteins selected as proof of concept, while strong candidates in terms of their interactions with telomerase and possession of similar yeast homologues, are also fairly important proteins. As such, a therapeutic targeting them may have off-target effects. However, with further analysis of the protein network, novel targets should appear that aren’t limited in this way.
Additionally, in finding ligands to be implemented into PROTAC molecules, only their binding affinities will be tested. This is the main characteristic for design of a PROTAC, however it misses the potential for discovery of inherent inhibitory action within the ligands. This could be improved by also designing inhibition assays for each of the ligands.
The cell line analysis to be conducted in the testing of the therapeutic also features a number of limitations that require further investigation. The nominal cell line, MFM-233, is a mammary carcinoma cell line isolated from a patient’s metastatic breast tissue. Breast cancer is an incredibly heterogeneous disease that unfortunately cannot be comprehensively studied in vitro in a single cell line. The American Type Culture Collection (ATCC) features a collection of 45 cell lines that provide a comprehensive representation of breast cancer (1). Further studies should include investigations into the therapeutic action of PROTACs on multiple cell lines.
The murine model adopted in this study provided a representation of the action of putative PROTAC ligands in vivo. Unfortunately, this can never be an accurate model for a human system. The need to use the severely immunocompromised NOD-SCID mice cannot accurately represent the immune landscape of human breast cancer patients. One caveat to this is that patients undergoing traditional chemotherapy are also likely to be immunodeficient (2). When our PROTAC therapeutics show successful therapeutic action in vitro and in vivo, studies in humans will be conducted. Following the aforementioned experiments, it is advisable to move on to initial clinical trials.
The yeast model employed by the maths team and biochemical verification is anticipated to be successful for our initial experiments. While it is easy and quick to grow, the two organisms have obvious differences. Homology between human and yeast proteins were maximised for the nominal protein targets. However, it would be inaccurate to assume that the action of a PROTAC would be the same in a human cell and a yeast cell. This is a further limitation of our proposed experiment.
Future Directions - Network Theory
The WalkTrap algorithm opens many doors for future research. Firstly, while we did minimal validation on the length of walk, considered as a parameter for the community detection algorithm, further experimentation on this would be valuable in understanding the algorithm. Secondly, with the WalkTrap community finding we ended up with large numbers of very small communities, including many consisting of only one node. Further research into understanding how we should interpret this–whether it reflects, for example, deficiencies in the algorithm or a failure of the STRING database in correctly capturing interactions–would allow us to better assess the strengths and weaknesses of the algorithm, relative to other community detection algorithms. Finally, the identified telomerase community is relatively large, and something that we could have tried is to perform WalkTrap again on the telomerase community, thereby potentially achieving finer resolution information on functional modules within the community and allowing more precise targeting.
Data in the STRING database have varying qualities; we took a threshold value of 700 out of 1000 to create our network. Experimentation with this value and how changing it changes our results could help determine how robust our results are. Furthermore, inclusion of other data sets such as mRNA transcription data could lend further biological insight which would help us find potentially more relevant results.
While our aim was to impact only one specific protein complex, the Saccharomyces cerevisiae PPIN is highly connected and the removal of any node could have diverse, unpredictable effects on the entire network. Further modelling to determine ‘off-target’ effects of knocking out proteins on neighbouring communities would be extremely valuable.