Sequence-based Prediction of the Cellular Toxicity Associated with Amyloid Aggregation within Protein Condensates

Attila Horváth, Michele Vendruscolo, Mónika FuxreiterBiochemistry 2022 · DOI

TL;DR

This paper uncovers the "amino acid code" that dictates the toxicity of aging protein condensates, specifically focusing on the TDP-43 protein implicated in neurodegenerative diseases. They found that mutations promoting amyloid aggregation actually decrease toxicity, while those favoring droplet formation increase it, and developed a predictive model for this toxicity.

What Problem Does This Paper Address?

Proteins can exist in various states within cells, including native, liquid-like "droplet" states, and aggregated "amyloid" states. The transition from reversible droplets to irreversible amyloid aggregates is linked to neurodegenerative diseases like ALS. Understanding what drives the toxicity during this maturation process is crucial for developing therapeutic strategies. While previous work identified regions prone to droplet formation or amyloid aggregation, the specific sequence features that dictate the cytotoxicity of these aging condensates remained unclear. This paper aims to decipher this sequence-based code for TDP-43 cytotoxicity.

Methods

The researchers leverage a "droplet landscape" framework, which maps protein regions based on their propensity to form droplets (PDP) and their multiplicity of binding modes (MBM). MBM quantifies how many different ways a protein region can interact, ranging from ordered (specific contacts) to disordered (flexible interactions).

Key methods include:

  • Deep Sequencing Data Analysis: They analyzed over 20,000 TDP-43 variants generated through deep sequencing.
  • FuzDrop and FuzPred: These existing computational tools were used to predict droplet-promoting propensity (PDP) and the multiplicity of binding modes (MBM) for protein sequences.
  • Extension of FuzDrop: The core novelty lies in extending the FuzDrop method to predict cytotoxicity. This involved developing a machine learning model (random forest) that uses changes in PDP (ΔPDP), amyloid-forming probability (ΔPAP), and MBM (ΔMBM) upon mutation as input features.
  • Statistical Analysis: Mann-Whitney tests were used to assess the statistical significance of differences in MBM between wild-type and mutant residues. Pearson's correlation coefficients were used to evaluate the performance of the predictive models.

Datasets

  • TDP-43 Deep-Sequencing Data: Over 20,000 TDP-43 variants with experimentally measured cytotoxicity were used. This dataset includes single and double missense mutants.
  • ALS-associated Mutants: A subset of these mutants were specifically identified as being associated with ALS.
  • Wild-type TDP-43 LC Domain: This served as the baseline for comparison with mutant sequences.

Key Insights and Ideas

  • Aggregation as a Protective Mechanism (for TDP-43): Counterintuitively, mutations that promote amyloid aggregation appear to decrease TDP-43 cytotoxicity. This suggests that forming ordered amyloid structures might sequester toxic species or prevent them from engaging in harmful interactions.
  • Droplet Formation Drives Cytotoxicity: Conversely, mutations that increase the propensity for droplet formation and perturb disordered interactions lead to increased cytotoxicity. This highlights the critical role of the liquid-like condensate state and its dynamic interactions in disease pathology.
  • MBM as a Key Predictor: The multiplicity of binding modes (MBM) is a crucial factor. Mutations that increase MBM, particularly in droplet-promoting regions, shift binding towards more ordered interactions, which is linked to higher cytotoxicity.
  • Sequence-Based Prediction is Feasible: The study demonstrates that the amino acid sequence alone contains sufficient information to predict the cytotoxicity of TDP-43 condensates, paving the way for in silico screening of potentially toxic mutations.

Key Results

  • Predictive Model Performance: The extended FuzDrop method achieved high correlation coefficients (r > 0.90) between experimental and predicted cytotoxicity for both single (r = 0.975) and double (r = 0.983) TDP-43 mutants. The model also performed well on ALS-associated mutants (r = 0.89).
  • Droplet-Promoting Residues are More Toxic: Variants classified as "droplet-promoting" (high PDP, low MBM) exhibited significantly higher cytotoxicity compared to "amyloid-promoting" (low PDP, high MBM) or "aggregation hot-spot" variants.
  • MBM Increase in Droplet Regions: Mutations in the droplet region of TDP-43 significantly increased MBM compared to wild-type, indicating a shift towards more ordered interactions and, consequently, higher cytotoxicity.

Limitations and Future Work

  • Model Organism: The cytotoxicity was assessed in yeast, and while the study suggests general molecular mechanisms, direct translation to human cells requires further validation.
  • TDP-43 Specificity: The findings are specific to TDP-43. Future work should investigate whether these principles apply to other proteins involved in protein condensation diseases.
  • Broader Application: The authors suggest that extending this approach to other proteins and disease contexts would be a valuable next step. Investigating whether the amino acid code for cytotoxicity is similar across different proteins is also an open question.

Summary updated

Abstract

Various neurological dysfunctions are associated with cytotoxic amyloid-containing aggregates formed through the irreversible maturation of protein condensates generated by phase separation. Here, we investigate the amino acid code for this cytotoxicity using TDP-43 deep-sequencing data. Within the droplet landscape framework, we analyze the impact of mutations in the amyloid core, aggregation hot-spot, and droplet-promoting residues on TDP-43 cytotoxicity. Our analysis suggests that TDP-43 mutations associated with low cytotoxicity moderately decrease the probability of droplet formation while increasing the probability of multimodal binding. These mutations promote both ordered and disordered binding modes, thus facilitating the conversion between the droplet and amyloid states. Based on this understanding, we develop an extension of the FuzDrop method for the sequence-based prediction of the cytotoxicity of aging condensates and test it over 20,000 TDP-43 variants. Our analysis provides insight into the amino acid code that regulates the cytotoxicity associated with the maturation of liquid-like condensates into amyloid-containing aggregates, suggesting that, at least in the case of TDP-43, mutations that promote aggregation tend to decrease cytotoxicity, while those that promote droplet formation tend to increase cytotoxicity.
BibTeX
@inproceedings{horvath2022sequence,
  author = {Attila Horváth and Michele Vendruscolo and Mónika Fuxreiter},
  title = {{Sequence-based Prediction of the Cellular Toxicity Associated with Amyloid Aggregation within Protein Condensates}},
  booktitle = {Biochemistry},
  year = {2022},
  doi = {10.1021/acs.biochem.2c00499},
}
Figure