Results - LRRC56
LRRC56
The wild-type structure of LRRC56 was predicted with a moderate level of confidence. The central Leucine-Rich Repeating structure is clearly visible and predicted with a high-degree of confidence, but the majority of the protein appears to be intrinsically disordered.
AlphaFold Confidence
N>C
Large intrinsically disordered regions are often a sign of liquid-liquid phase separation. Because of this and Lrrc56's strong localization to the DynAP, a liquid organelle, I used a suite of machine learning classifiers to predict the likelihood of this protein undergoing phase separation. For comparison, these tools were also applied to a clinical mutant isoform of Lrrc56 (L140P) and six known DynAP proteins. For a positive control, a well-characterized protein (Mex3c) known to phase separate in P granules, a different biomolecular condensate was used; GFP was used as a negative control.
deePhase v1, PSPredictor, and FuzDrop all gave similar results for most proteins. However, PSPredictor gave CFAP44 a lower score than the other two programs, classifying it as a non-liquid separating protein. catGRANULE gave unpredictable results; the positive control scored lower than the negative control, and the program's results were scaled differently than the other options. PSPer performed particularly poorly, giving low scores for all proteins including the positive control.
FuzDrop, which gave the highest phase-separation predictor score to LRRC56, also provided residue-specific droplet-promoting probabilities. Regions with high droplet-promoting probabilities are regions with a high-degree of intrinsic disorder as predicted by AlphaFold.
Lrrc56:Daap1 - AlphaFold Multimer Attempts
Protein structure prediction algorithms like AlphaFold are known for performing poorly on proteins with large IDRs, especially RNA-binding proteins, which are often involved in phase separation. Daap1, one of the six selected DynAP proteins, also has a high degree of intrinsic disorder. To see if AlphaFold would perform better when these two proteins were predicted together, I used AlphaFold2 Multimer to predict their heterodimer structure. Unfortunately, AlphaFold timed out after the fifth attempt to generate a top hit. Intermediate candidate results generated are shown above, colored by pLDDT score (AlphaFold Confidence) and by chain.
Lrrc56 mutant - patient allele Leu140Pro
While Lrrc56 is largely disordered, its eponymous repeating region is well-characterized and predicted with a high degree of confidence. The recently-reported disease-causing clinical allele consists of one Leucine to Proline missense mutation in this region. I used AlphaFold2 to predict the structure of this mutant isoform; the ordered region remained unchanged.
Lrrc56 Leu140 is highly conserved
AlphaFold Confidence
N>C
wild-type
p(Leu140Pro)
Zoomed view of Lrrc56 wild-type and mutant protein. Leucine 140 is highlighted in yellow for the wild-type protein at left; the same residue mutated to proline is highlighted at right.