Topology-Driven Discovery of Transmembrane Protein S-Palmitoylation

Abstract

Protein S-palmitoylation is a reversible lipophilic posttranslational modification regulating a diverse number of signaling pathways. Within transmembrane proteins (TMPs), S-palmitoylation is implicated in conditions from inflammatory disorders to respiratory viral infections. Many small-scale experiments have observed S-palmitoylation at juxtamembrane Cys residues. However, most large-scale S-palmitoyl discovery efforts rely on trypsin-based proteomics within which hydrophobic juxtamembrane regions are likely underrepresented. Machine learning- by virtue of its freedom from experimental constraints - is particularly well suited to address this discovery gap surrounding TMP S-palmitoylation. Utilizing a UniProt-derived feature set, a gradient boosted machine learning tool (TopoPalmTree) was constructed and applied to a holdout dataset of viral S-palmitoylated proteins. Upon application to the mouse TMP proteome, 1591 putative S-palmitoyl sites (i.e. not listed in SwissPalm or UniProt) were identified. Two lung-expressed S-palmitoyl candidates (synaptobrevin Vamp5 and water channel Aquaporin-5) were experimentally assessed. Finally, TopoPalmTree was used for rational design of an S-palmitoyl site on KDEL-Receptor 2. This readily interpretable model aligns the innumerable small-scale experiments observing juxtamembrane S-palmitoylation into a proteomic tool for TMP S-palmitoyl discovery and design, thus facilitating future investigations of this important modification.

Keywords: S-acylation; S-palmitoylation; gradient boosting; machine learning; transmembrane protein.

Figure 1.. Experimental “detectability” of TMP S-palmitoyl sites.

The murine transmembrane proteome was subjected to in silico trypsinization, followed by labeling of Cys-containing peptides based on whether they are (A & B) reported as S-palmitoylated in SwissPalm (magenta) or (C & D) within 20 amino acids of a transmembrane domain and thus proximal to the lipid bilayer (teal). The dashed rectangle represents a general detectible region for bottom-up proteomics of 700 – 3000 Da and mean Kyte-Doolittle hydrophobicity of −2 to +1. The pie charts indicate the fraction of Cys-containing peptides that fall within the detectable range based on whether the Cys sites are (B) reported in SwissPalm or (D) juxtamembrane in location.

Figure 2.. Establishing the training dataset and exploratory data analysis.

(A) Schematic of training data isolation and pre-processing. Data were filtered to include sites confirmed by site-directed mutagenesis or ³H-palmitate radiolabeling. The seven included species were H. sapiens (58.15%), M. musculus (23.65%), R. norvegicus (11.62%), A thaliana (2.59%), S. cerevisiae (1.93%), mutant proteins from Rodenburg et al. (0.73%) and B. taurus (0.60%). Locations of S-palmitoyl sites within (B) transmembrane and (C) cytoplasmic regions. Shown alongside each region is a jitterplot of relative Cys location in Topology where 0 and 1 represent the N- and C-terminal ends of each region.

Figure 3.. Training and hyperparameter tuning of TopoPalmTree.

Shown on each y-axis are mean AUC values from the receiver operator curve vs. (A) interaction depth, (B) number of trees and (C) shrinkage rate. Chosen hyperparameters are highlighted in red. (D) Performance on 10-fold cross validation with chosen hyperparameters of interaction depth = 18, ntree = 2000, shrinkage = 0.08 and m.minobsinode = 5. Statistics relevant to the final model fitting are shown in (D).

Figure 4.. TopoPalmTree validation with a holdout dataset and benchmarking to GPS-Palm.

A dataset of 30 viral S-palmitoylated proteins (containing 82 S-palmitoyl sites) was employed for holdout given complete lack of sequence similarity to training data. Performance shown as (A) precision-recall curve and (B) F1 score vs threshold. (C) Performance of TopoPalmTree at three different thresholds (Low: 0.25, Med: 0.50, High: 0.75) compared to GPS-Palm and the 3 available thresholds of Low, Med, High.

Figure 5.. Application of TopoPalmTree for S-palmitoyl site discovery.

(A) Rank order plot of TopoPalmTree score for each Cys of the murine TMP proteome (49,828 total). Thresholds (dashed lines) are shown at 0.75 (high), 0.50 (medium) and 0.25 (low) with associated number of inferred S-palmitoyl sites at each threshold value. (B) Pie chart of database site matches (UniProt and SwissPalm combined) from the high cutoff compared to the number inferred sites that have not been reported in either database. (C) Schematic of Vamp5 topology. (D) Vamp5 probability scores from TopoPalmTree vs GPS-Palm. (E) Acyl-RAC of Vamp5-flag Cys mutants in HEK293 cells. (F) Acyl-RAC of Vamp5-flag from HEK293 cells co-treated with vehicle (DMSO) or 2-BP for 18 h. (G) Acyl-RAC of endogenous Vamp5 in murine lung. (H) Schematic of Aqp5 topology. Shown are transmembrane domains (orange) and two intramembrane regions (black) that do not traverse the entire lipid bilayer. (I) Aqp5 probability scores from TopoPalmTree vs GPS-Palm. Given rare nature of intramembrane regions not represented in the training data, TopoPalmTree does not provide a probability score for Cys¹⁸² (asterisk). (J) Acyl-RAC of Aqp5-flag Cys mutants in HEK293 cells. (K) Acyl-RAC of Aqp5-flag from HEK293 cells co-treated with DMSO or 2-BP for 18 h. (L) Acyl-RAC of endogenous Aqp5 in murine lung.

Figure 6.. TopoPalmTree facilitates in silico “design” of an S-palmitoyl site.

(A) Schematic of KdelR2 along with (B) probability scores of its 3 native Cys residues. Shown in red is position 200, which is located in the C-terminal juxtamembrane region where 9 residues showed probability scores > 0.90 by in silico mutagenesis. (C) Acyl-RAC of WT vs T200C KdelR2-flag in HEK293 cells. Hydroxylamine-dependent pulldown is indicative of protein S-acylation status.