Summary of TurboID approach for protein labelling in all C. elegans neurons during learning.

(A) Workflow for mass spectrometry-based analysis. Biotin-depleted animals without (non-transgenic/Non-Tg, red) or with TurboID (transgenic, yellow) were exposed to 1 mM of exogenous biotin during conditioning by pairing food with ‘no salt’ (orange - trained) or ‘high salt’ (blue - control). >3000 worms were used per group /biological replicate (n = 5) – a small proportion of each group was tested in a chemotaxis assay to assess learning capacity, while the rest was subjected to sample preparation steps for mass spectrometry (see Experimental Procedures). Some harvested protein was probed via western blot for the presence of biotinylated proteins or V5-tagged TurboID (see panel C for representative image from replicate 1). (B) The graph shows chemotaxis assay data for Non-Tg/wild type (WT) and transgenic C. elegans following salt associative learning. Each data point represents a ‘chemotaxis index’ (CI) value for one biological replicate (n = 8). Each biological replicate includes three technical replicates (26-260 worms /technical replicate). Statistical analysis: Two-way ANOVA & Tukey’s multiple comparisons test (**** ≤ 0.0001; ns = non-significant). Error bars = mean ± SEM. (C) Western blots to visualise V5-tagged TurboID and biotinylated proteins. The left side shows V5-tagged TurboID visualised using 18 µg total protein from naïve worms per lane (39 kDa). Non-Tg protein lysates acted as a negative control. α tubulin was probed as a loading control. The right side shows biotinylated proteins visualised from 25 µg total protein per lane from control (C) or trained (T) worms with streptavidin-horseradish peroxidase (HRP). (D) Venn diagram comparing all proteins assigned an identity by MASCOT from peptides detected by mass spectrometry from transgenic worms. Values represent the number of proteins listed as detected in ‘TurboID, control’ (blue) and ‘TurboID, trained’ (orange). These lists were generated by first subtracting proteins identified in corresponding Non-Tg lists, and then comparing both control and trained TurboID lists. The overlap represents proteins unique to ‘TurboID, trained’ worms in ≥1 replicate/s that were also detected in ‘TurboID, control’ worms in ≥1 other replicate/s.

Molecular pathways previously implicated in associative learning are detected in our learning proteome.

Proteins detected from ‘TurboID, trained’ worm lysates by mass spectrometry are in bold with circles coloured as orange (‘assigned hits’ assigned protein identities by MASCOT) and/or blue (‘unassigned hits’ given protein identities by bulk BLAST searching, but not MASCOT). Darker colours mean the protein has been detected in more than one biological replicate (see legend).

Schematics for metabolic processes represented in the learning proteome.

The molecular pathways above are (A) carbohydrate metabolism (glycolysis & gluconeogenesis) and (B) fatty acid metabolism (via the tricarboxylic acid or TCA cycle). Each protein is a node in white (not detected by TurboID during learning), orange (an ‘assigned hit’), and/or blue (an ‘unassigned hit’) based on mass spectrometry data from ‘TurboID; trained’ worms. Darker colours mean the protein has been detected in more than one biological replicate (see legend).

Neuron-specific expression within the learning proteome.

Mass spectrometry runs (n = 5) were performed with the ThermoFisher Scientific Q-Exactive Orbitrap (‘QE’) and/or ThermoFisher Scientific Orbitrap Exploris (‘Exploris’), for technical reasons. There are six lists because replicate #3 was run on both mass spectrometers: corresponding protein lists are annotated as ‘3a’ and ‘3b’ respectively. The total ‘#Assigned hits’ versus ‘#All hits’ (assigned + unassigned hits) are shown in rows are listed above. The CeNGEN database (threshold = 2) was used to determine corresponding percentages for assigned hits versus all hits as ‘% Neuronal for assigned hits’ versus ‘% Neuronal for all hits’ (Taylor et al., 2021). The average percentages across all replicates were 91% for assigned hits only versus 89% for all hits.

Schematic for neural circuit containing neuron classes represented by the learning proteome.

Neuron classes represented by the learning proteome were identified using the gene enrichment tool from WormBase (Angeles-Albores et al., 2016) and the CeNGEN database (threshold = 2) (Taylor et al., 2021). Neurons are represented by pink triangles (sensory), orange pentagons (interneurons), and purple circles (motor neurons). Chemical synapse (black arrows) and gap junction (dotted arrows: grey for gap junctions only or yellow for synapses and gap junctions) information is provided using the software WormWeb (Bhatla, 2009). Learning regulators validated in this study are also represented: ACC-1 (brown rectangles), ACC-3 (pink squares) and LGC-46 (purple diamonds) are annotated above based on single neuron expression profiles from CeNGEN (Taylor et al., 2021). Notably, KIN-2 and F46H5.3, discussed in detail below, are expressed in all neurons shown except for DD.

C. elegans PKA regulatory subunit KIN-2 acts in neurons to regulate salt associative learning.

Salt chemotaxis behaviour was measured in the form of chemotaxis indices (CI) for naïve/untrained worms (grey circles), high-salt control (blue squares), and trained worms (orange triangles). This was done for (A & B) wild-type (WT) animals, (A & B) kin-2(ce179) mutants, and (B) transgenic worms with a WT background engineered to overexpress KIN-2 from the ce179 allele in all neurons (10-60% transgenic worms per technical replicate, both non-transgenic (-) and transgenic (+) siblings are plotted above). Each data point represents one biological replicate where (A) n = 5 and (B) n = 3 (one biological replicate was excluded from high-salt control and trained kin-2(ce179) groups due to insufficient sample size). (A) 32-487 worms and (B) 5-184 worms per technical replicate. Error bars = mean ± SEM. Annotations above graphs represent P-values from Two-way ANOVA and Tukey’s multiple comparison tests (**** ≤ 0.0001; *** ≤ 0.001; ** ≤ 0.01; * ≤ 0.05; ns = non-significant). (B) Statistical comparisons between WT trained and siblings in transgenic lines are in red (top row), between adjacent trained groups are in green (middle row), and between groups within each line in black (bottom row).

Summary of candidates assessed for their effect in learning.

The number (#) of biological replicates (total n = 5) in which each candidate was detected as an assigned hit (by the MASCOT software) or in assigned + unassigned hits (identified by bulk BLAST search) is provided under ‘# Biological replicates in TurboID trained’ and ‘ # Biological replicates in TurboID high-salt control’ columns. These values exclude proteins from non-transgenic trained and non-transgenic high-salt control groups, respectively. Orange highlights indicate candidates detected in more replicates in the TurboID-trained group. Candidates are also defined as ‘weak’ or ‘strong’ based on the frequency of detection across biological replicates.

Salt associative learning is dependent on arginine kinase F46H5.3 and not armadillo-domain containing protein C30G12.6.

Chemotaxis indices (CI) are shown for wild-type/WT animals versus mutants for (A) F46H5.3 (non-backcrossed with WT, n = 5), (B) F46H5.3 backcrossed with WT (n = 4), (C) C30G12.6 (non-backcrossed with WT, n = 5), and (D) C30G12.6 backcrossed with WT (n = 5). These animals were assessed for salt associative learning by preparing three groups for each line: naïve/untrained (grey circles), high-salt control (blue squares), and trained (orange triangles) (27-395 worms per technical replicate). Each data point is for one biological replicate each comprising of three technical replicates. Error bars = mean ± SEM. Statistical analyses were done by Two-way ANOVA and Tukey’s multiple comparison test (**** ≤ 0.0001; ** ≤ 0.01; * ≤ 0.05; ns = non-significant).

Learning regulators KIN-2 and F46H5.3 may modulate learning through calcium signalling pathways.

Pathway components present within the learning proteome are shown with protein names in bold. Darker colours mean the protein has been detected in more than one biological replicate (see legend). Orange and/or blue circles represent candidates that are ‘assigned hits’ and/or ‘unassigned hits’ respectively. Orange dotted arrows denote protein-protein interactions predicted by STRING (version 12.0), whereas black arrows are based on known interactions.

Western blot to quantify biotin-tagged protein levels following biotin exposure.

C. elegans used to generate this blot were either wild-type/non-transgenic/Non-Tg animals or transgenic worms expressing TurboID in all neurons. Animals were grown for at least two generations via their diet (i.e., E. coli bioB::kan MG1655), and then their biotin-depleted progeny (young hermaphrodite adults) were lysed by sonication for protein extraction. Whole worms were lysed immediately (-) or fed MG1655 supplemented with 1 mM biotin for 6 h (+) before lysis (+). Biotin-tagged proteins were probed with streptavidin-horseradish peroxidase (40 μg of total protein /lane). Here, the sum of areas containing biotin-tagged protein in each lane is interpreted as signal intensity for these proteins. By comparing this signal for untreated and treated lanes within each worm line, treatment was seen to increase signal (1.3-fold for non-Tg and 1.7-fold for transgenic C. elegans). The leftmost lane contains protein standards (kDa, sizes annotated on the left).

Design of chemotaxis assay plates to quantify salt chemotaxis behaviours.

(A) Method used to generate a salt concentration gradient (0-200 mM NaCl) on chemotaxis assay plates: 5 mm cubes of no-salt/salt-deficient agar with or without high-salt (200 mM NaCl) were placed on top of salt-deficient agar, 5 mm from the plate edge (6.0 cm diameter). Cubes were incubated overnight at 22 °C, or incubated sequentially for (1) 5.5 h at 37 °C, (2) 15 min at 15 °C, and then (3) 0.5-2 h at 22 °C. Each cube was replaced with a 1 µL spot of 0.1 M sodium azide <2 min before use. (B) Template used to quantify salt chemotaxis on chemotaxis assay plates: regions of high salt (orange) and low salt (blue) were defined as a circle (diameter = 2.4 cm). These regions were cut-off on their longitudinal axis since they were centred 0.5 cm from the edge of each plate. Worms were transferred onto the region of origin (green; 2 cm short axis and 2.4 cm long axis). Points where agar cubes or worms were transferred onto are represented by a small black circle.

Western blots to assess biotinylation in C. elegans by TurboID during memory encoding of salt associative learning.

Each lane contains whole worm lysate from wild-type/non-transgenic animals (Non-Tg, red) or TurboID-positive transgenic C. elegans (TbID, yellow). Worms were biotin-depleted for at least two generations via their diet (i.e., E. coli bioB::kan MG1655), and then treated with 1 mM of exogenous biotin for 6 h. This biotin treatment was paired with salt associative learning to generate high-salt control/C (food + salt, blue) and trained/T worms (food + no salt, orange) for protein extraction. Each panel corresponds to a biological replicate of mass spectrometry data with TurboID as annotated above (underlined in black). Each lane contains (A) 50 μg, (B, D) 40 μg, or (C) 20 μg of total protein, with biotin-tagged proteins visualised using streptavidin-horseradish peroxidase. For panel (C), unannotated lanes correspond to protein samples not relevant to this experiment. The leftmost lane for each image contains protein standards (kDa, sizes annotated on the left). Comparisons regarding signal intensity between each experimental group are in Table S2.

Analysis of subcellular localisations of assigned protein hits in the learning proteome.

Analyses based on gene ontology (GO) terms were performed with ClueGO (version 2.5.10) using Cytoscape (version 3.10.0). (A) Pie chart summarising the percentage of all proteins in ‘TurboID, trained’ protein lists assigned GO terms within different ‘cellular components’ (from Table S3). Abbreviations: ECM = extracellular matrix. (B-G) Protein-protein interaction (PPI) networks of proteins in the learning proteome assigned to cellular components most relevant to the nervous system: (B) Neuronal cell body/cytoplasm, (C) Cytoskeleton, (D) Cilia and dendrites, (E) Axon, (F) Pre-synapse, and (G) Vesicles. PPI interactions were exported from STRING (version 12.0) and visualised with Cytoscape. Arrows from networks in B-G annotate the relevant region on a neuron schematic. Highlighted proteins were further tested in this study. Known information on subcellular localisation and predictions based on homology were also used for several nodes (Gaudet et al., 2011, Takayanagi-Kiya et al., 2016, Lans et al., 2004, Arellano-Carbajal et al., 2011, Popovici et al., 2002). The consolidated data from GO term analyses can accessed on Table S5.

Analysis of molecular functions of assigned hits in the learning proteome.

Analyses using GO terms for molecular function were performed with ClueGO (version 2.5.10) using Cytoscape (version 3.10.0). Protein-protein interaction (PPI) networks generated for proteins assigned the GO term (A) G protein signalling, (B) Insulin and protein kinases, (C) Neurotransmission, (D) Protein synthesis, and (E) Protein degradation. PPI interactions were exported from STRING (version 12.0) to be visualised with Cytoscape using the protein lists in Table S3. Highlighted proteins were tested in this study using appropriate genetic mutants. Known information on molecular function and predictions based on homology were also used for several nodes (Putrenko et al., 2005, Park et al., 2000, Gottschling et al., 2017, Gaudet et al., 2011, Hayashizaki et al., 1998, Lans et al., 2004). The consolidated data from GO term analyses can accessed on Table S6.

Assessing overlap for proteins detected in trained worms across all biological replicates.

(A-B) Numbers represent (A) assigned proteins only or (B) all (assigned and unassigned) proteins detected with the Q Exactive (QE) (left) or the Exploris (right) in ‘TurboID, trained’ animals only; these proteins were identified by subtracting proteins also in ‘non-transgenic, trained’ and/or in ‘TurboID, control’ worms. Six protein lists are compared here since one replicate was run in both mass spectrometers. (C-D) Lists containing proteins specific to trained animals for each biological replicate were compared based on the mass spectrometer used for (C) assigned proteins only and (D) all proteins.

Behavioural testing for strong candidates identified by TurboID-based mass spectrometry experiments.

Worms with single mutations targeting molecular components in the following pathways were tested for their salt associative learning capacity, compared to wild-type (WT) worms: (A & B) Acetylcholine signalling: acr-1 (n = 3) and elp-1 (n = 2), (C) G protein signalling (gap-2, n = 5), (D & E) p38/MAPK signalling: uev-3 (n = 5) and fsn-1 (n = 3), (F) IGCAM (rig-4, n = 3), (G & H) Guanyl nucleotide exchange factors: aex-3 (n = 2) and tag-52 (n = 3), and (I-M) Other proteins of interest: elo-6 (n = 2), ift-139 (n = 6), tap-1 (n = 4), saeg-1 (n = 3), and ver-3 (n = 5). Each data point represents the average chemotaxis index (CI) per biological replicate (three technical replicates per biological replicate, 22-540 worms per technical replicate). Statistical analysis: Two-way ANOVA & Tukey’s multiple comparisons test (**** ≤ 0.0001; *** ≤ 0.001; ** ≤ 0.01; * ≤ 0.05; ns = non-significant). Error bars = mean ± SEM.

Behavioural testing for weak candidates in the learning proteome.

Weak candidates from TurboID-based mass spectrometry data were tested by assessing learning capacity for single mutant C. elegans, compared to wild-type (WT) worms. Each mutation targets a molecular component in one of the following pathways: (A, B, & C) Acetylcholine signalling: acc-1, acc-3, or lgc-46 (n = 5), (D, E, & F) Neurotransmission: gbb-2 (n = 5), glr-1 (n = 3), or maco-1 (n = 3), (G, H, & I) G protein signalling: gap-1, (n = 4), gpa-2 (n = 5), or rho-1 (n = 6), or (J) p38/MAPK signalling (dlk-1, n = 2). One biological replicate (n) constitutes each data point, with three technical replicates per biological replicate (23-372 worms per technical replicate). Error bars = mean ± SEM. Statistical analysis: Two-way ANOVA & Tukey’s multiple comparisons test (**** ≤ 0.0001; *** ≤ 0.001; ** ≤ 0.01; * ≤ 0.05; ns = non-significant).

Salt aversive learning is modulated by arginine kinase F46H5.3.

Worms were naïve, mock-conditioned with no salt + no food, or conditioned with salt + no food. Three technical replicates were generated per biological replicate (26-481 worms per technical replicate). (A) n = 4 and (B-E) n = 3 biological replicates. Each data point represents a chemotaxis index (CI) for an individual biological replicate. Statistical analysis: Two-way ANOVA & Tukey’s multiple comparisons test (**** ≤ 0.0001; *** ≤ 0.001; ** ≤ 0.01; * ≤ 0.05; ns = non-significant). Error bars = mean ± SEM.