An expanded toolkit of synaptic reporters to study C. elegans synapses.

(A) A schematic of reporters used to visualize C. elegans synapses. Chemical synapses were visualized using fluorophore-tagged CLA-1 or RAB-3 proteins to label presynaptic specializations, or synapse-specific reporters using transsynaptic tools such as NLG-1-based GRASP. Electrical synapses were visualized by directly tagging constituent innexin proteins with a fluorophore. (B) A schematic showing all neuron classes in the adult C. elegans hermaphrodite. Neurons for which new reporters are described in this paper are colored in orange. Neurons for which reporters have been previously published but are quantified with WormPsyQi in this paper are colored in green.

WormPsyQi robustly segments neurites and quantifies synapses.

(A) A flowchart of the WormPsyQi pipeline. The pipeline is automated except for the correction process, which can be bypassed if the initial segmentation result is good enough. (B) Representative input raw and output segmentation images for the interneuron pair AIB. AIB-specific presynaptic specializations and processes were visualized using GFP-tagged CLA-1 (otIs886) and a cytoplasmic marker (otEx8023). Predicted puncta in the lower panel are colored arbitrarily to represent discrete fluorescent signal. Cell bodies are excluded from the mask to exclude any colocalizing puncta in downstream steps. Images are maximum intensity projections of confocal Z-stacks. (C, D) Synapse segmentation accuracy is assessed with IoU score between the ground-truth label and the prediction label. Calculations were performed on six images of the GFP::CLA-1 reporter for neurons I5 (otEx7503). The images were labeled by four different experts. (C) The IoU score over the size of the training set shows that having one labeled image as a training set is enough to reliably predict synapses for the same conditions of images. In bars 1-3 (from left to right), one image was selected as the training set and then the given number patches in the column label were randomly sampled after the regular patch sampling step of training (see Methods) to further reduce the size of the training set. In bars 4-6, the number of images in the column label were selected as the training set. For each expert’s labels, all possible combinations of the training set were tested and the IoU score for the held-out test set is shown. (D) The average IoU score between different experts’ labels is taken as a benchmark (bar 1, leftmost). The IoU scores of WormPsyQi prediction are significantly greater than the benchmark IoU when the same expert’s labels used in the training step were taken as the ground-truth to compare with, both when 1 image and 5 images were used for training. If another expert’s labels were taken as ground-truth, then the IoU score of WormPsyQi prediction was lower but still demonstrated significant improvement compared to the benchmark IoU. P values were calculated using one-way ANOVA with Bonferroni correction for multiple comparisons ***P ≤ 0.001, **P ≤ 0.01, *P ≤ 0.05 and P > 0.05 not significant (ns).

WormPsyQi validates sexually dimorphic synapses in C. elegans.

(A) Subsets of hermaphrodite (red) and male (blue) connectivity diagrams, based on EM studies (Cook et al., 2019; White et al., 1986), showing adult stage sexually-dimorphic synapses analyzed for validating WormPsyQi. Synapses, depicted by arrows, were visualized using GRASP or iBLINC reporters generated either for previous studies (PHA>AVG, ADL>AVA, and PHB>AVG in Cook et al., 2019; Bayer et al., 2018; Oren-Suissa, Bayer, and Hobert, 2016) or this paper (PHB>AVA). Sensory, inter- and motor neurons are depicted as triangles, hexagons, and circles, respectively. (B) WormPsyQi validates sex-specific synapses in adult hermaphrodites and males. PHB>AVA, PHA>AVG, ADL>AVA, and PHB>AVG synapses were visualized using the transgenes otIs839, otIs630, otEx6829, and otIs614, respectively. Panels corresponding to each reporter show raw confocal images (top) and segmented neurites and synapses (bottom). Segmentation and quantification were performed using WormPsyQi. Synchronized day 1 adult animals were scored. Red – hermaphrodite, blue – male. P values were calculated using an unpaired t-test. ****P ≤ 0.0001, ***P ≤ 0.001, **P ≤ 0.01, and *P ≤ 0.05. In each dataset, a dot represents a single worm and lines represent median with 95% confidence interval. All raw and segmented images are maximum intensity projections of confocal Z-stacks. Scale bars = 10μm.

WormPsyQi is generalizable across diverse synaptic reporters targeting the synapses in the central nervous system

(A) Representative images of cell-specific synaptic reporters showing presynaptic specializations, as visualized with GFP-tagged CLA-1, for neuron classes ASK (otIs789), AFD (otEx7786), IL1 (otEx7363), IL2 (otIs815), RIB (otIs810), HSN (otIs788), AIB (otEx8023; otIs886), and PHB (otIs883). (B) Representative images of cell-specific synaptic reporters showing presynaptic specializations, as visualized with fluorescently-tagged RAB-3, for neuron classes PHA (otIs702) and ASK (otEx7231). In each panel, top (raw) and bottom (segmented) images represent both neuronal processes, based on a cytoplasmic marker, and synaptic puncta. There was no cytoplasmic marker in the PHB reporter used, so only puncta are shown. All images are maximum intensity projections of confocal Z-stacks. Quantification of the number of puncta, performed using WormPsyQi, is shown in corresponding graphs. In each graph, a dot represents a single worm and lines represent median with 95% confidence interval. L4 animals were scored unless otherwise noted. Scale bars = 10μm.

WormPsyQi can quantify synapses in variable GRASP and iBLINC reporters

(A) Representative images of NLG-1 GRASP-based synaptic reporters labeling the following synapses: ASK>AIA (otIs653), ADL>AIA (otEx7457), PHB>AVA (otIs839), AIB-SAA (otEx7809), SAA-AVA (otEx7811), BDU>HSN (otEx7759), and PHB>AVG (otIs614). (B) Representative images of iBLINC-based synaptic reporters labeling PHA>AVG (otIs630) and LUA>AVG (otEx6344) synapses. WormPsyQi vs. manual quantification is shown in corresponding plots; EM data is shown where available (Witvliet et al., 2021; Cook et al., 2019; White et al., 1986), but was excluded from statistical analyses due to small sample sizes (n = 1 - 3). In each dataset, a dot represents a single worm and lines represent median with 95% confidence interval. All raw and segmented images are maximum intensity projections of confocal Z-stacks. P values were calculated using an unpaired t-test. *P ≤ 0.05 and P > 0.05 not significant (ns). Scale bars = 10μm.

Systematic analysis of pharyngeal synapses using WormPsyQi.

CLA-1 or RAB-3 based presynaptic specializations of pharyngeal neurons (A) I5, (B) MC, (C) NSM, (D) I1, (E) I2, and (F) M4 analyzed using transgenes otEx7503, otEx7505, otEx7499, otEx7501, and otIs597, respectively. For each reporter, a corresponding graph shows manual vs. WormPsyQi quantification, along with EM synapse counts as described in Cook et al., 2020. Day 1 adults were scored, with some datasets previously published (Cook et al., 2020) and re-analyzed here. Masking was used prior to synapse prediction for all neurons except I2, where the cytoplasmic reporter was too dim for creating a continuous mask. P values were calculated using an unpaired t-test. ****P ≤ 0.0001, *P ≤ 0.05, and P > 0.05 not significant (ns). EM data was excluded from statistical analyses based on limited sample size (n=1). In each dataset, a dot represents a single worm and lines represent median with 95% confidence interval. All raw and segmented images are maximum intensity projections of confocal Z-stacks. Scale bars = 10μm.

An analysis of electrical synapses using WormPsyQi

INX-6::GFP (ot805) puncta are observed in the pharynx and the nerve ring. Nerve ring gap junctions are compared between L4 and dauer animals to validate previous observations (Bhattacharya et al., 2019). Given that outside of the pharynx, inx-6 is exclusively expressed in AIB interneurons in dauers, the puncta can be assigned to AIB axons as shown previously (Bhattacharya et al., 2019); in the absence of a cytoplasmic marker, WormPsyQi scores a limited number of puncta in L4 animals, but the dauer count significantly exceeds L4 count, as expected. Red arrows denote AIB-localizing electrical synapses and yellow arrows denote electrical synapses in the pharyngeal muscles. All raw and processed images are maximum intensity projections of confocal z-stacks. Statistical analysis was performed using an unpaired t-test. Nerve ring synapses are denoted by a yellow box; synapses outside the box are pharyngeal. In each dataset, a dot represents a single worm and the height of the bar represents median with 95% confidence interval. Images are maximum intensity projections of confocal Z-stacks. Scale bars = 10μm.

Developmental synapse addition in the CNS roughly scales with neurite length in strains analyzed.

(A) Representative quantification of the total nerve ring length of late-L1 and day 1 adult animals, as analyzed with an ASK-specific cytoplasmic transgene (srg-8p::tagRFP in reporter otIs789). Total ASK axon length in the nerve ring increases ∼2-fold between late-L1 and adulthood. Dashed lines mark axons in laterally-positioned worms. (B) Representative images of an ASK-specific CLA-1 reporter (analyzed using the transgene otIs789) at late-L1 and day 1 adult stages. The graph shows the number of puncta representing ASK-specific presynaptic specializations quantified using WormPsyQi. Number of puncta increases ∼2-fold, which is in proportion with the increase in axon length across development. (C-E) Representative images and WormPsyQi-based quantification of synapse addition using synapse-specific GRASP reporters visualizing (C) ASK>AIA (otIs653), (D) AIB>SAA (otEx7809) (E), and PHB>AVA (otIs839) synapses at early larval (L1 or L2) and L4 or adult (Ad) stages. P values were calculated using an unpaired t-test (A, B, D, E) or one-way ANOVA with Bonferroni correction for multiple comparisons (C). ****P ≤ 0.0001 and ***P ≤ 0.001. EM data (taken from Witvliet et al., 2021; Cook et al., 2019; White et al., 1986) was excluded from statistical analysis based on limited sample sizes (n = 1 - 4) which were out of proportion with reporter-based data. In each dataset, a dot represents a single worm and lines represent median with 95% confidence interval. All raw and segmented images are maximum intensity projections of confocal Z-stacks. Scale bars = 10μm.

Synapse addition in the enteric nervous system does not scale with neuron growth over developmental.

(A) Overview of pharynx and total body length elongation between L1 and adult stages based on previous findings (Witvliet et al., 2021; Shibata et al., 2016). (B-F) Representative images and quantification of number of presynaptic specializations at early larval and adult stages visualized using (B) GFP::RAB-3 in MC (otEx7505), (C) GFP::CLA-1 in I2 (otEx7497), (D) GFP::RAB-3 in M3 (otIs602), (E) GFP::RAB-3 in M4 (otIs597), and (F) GFP::CLA-1 in NSM (otEx7499) neurons. Manual counts of total CLA-1 or RAB-3 puncta are plotted for each reporter (black boxes). For neurons MC, M3, M4, and NSM, region of interest (ROI, yellow box) length was measured from the pharyngeal nerve ring to the most posterior CLA-1 or RAB-3 signal observed. For I2 neurons, ROI length was measured from the pharyngeal nerve ring to the tip of the anterior end of the pharynx based on brightfield. For NSM neurons, ROI length was measured from the pharyngeal nerve ring to the grinder in the terminal bulb, based on brightfield. For neurons M4 and NSM, additional quantification was performed for the pharyngeal nerve ring region (NR, blue box) and the isthmus (purple). P values were calculated using Mann-Whitney test. ****P ≤ 0.0001, ***P ≤ 0.001, **P ≤ 0.01, and P > 0.05 not significant (ns). In each dataset, a dot represents a single worm and lines represent median. All images are maximum intensity projections of confocal Z-stacks. Scale bars = 10μm.

Extracting subcellular and spatial synaptic features using WormPsyQi.

(A-C) Illustration of distribution profile of RAB-3 puncta in PHA, CLA-1 puncta in AIB, and CLA-1 and RAB-3 puncta in ASK neurons. Top: Schematic diagram of neurons (source: WormAtlas). Middle left: raw image data with contrast enhancement and maximum projection. Middle right: skeleton diagram of neurons. Position 0 indicates SOMA and 1 indicates the furthest axon terminal. Bottom: Probability density histogram of puncta distribution along the neurite process. Distribution profiles of worms at the same stage were combined to give a histogram (see Methods). (C) RAB-3 and CLA-1 puncta profiles are overlaid. (D) Example features describing developmental changes of ASK neuron’s connections. Top: puncta volume and puncta number per worm, from left to right. Bottom: puncta number density and puncta volume density per worm (calculated along the neurite domain covered by synapses). P values were calculated using one-way ANOVA with Bonferroni correction for multiple comparisons. ***P ≤ 0.001, **P ≤ 0.01, *P ≤ 0.05 and P > 0.05 not significant (ns). In each dataset, a dot represents a single worm. (E) Synaptic puncta density distribution in ASK neurons at different stages. For each worm, the synaptic volume distribution is normalized by axon length.

(A) WormPsyQi pipeline schematics with representative input and output. Input images are typically 3d multicolor stacks. Neurite segmentation process generates a 3d binary image labeling regions where synapses could be localized. Synapse segmentation process takes the input image and the neurite mask and generates a binary image labeling synapse. The example image is showing the connected components from the binary synapse label colored differently and overlaid with the neurite label in gray for visualization. Synapse quantification process extracts various synapse- and image-wise quantities and saves them as .csv format. (B) WormPsyQi main GUI. WormPsyQi provides a GUI to easily manage the processing parameters and the output data storage. (C) WormPsyQi GUI for reviewing and correcting synapse segmentation. Users can quickly scroll through the slices to visually inspect the quality of the synapse segmentation. False-positive synapses can be reviewed and rejected individually, or ROI can be selected to reject synapses in/out of the ROI.

(A) 2.5D U-Net architecture for neurite segmentation. The input is a chunk with seven slices and the output is the segmentation mask of the center slice, with intermediate convolution blocks that combine depth-wise spatial context in the concatenation path. The structure of the encoding block is detailed in panel A’. (B) Flowchart for synapse segmentation. Rectangles represent a process or action, and parallelograms represent an input, output, or intermediate data.

WormPsyQi performs optimally on images with a cytoplasmic marker and discrete puncta.

(A) Manual vs. WormPsyQi quantification for representative GFP::CLA-1 (for neurons AFD, ASK, HSN, PHB, IL1 and IL2) and GFP::RAB-3 (for neurons PHA and ASK) reporters. No significant difference was observed in all cases besides PHB, where the masking step was skipped. The bottom row denotes whether or not a neurite mask was used prior to synapse segmentation in the pipeline. P values were calculated using an unpaired t-test. **P ≤ 0.01 and P > 0.05 not significant (ns). In each dataset, a dot represents a single worm and lines represent median with 95% confidence interval. (B-C) Representative raw and segmented images and quantification of RAB-3 (B) and CLA-1 (C) reporters for ASK neurons. Panels C’ and D’ show zoomed in regions of ASK presynaptic specializations, depicting the relative difference in puncta resolution along the nerve ring. More ASK presynaptic specializations were scored in the CLA-1 reporter. (D) Images of a representative subset of reporters for neurons RME, SAA, and AIB, for which WormPsyQi could not segment and quantify puncta because of diffuse signal. Specializations were visualized using fluorescently-tagged presynaptic SNB-1 and RAB-3. All images are maximum intensity projections of confocal Z-stacks. Scale bars = 10μm.

WormPsyQi analysis of all presynaptic specializations in the pharynx.

Representative images and quantification of a pan-pharyngeal synaptic reporter (otIs785). Neuronal processes were labeled with ceh-34p::TagRFP and presynaptic specializations were visualized with ceh-34p::GFP::CLA-1. Raw (top), segmented (middle), and overlap (bottom) images are shown. Quantification of puncta in the whole pharynx (‘Total’) and posterior pharyngeal bulb (blue) are shown in the graph. In each dataset, a dot represents a single worm and lines represent median. Compared to the posterior bulb the spread is greater for total count, where synapse segmentation was suboptimal due to a high synapse density and inseparable processes. All images are maximum intensity projections of confocal Z-stacks. Muscle expression from ceh-34p is marked with an asterisk. Scale bars = 10μm.

Manual and WormPsyQi quantification show the same developmental trends in synapses in the pharynx.

Graphs showing manual vs. WormPsyQi puncta quantification for presynaptic specializations in pharyngeal neurons (A) I2, (B) M3, (C) NSM, and (D) MC in early larval (L1, L2) and late larval (L4) or adult (Ad) stages. Although WormPsyQi scores fewer puncta relative to manual counts in L1 and L2 animals, the trends between the two modes of scoring are consistent with modest synapse addition in neurons I2, M3, and MC, and significant addition in neurons NSM. P values were calculated using one-way ANOVA with Bonferroni correction. ****P ≤ 0.0001, **P ≤ 0.01, *P ≤ 0.05 and P > 0.05 not significant (ns). In each dataset, a dot represents a single worm and lines represent median with 95% confidence interval.

Distribution of CLA-1 puncta in individual worms.

Individual CLA-1 puncta distribution in cell-specific reporters for neurons PHA, AIB, and ASK (population-level quantification and statistics summarized in Figure 10). Each column in the graphs corresponds to an individual worm. Gray dots denote discrete puncta, and the size of dots corresponds to the synapse volume. The y-axis represents the normalized distance from the soma, with 0 being the closest and 1 being the farthest point in the region of interest. (A) PHA RAB-3 adults, (B) AIB CLA-1 L4s, (C) ASK CLA-1 L1s, (D) ASK CLA-1 L4s, (E) ASK RAB-3 L4s, and (F) ASK CLA-1 adults.

Extended features and representative two-dimensional scatter plots of ASK CLA-1 reporter across different developmental stages.

(A) Left: Average volume of synaptic puncta. Right: Average fluorescence intensity. Each dot in a dataset represents a single worm. P values were calculated using one-way ANOVA with Bonferroni correction for multiple comparisons ***P ≤ 0.001, and P > 0.05 not significant (ns). (B) Two examples illustrating two-dimensional synapse ultrastructure analysis. Color coding is consistent with Figure 10. Left: Total number vs. average volume of puncta at L1, L4, and adult stages. In this representation, L1 stage worms are clustered separately from L4 and adult stages, which are grouped together. Right: Number density vs. average intensity of puncta; developmental stages are not clearly differentiated in this feature space.