Serial Synapse Formation through Filopodial Competition for Synaptic Seeding Factors

Abstract

Following axon pathfinding, growth cones transition from stochastic filopodial exploration to the formation of a limited number of synapses. How the interplay of filopodia and synapse assembly ensures robust connectivity in the brain has remained a challenging problem. Here, we developed a new 4D analysis method for filopodial dynamics and a data-driven computational model of synapse formation for R7 photoreceptor axons in developing Drosophila brains. Our live data support a "serial synapse formation" model, where at any time point only 1-2 "synaptogenic" filopodia suppress the synaptic competence of other filopodia through competition for synaptic seeding factors. Loss of the synaptic seeding factors Syd-1 and Liprin-α leads to a loss of this suppression, filopodial destabilization, and reduced synapse formation. The failure to form synapses can cause the destabilization and secondary retraction of axon terminals. Our model provides a filopodial "winner-takes-all" mechanism that ensures the formation of an appropriate number of synapses.

Keywords: Development; Drosophila; axon; brain; filopodia; live imaging; modeling; synapse formation.

Figure 1:. 4D filopodia tracking reveals stochastic dynamics prior to synapse formation and rare ‘bulbous’ filopodia that stabilize one at a time during synapse formation

(A, B) Drosophila R7 photoreceptor axon terminals transition from a growth cone-like structure with multiple filopodia just prior to synapse formation at 50% pupal development (P50) to a smooth, adult terminal with 20-25 synapses (magenta: CD4-tdTomato, green: BrpD3-GFP). (C-D) A semi-automatic method for 4D filopodia tracking is based on the Amira filament editor. (E) Imaging protocol for >20 hour continuous time lapse and fast imaging at P40 and P60 for the same axon terminals. (F) R7 filopodia fall into short-lived and long-lived classes that both fit Poisson (stochastic) distributions at both P40 and P60. (G) Representative snapshots of an R7 axon terminal in the brain at P60 with a continuous stable bulb (yellow arrowhead). (H) Number of bulbous filopodia at P60. Separation into stable (middle) and transient (right) bulbs reveals that most time points contain 1-2 stable bulbs. The distributions can be fit with negative feedback (sold black lines), but not with Poisson product distributions (dotted lines). Scale bar: 2μm.

Figure 2:. One filopodium at a time accumulates synaptic seeding factors

(A-F) Localization in R7 photoreceptor terminals and filopodia for BrpD3-GFP (A, B), GFP-Syd-1 (C, D) and Liprin-α-GFP (E, F). Shown are two time points: P50 (A, C, E) and P70 (B, D, F). Yellow circles indicate filopodia with no measurable GFP signal, green circles weak signal, and blue circles clear accumulations. (A’-F’) show the single channel for the GFP-tagged proteins (green), and (A”-F”) show the single channel for the membrane tag CD4-tdTomato (magenta). Scale bar: 2 μm. (G) Quantification of filopodial accumulation of the three proteins. (H) Number of BrpD3 punctae per R7 terminal binned according to their lifetimes. R7 terminals were live imaged at 10 min resolution starting at P+50% + 22h in culture. Individual punctae were tracked for 5,5h to determine lifetimes (n = 5 terminals). Error bars denote SEM.

Figure 3:. A data-driven computational model predicts ‘serial synapse formation’ based on competition and negative feedback of bulbous filopodia

(A) Summary of the data-driven Markov state model from filopodial birth to synapse formation. All rates in blue are measured from live imaging data. Rates r₁ and r₂ denote the generation and retraction of filopodia, r₃ and r₄ denote the formation- and degeneration of a bulbous tip; r₅ denotes the stabilization of the bulbous tip and r₆ the formation of a synapse. (B) Estimation of time-dependent function required for the modeling from P40-P100 (40%-100% of pupal development). The reduction of filopodia was based on measured filopodial counts from fixed preparations (blue disks = average, error bars = standard deviation) and modelled by a time-dependent function f_F(t) (dashed red line) as outlined in the Methods section. The increased propensity to form bulbs on these filopodia (black dashed line) was estimated based on bulb measurements shown in panel (D) and as explained in the Methods section. (C) Output of Markov state model for filopodial dynamics based on measured rates according to the model in (A). Solid red lines indicate the median number of bulbs from the stochastic simulations, whereas dark grey areas denote the interquartile range (50% of the data) and light grey the 95% confidence range from the simulations. (D) Measured number of bulbous tips (disks = average, error bars = standard deviation). (E) Output of Markov state model for the development of bulbous tips. Black dotted lines: average number of bulbs; solid red line: median number of bulbs; grey confidence ranges as in (C). (F) Measured numbers of synapses between P40 and P100 (disks = average, error bars = standard deviation). (G) Output of Markov state model for synapse formation. Black dotted lines: average number of bulbs; solid red line: median number of bulbs; grey confidence ranges as in (C).

Figure 4:. Loss of synaptic seeding factors Syd-1 and Liprin-α causes a loss of inhibitory feedback and filopodial bulb destabilization

Analyses of filopodial dynamics and synapse formation for syd-1 (green), liprin-α (red) and control (blue). (A) Lifetime of bulbous filopodia. (B) Total number of bulbous filopodia per terminal per hour. (C) Average number of bulbous filopodia per time instance. (D) Number of concurrently existing bulbous filopodia per axon terminal per time instance observed in the data, simulated after inclusion of a feedback (+f1) and without a feedback (−f1). (E, F) Representative snapshots of syd-1 (E) and liprin-α (F) revealing only transient bulbs. (G-N) Markov state model simulation for syd-1 (G-J) and liprin-α (K-N) for the numbers of filopodia (G, K), transient bulbs (H, L), stable bulbs (I, M) and synapses (J, N). In all cases control traces from Fig. 3 are shown in yellow. Black dotted lines: mean number of bulbs; solid red line: median number of bulbs; dark grey denotes the interquartile range (50% of the data) and light grey the 95% confidence range. (O-R) Measurement of BrpD3 punctae in mutant axon terminals. (O’-R’) BrpD3 single channel. Scale bar: 2 μm. (S, T) Quantification of BrpD3 synapse numbers per terminal relative to control. n=18 and 16 (p = 0.0007). (T) Number of BrpD3 punctae per terminal with lifetimes greater than 3h in R7 axons live imaged for 4h at P+70% in wild-type (n=5) and liprin-α^E mutants (n=5). (U) Quantification of synapse numbers in syd-1^ΔRhoGAP flies. n = 45, 18 and 32 (p < 0.0001). Error bars denote SEM.

Figure 5:. Analysis of the Syd-1/Liprin-α pathway components reveal a role for Lar, but not Trio in bulb initiation

Analyses of filopodial dynamics and synapse formation for lar (orange), trio (magenta) and control (blue). (A) Lifetime of bulbous filopodia. (B) Total number of bulbous filopodia per terminal per hour. (C) Average number of bulbous filopodia per time instance. (D) Number of concurrently existing bulbous filopodia per axon terminal per time instance observed in the data, simulated after inclusion of a feedback (+f1) and without a feedback (−f1). (E, F) Representative snapshots of lar (E) and trio (F) revealing only transient bulbs. (G-N) Markov state model simulation for lar (G-J) and trio (K-N) for the numbers of filopodia (G, K), transient bulbs (H, L), stable bulbs (I, M) and synapses (J, N). In all cases control traces from Fig. 3 are shown in yellow. Black dotted lines: mean number of bulbs; solid red line: median number of bulbs; dark grey denotes the interquartile range (50% of the data) and light grey the 95% confidence range. (O-R) Measurement of BrpD3 punctae in mutant axon terminals. (O’-R’) BrpD3 single channel. Scale bar: 2 μm. (O-Q) Measurement of BrpD3 punctae in trio and control axon terminals. (O’-P’) BrpD3 single channel. (Q) Quantification of BrpD3-marked synapse numbers relative to control at P90. n=87 and 61, p= 0.67 (R) Schematic summary of protein functions during synapse formation.

Figure 6:. A computational model predicts axon retractions in lar, syd-1, and liprin-α, but not in trio

(A) Schematic of timeline during synapse formation, including continuous decline of transient filopodia, the first appearance of bulbs and the continuous increase in synapse numbers. (B) Measured R7 axon retraction rates. (C) Probability of R7 axon terminal retractions at P100 based on computational modeling of stabilization through a combination of transient filopodia and synapses. (D-G) Representative time-lapse snapshots from long-term live imaging of R7 axon stabilization and retraction in the four mutants. Dashed lines mark the wild-type R7 target layer (M6). Scale bars: 3 μm. (H-K) Computational modeling of predicted probabilistic axon retractions between P40-P100 for all four mutants (comp. to measured data in panel B).

Figure 7:. Serial Synapse Formation Model

The measured live dynamics and computational modeling suggest the following model: (1) stochastic filopodial exploration leads to synaptic capture via a cell surface receptor, e.g. Lar (2) early synaptic seeding factors (Syd-1 and Liprin-α) are recruited to the captured filopodium in an enlarged bulb; (3) secondary simultaneously forming bulbs are destabilized via the function of the RhoGEF Trio, thereby ensuring one synaptogenic filopodium at any given time; recruitment of the active zone protein Brp and synapse maturation occur after filopodial retraction back in the main axon terminal, allowing a new cycle of bulb formation and stabilization.