Synaptic mechanisms for motor variability in a feedforward network

Abstract

Behavioral variability often arises from variable activity in the behavior-generating neural network. The synaptic mechanisms underlying this variability are poorly understood. We show that synaptic noise, in conjunction with weak feedforward excitation, generates variable motor output in the Aplysia feeding system. A command-like neuron (CBI-10) triggers rhythmic motor programs more variable than programs triggered by CBI-2. CBI-10 weakly excites a pivotal pattern-generating interneuron (B34) strongly activated by CBI-2. The activation properties of B34 substantially account for the degree of program variability. CBI-10- and CBI-2-induced EPSPs in B34 vary in amplitude across trials, suggesting that there is synaptic noise. Computational studies show that synaptic noise is required for program variability. Further, at network state transition points when synaptic conductance is low, maximum program variability is promoted by moderate noise levels. Thus, synaptic strength and noise act together in a nonlinear manner to determine the degree of variability within a feedforward network.

Fig. 1. Feedforward circuit, CBI-10 activity during feeding, and motor programs elicited by CBI-10 and CBI-2.

(A) Schematic diagram of the feedforward circuit in the Aplysia feeding system: from command-like neurons, to pattern-generating neurons [e.g., protraction interneurons (PIs), B34 and B63 and retraction interneuron (RI), B64], and protraction (PM) and retraction (RM) motoneurons. Each cycle of the protraction-retraction sequence is mediated primarily by the alternating activity of B63 and B64. Connection symbols: open triangle, excitation; filled circle, inhibition; s, slow connections. (B) Morphology of a left CBI-10 revealed by carboxyfluorescein injection. Its main axon projects anteromedially, travels ventrally and turns back toward the soma, and then projects posteromedially before making a U-turn to the cerebral-buccal connective (CBC) (see fig. S1). Note the extensive processes in the medial region near the soma. The front image originated from three fluorescent images at different depths stacked together. The background image was a bright-field image stacked with the fluorescent image. AT, anterior tentacular; LLAB, lower labial nerve; ULAB, upper labial nerve; CPe, cerebral-pedal connective; CPl, cerebral-pleural connective. (C and D) CBI-10 in a semi-intact preparation. Feeding behavior was monitored by buccal mass “pressure.” (C) Upon delivery of dried seaweed near the mouth, CBI-10 was activated during feeding episodes. Completion of the feeding sequence was verified by exiting of seaweed from the cut end of the esophagus. (D) Stimulation of a single CBI-10 through DC current (bar) elicited four feeding-like responses. (E and F) Multiple cycles of motor programs elicited by CBI-10 (E) or CBI-2 (F) from a single in vitro preparation. Open bar, protraction; filled bar, retraction. (G to I) Paired t test comparing the coefficient of variation (CV) of protraction duration (G) (t₇ = 3.861, **P = 0.0062), duty cycle (H) (t₇ = 5.147, **P = 0.0013), or interspike interval (ISI) of B34 (I) (t₄ = 5.964, **P = 0.004) when CBI-10 or CBI-2 was stimulated at 10 Hz.

Fig. 2. Correlational and causal roles of pattern-generating interneurons (B34) in the generation of motor programs with different degree of variability.

(A) Activity of B34 during CBI-10 (n = 19) versus CBI-2 (n = 22) evoked programs (unpaired t test, t₃₉ = 5.136, ****P < 0.0001). (B and C) B34 activity was inversely correlated with the CV of protraction duration (B) (r = −0.6615, r² = 0.4376, P < 0.0001, n = 58) and duty cycle (C) (r = −0.6245, r² = 0.39, P < 0.0001, n = 58). Lines, linear regression lines; r, correlation coefficient. (D to G) Subthreshold depolarization of B34s (E) (bars under B34 recordings, n = 5) in the programs elicited by CBI-10 (8 Hz). Paired t test comparing the CVs under control conditions with those obtained with depolarization (dep) [(F) protraction duration: t₄ = 3.991, *P = 0.0163; (G) duty cycle: t₄ = 3.078, *P = 0.037]. (H to K) Hyperpolarization of B34s (I) (bars under B34 recordings, n = 8) in programs elicited by CBI-2 (5 Hz). Paired t test comparing the CVs under control conditions with those obtained with hyperpolarization (hyp) [(J) protraction duration: t₇ = 5.198, **P = 0.0013; (K) duty cycle: t₇ = 6.797, ***P = 0.0003]; error bars, SEM. c-B34, contralateral B34. Both ipsilateral and contralateral B34s were depolarized (E) or hyperpolarized (I), and the two B34s are coupled electrically (44). Thus, subthreshold depolarization of B34s (E) made the programs elicited by CBI-10 less variable, whereas hyperpolarization of B34s (I) made programs elicited by CBI-2 more variable.

Fig. 3. Strength and noise in synapses to B34 from CBI-10 versus CBI-2.

(A to C and E to G) Synaptic strengths and facilitation in physiology (A to C) and in the model (E to G). Both CBI-10 (A and E) and CBI-2 (B and F) stimulations (10 Hz for 2 s) elicited monosynaptic EPSPs in B34, but the EPSP amplitude was larger for CBI-2 than for CBI-10. (C and G) Group data from physiological experiments (n = 8) (C) and the model (n = 6) (G). Paired t test comparing average amplitudes of the EPSPs from CBI-10 versus CBI-2. In the physiology (C): first EPSPs, t₇ = 2.512, *P = 0.040; last EPSPs, t₇ = 6.462, ***P = 0.0003; in the model (G): first EPSPs, t₅ = 11.55, ****P < 0.0001; last EPSPs, t₅ = 10.39, ***P = 0.0001. Paired t test comparing average amplitudes of the first and last EPSPs. In the physiology (C): CBI-10, t₇ = 3.494, *P = 0.010; CBI-2, t₇ = 7.318, ***P = 0.0002; in the model (G): CBI-10, t₅ = 2.914, *P = 0.0332; CBI-2, t₅ = 7.579, ***P = 0.0006; error bars, SEM. (A, B, D to F, and H) Synaptic noise in physiological experiments (A, B, and D) versus the model (E, F, and H). CBI-10 or CBI-2 was stimulated at 10 Hz for 2 s for 10 times (only five alternate examples from the 10 trials are shown). In the physiology (D), the average CVs for the last EPSPs from CBI-10 to B34 were 0.221 (n = 6) and from CBI-2 to B34 were 0.194 (n = 6); paired t test, t₅ = 0.739, P = 0.493; n.s., not significant. Error bars, SEM. In the model (H), synaptic noise is modeled as a type of presynaptic noise, and the average CV for the last EPSPs from CBI-10 to B34 was 0.206 with SD of noise σ = 0.18 (n = 6) and from CBI-2 to B34 was 0.184 with σ = 0.15 (n = 6) (paired t test, t₅ = 1.135, P = 0.308), matching physiological data. For group data in (G) and (H), six simulations of 10 time series of EPSPs for 2 s were performed. The first simulation result was obtained with the default values of maximum conductance and plasticity parameter as defined in table S3. Five more sets of these two parameters were generated by adjusting them up or down randomly (up to 5% of their default values). The schematic diagram illustrates the circuit elements and their connections in the simulation. Connection symbols: open triangle, excitation.

Fig. 4. The feeding network model behaves like the biological network.

(A and C) Comparison of programs elicited by CBI-10 (A) and CBI-2 (C) with synaptic noise (CBI-10 and CBI-2 were stimulated at 10 Hz). (B) Depolarization of B34 (bar) during CBI-10 programs with noise made the programs less variable. (D to F) Comparison of CVs of protraction duration (D) (paired t test, t₉ = 7.963, ****P < 0.0001), duty cycle (E) (paired t test, t₉ = 6.073, ***P = 0.0002), and B34 ISI (F) (paired t test, t₉ = 14.97, ****P < 0.0001) from programs elicited by CBI-10 versus CBI-2. (G and H) Comparison of the CVs of protraction duration (G) (paired t test, t₉ = 4.855, ***P = 0.0009) and duty cycle (H) (paired t test, t₉ = 5.157, ***P = 0.0006) from programs elicited by CBI-10 when B34 was not depolarized (Normal) versus when B34 was depolarized (B34 dep). (I to L) Effects of B34 hyperpolarization on CBI-2–elicited programs. The programs were elicited by stimulation of CBI-2 at threshold frequency (6.5 Hz) (I and J). Hyperpolarization of B34 (B34 hyp) (bar) made the programs more variable (J). The current used to hyperpolarize B34 was set at 80% of the current that was just enough to completely prevent firing of B34. This was because 100% of the current prevents normal program generation in the model. B63 activity represents protraction, whereas B64 activity represents retraction. (K and L) Group data. Protraction duration: paired t test, t₉ = 5.579, ***P = 0.0003; duty cycle: paired t test, t₉ = 7.462, ****P < 0.0001. Error bars, SEM. All CVs were derived from averages of 10 simulations with a duration of 200 s each (there were five or more cycles in each simulation).

Fig. 5. Program variability in the computational model: Necessity of synaptic noise and roles of synaptic strength.

(A to H) Representative examples of motor programs elicited by CBI-10 or CBI-2 at different conductances [parameter analysis data are shown in (I) to (P)]. (A to D) With noise: Normal conductance from CBI-10 to B34, 0.68 and to B63, 1.59 (A) or high conductance from CBI-10 to B34, 1.1 and to B63, 1.59 (B); low conductance from CBI-2 to B34, 2.4 and to B63, 1.69 (C) or normal conductance from CBI-2 to B34, 3.4 and to B63, 1.69 (D). (E to H) Without noise: The same parameters as in (A) to (D). B63 activity represents protraction, whereas B64 activity represents retraction. (I to P) Parameter analyses of program variability for synaptic conductance from CBI-10 (I to L) or CBI-2 (M to P) to B34. The parameter (synaptic conductance from CBIs to B34) range was divided into orange (I to L) or purple (M to P) and white areas, with the data plotted as blue lines (with noise) or red lines (without noise). Orange area: No programs elicited by CBI-10; purple area: CBI-2–elicited programs were generated only by B63. White ones: Normal programs generated by both B34 and B63. Note the shift of the white areas with noise to the left relative to the data without noise for CBI-10 programs. One hundred data points were used for each plot. Each data point was derived from one simulation with a duration of 200 s (each simulation contains five or more programs). Labels in plots depict conductance parameters for examples shown in (A) to (H).

Fig. 6. Program variability in the computational model: Roles of different levels of synaptic noise.

(A to D) Effects of different levels of noise (SD, σ) on the CVs of CBI-10–elicited programs. CVs of protraction duration and duty cycles were the highest at moderate noise levels near transition points when the CBI-10 to B34 conductance was normal at 0.68 (A and B). CVs increased with noise levels when conductance was high at 1.1 (C and D), although the highest CVs were lower than the low CVs in (A and B). The red vertical dashed lines in (A and B) indicate the SDs used to generate data in Fig. 5 (I and J). Orange areas, no programs. (E to H) Motor programs elicited by CBI-2. The CBI-2 to B34 conductance was low at 2.4 (E and F) or normal at 3.4 (G and H). CVs were the highest at moderate noise levels near transition points when the CBI-2 to B34 conductance was low (E and F). CVs increased with noise levels when the CBI-2 to B34 conductance was normal (G and H), although the overall CVs were low compared with the data in (E) and (F). The red vertical dashed lines indicate the SDs used to generate data in Fig. 5 (M and N).