Different microcircuit responses to comparable input from one versus both copies of an identified projection neuron

Projection neurons that regulate microcircuit activity commonly occur as multiple copies of apparently equivalent neurons (Rosen et al., 1991; Blitz et al., 1999; Brodfuehrer and Thorogood, 2001; Hägglund et al., 2010; Betley et al., 2013; Bidaye et al., 2014; Gunaydin et al., 2014; Daghfous et al., 2016; Qiu et al., 2016; Li et al., 2017; Fino et al., 2018; Li and Soffe, 2019; Ruder and Arber, 2019). Although the consequences of their co-activation for microcircuit output and/or behavior are established in many systems, there appears to be no systematic comparison of a circuit response to co-activating all versus a defined subset of them. Such comparisons would help elucidate how these systems operate, and could inform whether continued normal behavior would be likely to occur after loss of some copies of a projection neuron due to injury or disease (Fink and Cafferty, 2016).

The microcircuit response to activating all or some copies of a projection neuron might be equivalent, particularly when those neurons act at least partly via peptide co-transmitters, because neurally released peptides diffuse broadly and have long-lasting effects on their targets (Marder, 2012; van den Pol, 2012; Nusbaum et al., 2017; Svensson et al., 2019). Alternatively, the microcircuit response might be skewed by the number of active projection neurons and their firing rates, for example because neuropeptide release often has a higher firing rate threshold than small molecule co-transmitters and its release can be a non-linear function of firing rate (Cazalis et al., 1985; Peng and Horn, 1991; Whim and Lloyd, 1994; Vilim et al., 1996, 2000; Arrigoni and Saper, 2014; Nusbaum et al., 2017), and modulatory actions can be concentration specific (Flamm and Harris-Warrick, 1986; Saideman et al., 2006; Fort et al., 2007; Poels et al., 2007, 2009; Dickinson et al., 2015; Blitz et al., 2019). There is also the additional challenge of establishing whether an apparently equivalent set of projection neurons are indeed functionally equivalent. In some cases, such sets of neurons were subsequently determined to diverge and influence overlapping or distinct sets of neurons (Lammel et al., 2011; Betley et al., 2013; Luo et al., 2018). There can also be distinct responses to activating all versus some copies of a projection neuron when they are bilaterally symmetrical and normally operate separately to mediate hemilateral movements (Shimazaki et al., 2019; Cregg et al., 2020).

Manipulating the activity of multiple copies of a projection neuron in a controlled manner is possible, for example by optogenetics, but it remains challenging to precisely manipulate a defined subset of them. Such precision is possible, however, in some smaller systems where identified projection neurons with a known influence on a target microcircuit are present as small copy numbers (Rosen et al., 1991; Frost and Katz, 1996; Blitz et al., 1999; Brodfuehrer and Thorogood, 2001; Mesce et al., 2008; Jeanne and Wilson, 2015).

One tractable system for such a study is the stomatogastric nervous system (STNS) of the crab Cancer borealis (Marder and Bucher, 2007; Daur et al., 2016; Nusbaum et al., 2017; Stein, 2017). This system contains two well-characterized microcircuits in the stomatogastric ganglion (STG) that generate the motor patterns for chewing (gastric mill microcircuit), and the pumping and filtering of chewed food (pyloric microcircuit) in vivo and in the isolated STNS. These circuits are readily accessible in the isolated STNS because most of the 26 STG neurons contribute to one or both of these microcircuits, their somata are relatively large (diameter ∼35–120 µm), and most microcircuit neurons occur as single copies (Marder et al., 2017). There are also six different identified projection neuron pairs that regulate gastric mill and/or pyloric circuit activity (Coleman et al., 1992; Norris et al., 1994, 1996; Blitz et al., 1999; Christie et al., 2004).

The best characterized projection neuron in the C. borealis STNS is modulatory commissural neuron 1 (MCN1) (Coleman and Nusbaum, 1994; Hedrich et al., 2011; Nusbaum et al., 2017). Each MCN1 projects an axon from one of the paired commissural ganglia (CoGs), through the inferior oesophageal nerve (ion) and stomatogastric nerve (stn), to innervate the STG and drive the gastric mill and pyloric rhythms (Coleman et al., 1992; Coleman and Nusbaum, 1994; Bartos and Nusbaum, 1997). MCN1 can be selectively driven by extracellular ion stimulation, and both MCN1s contain the same co-transmitters, influence the same STG circuit neurons, and drive the gastric mill rhythm by the same mechanism (Coleman et al., 1995; Bartos and Nusbaum, 1997; Bartos et al., 1999; Blitz et al., 1999; Wood et al., 2000; Stein et al., 2007; DeLong et al., 2009a,b; Nusbaum et al., 2017). Furthermore, despite being present in separate ganglia, both MCN1s are co-activated by all identified sensory and central nervous system (CNS) pathways (Beenhakker et al., 2004, 2005; Blitz et al., 2004, 2008; Christie et al., 2004; Hedrich et al., 2009; Blitz and Nusbaum, 2012; White et al., 2017).

Here, we examine whether the gastric mill microcircuit response to dual and single MCN1 stimulation, in preparations where the CoGs are removed and the MCN1 axon in the ion is selectively stimulated, is comparable when the two conditions have matched firing rates and patterns. We ensure this match by co-stimulating the two MCN1s in a one-to-one alternating pattern that produces an inter-stimulus interval equivalent to that used for the matched single MCN1 stimulation. These matched stimulations did not elicit equivalent gastric mill rhythms. For example, the gastric mill rhythm was more consistently coordinated during the dual MCN1 stimulations. These different outcomes likely resulted at least partly from the similar firing frequency of the gastric mill rhythm generator neuron LG (lateral gastric) across each set of matched single and dual MCN1 stimulations. This similar LG neuron activity was likely more effective in inhibiting each MCN1 during the dual stimulations, where their firing frequency was half that of the same neuron in the corresponding single stimulation. These results indicate that at least some neural circuits are optimally driven by co-activating all copies of circuit-driving projection neurons.

Adult male Jonah crabs (Cancer borealis Stimpson 1859) were obtained from commercial suppliers (Fresh Lobster, LLC; Marine Biological Laboratory; Ocean Resources, Inc.) and maintained in aerated, filtered artificial seawater at 10–12°C. Animals were cold anesthetized by packing in ice for at least 30 min before dissection. The foregut was then removed from the animal, after which the STNS was dissected from the foregut in physiological saline at 4°C. The dorsal connective tissue sheath of the STG was removed immediately prior to recording to facilitate access for intrasomatic recordings.

Cancer borealis physiological saline contained the following (mmol l⁻¹): 440 NaCl, 26 MgCl₂, 13 CaCl₂, 11 KCl, 10 Trizma base, 5 maleic acid and 5 glucose; pH 7.4–7.6. All preparations were superfused continuously with C. borealis saline (8−12°C).

Electrophysiology experiments were performed using standard techniques for this system (Beenhakker and Nusbaum, 2004). The isolated STNS (Fig. 1A) was pinned down in a silicone elastomer-lined (Sylgard 184, KR Anderson) Petri dish. Each extracellular nerve recording resulted from a pair of stainless-steel wire electrodes (reference and recording) whose ends were pressed into the Sylgard-coated dish. A differential AC amplifier (model 1700, A-M Systems) amplified the voltage difference between the reference wire (placed in the bath) and the recording wire [placed near an individual nerve and isolated from the bath by petroleum jelly (Vaseline, Lab Safety Supply)]. This signal was then further amplified and filtered (model 410 amplifier, Brownlee Precision). Extracellular nerve stimulation was accomplished by placing the pair of wires used to record nerve activity into a stimulus isolation unit (SIU 5, AstroNova Inc.) that was connected to a stimulator (model S88, AstroNova Inc.).

Intrasomatic recordings were made with sharp glass microelectrodes (8–20 MΩ) filled with potassium acetate (4 mol l⁻¹) plus KCl (20 mmol l⁻¹), or KCl alone (1 mol l⁻¹). Intracellular signals were amplified using Axoclamp 2B amplifiers (Molecular Devices), and then further amplified and filtered (model 410 amplifier). Current injections were performed in single-electrode discontinuous current-clamp (DCC) mode with sampling rates between 2 and 3 kHz. To improve visibility for intracellular recording, the STG was dorsally desheathed and viewed with light transmitted through a dark-field condenser (Nikon). STG neurons were identified on the basis of their axonal projections, activity patterns and interactions with other STG neurons (Weimann et al., 1991; Blitz et al., 2008). In the displayed extracellular nerve recordings, all action potentials of a given amplitude in a particular nerve are generated by the neuron indicated above a subset of those action potentials (Nusbaum et al., 2017).

Each ion was stimulated (duration per stimulus: 1 ms), just medial to the labral nerve (ln; Fig. 1A), to selectively activate the left MCN1 (MCN1_L) or the right MCN1 (MCN1_R) (Bartos and Nusbaum, 1997). The left and right CoG, and hence MCN1_L and MCN1_R, were defined based on the isolated STNS being pinned with the dorsal side facing up. There is only one other CoG projection neuron (MCN5) that innervates the STG which has an axon in the ion, and it has a higher stimulus threshold than MCN1 as well as having no direct influence on LG (Coleman et al., 1992; Norris et al., 1996; Blitz et al., 2019). Each suprathreshold ion stimulus elicited a single MCN1 action potential, as established by recording the resulting unitary, electrical excitatory postsynaptic potential (EPSP) in the LG neuron (Coleman et al., 1995). Using a voltage setting on the stimulator that was ≤0.5 V above threshold for eliciting MCN1 action potentials prevented or minimized co-stimulation of MCN5 (Bartos and Nusbaum, 1997). A tonic ion stimulation pattern (i.e. constant inter-stimulus intervals) was used in all these experiments.

To determine the relative influence of matched dual versus single MCN1 stimulation on the gastric mill rhythm, we used a range of MCN1 firing rates that spanned its activity range in response to input pathway stimulation. Specifically, we stimulated each MCN1 separately, at 10, 20 or 30 Hz using a tonic stimulation pattern. We compared these single stimulations to matched dual MCN1 stimulation during which the two MCN1s were each stimulated at 50% of the firing rate of the associated single stimulations (e.g. 5 Hz+5 Hz compared with 10 Hz). During dual stimulation, each MCN1 was activated to fire a single action potential alternately, such that their combined firing rate and pattern was the same as the matched single stimulation. To ensure the occurrence of the same MCN1 inter-spike interval during the matched single and dual MCN1 stimulations, we consistently tracked the presence of MCN1-evoked EPSPs in the intracellular LG neuron recordings.

In some experiments where the activity of the retraction phase neuron DG (dorsal gastric) persisted through the protraction phase of the gastric mill rhythm, the ability of the protractor neuron LG to regulate DG activity was examined by injecting LG with suprathreshold depolarizing current pulses during some of its bursts to increase its within-burst firing rate. In these same experiments, we also determined the impact of these LG current injections on the pyloric cycle period (see below). The LG current injections (+1 to +4 nA) were done in DCC, using brief depolarizing pulses (pulse duration: 30–50 ms) that each elicited a single action potential. Each within-burst current injection train was turned on and off manually, and was timed to occur during a normal LG burst for a duration similar to the LG burst duration during that particular gastric mill rhythm. These manipulations were performed in a subset of experiments during dual 10 Hz and single 20 Hz MCN1 stimulated gastric mill rhythms. Two of these LG current injections, separated by four to eight control LG bursts, were performed during each gastric mill rhythm.

In control experiments, the mechanosensory ventral cardiac neurons (VCNs) were activated by stimulating the dorsal posterior oesophageal nerve (dpon; duration per stimulus: 1 ms), either unilaterally or bilaterally, in preparations where the superior oesophageal nerves (sons) were bisected between the dpon and the stn. The VCNs innervate the ipsilateral CoG via the dpon and son, and their stimulation triggers a version of the gastric mill rhythm by eliciting a long-lasting activation of the CoG projection neurons MCN1 and CPN2 (commissural projection neuron 2) (Beenhakker et al., 2004; Beenhakker and Nusbaum, 2004). Bisecting the sons, through which the CPN2 axon projects to the STG, prevents CPN2 activity from influencing the STG, enabling VCN stimulation to influence the gastric mill microcircuit exclusively via its activation of MCN1 (Norris et al., 1994; Beenhakker and Nusbaum, 2004). These experiments included both bilateral (N=2) and unilateral VCN stimulations (N=3). In each experiment, the MCN1 firing rate resulting from VCN stimulation was determined on the stimulated side(s) and subsequently used as the ipsilateral ion stimulation frequency, to compare the gastric mill rhythm response to direct (ion stimulation) and indirect (dpon stimulation) activation of MCN1. All ion stimulation experiments were performed from September 2013 to March 2015. The dpon versus ion stimulation experiments were performed during February and March 2017.

Data were collected in parallel onto a chart recorder (AstroNova Everest model) and computer. Acquisition onto computer (sampling rate ∼5 kHz) used the Spike2 data acquisition and analysis system (Cambridge Electronic Design). Some analyses, including cycle period, burst durations, duty cycle, number of action potentials per burst and intraburst firing frequency were conducted on the digitized data using a custom-written Spike2 program (‘The Crab Analyzer’). To facilitate data analysis and improve clarity in some figures, a raw extracellular recording (e.g. dorsal gastric nerve, dgn) was duplicated with the stimulation artifacts digitally subtracted, reducing their amplitude or eliminating them. This is indicated where appropriate in each figure legend, and is illustrated in Fig. 2B. A custom-written script in Spike2 was used to digitally subtract the artifacts after manually inspecting the trace to verify that only that particular unit was selected for subtraction (Blitz and Nusbaum, 2008).

Unless otherwise stated, each data point in a dataset was derived by determining the mean for the analysed parameter from 8–20 consecutive gastric mill cycles. One gastric mill cycle was defined as extending from the onset of consecutive LG neuron action potential bursts (Beenhakker and Nusbaum, 2004; Wood et al., 2004). Thus the gastric mill cycle period was measured as the duration between the onset of two successive LG neuron bursts. The protractor phase was measured as the LG burst duration, whereas the retractor phase was measured as the LG interburst duration. A gastric mill rhythm-timed burst duration was defined as the duration between the onset of the first and last action potential within an impulse burst, during which no interspike interval was longer than 2 s (approximately twice the pyloric cycle period during the gastric mill rhythm and no more than half the duration of each gastric mill phase) (Beenhakker et al., 2004). The intraburst firing rate of a neuron was defined as the number of action potentials within a burst minus one, divided by the burst duration. The pyloric cycle period was determined as the duration between the onset of successive bursts in the pyloric dilator (PD) neuron (Fig. 1B). The PD neuron is a pyloric pacemaker neuron component (Marder and Bucher, 2007; Selverston and Miller, 1980).

To evaluate the influence of the LG neuron depolarizing current injections on the pyloric cycle period, the longest pyloric cycle that occurred after the start of the LG burst was selected for analysis during (a) each LG burst that received depolarizing current pulses, and (b) the two preceding LG bursts. The first pyloric cycle after LG burst onset was always the slowest one (i.e. longest cycle period) during the control LG bursts and was the slowest during 77% of the depolarized LG bursts, likely reflecting the fact that the LG instantaneous firing rate was often highest at the start of its burst. To compare these pyloric cycle periods between the depolarized and control LG bursts, we first determined and compared the ratios of the pyloric cycle period from the (a) depolarized LG burst divided by that from the preceding control LG burst, and (b) the same preceding control LG burst divided by that from the control LG burst two prior to the depolarized burst.

Data were plotted with Excel (version 2002; Microsoft) and MATLAB (version 8; MathWorks). Figures were produced using CorelDraw (version 13.0 for Windows). Statistical analyses were performed by comparing the overall mean of individual mean values for two different manipulated conditions, or control and manipulated groups, from N experiments (see Results for each N value) using Microsoft Excel, SigmaPlot 13.0 (SPSS Inc.) and MATLAB. Unless otherwise indicated, the presented N values represent the number of preparations. Comparisons were made to determine statistical significance using (a) repeated-measures ANOVA (RM-ANOVA), with the Holm–Šidák post hoc test when the RM-ANOVA P-value was <0.05, (b) RM-ANOVA on ranks, with post hoc Tukey’s test or chi-square test, (c) paired Student's t-test, (d) signed rank test, (e) chi-square test (different from the aforementioned post hoc chi-square test), (f) Fisher's exact test, (g) Wilcoxon signed rank test or (h) unpaired t-test. In all experiments, the effect of each manipulation was reversible, and there was no significant difference between the pre- and post-manipulation groups. Significant differences were determined to occur when P<0.05. Data are expressed as means±s.e.m.

The gastric mill rhythm is a two-phase motor pattern (protraction, retraction) that drives the rhythmic contraction of striated muscles in the middle (i.e. gastric mill) stomach compartment of decapod crustaceans (Heinzel, 1988; Heinzel et al., 1993; Diehl et al., 2013). The sequence of these muscle contractions causes the paired lateral teeth and medial tooth within the gastric mill to rhythmically move towards (protract) and away from (retract) the midline, macerating food moved into the gastric mill from the anterior, cardiac sac stomach compartment. The chewed food is then filtered and pumped through the posterior stomach compartment, the pylorus, to enter the midgut for nutrient absorption. The rhythmic chewing pattern is generated by the gastric mill central pattern generator circuit, an episodically active microcircuit in the STG that is driven by projection neurons, including the paired MCN1s, located in the CoGs (Fig. 1A,C) (Coleman and Nusbaum, 1994; Bartos et al., 1999; Beenhakker and Nusbaum, 2004; Blitz et al., 2008, 2019). The gastric mill rhythm is episodically active, in vivo and in vitro, because the projection neurons that drive it are episodically active, requiring excitatory drive from sensory neurons and other CNS neurons (Beenhakker et al., 2004; Blitz et al., 2004, 2008; Christie et al., 2004; Hedrich et al., 2009, 2011; Diehl et al., 2013).

There are eight gastric mill microcircuit neuron types, including four protractor motor neurons [LG, MG, GM (gastric mill), IC (inferior cardiac)] and four retraction phase neurons [one interneuron: Int1; three motor neurons: DG, VD (ventricular dilator), AM (anterior median)] (Fig. 1C) (Nusbaum et al., 2017). All but GM are present as single copies; there are four GMs. The neurons that are necessary and sufficient to generate the gastric mill rhythm (i.e. the rhythm-generating sub-circuit) during selective MCN1 activation includes the reciprocally inhibitory pair LG–Int1 plus the STG terminals of MCN1 (MCN1_STG) (Fig. 1D) (Coleman et al., 1995; Bartos et al., 1999; DeLong et al., 2009b; Nusbaum et al., 2017).

MCN1 is a multi-transmitter neuron that uses only its peptide co-transmitter CabTRP Ia (Cancer borealis tachykinin-related peptide Ia) to influence LG, causing a slowly developing metabotropic excitation by activating I_MI (modulator-activated inward current), while it uses only its small molecule co-transmitter GABA to influence Int1, causing a fast ionotropic excitation (Blitz et al., 1999; Wood et al., 2000; Stein et al., 2007). I_MI is a voltage-dependent depolarizing current with NMDA-like properties that is activated in STG neurons by many neuromodulators (Golowasch and Marder, 1992; Swensen and Marder, 2000, 2001; DeLong et al., 2009b; Rodriguez et al., 2013). MCN1 also has a functionally important electrical synapse with LG which strengthens the LG burst (Coleman et al., 1995). The MCN1 transmitter-mediated actions are limited to the retraction phase, because it receives fast glutamatergic inhibition at its STG terminals during protraction from LG (Coleman and Nusbaum, 1994). This LG-mediated inhibition triggers the transition from retraction to protraction and, by triggering decay of I_MI amplitude, enables the eventual transition from protraction to retraction (Fig. 1D) (Coleman et al., 1995; DeLong et al., 2009b). Here, we use the LG neuron burst to represent the protraction phase, and the LG neuron interburst burst to represent the retraction phase (Fig. 1B).

We determined how consistently fully coordinated gastric mill rhythms were generated from three sets of matched dual versus single MCN1 stimulations (5 Hz dual versus 10 Hz single; 10 Hz dual versus 20 Hz single; 15 Hz dual versus 30 Hz single) (Fig. 3). A fully coordinated gastric mill rhythm exhibits consistent alternating bursting between the protraction- and retraction-phase neurons (Fig. 1B,C). Most aspects of this patterning during the MCN1-driven gastric mill rhythm result directly from the microcircuit synapses made by the rhythm generator neurons LG and Int1 (Fig. 1C) (Bartos et al., 1999; White and Nusbaum, 2011). Consequently, most gastric mill neuron activity is tightly linked to that of LG and/or Int1. For example, IC neuron activity is consistently limited to the protraction phase while VD neuron activity is limited to the retraction phase (Figs 1B and 4). In contrast, the gastric mill rhythm-timed bursting pattern of the retractor neuron DG is an indirect consequence of LG neuron activity. Specifically, the DG neuron burst pattern results from the LG ionotropic inhibition of MCN1_STG, which weakens or eliminates MCN1 metabotropic excitation of DG (Fig. 1C,D) (Coleman and Nusbaum, 1994). As is evident in Fig. 2, in our matched MCN1 stimulation experiments DG neuron activity was not always limited to the retraction phase.

As shown in Fig. 3, fully coordinated gastric mill cycles occurred more consistently during 5 Hz dual MCN1 stimulation than 10 Hz single stimulation (fraction of coordinated cycles per gastric mill rhythm: 5 Hz dual, 0.91±0.08; 10 Hz_R single, 0.47±0.11; 10 Hz_L single, 0.45±0.11; RM-ANOVA, P<0.001; Holm–Šidák post hoc test: 5 Hz dual versus 10 Hz_R or 10 Hz_L, P=0.001; 10 Hz_R versus 10 Hz_L, P=0.90; N=10 preparations). Additionally, in eight of the 10 (80%) 5 Hz dual stimulation experiments, there was LG and DG burst alternation in every gastric mill rhythm cycle (Fig. 3). In contrast, a fully coordinated gastric mill rhythm occurred in only one of 10 (10%) experiments each for the 10 Hz_L and 10 Hz_R single stimulations (Fig. 3).

In contrast to the 5 Hz dual–10 Hz single stimulations, there was no difference in the mean fraction of gastric mill cycles exhibiting LG–DG alternation during the two higher matched pairs of MCN1 stimulation (10 Hz dual, 0.47±0.12; 20 Hz_L single, 0.29±0.13; 20 Hz_R single, 0.15±0.05; RM-ANOVA, P=0.09, N=11 preparations; 15 Hz dual, 0.36±0.14; 30 Hz_L single, 0.23±0.01; 30 Hz_R single, 0.06±0.05, RM-ANOVA, P=0.1, N=11) (Fig. 3). However, the median values of the 10 Hz dual and 20 Hz single matched pairs were offset considerably, due in part to there being a higher percentage of rhythms with no cycles of LG–DG alternation during the single stimulations [20 Hz single, 10/22 (45%); 10 Hz dual, 2/11 (18%); Fig. 3]. This difference was also present, albeit to a lesser degree, for the highest matched stimulations [30 Hz single, 15/22 (68%); 15 Hz dual, 6/11 (55%); Fig. 3].

There was also separation between the 10 Hz dual–20 Hz single MCN1 stimulations with respect to the percentage of experiments in which more than half of the cycles displayed LG–DG alternation (10 Hz dual, N=6/11, 55%; 20 Hz single: N=3/22, 14%; P=0.033, Fisher's exact test), but this was not the case for the 15 Hz dual–30 Hz single stimulations (P=0.186) (Fig. 3). However, unlike the consistently coordinated gastric mill rhythms that occurred during the dual 5 Hz stimulations, comparable full coordination for all cycles only occurred during two of 11 preparations (18%) for both 10 Hz dual and 15 Hz dual stimulations (Fig. 3). During the single MCN1 stimulations, full coordination during all cycles occurred in only two of 20 (10%) 10 Hz stimulations, two of 22 (9.1%) 20 Hz stimulations, and none of the 30 Hz (N=22) stimulations. Across all three sets of matched MCN1 stimulations, at least 50% of cycles exhibited LG–DG burst alternation in more of the dual (19/32; 59%) than the single (19/64; 30%) stimulations (P=0.01, chi-square test on contingency table).

During gastric mill rhythm cycles when LG–DG burst alternation did not occur, DG neuron activity persisted through the LG burst (Figs 2 and 4). When DG activity did continue through the protraction phase, its firing rate was consistently reduced relative to that during retraction (10 Hz single: P<0.001, N=9; 10 Hz dual: P=0.002, N=5; 20 Hz single: P<0.001, N=10; 15 Hz dual: P=0.015, N=4; 30 Hz single: P=0.008*, N=8; paired t-test, except *signed rank test). This DG firing rate reduction during protraction was greater during the dual 10 Hz and dual 15 Hz MCN1 stimulations than during their matched single stimulations (10 Hz dual versus 20 Hz single: N=5, P=0.027; 15 Hz dual versus 30 Hz single: N=4, P=0.037, paired t-test). The larger DG firing rate reduction during protraction when both MCN1s were co-stimulated occurred despite the fact that the retraction phase firing rate of DG was equivalent during the matched dual and single stimulations (10 Hz dual, 12.5±0.6 Hz, N=5; 20 Hz single, 12.0±0.5 Hz, N=8, P=0.83; 15 Hz dual, 14.8±0.8 Hz, N=4; 30 Hz single, 13.4±0.3 Hz, N=8, P=0.35; unpaired t-test).

The failure of DG activity to terminate during protraction was particularly pronounced during 30 Hz single MCN1 stimulations. These stimulations rarely elicited LG–DG alternation in 50% or more gastric mill rhythm cycles (N=4/22, 18%), and only once (N=1/22, 4.5%) elicited alternation during 75% of the cycles (Fig. 3). Most of these 30 Hz single stimulations (15/22; 68%) exhibited no cycles in which DG was silenced during protraction. This latter type of response did not occur as frequently during the other stimulation protocols (5 Hz dual, 0%; 10 Hz single, 10%; 10 Hz dual, 18%; 20 Hz single, 45%; 15 Hz dual, 55%) (Fig. 3).

There were also differences in several other gastric mill rhythm parameters between the matched dual and single MCN1 stimulations. For example, there was a longer gastric mill cycle period during the 5 Hz dual MCN1 stimulations than the 10 Hz single stimulations (N=14, P<0.001, one-way RM-ANOVA, Holm–Šidák post hoc test) (Figs 4A and 5A). Despite the rhythm being slower during the dual 5 Hz stimulation, there was no difference in protraction duration between the dual and single stimulations (N=13, P=0.14, RM-ANOVA) (Fig. 5B). Instead, the longer cycle period resulted from an increased retraction duration (N=14, P<0.001, RM-ANOVA on ranks, post hoc Tukey’s test) (Fig. 5C). This selective prolongation of retraction during the dual 5 Hz MCN1 stimulations decreased the LG neuron duty cycle relative to the matched single stimulations (N=14, P<0.01; RM-ANOVA, Holm–Šidák post hoc t-test) (Fig. 5D). The effects of 10 Hz_L and 10 Hz_R single MCN1 stimulations were indistinguishable for this and all other examined parameters (N=14, P>0.05) (Fig. 5).

At the higher MCN1 stimulation frequency sets (10 Hz dual versus 20 Hz single, N=19; 15 Hz dual versus 30 Hz single, N=18), protraction and retraction were both prolonged, and thus so was cycle period, during the dual MCN1 stimulations (Fig. 5A–C). In contrast, there was no distinction in these parameters between the single MCN1_L and MCN1_R stimulations at the same stimulation rate (Fig. 5A–C). The parallel increases in protraction and retraction duration during the higher dual stimulation rates resulted in there being no difference in the LG duty cycle between these conditions and their matched set of single MCN1 stimulations (Fig. 5C).

We also determined whether the LG neuron burst characteristics were differentially influenced by matched dual and single MCN1 stimulation. With respect to the number of LG spikes per burst, there was no difference at the lowest matched MCN1 stimulation rate (N=14, P=0.81, RM-ANOVA on ranks) (Fig. 5E). In contrast, during both of the higher matched stimulation rates, the dual stimulations consistently elicited more LG spikes per burst than their matched single MCN1 stimulations (10 Hz dual versus 20 Hz single: P<0.001, N=19; 15 Hz dual versus 30 Hz single: P<0.001, N=18; RM-ANOVA, Holm–Šidák post hoc test) (Fig. 5E). There was no change in the LG intraburst firing rate within any of the three sets of stimulation conditions (Fig. 5F). Instead, the increased LG spike numbers reflected the prolonged LG burst duration (Fig. 5B).

Because both MCN1s are co-activated by all identified input pathways, we also compared the gastric mill rhythms when both MCN1s were co-stimulated (10 Hz each) versus single MCN1 stimulation at the same frequency (N=13) (Fig. 6A,B). These stimulation conditions again elicited differences in some gastric mill rhythm parameters (Fig. 6B). For example, dual MCN1 stimulation (10 Hz each) elicited gastric mill rhythms exhibiting a longer cycle period (P=0.006) due to a prolonged protraction phase (P<0.001), with an increased LG duty cycle (P=0.003) containing more LG spikes per burst (P<0.001) generated at a higher firing rate (P<0.001) than during single MCN1 stimulation at 10 Hz (Fig. 6B). There was no difference in the retraction phase duration (P=0.455).

In addition to differences in the mean value of some gastric mill rhythm parameters between matched dual and single MCN1 stimulation frequencies, we assessed parameter variability across experiments in relation to the mean using the coefficient of variation (CV). Statistically significant differences occurred primarily during the 5 Hz dual versus 10 Hz single stimulations. For example, the CV across experiments for cycle period was higher for the 10 Hz single MCN1 stimulations (dual MCN1 versus: MCN1_L, P=0.028; MCN1_R, P=0.031; RM-ANOVA, Holm–Šidák post hoc test; N=14) (Fig. 7A). Some of the studied parameters whose mean values were not different nevertheless exhibited differences in the CV, as was the case for the protraction duration and number of LG spikes per burst (Figs 5 and 7). For example, the protraction duration CV during the single 10 Hz stimulations for MCN1_L was 0.27±0.01 s, while it was 0.17±0.01 s for the dual 5 Hz stimulation (N=14; P=0.045, RM-ANOVA, Holm–Šidák post hoc test). However, despite the 10 Hz stimulation of MCN1_R having the same protraction duration CV as MCN1_L (0.27±0.01), this value was not different from that of the matched 5 Hz dual stimulations (P=0.062, RM-ANOVA, Holm–Šidák post hoc test). With respect to the number of LG spikes per burst, single MCN1 stimulations at 10 Hz produced a higher CV across experiments than the dual 5 Hz stimulation (dual MCN1 versus: MCN1_L, P=0.03; MCN1_R, P=0.042; RM-ANOVA, Holm–Šidák post hoc test; N=14). In contrast, the CV for retraction duration, LG duty cycle and LG firing frequency was comparable for the matched dual (5 Hz each) and single (10 Hz) MCN1 stimulations (Fig. 7C,D,F). Few parameters exhibited CV differences between the higher frequency-matched MCN1 stimulations, and in no cases did both single stimulations differ from the matched dual stimulation (Fig. 7).

The fact that DG neuron activity was not always limited to the retraction phase in these experiments, but exhibited a reduced firing rate when active during protraction (Figs 2–4), suggested that the LG firing frequency during these rhythms weakened but did not eliminate MCN1_STG transmitter release. We used matched dual (10 Hz) and single (20 Hz) MCN1 stimulations to test this hypothesis by comparing the influence of natural and strengthened LG bursts on two targets that it only affects via its inhibition of MCN1_STG, including (a) DG neuron activity and (b) the pyloric cycle period. The pyloric cycle period is a useful assay because MCN1 directly excites the pyloric rhythm, reducing the pyloric cycle period, whereas LG activity only influences this rhythm (i.e. increases the pyloric cycle period) via its inhibition of MCN1_STG (Fig. 1C) (Bartos and Nusbaum, 1997).

When LG neuron activity was strengthened during matched dual (10 Hz) and single (20 Hz) MCN1 stimulations, its increased firing rate was not different across protocols (20 Hz_L: 13.9±0.4 Hz; 20 Hz_R: 13.4±0.4 Hz; 10 Hz_both: 13.1±0.6 Hz; N=5 experiments, 10 LG depolarizations per condition; P=0.111, RM-ANOVA). These increased firing rates were consistently higher than during the control LG bursts (P<0.001 for each protocol, RM-ANOVA, Holm–Šidák post hoc test), which ranged from 4.5 to 6.0 Hz.

For gastric mill rhythm cycles where DG neuron activity extended through the LG burst, enhanced LG activity eliminated this prolonged DG activity in 9/20 cycles during 20 Hz single MCN1 stimulations (45%; N=5 experiments, two LG current injections per MCN1 stimulation). In the remaining 11 cycles, the DG firing rate was weakened but not terminated. This extended DG activity was more consistently eliminated during the dual 10 Hz MCN1 stimulations (9/10 cycles 90%; n=5 experiments, two LG current injections per MCN1 stimulation; P=0.024, Fisher's exact test), suggesting that LG more effectively regulates MCN1 activity when MCN1 fires at a lower rate.

We evaluated the impact of an increased LG firing frequency on the pyloric cycle period by first normalizing the pyloric cycle period during strengthened LG firing to that during the preceding, control LG burst (‘LG depolarization ratio’; see Materials and Methods). We then compared this normalized value with a control value obtained by dividing the pyloric cycle period during the same, preceding control LG burst by the cycle period during the control LG burst two prior to the depolarized LG burst (‘control ratio’). The control pyloric cycle period ratios were comparable across all MCN1 stimulation protocols (N=10 stimulations per protocol, P=0.407, repeated measures ANOVA on ranks).

During all three sets of MCN1 stimulation, the LG depolarization ratio was larger than the control ratio [20 Hz_L: LG depolarization ratio, 1.11±0.03; control ratio, 1.00±0.03, P=0.004; 20 Hz_R: LG depolarization ratio, 1.18±0.06; control ratio, 0.99±0.01, P=0.004; 10 Hz_Both: LG depolarization ratio, 1.07±0.01; control ratio, 0.99±0.01, P=0.001; N=5, signed rank test (20 Hz_L,R), paired t-test (10 Hz_both)] (Fig. 8). The increased ratios that occurred during the LG current injections indicated that the pyloric cycle period at these times was prolonged relative to that during the control LG bursts. Consistent with the limited LG–DG alternation reported above (Figs 2–4), and the impact of an increased LG firing rate on the timing of DG neuron bursts, these pyloric rhythm results suggested that the strengthened LG bursts more strongly inhibit MCN1_STG transmitter release.

We used ion stimulation to selectively drive MCN1 despite the presence of another CoG neuron (MCN5) that projects to the STG through the ion because MCN1 has a lower stimulation threshold (see Materials and Methods). We nevertheless considered the possibility that some MCN5 stimulation contributed to our results by comparing the influence of ion stimulation with that resulting from activating MCN1 by stimulating the dpon, after bisecting the son medial to the dpon (see Materials and Methods) (Fig. 1A).

The MCN1 firing rate resulting from dpon stimulation in different experiments ranged from 9.5 to 26 Hz. To match the ion and dpon activation of MCN1 in each experiment, we first triggered a VCN activation of MCN1 and recorded the resulting gastric mill rhythm. In parallel, we determined the VCN-triggered MCN1 firing rate and used that same rate to subsequently stimulate the ion. We then followed each ion stimulation with a second dpon stimulation. Little or no MCN5 activity was evident in the ion recordings during the VCN-triggered gastric mill rhythms. We assayed the gastric mill rhythm response by analysing the same gastric mill rhythm parameters as above (Fig. 5) and determining the gastric mill rhythm-timed activity of the DG neuron.

The gastric mill microcircuit response to dpon and ion stimulation was comparable for five of the six analysed parameters, including cycle period, protraction and retraction duration, number of LG spikes per burst, and the LG duty cycle (N=5) (Fig. 9). The only analysed parameter with distinct values was the LG neuron firing frequency (dpon stimulation 1: 6.6±1.1 Hz; ion stimulation: 5.9±0.9 Hz; dpon stimulation 2: 6.5±1.0 Hz; N=5; dpon1 versus ion: P=0.03; dpon2 versus ion: P=0.04; dpon1 versus dpon2: P=0.67, RM-ANOVA, Holm–Šidák post hoc test) (Fig. 9F). Similarly, the matched MCN1 firing rates from dpon and ion stimulations resulted in comparable DG neuron activity patterns. In none of these experiments was DG activity limited to the retraction phase, nor were there any gastric mill rhythms in which the LG and DG bursts alternated in at least 50% of the cycles (N=5).

In this paper we determined that firing rate- and pattern-matched dual and single stimulation of the paired projection neuron MCN1 does not elicit equivalent gastric mill motor patterns. These different outcomes appear due, at least partly, to a mismatch in the firing rate of MCN1 and the gastric mill microcircuit neuron LG, which regulates MCN1 transmission in the STG via presynaptic inhibition (Fig. 1D) (Coleman and Nusbaum, 1994; Coleman et al., 1995). Specifically, the dual MCN1 stimulations more consistently elicited completely coordinated gastric mill cycles. This coordination distinction was particularly pronounced at the lowest MCN1 stimulation frequency comparison, during which the dual stimulations also exhibited lower variability in several parameters across experiments. This outcome suggests that acute loss of one MCN1 would compromise behavioral performance to the extent that effective chewing involves coordinated rhythmic protraction and retraction of the teeth (Heinzel et al., 1993; Diehl et al., 2013). Functional recovery after such a loss might be facilitated, however, by long-term compensatory mechanisms that are known to occur in some motor systems (Büschges et al., 1992; Sánchez et al., 2000; Sakurai and Katz, 2009; Fink and Cafferty, 2016; Sakurai et al., 2016; Brown and Martinez, 2018; Puhl et al., 2018).

Insofar as most gastric mill microcircuit neurons are also the motor neurons for the system, the other rhythm parameter differences that we identified suggest that differences would also occur in the response dynamics of at least some gastric mill muscles, producing changes in the timing and/or strength of teeth movements (Stein et al., 2006; Diehl et al., 2013). Whether these latter changes would compromise chewing behavior, however, remains to be determined. In all comparisons, there were no differences between the matched single stimulations (MCN1_L versus MCN1_R), which is consistent with previous evidence suggesting that the two MCN1s are functionally equivalent. There do not appear to be other studies that performed such a comparison on different copies of a projection neuron within the same preparations.

The role(s) of presynaptic regulation of projection- and sensory neuron inputs to microcircuits remains under-explored despite its established presence, often at multiple locations, in several neural systems (Nusbaum, 1994; Sillar and Simmers, 1994; Krieger et al., 1996; Cochilla and Alford, 1999; Westberg et al., 2000; Takahashi and Alford, 2002; Evans et al., 2003; Hurwitz et al., 2005; Barrière et al., 2008; Blitz and Nusbaum, 2008, 2012; Jing et al., 2011; Wang, 2012; McGann, 2013; Sirois et al., 2013; Blitz et al., 2019). For the MCN1–gastric mill rhythm, the pivotal role of LG neuron presynaptic inhibition of MCN1_STG suggests that this circuit design favors dual MCN1 activity over firing rate-matched single MCN1 activity. This suggestion is consistent with the greater effectiveness of this presynaptic action during the dual MCN1 stimulations, where each MCN1 fired at half the rate of the matched single MCN1 stimulations, and by our finding that the LG firing rate was equivalent within each matched set of MCN1 stimulations. This outcome resulted from the LG synaptic action reducing but not eliminating MCN1 transmitter release in these experiments, insofar as increasing the LG intraburst firing rate during both dual and single MCN1 stimulation more consistently limited DG neuron activity to the retraction phase, and further weakened the pyloric rhythm. LG only regulates these two events via its inhibitory action on MCN1_STG (Coleman and Nusbaum, 1994; Bartos and Nusbaum, 1997). This is the first indication in the biological system that the gastric mill rhythm-timed LG bursts reduce but do not eliminate MCN1_STG transmitter release, although a previous computational model did support this possibility (DeLong et al., 2009b).

The prolonged gastric mill cycle period during the dual MCN1 stimulations resulted from selectively prolonged retraction during dual 5 Hz MCN1 stimulation, whereas both phases were prolonged during the dual 10 Hz and 15 Hz MCN1 stimulations. This distinction suggests that different cellular/synaptic mechanisms underlie cycle period regulation during different MCN1 firing rates. Selectively prolonged retraction occurs when the accumulation rate of the voltage-dependent inward current I_MI in LG is reduced at times when there is little or no change in the strength of Int1-mediated inhibition (Beenhakker et al., 2005; DeLong et al., 2009a,b). During each retraction phase, MCN1-released CabTRP Ia peptide re-initiates a slow build-up of I_MI in LG, because the previous build-up had decayed during protraction (Fig. 1D). In the present study, this prolonged retraction phase may have resulted from a relatively low release rate of CabTRP Ia peptide when each MCN1 is firing at 5 Hz, a firing rate that commonly generates a near-threshold level of neuropeptide release (Vilim et al., 1996, 2000; Liu et al., 2011; Ding et al., 2019). MCN1-released CabTRP Ia peptide activates I_MI in LG, and 5 Hz is the approximate threshold firing frequency for the MCN1-driven gastric mill rhythm (Wood et al., 2000; Kirby and Nusbaum, 2007; DeLong et al., 2009b). Selectively prolonged retraction may also result from I_MI declining to a lower level at the end of each LG burst during the 5 Hz dual MCN1 stimulations relative to the matched 10 Hz single stimulations, owing to LG more strongly inhibiting MCN1_STG transmitter release at the lower MCN1 firing rate. After a stronger LG inhibition of MCN1_STG, more time would be needed for sufficient I_MI to build up and trigger the next LG burst (DeLong et al., 2009b).

How the dual 10 Hz and 15 Hz MCN1 stimulations prolonged both gastric mill rhythm phases relative to their matched single stimulations is less clear. However, MCN1 activity does prolong both phases when Int1 inhibition of LG is strengthened to a greater degree than the parallel rate of I_MI build-up (Beenhakker et al., 2005). At the higher MCN1 stimulation frequencies the dual stimulations may alter the balance of CabTRP Ia and GABA release, relative to their matched single stimulations, such that the MCN1 GABAergic excitation of Int1 becomes relatively more strengthened than the peptidergic excitation of LG. Neuropeptide and small molecule transmitter release can have different Ca²⁺ and/or firing rate dependencies (Whim and Lloyd, 1989; Peng and Zucker, 1993; Liu et al., 2011; Nusbaum et al., 2017). Also, as above, the more effective LG inhibition of MCN1_STG during the dual stimulations would be likely to more completely reduce I_MI amplitude at the end of each LG burst, contributing to a longer duration for I_MI build-up during each subsequent retraction phase.

In contrast to these experiments, where optimal generation of a fully coordinated gastric mill rhythm resulted from dual MCN1 stimulation at 5 Hz, the MCN1 firing rate is often considerably higher (25–30 Hz) during fully coordinated gastric mill rhythms triggered by the VCN or POC neurons in the complete STNS (Beenhakker and Nusbaum, 2004; Blitz and Nusbaum, 2012). However, the VCN- and POC-triggered gastric mill rhythms are driven by co-activating MCN1 and CPN2, another CoG projection neuron. Also, during these latter rhythms the firing pattern of both projection neurons is not tonic but is rhythmically coordinated with the gastric mill and pyloric rhythms, due to synaptic input from the gastric mill and pyloric microcircuit interneurons Int1 and AB. Microcircuit output-linked activity patterns in projection neurons that drive rhythmic motor patterns are common across systems (Weeks and Kristan, 1978; Rosen et al., 1991; Norris et al., 1994, 1996; Puhl et al., 2012; Grillner and El Manira, 2020). It remains to be determined whether some or all of these distinctions, as well as the rhythmic sensory feedback that would be present in vivo, compensate for the degraded coordinated activity that occurred under many of our experimental conditions in the isolated STG. Although rhythmic sensory feedback is not necessary for core rhythm generation in most motor systems, such feedback commonly sculpts the final motor pattern in vivo (Wolf and Pearson, 1988; Hooper et al., 1990; Büschges et al., 1992; Combes et al., 1999; Eisenhart et al., 2000; Marder and Bucher, 2001; Smarandache and Stein, 2007; Hedrich et al., 2009; Büschges et al., 2011; Akay et al., 2014). Additionally, regarding this latter issue, DG neuron-mediated muscle contraction activates the muscle stretch-sensitive neuron GPR (Katz et al., 1989). GPR activity, in turn, excites MCN1 and CPN2 in both CoGs, influences the gastric mill rhythm generator neurons in the STG, and can entrain the gastric mill rhythm (Blitz et al., 2004; Beenhakker et al., 2005). Thus, this sensory feedback pathway may normally provide a correction signal that maintains MCN1-driven gastric mill rhythm coordination in vivo.

The superior coordination of gastric mill rhythms driven by dual MCN1 stimulation provides a reasonable explanation for why all identified inputs to MCN1 influence both copies, despite their presence in separate ganglia (Beenhakker et al., 2004, 2005; Blitz et al., 2004, 2008; Christie et al., 2004; Hedrich et al., 2009; Blitz and Nusbaum, 2012; White et al., 2017). This system design also provides a cautionary note for efforts focused on enabling recovery of function after partial loss of pathways that drive behavior. Increasing the firing rate of the remaining copies in a compromised pathway might be expected to compensate for partial loss of some copy members, but this is not the only possible outcome. For example, as noted above, neurons often exhibit firing rate-dependent changes in neurotransmitter release that can alter the balance of their ionotropic and metabotropic actions, leading to qualitative changes in microcircuit output. Additionally, insofar as neurotransmitter release for many neurons is also regulated locally at axon terminals, the balance between the projection neuron firing rate and that of the regulating presynaptic neuron can be pivotal to maintaining a coordinated motor pattern, as is the case for MCN1. In conclusion, the gastric mill microcircuit is more effectively driven by co-activating the paired projection neuron MCN1 than by its firing rate- and pattern matched single stimulation, at least in the isolated nervous system.

We thank Mr Logan Fickling for helpful feedback on earlier versions of the manuscript and for assistance in preparing Fig. 3.