Consequences of acute and long-term removal of neuromodulatory input on the episodic gastric rhythm of the crab Cancer borealis

Abstract

For decades, the episodic gastric rhythm of the crustacean stomatogastric nervous system (STNS) has served as an important model system for understanding the generation of rhythmic motor behaviors. Here we quantitatively describe many features of the gastric rhythm of the crab Cancer borealis under several conditions. First, we analyzed spontaneous gastric rhythms produced by freshly dissected preparations of the STNS, including the cycle frequency and phase relationships among gastric units. We find that phase is relatively conserved across frequency, similar to the pyloric rhythm. We also describe relationships between these two rhythms, including a significant gastric/pyloric frequency correlation. We then performed continuous, days-long extracellular recordings of gastric activity from preparations of the STNS in which neuromodulatory inputs to the stomatogastric ganglion were left intact and also from preparations in which these modulatory inputs were cut (decentralization). This allowed us to provide quantitative descriptions of variability and phase conservation within preparations across time. For intact preparations, gastric activity was more variable than pyloric activity but remained relatively stable across 4–6 days, and many significant correlations were found between phase and frequency within animals. Decentralized preparations displayed fewer episodes of gastric activity, with altered phase relationships, lower frequencies, and reduced coordination both among gastric units and between the gastric and pyloric rhythms. Together, these results provide insight into the role of neuromodulation in episodic pattern generation and the extent of animal-to-animal variability in features of spontaneously occurring gastric rhythms.

the circuits that generate rhythmic motor patterns are particularly useful for the study of how circuit output depends on the interactions among circuit neurons and the modulatory inputs that affect them (Marder and Bucher 2007; Marder and Calabrese 1996; Nusbaum and Beenhakker 2002). Some motor patterns, such as those that drive swimming or walking, are episodic and are activated when needed by the animal. Others, such as vertebrate respiration, are ongoing throughout the animal's lifetime. The crustacean stomatogastric ganglion (STG) contains the motor neurons that are active during both the slow and episodic gastric rhythms and the faster, continuously active pyloric rhythm (Harris-Warrick et al. 1992; Marder and Bucher 2007; Mulloney and Selverston 1974a, 1974b; Selverston et al. 1976).

STG circuits have been particularly useful for the study of the acute actions of neuromodulators and neuromodulatory inputs (Harris-Warrick and Marder 1991; Marder 2012; Nusbaum and Beenhakker 2002). In vitro preparations of the stomatogastric nervous system (STNS) can be maintained for days or weeks in sterile saline (Hamood et al. 2015). This allows for the analysis of STG motor patterns over extended periods of time in the presence or absence of descending modulatory inputs (Hamood et al. 2015; Luther et al. 2003; Mizrahi et al. 2001; Thoby-Brisson and Simmers 1998, 2000, 2002).

The anatomy of the STNS allows for these modulatory inputs to the STG to be removed by a relatively simple procedure, as these inputs arrive through a single input nerve that can be severed without affecting the primary output nerves through which rhythmic STG outputs are carried. This provides the opportunity to study how stomatogastric rhythms respond to the loss of neuromodulatory inputs. This perturbation shares important features with spinal cord injury in humans. When the spinal cord is severed action potential propagation is lost through the lesion, and these isolated networks may develop pathological, spastic activity patterns that can cause significant difficulties for the patient (Beauparlant et al. 2013; Little et al. 1989). Recent work in other central pattern generator (CPG) systems has investigated how proper function might be restored after similar injuries and how underlying neuronal variability affects individual vulnerability to the loss of modulatory input (Harley et al. 2015; Sakurai et al. 2014; Sakurai and Katz 2009).

In the STG, the loss of modulatory inputs has primarily been studied using the continuously active pyloric rhythm. In a previous study we compared the pyloric rhythms generated under control conditions and after acute and long-term removal of neuromodulatory inputs (Hamood et al. 2015). Here we examine the activity of gastric units following this perturbation. The episodic gastric network may provide a better model for vertebrate spinal injury, where the affected CPGs for locomotion are episodically active. Although the gastric rhythm depends heavily on interactions with descending inputs (Bartos et al. 1999; Blitz et al. 1999; Blitz and Nusbaum 2012; Dickinson et al. 1988; Nadim et al. 1998; Nagy et al. 1988; Nusbaum and Beenhakker 2002), surprisingly, rhythmic activity in some gastric neurons persists in long-term recordings. Consequently, in this report we quantitatively describe a number of features of the rhythmic activity in gastric neurons in long-term (many days) recordings of STG preparations. Despite the fact that there are numerous studies of the gastric rhythm and its modulation, this is the first presentation of the variability within and across preparations in long-term recordings of Cancer borealis gastric neurons in the presence and absence of the influence of the descending neuromodulatory inputs.

MATERIALS AND METHODS

Animals.

Adult male C. borealis were purchased from Commercial Lobster (Boston, MA) and housed communally in artificial seawater at 10–12°C until dissection. Animals used in this study were obtained from January 2012 to March 2015.

Saline composition.

C. borealis physiological saline included (in mM) 440 NaCl, 11 KCl, 13 CaCl₂, 26 MgCl₂, 11 Trizma base, and 5 maleic acid, pH 7.4–7.5. For long-term experiments, this saline was supplemented with 5.5 mM dextrose, 50 μg/ml penicillin-streptomycin (GIBCO, reference no. 15140-148), and 500 ng/ml Fungizone (amphotericin B, Omega Scientific, Toronto, ON, Canada; catalog no. FG-70) as previously described (Hamood et al. 2015).

In vitro long-term recordings.

Petri dishes were disinfected by UV light for at least 30 min prior to use, and dissection tools were soaked in 70% ethanol to help ensure sterility of long-term preparations. On the first day of the experiments, usually within a few days of the arrival of the animals, crabs were anesthetized by storage in ice for 30 min prior to dissection. Methods were as previously described: intact STNSs, including the STG, the bilateral commissural ganglia (CoGs), and the esophageal ganglion (OG), were dissected and pinned out on a Sylgard-coated petri dish (Fig. 1A) and maintained in continuously superfused physiological saline containing antibiotics. Vaseline wells were built around the motor nerves containing the axons for the motor neurons of the pyloric and gastric rhythm, including the lateral ventricular nerve (lvn), the pyloric dilator nerve (pdn), the dorsal gastric nerve (dgn), and the lateral gastric nerve (lgn). Preparations were maintained in a humidity-controlled environmental chamber and held at 12°C. Fresh saline was continuously superfused at a rate of 300–500 ml/day. For decentralized experiments, decentralization was accomplished after at least 1 h of baseline recordings by transection of the stomatogastric nerve (stn), approximately midway between the STG and the OG (Fig. 1A). Dishes were then covered with Parafilm and left undisturbed for the duration of the experiments while recordings and saline superfusion continued. Fresh saline was pipetted into the Vaseline wells daily. This protocol reliably maintained healthy STNS recordings for long periods of time, lasting up to several weeks.

Fig. 1.

Baseline rhythms during gastric episodes. A: schematic diagram of the stomatogastric nervous system (STNS), as pinned out for extracellular recordings. Vaseline wells were built around at least the lateral gastric nerve (lgn), lateral ventricular nerve (lvn), and dorsal gastric nerve (dgn) for these experiments, allowing activity in the gastric and pyloric motor neurons to be monitored. These neurons have somata located in the stomatogastric ganglion (STG). Most preparations also included the pyloric dilator nerve (pdn), on which the bursting activity of the pyloric pacemaker can be monitored alone (recordings not shown). Arrow marks the location of nerve transection known as decentralization, performed in some experiments. This perturbation severs connections from modulatory inputs located in the paired commissural ganglia (CoGs) and the midline esophageal ganglion (OG). B–D: example recordings from 3 intact preparations at baseline displaying gastric rhythms including the lateral gastric (LG), dorsal gastric (DG), and gastric mill (GM) neurons. The GM neuron is the smallest-amplitude unit visible on the dgn in these traces. The faster, ongoing pyloric rhythm can be seen on the lvn. E: histogram of observed gastric frequencies for n = 31 animals with gastric rhythms at baseline. F: histogram of pyloric frequencies during gastric episodes for the same preparations (n = 30).

Data were acquired with a Digidata 1440 digitizer (Molecular Devices), a model 3500 extracellular amplifier (A-M Systems), and pCLAMP data acquisition software (Molecular Devices, version 10). Recordings always contained at least the lvn, lgn, and dgn, allowing for unambiguous detection of the burst properties for the lateral gastric (LG) and dorsal gastric (DG) neurons, and often contained the pdn for unambiguous detection of the pyloric frequency by the pyloric dilator (PD) neuron (Fig. 1, B and C).

Data analysis.

Data were analyzed off-line with custom scripts written in Spike2 (Cambridge Electronic Design, version 7) and MATLAB (MathWorks, version 2014b). These scripts allowed semiautomated detection of burst timing in DG, LG, and PD neurons, primarily by threshold-based spike detection from identified nerves. The LG neuron can be identified as the large-amplitude unit visible on the lgn and also on the lvn (Fig. 1, B–D). The DG and gastric mill (GM) neurons can be distinguished from each other by their spike amplitude on the dgn and from the continuously spiking anterior gastric receptor (AGR) sensory neuron by their bursting, sporadic activity pattern. Gastric episodes were then detected by scanning the detected bursts. A gastric episode was counted whenever more than three bursts were observed in both the LG and DG neurons, with a ≤60-s interburst interval. For some analyses, episodes were also detected if only one of these two neurons met this bursting criterion. These cases are clearly described when appropriate in results (specifically, only Fig. 5, C–G, and Fig. 10E display data from single-unit episodes). The choice of 60 s as a cutoff is arbitrary but was based on general agreement with episodes detected by eye. Phase relationships between gastric neurons were calculated using the LG neuron burst onset as a reference. Relative phase relationships of the DG neuron were counted any time there was precisely one detected DG neuron burst onset between the onset of a particular LG burst and the end of the following LG burst. This method limited detection of phase to regions of gastric activity in which such calculations appeared sensible, excluding more erratic activity patterns from the phase analysis.

Fig. 5.

Decentralization reduces gastric activity. A: example recording of a preparation with an ongoing gastric rhythm at the time of decentralization [stomatogastric nerve (stn) transected at arrow]. Activity in the gastric rhythm decreases immediately, with LG bursts ceasing and DG showing a reduced burst frequency. B: box plots and scatterplots show the overall % of time a gastric episode featuring at least DG and LG was observed for each preparation, plotted by group for n = 9 intact and n = 20 decentralized (D/C) preparations. Gastric episodes were more rare after decentralization. C: decentralized preparations show an increased % of time with no gastric activity in either the LG or DG neuron; here episodes containing only 1 of these 2 units were counted as gastric activity. D and E: when considering gastric episodes featuring either LG or DG, but not both, groups were not significantly different. Episodes with only 1 gastric unit were very rare for both groups. F and G: LG and DG neurons were less active after decentralization, when considering both 2-unit and single-unit episodes of gastric activity. For all plots, preparations with zero values are shown as 0.001%, to facilitate visualization on a log scale, and significance between groups is indicated by asterisks (*P < 0.05).

Fig. 10.

Coordination between and within rhythms is reduced after decentralization. A: box plots and scatterplots show % of overlap between LG and DG neurons for every long-term preparations in which at least 5 gastric episodes were observed (n = 7 intact, n = 6 decentralized). Specifically, these plots show the % of time during LG bursts that DG was also bursting. B: group data show the average overlap across all episodes for each preparation shown in A and also for the n = 31 baseline gastric rhythms previously shown. Decentralized rhythms had significantly more overlap between gastric unit bursts than intact and baseline preparations, which did not differ from each other (n.s.). C: box plots and scatterplots show the difference between gastric burst frequency as determined by LG and frequency determined by DG (difference index was computed as the absolute value of the frequency difference divided by the sum of the 2 frequencies) for the same preparations as A. D: group data show the average difference between LG and DG frequencies across all episodes for each preparation in C and also for the n = 31 baseline gastric rhythms. Decentralized rhythms had a significantly greater frequency difference than baseline and intact preparations, which did not differ from each other (n.s.). E: gastric episodes featuring both LG and DG activity are observed significantly more than expected by chance if LG and DG were independent of each other. This was true for both intact and decentralized preparations. F: box plots and scatterplots show the average distance to the nearest pyloric cycle integer value (absolute value) across all episodes for each long-term preparation displaying multiple gastric episodes. Coupling with the LG neuron in intact preparations was significantly greater than other groups, represented here by a lower average distance to the nearest integer. Significance of all group comparisons is indicated by asterisks (*P < 0.05, **P < 0.01, ***P < 0.001).

Summary data for gastric frequency were taken as the average of the observed LG and DG burst frequencies in a given episode; these individual neuron burst frequencies were taken as the median LG or DG frequency during the episode. Pyloric burst frequency and gastric phase relationships were also summarized for each episode by using the median values. We chose median values rather than mean values as the median is less sensitive to rare, spuriously detected outliers that are difficult to avoid in our large data set; the median also offers a better estimation of the average burst frequency during a gastric episode. When finding the within-animal average for these measures, averaged across all episodes, the mean was used, and these values were then used to compare differences by group.

Box plots were often used to visualize data. These plots show the median and first and third quartiles of the given data, while whiskers extend to the most extreme datum within 1.5 times the interquartile range from the nearest quartile.

Statistics.

Statistical significance of all tests was assessed at the P < 0.05 level, unless otherwise indicated. Significance of correlations between phases and frequencies was assessed with Spearman's rank correlation test, and significance was assessed at the P < 0.05 level. The significance of integer coupling between the pyloric and gastric rhythms was assessed with a shuffling procedure described below; the degree of integer coupling for each shuffled version of the pyloric and gastric data was assessed by computing the average absolute value difference from the nearest integer. Groups were compared on phase data (LG off, DG on, DG off) with the Watson-Williams test for circular data. Groups were compared for gastric on time by independent samples t-tests. For burst frequency, LG/DG overlap, LG/DG frequency difference, and integer coupling groups were compared by one-way ANOVA followed by Tukey's test to specify group differences. The significance of group proportions for gastric activity was assessed by χ²-test. All tests were computed in MATLAB (MathWorks, version 2014b).

RESULTS

Baseline gastric rhythms.

We first describe the spontaneous gastric rhythms observed across a population of 31 freshly dissected C. borealis STNSs. Figure 1A presents a schematic diagram of the STNS, as pinned out for the in vitro recordings used for all the experiments described in this work. Extracellular nerve recordings were obtained from at least the lvn, dgn, and lgn. These nerves allow activity in the pyloric and gastric rhythms to be monitored as shown in the examples in Fig. 1, B–D. The ongoing, triphasic pyloric rhythm can be observed on the lvn. The LG neuron can be seen firing bursts of action potentials both on the lgn and the lvn, while the DG neuron can be seen firing on the dgn. The GM neurons are also seen firing bursts on the dgn, out of phase with the DG neuron. Activity in GM neurons is sometimes difficult to observe with extracellular recordings of the dgn alone and thus was not always analyzed. The DG and LG neurons burst in an alternating pattern during episodes of gastric rhythm activity.

The frequencies of spontaneous gastric rhythms and simultaneously active pyloric rhythms are shown in Fig. 1, E and F. The average gastric rhythm frequency of 0.097 ± 0.026 Hz [standard deviation (SD)] is approximately an order of magnitude slower than corresponding pyloric rhythm frequencies (1.38 ± 0.26 Hz), consistent with previous reports. Similar to pyloric rhythms, gastric rhythm burst frequencies varied across a two- to threefold range (Fig. 1, E and F).

The gastric rhythm is known to exhibit different modes of behavior (Blitz et al. 2004; Heinzel 1988; Heinzel et al. 1993), which are thought to allow the stomach to process food of differing consistency. Although features of the in vitro activity pattern depend on which inputs are responsible for the activation of the modulatory inputs (Beenhakker et al. 2005, 2007; Blitz et al. 1999, 2004, 2008), we did not attempt to sort the observed spontaneous gastric rhythms into clusters. However, the features of the gastric rhythms examined and reported here appear to vary smoothly across the population. This likely reflects the fact that we did not attempt to activate specific modulatory inputs and therefore in the long-term experiments these descending modulatory inputs may have been themselves spontaneously active in various patterns.

Phase relationships are organized.

A quantitative examination of the phase relationships among the LG, DG, and GM neurons revealed consistent phase organization across animals, similar to that seen in the pyloric rhythm. Median phase relationships for these 31 preparations are summarized on the concentric circular phase plots in Fig. 2A. The GM and LG neurons tended to burst in a similar phase of the gastric rhythm, while the DG neuron fired out of phase with LG. We quantified gastric neuron phases relative to LG burst onset and plotted them for each preparation as a function of gastric frequency (Fig. 2B). Similar to the pyloric rhythm, the phase relationships of gastric neurons were well conserved across the range of gastric frequencies; only one phase relationship was significantly correlated with frequency (DG on, R = 0.44, P = 0.015, Spearman's rank correlation). However, phase relationships were correlated with each other across the animals in the baseline population (Fig. 2C). LG off phase was significantly correlated with DG on phase (R = 0.79, P < 0.001), DG off phase (R = 0.52, P = 0.003), and GM off phase (R = 0.79, P < 0.001).

Fig. 2.

Baseline gastric phase relationships. A: circular plots of median phase relationships among LG, DG, and GM neurons for n = 31 preparations with a gastric rhythm at baseline. Each concentric circle represents the on or off phase of a particular component of the gastric rhythm, with a point on the circle representing the median value for a particular preparation. Solid lines represent circular mean vectors for each relationship (lengths scaled by largest radius on plot; longer lines indicate lower variance). Phases were quantified relative to LG burst onset. B: phase relationships for the same n = 31 preparations as a function of gastric frequency. Gastric phase was roughly conserved across frequency at the population level. Lines represent linear fits but are only significant where indicated by asterisk. C: gastric phase relationships correlate. Here phases of DG and GM neurons are plotted relative to LG off phase within each preparation. Lines represent linear fits, while significance is indicated by asterisks (*P < 0.05, **P < 0.01, ***P < 0.001).

Frequencies of gastric and pyloric rhythms are coordinated.

Previous experimental and theoretical work has described a relationship between the gastric and pyloric rhythms by which gastric cycles are integer-coupled to pyloric cycles, such that when counting the number of complete pyloric cycles that have occurred during each gastric cycle these counts tend to cluster near integer values (Bartos et al. 1999; Bucher et al. 2006; Nadim et al. 1998). Examples of this are shown for three preparations in Fig. 3. For each of these preparations, we quantified the number of pyloric cycles (including partial cycles) between the start of each burst for LG and DG separately. We then plotted the distribution of cycle counts as histograms, using a bin size of 0.1 cycles. Consistent with previous reports, these values tended to cluster around integer values for the LG neuron (Fig. 3, A1, B1, and C1). However, this relationship was much weaker for the DG neuron (Fig. 3, A2, B2, and C2). We evaluated the significance of these couplings with a previously described shuffling procedure (Bucher et al. 2006). First, we measured the total distance to the nearest integer (absolute value) for each gastric neuron burst. Then we shuffled the pyloric periods 1,000 times, repeated the calculation each time, and asked how many of these 1,000 had greater nearest-integer distances than the original data, obtaining a P value. When quantified in this way, 26 of 30 baseline preparations for which pyloric and gastric data were available showed significant integer coupling with the LG neuron at the stringent P < 0.01 significance level. In contrast, only 3 of 30 DG neurons were significantly integer-coupled to the pyloric rhythm. We also confirmed that LG neurons were more strongly coupled than DG neurons by computing the average integer distance for these neurons for each baseline preparation and plotted these data in Fig. 3D. LG neuron periods, on average, were significantly closer to integer multiples of ongoing pyloric rhythm periods than were those of DG neurons (P < 0.001).

Fig. 3.

Coordination between gastric and pyloric rhythms. A1, B1, and C1: histograms plot the number of pyloric cycles [measured by pyloric dilator (PD) burst onsets], including partial cycles, between each burst onset for the LG neuron during episodes of the gastric rhythm from 3 separate preparations at baseline (bin size 0.1 Hz). These data cluster near integer values. The significance of this integer coupling between gastric and pyloric rhythms was quantified with a shuffling procedure, and P values are indicated in insets. A2, B2, and C2: histograms from the same 3 preparations show integer coupling measured in the same way, but from the DG neuron. These preparations all show significant coupling of the pyloric rhythm to the LG neuron but not the DG neuron. D: box plots and scatterplots showing the average distance to the nearest integer for each preparation as measured by LG and DG. The LG neuron burst onset was significantly more coupled to the pyloric rhythm than the DG neuron (***P < 0.001). E: gastric and pyloric rhythm frequencies were correlated across preparations.

We next assessed the coordination between the pyloric and gastric rhythms by asking whether the frequencies of the two rhythms were correlated across animals. Previous work in lobster found no relationship by linear regression analysis (Bucher et al. 2006). Because we found linear regression to be sensitive to the choice of period or frequency to describe the pace of observed rhythms, we evaluated this relationship with Spearman's rank correlation (Fig. 3E). We find that the pyloric and gastric rhythms are correlated in frequency (R = 0.56, P = 0.002). Because this correlation was so strong, we found significant, though numerically different, relationships by linear regression on period or frequency as well. We have chosen to display data as frequency, as we consider rhythms substantially near zero frequency to be qualitatively behaviorally similar; this is reflected in frequency data, but not period.

Long-term recordings.

The gastric rhythms described thus far were spontaneously active at baseline, with one analyzed episode per animal. However, it is possible to maintain preparations of the STNS with continuous extracellular recordings for time periods lasting up to several weeks (Hamood et al. 2015). Because the gastric rhythm is episodic in nature, starting and stopping over time, it is possible to study episodes of activity within preparations across time, and to compare this activity across groups after a perturbation such as decentralization. We performed these recordings while monitoring extracellular gastric activity for nine intact preparations of the STNS, maintained for many days in physiological saline, and quantified features of each detected gastric episode as we did for the baseline population of gastric rhythms. To detect gastric episodes, we recorded an episode of gastric activity each time at least four bursts of both the LG and DG neurons occurred with no greater than a 60-s interburst interval (see Table 1 for details). We did not quantify the GM neuron for these long-term preparations, as it was usually too difficult to reliably detect the low-amplitude GM signal in the recordings from the long-term experiments.

Table 1.

Detailed episode detection statistics

Condition	Total Time Analyzed, h	Detected Two-Unit Episodes	Mean Episode Duration, s	SD of Duration, s
Intact	284.4	120	330.7	367.7
Intact	269.6	180	595.1	798.4
Intact	96	0
Intact	71.5	125	297.1	233
Intact	7	4	90.8	22.6
Intact	81.8	8	304.4	297
Intact	142.9	18	470.5	186.2
Intact	68.7	173	923.4	1,859.1
Intact	94.5	11	4,049.5	11,578.8
Decentralized	285.5	0
Decentralized	155	0
Decentralized	297	243	272	168.3
Decentralized	253.4	112	342.5	424.3
Decentralized	141.7	0
Decentralized	241.2	0
Decentralized	225	0
Decentralized	112.5	0
Decentralized	233.9	0
Decentralized	93.7	0
Decentralized	336.1	0
Decentralized	402.2	42	1,387.7	2,389
Decentralized	287.2	1	257.4	0
Decentralized	210.1	0
Decentralized	120.9	0
Decentralized	70.7	0
Decentralized	119	78	342.8	573.8
Decentralized	90	0
Decentralized	163.6	98	400.4	591.8
Decentralized	113.5	16	479.8	947.2

The total number of detected episodes for each preparation examined is shown, along with descriptive statistics of episode durations. The total time analyzed was sometimes less than the total time that preparations were maintained, because of variable recording quality and occasional computer failures. The average episode duration was not significantly different between intact and decentralized preparations (P = 0.45, independent samples t-test).

Intact preparations are stable across time.

Figure 4 shows a gastric episode from an intact preparation from the first day of the experiment (Fig. 4A) and another from the sixth day (Fig. 4B). While there are observable differences between these two recordings, the general features of the gastric rhythm, such as frequency and relative phase relationships between the DG and LG neurons, are roughly maintained. To visualize the stability of gastric activity across time, we quantified the episodes observed for each preparation of the STNS and binned these episodes by the day in which they occurred. We then summarized the group data by averaging across all episodes each day for each preparation and pooled these data in Fig. 4. Because of the episodic nature of the gastric rhythm, we did not attempt to statistically test these differences across time; the number of preparations exhibiting a gastric rhythm on a particular day was often low and variable. The subset of preparations with detected gastric episodes was also different across days. However, an examination by day of gastric frequency (Fig. 4C), pyloric frequency during gastric episodes (Fig. 4D), and the overall percentage of time during which gastric activity was detected (Fig. 4E) revealed no apparent differences across time. Phase relationships between the LG and DG neurons (Fig. 4F) were more variable; however, these units maintained an out-of-phase pattern of bursting activity. Thus we conclude that activity in the gastric network of intact preparations of the STNS remains relatively stable as preparations are kept healthy across days.

Fig. 4.

Gastric rhythms are stable across time in preparations with the STNS left intact. A: example gastric rhythm recorded from a preparation on the 1st day of the experiment. B: example gastric rhythm from same preparation, recorded on the 6th day of continuous recording, under conditions of continuous saline superfusion. Rhythms appear qualitatively similar across time. C: box plots and scatterplots of gastric frequencies for n = 9 long-term intact preparations. Data were binned by day, and the average value across observed episodes was plotted for each preparation. Preparations with no gastric episodes during a given time period are not shown. The subset of preparations that were active varied across days. D: same plot as C but for the pyloric frequency observed within gastric episodes. E: box plots and scatterplots show % of time for each preparation that an active gastric rhythm was observed. Preparations with no gastric episodes during a given time period are shown as 0.001% to facilitate visualization on a log scale. F: gastric phase relationships between the LG and DG neurons binned by day. Each point represents the mean indicated phase relationship of all detected gastric episodes in a particular preparation on the indicated day. Lines show the circular mean vector for each phase relationship (lengths scaled by largest radius on plot; longer lines indicate lower variance).

Decentralization reduces gastric activity.

While gastric activity was relatively infrequent for intact preparations, it was observed even more rarely for decentralized preparations. Figure 5A shows an example recording demonstrating the acute effect of transection of the stn on an ongoing gastric rhythm. In this example, gastric activity immediately became drastically reduced, as the LG neuron stopped bursting and the DG neuron showed a much reduced frequency, which trailed off to zero over the course of several minutes. This immediate cessation of activity was typical and is perhaps one reason why gastric rhythms have been thought not to occur after the loss of neuromodulatory inputs. We maintained 20 STNS preparations for many days after decentralization (see Table 1 for details) and detected and quantified episodes of gastric activity as described above. For this analysis we also included episodes of activity that featured at least four bursts of either DG or LG neuron singly, to more completely describe the sparse activity of gastric neurons under these conditions. All observed gastric episodes for each preparation were considered.

Gastric activity was greatly reduced across time for decentralized preparations. After decentralization, the percentage of time gastric episodes involving both DG and LG were detected was significantly reduced (Fig. 5B; P = 0.011). To determine the extent to which DG and LG were affected by decentralization individually, we then considered the percentage of time that activity involving only one unit, DG or LG, was detected. Combining these data, we find that the percentage of time with no detected gastric unit activity in either unit was significantly increased after decentralization (Fig. 5C; P = 0.011). However, there was no significant difference in the amount of time that episodes of only LG (Fig. 5D; P = 0.90) or only DG (Fig. 5E; P = 0.61) were detected, with these “half-episodes” being extremely rare for both groups. When these data were combined to compute the total percentage of time active for each neuron individually, decentralization reduced the total active time of the LG neuron (Fig. 5F; P = 0.011) and the DG neuron (Fig. 5G; P = 0.011). Overall, 8 of 9 intact preparations displayed gastric episodes including both the DG and LG neurons in the days after dissection from the animal, compared with 7 of 20 decentralized preparations (P = 0.007).

While decentralization greatly lowered the number and overall duration of observed gastric episodes, it did not eliminate these episodes completely. Figure 6 demonstrates example recordings of the lvn, dgn, and lgn for four decentralized preparations of the STNS. While these episodes contain rhythmic bursting in both the DG and LG neurons, they are highly variable in other ways. Some of these episodes show fast ongoing pyloric activity (Fig. 6B), while some show barely any pyloric activity at all (Fig. 6C). Some episodes show frequencies in the approximate range of intact preparations (Fig. 6A), while others are remarkably slow (Fig. 6C) and some seem to show the different components of the gastric rhythm bursting at unrelated frequencies (Fig. 6D). Additionally, the phase relationships evident in these examples are highly variable, and the antagonistic activity pattern most often observed for intact preparations of the STNS is disrupted, with the DG and LG neurons often firing near-simultaneous bursts (Fig. 6A).

Fig. 6.

Examples of gastric episodes from long-term preparations. A–D: examples from 4 different preparations of detected gastric episodes featuring bursting activity in both LG and DG neurons. Gastric activity in decentralized preparations was more rare, slower, and generally displayed less organized and consistent phase relationships among component neurons.

Decentralized gastric episodes are less correlated within animals.

Because the gastric rhythm is episodic within animals, turning on and off across time, long-term recordings allow the determination of the relationships between phases and frequencies, previously described for the population of baseline rhythms, within each animal. Figure 7 shows the results of this analysis for two characteristic examples of intact preparations and two characteristic examples of decentralized preparations that displayed many episodes of gastric activity. For intact preparations, strong within-preparation correlations were observed between gastric frequency and the relative phase relationships of DG and LG neurons (Fig. 7, A1 and B1), an interesting finding considering these correlations were weak or nonsignificant for the population at baseline. However, for the two example decentralized preparations, only LG off was significantly correlated with frequency, and the sign of this correlation was reversed (Fig. 7, C1 and D1). Similarly, for the two intact examples, significant within-animal correlations were observed between LG off phase and the on and off phases of the DG neuron (Fig. 7, A2 and B2); however, these same relationships were nonsignificant for the decentralized examples (Fig. 7, C2 and D2). In general, the episodes of gastric activity observed for these two preparations were much more variable, erratic, and phase-disorganized than for intact preparations. In contrast, neither the intact nor decentralized long-term preparations seemed to show strong correlations between gastric and pyloric frequencies, as were observed in the baseline population (Fig. 7, A3, B3, C3, and D3), although in some cases this correlation was significant both for intact and decentralized experiments (Fig. 7, A3 and C3).

Fig. 7.

Within-preparation phase and frequency relationships of gastric episodes. A1 and B1: phase relationships as a function of gastric frequency for 2 example preparations of the STNS left intact and recorded across many days. Each point represents the median value of the indicated phase relationship for 1 gastric episode. These preparations exhibited many episodes of gastric activity featuring bursting activity of both LG and DG neurons; all detected episodes for which phase could be quantified are shown (A, n = 108; B, n = 170). A2 and B2: relative phase relationships between LG and DG neurons for the same 2 preparations. Each point represents the median value of the indicated relationship for 1 gastric episode; all episodes for which phase could be quantified are shown. A3 and B3: pyloric and gastric frequency relationships during all observed gastric episodes, within the same 2 intact preparations. C and D: frequency and phase relationships shown in the same way for 2 example decentralized preparations that exhibited many episodes of gastric activity (C, n = 183; D, n = 94). For all plots, lines indicate linear fits but are only significant where indicated by asterisks (*P < 0.05, **P < 0.01, ***P < 0.001).

Decentralized episodes are slower and more variable.

We quantified the differences observed across animals and conditions in observed gastric episodes and summarized these data in Fig. 8. Figure 8A shows scatterplots and box plots of all detected gastric episodes for each preparation for which at least three gastric episodes were observed. In Fig. 8A, far left, is a similar plot for the population of 31 baseline gastric rhythms, for comparison. Group data for frequency are shown in Fig. 8B, with each preparation summarized by its median episode frequency across all episodes. Decentralization reduced the frequency of observed gastric episodes (groups compared by 1-way ANOVA, P < 0.001), relative to intact controls (Tukey's test, P = 0.012) and the population of baseline rhythms (Tukey's test, P < 0.001); however, there was no significant difference between long-term intact preparations and the baseline population (Tukey's test, P = 0.42). Similarly, the frequency of the pyloric rhythm during gastric episodes was reduced after decentralization relative to intact controls (ANOVA, P < 0.001; Tukey's test, P < 0.001; Fig. 8, C and D) and the baseline population (Tukey's test, P < 0.001), while the intact controls did not differ from baseline rhythms (Tukey's test, P = 0.43). Thus long-term preparations of the intact STNS display episodes of gastric activity that continue to be similar to baseline examples across time, while those of decentralized preparations are frequency reduced, similar to the effects observed for the pyloric rhythm (Hamood et al. 2015).

Fig. 8.

Episodes of gastric activity are quantitatively different after decentralization. Episodes considered here featured bursting activity in both LG and DG neurons. A: box plots and scatterplots of gastric frequency for each detected gastric episode in every long-term preparation for which at least 5 episodes were observed (n = 7 intact, n = 6 decentralized). B: group data show the average gastric frequency across all episodes for each preparation shown in A and also for the n = 31 baseline gastric rhythms previously shown. Decentralized rhythms were slower relative to baseline and intact rhythms, which did not differ from each other. C: box plots and scatterplots of pyloric frequency within gastric episodes for the same preparations and episodes as A. D: group data show the average pyloric frequency within gastric episodes across all episodes for each preparation shown in B and also for the n = 31 baseline gastric rhythms previously shown. Decentralized pyloric rhythms were slower than baseline and intact rhythms, which did not differ from each other. E: within-preparation variability across all observed gastric episodes in frequency and phase relationships among gastric neurons. Frequency variability is represented by the coefficient of variation (CV), while phase variability is represented by circular variance. F: across-preparation variability, measured similarly, considering the average values across all episodes for each preparation. Significance of group differences is indicated by asterisks (*P < 0.05, **P < 0.01, ***P < 0.001); n.s., not significant.

In the pyloric rhythm, decentralization also increases the variability of observed rhythms, viewed both within and across animals (Hamood and Marder 2015). We tested the variability of the gastric rhythm within each decentralized preparation by calculating the coefficient of variation (CV) of the gastric and pyloric rhythm frequencies and the circular variance of the phase relationships between the LG and DG neurons (Fig. 8E). We did not observe a significant increase in within-preparation variability for the frequency of the gastric rhythm directly, although a trend toward increasing variability was apparent (P = 0.049, uncorrected for multiple comparisons), or for LG off phase (P = 0.35). Within-preparation variability was significantly increased for in-episode pyloric frequency (P = 0.012), DG on phase (P = 0.003), and DG off phase (P = 0.007). We next quantified variability across the population of preparations in which gastric episodes were observed by calculating the CV and circular variance of these same features across animals (Fig. 8F). In contrast to the pyloric rhythm, some features of the gastric rhythm did not seem more variable across animals after decentralization, including gastric burst frequency and LG off phase. However, variability was greatly increased for DG on phase, DG off phase, and pyloric frequency, the latter of which is consistent with previous work that did not only consider pyloric rhythms during gastric episodes.

Average gastric phase relationships are altered by decentralization.

Figure 9 summarizes gastric phase relationships for the six intact preparations featuring the largest number of gastric episodes (Fig. 9A) and the six decentralized preparations that featured multiple gastric episodes (Fig. 9B). While the previously described increase in phase variability within and across animals is readily apparent in these individual examples, we assessed the average change across animals in gastric phase by summarizing each preparation by its mean gastric phases in Fig. 9C. Decentralized preparations were significantly different from intact preparations in average phase of LG off (P = 0.015), DG on (P < 0.001), and DG off (P = 0.011). Decentralized preparations also differed from the population of baseline rhythms on those three features (all P < 0.001). Conversely, long-term intact preparations were not significantly different from the baseline population on any of these three measures of phase (LG off, P = 0.76; DG on, P = 0.069; DG off, P = 0.72). Thus, similar to what we have shown for the pyloric rhythm, long-term intact preparations of the STNS continue to display gastric rhythms in many ways statistically indistinguishable from baseline rhythms, while decentralized preparations rapidly diverge and are robustly different from both groups.

Fig. 9.

Gastric phase relationships are altered after decentralization. A: phase plots of LG and DG for observed gastric episodes in the 6 intact preparations with the greatest number of observed episodes. Each point represents the median of the indicated phase relationship for 1 gastric episode; all episodes for which phase could be quantified are shown. Lines represent the circular mean vector of each group (lengths scaled by largest radius on plot; longer lines indicate lower variance). B: similar phase plots of LG and DG for every observed gastric episode for which phase could be quantified in the 6 decentralized preparations that displayed multiple episodes of 2-unit gastric activity. C: group plots of average gastric phase for long-term intact (top), long-term decentralized (bottom), and baseline (right) preparations. Gastric phase relationships in decentralized preparations were significantly different (*P < 0.05, ***P < 0.001) from intact and baseline preparations, which did not differ from each other (n.s.).

Reduced coordination between rhythms.

While decentralized preparations produced slower, more variable rhythms, we also observed apparent differences in the coordination of LG and DG neurons. We considered this coordination both in terms of the loss of the normal antagonistic bursting behavior and also in their sometimes divergent frequencies within episodes (Fig. 6). We quantified these differences and also assessed the degree to which coordination between the gastric and pyloric rhythms was affected by this perturbation (Fig. 10). To assess the degree to which LG and DG neurons were bursting in their normal antagonistic pattern, we quantified the percentage of time in each LG burst for which DG was also bursting, and termed this the “LG/DG overlap.” Figure 10A shows the calculated overlap for each detected episode in every intact and decentralized preparation of the STNS and includes the baseline population at the far left for comparison; group data are shown in Fig. 10B. Decentralized preparations had increased overlap between LG and DG neuron bursts compared with both long-term intact preparations (ANOVA, P < 0.001; Tukey's test, P ≤ 0.001) and the baseline population (Tukey's test, P < 0.001). In contrast, intact preparations did not differ from the baseline group (Tukey's test, P = 0.99). We also quantified the degree to which the observed frequencies of LG and DG neurons within each episode differed, by calculating the absolute value of the difference between these two frequencies and dividing by the sum of these two frequencies, yielding a difference measure that ranges from 0 to 1. This measure is shown for all observed episodes in each preparation in Fig. 10C and across groups in Fig. 10D. Here too decentralized preparations were significantly different from both long-term intact preparations (ANOVA, P < 0.001; Tukey's test, P < 0.001) and the population of spontaneous baseline rhythms (Tukey's test, P < 0.001). Long-term intact preparations did not differ from the baseline population (Tukey's test, P = 0.89).

The general lack of coordination between DG and LG neurons in decentralized preparations led us to consider whether their activity patterns were totally independent or whether they were preferentially activated together. To address this question, we used the total percentage of time activity was detected in the LG and DG neuron individually, as computed in Fig. 5, F and G. These percentages include both the fraction of time an episode was detected with both units and the fraction of time an episode was detected featuring each unit individually. We multiplied these percentages together for each preparation, obtaining an expected value for the percentage of time a gastric episode featuring both units would be detected if the activity of these two neurons were completely unrelated. We then compared these values to the observed values for both intact and decentralized preparations (Fig. 10E). Gastric activity featuring both units was observed significantly more than expected by chance for both the intact (P = 0.033, paired t-test) and decentralized (P = 0.015, paired t-test) populations. Thus, although decentralized rhythms feature a marked lack of coordination between gastric units, these units are still being coactivated by some STG-intrinsic mechanism.

Finally, we assessed the degree to which coordination between the pyloric and gastric rhythms was affected by decentralization by quantifying the integer coupling between the LG and DG neurons and the number of pyloric cycles between each burst, as shown for the baseline population in Fig. 3. For each episode, we calculated the number of pyloric cycles per LG and DG burst (including partial cycles) and summarized each by the average distance to the nearest integer value across all bursts (Fig. 10F). For intact preparations, this distance was lower for LG bursts than DG bursts (ANOVA, P = 0.005; Tukey's test, P = 0.006), consistent with the stronger coupling to the LG neuron observed in the baseline population. However, for decentralized preparations there was no difference between coupling of LG and DG neurons (Tukey's test, P = 0.99). LG coupling was also reduced on average in decentralized preparations relative to those left intact (Tukey's test, P = 0.026). Episodes of gastric activity in decentralized preparations are less coordinated both within gastric units and also with simultaneously active pyloric rhythms.

Changes between groups persist across time.

Changes following decentralization have been shown for intrinsic properties of pyloric neurons (Khorkova and Golowasch 2007; Temporal et al. 2014; Thoby-Brisson and Simmers 2002), although recent work has shown that pyloric activity does not recover to levels resembling preparations left intact across time (Hamood et al. 2015). The degree to which similar compensatory regulation of excitability might occur in the gastric neurons is unknown, and may be of particular interest considering that the episodic nature of gastric activity presents an apparent challenge to regulatory schemes that depend on an ongoing readout of neuronal activity. To assess the degree to which gastric activity might recover across time after decentralization, we performed a day-by-day analysis of decentralized gastric episodes similar to that previously shown for intact preparations (Fig. 11). We did not perform a statistical analysis of the significance of these data because of the episodic nature of the activity and the variable and small population displaying rhythms at any given time point. However, this analysis reveals no obvious differences in gastric frequency (Fig. 11A) or pyloric frequency during gastric episodes (Fig. 11B). As with long-term intact preparations, gastric phase relationships were more variable (Fig. 11D). Activity-dependent regulation might be most apparent by an increase in the percentage of time gastric episodes are observed, but no obvious changes are visible here, either (Fig. 11C); however, a slight trend toward increasing gastric activity across time is present in the data.

Fig. 11.

Gastric rhythms remain altered across time after decentralization. A: box plots and scatterplots of gastric frequencies observed for n = 20 long-term intact preparations, 7 of which displayed gastric episodes at some point after decentralization. Data were binned by day, and the average value across observed episodes was plotted for each preparation. Preparations with no gastric episodes during a given time period are not shown. The subset of preparations which were active varied across days. B: same plot as A but for the pyloric frequency observed within gastric episodes. C: box plots and scatterplots show % of time for each preparation that an active gastric episode was observed. Preparations with no gastric episodes during a given time period are shown as 0.001% to facilitate visualization on a log scale. D: gastric phase relationships between the LG and DG neurons binned by day. Each point represents the mean indicated phase relationship of all detected gastric episodes in a particular preparation on the indicated day. Lines show the circular mean vector for each phase relationship (lengths scaled by largest radius on plot; longer lines indicate lower variance).

DISCUSSION

It has been known for many years that denervation or long-term perturbations of activity can change the receptor number and distribution on target neurons and muscles (Axelsson and Thesleff 1959; Kilman et al. 2002; Moore et al. 1971; Thesleff 1979; Thesleff and Quastel 1965; Turrigiano 2011). This presents an interesting challenge for understanding how excitability and responses to modulators are controlled in neurons that may only be episodically activated, sometimes with days or weeks between epochs of activity. It is generally thought that the episodic gastric rhythm is specifically activated by descending neuromodulatory neurons, triggered by specific sets of sensory inputs (Beenhakker et al. 2005, 2007; Blitz et al. 1999, 2004, 2008; Combes et al. 1999; Dando and Selverston 1972; DeLong and Nusbaum 2010; Nusbaum and Marder 1989a). Because the period of time between feedings may vary considerably in the wild, it is also possible that the mechanisms governing the activation of the gastric rhythm must be robust to a very intermittent activation schedule.

Rich modulation of the gastric network.

This work investigates the combined effects of the loss of many modulatory inputs to the STG on the gastric rhythm. Projections to gastric neurons make direct synapses onto core neurons of the gastric CPG such as LG and interneuron 1 (Int1) (Coleman et al. 1995; Norris et al. 1994). These modulatory projections also receive feedback connections from STG neurons, the synapses of which may exist inside the STG or in the anterior ganglia containing the modulatory somata (Blitz and Nusbaum 2012; Nusbaum et al. 1992). Previous work has demonstrated a rich diversity of roles for these projection neurons in driving and shaping gastric behavior, as well as influencing interactions between the pyloric and gastric rhythms (Blitz and Nusbaum 2008; White and Nusbaum 2011; Wood et al. 2004). Distinct input pathways have been shown at times to activate similar gastric patterns and at other times to activate different versions of gastric rhythm, which depend on feedback projections back to the CoGs (Blitz and Nusbaum 2012; Saideman et al. 2007; Wood et al. 2000). Sensory inputs from the stomach are also important for shaping gastric activity both in the intact animal and in vitro (Beenhakker et al. 2004, 2005; Blitz et al. 2004; Combes et al. 1995; Goeritz et al. 2013; Hedrich et al. 2011; Katz et al. 1989; Katz and Harris-Warrick 1989, 1991). Thus decentralization might represent a more dramatic perturbation for the gastric network than for the pyloric circuit. Indeed, without drive from these descending inputs to the STG, it may be considered surprising that the gastric network is active at all, and the source of STG-intrinsic activation of gastric neurons is unknown.

Variability in spontaneous gastric rhythms.

Our work provides an extensive description of variability in the population of spontaneously observed gastric rhythms. Baseline variability across animals in measured features of the gastric rhythm was comparable with that previously observed in the pyloric rhythm (Hamood and Marder 2015); for example, gastric frequencies varied across a roughly threefold range. But, as in the pyloric rhythm, the relative timings of activation in the motor neurons of the gastric rhythm were conserved across frequencies. This is not surprising considering that the relative activation of different muscles involved in the rhythmic contractions of teeth is of critical importance for effective behavior (Hartline and Maynard 1975; Heinzel 1988) and thus likely to be an important target of homeostatic regulatory schemes in this network. However, this variability was also constrained, as relative phase relationships were highly correlated across animals.

This analysis benefited from our ability to track the gastric rhythm within individual preparations for several days. Unlike the continuously active pyloric rhythm, the gastric rhythm alternates between on and off periods, even for the most active preparations. Each gastric episode typically lasts <30 min, and thus these episodes offer discrete samples of gastric activity that can be cleanly measured and compared across time. Our experiments suggest that within-animal variability in the gastric rhythm exceeds that of the pyloric rhythm. Of all features examined for within-preparation variability in gastric episodes, the least variable was pyloric rhythm frequency in intact preparations. For example, the median within-preparation CV of gastric frequency was more than twice that of the pyloric frequency (Fig. 8E). Interestingly, in the decentralized population, within-preparation variability in the pyloric frequency seemed to match that of gastric frequency.

Across preparations, variability in gastric rhythms from preparations left intact over time remained at levels comparable to that observed at baseline (Fig. 8F). This variability was also similar to that observed for pyloric frequency. Additionally, intact preparations varied in the degree to which they expressed spontaneous gastric activity at all. Some preparations displayed gastric rhythms never or only very rarely across many days of recordings, while others were consistently active (Fig. 5B). How this variability arises in network output is unclear, although many sources of variability have been identified in the STG that could contribute to these observations, including variable ion channel expression (Schulz et al. 2006, 2007), synaptic strength (Goaillard et al. 2009; Johnson et al. 1991; Thirumalai et al. 2006), and response to neuromodulation (Nusbaum and Marder 1989b; Spitzer et al. 2008). Decentralization increased the across-preparation variability of relative phase relationships between LG and DG neurons but did not seem to have the same effect on gastric frequency, in contrast to the pyloric rhythm.

Response of an episodic CPG to loss of neuromodulation.

In the continuously active pyloric rhythm decentralization reduces activity, rapidly decreasing the frequency of all preparations, sometimes to zero. After this, preparations continue to exist in a low-frequency, high-variability state (Hamood et al. 2015). However, many observable features of the pyloric rhythm remain intact. For example, the PD neuron continues to exhibit strong bursting behavior in most preparations, and LP and PD neurons, while showing altered phase relationships, continue to fire in roughly the same order, maintaining an antagonistic relationship between their burst timings. In contrast, decentralized gastric activity patterns are more profoundly affected. While these rhythms slow down to a similar extent, they are much more rarely observed, being seen in only 7 of 20 long-term preparations examined here. Furthermore, phase relationships are more severely altered, as components of the gastric rhythm sometimes fire in patterns that ignore the activity of other components of the network.

This disorganized activity pattern is reminiscent of the spastic activity patterns sometimes seen after spinal cord injury, where episodic pattern generators associated with locomotion may develop pathological activity patterns (Beauparlant et al. 2013; Ditunno and Formal 1994; Siddall et al. 1997). While we did not observe any significant effects of time that might be interpreted as functional recovery, decentralized gastric activity was intermittent and highly variable. It is probable that some plasticity mechanisms are operating during the time course of our experiments, and these mechanisms may be unable to function properly in the absence of neuromodulation, perhaps leading to aberrant compensation. In the in vivo STG, decentralization has less dramatic effects than it does in the in vitro preparation, but hormonal modulation is still available to STG networks that may facilitate continued function (Rezer and Moulins 1983, 1992).

Coupling between rhythms.

Often considered separately, the pyloric and gastric CPGs are contained in the same ganglion and are richly interconnected, even sharing some of the same component neurons (Dickinson 1995; Weimann et al. 1991; Weimann and Marder 1994). Several types of interactions have been previously described, including gastric-timed modulation of pyloric activity (Clemens et al. 1998b; Mulloney 1977), precise coupling between the periods of the two rhythms (Bucher et al. 2006), and frequency modulation of the gastric rhythm by the pyloric rhythm (Bartos et al. 1999; Nadim et al. 1998). Some of these interactions have been shown to depend on intact connections to anterior modulatory ganglia (Blitz and Nusbaum 2012), while others are STG intrinsic (Clemens et al. 1998a). Behavior and modulation can also affect the degree to which these interactions are observed: in lobsters, feeding resulted in changes in intercircuit interactions (Clemens et al. 1998b). Here we demonstrate relationships between pyloric and gastric frequencies in spontaneous preparations at baseline both in terms of the integer coupling between the rhythms and by a previously unreported, robust correlation between the frequencies of the two rhythms (Fig. 3E). Although such a relationship would appear natural given all of these previous observations, it was not found in an earlier study using lobsters (Bucher et al. 2006).

In the long-term preparations, integer coupling remained high for LG neurons from intact preparations, suggesting that the mechanisms of rhythm generation in these experiments remained stable. However, the loss of modulatory input eliminated this coupling, or at least reduced it to the low levels observed for the DG neuron in intact preparations. Integer coupling was not different across animals for intact DG neurons, decentralized DG neurons, and decentralized LG neurons. This loss of coupling is not surprising, as it depends on feedback from projection neurons with somata anterior to the stn transection in these experiments. All of these data support previous interpretations of the mechanism behind this interaction (Bartos et al. 1999; Nadim et al. 1998).

While many features of the STNS motor patterns are conserved in many crustacean species, the shape of the stomach varies across species. Notably, crabs are flat relative to lobsters, and the actual musculature has variable configurations (Maynard and Dando 1974; Meyrand and Moulins 1986; Tazaki and Tazaki 2000). There are many examples of changes in modulatory neuron function and cotransmitter complement across even closely related species (Beltz et al. 1984; Katz and Harris-Warrick 1999; Meyrand et al. 2000; Skiebe 1999; Skiebe and Ganeshina 2000). Consequently, it is not surprising that previous work has demonstrated that the coupling between spontaneous pyloric and gastric rhythms across species varies (Bucher et al. 2006; Clemens et al. 1998b; Hartline and Maynard 1975). Because these interactions can be affected by modulation, we anticipate that integer coupling may be present only in specific sets of circumstances, even in the species in which it appears (Bartos et al. 1999).

Long-term regulation of an episodic activity pattern.

We did not find any evidence for qualitative changes in the character of the activity patterns over time, for either intact or decentralized preparations. However, there did seem to be a trend for increasing numbers of gastric episodes over time in the decentralized population. How long-term activity may be regulated in an episodic pattern generator remains an interesting question and poses challenges compared with a continuously active pattern generator like the pyloric CPG. When a network is continuously active, any form of time integration of activity can reliably measure the degree to which the network is active (LeMasson et al. 1993; Liu et al. 1998; O'Leary et al. 2013, 2014), using calcium targets to regulate features such as burst frequency and duty cycle. However, for episodic pattern-generating networks, long periods of inactivity are to be expected. Any regulatory scheme designed to act on activity must somehow refrain from making extensive changes to network architecture during these quiescent periods, lest the rhythm behave improperly when next activated. Indeed, the population of intact preparations examined here showed large variability in the amount of time spent with active gastric rhythms during the course of our experiments. Yet across animals gastric rhythms for intact preparations continue to look similar to those seen immediately after dissection. While there exist theoretical frameworks for activity-dependent regulation of conductances (O'Leary et al. 2013, 2014) and synapses (Soto-Trevino et al. 2001) that may apply to the pyloric rhythm in a straightforward manner, extending these to an episodic pattern generator is an important challenge for future work.

Neuromodulation affects the STG at every level: from intrinsic conductances, to synaptic strengths, to the propagation of action potentials along axons to their motor targets (Bucher and Goaillard 2011; Marder et al. 2014). Within this complex web of interactions exist networks of highly variable underlying parameters, which are nonetheless able to produce stereotyped and behaviorally critical output patterns across animals (Goaillard et al. 2009; Marder 2011, 2012; Marder and Goaillard 2006). The molecular and cellular mechanisms that allow the nervous system to produce and maintain long-term stable motor patterns that are capable of responding to various environmental perturbations (Marder et al. 2015) await future research.

GRANTS

This work was supported by National Institutes of Health Grants F31 NS-080420 (A. W. Hamood) and MH-46742 (E. Marder).

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the author(s).

AUTHOR CONTRIBUTIONS

Author contributions: A.W.H. and E.M. conception and design of research; A.W.H. performed experiments; A.W.H. analyzed data; A.W.H. and E.M. interpreted results of experiments; A.W.H. prepared figures; A.W.H. drafted manuscript; A.W.H. and E.M. edited and revised manuscript; A.W.H. and E.M. approved final version of manuscript.