Author response: Metrics of high cofluctuation and entropy to describe control of cardiac function in the stellate ganglion
Year: 2022
Type: peer-review
Abstract: Article Figures and data Abstract Editor's evaluation Introduction Results Discussion Materials and methods Appendix 1 Data availability References Decision letter Author response Article and author information Metrics Abstract Stellate ganglia within the intrathoracic cardiac control system receive and integrate central, peripheral, and cardiopulmonary information to produce postganglionic cardiac sympathetic inputs. Pathological anatomical and structural remodeling occurs within the neurons of the stellate ganglion (SG) in the setting of heart failure (HF). A large proportion of SG neurons function as interneurons whose networking capabilities are largely unknown. Current therapies are limited to targeting sympathetic activity at the cardiac level or surgical interventions such as stellectomy, to treat HF. Future therapies that target the SG will require understanding of their networking capabilities to modify any pathological remodeling. We observe SG networking by examining cofluctuation and specificity of SG networked activity to cardiac cycle phases. We investigate network processing of cardiopulmonary transduction by SG neuronal populations in porcine with chronic pacing-induced HF and control subjects during extended in-vivo extracellular microelectrode recordings. We find that information processing and cardiac control in chronic HF by the SG, relative to controls, exhibits: (i) more frequent, short-lived, high magnitude cofluctuations, (ii) greater variation in neural specificity to cardiac cycles, and (iii) neural network activity and cardiac control linkage that depends on disease state and cofluctuation magnitude. Editor's evaluation The paper thoroughly examines the role of the stellate ganglia in the control of cardiac rhythmicity. The empiric findings on stellate-ganglia-mediated regulation of cardiac activity in the presence of chronic cardiac failure, in particular, are translationally relevant and could be the basis of future drug-based interventions. https://doi.org/10.7554/eLife.78520.sa0 Decision letter Reviews on Sciety eLife's review process Introduction Neural control of cardiac function involves adaptive adjustment of mechanical and electrical activity to meet the organism's demand for blood flow. This cardioneural control scheme consists of neural populations in the central, peripheral, and intrinsic cardiac nervous systems. Interactions among components of the cardiac nervous system highlight that these neural populations work in concert, rather than as independent, singular processing units (Ardell et al., 2016). From an information processing standpoint, the operation of these interconnected neural networks has evolved to coordinate cardiac function on a beat-by-beat basis, producing the 'functional' outputs of this control scheme such as blood pressure, heart rate, or respiratory pressure (RP) and rate. Localized adaptations in the cardioneural network in response to pathology can cause an evolution of global network properties with heightened risk of poor outcomes without measurable evidence from these functional outputs (Deyell et al., 2015; Kember et al., 2013). There is a current focus on understanding cardioneural network processing within the stellate ganglion (SG), a collection of nerves serving as the major source of sympathetic input to the heart (Mehra et al., 2022). The SG (located in either side of the neck) operates as an integrative layer within the control hierarchy where it processes central cardiac inputs to the heart, receives cardiac feedback, and projects efferent control outputs to the heart. In pathological states such as heart failure (HF), morphological and neurochemical remodeling of SG neurons have been reported in both animal models (Ajijola et al., 2013; Han et al., 2012; Ajijola et al., 2015; Nakamura et al., 2016) and in humans (Ajijola et al., 2020; Ajijola et al., 2012b). Due to its key role in proarrhythmic neural signaling and convenience in surgical accessibility, clinical interventions targeting SG are used to treat various cardiovascular conditions (Vaseghi et al., 2012; Vaseghi et al., 2017; Ajijola et al., 2012a). It has also been established that an enhanced cardiac sympathetic afferent reflex contributes to sympathoexcitation and pathogenesis of HF (Wang and Zucker, 1996; Ma et al., 1997; Chen et al., 2015; Wang et al., 2017; Wang et al., 2008; Wang et al., 2014; Gao et al., 2005; Gao et al., 2007). Despite these novel interventions and general understanding, SG clinical therapy will remain largely unexplored without greatly improved understanding of SG neuronal information processing in healthy versus pathological states. Prior studies examining the SG neural activity have been limited to in vivo extracellular recordings (Armour, 1983; Armour, 1986; Armour et al., 1998; Yoshie et al., 2020; Yoshie et al., 2018). Recently, we explored network processing of cardiopulmonary transduction by SG neuronal populations in healthy porcine, defining a novel metric 'neural specificity' that measures specificity of neural firing patterns to cardiopulmonary signals (Sudarshan et al., 2021). This metric is contrastive and a measure of the difference between the probability density function (PDF) of neural 'sampling' of a control target relative to the same in the random sampling limit. While the target, left ventricular pressure (LVP) considered here is periodic this is not a necessary condition for use of the specificity metric; it is also applicable to aperiodic signals in an event-based fashion. In the current work, we investigate differences in information transfer between control and HF porcine models with multi-channel electrode arrays. We first uncover network-level spatiotemporal dynamic signatures by quantifying short-lived high cofluctuation events in neural activity. Second, we study coherence and consistency in the evolution of neural specificity with respect to the control target. Third, we expose differences in neural specificity and its coherence and consistency, via entropy, inside and outside cofluctuation events. These differences are considered for control and HF models and quantify differences in the maintenance of function between these groups. Results Neural activity was measured over 16 channels along with simultaneous LVP for approximately 6 hr of continuous recordings per animal. Representative neural activity recording for a single channel, LVP, and representative spike trains are displayed for control and HF animals in Figure 6A. A total of 17 Yorkshires (6 control, 11 HF; Figure 6D) underwent the terminal experiment described in Figure 6E. Upon the signal processing pipeline described above, we computed two event rate measures per animal as the final product representing the cofluctuations (ERMEAN, ERSTD). As the metric representing the neural specificity, we computed two entropy measures per channel (EntropyMEAN, EntropySTD), resulting in a total of 16 EntropyMEAN and 16 EntropySTD per animal. Finally, we used these metrics to quantify: (i) neural population dynamics (i.e., ERMEAN, ERSTD), (ii) neural specificity to target LVP, or cardiac control (i.e., EntropyMEAN, EntropySTD), and (iii) linkage between neural population dynamics and specificity (i.e., EntropyMEAN,EVENT, EntropySTD,EVENT). SG in HF exhibits high event rate Figure 1A-B shows event rate outcomes grouped by HF models and controls. HF animals show significantly higher event rates compared to control animals for both ERMEAN (p=0.011, effect size d=1.59, ERMEAN,HF=0.0012evts/s, ERMEAN,Controls=0.0002evts/s) and ERSTD (p=0.023, d=1.48, ERSTD,HF=0.001evts/s, ERSTD,Controls=0.0001evts/sec). The cofluctuation time series for each animal is depicted in Figure 3, where the event time series are computed. The 'events' or short-lived intervals where high cofluctuations exist are shown as level 1, leading to the event time series in Figure 4. We observe that the cofluctuations are more localized in HF animals with greater heterogeneity. Figure 1 Download asset Open asset Event rate (ER) and entropy results between control and heart failure (HF) animals (horizontal axes). White triangles indicate data points. (A, B) HF group animals show higher ERMEAN and ERSTD compared to control group (p<0.05). (C, D) HF group animals show higher entropy variability (EntropySTD, Equation 2, p=0.008, in (D)), and no difference in EntropyMEAN (Equation 2, p=0.08, in (C)). For ER, p values are from two-sample t-test or Wilcoxon rank-sum tests, depending on normality. For entropy, p-values are from linear mixed effects (LMEs, Equation 2) detailed in Materials and methods. HF animal models have heavy-tailed cofluctuation distributions We qualitatively explored the statistical distribution of the cofluctuation time series. Figure 2 shows log-normal fits for each animal group for CofluctuationMEAN and CofluctuationSTD time series, along with 68% confidence interval (CI) bounds, mean of fit (μFIT) and standard deviation of fit (σFIT). Control animals (Figure 2A–B) exhibit narrow CIs, lower (μFIT) and (σFIT) values, and tighter log-normal fits. In contrast, HF animals (Figure 2C–D, Figure 3, Figure 4) exhibit wider CIs, higher (μFIT) and (σFIT) values, and poorer log-normal fits. Of note, HF animals have heavy tails ranging further outside of confidence bounds. Figure 2 Download asset Open asset Cofluctuation histograms (calculated from mean or standard deviation of sliding spike rate, referred as CofluctuationMEAN and CofluctuationSTD, respectively) and log-normal fits for each animal group. μFIT and σFIT are the respective mean and standard deviation (STD) of fitted distribution, used for 68% confidence interval bounds. (A, B) Control animals have narrower bounds and represent a better fit to log-normal distribution. (C, D) Heart failure (HF) animals display more heavily skewed distributions that indicate heavy tails. Figure 3 Download asset Open asset Cofluctuations time series at convergent C values for each animal. (A) Cofluctuations from coactivity calculation from mean of sliding spike rate for control animals. (B) Cofluctuations from coactivity calculation from mean of sliding spike rate for heart failure (HF) animals. (C) Cofluctuations from coactivity calculation from standard deviation of sliding spike rate for control animals. (D) Cofluctuations from coactivity calculation from standard deviation of sliding spike rate for HF animals. Figure 4 Download asset Open asset Events time series at convergent (C,T) pairs for each animal. (A) Events from coactivity calculation from mean of sliding spike rate for control animals. (B) Events from coactivity calculation from mean of sliding spike rate for heart failure (HF) animals. (C) Events from coactivity calculation from standard deviation of sliding spike rate for control animals. (D) Events from coactivity calculation from standard deviation of sliding spike rate for HF animals. SG shows greater variation in neural specificity to LVP in HF We next examined the neural specificity to LVP, quantified by entropy measures in Equation 2. Figure 1C–D shows EntropyMEAN and EntropySTD, grouped by animals. Compared to the control group, SG of HF animals exhibited significantly higher EntropySTD (variation in entropy, Figure 1D, adjusted β=0.01 n.u., 95% CI=±0.01 n.u., dRM=0.73, p=0.009). However, there is no significant difference in EntropyMEAN (mean entropy) between animal groups. (Figure 1C, β=0.04 n.u., ±0.05 n.u., dRM=0.82, p=0.087). Neural network activity and cardiac control linkage depends on animal group and cofluctuation magnitude We explored the nature of cardiac control inside and outside short-duration regions of high cofluctuation, that is, 'events', characterized by strongly coherent stellate neural activity patterns. Insight into how these events may be relevant to cardiac control is considered here in the context of how control differs inside and outside events and termed 'event entropy'. First, we studied the extent to which event entropy differs inside and outside of events (Figure 5A, C, event type as fixed effect in Eq. (2)). Second, we studied whether event entropy is sensitive to the animal type characterized here as control or HF (Figure 5B, D, animal type as fixed effect in Eq. (9)). Figure 5 Download asset Open asset Event entropy Equation 2 investigation involved consideration of entropy values inside and outside of event regions. (A) There is significant difference between in EntropyMEAN,EVENT and EntropySTD,EVENT across all animals (p=0.0006). (B) There is no significant difference in EntropyMEAN,EVENT between animal groups (p=0.07). (C) There is significant difference in EntropySTD,EVENT between events and non-events across all animals (p<0.0001). (D) There is significant difference in EntropySTD,EVENT between animal groups (p<0.012). Regardless of the animal group, EntropyMEAN,NON−EVENT significantly exceeds EntropyMEAN,EVENT (Figure 5A, β=0.007 n.u., ±0.004 n.u., dRM=0.07, p<0.001). Similarly, EntropySTD,NON−EVENT significantly exceeds EntropySTD,EVENT (Figure 5C, β=0.01 n.u., ±0.002 n.u., dRM=0.29, p<0.001). An examination of the contribution of each animal group showed no significant difference between groups for EntropyMEAN,EVENT (Figure 5B, β=0.06 n.u., ±0.0 n.u., dRM=1.13, p=0.07). On the other hand, HF animals exhibited an increase in EntropySTD,EVENT compared to control animals (Figure 5D, β=0.02 n.u., ±0.02 n.u., dRM=0.75, p=0.012). These analyses imply that the linkage between neural network function and cardiac control differ inside and outside of cofluctuation events and between animal groups in the SG. Discussion In this work, we performed a novel investigation of SG neural population dynamics and neural specificity to continuous LVP in control and HF Yorkshire pigs. The methods in this work are intended to measure the way population neural activity relates to closed-loop control of a target and how that computation changes in diseased states. This was applied here to closed-loop control of cardiac output where the assumed target was LVP. The methods in this work involved. Neural specificity: A measure of bias in neural activity toward 'sampling' of specific target states. The target specificity is a contrastive measure that compares neural sampling of a target relative to random sampling of the same target. Neural specificity coherence: Entropy of neural specificity was used to measure coherence of neural specificity as a function of time. Cofluctuation events: The degree of coactivity in the dynamics of the mean and its standard deviation was measured between pairs of channels from minimum to maximum physical separation and this exposed short-duration 'events' when cofluctuation was unusually high. Event entropy: Functional significance of cofluctuation events was evaluated by comparing differences in the degree of neural specificity coherence inside and outside of events. Prevalence of short-lived cofluctuations in SG activity in HF In prior work, we identified neural specificity toward near-peak systole of the LVP waveform in control animals (Sudarshan et al., 2021). Application of this metric and the construction of a related coherence measure provided insight into differences in neural processing dynamics between control and HF animals. Our results show that cardiac control exerted within diseased states has greater variation in entropy and thus less consistency for HF animals compared to control animals. This finding may extend to other pathologies for which the cardiac control hierarchy is disrupted. Neural network activity is linked to cardiac control Based on the effect size (dRM), event entropy magnitude appears to be higher with greater variation observed in HF animals compared to control animals (Figure 5B-D). This implies a level of increased unpredictability and increased difficulty in cardiac control for animals in HF over control animals. A limitation of this result is that the effect sizes for event versus non-event comparisons are small to medium, which potentially indicates a larger study is necessary to better understand the physiological contributions from event type. Another limitation of the study lies in the absence of multiple-class pathologies (i.e., different HF models or other reproducible models) and in the absence of stratified pathologies (i.e., animal models with varying degrees of HF). Measurement of these neurocardiac metrics during slow, quasi-static application of clinically relevant stressors (Akeju and Brown, 2017; Chamadia et al., 2019) should provide unique opportunities to investigate unresolved questions. Future studies should focus on expanding the data set to examine how these metrics change with varying pathologies or varying disease models. We also cannot exclude possible effects of general anesthesia, open chest, and open pericardial effects on our findings, though the effects are likely consistent across the groups studied in the same manner. Conclusion In this study, we looked, for the first time to our knowledge, at long-term studies of in vivo cardiac control in baseline states. The baseline states provide unique signatures that differentiate animals with HF and controls. We discovered the inputs (i.e., neural signals) and outputs (i.e., blood pressure) are linked, which led us to develop metrics to analyze the dynamical state of this networked control (Gurel et al., 2022). The primary observation has been that event-based processing within the SG and its relationship to cardiac control is strongly modified by HF pathology. Our analysis is pointing to HF being best considered as a spectrum rather than a binary state. The magnitude of cofluctuation and neural specificity may give us a measure of the degree of HF and insight into the extent to which cardiac control is compromised with respect to neural specificity and/or cofluctuation. Future therapies may benefit from being able to infer the degree of HF in terms of neural markers as represented in this work, in a less invasive way. Intriguing connections involve the alignment of our work with a growing consensus in neuroscience. Spatiotemporal changes in neural activity and linkages with control targets are associated with behavioral changes and the onset and development of specific pathologies. For instance, spatiotemporal brain-wide cofluctuations were reported to reveal major depression vulnerability (Hultman et al., 2018). Neural ensembles were linked to visual stimuli in mice (Miller et al., 2014). Another study reported that brain's functional connectivity is driven by high-amplitude cofluctuations and that these cofluctuations encode subject-specific information during experimental tasks (Zamani Esfahlani et al., 2020). Similar cofluctuations were also reported to inform olivary network dynamics in the form of state changes in learning new motor patterns in mice (Wagner et al., 2021). Unique co-activation patterns in spontaneous brain activity indicated a signature for conscious states in mice (Gutierrez-Barragan et al., 2022). Global brain activity has also been linked to higher-level social behaviors (Mague et al., 2022). These parallel conclusions in cardiac and neuroscience studies indicate similar experimental methods used to measure neural integration relative to control targets. Such measurements may be instrumental to design and assess the efficacy of neurally based clinical interventions both at the level of the brain and the SG. Materials and methods Animal experiments Request a detailed protocol Figure 6 presents the conceptual overview and study design. The study was performed under a protocol approved by the University of California Los Angeles (UCLA) Animal Research Committee (ARC), in compliance with the UCLA Institutional Animal Care and Use Committee (IACUC) guidelines and the National Institutes of Health (NIH) Guide for the Care and Use of Laboratory Animals (Protocol: ARC 2015-022). Figure 6D–E summarizes the studied animal groups and experimental pipeline. Male Yorkshire pigs (n=17) weighing 57.5±12kg(mean±SD) were studied as control (n=6) and HF model (n=11) groups. For SG neural data collection, the animals were sedated with tiletamine and zolazepam (Telazol, 4–8 mg/kg) intramuscularly, intubated, and maintained under general anesthesia with inhaled isoflurane (2%). Continuous intravenous saline (8-10ml/kg/h) was infused throughout the protocol and animals were temperature maintained using heated water blankets (37−38oC). Figure 6 Download asset Open asset Experimental workflow and overall concept. (A) A linear microelectrode array (LMA) was inserted to the left stellate ganglion (SG) for each animal. A total of 16 channels of neural activity were collected along with simultaneous left ventricular pressure (LVP). Representative spike trains are displayed for each animal group. (B) Specifications of the LMA. (C) Conceptual representation of this work. SG receives efferent and afferent information from the spinal cord's intermediolateral complex (IML) and dorsal root ganglion, respectively, and transmits efferent information to the heart. In this work, we investigate neural activity and its relationship to cardiac function as represented by a control target such as LVP. (D) Among 17 Yorkshire pigs, 11 had heart failure induced by ventricular pacing, and 6 were in control group. SG recordings were collected at terminal experiments for both groups. (E) Experimental flow descripting the surgical for the recordings. dorsal by an the of the was performed to have a of the (Figure The was to expose the heart and both stellate surgical animals were to anesthesia for or to effect for with for in vivo neural recordings from the left SG. The left was and a pressure was inserted to LVP. and were and at blood were at to experimental the of the animals were under of isoflurane and cardiac was The HF model was with as described et al., and in Figure animals had a of hr and chronic was from the This process ventricular which to also as et al., 2016). the of was and the animals have been to have as HF at surgical described in Figure were and extracellular recordings were from the left SG, shown in Figure 6A. It should be that a of HF animals (n=6) underwent an application of to study its effects on the of as a However, no significant effect of was in any of the and et al., 2021). in this work, we HF animals with HF animals. We the the by both structural and functional data et al., 2021). was with studies of the left and dorsal root ganglion a of afferent was significantly within the in the for the group. the of cardiac was by the significant of neurons in Functional was by the response to the of channel and The group had a significantly lower pressure (LVP) response in the application of and that of cardiac sympathetic afferent reflex was by application in each SG neural recordings and experimental protocol Request a detailed protocol For each a microelectrode array was inserted in the of the left SG (Figure The of a of that recording and a of (Figure a total of 16 recording with by A microelectrode was used to of and the The signals were to a data and using data were in and in spike rate within of electrode a time of approximately 3 hr is the (Sudarshan et al., 2021). It should be that our study with recordings of the neural populations to the electrode The studies into cardiac nervous system used for which the target neurons should be and should be used for neural these we used electrode to the behaviors of SG neural This experimental from to recording has in the in and neuroscience an experimental to the behaviors of neural populations (Gurel et al., 2022). processing and analysis processing pipeline Request a detailed protocol A of the signal processing pipeline is in Figure In is used to the coactivity as the collection of between all possible channel The coactivity is computed at each and associated with a of neural activity (Figure This computation a sliding of coactivity referred to as the time Figure Download asset Open asset processing events of high cofluctuation in the coactivity time series are using two (i) the coactivity time series is to a time at each the of coactivity a C is and (ii) 'events' are as when of the cofluctuation time series a The used to the (C,T) detailed in this event and for the computation of the event rate (ER) mean and standard deviation which are used in the statistical These cofluctuation events are regions that expose in neural processing within the SG. These events are linked to function the consideration of how neural specificity differs inside and outside cofluctuation events in control and HF animals. The relationship between a control target such as LVP and neural activity at each channel is quantified via a varying neural specificity (Sudarshan et al., Figure The neural specificity is contrastive it is the difference between the of neural sampling of a target and the same from random The neural activity in the SG is to be a of and activity from neurons with inputs from It in this that we neural when we observe the specificity to the target or the random sampling limit. Neural specificity is a signal measured across target states at each channel as a function of time. This is for each channel, to a time series by its coherence in terms of The evolution of coherence in time to the dynamics or consistency of neural information each signal processing is provided in this Appendix 1 the of the As in the signal processing our measures are event rate, entropy, and event These metrics are
Source: eLife
Language: English
Subfield: Cardiology and Cardiovascular Medicine
Field: Medicine
Domain: Health Sciences
Open Access status: gold
Estimated APC paid: $2,500
OpenAlex ID: W4313470883