Laser ablation of Dbx1 neurons in the pre-Bötzinger complex stops inspiratory rhythm and impairs output in neonatal mice

1) A major concern is how the authors seem to extrapolate the findings on the numbers of neurons that when ablated lead to rhythm termination to the in vivo state. The conclusion that the “preBötC core is very sensitive to just a few constituent neurons (i.e. <100)” may not be valid for the in vivo state of the preBötC network, a caveat that the authors should discuss in the text. Their own results with SP injection suggests that more excitation and/or modulation in vivo may make the rhythm generator immune to loss of 15% of the Dbx1 Interneurons. They should also draw back on their extrapolation of how these results relate to the loss of respiratory function in disease sates in vivo.

We addressed this critique in response to Reviewer #2 below. The most important revisions to the text, which pertain to this critique, are now found in the Discussion section.

2) The effects deletions of Dbx1 neurons on motor discharge appear to be only on amplitude and not on patterning of the motor outflow. By pattern the reviewers mean phase relations of MNs not just the number of MNs activated or their spike frequency. Careful rewriting should make this distinction clear. The authors do not demonstrate experimentally that these neurons are NOT part of the rhythm-generating kernel of excitatory neurons and hence are “non-rhythmogenic” Dbx1 premotoneurons or directly that they are premotor. This inference/hypothesis came from the excitatory network model. It is a reasonable hypothesis but at present we don't know enough about the spatial organization and projection patterns of these Dbx1 neurons. It is possible that some of these neurons have axon collateral projections within the preBötC and also contralaterally in addition to ipsilateral projections to the hypoglossal motor nucleus. Also the authors have not shown us that the 2 cells that they identified actually have synaptic connections with motor neurons.

The Reviewing editor is correct. We did not demonstrate that Dbx1 preBötC neurons with premotor functionality are not also part of the rhythmogenic core, but the model-experiment disparity suggests that this is the case. That is, we had to delete approximately twice as many cells in the experiments as in the simulations of a purely rhythmogenic network to cause rhythm cessation. If some Dbx1 neurons were simultaneously part of the rhythmogenic and premotor subsets, then two predictions would arise: 1) we would expect the experiments to decline in output amplitude, but not the simulations (a pure rhythmogenic setup), and 2) we would expect both simulations and experiments to require similar ablation tallies to stop the rhythm. Prediction 2 was clearly not observed.

It is also correct that we did not demonstrate synaptic connections with XII motoneurons. Our evidence for premotor function is based on: i) a decline in XII output after laser ablation, ii) antidromic activation from the XII nucleus. (Note, the XII nucleus is virtually homogenous containing >95% motoneurons and only 5% GABAergic interneurons. Thus, it is highly probable that any axons antidromically activated via stimulation within the XII nucleus synapse on XII motoneurons.) And, iii) biocytin reconstruction showing axon trajectory toward/into the XII nucleus. These are indirect demonstrations, but taken together make a strong case in our view.

We further address this critique of ‘pattern’ in response to Reviewers #1 and #3, below.

3) We have appended the original reviews because they emphasize different aspects of the paper but are non-conflicting. They can be fruitfully used to revise the manuscript.

Reviewer #1

[…] This paper is clearly written and meticulously documented with convincing data. Each figure conveys important data and the Discussion appropriately points out potential clinical implications. This work is a significant and necessary step forward from previous laser ablation studies by the same lab that identified potential rhythmogenic targets by Ca imaging [Hayes JA, Wang X, Del Negro CA. 2012. Cumulative lesioning of respiratory interneurons disrupts and precludes motor rhythms in vitro. PNAS 109(21): 8286-91].

We thank Reviewer #1 for a supportive assessment.

Reviewer #2

[…] One major concern to be addressed is that the quantification of the size of the Dbx1 preBötC population required to maintain rhythmogenic function in vitro (∼15% of the total Dbx1 population) may not be representative of the situation operating in vivo because of major differences in the level of Dbx1 neuron excitation and hence ability of this network population to generate inspiratory rhythm in different states. There are sources of excitatory/neuromodulatory drive to the preBötC operating in vivo (e.g. RTN) that are absent in the slice preparations and the slices are in fact in a low excitation state (i.e., compared to the in vivo mouse, very low inspiratory discharge frequency dependent on elevated extracellular potassium for stable slow rhythm generation). The conclusion that the “preBötC core is very sensitive to just a few constituent neurons (i.e. <100)” (and the authors suggestion that this may have ramifications for the loss of respiratory function in disease states) may not be valid for the in vivo state of the preBötC network, a caveat that the authors should discuss in the text. Have the authors effectively varied the state of excitation for example in the model network to gain insight into how level of excitation affects the critical mass of neurons required to sustain rhythmogenic function?

Recognizing that measurements in vitro may not map directly to the preBötC in vivo, we acknowledge in the edited version that our quantifications apply for in vitro conditions only. Furthermore, we acknowledge the caveats inherent in extrapolating our in vitro measurements to the in vivo condition. Some examples:

The sentence quoted above by Reviewer #2, and three additional sentences were modified as follows, “… the preBötC core in vitro, is very sensitive to the loss of just a few constituent interneurons (i.e., <100).” This sensitivity to neuron loss may be accentuated in reduced slice preparations lacking excitatory and neuromodulatory drive from the rostral medulla and pons as well as peripheral chemosensory and mechanosensory vagal sensory feedback. Extrinsic sources of drive raise preBötC network excitability and enhance respiratory rhythm. An acute sensitivity to neuron loss, such as we report for slices, may not apply to the preBötC network in vivo, but this remains to be tested via quantitative cellular ablation experiments with physiological monitoring in intact animal models. (Discussion section)

And also in the Discussion section, “… unraveling the cellular and synaptic mechanisms of rhythmogenesis and motor output could be exploited to restore functionality in lesioned slices and, to the extent that our observations apply in vivo, to develop clinical therapies that bolster respiratory function in pathological conditions of animal models or human patients.”

The results with the excitatory network model are instructive regarding perturbations of population burst frequency vs. amplitude with simulated single-cell ablation. The authors state (p. 8, para. 1) that some of their test simulation results substantiate that the model networks well represent the neonatal mouse preBötC in vitro.” In the experiments of Kam et al. 2013 cited, network bursts were generated after a substantial delay (200 - 300 ms) from the onset of neuronal activation. Does the present network model predict this temporal feature? Also the cellular burst generation and terminations mechanisms represented in the Rubin-Hayes model have not been verified experimentally, which should be noted in the Discussion. On the other hand, it could be that the type of test results used to make the claim about the model as a good representation may not be particularly sensitive to the actual burst generation and termination mechanisms incorporated.

Almost exactly matching the experimental results of Kam et al. (J Neurosci 33: 3332-8, 2013), our network-wide evoked bursts emerged with a delay of 200-300 ms. For example, in our Fig. 6B simulations, the five-cell stimulation evoked a network burst with a delay of 300.3 ms; and the six-cell stimulation evoked a network burst with a delay of 215.2 ms. 6 We respectfully point out that most of the mechanisms for burst initiation and termination in the Rubin-Hayes mathematical model (Rubin et al. PNAS 106: 2939-44, 2009) have in fact been experimentally tested.

Recurrent excitation: its role in initiating the inspiratory burst phase has been documented numerous times, including two canonical papers (Funk et al. J Neurophysiol 70: 1497-1515, 1993; Wallén-Mackenzie et al. J Neurosci 26: 12294-307, 2006) and three very recent papers (Kam et al. J Neurosci 33: 3332-8, 2013 [which used glutamate un-caging to trigger recurrent excitation]; Carroll & Ramirez. J Neurophysiol 109: 285-95, 2013; Carroll et al. J Neurophysiol 109: 296-305, 2013).

CAN current: Prof. Ramirez’ group (Univ. Wash., Seattle) was the first to document the role of ICAN in preBötC bursting (Thoby-Brisson & Ramirez. J Neurophysiol 86: 104-12, 2001), and then my lab characterized the current in some biophysical detail (Pace et al. J Physiol 582: 113-25, 2007; Crowder et al. J Physiol 582: 1047-58, 2007). Subsequently Dr. S. Mironov examined the respiratory burst-generating role of ICAN in a sequence of papers (Mironov J Physiol 586: 2277-91, 2008; Mironov & Skorova. J Neurochem 117: 295-308, 2011; Mironov 591: 1613-30, 2013). Finally, my group recently demonstrated the influence of ICAN specifically in Dbx1 preBötC neurons (Picardo et al. J Physiol 591: 2687-703, 2013).

Burst termination: We demonstrated the role of the electrogenic Na-K ATPase pump current (Ipump) in burst termination in two papers (Del Negro et al. J Physiol 587: 1217-31, 2009; Krey et al. Front Neural Circuits 4: 124, 2010). Furthermore, the paper by Krey et al. also showed the contributions of Na^+-dependent K+ currents (INa-K) and ATP-dependent K+ currents (IK-ATP), which are additional activity-dependent outward currents that can contribute to burst termination as predicted by the Rubin-Hayes modeling framework (see Fig. 4 of that paper).

Cellular ‘pacemaker’ activity: In the original Rubin-Hayes model, intrinsic bursting was precluded by the parameters. In a follow-up study, we explored parameter space in the model and demonstrated that the Rubin-Hayes model could show Na⁺or Ca²⁺mediated intrinsic bursting activity (Dunmyre et al. J Comput Neurosci 31: 305-28, 2011), which was shown experimentally several times (e.g., Thoby-Brisson & Ramirez. J Neurophysiol 86: 104-12, 2001; Pena et al. Neuron 43: 105-17, 2004; Del Negro et al. J Neurosci 25: 446-53, 2005).

The only aspect of the Rubin-Hayes model that has not yet been experimentally tested is the role of short-term synaptic depression, which also aids in burst termination in the model. My lab has been working on that problem for the last year and we are writing up the study now (although obviously we cannot cite it yet).

In the final sentence regarding the modeling, Reviewer #2 rightly questions whether the specific cellular mechanisms of the Rubin-Hayes model are important for laser ablation simulations. Presumably, if the laser ablation approach interrogates network properties, then the properties of the constituent cells may not matter. However, in our view, since the experiments are actually performed on preBötC interneurons, then the simulations should employ models that well represent preBötC neurons. Therefore, we would substitute our model for a Morris-Lecar generalized model or a stripped-down Hodgkin-Huxley generic model. For the preBötC, Rubin-Hayes is a model that we developed and thus advocate, but another contemporary model of preBötC neurons was published by Rybak’s group (Jasinski et al. Eur J Neurosci 37: 212-30, 2013), which builds on a model developed by Butera’s group (Toporikova & Butera. J Comput Neurosci 30: 515-28, 2011). We are in the process of implementing this Jasinski-Rybak model to repeat the simulations in Figure 6 and Figure 6–figure supplement 1. Unfortunately the published 7 paper by Jasinski et al. contains several typos and does not provide initial conditions and the code for the Jasinksi-Rybak model is not in the public domain – so it has taken my team several months just to correctly implement the single-cell version of the model and correctly reproduce the figures of their paper (to make sure the model is setup correctly). We have now begun replicating the simulations of the present study submitted to eLife, and we intend to publish the simulations of laser ablation using Jasinski-Rybak preBötC neuron models in a forthcoming project. We anticipate that Jasinski-Rybak will give similar results to Rubin-Hayes when integrated into network simulations and subjected to piecewise cellular ablations.

An experimental issue to be addressed is that there is no information provided for the five slices studied in ablation experiments on the number of neurons ablated on each side of the slice. In the Materials and methods the authors state that “The domain can be bilaterally distributed like the preBötC and ventral respiratory column” and bilateral target maps are shown, but the authors should state how the targets were actually distributed in the experiments and any potential impact on the ablation results should be discussed.

We created a new supplementary figure (Figure 2–figure supplement 2), which plots lesion tallies for preBötC-surface and control slices. It shows the total ablation tally as well as the individual tallies per side.

We now acknowledge the bilateral nature of target detection and laser ablation in the Results section..

Because Dbx1 preBötC neurons so often show commissural projections (Bouvier et al. Nat Neurosci 13: 1066-74, 2010; Gray et al. J Neurosci 30: 14883-95, 2010; Picardo et al. J Physiol 591: 2687-703, 2013; and Figure 8A of the present paper) we favor the null hypothesis that the connectivity of Dbx1 preBötC neurons is the same unilaterally and bilaterally. In that case, it would make no difference whether a Dbx1 neurons was ablated on side 1 or 2; the effect on network function would be the same.

Reviewer #3

[…] While the paper is clearly written, easy to follow and the figures explicit I have a few concerns:

First the degree of novelty: the elegant method consisting in lesioning specific neuronal populations within a network and more specifically the respiratory network of the preBotzinger complex while monitoring the resulting motor output has been already presented (including all the methodological controls), (Hayes, 2012). And the important role of Dbx1-derived cells in constituting the core of the preBotzinger respiratory network has also been recently published (Bouvier et al., 2010; Gray et al., 2010). Therefore the experiments dedicated to examine the effects of laser ablation of Dbx-derived respiratory neurons on the respiratory network function seem to only complement/confirm previous already published results, but do not really bring any novel results susceptible to significantly improve the current knowledge on respiratory rhythmogenesis.

It is true that Bouvier et al. (Nat Neurosci 13: 1066-74, 2010) and Gray et al. (J Neurosci 30: 14883-95, 2010) tested the rhythmogenic role of Dbx1-derived interneurons. It is also true that our first laser ablation study (Hayes et al. PNAS 109: 8286-91, 2012) estimated core rhythm generator size, and established methodological controls. Nevertheless, the present study is novel and important in our view. Hayes et al. (2012) identified neurons by Ca²⁺ imaging, which cannot differentiate glutamatergic rhythm-generating neurons from GABA/glycinergic non-rhythmogenic neurons because both neuron classes oscillate in sync with the inspiratory rhythm. Moreover, Bouvier et al. and Gray et al. did not attempt to quantify the size of the Dbx1 preBötC core population or test the robustness of its functionality. Therefore, the novelty of the present study is that we test the obligatory rhythmogenic role of the Dbx1 population while quantifying its size, and in doing so we avoid counting inhibitory neurons in our measurements.

An additional element of novelty in this study is that we found that Dbx1 preBötC neurons also serve in a premotor capacity. Bouvier et al. and Gray et al. did not discover that overlapping role for Dbx1 preBötC neurons. This last point is not just a new piece of information but rather a significant paradigm shift because in the respiratory CPG a single cardinal class of interneurons serves rhythm-generating and premotor functions, whereas these functions are ordinarily attributed to separate populations.

Second I have a problem with the use of the word “pattern”. This might be only a semantic problem but it should be made clear what the authors mean with pattern. To my point of view “pattern” means the respective and relative organization of the different phases of a rhythmic activity. For example for the respiratory behavior this could include expiration vs inspiration phases, the distinction between the different sub-phases of the inspiration activity.... But, to me, the amplitude of the hypoglossal motor output recorded from a slice preparation is not a parameter that can be associated with patterning. If we agree on this, then part of the conclusions of the paper are incorrect and should be substantially revised.

The word ‘pattern’ was changed to ‘amplitude’, ‘magnitude, or ‘output’ as appropriate. The specific instances are marked by red text in the revised manuscript, mainly in the Abstract, Introduction, and Discussion. In all other instances the word ‘pattern’ is only used in reference to ‘central pattern generator’, axon projection pattern, or spiking pattern. Reviewer #1 raised a similar concern and we provide the same response.

In the Introduction the authors claim that “rhythmogenic preBotzinger neurons are distinguished by glutamatergic neurotransmitter” and that “GABA- and glycinergic neurons in the preBotC influence sensory integration but are non-rhythmogenic because blocking inhibition does not prevent the respiratory rhythm to be generated”. This is not completely true as Morgado-Valle (2010) provided evidence that some rhythmically active glycinergic neurons of the preBotC exhibit pacemaker properties and Winter (2009) showed that glycinergic neurons are significantly integrated in the preBotC network as they constitute half of the rhythmically active neuron in this network. The authors should at least site these work here.

Morgado-Valle et al. (J Neurosci 30: 3634-39, 2010) showed glycinergic preBötC neurons with pacemaker properties, but that does not make such neurons rhythmogenic. In fact, Morgado-Valle et al. used the existence of glycinergic preBötC neurons with pacemaker activity to argue that pacemaker activity per se is not obligatory for rhythm generation (since it is so well established that chloride-mediated synaptic inhibition, and inhibitory neurons in the preBötC, are dispensable for respiratory rhythm in slice preparations).

The paper by Winter et al. (Pflügers Archiv 458: 459-69, 2009) was cited already in the Introduction on p. 3. We also cite a large number of studies in our Introduction showing that chloride-mediated inhibition, and thus GABA- and glycinergic interneurons, are non-rhythmogenic in vitro. In vivo studies such as Büsselberg et al. (Neurosci Lett 316: 99-102, 2001) and more recently Jaczewski et al. (J Neurosci 33: 5454-65, 2013) also show that synaptic inhibition in the respiratory network is non-rhythmogenic, but rather shapes the final output pattern of respiration and influences sensorimotor integration. The Introduction now states more clearly what the role of inhibitory neurons is, “…GABA- or glycinergic neurons (Kuwana et al., 2006; Winter et al., 2009), which influence sensory integration as well as coordinated patterns of muscle contraction for respiratory behaviors, but are non-rhythmogenic…”

In the legend of Figure 1 and Figure 4 it is written that the red dots in panel C correspond to Dbx1- neurons and the blue ones to non-Dbx1 neurons. How does the system detect the non-Dbx1 neurons? I understood that the cell to be deleted were selected on the basis of fluorescent protein expression. By definition the non-Dbx1 neurons should not be detected because they are not tdTom positive. Thus how can they be identified by the system? Please clarify this.

We apologize for unclear wording in the figure legends. Indeed, blue ROIs are not non-Dbx1 neurons. Rather, blue ROIs are rejected fluorescent targets. As we explain in the Figure 1—figure supplement 2, tdTomato-expressing ROIs are rejected as targets if they are insufficiently circular or have been previously detected in a deeper z-plane. The figure legends for 1 and 4 now read, “(C) Mask of targets showing validated Dbx1 interneuron targets (red) and regions of fluorescence that do not pass muster and were rejected as targets (blue).” We thank Reviewer #3 for pointing out this flaw in our figure legends.

It is stated that the fluorescent neurons were preferentially selected in the dorsal part of the preBotC. Why this part more specifically? If similar experiments are performed exclusively in the ventral part of the preBotC is the ablation-induced effect on XII motor output amplitude still observed? Indeed Funk (1993), Koizumi (2013) and the present work locate the inspiratory premotoneurons in the dorsal aspect of the preBotC region, so one would expect that when ablations are performed in the ventral region of the preBotC then the consequences on motor output should be less. Did the authors ever observed a difference on results obtained depending on the ventral/dorsal position of the ablation domain? Such experiments would validate part of the results presented here and would be in complete agreement with the model used in this study.

The patch-clamp recordings whose methods were described in the first submission were undertaken specifically to test whether Dbx1 preBötC neurons had projections to the XII nucleus, as would be expected for premotor interneurons. Reviewer #3 correctly points out that the dorsal preBötC would be the most likely position for such neurons based on Funk et al. (1993) and Koizumi et al. (2008 and 2013). Therefore, we targeted our patch recordings in the dorsal preBötC for that specific purpose, to look for Dbx1 neurons with putative premotor functionality.

If the domain for single-cell cumulative laser ablations were to be confined to the ventral preBötC, then one would quite logically predict that rhythmic frequency and stability would be perturbed, but effects on motor output amplitude would be minimized. Indeed, that would be consistent with the model we propose. We have not yet done the experiment in that way. Rather, we have taken the reverse approach and selectively laser ablated Dbx1 neurons with the domain set to the dorsal preBötC and even farther dorsally to the intermediate reticular formation (IRt). The prediction would be to degrade the amplitude of motor output without affecting frequency, and we have observed that result. These follow-up laser-ablation studies are not yet ready for publication and exceed the scope of the present manuscript in our view.