Author response:
Reviewer #1 (Public review):
Summary:
This manuscript by Guo and Uusisaari describes a series of experiments that employ a novel approach to address long-standing questions on the inferior olive in general and the role of the nucleoolivary projection specifically. For the first time, they optimized the ventral approach to the inferior olive to facilitate imaging in this area that is notoriously difficult to reach. Using this approach, they are able to compare activity in two olivary regions, the PO and DAO, during different types of stimulation. They demonstrate the difference between the two regions, linked to Aldoc-identities of downstream Purkinje cells, and that there is co-activation resulting in larger events when they are clustered. Periocular stimulation also drives larger events, related to co-activation. Using optogenetic stimulation they activate the nucleoolivary (N-O) tract and observe a wide range of responses, from excitation to inhibition. Zooming in on inhibition they test the assumption that N-O activation can be responsible for suppression of sensoryevoked events. Instead, they suggest that the N-O input can function to suppress background activity while preserving the sensory-driven responses.
Strengths:
This is an important study, tackling the long-standing issue of the impossibility to do imaging in the inferior olive and using that novel method to address the most relevant questions. The experiments are technically very challenging, the results are presented clearly and the analysis is quite rigorous. There is quite a lot of room for interpretation, see weaknesses, but the authors make an effort to cover many options.
Weaknesses:
The heavy anesthesia that is required during the experiment could severely impact the findings. Because of the anesthesia, the firing rate of IO neurons is found to be 0.1 Hz, significantly lower than the 1 Hz found in non-anesthetized mice. This is mentioned and discussed, but what the consequences could be cannot be understated and should be addressed more. Although the methods and results are described in sufficient detail, there are a few points that, when addressed, would improve the manuscript.
We sincerely thank the reviewer for their encouraging comments and recognition of our study’s significance. We fully acknowledge the confounding effects of the deep anesthesia used in our experiments, which was necessary to ensure the animals’ welfare while establishing this technically demanding methodology. We elaborate on these effects below and will further clarify them in the revised manuscript.
Ultimately, the full resolution of this issue will require recordings in awake animals, as we consider our approach an advancement from acute slice preparations but not yet a complete representation of in vivo IO function. However, key findings from our study—such as amplitude modulation with co-activation and the potential role of IO refractoriness in complex spike generation—could be further explored in existing cerebellar cortical recordings from awake, behaving animals. We hope our work will motivate re-examination of such datasets to assess whether these mechanisms contribute to overall cerebellar function.
Reviewer #1 (Recommendations for the authors):
On page 10 the authors indicate that 2084 events were included for DAO and 1176 for PO. Is that the total number of events? What was the average and the range per neuron and the average recording duration?
Thank you for pointing out lack of clarity. The sentence should say "in total, 2084 and 1176 detected events from DAO and PO were included in the study". We will add the averages and ranges of events detected per neuron in different categories, as well as the durations of the recordings (ranging from 120s to 270s) to the tables.
On page 10 it is also stated that: "events in PO reached larger values than those in DAO even though the average values did not differ". Please clarify that statement. Which parameter + p-value in the table indicates this difference?
Apologies for omission. Currently the observation is only visible in the longer tail to the right in the PO data in Figure 2B2. We will add the range of values (3.0-75.2 vs 3.1-39.6 for PO and DAO amplitudes, respectively) in text and the tables in the revision.
Abbreviating airpuff to AP is confusing, I would suggest not abbreviating it.
Understood. We will change AP to airpuff in the text. In figure labels, at least in some panels, the abbreviation will be necessary due to space constraints.
What type of pulse was used to drive ChrimsonR? Could it be that the pulse caused a rebound-like phenomenon with the pulse duration that drove the excitation?
As described on line 229 and in the Methods, we used 5-second trains of 5-ms LED light pulses. Importantly, these stimulation parameters were informed by our extensive in vitro examination of various stimulation patterns (Lefler et al., 2014), which consistently produced stable postsynaptic responses without inducing depolarization or rebound effects. Additionally, Loyola et al. (2024) reported no evidence of rebound activity in IO cells following optogenetic activation of N-O axons in the absence of direct neuronal depolarization. We will incorporate these considerations into the discussion, while also acknowledging that unequivocal confirmation of “direct” rebound excitation would require intracellular recordings, such as patch clamp experiments.
The authors indicate that the excitatory activity was indistinguishable in shape from other calcium activity, but can anything be said about the timing (the scale bar in Figure 4A2 has no value, is it the same 2s pulse)?
Apologies for oversight in labeling the scale bar in Figure 4A2 (it is 2s). While we deliberately refrain from making strong claims regarding the origin of the NO-evoked spikes, their timing can be examined in more detail in Figure 4 - Supplement 1, panels C and D. We will make sure this is clearly stated in the revised text.
Did the authors check for accidental sparse transfection with ChrimsonR of olivary neurons in the post-mortem analysis?
Good point! However, we have never seen this AAV9-based viral construct to drive trans-synaptic expression in the IO, nor is this version of AAV known to have the capacity for transsynaptic expression in general.
No sign of retrograde labeling (via the CF collaterals in the cerebellar nuclei) was seen either. Notably, the hSyn promoter used to drive ChrimsonR expression is extremely ineffective in the IO. Thus, we doubt that such accidental labeling could underlie the excitatory events seen upon N-O stimulation. We will add these mentions with relevant references to the discussion of the revised manuscript.
On page 18 the authors state that: "The lower SS rate was attributed to intrinsic factors of PNs, while the reduced frequency of CSs was speculated to result from increased inhibition of the IO via the nucleo-olivary (N-O) pathway targeting the same microzone." I think I understand what you mean to say, but this is a bit confusing.
Agreed. We will rephrase this sentence to clarify that a lower SS rate in a given microzone may lead to increased activation of inhibitory N-O axons that target the region of IO that sends CF to the same microzone.
Is airpuff stimulation not more likely to activate PO dan DAO because of the related modalities (more face vs. more trunk/limbs?), and thereby also more likely to drive event co-activation (as it is stated in the abstract).
We agree that the specific innervation patterns of different IO regions likely explain the discrepancy between previous reports of airpuff-evoked complex spikes in cerebellar cortical regions targeted by DAO and the absence of airpuff responses in the particular region of DAO accessible via our surgical approach. As in the present dataset virtually no airpuff-evoked events were seen in DAO regions, we are unable to directly compare airpuff-evoked event co-activation between PO and DAO. The higher co-activation for PO was observed for "spontaneous" activity.
The Discussion addresses the question of why N-O pathway activation does not remove the airpuff response.
Given the potentially profound effect, I would propose to expand the discussion on the role of aneasthesia, including longer refractory periods but also potential disruption of normal network interactions (even though individually the stimulations work). Briefly indicating what is known about alpha-chloralose would help interpret the results as well.
We fully agree that the anesthetic state introduces confounding factors that must be considered when interpreting our results. We will expand the discussion to address how anesthesia, particularly alphachloralose as well as tissue cooling, may contribute to prolonged refractory periods and potential disruptions in normal network interactions. However, we recognize that certain aspects cannot be fully resolved without recordings in awake animals. For this reason, we characterize our preparation as an "upgraded" in vitro approach rather than a fully representative in vivo model.
Please clearly indicate that the age range of P35-45 is for the moment of virus injection and specify the age range for the imaging experiment.
Apologies for the oversight. We will indicate these age ranges in the results (as they are currently only specified in Methods). The P35-45 range refers to moment of virus injection.
The methods indicate that a low-pass filter of 1Hz was used. I am sure this helps with smoothing, but does it not remove a lot of potentially interesting information. How would a higher low-pass filter affect the analysis and results?
We acknowledge that applying a 1 Hz low-pass filter inevitably removes high-frequency components, including potential IO oscillations and fine details such as spike "doublets." However, given the temporal resolution constraints of our recording approach, we prioritized capturing robust, interpretable events over attempting to extract finer features that might be obscured by both the indicator kinetics and imaging speed.
While a higher cut-off frequency could, in principle, allow more precise measurement of rise times and peak timings, it would also amplify high-frequency noise, complicating automated event detection and reducing confidence in distinguishing genuine neural signals from artifacts. Given these trade-offs, we opted for a conservative filtering approach to ensure stable event detection. Future work, particularly with faster imaging rates and improved sensors (GCaMP8s) will be used to explore the finer temporal structure of IO activity. We will deliberate on these matters more extensively in the revised discussion.
Reviewer #2 (Public review):
The authors developed a strategy to image inferior olive somata via viral GCaMP6s expression, an implanted GRIN lens, and a one-photon head-mounted microscope, providing the first in vivo somatic recordings from these neurons. The main new findings relate to the activation of the nucleoolivary pathway, specifically that: this manipulation does not produce a spiking rebound in the IO; it exerts a larger effect on spontaneous IO spiking than stimulus (airpuff)-evoked spiking. In addition, several findings previously demonstrated in vivo in Purkinje cell complex spikes or inferior olivary axons are confirmed here in olivary somata: differences in event sizes from single cells versus co-activated cells; reduced coactivation when activating the NO pathway; more coactivation within a single zebrin compartment.
The study presents some interesting findings, and for the most part, the analyses are appropriate. My two principal critiques are that the study does not acknowledge major technical limitations and their impact on the claims; and the study does not accurately represent prior work with respect to the current findings.
We thank the reviewer for recognising the value of the findings in our "reduced" in vivo preparation, and apologize for omissions in the work that led to critique. We will elaborate on these matters below and prepare a revised manuscript.
The authors use GCaMP6s, which has a tau1/2 of >1 s for a normal spike, and probably closer to 2 s (10.1038/nature12354) for the unique and long type of olivary spikes that give rise to axonal bursts (10.1016/j.neuron.2009.03.023). Indeed, the authors demonstrate as much (Fig. 2B1). This affects at least several claims:
a. The authors report spontaneous spike rates of 0.1 Hz. They attribute this to anesthesia, yet other studies under anesthesia recording Purkinje complex spikes via either imaging or electrophysiology report spike rates as high as 1.5 Hz (10.1523/JNEUROSCI.2525-10.2011). This discrepancy is not acknowledged and a plausible explanation is not given. Citations are not provided that demonstrate such low anesthetized spike rates, nor are citations provided for the claim that spike rates drop increasingly with increasing levels of anesthesia when compared to awake resting conditions.
We fully acknowledge that anesthesia is a major confounding factor in our study. Given the unusually invasive nature of our surgical preparation, we prioritized deep anesthesia to ensure the animals’ welfare. This, along with potential cooling effects from tissue removal and GRIN lens contact, likely contributed to the observed suppression of IO activity.
We recognize that reported complex spike rates under anesthesia vary considerably across studies, and we will expand our discussion to provide a more comprehensive comparison with prior literature. Notably, different anesthetic protocols, levels of anesthesia, and recording methodologies can lead to widely different estimates of firing rates. While we cannot resolve this issue without recordings in awake animals, we will clarify that our observed rates likely reflect both the effects of anesthesia and specific methodological constraints. We will also incorporate additional references to studies examining cerebellar activity under different anesthetic conditions.
More likely, this discrepancy reflects spikes that are missed due to a combination of the indicator kinetics and low imaging sensitivity (see (2)), neither of which are presented as possible plausible alternative explanations.
We acknowledge that the combination of slow indicator kinetics and limited optical power in our miniature microscope setup constrains the temporal resolution of our recordings. However, we are confident that we can reliably detect events occurring at intervals of 1 second or longer. This confidence is based on data from another preparation using the same viral vector and optical system, where we observed spike rates an order of magnitude higher.
That said, we do not make claims regarding the presence or absence of somatic events occurring at very short intervals (e.g., 100-ms "doublets," as described by Titley et al., 2019), as these would likely fall below our temporal resolution. We will clarify this limitation in the revised manuscript to ensure that the constraints of our approach are fully acknowledged.
While GCaMP6s is not as sensitive as more recent variants (Zhang et al., 2023, PMID 36922596), our previous work (Dorgans et al., 2022) demonstrated that its dynamic range and sensitivity are sufficient to detect both spikes and subthreshold activity in vitro. Although the experimental conditions differ in the current miniscope experiments, we took measures to optimize signal quality, including excluding recordings with a low signal-to-noise ratio (see Methods). This need for high signal fidelity also informed our decision to limit the sampling rate to 20 fps. In future work, we plan to adopt newer GCaMP variants that were not available at the start of this project, which should further improve sensitivity and temporal resolution.
Many claims are made throughout about co-activation ("clustering"), but with the GCaMP6s rise time to peak (0.5 s), there is little technical possibility to resolve co-activation. This limitation is not acknowledged as a caveat and the implications for the claims are not engaged with in the text.
As noted in the manuscript (L492-), "interpreting fluorescence signals relative to underlying voltage changes is challenging, particularly in IO neurons with unusual calcium dynamics." We acknowledge that the slow rise time of GCaMP6s ( 0.5 s) limits our ability to precisely resolve the timing of co-activation at very short intervals. However, given the relatively slow timescales of IO event clustering and the inherent synchrony in olivary network dynamics, we believe that the observed co-activation patterns remain meaningful, even if finer temporal details cannot be fully resolved.
To ensure clarity, we will expand this section to explicitly acknowledge the temporal resolution limitations of our approach and discuss their implications for interpreting co-activation. While the precise timing of individual spikes within a cluster may not be resolvable, the observed increase in event magnitude with coarse co-activation suggests that clustering effects remain functionally relevant even when exact spike synchrony is not detectable at millisecond resolution.
This finding is consistent with the idea that co-activation enhances calcium influx, leading to larger amplitude events — a relationship that does not require perfect temporal resolution to be observed. The fact that this effect persists across a broad range of clustering windows (as shown in Figure 2 Supplement 2) further supports its robustness. While we cannot make strong claims about precise spike timing within these clusters nor about the mechanism underlying enhanced calcium signal, our results demonstrate that co-activation may influence IO activity in a quantifiable way. We will clarify these points in the revised manuscript to ensure that our findings are appropriately framed given the temporal constraints of our imaging approach.
The study reports an ultralong "refractory period" (L422-etc) in the IO, but this again must be tempered by the possibility that spikes are simply being missed due to very slow indicator kinetics and limited sensitivity. Indeed, the headline numeric estimate of 1.5 s (L445) is suspiciously close to the underlying indicator kinetic limitation of 1-2 s.
Our findings suggest a potential refractory period limiting the frequency of events in the inferior olive under our recording conditions. This interpretation is supported by the observed inter-event interval distribution, the inability of N-O stimulation to suppress airpuff-evoked events, and lower bounds reported in earlier literature on complex spike intervals recorded in awake animals under various behavioral contexts. Taking into account the likely cooling of tissue, a refractory period of 1.5s is not unreasonable. Of course, we recognize that the slow decay kinetics of GCaMP6s may cause overlapping fluorescence signals, potentially obscuring closely spaced events. This is in line with data presented in the Chen et al 2013 manuscript describing GCaMp6s (PMID: 36922596; Figure 3b showing events detected with intervals less than 500 ms).
The consideration of refractoriness only arose late in the project while we were investigating the explanations for lack of inhibition of airpuff-evoked spikes. Future experiments, particularly in awake animals, will be instrumental in validating this interpretation. To ensure that the refractory period is understood as one possible mechanism rather than a definitive explanation, we will rephrase the discussion to clarify that while our data are compatible with a refractory period, they do not establish it conclusively.
The study uses endoscopic one-photon miniaturized microscope imaging. Realistically, this is expected to permit an axial point spread function (z-PSF) on the order of 40um, which must substantially reduce resolution and sensitivity. This means that if there *is* local coactivation, the data in this study will very likely have individual ROIs that integrate signals from multiple neighboring cells. The study reports relationships between event magnitude and clustering, etc; but a fluorescence signal that contains photons contributed by multiple neighboring neurons will be larger than a single neuron, regardless of the underlying physiology - the text does not acknowledge this possibility or limitation.
We acknowledge that the use of one-photon endoscopic imaging imposes limitations on axial resolution, potentially leading to signal contributions from neighboring neurons. To mitigate this, we applied CNMFe processing, which allows for the deconvolution of overlapping signals and the differentiation of multiple neuronal sources within shared pixels. However, as the reviewer points out, if two neurons are perfectly overlapping in space, they may be treated as a single unit.
To clarify this limitation, we will expand the discussion to explicitly acknowledge the impact of one-photon imaging on signal separation and to emphasize that, while CNMFe helps resolve some overlaps, perfect separation is not always possible. As already noted in the manuscript (L495-), "the absence of optical sectioning in the whole-field imaging method can lead to confounding artifacts in densely labeled structures such as the IO’s tortuous neuropil." We will further elaborate on how this factor was considered in our analysis and interpretation.
Second, the text makes several claims for the first multicellular in vivo olivary recordings. (L11; L324, etc).
I am aware of at least two studies that have recorded populations of single olivary axons using two-photon Ca2+ imaging up to 6 years ago (10.1016/j.neuron.2019.03.010; 10.7554/eLife.61593). This technique is not acknowledged or discussed, and one of these studies is not cited. No argument is presented for why axonal imaging should not "count" as multicellular in vivo olivary recording: axonal Ca2+ reflects somatic spiking.
We appreciate the reviewer’s point and acknowledge the important prior work using two-photon imaging to record olivary axonal activity in the cerebellar cortex. However, while axonal calcium signals do reflect somatic spiking, these recordings inherently lack information about the local network interactions within the inferior olive itself.
A key motivation for our study was to observe neuronal activity within the IO at the level of its gap-junctioncoupled local circuits, rather than at the level of its divergent axonal outputs. The fan-like spread of climbing fibers across rostrocaudal microzones in the cerebellar cortex makes them relatively easy to record in vivo, but it also means that individual imaging fields contain axons from neurons that may be distributed across different IO microdomains. As a result, while previous work has provided valuable insight into olivary output patterns, it has not allowed for the examination of coordinated somatic activity within localized IO neuron clusters.
With apologies, we recognize that this distinction was not sufficiently emphasized in our introduction. We will clarify this key point and ensure that the important climbing fiber imaging studies are properly cited and contextualized in the revised manuscript.
Reviewer #2 (Recommendations for the authors):
The authors state: "we found no reports that examined coactivation levels between Z+ and Z- microzones in cerebellar complex spike recordings" (L359). Multiple papers (that are not cited) using AldolaceC-tdTomato mice with two photon Purkinje dendritic calcium imaging showed synchronization (at similar levels) within but not across z+/z- bands. (2015 10.1523/JNEUROSCI.2170-14.2015, 2023 https://doi.org/10.7554/eLife.86340).
We apologize for the misleading phrasing. We will rephrase this statement to: "While complex spike coactivation within individual zebrin zones has been extensively studied (references), we found no reports directly comparing the levels of intra-zone co-activation between Z+ and Z microzones."
Additionally, we will ensure that the relevant studies demonstrating synchronization within zebrin zones, as well as (lack of) interactions between neighboring zones, are properly cited and discussed in the revised manuscript.
The figures could use more proofreading, and several decisions should be reconsidered:
Normalizing the amplitude to maximum is not a good strategy, as it can overemphasize noise or extremely small-magnitude signals, and should instead follow standard convention and present in fixed units (3A2, 4B2, and even 2C).
As noted earlier, we have excluded recordings and cells with high noise or a low signal-to-noise ratio for event amplitudes, ensuring that such data do not influence the color-coded panels. Importantly, all quantitative analyses and traces presented in the manuscript are normalized to baseline noise level, not to maximal amplitude, ensuring that noise or low-magnitude signals do not skew the analysis.
The decision to use max-amplitude normalization in color-coded panels was made specifically to aid visualization of temporal structure across recordings. This approach allows for clearer comparisons without the distraction of inter-cell variability in absolute signal strength. However, we recognize the potential for confusion and will revise the Results text to explicitly clarify that the color-coded visualizations use a different scaling method than the quantitative analyses.
x axes with no units: Figures 2B2, 2E1, 3B2, 3C2, 5B2, 5C2, 5D2.
No colorbar units: 5A3 (and should be shown in real not normalized units).
No y axis units: 5D1.
No x axis label or units: 5E1.
5E3 says "stim/baseline" for the y-axis units and then the first-panel title says "absolute frequencies" meaning it’s *not* normalized and needs a separate (accurate) y-axis with units.
Illegibly tiny fonts: 2E1, 3E1, etc.
We will correct all these in the revised manuscript. Thank you for careful reading.
