Toothy: an interactive platform for dentate spike curation

Reviewer #1 (Public review):

Summary:

Esfahany et al. describe a new platform (Toothy) to identify and analyze dentate spikes and sharp wave ripples from silicon probe electrophysiology data. The goal is to facilitate and standardize the extraction of DS1 and DS2 events, which have highly variable properties across recordings from different labs. The manuscript describes the basic workflow of the Toothy pipeline, including loading data, assigning channels along a linear probe, customizing parameters, selecting ideal channels for analysis, and classifying DS1 and DS2 events.

Strengths:

The manuscript is clear and easy to follow and does a good job of describing the platform. Overall, this will be a useful analysis pipeline that can help to standardize DS analysis across labs and datasets.

Weaknesses:

The current version has several bugs that prevent analysis, and the documentation of analysis parameters needs to be improved.

(1) In limited testing, the pipeline had several bugs, and I was not able to complete the full analysis of a dataset. Loading data from .mat or .npy files gave errors (it seemed that the metadata was not loaded correctly from the pop-up window). I was able to load a .nwb file, which worked well. The probe configuration tool was a bit difficult to understand, and there was not much documentation to help, although it worked when simply entering the x-y coordinates of the channels. It also crashed several times while trying to make a probe configuration due to it trying to save when a small typo was briefly entered. The initial analysis worked well, and the auto-selected channels matched our recording notes and seemed appropriate. DSs and ripples were extracted. An error came when trying to classify DSs, and the program repeatedly crashed across a variety of parameters. Overall, parts of the pipeline worked well, but others had significant bugs that need to be addressed.

(2) The authors should provide test data that can be run through the pipeline. Ideally, this could use a variety of data types, probes, and conditions so that it is clear how they differ.

(3) There are a lot of parameters that can be adjusted, but very little information about how they are chosen and what goes into parameter selection for a dataset. Additional documentation with more information on adjustable parameters, channel selection, and best practices would help improve the utility of the tool. Ideally, this could also integrate citations (either in the manuscript or documentation) to support some of the choices made during parameter selection.

(4) There is no validation presented against other analysis methods or datasets. While there is no ground truth of when DSs occur, this may limit the ability of this tool to become the standard for DS analysis. A section comparing the analysis used in the pipeline to other published analyses would be helpful.

(5) In the manuscript, it would be helpful to further describe the rationale for initially detecting DSs and SPW-Rs on all channels, when they are network events that occur across channels.

(6) A section on what hardware and software are necessary to run the pipeline should be added.

Reviewer #2 (Public review):

Summary:

This work provides an open-source, Python-based, graphical user interface for curating the detection and classification of dentate spikes (DSs) from hippocampal local field potential (LFP) recordings. The tool may also be used to detect, but not classify, sharp wave-ripples (SPW-Rs). The tool utilizes previously published Python packages for loading LFP files and creating experiment-specific probe objects. Detection and classification parameters are clearly defined and logged in a parameter file before starting processing. Once LFP data has been mapped to the probe object, event detection occurs across all channels. DSs are detected as qualifying peaks in the filtered DS band LFP, while SPW-Rs are detected as qualifying peaks in the filtered ripple band amplitude envelope. An initial curation step allows visualization of the LFP, instantaneous current source density (CSD), and depth-by-frequency band power plots for determining the approximate channel locations of key anatomical regions (i.e., CA1, the hippocampal fissure, and the hilus of the dentate gyrus). The optimal channel for detection is further refined in the next step by comparing event waveforms and quality metrics across channels. Artifacts and noisy waveforms can also be manually excluded during this step. Finally, DSs detected from the optimal channel are classified by computing the CSD profile around events and then clustering the first two principal components of all CSDs. The authors claim that this customizable tool will standardize DS detection and classification.

Strengths:

Toothy's detection and classification algorithms are appropriate and well-validated in the literature. The ability to change many parameters, the CSD calculation method, and clustering algorithm is helpful for precise replication of methodology that has varied previously. Default parameters optimized for mouse recordings provide a standardized starting point for rodent researchers.

The authors' commitments to transparency and user-friendliness are to be commended (e.g., clear instructions, defined and logged parameters, multiple visualization options, etc.) and are likely to be appreciated by new users. Researchers with little-to-no coding experience should find this tool especially powerful for jumpstarting their own DS analyses.

While not the focus of the paper, the capability to detect SPW-Rs provides an additional use case for Toothy and streamlines simultaneous analysis of SPW-Rs and DSs.

Weaknesses:

I encountered unexpected errors while trying to load LFP data into Toothy for testing, indicating that the "data ingestion" stage of Toothy requires minor code revision.

Toothy's utility for recordings that do not produce an LFP depth profile is unclear. According to the authors, Toothy allows probe designs with irregular spatial sampling (e.g., tetrodes) to be used. However, recording from a linear probe with electrodes spanning from approximately the hippocampal fissure to the hilus of the dentate gyrus is required for Toothy's full functionality. For example, Toothy uses a DS type classification algorithm that relies on sufficiently sampled CSD depth profiles that tetrode recordings cannot provide. As such, usage is currently restricted to detection only for certain recording setups.

The documentation on Toothy's output could be improved. Specifically, the work does not state which files different data are saved to or list the properties saved per detected event. Furthermore, the work does not discuss the potential importance of DS properties that are saved besides those related to the timing of the DS and its type.

Reviewer #3 (Public review):

Summary:

Esfahany et al present a novel, UI-based tool to detect dentate spikes from hippocampal local field potential recordings, called Toothy. Toothy is easily accessible, compatible with many popular recording formats, and guides users entirely via UI through the dentate spike curation and analysis process. The functional and interactive visualizations enable users to gain a detailed understanding of their data and rigorously analyze dentate spike phenomena. This tool will be broadly useful for anyone who studies hippocampal electrophysiology. Furthermore, by expanding access to dentate spike analysis, it may encourage more scientists to explore this understudied but critical phenomenon.

Strengths:

(1) Toothy provides several ways for users to interact directly with parameters, revealing the ramifications of these choices. Most parameters are adjustable and made obvious via a UI panel. Their effects are then visualized across channels and individual events. This will help users think critically when selecting parameters.

(2) Toothy is fully UI-based and pip-installable, lowering the barrier to entry far below what most electrophysiology analysis tools offer.

(3) The channel selection tool is broadly useful for identifying DG hilus and CA1 pyramidal locations. Since subregional and laminar localization of electrode sites is critical to correctly interpret hippocampal recordings, this tool could be more generally used to identify site locations across the hippocampus.

Weaknesses:

(1) The rationale behind parameter choices is not explained. In order to function "not as a black-box detector", as the authors state, all initial parameter choices should be explained with citations. If possible, these citations would also be available from Toothy directly, alongside citations describing alternative parameter choices. This will help users make informed choices. For instance, a user analyzing data from rats would need to adjust the default ripple frequency band upwards (150-250Hz), and would benefit from guidance to adjust this properly.

(2) The Results describe the functions of Toothy from the perspective of the user, but there is no Methods section describing what Toothy does between UI displays. This would allow readers to compare the tool directly to analysis pipelines as described in the Methods sections from other papers. Particular attention should be paid to justifying the analysis decisions that cannot be changed by the user, such as detecting events off of a single representative channel instead of across a consensus of multiple channels.

(3) It's unclear whether or how Toothy evaluates data quality to confirm that its analyses return interpretable results. At a minimum, the tool should confirm adequate sampling rate (e.g. <=1kHz) and inter-site spacing for CSD (e.g. <=50um).

(4) The paper does not put Toothy into context among the other common open-source electrophysiology analysis toolboxes. Consider Rippl-AI (Navas-Olive & Rubio et al, 2024) or pynapple (Viejo et al, 2023), to give a few examples. The paper would be strengthened by addressing how Toothy extends beyond the capacities of these other tools and how Toothy can be integrated into a workflow that also uses these other tools.

Author response:

We are very happy that our work was positively received by the reviewers and editors and we are looking forward sharing our results via eLife.

In addition, we have added more details on the generation of the brainbow lines, and we have submitted the respective plasmids to Addgene and give the respective IDs. Some additional minor changes were done to make the text more clear.

We have to confess that we were not completely happy with the term “solid” for the strength of evidence. That assessment seems to be based on the fact that we reached no single cell resolution with our brainbow imaging. Admittedly, this is true and we have openly discussed this technical restriction. However, to our knowledge, within arthropods transgenic single-neuron marking had been restricted to Drosophila melanogaster, i.e. the most advanced genetic model system. Our work presents the first generation of respective transgenic lines and the first application of the brainbow system in any arthropod apart from Drosophila. Therefore, we tend to consider our methodological approach to go beyond current state of the art. We would also like to point out an additional aspect of our technical approach: We transgenically marked a subset of neural cells expressing a given transcription factor. This allowed us relating projection patterns and the expression of classic neural markers (e.g. neuropeptides) to the expression of a transcription factor (i.e. a molecular subset of neurons). We are not aware of other arthropod studies doing this (apart from flies, of course).

In summary, we wonder whether the chosen level of strength of evidence might be reconsidered from the perspective of emerging model organisms and in comparison with other recent studies published in eLife, where the well-known classic neural markers were used to determine spider brain morphology.

Public Reviews:

Reviewer #1 (Public review):

Summary:

Pang et al. investigated the expression pattern of the transcription factor foxQ2II in an adult beetle brain. They find nine distinct clusters, with many neurons expressing Glut/ChaT and dopamine. Some of the dopamine neurons resemble cell types described in Drosophila. Several neurons seem to project to prominent higher brain regions such as the MB and CX, and might even connect to both.

Strengths:

The authors use state-of-the-art labeling techniques for the analysis of individual cell types, such as beetle brainbow, to investigate the until now unknown expression of the transcription factor in the adult beetle brain.

We would add that this work establishes and introduces brainbow for the first time in an arthropod outside Drosophila melanogaster and that we are the first (outside flies) to relate the expression of a neural transcription factor with neural projection and neurotransmitter content.

Rigorous cell reconstruction and image analysis revealed a better understanding of the anatomy of the labeled cells.

Weaknesses:

The brainbow labeling seems to include all cells labeled by the enhancer trap line, as well as the ones not expressing foxQ2II. Thus, it is unclear how useful this data is to compare individual cells to other insects.

We kindly disagree with the first statement: not all cells of the enhancer trap are labelled but a subset. Therefore, we call it “sparse labelling” in our manuscript while admittedly we do not reach “single cell labelling”, which admittedly limits both precision and use.

The functional relevance of this transcription factor in the adult brain cell is still unknown. It is therefore unclear if the described neurons have any specific function and if they require this transcription factor for normal function.

Previously, we found that this gene has an important function in neural development during embryogenesis. Actually, we have done extensive RNAi experiments to test for an effect during postembryonic development. We found surprisingly small defects when looking at the projection patterns of several imaging lines. However, we found some changes in behaviour. Given the extensive data presented in the current paper, we decided to publish these functional data (another 12 figures/suppl. figures) separately.

We also note that the identity/function of neurons is determined by a mix of transcription factors. Disentangling the individual role of each of those transcription factors would indeed be an exciting question and a major endeavour.

Overall, the neural reconstructions are missing single-neuron details; it is difficult to compare the shown cell types to specific cell types in Drosophila based on the presented data, and this finding remains speculative.

Indeed, we do not reach single-cell resolution, which is below the standards of fly neurobiology. However, compared with other arthropods we reach a unique level of precision. Specifically, we are able to relate the expression of a transcription factor to neural projection and neurotransmitter content.

We also think that combining our transgenic line with dopamine expression was sufficient to compare the labelled cells to fly neurons. We feel that most homology assessments of neurons across such large evolutionary distances will remain hypothetical to some degree.

Reviewer #2 (Public review):

Summary:

The authors provide the first thorough profiling of neurons in Tribolium characterized by the expression of the transcription factor foxQ2, which will be useful for developmental neurobiology. They use state-of-the-art methods convincingly to not only identify the neurons, but also to further characterize them anatomically and neurochemically.

Strengths:

Thorough and meticulous application of state-of-the-art anatomical methods in a nonstandard laboratory organism.

Thank you for this encouraging comment.

Weaknesses:

No weaknesses were identified by this reviewer.

Comments:

I don't really have any major suggestions at all. Loved the work.

There is only one tiny nitpicking aspect:

P21: "Biogenic amines are involved in learning and memory and setting arousal threshholds (Davis, 2023), which are functions performed by the mushroom bodies and related to the function of the central complex in goal directed navigation, respectively."

MBs mainly process olfactory memory. At least in Drosophila, most other kinds of memories are being supported elsewhere.

https://pubmed.ncbi.nlm.nih.gov/10454381/

such as, e.g., visual pattern learning in the CX

https://pubmed.ncbi.nlm.nih.gov/16452971/

or motor learning in motor neurons

https://pubmed.ncbi.nlm.nih.gov/38779314/

or ventral ganglion, antennal lobes, and median bundle for place learning:

https://pubmed.ncbi.nlm.nih.gov/10706599/

If the authors focus on MBs, this sentence ought to reflect the fact that the function of the MBs is much narrower than the current sentence appears to suggest.

Thanks for this clarification – we have rephrased:

"Biogenic amines are involved in learning and memory and setting arousal thresholds (Davis, 2023). This relates to the mushroom bodies’ function in olfactory memory, and the function of the central complex in visual pattern learning and goal-directed navigation, respectively."