Deep anatomical and ultrastructural classification of neurons in the zebrafish olfactory bulb

Reviewer #1 (Public review):

Summary

The authors set out to address a critical gap in olfactory bulb research: the lack of a comprehensive annotation of neuron types and their connectivity in the adult zebrafish. Using a large serial block-face scanning electron microscopy volume, they reconstructed 459 neurons and performed a detailed morphological and ultrastructural classification. They identified 13 major neuron subclasses, including two projection neuron types (mitral cells and ruffed cells) and 11 interneuron types. A key finding is that despite the diffuse appearance of the lateral olfactory bulb neuropil, it is organized into discrete, segregated glomeruli. The authors also performed targeted synapse annotation, revealing systematic and selective connectivity patterns between projection neurons and specific interneuron subnetworks, and discovered distinct microcircuit motifs involving reciprocal and unidirectional connectivity. This work provides a crucial anatomical framework for understanding the circuit basis of olfactory computations.

Strengths

(1) The volumetric EM reconstructions capture fine ultrastructural features (e.g., spine shapes, neurite caliber variations, and varicosity morphology) at nanometer resolution across hundreds of neurons. The descriptive detail and 3D rendering provided for each subclass are remarkably thorough.

(2) The study goes beyond simple morphological descriptions by integrating ultrastructural features, such as spine shape, neurite diameter, and synaptic arrangements, to define neuron classes. This provides a more functionally relevant taxonomy. The classification was also cross-validated by independent neuroscientists.

(3) The targeted synapse annotation translates anatomy into testable circuit hypotheses, revealing distinct connectivity motifs (reciprocal vs. unidirectional, MC-selective vs. RC-selective) that provide structural substrates for computational functions such as normalization.

Weaknesses

(1) It is important to provide an estimate of the total neuronal population of the adult zebrafish OB and justify that their sample size is sufficient to claim a "comprehensive" neuron type classification.

(2) The classification was based on morphology and ultrastructure. If feasible, we recommend an unsupervised clustering analysis using only the synaptic connectivity matrix for fully reconstructed neurons to test whether connectivity patterns recapitulate the 13 morphological subclasses. Besides, cross-validation by independent annotators is great. However, when the annotators disagreed on a neuron's subclass, how was it resolved?

(3) A summarized reference table that includes features for each subclass is highly recommended (the main text cited several tables; however, I did not find them). This would facilitate community adoption of the classification.

(4) The schematic in Figure 20 could be optimized to reflect synaptic connections among 13 subclasses with connection strength annotated.

Reviewer #2 (Public review):

Summary:

The authors have used a previously published SBEM dataset of the adult zebrafish olfactory bulb and carefully reconstructed the fine structure of a large number (459) of neurons. Through manual classification of cells into 13 different anatomical classes, they provide a carefully annotated library of cells. Classifications have been validated by comparing the labeling of 5 human annotators, which revealed a large degree of reproducibility. Synaptic identification enables assignment of connectivity, revealing a major contribution of reciprocal connections (two-way synapses) that may support key computational features during olfactory processing. In general, the work provides a clean descriptive account of the organization of the zebrafish olfactory bulb, which should help to better understand principles of olfactory processing in vertebrates.

Strengths:

The authors provide a library of neurons in the olfactory bulb of adult zebrafish that will be highly useful to the broader circuit neuroscience community. Figures of classified cell types have been convincingly arranged to appreciate the high quality of the EM volume and reconstructions. Cells are easily accessible through a web interface to facilitate the assessment of classifications. An important result is that anatomical structures in zebrafish clearly follow the glomerular arrangement known in mammals, which argues against a microglomerulus model previously proposed for the zebrafish olfactory bulb.

Weaknesses:

The paper uses Wanner et al. 2016 as a main base dataset to generate a descriptive library of cells, in a limited volume (18%) of the zebrafish olfactory bulb. Annotations of cell types and synapses are largely human-based. Using a multi-human validation strategy helps to assess labeling, but it would benefit from more rigorous statistics, in particular for the synapse annotations. In general, the paper would be easier to read with a glossary of cell types. Some features are quite intriguing, like the cilia structures in some of the somata or the ruffs, but discussion of the potential computational roles of these fine structures is missing.

Reviewer #3 (Public review):

Summary:

The authors aimed to build a comprehensive understanding of the cell classes that make up the adult zebrafish olfactory bulb (OB) and their connectivity. Using an EM volume covering the three OB layers and several glomeruli, they aim to reconstruct a sufficient sample of all cell classes in the OB. Based on 459 proofread reconstructions, they identify 13 cell classes and a small number of subclasses and offer detailed expert neuroanatomical arguments, largely qualitative, for why these classes are distinct. While they do not have a synapse classification, they further build a manual wiring diagram of connectivity based on these reconstructions, which together offer a strong baseline understanding of the organization of the OB and useful properties to build into a future quantitative assessment of fish neuroanatomy in general.

Strengths:

The paper offers a powerful example of why expert assessment remains a vital tool in neuroanatomy. While quantification is deeply important to scale data, the kind of careful multimodal human investigation of neuronal shape, connectivity, and ultrastructure forms a key basis for discovering and differentiating neuronal cell types. The authors use multi-expert classification to validate qualitative assessments as well, which is a good way to handle cross-individual uncertainty. The descriptions of each type and their relationships are generally quite thorough, and there are enough cells to offer a satisfying baseline categorization. Anyone studying zebrafish OB is likely to get a lot out of this paper.

Weaknesses:

While I am a strong believer in expert neuroanatomical intuition alongside quantitative neuroanatomical approaches to cell typing (especially given the manageable number of cells and cell types), I would have liked a section explaining why this approach was taken as opposed to a more quantitative one from the beginning. I think that this is a case study in where the expert qualitative approaches are useful, especially with fascinating details like the unusual hand-shaped spines. A lot of the strength of the paper comes from this philosophy, in my view, and it would be nice to hear it elucidated by the authors. I would also be curious if the authors came to any conclusions about what might have been possible with a more data-driven approach, although this may be beyond the scope of the work or not terribly interesting.

There were some aspects of the cell typing that I didn't totally understand. Two of the most common issues with cell typing are that (1) it is unclear if cell types are discrete and (2) unusual cells appear that don't have enough peers to classify into an unambiguous type. Did those happen here? For example, it was not obvious to me how it was decided that mitral cells fell into three subclasses: small, medium, and large, and not a continuum from small to large. Similarly, some interneuron classes had subclasses. What anatomical qualities suggested that a split was a subclass-level split vs a class-level split? Were there any cells that were impossible to classify into a group, or was everything actually tidy? How confident are the authors that they have captured the complete diversity of the OB from this sampling?

I generally appreciated the numerous detailed figures and found them generally clear and informative. However, as someone who is not deeply familiar with zebrafish olfactory bulb organization, I would have liked to have at least some global context for each cell class in terms of how typical cells relate to the layers and glomeruli. Figure 18B does a nice job of this with many cells, but it would be clearer to see this along the way with the individual classes (and subclasses) as well.

For the connectivity analysis, I commend the authors for doing as much manual synapse annotation as they did, but the lack of completeness makes it less clear how to interpret the connectivity findings. In particular, how complete is the current assessment? For example, when describing connectivity between pairs of cells, how were the pairs selected? Did they necessarily have contact between the meshes, or innervate the same glomerulus, or were they just two cells of the same type in any location? This is useful to understand how to interpret the connectivity fractions measured.

Author response:

We thank the reviewers for their constructive feedback.

Reviewer 1:

(1) We will provide an estimate of total neuron numbers.

(2) We will explore the possibility to extend cell type classification by unsupervised clustering of connectivity but it may not be possible to obtained a meaningful clustering based on the available connectivity data. The main issue is that connectivity information is available only for subsets of neurons. The underlying reason is that synapses were annotated manually and a complete manual annotation of the entire dataset would be an excessive task. In principle, this problem could be solved by automation but networks for automated annotation of synapses in other brain areas do not produce reliable results in the olfactory bulb, most likely because many neurites are both pre-and postsynaptic (further remarks below).

(3) We are happy to supply the missing tables and to provide a glossary/table summarizing the features of the different neuron types.

(4) Thank you; we will consider the suggestion to optimize Fig. 20.

Reviewer 2:

We agree that more a more comprehensive analysis of synaptic connectivity would be desired. The underlying problem, however, is that scaling synapse annotation requires automation but networks used for synapse annotation in other brain areas failed to produce reliable results in the olfactory bulb. The underlying reason is most likely that neurites of many olfactory bulb neurons are both pre- and postsynaptic; their ultrastructure is therefore different from classical axons or dendrites, e.g., vesicles are frequently found in close proximity on both sides. We therefore annotated synapses manually, which is the reason why synapse annotation was restricted to subsets of neurons. We will explain this situation in the revisions. Moreover, we will explore further approaches for automated cell type classification using both morphological and synaptic features.

We shall be happy to include a glossary of cell types.

We are also happy to expand the discussion of ultrastructural observations. Concerning primary cilia, we do not have a strong hypothesis concerning their function, neither based on our observations nor based on the literature.

Reviewer 3:

Following the reviewer’s suggestion we will be happy to include a more in-depth discussion of why our analysis is based primarily on human annotations and observations rather than more automated classification.

As suggested by the reviewer we are planning to include additional figures and/or tables to show the relations between neuron types and layers, and to provide more summary information about cell types.

We will also explain in more detail how neurons were chosen for connectivity analysis, with a specific focus on the assessment of connectivity fractions.