Review of Dake Huang et al.

Summary
In this study, the authors characterized the axonal terminations originating from chemically identified neuronal subpopulations of mouse parabrachial neurons using a virally driven tracer that labels the presynaptic marker, synaptophysin. The authors present the results of their studies for three different strains of mice and use a machine learning algorithm to analyze the synaptic boutons they immunolabeled for each strain, and to illustrate these in a series of drawings through the mouse brain. The authors provide a large and impressive dataset in this study, beautiful images, and careful documentation of the bouton distributions at a gross neuroanatomical, if not sub-regional, level. The manuscript is generally well written and the data elegantly presented. The strengths of the study lie in the data presentation and in the meticulous attention to histology paid by the authors. There are some concerns about consistency in nomenclature, descriptions of how Nissl-series were used, the rationale for the use of a specific transgenic line, and some aspects of the machine learning and bouton analyses.

Overall, however, the work presented lays substantial groundwork for future experimentation targeting the glutamatergic output projections from the parabrachial nucleus and the subsets of genetic profiles in this neuron population. Descriptions of the efferent projection patterns, both previously described and novel, provide interesting functional perspectives of the parabrachial nucleus and the cell types targeted that will be valuable in forthcoming investigation. The end-to-end two-step algorithm provides a sophisticated way of accelerating data quantification and can easily be implemented in labs across the discipline. Moreover, its ability to output multiple formats (i.e., png, svg, csv) allows for a wide array of analyses that is well suited to accommodate the needs of individual labs.

Major concerns
1. The authors may be inadvertently sowing confusion by utilizing mouse neuroanatomical nomenclature that derives from multiple, mutually conflicting sources. For example, throughout the study, the authors explicitly separate the preoptic area from the hypothalamus, and the hypothalamus from the diencephalon. This is a convention adopted by Franklin and Paxinos in keeping with their alignment with Luis Puelles’s “prosomeric”-based nomenclature for the mouse. (See, for example, p. 225 of Chapter 8 in The Mouse Nervous System, published in 2012 by Watson, Paxinos, and Puelles). However, in other locations, the authors cite Dong, 2008, which is the initial edition of the Allen Reference Atlas of the mouse brain. This latter work squarely places the preoptic area within the hypothalamus and that structure, in turn, within the diencephalon. It is suggested that the authors should decide which designated nomenclature system they are using, and to cite and use it explicitly and consistently throughout their work. Also, the sources of all abbreviations in their list should be consistent with that atlas source and the authors should state this somewhere in the text.
2. In their Methods describing the plotting of boutons for illustration, the authors stated: “In separate layers, we aligned a Nissl-counterstained image of each section to trace the brain borders, major white matter tracts, and cerebral ventricles to generate the illustrations shown in the figure.” Strictly speaking, identifying the brain borders, white matter tracts, and ventricles are actions that do not necessarily take advantage of the fact that the section was Nissl-stained. In other words, Nissl-stained sections would provide much more information than it seems was utilized by the authors, which is curious. Did the authors attempt to align the cytoarchitectonic boundaries of the neuronal populations (bounded brain regions and subregions) identified by Nissl-staining as they plotted their boutons for illustration? If they did, this should be mentioned. If they did not, how did the authors identify specific brain regions for labeling (e.g., APir, PSTN, etc.).
For example, the authors also stated: “We identified Syp-mCherry-labeled boutons in NiDAB-labeled images (without Nissl counterstaining), then compared these images side-by-side with subsequent cytoarchitectural images of the same sections after Nissl counterstaining.” This gets to the issue, but again, it is not clear what specifically was being compared between the NiDAB-labeled sections and the Nissl-stained sections.
3. The rationale for using a GFP reporter mouse for Vglut2, which would trigger expression in all neurons that currently express or have had previous expression of VGlut2 is not totally clear. Structures such as the locus coeruleus are labeled, and the authors state in their figure legend (but not main text) that LC neurons in the adult mouse are VGlut2 negative. Based on these results, for the parabrachial nucleus, it becomes unclear if VGlut2 is in fact expressed in all parabrachial sub-regional neurons in the adult mouse using this reporter system, or alternatively, that there is an artifact where expression is evident even in the parabrachial neurons that no longer express bona fide VGlut2 in adulthood.
4. Regarding the algorithm-generated bouton plots, delineation of region boundaries using the cytoarchitecture of the Nissl-stained tissue sections would validate the location and spatial distribution of the plotted data. Parcellation of the regions labeled would also be a key step in preparing the generated plots to be mapped to a standardized mouse brain atlas.
5. The significance and rationale as to why the authors measured the long-axis diameters of boutons for two regions (lateral VPpc, parafascicular nucleus) is unclear. What does this analysis reveal apart from morphological differences, and why were only these two regions chosen?
6. Are there negative controls that the authors performed for the RNAScope studies?
7. It would be beneficial to revise the color scheme of the density map in Figure 6 so as to not confuse the reader that the densities are comparable across cases/columns; the density measure of one cannot be directly compared to the density of another if the injections are not identical in size and location.
8. It would be immensely helpful to have higher resolution images of the sections in the Supplementary figure so that readers can independently use the data to map the locations of the injection sites to their own atlas formats, and thereby contextualize the sites with their own datasets. Greater resolution would enable better discernment of Nissl-delimited boundaries in the sections.

Minor concerns
● Title. The title should state that the projections being studied were in the mouse.
● Abstract. The authors state, ”To do this, we used a highly sensitive, Cre-dependent anterograde tracer, Synaptophysin-mCherry, in three different strains of Cre-driver mice.” For clarification it is recommended that the authors state that this is an AAV-mediated vector.
● Page 4. The authors state: “In mice, the efferent projections of PB Foxp2 neurons remain unknown, and we lack a brainwide map of the overall pattern of PB output projections in this species.” Strictly speaking, mice does not refer to a species, but a taxon, as “mice can refer to any of several different species.
● Page 5. “We used n=15 male and female mice” - consider indicating quantity of each sex.
● Page 5. “The pipette was left in place for an additional 3-5 minutes” - should be changed to 3–5 minutes using an en dash in addition to other time spans referenced using a hyphen (3-5 weeks on page 6 and throughout).
● Page 6. “i.p. 150-15 mg/kg” is a rather unclear range. Please correct.
● Page 6. “The sections were washed 3x in PBS” - should be changed to 3✕ indicating multiplicity. Changes needed throughout in reference to washing.
● Page 7. “ddH2O” and “H2O2” should have subscripted numerals. Also, “biotin-avidin” is typically written as “avidin-biotin” (in fact, “ABC Kit” refers to “avidin-biotin complex kit”).
● Page 9. “We used a 20x or 40x objective” - should be changed to ✕20 or ✕40.
● Page 9. “All slides were imaged” - consider replacing “imaged” with “photographed” or “scanned”.
● Page 9. “…Figure 3.” The in-text figure reference should be referencing Figure 4.
● Page 9. “To compare the spatial distribution of injections in the PB region from each genotype, we plotted the core cluster of Syp-mCherry-expressing neurons in every case onto three template levels of the PB (approximate bregma levels -5.0, -5.2, -5.4) (Franklin & Paxinos, 2013; Geerling et al., 2016).” As written, the sentence implies that digital templates of one of the specified atlases was used at those coordinates to produce the drawings and/or that the drawings were on digital atlas templates. However, a careful examination of the figure in which these data are presented reveals that the authors probably meant three levels that closely matched templates in those atlases, and that the drawings were actually made onto traced outlines of fiducials as they appear in the Nissl series. The authors should reword this to clarify their description.
● Page 9. The “Franklin & Paxinos, 2013” reference should read “Paxinos & Franklin, 2013”, since that edition of the atlas has the authors listed in the latter arrangement. The references list should also revise the entry so that the correct order is listed.
● Page 10. “To plot Syp-mCherry labeled boutons for illustrations, we developed a neural network-based Python script.” - consider revising to “We developed a Python script to plot Syp-mCherry-labeled boutons for illustrations” to prevent redundancy with the third sentence of the paragraph which again mentions the neural-network nature of the algorithm.
● Page 10. Consider clarifying the amount or source of negative training examples for the development of the convolutional network and whether it is a balanced training set (173,934 positive and 173,934 negative).
● Page 10. "which outputted a probability of the potential bouton being a true bouton" - consider revising to “the probability of the potential bouton being labeled as true bouton by a human annotator" because the model does not know what a “true” bouton is.
● Page 10. Consider commenting on the performance of the trained model on a test set to justify its use in the analysis pipeline.
● Page 10. “A full resolution TIFF was exported” - consider including the resolution of nanometers per pixel.
● Page 10. “Then searches for pixels that are 3 grayscale values darker” - consider revising to match the github documentation, “detects potential boutons by looking for groups of pixels that are darker than the surrounding tissue.”
● Page 11. Consider briefly mentioning the ability to output confirmed boutons as svg and csv formats as it may be of importance to other labs that, for example, use vector graphics for their analysis.
● Page 35. “data fig. presented” - consider revising to “figure”.

I thank the authors for clarifying their text to address most, if not all, of my concerns. I have no further suggestions for revision. However, I will leave the authors with one final thought regarding their not choosing a particular ontological framework for neuroanatomical regions (my point 1):

As the rock band Rush famously sang in their song, "Freewill": "If you choose not to decide, you still have made a choice".

At any rate, I commend the authors on a fine study.