Toward an integrated classification of neuronal cell types: morphoelectric and transcriptomic characterization of individual GABAergic cortical neurons

Toward an integrated classification of neuronal cell types: morphoelectric and transcriptomic characterization of individual GABAergic cortical neurons Gouwens Sorensen Baftizadeh Budzillo Lee Jarsky Alfiler Arkhipov

Tag layer of the DHSS Meta-Tagging Project. The article text is reproduced from the staged copy named in sourceDesc and is not altered; the tags are held separately in standOff so that overlapping and nested spans are preserved.

Toward an integrated classification of neuronal cell types: morphoelectric and transcriptomic characterization of individual GABAergic cortical neurons 2020 bioRxiv 10.1101/2020.02.03.932244 Gouwens2020-toward-an-integrated-classification-of-neuro Gouwens2020-toward-an-integrated-classification-of-neuro.txt, the project's staged copy 097349e00ad2968c3d6a9ca18dad12f4aad554c4 not-redistributable neuroscience

the paper's own words

the tag layer; a sentence written by the tagger about the paper, always anchored to a quotation but not itself a quotation

0 spans located by searching the stored text for the quote's exact words, allowing any whitespace between them existence Quotation withheld because this record is not cleared for redistribution. cortical GABAergic interneuron cell types Quotation withheld because this record is not cleared for redistribution. electrophysiology electrophysiology Quotation withheld because this record is not cleared for redistribution. mechanisms the noisy modality is not clustered on its own but MAPPED onto a reference taxonomy built from the clean one, and the stated purpose is to minimise the contribution of contamination from non-neuronal cells and nearby excitatory neurons Quotation withheld because this record is not cleared for redistribution. mechanisms the mapping is bootstrapped rather than computed once - a hundred iterations, each using only seventy per cent of the available genes and of the reference cells - so that a cell's assigned type carries a confidence instead of being a bare label Quotation withheld because this record is not cleared for redistribution. mechanisms AMBIGUITY THAT BELONGS TO THE TAXONOMY IS SEPARATED FROM AMBIGUITY THAT BELONGS TO THE MEASUREMENT: a reference matrix records how often each type is confused with each other type in the clean data, and a cell's observed confusion is scored against that expectation rather than against zero Quotation withheld because this record is not cleared for redistribution. mechanisms two numerical cut-offs then decide which cells are set aside - a divergence above two, and, among the survivors, a correlation below one half between the cell's expression and the average expression of the types Quotation withheld because this record is not cleared for redistribution. mechanisms the per-type analyses are then run only on the subset with the highest-confidence mapping, and the stated reason is to limit the effect of mapping ambiguities due to technical issues Quotation withheld because this record is not cleared for redistribution. mechanisms the mapping is checked four separate ways rather than asserted: against a method that does not use the tree's hierarchy at all, against cells from transgenic lines labelling known types, against marker-gene expression levels, and against the layer distribution the dissociated cells show Quotation withheld because this record is not cleared for redistribution. mechanisms whether two named types are electrophysiologically distinct is made TESTABLE rather than assumed: a two-component model, one component per type, is fitted against a single-component model of the two merged, and how much better the split describes the data is then measured as a log-likelihood ratio between the two models Quotation withheld because this record is not cleared for redistribution. mechanisms correspondence between the two classifications is found by spectral co-clustering of the two type sets, with agreement given a stated threshold - at least two thirds of a type's cells falling inside its co-cluster Quotation withheld because this record is not cleared for redistribution. mechanisms THE FINAL TYPES ARE EDITED CO-CLUSTERS, NOT CO-CLUSTERS: the poorly-agreeing types are removed first, and the remaining co-clusters are split further wherever the transcriptomic differences inside one are large - so the twenty met-types are the output of a procedure with two further editing steps applied AFTER the clustering, not of the clustering itself Quotation withheld because this record is not cleared for redistribution. Too few morphological reconstructions exist for the Vip, Lamp5 and Sncg subclasses, so the number of met-types may change. what this limits: the met-type count for three named subclasses: WE DO NOT YET HAVE A SUFFICIENTLY LARGE NUMBER OF MORPHOLOGY RECONSTRUCTIONS FOR THE Vip, Lamp5 AND Sncg SUBCLASSES Quotation withheld because this record is not cleared for redistribution. Brain-slice preparation artifacts cannot be ruled out as a source of the observed morphological differences. what this limits: the morphological differences the classification rests on: WE CANNOT RULE OUT THE POSSIBILITY THAT ARTIFACTS OF THE BRAIN SLICE PREPARATION CONTRIBUTE TO OR EXPLAIN OBSERVED MORPHOLOGICAL DIFFERENCES Quotation withheld because this record is not cleared for redistribution. Restricting analysis to highly consistent mappings may impose discrete types on what is really continuous variation. what this limits: the discreteness of the types themselves: SOME OF THE AMBIGUITY IN TYPES COULD BE BIOLOGICAL RATHER THAN TECHNICAL, SUCH THAT FOCUSING OUR ANALYSIS THIS WAY MAY OVERLY-DISCRETIZE A LANDSCAPE WITH MORE CONTINUOUS VARIATION BETWEEN CELLS AND CELL TYPES Quotation withheld because this record is not cleared for redistribution. The field is stated to have no consensus number, or even definition, of neuronal cell types. what this limits: any count of neuronal cell types taken as settled: the authors state the field has not readily converged on a consensus number, or even definition, of neuronal cell types, and that efforts to determine one have been limited by well-documented challenges Quotation withheld because this record is not cleared for redistribution. Two modalities counting the same area's interneuron types differ by more than a factor of two. what this limits: any number of GABAergic cell types quoted for mouse VISp: the transcriptomic estimate is 60 t-types and the morpho-electric estimate 26 me-types, and the authors state such discrepancies are not unique to VISp Quotation withheld because this record is not cleared for redistribution. A classical cell-type name has no absolute definition, and by the loose reading every met-type would qualify. what this limits: any classification of a neuron as a Martinotti cell: the authors state they are not aware of an absolute definition in terms of total amount of axon in L1, that it is generally described as significant, and that all the Sst met-types contain at least one cell that could be considered one Quotation withheld because this record is not cleared for redistribution. The combined classifier's stated improvement is over the morphology arm, not the electrophysiology arm. what this limits: any claim that combining modalities improved t-type prediction here: electrophysiology alone gave 57%, morphology alone 41%, and the combination 57% Quotation withheld because this record is not cleared for redistribution. Type frequencies differ between the two preparations despite the same transgenic driver lines. what this limits: any comparison of t-type proportions between Patch-seq and dissociated-cell data: the difference persisted despite using the same, relatively specific transgenic driver lines, and the authors state it may reflect gene expression differences in slice versus dissociated cell preparations Quotation withheld because this record is not cleared for redistribution. the reference cells split into an 80 per cent training set and a 20 per cent test set before the classifier was fitted eliminates: a cell-type accuracy earned on the cells the classifier was fitted to - the held-out fifth is the only part of the data that can report performance · outcome (stated by the tagger, no quotation): held Quotation withheld because this record is not cleared for redistribution. t-types t-types Quotation withheld because this record is not cleared for redistribution. me-types me-types Quotation withheld because this record is not cleared for redistribution. Martinotti cell Martinotti cell Quotation withheld because this record is not cleared for redistribution. met-types met-types Quotation withheld because this record is not cleared for redistribution. co-clusters co-clusters Quotation withheld because this record is not cleared for redistribution. consensus number consensus number Quotation withheld because this record is not cleared for redistribution. t-type classifier t-type classifier Quotation withheld because this record is not cleared for redistribution. a Martinotti cell - A NAMED TYPE WHOSE DEFINING QUANTITY HAS NO THRESHOLD: the quantity is total amount of axon in L1, the criterion on it is the word significant, the authors state they are not aware of an absolute definition, and all the Sst met-types contain at least one cell that could therefore be considered one Quotation withheld because this record is not cleared for redistribution. a met-type - A CO-CLUSTER THAT WAS THEN EDITED: spectral co-clusters of t-types and me-types were the starting point, t-types with poorer agreement were removed, and co-clusters were split further where transcriptomic differences were large, so the twenty is the output of a procedure with two judgement steps rather than a count of found groups Quotation withheld because this record is not cleared for redistribution. the counted number of GABAergic cell types - MODALITY-RELATIVE: for one cortical area of one species it is 60 by transcriptomics and 26 by morpho-electric features, a difference of more than a factor of two, and the authors state other areas have likewise arrived at varying numbers Quotation withheld because this record is not cleared for redistribution. extends → Tasic et al. (2018/2016) transcriptomic taxonomy and Gouwens et al. (2019) morpho-electric types Quotation withheld because this record is not cleared for redistribution. builds-on → Tasic et al. (2018) dissociated-cell transcriptomic reference dataset Quotation withheld because this record is not cleared for redistribution. t-types —contrasted-with→ me-types t-types me-types differs in: how many kinds each yields for one cortical area of one species: 60 against 26 · strength of the relation: 1 Quotation withheld because this record is not cleared for redistribution. met-types —operationalised-as→ co-clusters met-types co-clusters holds when: after t-types with poorer agreement were removed and co-clusters were split further where transcriptomic differences were large · strength of the relation: 1 Quotation withheld because this record is not cleared for redistribution. sparse principal component analysis sparse principal component analysis Quotation withheld because this record is not cleared for redistribution. principal component analysis principal component analysis Quotation withheld because this record is not cleared for redistribution. likelihood ratio test likelihood ratio test Quotation withheld because this record is not cleared for redistribution. bootstrap bootstrap Quotation withheld because this record is not cleared for redistribution. statistical-significance statistical-significance Quotation withheld because this record is not cleared for redistribution. random forest classifier random forest classifier Quotation withheld because this record is not cleared for redistribution. cross-validation cross-validation Quotation withheld because this record is not cleared for redistribution. discriminability index (effect size) discriminability index (effect size) about this quote: DAMAGED GLYPHS IN THIS QUOTE: (cid:48). These are raw PDF font glyph indices that leaked out in place of the characters they stand for - usually a mathematical operator. The quote is verbatim to the staged text and its anchor is sound; the STAGED TEXT is what is damaged. NOT repaired: the cid-to-character mapping is defined by the font embedded in that particular PDF, so guessing it would produce a clean and unverifiable quote. Repair needs the source PDF and is a re-staging job. Quotation withheld because this record is not cleared for redistribution. correlation correlation Quotation withheld because this record is not cleared for redistribution. Kullback-Leibler divergence Kullback-Leibler divergence Quotation withheld because this record is not cleared for redistribution. Jensen-Shannon divergence Jensen-Shannon divergence Quotation withheld because this record is not cleared for redistribution. standard error of the mean standard error of the mean Quotation withheld because this record is not cleared for redistribution. resampling / subsampling for cluster stability resampling / subsampling for cluster stability Quotation withheld because this record is not cleared for redistribution. The combined data support at least twenty GABAergic interneuron met-types whose transcriptomic, electrophysiological and morphological properties agree. Quotation withheld because this record is not cleared for redistribution. Nine met-types were resolved within the Sst subclass. Quotation withheld because this record is not cleared for redistribution. Sst neurons respect laminar boundaries, and their dominant laminar target may distinguish them more reliably than the presence of axon in layer 1. Quotation withheld because this record is not cleared for redistribution. Most transcriptomic types differ systematically from one another in their electrophysiological properties. Quotation withheld because this record is not cleared for redistribution. Adding morphological to electrophysiological features raised t-type classifier accuracy to 57%. Quotation withheld because this record is not cleared for redistribution. A neuron's specific t-type is frequently hard to predict from its morphological and/or physiological features. Quotation withheld because this record is not cleared for redistribution. Somata of many GABAergic t-types occupy specific laminar or sublaminar positions, giving an anatomical correlate to purely transcriptomic groupings. Quotation withheld because this record is not cleared for redistribution. Individual cells' axon depth histograms correlate positively with their t-type average, indicating consistent innervation patterns within a type. Quotation withheld because this record is not cleared for redistribution. met-type met-type Quotation withheld because this record is not cleared for redistribution. t-type t-type Quotation withheld because this record is not cleared for redistribution. “inconsistent” cells “inconsistent” cells Quotation withheld because this record is not cleared for redistribution. Patch-seq recording technique Patch-seq recording technique Quotation withheld because this record is not cleared for redistribution. transcriptomic type transcriptomic type Quotation withheld because this record is not cleared for redistribution. intrinsic electrophysiological features intrinsic electrophysiological features Quotation withheld because this record is not cleared for redistribution. morphological features morphological features Quotation withheld because this record is not cleared for redistribution. soma location soma location Quotation withheld because this record is not cleared for redistribution. 3,708 — cortical GABAergic interneurons whose transcriptomic and electrophysiological data passed quality control; 350 of them were morphologically reconstructed Quotation withheld because this record is not cleared for redistribution. 3,417 of the 3,708 — cells mapped to one or more reference transcriptomic types with better than 70% confidence Quotation withheld because this record is not cleared for redistribution. >90% — accuracy of the random forest classifier predicting transcriptomic subclass from electrophysiological features Quotation withheld because this record is not cleared for redistribution. 57% — accuracy of the classifier predicting individual t-type from electrophysiological features Quotation withheld because this record is not cleared for redistribution. 41% — accuracy of the classifier predicting t-type from morphological features alone Quotation withheld because this record is not cleared for redistribution. 57% — accuracy of the classifier when electrophysiological and morphological features were combined - no better than electrophysiology alone Quotation withheld because this record is not cleared for redistribution. 83% — share of the authors' Patch-seq recordings made in primary visual cortex, the remainder in surrounding higher visual areas Quotation withheld because this record is not cleared for redistribution. (60) and (26) — previously estimated numbers of VISp GABAergic transcriptomic types versus morpho-electric types, the discrepancy this study set out to resolve Quotation withheld because this record is not cleared for redistribution. twenty — met-types produced by the co-clustering plus transcriptomic-splitting procedure Quotation withheld because this record is not cleared for redistribution.

The full text of this article is not redistributable, so it is not included here. The evidence sentences are withheld with it; the tag layer and its grounding status remain. The article itself is at https://doi.org/10.1101/2020.02.03.932244.