DANDI:000023

Claims 33

From the paper 10.1101/2020.03.31.018820 · claims file · checked 2026-10-22: ready

• features-and-labels-link

  The dandiset's recordings can be matched to the paper's per-cell table (t-type and depth) by specimen ID, and the 18 features computed from them agree with the paper's feature table.

  How it will be judged

  Cells are matched by the specimen ID (SpecimenID in the code's cell table). The 18 features are computed from each matched recording with IPFX, after the sweep QC of the Methods. For each feature, the median over matched cells of the absolute difference from the code's table relative to the table's absolute value is taken (10%, the protocol's default tolerance, as the bound). Reproduced when at least 283 of the 385 cells match and all 18 features are within 10%; partially reproduced when cells match and 9 to 17 features are within 10%; not reproduced otherwise. The number of matched cells, and how many of them carry a t-type or depth in the files themselves, is reported. The code's table is from the revision (316 human cells), so cells beyond the preprint's 283 are expected.

• cohort-cells

  385 Patch-seq neurons passing transcriptomic QC mapped to the five supragranular glutamatergic t-types, and 283 of them had recordings complete enough to compute electrophysiological features.

  A total of 385 neurons that passed transcriptomic data QC mapped with higher confidence to the five supragranular human glutamatergic t-types than to any other neuron type. ... Most neurons (283) also had sufficiently complete recordings to calculate electrophysiological features.

  Results, Patch-seq pipeline for human neurosurgical tissue analysis, paragraph 2

  How it will be judged

  Counts are records, so the tolerance is zero. The 385 is counted as the cells of the code's table that have a recording in the dandiset; the 283 as those whose recordings give the 18 features with IPFX (the paper does not say how many features may be missing; readings are (a) at least one feature, (b) all 18, (c) no more than 3 of 18 missing, the rule the Methods give for the projections). Reproduced when both counts match under one reading; partially reproduced when one does. The dandiset may hold other cells; they are counted and reported apart.

• cohort-donors

  The neurons came from neurosurgical specimens of 90 tissue donors.

  across human neurosurgical specimens from 90 different tissue donors

  Discussion, paragraph 2

  How it will be judged

  Donors are counted as distinct subjects. The paper allows two readings, (a) all subjects in the dandiset, or (b) the donors of the 385 cells of the code's table that are in the dandiset. Reproduced when either gives 90. The code's pathology table also lists 90 donors, which is reported.

• cohort-type-counts

  The electrophysiology figures show 66 LTK, 25 GLP2R and 136 FREM3 neurons, and 21 deep FREM3, 17 CARM1P1 and 37 COL22A1 neurons.

  Bottom panels compare the intrinsic electrophysiological responses for 66 LTK, 25 GLP2R and 136 FREM3 neurons. ... Electrophysiological description of the intrinsic electrophysiology responses of 21 deep FREM3, 17 CARM1P1 and 37 COL22A1 neurons.

  Figure 4a legend; Figure 6b legend

  How it will be judged

  Counts per t-type of the cells with features, under the readings of cohort-cells for which cells have features; deep FREM3 also needs a normalized depth above 0.575. Reproduced when all six match under one reading; partially reproduced when at least three do.

• fig1a-thickness

  Human L2 and L3 together (1.23 +/- 0.15 mm) are on average 1.16 times thicker than the whole mouse cortex (1.06 +/- 0.01 mm).

  The combined thickness of the relatively thin L2 and very thick L3 in human cortex (1.23 ± 0.15 mm) is on average 1.16 times greater than the thickness of the entire mouse cortex (1.06 ± 0.01 mm; Fig. 1a ).

  Results, Human supragranular cortex is more diverse than in mouse, paragraph 1; Figure 1a

  How it will be judged

  Tolerances are the reported spreads (SD across donors). Not tested on the dandiset.

• fig1b-density

  Human supragranular neuron density is 27.7 +/- 4.5 thousand cells/mm3, highest in L2 and halving to a minimum in mid L3, while mouse VISp L2/3 has about 6 times that density (165 +/- 24.9 thousand cells/mm3), uniform with depth.

  the average density of human neurons was 27.7 ± 4.5 thousand cells/mm 3 ( Fig. 1b ; left). This distribution is not homogeneous, with higher density in L2 that decreases by half to reach a low point in mid L3 ( Fig. 1b ; right). In contrast, supragranular layers of mouse VISp show 6 times the neuronal density of human (165 ± 24.9 thousand cells/mm 3 ) with a homogeneous distribution across cortical depth.

  Results, Human supragranular cortex is more diverse than in mouse, paragraph 1; Figure 1b

  How it will be judged

  Tolerances are the reported spreads. Method per the Methods (20 bins over L2/3, first and last dropped, one section per donor). Not tested on the dandiset.

• fig1c-soma-area

  Mean soma area doubles from L2 to deep L3 in human but is uniform across mouse L2/3; the interquartile ranges are 44 to 96 um2 in mouse and 107 to 253 um2 in human, and the largest human somata exceed 850 um2.

  the average cross-sectional area of neuron somata doubles from L2 to deep L3 in human supragranular cortex in a graded fashion but in mouse is remarkably uniform across the depth of supraganular cortex ( Fig. 1c , left). The interquartile range of mouse and human somata were 44 - 96μm 2 and 107 - 253μm 2 , respectively, with the largest human somata exceeding 850μm 2 .

  Results, Human supragranular cortex is more diverse than in mouse, paragraph 1; Figure 1c

  How it will be judged

  Not tested on the dandiset.

• fig1c-soma-variation

  The variation of soma size in deep L3 is four times higher in human than in mouse.

  variation in deep L3 neuron soma size is four-fold higher in human compared to mouse ( Fig. 1c , right)

  Results, Human supragranular cortex is more diverse than in mouse, paragraph 1; Figure 1c right

  How it will be judged

  Not tested on the dandiset. The bins that make up deep L3 are not stated.

• fig1f-frem3-heterogeneity

  The human FREM3 t-type has more within-type transcriptomic heterogeneity (variance explained by PC1 above permuted data) than any other human or mouse supragranular t-type, which are similar to one another.

  This analysis confirms high heterogeneity in FREM3 , with lower, generally similar, values for all other human and mouse t-types ( Fig. 1f ).

  Results, Human supragranular cortex is more diverse than in mouse, paragraph 3; Figure 1f

  How it will be judged

  Method per the Methods (80 cells per type, 80 most variable genes, 100 repeats against gene-shuffled data). Not tested on the dandiset.

• fig1g-deep-types-de-genes

  The homologous types are similarly discrete from one another in mouse and in human, while CARM1P1 and COL22A1 have many more differentially expressed genes against the homologous types than the homologous types have against one another, in either species.

  Homologous types are similarly discrete from one another in both mouse in human, whereas the deep L3 CARM1P1 and COL22A1 t-types had many more DE genes when compared to the homologous t-types ( Fig. 1g ).

  Results, Human supragranular cortex is more diverse than in mouse, paragraph 3; Figure 1g

  How it will be judged

  Not tested on the dandiset.

• fig3a-platform-correlation

  Human Patch-seq cells and dissociated nuclei of matched t-types are highly correlated in gene expression (R = 0.85), more than mouse cells and human nuclei (R = 0.57); 177 genes are enriched in nuclei and 2,670 at least four-fold in Patch-seq.

  Expression data from dissociated mouse whole cells and human nuclei were moderately correlated (R=0.57, p~0, Fig. 3a , left), ... dissociated nuclei and Patch-seq cells from matched human t-types were highly correlated (R=0.85, p~0, Fig. 3a , right). Relatively few genes (177 genome-wide) showed enriched expression in dissociated nuclei relative to Patch-seq cells ... we identified 2,670 genes with at least four-fold enrichment in Patch-seq

  Results, High-confidence Patch-seq neuron t-type mapping, paragraph 2; Figure 3a

  How it will be judged

  Not tested on the dandiset.

• fig2d-pathology-scores

  Average pathology scores over six markers ranged from 0.06 to 1.83 across donors, and most cases had averages below 1.0, a range not considered overtly pathological.

  images from the case with the lowest average score (mean of 6 marker scores = 0.06) are contrasted with images from the case with the highest average score (mean of 6 marker scores = 1.83) ( Fig. 2c ). ... However, this was rare, and most cases had average scores <1.0 ( Fig. 2d )

  Results, Detailed histological assessment of human surgical tissue, paragraph 2; Figure 2c and 2d

  How it will be judged

  The average is of GFAP, IBA1, Ki-67, SMI-32, NeuN and Nissl (cellularity) scores, as the code's pathology_analysis.ipynb sums them. "Most" is read as more than half of the donors with all six scores. Not tested on the dandiset.

• edfig1-marker-correlation

  The six pathology markers are very weakly correlated with one another across donors, with only Ki-67 and Nissl (cellularity) modestly correlated.

  we found very low correlation of the 6 cellular markers with each other, with only Ki-67 and Nissl (cellularity) being modestly correlated ( Extended Data Fig. 1 )

  Results, Detailed histological assessment of human surgical tissue, paragraph 2; Extended Data Figure 1

  How it will be judged

  Pearson correlation across donors (Extended Data Fig. 1 legend). The paper gives no threshold for "modest"; reproduced when the Ki-67 and Nissl pair has the largest absolute correlation of the 15 pairs. Not tested on the dandiset.

• fig2ef-pathology-ephys

  Of 18 electrophysiological features compared between neurons from donors with low (0 to 1) and high (1 to 3) GFAP or IBA1 scores, only resting membrane potential under IBA1 differs, neurons from high-IBA1 donors being about 5 mV more hyperpolarized; the other 17 IBA1 and all 18 GFAP comparisons do not differ.

  Among the 18 electrophysiological features analyzed for GFAP and IBA1, only one feature (resting membrane potential, RMP) related to IBA1 was significantly different between low versus high groups. Neurons in the high IBA1 group were approximately 5 mV more hyperpolarized at rest than neurons in the low IBA1 group ( Fig. 2f ). The remaining 17/18 features for IBA1 and all 18 features for GFAP were not statistically different between low and high groups

  Results, Detailed histological assessment of human surgical tissue, paragraph 3; Figure 2e and 2f

  How it will be judged

  Partly a claim of absence, judged on the paper's test. Cells are the supragranular glutamatergic neurons (the code's 385-cell table) with all 18 features defined (the code drops cells with any missing feature). The paper's text says "all supragranular glutamatergic neurons"; the original pathology notebook also keeps only temporal cortex cells of the hIVSCC-MET project, from a table not in the repository. So the readings are (a) temporal cells only (region temporal in the code's cell table), (b) all regions; the group counts above say which the paper used. Each donor's GFAP and IBA1 scores are the columns GFAP Cortex and Iba1 Cortex of the code's data/pathology_scoring.csv (the mean of three neuropathologists), low being at most 1 and high above 1 (the code; the text gives 0-1 and 1-3). The test is the code's, scipy's Mann-Whitney U with its default of scipy 1.4 (the pinned version), whose p is half the two-sided p, at the Bonferroni level 0.05/18 per marker (Figure 2f legend and the code). The paper does not say whether the 5 mV is a difference of group medians or of group means; both are computed, and the size part of the outcome is the one both give (methods_underspecified for the size when they differ). Reproduced when, under one reading of the cells, RMP under IBA1 is the only significant comparison of the 36, with the high group more hyperpolarized and the difference within tolerance; partially reproduced when the significance pattern holds but the size is outside tolerance; not reproduced when any other comparison is significant or RMP under IBA1 is not. The two-sided Mann-Whitney test is shown as a variant, as is the updated notebook's result.

• fig3d-depth-by-type

  LTK neurons lie mainly in L2 and at the L2/L3 border, GLP2R mainly in upper L3, FREM3 span L2 and L3 and continue into L4, and CARM1P1 and COL22A1 lie almost entirely in deep L3 and along the L3/L4 border.

  Human LTK neurons were found primarily in L2 and in the border region of L2 and L3. GLP2R neurons were found primarily in upper L3, with some neurons found in L2. FREM3 neurons spanned L2 and L3, continuing into L4, consistent with their heterogeneous gene expression profile. CARM1P1 and COL22A1 were found almost entirely in deep L3 and along the L3/L4 border.

  Results, High-confidence Patch-seq neuron t-type mapping, paragraph 4; Figure 3d

  How it will be judged

  Normalized L2/3 depth per cell (0 at the L1/L2 border, 1 at the L3/L4 border; Figure 3d legend and Figure 1 legend). The L2/L3 border is not recorded per cell, so L2 is judged by order. Deep L3 is depth above 0.575 (the paper's density minimum) and L4 depth above 1. Reproduced when all of these hold; the median depth of LTK is below that of GLP2R; the GLP2R median is below 0.575; FREM3 includes cells shallower than the LTK median and cells deeper than 1; and, for CARM1P1 and COL22A1 each, "almost entirely" holds. The paper gives no number for "almost entirely"; the readings are (a) at least 90% and (b) at least 80% of cells deeper than 0.575. They are thresholds, so the outcome is the one both give, and the claim is methods_underspecified when they differ. Partially reproduced when the order of medians holds but not the rest. All cells of the code's table with a depth are used, whether or not they have features.

• fig4e-human-ephys-anova

  Among the three human homologous t-types (LTK, GLP2R, FREM3), 3 of 18 electrophysiological features explain more than 10% of variance by t-type (input resistance, f-I slope and rheobase), at FDR below 1e-5, with a maximum R2 of 0.16.

  For electrophysiological features ( Fig. 4e ), 3/18 showed differences between t-types that explained >10% of feature variance ( R 2 >0.1) for both human and mouse (FDR<0.05 for mouse features, <10 −5 for human). ... while the human types showed distinct firing properties (f-I slope and rheobase) in addition to input resistance, with a maximum R 2 =0.16.

  Results, Increased morpho-electric specialization in human L2-3 t-types, paragraph 5; Figure 4e

  How it will be judged

  The t-type ANOVA on LTK, GLP2R and all FREM3 cells (the preprint's Methods say the three homologous t-types; the code's revision uses superficial FREM3 only, depth below 0.575, shown as a variant). Reproduced when exactly the three features named have R2 above 0.1, each at FDR below 1e-5, and the maximum R2 is within tolerance; partially reproduced when at least two of the three named features have R2 above 0.1 at FDR below 0.05.

• fig4e-mouse-ephys-anova

  Among the three mouse L2/3 t-types, 3 of 18 electrophysiological features explain more than 10% of variance (input resistance, AP width and AP downstroke rate), at FDR below 0.05, with a maximum R2 of 0.12, below the human maximum.

  However, the mouse types were distinct in input resistance and two related AP shape features (width and downstroke) with a maximum R 2 =0.12

  Results, Increased morpho-electric specialization in human L2-3 t-types, paragraph 5; Figure 4e

  How it will be judged

  The t-type ANOVA on the mouse Adamts2, Rrad and Agmat cells. The mouse recordings are not in this dandiset and the paper does not say where they are (ref. 32 is the mouse Patch-seq study); the code's feature and cell tables hold their features and t-types. Reproduced when exactly the three features named have R2 above 0.1 at FDR below 0.05, the maximum R2 is within tolerance and is below the human maximum of fig4e-human-ephys-anova; partially reproduced when two of the three hold.

• fig4f-morph-anova

  16 of 60 morphological features have R2 above 0.15 between the human homologous t-types (FDR below 1e-3), against 12 of 60 for mouse (10 of them at FDR below 0.05).

  For morphological features ( Fig. 4f ), 16/60 features had R 2 >0.15 between the human t-types (FDR<10 −3 ), compared to 12/60 for the mouse t-types (with 10/12 significant at FDR<0.05).

  Results, Increased morpho-electric specialization in human L2-3 t-types, paragraph 5; Figure 4f

  How it will be judged

  The t-type ANOVA on morphological features. Not tested on the dandiset.

• fig4-glp2r-sag

  GLP2R neurons, like LTK neurons, lack firing rate adaptation and have higher input resistance, but unlike LTK neurons (which have no sag) they have pronounced sag.

  Electrophysiologically, GLP2R neurons exhibited LTK -like electrophysiology, lacking adaptation and higher input resistance, but differed from LTK neurons in that they had pronounced sag.

  Results, Increased morpho-electric specialization in human L2-3 t-types, paragraphs 1 and 2; Figure 4a

  How it will be judged

  Sag (sag ratio at about -100 pA) and adaptation (adapt_mean) compared between LTK and GLP2R with the pairwise test over the three homologous types (the paper's post hoc test for Figure 4e; the paper shows this finding only as traces). Reproduced when GLP2R sag is significantly higher than LTK sag (FDR below 0.05) and adaptation does not differ between them; partially reproduced when sag is higher but adaptation also differs. The higher input resistance is not judged, since the paper does not say what it is higher than (FREM3 is the likely comparison); the input resistance of the three types is reported.

• edtable1-species-ephys

  Between human and mouse homologous t-types, 17 of 18 electrophysiological features differ at FDR below 0.05.

  For homologous t-types, results are shown for unpaired t-tests across species for each feature ... 17out of 18 features are significant at FDR<0.05.

  Extended Data Table 1 legend

  How it will be judged

  Human LTK, GLP2R and FREM3 cells against mouse Adamts2, Rrad and Agmat cells (mouse features from the code's table), one test per feature, Benjamini-Hochberg over 18. The legend does not say whether the t-test assumes equal variances; readings are (a) Student's and (b) Welch's. A feature counts when its BH-corrected p is below 0.05 (the legend). Reproduced when both readings give 17; the claim is methods_underspecified when they differ in outcome. The code's test (scipy's Mann-Whitney with the scipy 1.4 default, on cells with all 18 features; AP width the one feature not significant) is shown as a variant, as is Cohen's d per feature.

• edfig6-classifier-human-ephys

  A random forest predicts the homologous human t-type from electrophysiological features with 67% accuracy.

  Homologous human t-types were predicted with 67% accuracy using electrophysiological features, while mouse t-types were predicted with 54% accuracy.

  Results, Increased morpho-electric specialization in human L2-3 t-types, paragraph 6; Extended Data Figure 6

  How it will be judged

  As the Methods (Analysis of features by t-type and species, paragraph 3) and the code (classifier_and_species_analysis.ipynb). LTK, GLP2R and FREM3 cells with all 18 features (the code drops any cell with a missing feature), features standardized over all cells, 1000 stratified random splits into 70% training and 30% test, a random forest of 600 trees (sklearn, with the code's settings class_weight balanced, max_features sqrt, min_samples_leaf 2), accuracy on the test set averaged over the splits. Reproduced when the mean accuracy is within tolerance. Accuracy with shuffled t-type labels (Methods) is reported for comparison. The code runs 10 splits (n_repeats = 10), not the Methods' 1000; the Methods are followed, and the number of splits changes only the precision of the mean.

• edfig6-classifier-mouse-ephys

  The same classifier predicts mouse L2/3 t-types from electrophysiological features with 54% accuracy, lower than for human.

  Homologous human t-types were predicted with 67% accuracy using electrophysiological features, while mouse t-types were predicted with 54% accuracy.

  Results, Increased morpho-electric specialization in human L2-3 t-types, paragraph 6; Extended Data Figure 6

  How it will be judged

  As edfig6-classifier-human-ephys, on mouse Adamts2, Rrad and Agmat cells (features from the code's table). Reproduced when the mean accuracy is within tolerance and below the human accuracy.

• edfig6-classifier-morphoelectric

  Combining morphological and electrophysiological features raises accuracy slightly, to 69% for human and 60% for mouse neurons.

  Classifying based on morphological and electrophysiological features combined resulted in slightly improved performance of 69% for human neurons and 60% for mouse neurons

  Results, Increased morpho-electric specialization in human L2-3 t-types, paragraph 6; Extended Data Figure 6

  How it will be judged

  Not tested on the dandiset alone; it needs the morphological features.

• fig5c-frem3-ephys-depth

  In FREM3 neurons, 9 of 18 electrophysiological features correlate with normalized depth at FDR below 0.05 (7 at FDR below 0.01); the strongest are increasing sag, increasing AP upstroke/downstroke ratio and decreasing latency at rheobase.

  Nine out of 18 physiological features were significantly correlated with depth (FDR<0.05, 7/18 FDR<0.01; Supplementary Data 1). The three strongest electrophysiology correlations with depth were the increase in sag and AP (Action Potential) upstroke / downstroke ratio and the decrease in AP latency at rheobase ( Fig. 5c ).

  Results, Human FREM3 t-type exhibits depth-dependent morpho-electro-transcriptomic variation, paragraph 2; Figure 5c

  How it will be judged

  The depth regression on all FREM3 cells with a normalized depth, per feature, BH over the 18. "Strongest" is read as the largest R2, the measure the Methods report. Reproduced when both counts are within tolerance and the three features with the largest R2 are sag and upstroke/downstroke ratio (positive slope) and latency (negative slope), each at FDR below 1e-7; partially reproduced when those three are significant at FDR below 0.05 with those signs but the counts or the ranking differ.

• fig5b-frem3-morph-depth

  In FREM3 neurons, 37 of 58 morphological features correlate with depth at FDR below 0.05 (28 at FDR below 0.01), including apical height, maximum basal dendrite length and soma radius.

  37 out of 58 morphological features were correlated with depth (FDR<0.05, 28/58 FDR<0.01).

  Results, Human FREM3 t-type exhibits depth-dependent morpho-electro-transcriptomic variation, paragraph 2; Figure 5b

  How it will be judged

  Not tested on the dandiset.

• fig5d-frem3-gene-depth

  FREM3 gene expression varies with soma depth; 790 genes correlate with depth at FDR below 0.05, enriched for synaptic transmission, cell migration and neuron projection morphogenesis.

  790 genes were correlated with depth (FDR<0.05) and GO (gene ontology) enrichment analysis on this gene set revealed a variety of significantly enriched functional categories ( Fig. 5d , Supplementary Data 1)

  Results, Human FREM3 t-type exhibits depth-dependent morpho-electro-transcriptomic variation, paragraphs 1 and 2; Figure 5a and 5d

  How it will be judged

  Not tested on the dandiset. The code's gene_depth_analysis.ipynb counts 1917 genes at FDR below 0.01 from its revision's results, so its table is not the preprint's.

• fig6-deep-ephys-anova

  The three deep L3 types (deep FREM3, CARM1P1, COL22A1) differ in 15 of 18 electrophysiological features (FDR below 0.05), with R2 above 0.4 for 4 of them, while no feature reaches R2 0.4 among the three homologous types.

  ANOVA differences were seen in 15 out of 18 electrophysiological features and 24 out of 60 morphological features (FDR<0.05) (Supplementary Data 1). Many of these features showed very large effect sizes, with the variance explained by type surpassing 40% ( R 2 >0.4) for 4 electrophysiological and 20 morphological features, while no features met this threshold for the three homologous types (maximum R 2 =0.26).

  Results, Deep L3 human glutamatergic t-types are morpho-electrically distinct, paragraph 4

  How it will be judged

  The t-type ANOVA on deep FREM3, CARM1P1 and COL22A1 cells (21, 17 and 37 neurons in the Figure 6b legend; CARM1P1 and COL22A1 are not cut by depth, as in the code). A feature differs when its BH-corrected p is below 0.05. Reproduced when both counts are within tolerance and no electrophysiological feature has R2 above 0.4 in fig4e-human-ephys-anova; partially reproduced when one count is within tolerance. The maximum R2 of 0.26 for the homologous types is not tested, as the text does not say which modality gives it and the electrophysiological maximum is 0.16.

• fig6-deep-morph-anova

  The deep L3 types differ in 24 of 60 morphological features (FDR below 0.05), with R2 above 0.4 for 20 of them.

  ANOVA differences were seen in 15 out of 18 electrophysiological features and 24 out of 60 morphological features (FDR<0.05)

  Results, Deep L3 human glutamatergic t-types are morpho-electrically distinct, paragraph 4; Figure 6d

  How it will be judged

  Not tested on the dandiset.

• fig6c-carm1p1-upstroke-rin

  CARM1P1 neurons have a faster AP upstroke than deep FREM3 and COL22A1 neurons, and a higher input resistance than deep FREM3 neurons.

  Electrophysiologically, CARM1P1 neurons exhibited a faster action potential upstroke than the other deep types ( Fig. 6c ). This extensive dendritic branching did not predict a lower input resistance, which was higher than deep FREM3 neurons ( Fig. 6c ).

  Results, Deep L3 human glutamatergic t-types are morpho-electrically distinct, paragraph 2; Figure 6c

  How it will be judged

  The pairwise test over the three deep types (Figure 6c legend; BH over the three pairs of a feature, as the code's plot_mw_bars). "Faster upstroke" is read as (a) the AP upstroke rate (upstroke_long_square) or (b) the upstroke/downstroke ratio, the feature the code plots for the deep types; both are run. The upstroke part is the one both readings give, and is methods_underspecified when they differ. Reproduced when CARM1P1 upstroke is significantly higher than both deep FREM3 and COL22A1, and CARM1P1 input resistance is significantly higher than deep FREM3; partially reproduced when one of the two parts holds.

• fig6c-col22a1-rin-fi-sag

  COL22A1 neurons have very high input resistance and a steeper f-I gain than the other deep L3 types, and less sag than CARM1P1 neurons.

  COL22A1 neurons exhibited very high input resistance ( Fig. 6c ) and thus were the most responsive to current injection, displaying a steeper firing frequency to current input gain relative to the other deep L3 t-types. Interestingly, COL22A1 neurons showed a smaller amount of sag than CARM1P1 neurons ( Fig. 6b )

  Results, Deep L3 human glutamatergic t-types are morpho-electrically distinct, paragraph 3; Figure 6b and 6c

  How it will be judged

  The pairwise test over the three deep types, for input resistance, f-I slope and sag. The sag and f-I gain findings are shown in the paper only in Figure 6b (traces, and f-I curves normalized to rheobase), with no test; the post hoc Mann-Whitney test of the Methods (Analysis of features by t-type and species, paragraph 2) and Figure 6c is used for them too, on the sag and fi_fit_slope features. Reproduced when COL22A1 input resistance and f-I slope are each significantly higher than both deep FREM3 and CARM1P1, and COL22A1 sag is significantly lower than CARM1P1 sag; partially reproduced when at least two of the three features hold.

• fig6a-deep-dendrites

  CARM1P1 neurons have the largest total dendritic length of the L2 and L3 t-types, with extensive proximal oblique and basal branching, while COL22A1 neurons have very sparse branching.

  CARM1P1 neurons had the largest total dendritic length of all the L2 and L3 t-types, despite having a shorter apical dendrite length on average. ... COL22A1 had very sparse dendritic branching ( Fig. 6a ).

  Results, Deep L3 human glutamatergic t-types are morpho-electrically distinct, paragraphs 2 and 3; Figure 6a

  How it will be judged

  Not tested on the dandiset.

• fig6e-deep-de-genes

  219 marker genes distinguish deep FREM3, CARM1P1 and COL22A1, 77 of them also found in Patch-seq cells, enriched for axon, synapse, calcium ion binding and extracellular matrix terms.

  we used genesorteR 44 with slightly relaxed parameters (quant = 0.7) to identify DE genes selective for one or two of these three t-type sets, and found 219 such marker genes ( Extended Data Fig. 9 ). ... Furthermore, 77 of these 219 marker genes (including four genes shown in Fig. 6e ) were also defined as marker genes by Patch-seq

  Results, Deep L3 human glutamatergic t-types are morpho-electrically distinct, paragraphs 5 and 6; Figure 6e; Extended Data Figure 9

  How it will be judged

  Not tested on the dandiset.

• fig6f-smi32

  Large FREM3 and CARM1P1 neurons in deep L3 are SMI-32 immunoreactive while COL22A1 neurons are not, and NEFH, the gene coding the SMI-32 epitope, is higher in deep FREM3 and CARM1P1 than in COL22A1 and the superficial types.

  The large FREM3 and CARM1P1 neurons were SMI-32ir, while COL22A1 neurons were not SMI-32ir. Furthermore, the gene coding SMI-32, NEFH, also shows increased expression in deep FREM3 and CARM1P1 relative to COL22A1 and all the superficial glutamatergic t-types ( Fig. 6e ).

  Results, Deep L3 human glutamatergic t-types are morpho-electrically distinct, paragraph 7; Figure 6e and 6f

  How it will be judged

  Not tested on the dandiset. The paper gives no counts of labelled cells.

Pipeline

Paper

done (1 of 1)

Claims

done

Claims check

done (ready)

Attempt

todo

Questions

none

Verified

none