What was used to reverse transcribe poly(A) RNA and amplify?

TheSMART-seq v4 (SSv4) Ultra Low Input RNA Kit for Sequencing (Takara 634894) was used to reverse transcribe poly(A) RNA and amplify full-length cDNA.

How many bins were used to compute the dispersion for each feature?

Highly variable isoforms and genes were identified by first computing the dispersion for each feature, and then binning all of the features into 20 bins.

How many components were used to visualize the NCA data?

To visualize the SMART-seq data with predefined cluster labels produced via a joint analysis with many other data types the authors performed NCA59 on the full scaled log(TPM + 1) matrix using the subcluster labels, to ten components.

What was the GTF and the GRCm38 genome fasta file?

The GTF and the GRCm38 genome fasta file (https://github.com/ pachterlab/BYVSTZP_2020/releases/tag/biorxiv_v1), provided by the BICCN consortium, were used to create a transcriptome fasta file, transcripts-to-genes map, and kallisto index using kb ref -i index.idx, -g t2g.txt -f1 transcriptome.fa genome.fa genes.gtf.

Why did the authors exclude the L5 IT subclass from their analyses?

Without being able to rule out that the low correlation for L5 IT cells across the technologies was due to confounding between batch and sex in the dataset, the authors decided to excluded the subclass from their analyses.

(Open Access) Isoform specificity in the mouse primary motor cortex (2020) | A. Sina Booeshaghi

Q: What contributions have the authors mentioned in the paper "Isoform cell-type specificity in the mouse primary motor cortex" ?

Additionally, the authors show that isoform specificity helps to refine cell types, and that a multi-platform analysis of single-cell transcriptomic data leveraging multiple measurements provides a comprehensive atlas of transcription in the mouse primary motor cortex that improves on the possibilities offered by any single technology.

Q: What is the SMART-seq method used to measure cell type specificity?

SMART-seq19 is an scRNA-seq method that produces full-length reads, enabling the quantification of individual isoforms of genes with the expectation-maximization algorithm20.

Q: what is the mRNA isoform in the human frontal lobe?

Unique transcriptome patterns of the white and grey matter corroborate structural and functional heterogeneity in the human frontal lobe.

Q: How many cells were assayed with MERFISH?

The authors analysed 6,160 mouse MOp cells assayed with SMART-seq, 280,327 cells assayed with MERFISH, and 94,162 cells assayed with 10x Genomics Chromium v3.

Q: What is the likely explanation for the differences in isoforms between classes?

These cases are likely to be instances where isoform shifts between cell types are a result of differential splicing—that is, the result of a post-transcriptional program.

Nature | Vol 598 | 7 October 2021 | 195

Article

Isoform cell-type specificity in the mouse

primary motor cortex

A. Sina Booeshaghi

, Zizhen Yao

, Cindy van Velthoven

, Kimberly Smith

, Bosiljka Tasic

Hongkui Zeng

& Lior Pachter

3,4

✉

Full-length SMART-seq

single-cell RNAsequencing can be used to measure gene

expression at isoform resolution, making possible the identication of specic

isoform markers for dierent cell types. Used in conjunction with spatial RNA capture

and gene-tagging methods, this enables the inference of spatially resolved isoform

expression for dierent cell types. Here, in a comprehensive analysis of 6,160 mouse

primary motor cortex cells assayed with SMART-seq, 280,327 cells assayed with

MERFISH

and 94,162 cells assayed with 10x Genomics sequencing

, we nd examples

of isoform specicity in cell types—including isoform shifts between cell types that

are masked in gene-level analysis—as well as examples of transcriptional regulation.

Additionally, we show that isoform specicity helps to rene cell types, and that a

multi-platform analysis of single-cell transcriptomic data leveraging multiple

measurements provides a comprehensive atlas of transcription in the mouse primary

motor cortex that improves on the possibilities oered by any single technology.

Transcriptional and post-transcriptional control of individual iso-

forms of genes is crucial for neuronal differentiation

4–8

, and isoforms

of genes have been shown to be specific to cell types in mouse and

human brains

9–14

. It is therefore not surprising that dysregulation of

splicing has been shown to be associated with several neurodevel-

opmental and neuropsychiatric diseases

6,15,16

. Thus, it is of interest to

study gene expression in the brain at single-cell and isoform resolution.

Nevertheless, current single-cell studies aiming to characterize

cell types in the brain using single-cell RNA sequencing (scRNA-seq)

have relied mostly on gene-level analysis. This is, in part, owing to the

nature of the data produced by the highest-throughput single-cell

methods. Popular high-throughput assays such as Drop-seq

, 10x

Genomics Chromium

and inDrops

produce 3′-end reads that are,

in initial pre-processing, collated by gene to produce per-cell gene

counts. SMART-seq

is an scRNA-seq method that produces full-length

reads, enabling the quantification of individual isoforms of genes with

the expectation-maximization algorithm

. However, such increased

resolution comes at the cost of throughput. SMART-seq requires cells

to be deposited in wells, thus limiting the throughput of the assay. In

addition, SMART-seq requires more sequencing per cell

The trade-offs are evident in analysis of scRNA-seq data from the pri-

mary motor cortex (MOp) produced by the BRAIN Initiative Cell Census

Network (BICCN)

. We examined 6,160 (filtered) SMART-seq v4 cells

and 94,162 (filtered) 10x Genomics Chromium (10xv3) cells (Extended

Data Fig.1, Fig.2a, b) and found that while 10xv3 and SMART-seq are

equivalent in defining broad classes of cell types, 3′-end technol-

ogy that can assay more cells can identify some rare cell types that

are missed at lower cell coverage (Extended Data Fig.2a). Overall, 56

clusters with gene markers could be identified in the 10xv3 data but

not in the SMART-seq data, whereas only 39 clusters with gene markers

could be identified in the SMART-seq data and not the 10xv3 data—this

differential is consistent with previously reported comparisons of

10x Genomics Chromium and SMART-seq clusters

21,23

. However, while

SMART-seq has lower throughput than some other technologies, it has

a notable advantage: because it probes transcripts across their entire

length, SMART-seq makes possible isoform quantification and the

detection of isoform markers for cell types that cannot be detected

with 3′-end technologies (Extended Data Figs.2b, c). Moreover, the

uniformity of read coverage of SMART-seq data

and its quantifica-

tion with state-of-the-art tools

yields higher sensitivity than other

methods, which can make possible refined cell-type classification.

To take advantage of the complementary strengths of these differ-

ent platforms, we introduce an approach to scRNA-seq that links the

SMART-seq resolved isoforms to the 10x Chromium defined cell types,

and merges this information with spatial transcriptomic measurements

obtained by MERFISH

(Fig.1). In addition to revealing extensive iso-

form diversity and cell-type specificity in the MOp, we find evidence

for previously missed transcriptionally distinct cell subtypes in the

MOp. Our results extend the notion of a single-cell database beyond a

list of gene markers, and we produce a gene-isoform-space single-cell

atlas of the MOp using the combined 10xv3, SMART-seq and MERFISH

data. Our methods are open source, reproducible, easy to use and con-

stitute an effective workflow for leveraging full-length scRNA-seq data

in combination with data from other technologies.

Isoform markers for cell types

To identify isoform markers of cell types, we first sought to visualize

our SMART-seq data using gene-derived cluster labels from the BICCN

analysis (Methods). Rather than layering cluster labels on cells mapped

https://doi.org/10.1038/s41586-021-03969-3

Received: 9 March 2020

Accepted: 27 August 2021

Published online: 6 October 2021

Open access

Check for updates

Department of Mechanical Engineering, California Institute of Technology, Pasadena, CA, USA.

Allen Institute for Brain Science, Seattle, WA, USA.

Division of Biology and Biological

Engineering, California Institute of Technology, Pasadena, CA, USA.

Department of Computing and Mathematical Sciences, California Institute of Technology, Pasadena, CA, USA.

✉

e-mail: lpachter@caltech.edu

196 | Nature | Vol 598 | 7 October 2021

Article

to 2D with an unsupervised dimensionality-reduction technique such

as t-distributed stochastic neighbour embedding

(t-SNE) or uniform

manifold approximation and projection

(UMAP), we used a super-

vised learning approach to project cells so that they are best separated

according to BICCN consortium

annotations using neighbourhood

component analysis (NCA). This method produces meaningful rep-

resentation of the global structure of the data (Fig.2b), without over-

fitting (Supplementary Fig.1a). Analysis of the projections revealed

a batch effect in the 10xv3 data—which we addressed by restricting

analysis to a single batch—and minimal evidence of a batch effect in

the MERFISH data (Methods, Supplementary Fig.2a, b).

Next, motivated by the discovery of genes exhibiting differential

exon usage between glutamatergic and GABAergic (γ-aminobutyric

acid-producing) neurons in the primary visual cortex

, we performed a

differential analysis between these two classes of neurons. We searched

for significant shifts in isoform abundances in genes whose expression

was stable across cell types (Methods). We discovered 398 such isoform

markers belonging to 310 genes (Supplementary Table1). Figure2c

shows an example of such an isoform from the oxidative resistance 1

(Oxr1) gene, which is known to be essential for protection against oxi-

dative stress-induced neurodegeneration

28,29

. While we see no change

in gene expression of Oxr1 between these two neuron types, we find

that among the 16 isoforms of the gene, one of them, Oxr1-204, is more

highly expressed in glutamatergic neurons. The Oxr1 gene undergoes

an isoform shift in GABAergic neurons, where the expression of the

Oxr1-204 isoform is significantly lower, suggesting distinct subcellular

isoform localization in the two neuron types

. A gene-level analysis is

blind to this isoform shift (Fig.2c, top right).

We hypothesized that there are genes exhibiting cell-type isoform

specificity at all levels of the MOp cell ontology. However, detection of

such genes and their associated isoforms requires meaningful cell-type

assignments and accurate isoform quantifications. To assess the reli-

ability of the SMART-seq clusters produced by the BICCN

, we exam-

ined the correlation in gene expression by cluster with an orthogonal

scRNA-seq technology, the 10xv3 3′-end assay. We clustered 94,162

10xv3 cells, also derived from the MOp, using the same method as the

SMART-seq cells (Methods). The clustering method generates three

hierarchies of cells: classes, subclasses and clusters. The SMART-seq

data have 2 major classes (glutamatergic and GABAergic), 18 subclasses

that subdivide the classes, and 62 clusters that subdivide the subclasses.

The 10x data similarly contain three hierarchies of cells: two major

classes (glutamatergic and GABAergic), 21 subclasses and 85 clusters.

We found high correlation of gene expression between the two assays

at the subclass and cluster levels (Extended Data Fig.3).

Next, we assessed the accuracy of the SMART-seq isoform quantifica-

tion and its concordance with 10xv3 quantifications of isoforms. Since

not all isoforms can be quantified from 10xv3 3′-end data, we examined

only isoforms containing some unique 3′ UTR sequence. This enabled

us to validate the isoform quantifications using a different method

(Methods). To extract isoform quantifications from 10xv3 data in cases

where there was a unique 3′ sequence, we relied on transcript compat-

ibility counts

produced by pseudoalignment with kallisto

. We were

able to validate the SMART-seq isoform shift predictions at both the

subclass and cluster levels (Extended Data Fig.4). The isoform abun-

dance correlations are slightly lower than those for gene abundance

estimates (Extended Data Fig.3), but sufficiently accurate to identify

significant isoform shifts, consistent with benchmarks showing that

isoforms can be quantified accurately from full-length bulk RNA-seq

Having validated the cluster assignments and isoform abundance

estimates, we tested for isoform switches for 16 cell subclasses exclud-

ing low quality cells (example in Fig.2d), and then for 48 distinct clusters

for subclasses that have more than one cluster (example in Fig.2e) and

more than 5 cells per cluster (Methods). At the higher level of 16 cell

subclasses, we found a total of 654 isoforms from 550 genes within

the glutamatergic class and 381 isoforms from 332 genes within the

GABAergic class exhibiting isoform shifts among the 16 cell subclasses

despite constant gene abundance (Supplementary Table2a, b). There

are several notable examples of isoform shifts at this level. For exam-

ple, we find a shift in the Snap25-202 isoform, whose expression has

been specifically shown to be correlated with age and to differentially

3′

Gene tagging

Drop-seq, inDrops or 10x Chromium

Smart-seq, Oxford Nanopore or

Pacic Biosciences

Detector

3′

5′

Isoform sequencing

Spatial capture

seqFISH, MERFISH or 10x Visium

Genes Isoforms

Gene

space

Isoform

space

Physical

space

Cell type

identication

Isoform

renement

Spatial

tethering

Fig. 1 | Measuring RNA with multiple platforms. RNA is measured using

gene-tagging techniques such as the 10x Chromium scRNA-seq protocol,

isoform sequencing techniques such as SMART-seq, and spatial RNA-capture

techniques such as MERFISH. High-cell-throughput gene tagging enables

cell-type identification with marker genes and deep full-length isoform

sequencing enables cell-type marker refinement at the isoform level. Spatial

RNA capture coupled with gene tagging and isoform sequencing enables

spatial resolution of cell-type-specific isoform markers. The multi-method

procedure for sampling RNA enables inference of spatially resolved cell-type-

specific isoforms that no single technique could achieve independently

Nature | Vol 598 | 7 October 2021 | 197

regulate synaptic transmission and synaptic plasticity at central syn-

apses

34,35

. This isoform marks the L6b subclass (Fig.2d). At the cluster

level, we found 923 isoforms from 823 genes exhibiting isoform shifts

among the 48 clusters passing filter despite constant gene abundance

(Supplementary Table3). Another isoform of notethat marks the L6b

Ror1_ 1 cluster, a subset of cells in the L6b subclass, is the Stxbp2-207

isoform whose gene Stxbp2 has previously been detected in the sub-

thalamic nucleus and the posterior hypothalamus

Assaying both male and female mice enabled us to examine

sex-specific effects in all subclasses except for the L5 IT, which was

excluded owing to batch effect (Supplementary Fig.4). In total, these

subclasses exhibited 418 sex-specific isoforms, averaging 40 iso-

forms per subclass (Supplementary Table7). Unlike a recent study

that reported a sex-specific cell type in the ventromedial nucleus of

the hypothalamus

, we do not find any sex-specific subclasses. How-

ever, we observed several autosomal isoforms that were differentially

expressed between male and female mice. Among these, the Shank1-203

isoform is differential in Vip neurons, a finding that refines previous

data showing that Shank1, which has been shown to localize in Purkinje

cells in the cortex

, is a sex-specific gene whose expression is regulated

by sex hormones

We also investigated instances where clusters could be refined

according to isoform expression. After reclustering each 10xv3-derived

cluster using SMART-seq isoform quantifications (seeMethods), we

found that 12 clusters can be split by isoforms. Examining the L6 CT

Grp_1 cluster, we find that the average effect size for differential iso-

forms that split the cluster into two sub-clusters is higher than that for

genes (Extended Data Fig.5). One isoform in particular, which splits

the L6 CT Grp_1 cluster, is a protein-coding isoform of the amyloid

precursor protein gene (App). Dysregulation of splicing for isoforms

of App have been associated with disease pathogenesis in Alzheimer

disease models

. Our findings show that isoform-level expression can

help refine cell types in the mouse MOp beyond what is possible using

gene-level expression estimates.

Along with isoforms detectable as differential between cell types

without change in gene abundance, we identified isoform markers

for the classes, subclasses, and clusters in the MOp ontology that are

differential regardless of gene expression. We found 5,658 isoforms

belonging to 3,132 genes that are specific to the glutamatergic and

GABAergic classes (Fig.3, Supplementary Table4), 7,588 isoforms

belonging to 4,171 genes within the glutamatergic class and 4,359 iso-

forms belonging to 2,614 genes within the GABAergic class exhibiting

isoform shifts specific to subclasses (Supplementary Table5a, b), and

for the 48 clusters passing filter, 3,171 isoforms belonging to 2,461 genes

exhibiting isoform shifts in clusters (Supplementary Table6). Together,

these form an isoform catalogue of the MOp (Supplementary Fig.5a, b).

Spatial isoform specificity

Spatial scRNA-seq methods are not currently well suited to directly

probing isoforms of genes owing to the number and lengths of probes

required—however, spatial analysis at the gene level can be refined

to yield isoform-level results by extrapolating SMART-seq isoform

quantifications (Fig.3, Supplementary Fig.5c).

Figure4a, b shows an example of a gene, Pvalb, for which the

SMART-seq quantification reveals that of the two isoforms of the gene,

only one, Pvalb-201, is expressed. Moreover, thiseffect is specific to

the Pvalb cell subclass (Fig.3). In an examination of MERFISH spa-

tial single-cell RNA-seq derived from 64 slices from the MOp region

(Extended Data Fig.6a), the Pvalb subclass, for which Pvalb is a marker,

can be seen to be dispersed throughout the motor cortex spanning all

layers (Extended Data Fig.6b). While MERFISH probes only measure

abundance of Pvalb at the gene level (Fig.4c), extrapolation from the

SMART-seq quantifications can be used to refine the MERFISH result to

reveal the spatial expression pattern of the Pvalb-201 isoform.

This extrapolation can be done systematically. To build a spatial

isoform atlas of the MOp, we identified differentially expressed genes

from the MERFISH data (Supplementary Table8a, b) and for each of

them checked whether there were SMART-seq isoform markers (from

Supplementary Table5a, b). An example of the result is shown in Fig.3,

which displays one gene for each cluster, together with the isoform

label specific to that cluster and the spatial location of the specific

cluster within a slice of the mouse MOp.

SMART-Seq10xv3 MERFISH

Reads

15,229,289,828 22,696,236,495 NA

Cells

6,295 94,162 280,327

Genes

10,333 5,891 72

Isoforms

20,319 NA NA

Classes

443

Subclasses

18 22 25

Clusters

62 14799

Cluster ClassSubclass

Average detected per cell

Differential isoform

log(TPM + 1)

Lamp5: 360

Sncg: 77

Vip: 607

Sst: 406

Pvalb: 521

L5 IT: 1,561

L5/6 NP: 207

L5 PT: 12

L2/3 IT: 482

L6b: 570

VLMC: 6

L6 IT: 395

L6 CT: 904

L6 IT Car3: 5

Endo: 7

Astro: 9

SMC: 19

Low quality : 12

GlutamatergicGABAergic

Oxr1-204 Oxr1

Snap25-202 Snap25

L2/3 IT L5 PT L5/6 NP L6 CT L6 IT L6 IT Car3 L6b

Stxbp2-207

Stxbp2

L6b

Kcnip1

L6b

Kcnip1

L6b Ror1_1

L6b Ror1_2

L6b Rprm

L6b Shisa6

GABAergic

Glutamatergic

Fig. 2 | Isoform specificity in the absence of gene specificity. a, Overview of

the data analysed. The clustering method used by the BICCN consortium

generates three hierarchies of cells: classes, subclasses and clusters. NA, not

applicable. b, A t-SNE of 10 neighbourhood components of 6,160 SMART-seq

cells coloured according to subclass. Astro, astrocytes; CT, corticothalamic;

endo, endothelial; IT, intratelencephalic; NN, non-neuronal; NP,

near-projecting; PT, pyramidal tract; SMC, smooth muscle cells; VLMC,vascular

lepotomeningeal cells. c, Example of a gene with an isoform specific to the

glutamatergic class. The Oxr1-204 isoform distribution in log1p(transcripts per

million (TPM)) across cells (left) superimposed on the t-SNE of the cells. The

cells belonging to the glutamatergic class are circled. The violin plots of

the gene and isoform distributions show that the gene is not differentially

expressed but the isoform is (right). d, Example of a gene with an isoform

specific to the L6b subclass. The Snap25-202 isoform distribution across cells

(left) superimposed on the t-SNE of the cells. The cells belonging to the L6b

subclass are circled. The violin plots of the gene and isoform distributions show

that the gene is not differentially expressed but the isoform is (right).

e, Example of a gene with an isoform specific to the L6b Ror1_1 cluster. The

Stxbp2-207 isoform distribution in log1p(TPM) across cells (left) superimposed

on the t-SNE of the cells. The cells belonging to the L6b Ror1_1 cluster are

circled. The violin plots of the gene and isoform distributions show that the

gene is not differentially expressed but the isoform is (right). *P < 0.01 between

the group and its complement. In violin plots, white circles represent the mean

and white bars represent the s.d.

198 | Nature | Vol 598 | 7 October 2021

Article

We hypothesized that the mouse MOp exhibits changes in isoform

expression associated with the physical location of cells

41,42

. To deter-

mine whether there are isoforms that increase or decrease in expression

along the depth of the motor cortex, we first estimated the position

of the various layers in the glutamatergic subclasses (Extended Data

Fig.7a, Methods), performed weighted least-squares regression on

the centroids of the subclasses and inferred isoform expression from

the SMART-seq data. While we find many isoforms that exhibit a sig-

nificant change in expression across the depth (Extended Data Fig.7b,

c, Supplementary Table8c), none of the isoforms that pass our filter

exhibit a monotonic change with respect to the mean. This suggests

that non-linear models may be better suited to study isoform variability

across the depth of the mouse MOp.

While direct measurement of isoform abundance may be possible

with spatial RNA-seq technologies such as SEQFISH

or MERFISH

such resolution would require dozens of probes to be assayed per

gene (Supplementary Fig.6), each of which is typically tens of base

pairs in length. Thus, while isoforms can theoretically be detected

in cases where they contain large stretches of unique sequence, the

technology is currently prohibitive for assaying most isoforms, mak-

ing the extrapolation procedure described here of practical relevance

(Supplementary Table9).

Splicing markers

Isoform quantification of RNA-seq can be used to distinguish shifts in

expression between transcripts that share transcriptional start sites

and shifts due to the use of distinct transcription start sites (TSSs).

Investigating such differences can, in principle, shed light on tran-

scriptional versus post-transcriptional regulation of detected isoform

shifts

43,44

. For example, in the glutamatergic class, Ptk2b (Extended

Data Fig.8a) exhibits differential expression of transcripts between

start sites (Extended Data Fig.8b). This gene is known to be associ-

ated with Alzheimer’s disease and its transcript usage is mediated

by genetic variation

. We find that isoforms sharing the preferential

start site exhibit no discernible difference in expression (Extended

Data Fig.8c), suggesting that the observed differences result from

cell-type-specific transcriptional, rather than post-transcriptional

regulation. We identified 1,971 isoforms from 128 groups of TSSs where

the TSSs are preferentially expressed in either GABAergic, glutamater-

gic or non-neuronal classes, even when the expression of isoforms

contained within the TSS is constant (Supplementary Table10a, c, d).

Such cases are likely to be instances where the TSS shifts between cell

types are a result of differential promoter usage—that is, the result of

a transcriptional program.

We also examined post-transcriptional programs (Supplementary

Table10b), instances where the TSSs are not differential between

classes, but where there are isoform shifts within TSSs between classes.

We find 31 isoforms from 28 TSS groups that are differential between

classes when the TSS group is not. One such example is expression

of isoforms Rtn1-201 and Rtn1-203, which share the same TSS in the

Rtn1 gene. The glutamatergic class exhibits preferential expression

of Rtn1-201, which was previously shown to be expressed in grey mat-

ter

, whereas the GABAergic class does not (Extended Data Fig.9).

These cases are likely to be instances where isoform shifts between

cell types are a result of differential splicing—that is, the result of a

post-transcriptional program.

Spon1

Bcl11b

C1ql3

Olfm3

Syndig1

Sema5a

Nr4a2

Lamp5

Cxcl14

Vip

Sst

Pvalb

Cldn5

MERFISH TPM

100,000

20,000

50,000

100,000

250,000

20,000

250,000

150,000

100,000

25,000

500,000

L5/6 NP (3,482)

L5 PT (6,759)

L2/3 IT (33,757)

L6b (3,487)

L6 IT (12,475)

L6 CT (23,878)

L6 IT Car3 (1,563)

Lamp5 (3,095)

Sncg (477)

Vip (2,789)

Sst (4,853)

Pvalb (7,885)

Endo (18,183)

201

202

201

202

201

Fig. 3 | Isoform atlas. Spatial isoform atlas of the MOp. The scatter plots (left)

show the locations of cells (black dots) in the indicated subclass within a single

representative slice of the mouse MOp as assayed by MERFISH. Right, each

column corresponds to a marker gene in the MERFISH datasetand each row

corresponds to a subclass (number of cells in parentheses) in which one

isoform (labelled on the diagonal) was differential in the SMART-seq dataset for

that subclass. This spatial isoform inference links isoform expression from the

SMART-seq data with the physical locations of cells expressing that isoform

from the MERFISH data. The normalized gene expression values are plotted for

each subclass–gene pair in TPM. The white circles in the violin plots represent

the mean and the white bars represent the s.d.

Pvalb-201 Pvalb-202

y (μm)

x (μm)

Fig. 4 | Spatial extrapolation of isoform expression. a, b, Expression of the

Pvalb-201 (a) and Pvalb-202 (b) isoforms in log1p(TPM) units for each cell

superimposed on the NCA–t-SNE plot, as assayed by SMART-seq. c, Spatial

expression of the Pvalb-201 isoform across 64 slices from the MOp, as

extrapolated from probes for the Pvalb gene assayed by MERFISH. Each cell

represented by MERFISH data is coloured by its expression of Pvalb in

normalized counts.

Nature | Vol 598 | 7 October 2021 | 199

Discussion

Our spatially resolved isoform atlas of the mouse MOp expands on

previously identified gene markers, extending the catalogue to isoform

markers for cell types characterized by the BICCN

. Our approach

leverages distinct strengths of different technologies, using the iso-

form resolution of SMART-seq in conjunction with the complementary

cell depth obtainable with 10x Genomics technology and the spatial

resolution produced with MERFISH to spatially place cell-type isoform

markers. This validated approach, in which we leverage technologies

that are broadly consistent (Extended Data Fig.10) yet complemen-

tary in their strengths, is important because isoform specificity could

helpinexplaining the molecular basis ofmorphological differences.

For example, Pvalb cells observed in hippocampal Pvalb interneu-

rons cannot be distinguished morphologically solely on the basis of

gene-level analysis

9,47,48

Spatial isoform markers also enable more targeted assays for ‘auto-

matic expression histology’and can facilitate investigation ofthe

functional significance of cell-type isoform specificity.Recently devel-

oped experimental methods for this purpose—for example, isoform

screens

—are a promising direction and will be key to understanding

the significance of the vast isoform diversity in the brain

Online content

Any methods, additional references, Nature Research reporting sum-

maries, source data, extended data, supplementary information,

acknowledgements, peer review information; details of author contri-

butions and competing interests; and statements of data and code avail-

ability are available at https://doi.org/10.1038/s41586-021-03969-3.

1. Picelli, S. etal. Full-length RNA-seq from single cells using Smart-seq2. Nat. Protoc. 9,

171–181 (2014).

2. Chen, K. H., Boettiger, A. N., Mofitt, J. R., Wang, S. & Zhuang, X. Spatially resolved, highly

multiplexed RNA proiling in single cells. Science 348, aaa6090 (2015).

3. Zheng, G. X. Y. etal. Massively parallel digital transcriptional proiling of single cells.

Nat. Commun. 8, 14049 (2017).

4. Weyn-Vanhentenryck, S. M. etal. Precise temporal regulation of alternative splicing

during neural development. Nat. Commun. 9, 2189 (2018).

5. Walker, R. L. etal. Genetic control of gene expression and splicing in the developing

human brain. Preprint at https://doi.org/10.1101/471193 (2018).

6. Porter, R. S., Jaamour, F. & Iwase, S. Neuron-speciic alternative splicing of transcriptional

machineries: implications for neurodevelopmental disorders. Mol. Cell. Neurosci. 87,

35–45 (2018).

7. Lukacsovich, D. etal. Single-cell RNA-seq reveals developmental origins and ontogenetic

stability of neurexin alternative splicing proiles. Cell Rep. 27, 3752-3759.e4 (2019).

8. Song, Y. etal. Single-cell alternative splicing analysis with expedition reveals splicing

dynamics during neuron differentiation. Mol. Cell 67, 148-161.e5 (2017).

9. Que, L., Winterer, J. & Földy, C. Deep survey of GABAergic interneurons: emerging

insights from gene-isoform transcriptomics. Front. Mol. Neurosci. 12, 115 (2019).

10. Huang, C.-C., Lin, Y.-S., Lee, C.-C. & Hsu, K.-S. Cell type-speciic expression of eps8 in the

mouse hippocampus. BMC Neurosci. 15, 26 (2014).

11. Zhang, Y. etal. An RNA-sequencing transcriptome and splicing database of glia, neurons,

and vascular cells of the cerebral cortex. J. Neurosci. 34, 11929–11947 (2014).

12. Sugino, K. etal. Mapping the transcriptional diversity of genetically and anatomically

deined cell populations in the mouse brain. eLife 8, e38619 (2019).

13. Bittar, P. G., Charnay, Y., Pellerin, L., Bouras, C. & Magistretti, P. J. Selective distribution of

lactate dehydrogenase isoenzymes in neurons and astrocytes of human brain. J. Cereb.

Blood Flow Metab. 16, 1079–1089 (1996).

14. Tasic, B. etal. Adult mouse cortical cell taxonomy revealed by single cell transcriptomics.

Nat. Neurosci. 19, 335–346 (2016).

15. Gandal, M. J. etal. Transcriptome-wide isoform-level dysregulation in ASD, schizophrenia,

and bipolar disorder. Science 362, eaat8127 (2018).

16. Petri, S. etal. The mRNA expression of AMPA type glutamate receptors in the primary

motor cortex of patients with amyotrophic lateral sclerosis: an insitu hybridization study.

Neurosci. Lett. 360, 170–174 (2004).

17. Macosko, E. Z. etal. Highly parallel genome-wide expression proiling of individual cells

using nanoliter droplets. Cell 161, 1202–1214 (2015).

18. Klein, A. M. etal. Droplet barcoding for single-cell transcriptomics applied to embryonic

stem cells. Cell 161, 1187–1201 (2015).

19. Ramsköld, D. etal. Full-length mRNA-seq from single-cell levels of RNA and individual

circulating tumor cells. Nat. Biotechnol. 30, 777–782 (2012).

20. Arzalluz-Luque, Á. & Conesa, A. Single-cell RNAseq for the study of isoforms-how is that

possible? Genome Biol. 19, 110 (2018).

21. Seirup, M. etal. Reproducibility across single-cell RNA-seq protocols for spatial ordering

analysis. PLoS ONE 15, e0239711 (2020).

22. Yao, Z. etal. A transcriptomic and epigenomic cell atlas of the mouse primary motor

cortex. Nature https://doi.org/10.1038/s41586-021-03500-8 (2021).

23. Wang, X., He, Y., Zhang, Q., Ren, X. & Zhang, Z. Direct comparative analyses of 10x

Genomics Chromium and Smart-seq2. Genomics Proteomics Bioinfomatics https://doi.org/

10.1016/j.gpb.2020.02.005 (2021).

24. Bray, N. L., Pimentel, H., Melsted, P. & Pachter, L. Near-optimal probabilistic RNA-seq

quantiication. Nat. Biotechnol. 34, 525–527 (2016).

25. Zhang, M. etal. Spatially resolved cell atlas of the mouse primary motor cortex by

MERFISH. Nature https://doi.org/10.1038/s41586-021-03705-x (2021).

26. Maaten, L. V. D. & Hinton, G. Visualizing data using t-SNE. J. Mach. Learn. Res. 9,

2579−2605 (2008).

27. Becht, E. etal. Dimensionality reduction for visualizing single-cell data using UMAP.

Nat. Biotechnol. 37, 38–44 (2018).

28. Oliver, P. L. etal. Oxr1 is essential for protection against oxidative stress-induced

neurodegeneration. PLoS Genet. 7, e1002338 (2011).

29. Volkert, M. R. Preventing neurodegeneration by controlling oxidative stress: the role of

OXR1. Front. Neurosci. 14, 611904 (2020).

30. Wu, Y., Davies, K. E. & Oliver, P. L. The antioxidant protein Oxr1 inluences aspects of

mitochondrial morphology. Free Radic. Biol. Med. 95, 255–267 (2016).

31. Tasic, B. etal. Shared and distinct transcriptomic cell types across neocortical areas.

Nature 563, 72–78 (2018).

32. Ntranos, V., Yi, L., Melsted, P. & Pachter, L. A discriminative learning approach to

differential expression analysis for single-cell RNA-seq. Nat. Methods 16, 163–166

(2019).

33. Sarantopoulou, D. etal. Comparative evaluation of full-length isoform quantiication from

RNA-seq. BMC Bioinformatics 22, 266 (2021).

34. Bark, I. C., Hahn, K. M., Ryabinin, A. E. & Wilson, M. C. Differential expression of SNAP-25

protein isoforms during divergent vesicle fusion events of neural development. Proc. Natl

Acad. Sci. USA 92, 1510–1514 (1995).

35. Irfan, M. etal. SNAP-25 isoforms differentially regulate synaptic transmission and

long-term synaptic plasticity at central synapses. Sci. Rep. 9, 6403 (2019).

36. Wallén-Mackenzie, Å. etal. Spatio-molecular domains identiied in the mouse

subthalamic nucleus and neighboring glutamatergic and GABAergic brain structures.

Commun. Biol. 3, 338 (2020).

37. Kim, D.-W. etal. Multimodal analysis of cell types in a hypothalamic node controlling

social behavior. Cell 179, 713-728.e17 (2019).

38. Böckers, T. M. etal. Differential expression and dendritic transcript localization of Shank

family members: identiication of a dendritic targeting element in the 3 untranslated

region of Shank1 mRNA. Mol. Cell. Neurosci. 26, 182–190 (2004).

39. Berkel, S. etal. Sex hormones regulate SHANK expression. Front. Mol. Neurosci. 11, 337

(2018).

40. Zhang, Y., Thompson, R., Zhang, H. & Xu, H. APP processing in Alzheimer’s disease.

Mol. Brain 4, 3 (2011).

41. Rash, B. G. & Grove, E. A. Area and layer patterning in the developing cerebral cortex.

Curr. Opin. Neurobiol. 16, 25–34 (2006).

42. Sansom, S. N. & Livesey, F. J. Gradients in the brain: the control of the development of form

and function in the cerebral cortex. Cold Spring Harb. Perspect. Biol. 1, a002519 (2009).

43. Trapnell, C. etal. Transcript assembly and quantiication by RNA-seq reveals unannotated

transcripts and isoform switching during cell differentiation. Nat. Biotechnol. 28, 511–515

(2010).

44. Yi, L. etal. Gene-level differential analysis at transcript-level resolution. Genome Biol. 19,

53 (2018).

45. Raj, T. etal. Integrative transcriptome analyses of the aging brain implicate altered

splicing in Alzheimer’s disease susceptibility. Nat. Genet. 50, 1584–1592 (2018).

46. Mills, J. D. etal. Unique transcriptome patterns of the white and grey matter corroborate

structural and functional heterogeneity in the human frontal lobe. PLoS ONE 8, e78480

(2013).

47. Klausberger, T. & Somogyi, P. Neuronal diversity and temporal dynamics: the unity of

hippocampal circuit operations. Science 321, 53–57 (2008).

48. Harris, K. D. etal. Classes and continua of hippocampal CA1 inhibitory neurons revealed

by single-cell transcriptomics. PLoS Biol. 16, e2006387 (2018).

49. Thomas, J. D. etal. RNA isoform screens uncover the essentiality and tumor-suppressor

activity of ultraconserved poison exons. Nat. Genet. 52, 84–94 (2020).

50. Karlsson, K. & Linnarsson, S. Single-cell mRNA isoform diversity in the mouse brain.

BMC Genomics 18, 126 (2017).

51. Beerenwinkel, N., Pachter, L. & Sturmfels, B. Epistasis and shapes of itness landscapes.

Stat. Sinica 17, 1317–1342 (2007).

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in

published maps and institutional afiliations.

Open Access This article is licensed under a Creative Commons Attribution

4.0 International License, which permits use, sharing, adaptation, distribution

and reproduction in any medium or format, as long as you give appropriate

credit to the original author(s) and the source, provide a link to the Creative Commons license,

and indicate if changes were made. The images or other third party material in this article are

included in the article’s Creative Commons license, unless indicated otherwise in a credit line

to the material. If material is not included in the article’s Creative Commons license and your

intended use is not permitted by statutory regulation or exceeds the permitted use, you will

need to obtain permission directly from the copyright holder. To view a copy of this license,

visit http://creativecommons.org/licenses/by/4.0/.

Isoform specificity in the mouse primary motor cortex

Figures

Citations

Benchmarking spatial and single-cell transcriptomics integration methods for transcript distribution prediction and cell type deconvolution

What is a cell type and how to define it?

RNA velocity unraveled

Depth normalization for single-cell genomics count data

A single-cell guide to retinal development: Cell fate decisions of multipotent retinal progenitors in scRNA-seq

References

Visualizing Data using t-SNE

Transcript assembly and quantification by RNA-Seq reveals unannotated transcripts and isoform switching during cell differentiation

Near-optimal probabilistic RNA-seq quantification

A robust and high-throughput Cre reporting and characterization system for the whole mouse brain

Genome-wide atlas of gene expression in the adult mouse brain

Related Papers (5)

Analysis and design of RNA sequencing experiments for identifying isoform regulation

Equivalent high-resolution identification of neuronal cell types with single-nucleus and single-cell RNA-sequencing

STAR: ultrafast universal RNA-seq aligner

RNA velocity of single cells

Single-nucleus RNA and ATAC sequencing reveals the impact of chromatin accessibility on gene expression in Arabidopsis roots at the single-cell level.

Frequently Asked Questions (13)

Q1. What contributions have the authors mentioned in the paper "Isoform cell-type specificity in the mouse primary motor cortex" ?

Q2. What was used to reverse transcribe poly(A) RNA and amplify?

Q3. How many bins were used to compute the dispersion for each feature?

Q4. What is the SMART-seq method used to measure cell type specificity?

Q5. How many components were used to visualize the NCA data?

Q6. what is the mRNA isoform in the human frontal lobe?

Q7. What is the effect of the MERFISH probes on the expression of Pvalb?

Q8. How many cells were assayed with MERFISH?

Q9. What is the likely explanation for the differences in isoforms between classes?

Q10. What was the GTF and the GRCm38 genome fasta file?

Q11. Why did the authors exclude the L5 IT subclass from their analyses?

Q12. What is the way to detect the isoforms in the merfish data?

Q13. How many isoforms are found in the tsss?