Transcriptomic Diversity of Cell Types Across The

bioRxiv preprint doi: https://doi.org/10.1101/2022.10.12.511898; this version posted October 14, 2022.
The copyright holder for this preprint

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
RESOURCE available under aCC-BY-NC-ND 4.0 International license.
Transcriptomic diversity of cell types

across the adult human brain
Kimberly Siletti1*, Rebecca Hodge2†, Alejandro Mossi Albiach1†, Lijuan Hu1, Ka Wai Lee1, Peter Lönnerberg1, Trygve Bakken2,
Song-Lin Ding2, Michael Clark2, Tamara Casper2, Nick Dee2, Jessica Gloe2, C. Dirk Keene3, Julie Nyhus2, Herman Tung2, Anna
Marie Yanny2, Ernest Arenas1, Ed S. Lein2, Sten Linnarsson1*
1
Karolinska Institute, Stockholm, Sweden 2Allen Institute for Brain Science, Seattle, Washington 3Department of Laboratory Medicine and Pathology, University
of Washington, Seattle, Washington, USA †These authors contributed equally to this work. *Corresponding authors.
The human brain directs a wide range of complex behaviors ranging from fine motor skills to abstract intelligence
and emotion. However, the diversity of cell types that support these skills has not been fully described. Here we used
high-throughput single-nucleus RNA sequencing to systematically survey cells across the entire adult human brain in
three postmortem donors. We sampled over three million nuclei from approximately 100 dissections across the forebrain,
midbrain, and hindbrain. Our analysis identified 461 clusters and 3313 subclusters organized largely according to de-
velopmental origins. We found area-specific cortical neurons, as well as an unexpectedly high diversity of midbrain and
hindbrain neurons. Astrocytes also exhibited regional diversity at multiple scales, comprising subtypes specific to the tel-
encephalon and to more precise anatomical locations. Oligodendrocyte precursors comprised two distinct major types
specific to the telencephalon and to the rest of the brain. Together, these findings demonstrate the unique cellular composi-
tion of the telencephalon with respect to all major brain cell types. As the first single-cell transcriptomic census of the entire
human brain, we provide a resource for understanding the molecular diversity of the human brain in health and disease.
Correspondence to:
K.S. (kimberly.siletti@ki.se / @kimsil)
S.L. (sten.linnarsson@ki.se / @slinnarsson / linnarssonlab.org)
The mammalian brain develops from four ultimately retained 606 high-quality samples covering 106
compartments—the telencephalon, diencephalon, dissections across ten brain regions, including the most
midbrain, and hindbrain—that give rise to the major anterior part of the cervical spinal cord in one donor (Fig.
structures that underlie the remarkable capabilities of the S1A; Table S1). We sampled all but twelve dissections
mature brain. The telencephalon contains the cerebral from three postmortem donors, although dissections
cortex, hippocampus, and cerebral nuclei and is therefore were sometimes combined differently across donors to
best studied. Indeed the cortex and hippocampus are achieve this goal, e.g., PAG became PAG-DR in one donor.
widely known as the centers of cognition and memory, We moreover enriched for neurons using fluorescence-
respectively, and the cerebral nuclei comprise diverse sets activated cell sorting (FACS) and aimed to collect 90%
of neuronal clusters that process fear and emotion (the neurons and 10% non-neuronal cells from each sample.
amygdala), as well as movement- and motivation-related We were not able to achieve this proportion in many non-
information (the basal ganglia). The diencephalon gives cortical regions due to an abundance of oligodendrocytes.
rise to the hypothalamus and thalamus, which maintain Total genes captured per cell therefore also varied across
homeostasis and relay sensory formation, respectively. dissections partly due to biological variation in cell size;
Posteriorly, the midbrain and hindbrain contain many for example, non-neuronal cells are small and yield
nuclei that regulate vital functions like cardiac, respiratory, fewer molecules, e.g., A35r (Fig. S1B). We additionally
and motor functions. The hindbrain comprises the pons, noted variability across donors, likely reflecting tissue
medulla, and cerebellum, the last of which uniquely quality (Fig. S1C). We captured a median 6224 cells from
emerges from an outgrowth of the developing hindbrain each sample and then filtered cells based on their total
called the rhombic lip. In recent years single-cell molecules—as counted by unique molecular identifiers
sequencing has been used to survey the cell-type diversity (UMIs)—and percentages of unspliced RNA, as well as
present across the entire mouse brain, as well as in select doublet scores (Fig. S1D-E).
regions and cell types of the human brain, especially the We then pooled high-quality cells from all
cortex (1–5). Nevertheless, cell-type diversity across most dissections into a single dataset for graph-based clustering
of the expansive human brain remains unexplored. (Methods). Harmony was used to integrate cells across
We therefore finely dissected tissue from across donors (6). The resulting clusters were split into two datasets
the telencephalon, diencephalon, midbrain, and hindbrain based on the dendrogram that largely corresponded to
and performed single-nucleus RNA sequencing (Fig. 1A; neurons and non-neuronal cells, and each dataset was
Methods). We set out to sample 100 anatomically distinct re-clustered. We used the Paris algorithm to build a
locations (hereafter “dissections”) based on the Allen Brain dendrogram from these clusters (7), and then cut the
Atlas, each in three donors and in technical duplicates. We dendrogram to identify “superclusters” that corresponded
bioRxiv preprint doi: https://doi.org/10.1101/2022.10.12.511898; this version posted October 14, 2022. The copyright holder for this preprint
available under aCC-BY-NC-ND 4.0 International license.
A Limbic cortex
106 dissections B Deep-layer NP
3 369 219 cells
ACC A23 A29 A30
Eccentric MSN
TH-TL Parietal cortex Deep-layer CT/6b
A5-7 A40 A43 S1C
Frontal cortex Basal forebrain MSN
A13 A14 A25 A32 CaB GPe GPi NAC Pu
A44 A45 A46 SEP SI Cla
FI M1C
Thalamus Hippocampus
ANC CM CM-Pf CA4 Thalamic
ETH LG LP LP-VPL Misc. excitatory Splatter
MD MD-Re MG Pul
STH VA VLN VPL
Amygdala
excitatory
Hippocampus
CA1-3
Mammillary
Occipital cortex body
A19 Pro V1C V2 Lower
Deep-layer IT rhombic lip
Hippocampus Midbrain-
DG derived inhibitory
Paleocortex Hippocampus CGE interneurons
AON Pir CA1 CA2 CA3 CA4
Temporal cortex DGC DGR DGU Sub
A1C A35 A36 A38 ITG Upper
Idg Ig LEC MEC MTG rhombic lip
STG TF Cerebellum
Hypothalamus Midbrain CBL CBV CbDN Upper-layer IT
HTHma HTHtub HTHso IC PAG PAG-DR LAMP5-LHX6
HTHpo MN PTR RN SC SN & Chandelier
Amygdala Spinal cord OPC
Pons SpC Microglia MGE interneurons
BL BM CEN CMN
CoA La BNST DTg PB PN PnAN
PnEN PnRF Fibroblast
Cerebral cortex Medulla
Hippocampus Oligodendrocyte lar Bergmann
Cerebral nuclei IC PAG PAG-DR
scu
Va
Hypothalamus PTR RN SC SN
Region
Thalamus Ependymal
Midbrain Cerebellar
Pons inhibitory
Cerebellum
Medulla Choroid Astrocyte 31 superclusters
Spinal cord plexus
# cells
Cerebral cortex
Hippocampus
Basal forebrain
Amygdala
Hypothalamus
Thalamus
Midbrain
Pons
Cerebellum
Medulla
Spinal cord
Supercluster
INA
SLC17A6
SLC17A7
SLC32A1 Max
PTPRC
CLDN5
ACTA2
LUM
PDGFRA
SOX10
PLP1
AQP4
FOXJ1 Min
TTR 0
Donors
36
# subclusters
per cluster 0
Σ=3313
0 Clusters 460
Figure 1 | Single-nucleus RNA sequencing reveals transcriptomic diversity across the adult human brain. (A) RNA sequencing was performed
on 106 dissections replicated in three donors. Schematic shows the location of each dissection across the brain. Abbreviations correspond to the
Allen Brain Reference Atlas or cortical Brodmann areas. (B) Neurons (right) and non-neuronal cells (inset left) were analyzed separately, and cells on
each t-SNE are colored by supercluster. Abbreviations: CA = Cornu Ammonis area; CGE = caudal ganglionic eminence; CT = corticothalamic; DG
= dentate gyrus; IT = intratelencephalic; NP = near-projecting; MGE = medial ganglionic eminence; MSN = medium spiny neuron; OPC = oligo-
dendrocyte precursor cell. (C) Each supercluster in (B) was analyzed separately, yielding the 461 clusters arranged in the dendrogram. Attributes are
plotted for each cluster from top to bottom: bars represent the number of cells; color depth represents the proportion of cells that derive from each
brain region; color supercluster is colored as in (B); heatmap shows expression levels for genes that mark neurons (INA), glutamatergic (SLC17A6,
SLC17A7) and GABAergic (SLC32A1) neurons, immune cells (PTPRC), vascular cells (CLDN5, ACTA2), fibroblasts (LUM), OPCs (PDGFRA,
SOX10), oligodendrocytes (SOX10, PLP1), astrocytes (AQP4), ependymal cells (FOXJ1), and choroid plexus (FOXJ1, TTR); color depth represents
the proportion of cells that derive from each donor; bars represent the number of subclusters produced by analyzing the cluster separately.
Siletti et al. Transcriptomic diversity of cell types across the adult human brain 2
Region
A
Deep-layer IT
Spinal cord
Medulla
Cerebellum
Pons
Midbrain
Thalamus
Hypothalamus
Cerebral nuclei
Hippocampus
Cerebral cortex
Deep-layer NP
Upper-layer IT
MGE interneuron
Amygdala excitatory
CGE interneuron
LAMP5-LHX6 and Chandelier
Miscellaneous
Deep-layer CT/6b
Hippocampus CA1-3
MSN
Eccentric MSN
Hippcampus DG
Hippocampus CA4
Thalamic excitatory
Midbrain-derived inhibitory
Splatter
Mammillary body
Lower rhombic lip
Upper rhombic lip
Cerebellar inhibitory
A13
A14
A19
A1C
A23
A25
A29-A30
A32
A35-A36
A38
A40
A43
A44-A45
A46
A5-A7
ACC
AON
FI
ITG
Idg
Ig
LEC
M1C
MEC
MTG
Pir
Pro
S1C
STG
TF
TH-TL
V1C
V2
CA1
CA1-2R
CA1C-CA3C
CA1R-CA2R-CA3R
CA1U-CA2U-CA3U
CA2U-CA3U
CA3R
CA4C-DGC
DGR-CA4
DGR-CA4Rpy
DGU-CA4Upy
Sub
Cla
CaB
GPe
GPi
NAC
Pu
SEP
SI
BL
BM
CEN
CMN
CoA
La
BNST
HTHma
HTHma-HTHtub
HTHpo
HTHpo-HTHso
HTHso
HTHso-HTHtub
HTHtub
MN
ANC
CM
ETH
LG
LP
LP-VPL
MD
MD-Re
MG
Pul
STH
VA
VLN
VPL
IC
PAG
PAG-DR
PTR
RN
SC
SN
SN-RN
CBL
CbDN
DTg
PB
PN
PnAN
PnEN
PnRF
IO
MoAN
MoRF-MoEN
MoSR
SpC
CBV
CM-Pf
D
A29-A30
B V1C
S1C
VIPR2 ANXA2 GUCA1C
TRPC3 SMIM32 PRELP SLC15A5 ITGA11
VWA2 VWA2 V2 A23
LAMA2 STG
Upper IT DPP4 Deep IT Deep NP Deep CT/6b
A5-A7 A19
Cerebral cortex
1,305,075 cells
M1C
Up
pe
GABA Pro
r-l
A40
133
134
128
138
135
122
123
124
125
120
121
129
130
131
126
127
142
143
144
141
140
139
137
145
146
147
148
149
152
151
150
83
91
90
94
93
92
95
96
86
87
88
89
85
97
104
103
99
98
106
105
101
102
112
100
110
109
108
107
Glutamate
a
ye
A1C
rI
ERG A13 A35r FI Pro A46 A43 TF
Tn
SCN5A A14 A38 ITG S1C ITG
FAM9B
eu
A19 A40 Idg STG A35-A36
A1C A43 Ig TF A44-A45 MTG
ro
Amygdala excitatory Miscellaneous LAMP5-LHX6 & Chandelier Ig
LEC TH-TL
n
A23 A44-A45
s
A25 A46 M1C V1C A13 TH-TL
A29-A30 A5-A7 MEC V2
A32 ACC MTG Non-cortical
113
114
117
116
178
177
132
404
164
165
166
168
167
2
1
Pir
419
407
406
405
172
156
157
158
161
159
160
154
174
175
267
269
270
273
271
272
274
275
264
265
266
A35-A36 AON
A34
C
Pir
NPY 0.6 Inhibitory
WNT16 PAWR NPR3 CORT A14 Idg
GLP1R TLL2 VIP Excitatory A25 A38
MGE interneurons CGE interneurons Splatter Density
AON
0
LEC
-10 0 6
313
428
238
237
235
409
403
289
290
291
292
296
295
294
293
285
284
283
282
279
280
281
288
287
286
278
277
276
258
257
256
255
240
260
259
236
262
261
263
243
241
242
249
250
239
248
252
251
253
254
246
245
244
(
Log2 Observed
Expected ) ACC FI
MEC
Figure 2 | Superclusters and clusters are differentially distributed across brain regions. (A) Dot plot represents supercluster distributions across
dissections. Within each column, dot size is proportional to the percentage of cells from the indicated dissection that belong to each supercluster.
Dot color indicates region according to the legend. (B) Stacked bar plots show the relative contributions of neurons from each cortical dissection,
colored according to the legend. Only clusters larger than 100 cells with at least 1% contribution from the cerebral cortex are shown. A Uniform
Manifold Approximation and Projection (UMAP) embedding is also shown for select clusters, where non-cortical cells are colored grey. Clusters
are labeled with two of their most enriched genes. Red labels indicate clusters expressing inhibitory GABAergic vesicular transporter SLC32A1; blue
clusters express excitatory glutamatergic vesicular transporter SLC17A6 or SLC17A7. (C) Histograms show the observed number of cells divided by
the expected number of cells (given the number of cells sampled overall and for each region), for each cluster and dissection. (D) Neighbor graph
represents upper-layer IT cells from all cortical dissections. Node size is proportional to the number of cells. The weight of an edge between two dis-
sections is proportional to the percentage of cells in the dissections that are neighbors.
to islands on a two-dimensional representation calculated and hindbrain neurons. We additionally analyzed each
by t-distributed stochastic neighbor embedding (t-SNE, cluster separately to produce a more detailed but less
Fig. 1B). Neurons and non-neuronal cells were divided robust clustering that contained 3313 clusters, which we
into 21 and 10 superclusters, respectively. Each was subsequently refer to as “subclusters” (Fig. 1C).
manually annotated based on the literature and their The distributions of superclusters across
regional composition (Table S2) with one exception, a dissections exposed three experimental caveats of our
supercluster that contained neurons from most brain work (Fig. 2A). First, dissections were not always replicable
regions. In the tradition of naming neurons for their across donors, partly because their anatomical borders
shapes, we named them splatter neurons based on their were challenging to distinguish. For example, although
appearance on the two-dimensional embedding. Each telencephalic superclusters contained some hypothalamus
supercluster was separately reanalyzed, resulting in a final and thalamus cells, we suspect that these cells derived
set of 461 clusters. A dendrogram of these clusters mirrored from telencephalic tissue bordering these regions. Second,
previous work in the mouse: neurons split separately from we sequenced multiple dissections together at once, and
non-neuronal cells and into two clades (Fig. 1C) (1). One sequencing reads were occasionally misassigned across
clade contained telencephalic excitatory neurons that samples (8). Although we implemented a preprocessing
express glutamatergic genes like SLC17A6 and SLC17A7. pipeline that computationally filtered these “index-
The other contained telencephalic inhibitory neurons hopping” reads (Methods), the filter was likely imperfect.
that express GABAergic genes like SLC32A1, as well as Some cells from the globus pallidus (GPi) clustered with
all excitatory and inhibitory diencephalic, midbrain, hindbrain neurons (Fig. 2A) but contained fewer UMIs
than similar cells. These cells also derived from a sample contained putative Layer-4 clusters (Fig. 2B; #133, 138,
sequenced together with a hindbrain dissection, implying 141). Layer-5 extratelencephalic neurons (ET) were found
read misassignment. Third, our computational approach in the miscellaneous supercluster (Fig. 2B; #113-114, 116-
to determine superclusters left some rare cell types poorly 118). One Layer-5 ET cluster—found mostly in frontal
connected on the graph. These cells were dissimilar to insula and anterior cingulate cortex—expressed von
one another but ended up in a single supercluster called Economo markers like BMP3 and ITGA4 (Fig. 2B; #117)
miscellaneous. For example, this supercluster contained (13). Although we also observed splatter neurons in the
lymphocytes that were distinct from other cell types in the cortex, most derived from the anterior olfactory nucleus
brain. (AON), a dissection that likely contained some basal-
Nevertheless, supercluster distributions revealed ganglia tissue. Only one splatter cluster contained cells
large-scale similarities among adult neurons that reflected from across the cortex and might represent SST-CHODL
their developmental history (Fig. 2A). Superclusters long-range projecting neurons, expressing SST, NPY, and
were largely exclusive to either telencephalic or non- high levels of NOS1 (Fig. 2B; #235) (14).
telencephalic dissections, and some were additionally Strikingly, layer-specific excitatory neurons
specific to regions like the hippocampus, cerebral nuclei, varied more across dissections than interneurons (Fig.
thalamus, or hypothalamus. Additionally, the presence of 2C). Most interneuron clusters contained cells from
three superclusters in multiple regions reflected migration all cortical dissections, and the dissections were nearly
during development. First, cortical interneurons migrate indistinguishable on a Uniform Manifold Approximation
from the medial and caudal ganglionic eminences and Projection (UMAP) embedding of these clusters. In
(MGE and CGE, respectively) during development. contrast, some excitatory clusters were enriched in specific
Correspondingly, MGE and CGE interneurons distributed dissections, such as visual cortex (V1C), Brodmann areas
across the telencephalon. Our data suggested that these 5-7 (A5-A7), frontal insula (FI), anterior cingulate cortex
cells migrate more widely than appreciated; hypothalamus, (ACC), or entorhinal cortex (A35-A36, MEC, and LEC).
thalamus, and midbrain contained cells from this These specializations occurred in different superclusters:
supercluster. This observation might reflect imprecise whereas ACC cells were enriched in one deep-layer NP
dissection or transcriptomic convergence during cluster (#85), A5-A7 cells were enriched in two upper-
development; some of the same transcription factors like layer IT and one miscellaneous Layer-5 ET cluster (#128,
LHX6 generate inhibitory neurons in the hypothalamus 131, 113). V1C exhibited distinctive deep-layer CT/6b
(9). However, neuronal migration from the ganglionic and upper-layer IT clusters (#107, 133), the latter of
eminences into the thalamus occurs specifically in human which we annotated as V1C’s distinctive TRPC3+ Layer
tissue, raising the possibility that some of these cells have 4 (15). Entorhinal and piriform cortex (Pir) not only
indeed migrated (10). Second, a supercluster of inhibitory contained distinctive deep-layer IT and deep-layer CT/6b
neurons was distributed mainly in the thalamus and clusters, but also neurons that clustered with amygdala
midbrain, likely corresponding to the SOX14+ neurons excitatory cells (Fig. 2B). These regions also contained
that migrate from the embryonic midbrain to populate the distinct interneuron clusters (Fig. 2B; #236, 246, 276).
thalamus (11). Our data suggested that some of these cells Notably, entorhinal and piriform cortex is considered
also migrate to the hypothalamus and pons; accordingly, part of the more evolutionarily conserved allocortex
we localized marker genes for these cells in the pons with that contains only three cellular layers. To more broadly
RNAScope in situ hybridization (Fig. S2). Third, two assess transcriptomic variation across different cortical
superclusters appeared to derive from the rhombic lip, dissections, we visualized cortical upper-layer IT neurons
which produces not only the cerebellum but also neurons as a graph where each node represented a dissection, and
that migrate to specific nuclei in the pons and medulla each edge represented the percentage of nearest neighbors
(12). One of these superclusters contained cerebellar shared between two dissections (Fig. 2D). Cortical
granule neurons that derive from the upper rhombic lip. dissections ranged transcriptomically between entorhinal
The second contained neurons from the pontine nucleus and visual cortex, and dissections from nearby anatomical
that derive from the lower rhombic lip. locations often neighbored one another, revealing a
We used supercluster and cluster labels to explore transcriptional gradient across the cortex.
the cellular composition of each brain region. Most of Other telencephalic regions exhibited both cortex-
the cortex is organized in six layers that contain diverse related and region-specific superclusters. For example, the
interneurons and excitatory neurons that project within hippocampus contained MGE and CGE interneurons in
and outside the telencephalon. Most cortical cells were addition to three superclusters related to its anatomical
accordingly MGE- and CGE-derived interneurons or subdivisions: dentate gyrus (DG) and cornu ammonis
layer-specific excitatory neurons: intratelencephalic subfields 1-4 (CA1-3, CA4) (Fig. 2A; Fig. S3A). We also
(upper-layer and deep-layer IT), near-projecting (deep- observed some layer-specific excitatory neurons that
layer NP), and corticothalamic (deep-layer CT/6b) (Fig. mostly derived from the cortex-adjacent subiculum. The
2A). Although cortical layers two through four are cerebral nuclei were more heterogeneous in accordance
commonly considered “upper” and layers five and six with their diverse neuronal circuitry. Claustrum contained
“deep,” both upper-layer IT and deep-layer IT superclusters mostly cortical superclusters, whereas the basal ganglia—
Figure 3 | Splatter neurons are highly complex

A Splatter neurons B D
25
5
0
SLC17A6
SLC32A1 Splatterand likely undersampled. (A) Splatter neurons are
Estimated number of components
Explained variance ratio (Component 1/ Component 2)

Deep-layer IT
92 clusters Upper-layer ITcolored on the t-SNE by cluster. (B-C) Cells are
Amygdala excitatory
291,883 cells Miscellaneous
MGE interneurons colored by expression levels of the gene they express
CGE interneurons
Thalamic excitatory
Deep-layer CT/6b most highly among those in the legend. Darker
MSN
Upper rhombic lip
Cerebellar inhibitory
color indicates higher expression; gray indicates no
Hippocampus CA1-3
LAMP5-LHX6 and Chandelier expression. (C) Arrows indicate clusters with high
Hippocampus DG
AVP and SLC6A5 expression, likely true vasopres-
C AVP Hippocampus CA4
Eccentric MSN
SLC5A7
SLC6A5
Deep-layer NP
Midbrain-derived inhibitory
sin-releasing and cholinergic clusters, respectively.
Mammillary body
Lower rhombic lip (D) For each supercluster, the bar represents the
Oligodendrocyte
optimal number of principal components to
Microglia
Astrocyte
OPC
Vasculardescribe the dataset (left) or the ratio between the
Bergmann
COP
Fibroblast
fractions of variance that the first two principal
Choroid plexus
Ependymal components explain (right). Blue bars indicate su-
Serotonergic
perclusters downsampled to 50,000 cells for
E Parabrachial nuclei (Pons) analysis; grey superclusters contain fewer than
F147 subclusters G H I H18.30.001 50,000 cells. Purple bar denotes a subset of splatter
70000
LMX1A
NR4A2
1000
H19.30.001
H19.30.002 cells, serotonergic neurons. (E) For each superclus-
PHOX2B
Number of subclusters
10 Cortical neurons
LHX1
Median cluster size
PB non-splatter
PAX5
PB splatter ter, total clusters is plotted against total subclusters
Cell density
(left) or mean cluster size (right). Splatter is orange.
35000
500
(F) Splatter cells from the parabrachial nuclei are

0
colored by cluster on the t-SNE. (G) Same as (B). (H)
0 100 0 100
0.25 0.50 0.75 For cortical neurons, parabrachial splatter neurons,
Distance to
Number of clusters furthest neighbor and all other parabrachial neurons, histograms
summarize the distributions of Euclidean distances
J 14 subclusters K MoRF-MoEN L
Serotonergic neurons
Subcluster
Region between cells and their 25th nearest neighbors.
IO PAG IC FEV
SLC6A4
Distances are normalized by the dataset’s maximum
PnRF HMX3
EN1 distance. (I) Cells are colored by donor. (J) Seroton-
MoSR DTg
SLC17A8 ergic neurons are colored by cluster on the t-SNE.
HOXB2
SN-RN
PnEN PB GAD1
GAD2 (K) A graph represents the dissections that
HTHma
SLC32A1
MET constitute 95% of the serotonergic neurons shown
HTHma-HTHtub
TACR3
P2RY1 in (J). Node size is proportional to the number of
0 Min Max CALB1 CALB1
cells from each dissection. The weight of an edge
PAX5 CALB1 between two dissections is proportional to the
M 8 subclusters N O P SEMA3D, GEM
Dopaminergic neurons
SOX6 VIP CALB1

RN CALB1
CALB1
percentage of cells in the dissections that are
HTHma GAD2 PRLHR GCGR
CALB1 CALB1 neighbors. (L) A heatmap represents expression
NEUROD6, PPP1R17 NPW GAD2
SN
SOX6 EBF2 levels across serotonergic cells for select genes. Top
SN-RN
PART1, GFRA2 doublets
bar indicates the subcluster for each cell, colored as
SOX6
AGTR1 in panel (J). The second bar indicates the region of
PAG SOX6 origin for each cell, colored as in panel (K). (M) Do-
PAG-DR LPL
CALB1 CALB1 GAD2 paminergic neurons are colored by subcluster on
CRYM, CALCR CERKL CALCRL
the t-SNE. (N) Same as (K) but for dopaminergic
neurons. (O) Same as (B). (P) Subclusters in (M) were split and further reanalyzed to produce finer subtypes. Cells are colored by subtype.
caudate, putamen, nucleus accumbens, and the external appeared particularly distinct and contained only MSNs
and internal segments of the globus pallidus (CaB, and splatter neurons. Indeed most amygdaloid splatter
Pu, NAc, GPe, Gpi)—contained mostly medium spiny neurons derived from BNST.
neurons (MSNs, Fig. 2A). Interestingly, we observed two Outside the telencephalon, the thalamus
MSN superclusters, one of which we putatively annotated contained three major superclusters: a region-specific
as the recently described CASZ1+ “eccentric” MSNs (Fig. thalamic excitatory supercluster, midbrain-derived
S3B-C) (5). MSN clusters were relatively specific to— inhibitory neurons, and splatter neurons, as well as the
although differentially distributed across (Fig. S3A)—the cortical interneurons detailed earlier (Fig. S3A). Similar to
basal ganglia, whereas eccentric MSNs were more evenly the amygdala, thalamic excitatory cells varied more across
distributed throughout the cerebral nuclei (Fig. 2A). Most dissections than midbrain-derived inhibitory clusters
eccentric MSNs expressed only DRD1, but many in the which contained cells from all thalamic dissections.
basal ganglia co-expressed DRD1 and DRD2, suggesting We also observed MSNs, amygdala excitatory neurons,
regional specialization of a more broadly distributed cell and hippocampal DG neurons in the thalamus that we
type (Fig. S3D-E). In addition to MSNs, the amygdala attributed to dissection imprecision. Among all regions
contained cortical superclusters—interneurons and some profiled, the cerebellum was least complex; although
layer-specific excitatory neurons—and its own excitatory we observed some granular-, molecular-, and purkinje-
supercluster. Both MSNs and amygdala excitatory layer interneurons, our samples were dominated by the
neurons were differentially distributed across amygdala highly abundant upper rhombic lip granule neurons (Fig.
dissections, implying that individual nuclei exhibit unique 2A). Finally, the hypothalamus, midbrain, and other
neuronal compositions (Fig. S3A). The bed nucleus of the hindbrain regions contained some abundant, relatively
stria terminalis (BNST), considered extended amygdala, homogeneous neurons—mammillary neurons, midbrain-
derived inhibitory neurons, and rhombic lip-derived inputs like taste, respiration, and thermoregulation. In
neurons, respectively—but most other neurons in these addition to glia, our parabrachial dissections contained
regions were splatter neurons. cells from five neuronal superclusters consistent with
Splatter neurons represented a strikingly their position bordering the midbrain: midbrain-derived
heterogenous group of cells that did not share a highly inhibitory, upper rhombic lip, lower rhombic lip, cerebellar
specific, broadly expressed gene module. 291,883 cells inhibitory, and splatter (Fig. S4K-N). We found 147 splatter
were distributed over 92 clusters, although many cells subclusters with more than five cells in the parabrachial
fell into several large clusters centrally located on the nuclei (Fig. 3F; Fig. S4O). Our analysis captured three
embedding that expressed no specifically enriched genes populations known to be spatially restricted within the
(Fig. 3A). Except for a subset of hypothalamic cells, most murine parabrachial nuclei: FOXP2+ neurons, LMX1B+
splatter neurons did not differ from other neurons in terms neurons, and PAX5+ neurons (16, 17). The former two
of quality metrics like total UMI counts or unspliced RNA populations are more specifically labeled in our dataset by
fractions (Fig. S4A-B). The splatter supercluster uniquely NR4A2 and LMX1A (Fig. 3G). LMX1A+ cells include the
comprised both inhibitory and excitatory cells: 22.1% of two previously described CALCA+ and SATB2+ positive
cells expressed the GABA transporter SLC32A1, 39.3% of populations (Fig. S4P). PHOX2B moreover identifies
cells expressed the predominant glutamatergic transporter populations ventral and medial to the parabrachial nuclei
SLC17A6, and 1.4% of cells co-expressed these two genes in mice (Fig. 3G) (16). Our observations therefore suggest
(Fig. 3B; Fig. S4C). The dataset moreover included broad conservation of parabrachial composition across
cholinergic (SLC5A7+) and glycinergic (SLC6A5+) mice and humans.
neurons, as well as more than half of all hypothalamic However, we observed more GABAergic neurons
neurons, many of which expressed low levels of the than expected in the parabrachial nuclei, which contain
hormone vasopressin (AVP, Fig. 3C; Fig. S4D-E). A gene primarily glutamatergic neurons (Fig. S4Q). Most
ontology analysis of the most variable genes suggested that GABAergic neurons belonged to a centrally located
these cells were differentiated in particular by extracellular cluster that broadly expressed LHX1 (Fig. 3G). Suspecting
matrix- and synapse-related genes (Fig. S4F), reflecting that these neurons originated from neighboring nuclei, we
the diversity of regions observed and neurotransmitters used our complete dataset to identify the most common
expressed in this supercluster. dissection for each splatter subcluster that contained more
To assess the complexity of these cells relative to than 20 parabrachial cells. Although 78 clusters contained
other superclusters, we estimated the number of principal primarily parabrachial cells, seven regions predominated
components necessary to represent each supercluster by the remaining clusters (Fig. S4R-S). Two predominantly
performing PCA on permuted datasets. We then compared hypothalamic clusters contained few UMIs and were
the variance explained by each component in the original therefore likely debris (Fig. S4O,S). Many LHX1+ cells
and permuted datasets (Fig. S4G; Methods). Nearly belonged to clusters populated by cells from the dorsal
twice as many principal components were necessary to tegmental nucleus (DTg), a region of the brain involved
capture splatter neurons than other superclusters (Fig. in spatial navigation which contains many GABAergic
3D). However, six times the variance was explained cells. The LHX1+ neurons might then represent regions
by the first principal component than the second, a bordering the parabrachial nuclei like the lateral DTg.
number more similar to homogeneous superclusters These observations emphasize that neuronal diversity
like upper rhombic lip, oligodendrocytes, and OPCs (Fig. varies considerably even across neighboring nuclei.
3D). Splatter cells therefore exhibit highly variable but Our analysis of parabrachial splatter neurons
relatively uncorrelated expression. To better capture this suggests three broader properties of the highly
heterogeneity, we performed all further analysis using heterogeneous splatter supercluster. First, these neurons
the subcluster labels for these cells (Fig. S4H). Although were more heterogeneous than other superclusters sampled
the median number of cells in each splatter cluster was in the same dissection. Splatter neurons in the parabrachial
smaller than for many other neuronal superclusters, the nuclei exhibited greater distances from their 25th nearest
92 clusters yielded 1145 subclusters, over 900 more than neighbors than did neurons from other superclusters,
any other supercluster (Fig. 3E). In agreement with our whose distances were more like those between cortical
findings across the whole brain, splatter neurons were most neurons (Fig. 3H; Fig. S4N). Secondly, splatter neurons
similar across neighboring anatomical and developmental might not reflect their developmental origins, unlike other
compartments, reflected both by the proportions of superclusters. Many FOXP2+ parabrachial neurons derive
subclusters shared across regions and the distributions of from the rhombic lip (16), but we found that most FOXP2+
cells from each region across the t-SNE (Fig. S4I-J). neurons clustered with splatter neurons—rather than with
Splatter neurons were nevertheless highly complex one of the rhombic-lip derived superclusters—implying
even within individual dissections. We investigated these that some splatter neurons are also rhombic-lip derived.
neurons in the parabrachial nuclei, where we sampled Our parabrachial dissection also contained FOXP2+ lower
a particularly high proportion of splatter neurons. The rhombic lip neurons, but these cells were likely too few in
parabrachial nuclei are located in the pons and form a number to represent all rhombic lip-lineage neurons in
relay center for interoceptive and exteroceptive sensory the parabrachial nuclei. Lastly, our analysis suggested that
the immense complexity of splatter neurons reflects highly ten subtypes and two classes (SOX6+ and CALB1+) of
reproducible cell types. After recalculating the t-SNE dopaminergic neurons (4). Using a classifier trained on
without batch correction, the embedding demonstrated Kamath’s data, we found that seven of our 14 TH+ subtypes
striking integration across donors, particularly across corresponded to eight previously defined subtypes (Fig.
the donors from which most parabrachial neurons were S5P-S; Methods). Our broader sampling allowed the
sampled (Fig. 3I). The fine structure of splatter neurons detection of seven new TH+ subtypes including one in the
was therefore highly reproducible between donors and substantia nigra (SOX6, LPL) and four CALB1+ subtypes
likely reflects true cell-type specialization. (PAX5, NPW, GCCR, VIP) in the periaqueductal grey and
Because many splatter clusters were defined by ventral tegmental area surrounding the red nucleus (Fig.
their neurotransmitter expression, we also investigated 3P; Fig. S5O). Similar to serotonergic neurons, CALB1
subcluster diversity by neurotransmitter type. Serotonergic subtypes in the periaqueductal grey expressed different
neurons are found across the brain stem and generally combinations of neuropeptides like VIP, NPPC, NPW and
express the transcription factor FEV and the serotonin PENK (Fig. S5T). Finally, we found two GAD2+ clusters
reuptake transporter SLC6A4. In our analysis, FEV and (CALCR and EBF2) that defined the molecular signature
SLC6A4 were highly expressed in cluster 397, which of a third class of human TH+ neurons that co-express
comprised 14 subclusters that derived primarily from the GABAergic markers (Fig. S5O,U) and can release GABA
pons, midbrain, medulla, and hypothalamus (Fig. 3J-K; (20). Our results are thus consistent with midbrain TH+
Fig. S5A-I). All cells expressed FEV except a small cluster neurons exhibiting a range of transcriptional profiles from
of 20 PAX6+ cells and some PITX2+ hypothalamus cells dopaminergic to GABAergic phenotypes.
(Fig. 3L; Fig. S5I). As previously described in mice, all To explore the extensive cellular heterogeneity
subclusters could be grouped into rostral- (HMX3+) and we observed in the midbrain and hindbrain, we used
caudal- (HOX+) derived populations, and rostral neurons spatial transcriptomics to characterize gene-expression
could be further split into EN1+ and EN1- populations distributions across three tissue sections A, B, and C (Fig.
(Fig. 3L) (18). Similar to a recent single-cell study of 4; Fig. S6). Enhanced ELectric (EEL) fluorescence in-
mouse serotonergic neurons in the midbrain raphe nuclei situ hybridization (FISH) enabled us to profile 445 genes
(19), our analysis revealed combinatorial expression of in each section and thereby define anatomical locations
glutamatergic and GABAergic genes SLC17A8, GAD1, and based on their cellular composition (15). Dopaminergic
GAD2 in addition to neuropeptides like PENK and TRH cells demarcated the parabrachial pigmented nucleus (PBP,
(Fig. 3L; Fig. S5G-I). We also similarly observed FEV+ cells Fig. 4A: A1), substantia nigra (SN, Fig. 4A: A2,A3), and
with little SLC6A4 expression, which the authors speculate ventral tegmental area (VTA, Fig. 4B: B1,B2). GABAergic
might indicate neurotransmitter plasticity. However, we cells within the substantia nigra further distinguished
additionally observed a substantial HMX3+ population pars compacta and pars reticulata; the former lacked
that expressed neither SLC17A8, GAD1, nor GAD2, as well GABAergic cells. Glutamatergic PVALB+ cells, molecularly
as a population that expressed the GABAergic transporter distinct from the GABAergic PVALB+ neurons scattered
SLC32A1, suggesting co-release of GABA and serotonin. throughout the VTA, occupied the red and pontine
These observations might represent either region- or nuclei (Fig. 4A-B: B3,B13; Fig. S6A) (21). Additionally,
species-specific specializations. Notably, one subcluster noradrenergic cells marked the locus coeruleus (Fig. 4C:
in our analysis appeared to correspond to the distinct C2). Although we profiled some of these locations with
P2RY1+MET+TACR3+ cluster located near the cerebral single-nucleus RNA sequencing, gene detection was too
aqueduct in mice (19). This subcluster contains mostly sparse to systematically localize our clusters. However, we
cells from the periaqueductal gray, suggesting this cell identified two spatial populations that matched RNA-seq
type and its unique electrophysiological and projection clusters. First, a dense cluster of glutamatergic PVALB+
properties are conserved in humans. neurons in the pontine nucleus also expressed HOPX and
We also further investigated splatter cluster 395 therefore corresponded to lower rhombic lip neurons, as
that was enriched in dopaminergic markers TH, SLC6A3, expected (Fig. S6C). Secondly, CALB2+ dopaminergic
and DDC and comprised eight subclusters containing cells in the parabrachial pigmented nucleus co-expressed
mostly midbrain—substantia nigra, red nucleus, and dopaminergic and GABAergic markers, similar to a
periaqueductal grey—and a few hypothalamic neurons (Fig. dopaminergic subtype we sequenced (Fig. S6A).
3M-N; Fig. S5J-N). SOX6, CALB1 and GAD2 expression The EEL data additionally mirrored our RNA-seq
defined three major classes of these neurons, a subset of analysis of splatter neurons more broadly. Although we
which we reanalyzed and further divided into 14 subtypes did not dissect VTA for sequencing, rodent VTA neurons
(Fig. 3O-P; Fig. S5O; Methods). We identified one small resemble splatter neurons in their combinatorial expression
subtype as doublets based on oligodendrocyte-marker of neurotransmitters, neuropeptides, and their receptors
expression, highlighting the challenges of performing (22). Indeed, GABAergic and glutamatergic neurons in the
quality control within this complex supercluster. The pons and VTA combinatorially co-expressed a number of
remaining subtypes revealed greater molecular complexity neuropeptides (Fig 4. B-C: B3-13,C1), and most CALB1+
than previously reported by Kamath et al., who sorted dopaminergic cells in the VTA also expressed GABAergic
NR4A2+ nuclei from the substantia nigra and found genes (Fig. 4B: B1,B2). Some neuronal populations
GABA B C LC Figure 4 | EEL-FISH reveals spatial distributions

TH
A B TPH2 C1 of gene expression in the midbrain and hindbrain.
PVALB
PENK Pons
C
Panels A, B, and C each correspond to a tissue
y
NPY C2
SST
A
NTS
TAC1 B10 5 mm section. Cells are depicted by dots in each panel; the
CARTPT
SPX
ADCYAP1
GABA
TH
CALB2
A1 GABA
TH
CALB2
A2
x
GABA
TH
CALB2
A3 GABA
SST
GHR
A4
color and size of the dot correspond to a gene and
KLHL1 KLHL1 that gene’s expression level, respectively. Dots
50μm
overlap when multiple genes are expressed in the
RN
B5
B12 GABA
TH
B1 GABA
TH
B2 GABA
PVALB
B3 GABA
SST
B4 same cell; multiple genes expressed in a cell at similar
CALB2
CALB1
CALB2
CALB1 levels will not be visible. Some cells are therefore
A4 B9
B1
B8 indicated with an arrowhead and inset in A1 - C2,
where each dot corresponds to a molecule rather
y
GABA B5 GABA B6 GABA B7 GABA B8

B4
B7
B6
PENK NPY PENK
NPY
TAC1
than a cell. Tissue sections include (A) red nucleus
A1
A2 B2 (RN), parabrachial pigmented nuclei (PBP),
PBP SNpr A3
substantia nigra pars reticulata (SNpr), substantia
B11
VTA GABA
PENK
B9 GABA
CARTPT
B10 GABA
SPX
B11 GABA
NTS
B12
nigra pars compacta (SNpc), and pons; (B) ventral
SNpc NPY
B3 SST tegmental area (VTA) and pons; and (C) pons and
locus coeruleus (LC).
GABA B13 GABA B14 GABA C1 TH C2
Pons B14 SLC17A6
PVALB
TPH2
SLC6A4
SLC6A4
TPH2
SLC18A2
SLC6A2
B13 PENK
Pons
2.5 mm
exhibited specific spatial distributions: SST+ GABAergic with their UMAP position and perhaps our observation
cells crossed the VTA axially, whereas NPY+ GABAergic that more genes are upregulated than downregulated
and PENK+ GABAergic cells distributed sagittally (Fig. in these cells (6,174 and 5,128 genes respectively). One
4B, Fig. S6B). Serotonergic neurons scattered along the distinct oligodendrocyte cluster largely derived from a
midline in the VTA and the pons (Fig. 4B: B14, Fig. 4C: single dissection: 88% of the cells in this cluster were from
C1). TAC1+ GABAergic neurons were instead distributed the hypothalamic mammillary nucleus in one donor.
broadly throughout the tissue (Fig. 4B: B8). Notably, Moreover only two of the top 20 enriched genes were
some populations were very rare. A spatially isolated protein coding (MDFIC2, COL10A1), and neither gene
cell situated in the substantia nigra pars reticulata co- was highly specific. This cluster therefore likely resulted
expressed GHR and SST (Fig. 4A: A4). Fewer than three from donor-specific or quality-related effects (Fig.
cells highly expressed peptides like ADCYAP1, CARTPT, S7B-F). The remaining oligodendrocytes were smoothly
NPX, or NTS (Fig 4B: B10-12; Fig. S6B). Our results distributed along the Type 1-to-Type 2 manifold.
therefore imply not only that splatter neurons encompass Because oligodendrocyte cells are relatively
both broadly distributed and spatially restricted cell types, homogenous and therefore not necessarily well
but also that their complexity reflects under-sampling. characterized by clustering analysis, we used Latent
Cells across the midbrain and hindbrain are defined Dirichlet Allocation (LDA) to identify transcriptional
by combinatorial neuropeptide and neurotransmitter modules that varied across these cells. LDA is most
expression, and specific combinations might be observed commonly used to identify groups of words—referred
in few cells that are difficult to sample with single-nucleus to as “topics”—that frequently appear together in text
RNA sequencing from single or few donors. documents. The method analogously reveals co-expressed
We next analyzed the non-neuronal cells in our genes in single-cell data (24). We fit an LDA model for
dataset, which included vasculature, immune cells, and 15 topics (k=15) to the oligodendrocyte data and identified
non-neuronal neuroepithelial cells called glia that support representative genes for each topic (Fig. S7G-L; Methods).
brain function in diverse ways: astrocytes maintain Mitochondrial genes represented Topic 3, indicating low-
brain homeostasis; oligodendrocytes myelinate neurons quality cells that we did not detect with other metrics.
to facilitate electrical impulses. Three superclusters Four other topics peaked successively along the Type
corresponded to the oligodendrocyte lineage: eight 1-to-Type 2 manifold and thus resembled intermediate
oligodendrocyte clusters, six committed oligodendrocyte stages along a maturation axis; correspondingly, these
precursors (COPs), and three oligodendrocyte precursor topics were lead by a topic well-represented by OPALIN
cells (OPCs). One oligodendrocyte, four COP, and one OPC and ended with a topic well-represented by RBFOX1 (Fig.
cluster were likely low quality or doublets and excluded S7I-L). LDA additionally enables the user to score new
from further analysis (#41-43, 36, 75) The remaining cells for a given gene topic. Because we did not observe
clusters were projected onto a single UMAP embedding Rbfox1 expression in mouse oligodendrocytes, we scored
that mirrored lineage differentiation (Fig. 5A). The myelin-forming and mature mouse oligodendrocytes
dendrogram grouped eight oligodendrocyte clusters into for the RBFOX1-correlated Topic 6 (1). A subset of
two major types that expressed either OPALIN or RBFOX1, primarily non-telencephalic oligodendrocytes (MOL3
as previously described (Type 1 and 2, respectively, Fig. 5B- and some MOL2 cells) scored highest for Topic 6 (Fig.
C; Fig. S7A; Table S3) (23). RBFOX1+ oligodendrocytes S7M). Among the genes most correlated with Topic 6 in
are thought to represent a mature end state, consistent mice, VAT1L, NKX2-8 (mouse Nkx2-9), and CD44 were
Figure 5 | Oligodendrocyte-lineage cells differ

A
Oligodendrocyte lineage B Oligodendrocytes between the telencephalon and the rest of the
OPALIN RBFOX1
adjusted p-value
COP brain. (A) Oligodendrocytes, COPs, and OPCs are
200
OLIGO arranged on a UMAP embedding. Cells are grayscale-
colored by cluster. Pie charts summarize each cluster’s
regional composition according to the legend. (B) A
0
Type1 Type2 −5 0 5
log2 fold change dendrogram of oligodendrocyte clusters was cut to
C OPALIN RBFOX1 CD44 Max
identify two transcriptomic types. Type 1 and Type 2
cells are colored on the t-SNE (left). Volcano plot
shows genes differentially expressed between Type 1
and Type 2 with q-value less than 1e-5 and mean
Min expression greater than 0.02 (right). (C) Cells are
OPC NELL1
0
colored by gene expression. (D) Same as (B) but for
OPCs OPCs. (E) Same as (C). (F) Bar heights represent the
PAX3
D LHX2
adjusted p-value
Region
HES1 proportions of Type 1 and Type 2 (bottom and top,
200
IRX5 respectively) oligodendrocytes and OPCs (left and
Spinal cord
Medulla
Cerebellum
Pons
Midbrain
Thalamus
Hypothalamus
Cerebral nuclei
Hippocampus
Cerebral cortex
17 clusters right, respectively) found in each region. Colors

596,090 cells
0
correspond to (A). (G-I) Each dot represents a
Type1 Type2 −5 0
log2 fold change
5
dissection colored by its region of origin as in (A).
F Oligodendrocytes OPCs E HES5 IRX5 GPC5 Max
Dots on the left are telencephalic dissections; dots on
the right are from the rest of the brain. Dots within
Type 1 Type 2
each group are jittered randomly along the x-axis for

visualization. Dot height represents a dissection’s (G)
percentage of cycling OPCs, (H) ratio of COPs to
Min
0 OPCs, or (I) ratio of oligodendrocytes to OPCs. Stars
G Cycling OPCs
H COPs / OPCs I Oligos / OPCs J Oligos / OPCs
(EEL)
indicate a statistically significant difference between
* * * the two groups (Mann-Whitney, p-value < 1e-4). The
0.010
y-axis in (G) was trimmed to 0.01 for visualization;

10
Ratio
Percent
one telencephalic dissection contained 0.04% cycling

20
0.05
Ratio
Ratio
0.005
cells. (J) Bar heights represent the ratios of

0
6 G C Y A s
A4 MT V1 AMN-RN VT Pon oligodendrocytes to OPCs in each tissue section
0.00
S
profiled with EEL.
0
strikingly specific in both species, suggesting these genes components explained most of their variance (Fig. 3D)—
are conserved markers of Type 2 oligodendrocytes (Fig. but are not homogeneous across the brain.
5B; Fig. S7N). Overexpression of the hyaluronan receptor We accordingly further assessed the four OPC
CD44 in oligodendrocytes promotes demyelination, which clusters (Fig. S8C). One Type 2 cluster was 54% mammillary
might signify myelin turnover (25). nucleus, reminiscent of an oligodendrocyte cluster (Fig.
The UMAP and nearest-neighbor graph suggested S8D-F). Although all three donors contributed to this
that the three COP clusters constituted a lineage trajectory cluster—207, 464, and 257 cells each—only FMO3 of the
linking OPCs to mature oligodendrocytes. In contrast, the top 20 enriched genes was protein-coding, consistent with
four OPC clusters were characterized by striking regional a possible quality effect. Clustering within Type 1 cells
specificity rather than maturation state (Fig. 5A; Fig. S7O). appeared to be driven by two expression gradients that
We used the cluster dendrogram to group OPCs into two prevented our analysis pipeline from merging the clusters.
major types and found each enriched within and outside Although the first gradient was correlated with total UMIs
the telencephalon (Type 1 and Type 2, respectively; Fig. per cell, not all genes were uniformly upregulated in larger
5D). These populations likely correspond to the NELL1+ cells. Telencephalic-specific transcription factor FOXG1
cortical and PAX3+ spinal-cord populations described was evenly expressed across Type 1 OPCs, whereas the
in a recent preprint (26). Type 1 and Type 2 OPCs secreted protein NELL1 and cell-adhesion molecule
differentially expressed over 155 genes with a greater CNTN5 increased smoothly (Fig. S8G-J). The second
than three-fold change in either population, including gradient was defined by the proteoglycan GPC5, previously
axon-guidance molecules, region-specific transcription identified as a marker for grey-matter astrocytes (Fig.
factors, and Notch signals (Fig. S8A; Table S4). Although 5E). OPC heterogeneity within the telencephalon might
our previous work identified only one OPC cluster in the therefore reflect interactions with the local environment.
adolescent mouse, experimental differences likely explain Because OPCs were distinct inside and outside
the discrepancy; we sampled fewer cells in mice using a the telencephalon, we examined more closely the
single-cell chemistry with lower sensitivity. We reanalyzed regional distributions of oligodendrocytes. Type 1
the data and found that type-specific markers Hes5 and oligodendrocytes predominated in the telencephalon,
Irx5 were expressed mutually exclusively in mouse OPCs whereas the proportion of Type 2 oligodendrocytes
(Fig. S8B). This observation highlights that OPCs exhibit progressively increased toward the brain’s more posterior
relatively little transcriptomic variation—few principal regions (Fig. 5F). Because Type 1 oligodendrocytes were
Figure 6 | Astrocytes exhibit transcriptomic

A Type 1 Type 2 B GFAP D Cerebral cortex
Max
Hippocampus
Cerebral nuclei Astrocytes
diversity across brain regions. (A) A dendrogram
13 clusters of astrocyte clusters was cut to identify two
Hypothalamus
Region
Thalamus
Midbrain
Pons 155,025 cells transcriptomic types colored on the t-SNE. (B) Cells
Cerebellum
Min
Medulla
Spinal cord
are colored by gene expression. (C) Cells are colored
0
C Cortical astrocytes by expression levels of the gene they express most
Interlaminar WIF1
highly among those in the legend. Darker color
marks LMO2
TNC indicates higher expression; gray indicates no
ocytes, expression. Each gene is enriched in one of the
otoplasmic cortical astrocyte types indicated on the t-SNE. (D)
Grey-matter
Cells are colored by their regions of origin. (E) A
White-matter graph represents astrocyte similarities across the
Brain region White matter Gene topics brain. Most nodes represent brain regions; cerebral
nuclei are broken up into finer regions or dissections.
E Claustrum F Score I Score Score by Region TBX5
5 Node size is proportional to the number of cells
Cerebral cortex Rhinal cortex
from each region or dissection. The weight of an
Topic 4
Amygdala
Hippocampus
SI Basal edge between two nodes is proportional to the
nuclei percentage of cells in the regions or dissections that
Hypothalamus BNST G Score by Type J 0 0.54 FOSB
SEP Globus pallidus 1 5 are neighbors. Abbreviations include: BNST = bed
Topic 9
Thalamus nucleus of the stria terminalis; SEP = septal nuclei; SI

Midbrain 0 = substantia innominata and nearby nuclei. (F) Cells
0 0.52
Pons
H FYB2 K 5
CRYAB are colored by their scores for a gene topic that
Max
Medulla
Topic 11
Cerebellum mirrors GFAP expression and therefore likely reflects

Min
Spinal cord white-matter association. (G) Violin plots

0 summarize score distributions for Type 1 and Type 2
0
0.80
astrocytes colored as in (A). (H) Cells are colored by
gene expression. (I-K) Cells are colored by their scores for each gene topic (left). Histograms summarize the score distributions for each region,
colored as in (C) (middle). Cells are colored by expression of a topic-representative gene (right).
also putatively less mature than Type 2, we wondered neurons and their extracellular environment (Fig. S9A).
whether the differential distributions of OPCs and Two major astrocyte types are present in the mouse brain:
oligodendrocytes along the anterio-posterior axis partly a telencephalic and a non-telencephalic type, each of
reflected different rates of oligodendrocyte turnover inside which additionally contains grey-matter (Gfap-negative)
and outside the telencephalon. Telencephalic OPCs were and white-matter (Gfap-positive) astrocytes. The cortex
indeed more proliferative than non-telencephalic OPCs additionally contains interlaminar astrocytes that are
(Fig. 5G). We therefore investigated the ratios of OPC, specific to Layer 1 (2). Our clustering analysis showed
COP and oligodendrocyte populations across dissections, that this molecular architecture is conserved in human
reasoning these ratios would reflect relative turnover astrocytes. We identified 13 astrocyte subpopulations that
rates. We found an elevated ratio of COPs to OPCs in organized into a telencephalic and non-telencephalic type
the telencephalon, indicating more frequent initiation of (Type 1 and 2, respectively), each containing GFAP low
differentiation (Fig. 5H). The ratio of oligodendrocytes and high populations (Fig. 6A-B; Fig. S9B-D). Type 1
to OPCs was instead higher in the brainstem, indicating included the expected cortical populations: WIF1+ grey-
lower turnover of mature cells relative to the progenitor matter, TNC+ white-matter, and LMO2+ interlaminar
population (Fig. 5I). EEL data confirmed the brain stem’s astrocytes. The latter appeared especially distinct on the
elevated ratio of oligodendrocytes to OPCs, except for a embedding (Fig. 6C).
section that originated from an anterior position in the We observed additional regional heterogeneity
substantia nigra and red nucleus (SN-RN, Fig. 5J). Taken within both Type 1 and Type 2 astrocytes (Fig. 6A).
together, these observations suggest that telencephalic Astrocyte clusters were distributed uniquely across most
OPCs are more proliferative, differentiate more frequently, brain regions, with astrocyte composition most similar
and form a larger population relative to oligodendrocytes. across neighboring developmental and anatomical
Elevated OPC proliferation might therefore more compartments (Fig. S9E-F). The cortex, for example,
frequently replenish the oligodendrocyte population in contained two gray-matter clusters (#52 and #54),
the telencephalon, lowering the resident proportion of one enriched in neocortex and another in allocortex,
more mature Type 2 oligodendrocytes. mirroring differences across cortical neurons. Whereas
Whereas oligodendrocytes likely occupied a the former was mostly specific to the cortex, the latter was
continuum of maturing states, mature astrocytes exhibited also found in the hippocampus and amygdala. The two
more sharply delineated heterogeneity across the brain. other cortical clusters—interlaminar (#58) and fibrous
A GO-enrichment analysis of the most variable genes (#61) astrocytes—were more evenly distributed across
in these cells revealed synaptic-transmission, axon- the entire cortex, perhaps reflecting less specialization
guidance, and extracellular-matrix terms, highlighting for specific neuronal circuitry. Hippocampal astrocytes
that astrocytes are specialized to interact with nearby were found not only in all four cortical clusters, but also
two largely hippocampus-specific clusters (#53 and #57). and well represented by high FOS and JUN expression
Astrocytes in the cerebral nuclei showed especially high (Fig. 6J). FOS is activated downstream of a range of
diversity, reflecting the diversity of neuronal circuitry cellular processes including ischemia, which would agree
this brain region encompasses (Fig. S9G). Whereas most with our observation that globus pallidus and substantia
amygdala astrocytes clustered with allocortical astrocytes nigra dissections contained unusually high numbers
(#54), striatal astrocytes formed their own PENK+ cluster of microglia (Fig. S1B). Pathological conditions might
highly distinct on the t-SNE (#55). Substantia innominata therefore have driven our classification of many globus
cells were most similar to striatum, but the claustrum to pallidus astrocytes as Type 2. We identified two other
cortex and the septal nuclei to the hypothalamus. BNST topics best represented by genes previously characterized
astrocytes clustered with amygdala, hippocampal, and as potential markers for reactive astrocytes (27). Topic 9
striatal cells. Most surprisingly, globus pallidus astrocytes was elevated in all fibrous astrocytes and correlated with
resembled midbrain and thalamus astrocytes and NR4A1 and CRYAB expression (Fig. 6K). Topic 8 was
therefore represented the only major telencephalic Type most elevated in a few tiny groups on the embedding and
2 cells. Other Type 2 cells derived from the diencephalon, correlated with inflammatory-response genes CHI3L1,
midbrain, and hindbrain. Hindbrain astrocytes were SERPINA3, and CCL2 (Fig. S10E). Together, these results
largely distributed across two clusters, one dominated emphasize both the regional heterogeneity of astrocytes
by medulla and spinal cord, and the other containing and the diverse molecular programs with which they react
cerebellar astrocytes. Pontine astrocytes were found in to their environments.
both of these clusters. Importantly, our clustering analysis Here we have presented the first survey of
did not fully discriminate the regional heterogeneity we transcriptomic cell-type diversity across the entire human
observed on the t-SNE. Some midbrain, hypothalamus, and brain. Our study highlights some of the challenges
thalamus cells that co-clustered with pontine astrocytes associated with profiling human tissue: we likely
are therefore likely transcriptionally distinct (compare observed pathological states like the response to ischemia
Fig. 6D and Fig. S9B). Transcriptomic variation among (e.g. astrocytes, Fig. 6), and clusters were not always
grey- and white-matter astrocytes possibly confounded reproducible across donors, partly due to the difficulties of
the analysis, particularly because astrocyte clusters were replicating anatomical dissections (e.g. hypothalamus, Fig.
distinguished mostly by combinatorial gene-expression 2). How cell types vary across donors will be an exciting
gradients rather than binary markers (Fig. S9H). avenue for future research. Our study also highlights a
To distinguish heterogeneity that our clustering remarkable gap in our current understanding of neuronal
analysis failed to capture, we again used LDA to identify diversity. The telencephalon was largely populated by
gene topics differentially distributed across astrocytes. abundant, relatively homogeneous cell types like layer-
First, we fit an LDA model with only two topics (k=2) and specific excitatory neurons, cortical interneurons, and
recovered a topic that mirrored GFAP expression (Fig. 6B,F). MSNs. In contrast, the molecular architecture of splatter
We used this topic to score astrocytes for their likelihood neurons—encompassing many tiny, well-defined clusters
to be white-matter associated. The “white-matter score” and larger highly heterogenous clusters—likely indicates
fell in a gradient across both Type 1 and Type 2 astrocytes, that we are highly under-sampled in regions like the
as well as within individual clusters like interlaminar hypothalamus, midbrain, and hindbrain, which control
astrocytes (#5). Notably, the score was differentially innate behaviors and autonomous physiological functions.
distributed across Type 1 and Type 2 astrocytes (Fig. Our previous work did not uncover equivalent complexity
6F,G), which might suggest that we sampled mostly white- in the mouse, but here we have identified subclusters that
matter astrocytes outside the telencephalon. Consistent are likely conserved across species, suggesting that deeper
with this hypothesis, white-matter scores decreased along sampling will reveal similar neuronal complexity in the
the antero-posterior axis in a trend that mirrored expected mouse. Sampling depth however might not fully explain
myelination levels (Fig. S10A). However, these types also the striking complexity of these neurons; splatter neurons
appeared transcriptionally diverse across regions; FYB2, uniquely comprised both GABAergic and glutamatergic
for example, was highly specific to Type 1 white-matter neurons that did not separate on the embedding, and
astrocytes (Fig. 6F). they expressed neuropeptides and other neuronal genes
We next fit an LDA model with 30 topics (k=30). in unusually complex combinatorial patterns. Our results
Although many topics correlated with individual clusters, additionally underscore how strongly developmental
we also identified gene topics that scored highly within and origins shape adult transcriptomic types. Both astrocytes
across clusters. Topics 5 and 29 appeared to differentiate and OPCs exhibited telencephalic and non-telencephalic
Type 2 grey- and white-matter astrocytes (Fig. S10B). types that transitioned in the hypothalamus (compare
Topic 22 instead revealed an expression gradient of Fig. S7O with Fig. S9E). Some superclusters reflected
growth factor HGF within cortical grey-matter astrocytes migratory cell types that appeared in multiple dissections,
(Fig. S10D), and Topic 4 was best represented by the including some not previously observed. A thorough
transcription factor TBX5, which is highly enriched in understanding of region-specific cellular diversity and its
hippocampus-specific astrocytes (Fig. 6I, clusters #53 and development will therefore be key for treating disease. For
#57). Notably, Topic 9 was elevated in the globus pallidus example, the unique composition of the telencephalon’s
oligodendrocyte lineage might be relevant for diseases like 10. K. Letinic, P. Rakic, Telencephalic origin of human thalamic GABAergic
neurons. Nat Neurosci. 4, 931–936 (2001).
multiple sclerosis. Our work therefore provides a critical
foundation for exploring the diverse neural circuitry 11. P. Jager, Z. Ye, X. Yu, L. Zagoraiou, H.-T. Prekop, J. Partanen, T. M. Jessell,
across the brain and its implications for human health. W. Wisden, S. G. Brickley, A. Delogu, Tectal-derived interneurons contribute
to phasic and tonic inhibition in the visual thalamus. Nat Commun. 7, 13579
(2016).
References 12. C. F. Kratochwil, U. Maheshwari, F. M. Rijli, The Long Journey of Pontine
1. A. Zeisel, H. Hochgerner, P. Lönnerberg, A. Johnsson, F. Memic, J. van Nuclei Neurons: From Rhombic Lip to Cortico-Ponto-Cerebellar Circuitry.
der Zwan, M. Häring, E. Braun, L. E. Borm, G. L. Manno, S. Codeluppi, A. Front Neural Circuit. 11, 33 (2017).
Furlan, K. Lee, N. Skene, K. D. Harris, J. Hjerling-Leffler, E. Arenas, P. Ernfors,
U. Marklund, S. Linnarsson, Molecular Architecture of the Mouse Nervous 13. R. D. Hodge, J. A. Miller, M. Novotny, B. E. Kalmbach, J. T. Ting, T. E.
System. Cell. 174, 999-1014.e22 (2018). Bakken, B. D. Aevermann, E. R. Barkan, M. L. Berkowitz-Cerasano, C. Cobbs,
F. Diez-Fuertes, S.-L. Ding, J. McCorrison, N. J. Schork, S. I. Shehata, K. A.
2. R. D. Hodge, T. E. Bakken, J. A. Miller, K. A. Smith, E. R. Barkan, L. T. Smith, S. M. Sunkin, D. N. Tran, P. Venepally, A. M. Yanny, F. J. Steemers, J.
Graybuck, J. L. Close, B. Long, N. Johansen, O. Penn, Z. Yao, J. Eggermont, W. Phillips, A. Bernard, C. Koch, R. S. Lasken, R. H. Scheuermann, E. S. Lein,
T. Höllt, B. P. Levi, S. I. Shehata, B. Aevermann, A. Beller, D. Bertagnolli, K. Transcriptomic evidence that von Economo neurons are regionally specialized
Brouner, T. Casper, C. Cobbs, R. Dalley, N. Dee, S.-L. Ding, R. G. Ellenbogen, extratelencephalic-projecting excitatory neurons. Nat Commun. 11, 1172
O. Fong, E. Garren, J. Goldy, R. P. Gwinn, D. Hirschstein, C. D. Keene, M. (2020).
Keshk, A. L. Ko, K. Lathia, A. Mahfouz, Z. Maltzer, M. McGraw, T. N. Nguyen,
J. Nyhus, J. G. Ojemann, A. Oldre, S. Parry, S. Reynolds, C. Rimorin, N. V. 14. B. Tasic, Z. Yao, L. T. Graybuck, K. A. Smith, T. N. Nguyen, D. Bertagnolli,
Shapovalova, S. Somasundaram, A. Szafer, E. R. Thomsen, M. Tieu, G. Quon, J. Goldy, E. Garren, M. N. Economo, S. Viswanathan, O. Penn, T. Bakken, V.
R. H. Scheuermann, R. Yuste, S. M. Sunkin, B. Lelieveldt, D. Feng, L. Ng, A. Menon, J. Miller, O. Fong, K. E. Hirokawa, K. Lathia, C. Rimorin, M. Tieu,
Bernard, M. Hawrylycz, J. W. Phillips, B. Tasic, H. Zeng, A. R. Jones, C. Koch, R. Larsen, T. Casper, E. Barkan, M. Kroll, S. Parry, N. V. Shapovalova, D.
E. S. Lein, Conserved cell types with divergent features in human versus mouse Hirschstein, J. Pendergraft, H. A. Sullivan, T. K. Kim, A. Szafer, N. Dee, P.
cortex. Nature. 573, 61–68 (2019). Groblewski, I. Wickersham, A. Cetin, J. A. Harris, B. P. Levi, S. M. Sunkin, L.
Madisen, T. L. Daigle, L. Looger, A. Bernard, J. Phillips, E. Lein, M. Hawrylycz,
3. T. E. Bakken, N. L. Jorstad, Q. Hu, B. B. Lake, W. Tian, B. E. Kalmbach, K. Svoboda, A. R. Jones, C. Koch, H. Zeng, Shared and distinct transcriptomic
M. Crow, R. D. Hodge, F. M. Krienen, S. A. Sorensen, J. Eggermont, Z. Yao, cell types across neocortical areas. Nature. 563, 72–78 (2018).
B. D. Aevermann, A. I. Aldridge, A. Bartlett, D. Bertagnolli, T. Casper, R. G.
Castanon, K. Crichton, T. L. Daigle, R. Dalley, N. Dee, N. Dembrow, D. Diep, 15. L. E. Borm, A. M. Albiach, C. C. A. Mannens, J. Janusauskas, C. Özgün,
S.-L. Ding, W. Dong, R. Fang, S. Fischer, M. Goldman, J. Goldy, L. T. Graybuck, D. Fernández-García, R. Hodge, F. Castillo, C. R. H. Hedin, E. J. Villablanca,
B. R. Herb, X. Hou, J. Kancherla, M. Kroll, K. Lathia, B. van Lew, Y. E. Li, C. S. P. Uhlén, E. S. Lein, S. Codeluppi, S. Linnarsson, Scalable in situ single-cell
Liu, H. Liu, J. D. Lucero, A. Mahurkar, D. McMillen, J. A. Miller, M. Moussa, profiling by electrophoretic capture of mRNA using EEL FISH. Nat Biotechnol,
J. R. Nery, P. R. Nicovich, S.-Y. Niu, J. Orvis, J. K. Osteen, S. Owen, C. R. 1–10 (2022).
Palmer, T. Pham, N. Plongthongkum, O. Poirion, N. M. Reed, C. Rimorin, A. 16. S. Karthik, D. Huang, Y. Delgado, J. J. Laing, L. Peltekian, G. N. Iverson, F.
Rivkin, W. J. Romanow, A. E. Sedeño-Cortés, K. Siletti, S. Somasundaram, J. Grady, R. L. Miller, C. M. McCann, B. Fritzsch, I. Y. Iskusnykh, V. V. Chizhikov,
Sulc, M. Tieu, A. Torkelson, H. Tung, X. Wang, F. Xie, A. M. Yanny, R. Zhang, J. C. Geerling, Molecular ontology of the parabrachial nucleus. J Comp Neurol.
S. A. Ament, M. M. Behrens, H. C. Bravo, J. Chun, A. Dobin, J. Gillis, R. 530, 1658–1699 (2022).
Hertzano, P. R. Hof, T. Höllt, G. D. Horwitz, C. D. Keene, P. V. Kharchenko,
A. L. Ko, B. P. Lelieveldt, C. Luo, E. A. Mukamel, A. Pinto-Duarte, S. Preissl, 17. J. L. Pauli, J. Y. Chen, M. L. Basiri, S. Park, M. E. Carter, E. Sanz,
A. Regev, B. Ren, R. H. Scheuermann, K. Smith, W. J. Spain, O. R. White, C. G. S. McKnight, G. D. Stuber, R. D. Palmiter, Biorxiv, in press,
Koch, M. Hawrylycz, B. Tasic, E. Z. Macosko, S. A. McCarroll, J. T. Ting, H. doi:10.1101/2022.07.13.499944.
Zeng, K. Zhang, G. Feng, J. R. Ecker, S. Linnarsson, E. S. Lein, Comparative
18. C. J. Wylie, T. J. Hendricks, B. Zhang, L. Wang, P. Lu, P. Leahy, S. Fox, H.
cellular analysis of motor cortex in human, marmoset and mouse. Nature. 598,
Maeno, E. S. Deneris, Distinct Transcriptomes Define Rostral and Caudal
111–119 (2021).
Serotonin Neurons. J Neurosci. 30, 670–684 (2010).
4. T. Kamath, A. Abdulraouf, S. J. Burris, J. Langlieb, V. Gazestani, N. M. Nadaf,
19. B. W. Okaty, N. Sturrock, Y. E. Lozoya, Y. Chang, R. A. Senft, K. A. Lyon,
K. Balderrama, C. Vanderburg, E. Z. Macosko, Single-cell genomic profiling of
O. V. Alekseyenko, S. M. Dymecki, A single-cell transcriptomic and anatomic
human dopamine neurons identifies a population that selectively degenerates
atlas of mouse dorsal raphe Pet1 neurons. Elife. 9, e55523 (2020).
in Parkinson’s disease. Nat Neurosci. 25, 588–595 (2022).
20. N. X. Tritsch, A. J. Granger, B. L. Sabatini, Mechanisms and functions of
5. A. Saunders, E. Z. Macosko, A. Wysoker, M. Goldman, F. M. Krienen, H.
GABA co-release. Nat Rev Neurosci. 17, 139–145 (2016).
de Rivera, E. Bien, M. Baum, L. Bortolin, S. Wang, A. Goeva, J. Nemesh, N.
Kamitaki, S. Brumbaugh, D. Kulp, S. A. McCarroll, Molecular Diversity and 21. D. M. Roccaro-Waldmeyer, F. Girard, D. Milani, E. Vannoni, L. Prétôt,
Specializations among the Cells of the Adult Mouse Brain. 174, 1015-1030.e16 D. P. Wolfer, M. R. Celio, Eliminating the VGlut2-Dependent Glutamatergic
(2018). Transmission of Parvalbumin-Expressing Neurons Leads to Deficits in
Locomotion and Vocalization, Decreased Pain Sensitivity, and Increased
6. I. Korsunsky, N. Millard, J. Fan, K. Slowikowski, F. Zhang, K. Wei, Y.
Dominance. Front Behav Neurosci. 12, 146 (2018).
Baglaenko, M. Brenner, P.-R. Loh, S. Raychaudhuri, Fast, sensitive and accurate
integration of single-cell data with Harmony. Nature methods. 16, 1289–1296 22. R. A. Phillips, J. J. Tuscher, S. L. Black, E. Andraka, N. D. Fitzgerald, L.
(2019). Ianov, J. J. Day, An atlas of transcriptionally defined cell populations in the rat
ventral tegmental area. Cell Reports. 39, 110616 (2022).
7. T. Bonald, B. Charpentier, A. Galland, A. Hollocou, Hierarchical Graph
Clustering using Node Pair Sampling. Arxiv (2018). 23. S. Jäkel, E. Agirre, A. M. Falcão, D. van Bruggen, K. W. Lee, I. Knuesel, D.
Malhotra, C. ffrench-Constant, A. Williams, G. Castelo-Branco, Altered human
8. R. Farouni, H. Djambazian, L. E. Ferri, J. Ragoussis, H. S. Najafabadi, Model-
oligodendrocyte heterogeneity in multiple sclerosis. Nature. 566, 543–547
based analysis of sample index hopping reveals its widespread artifacts in
(2019).
multiplexed single-cell RNA-sequencing. Nature communications. 11, 2704–8
(2020). 24. P. Bielecki, S. J. Riesenfeld, J.-C. Hütter, E. T. Triglia, M. S. Kowalczyk, R.
R. Ricardo-Gonzalez, M. Lian, M. C. A. Vesely, L. Kroehling, H. Xu, M. Slyper,
9. D. W. Kim, K. Liu, Z. Q. Wang, Y. S. Zhang, A. Bathini, M. P. Brown, S. H.
C. Muus, L. S. Ludwig, E. Christian, L. Tao, A. J. Kedaigle, H. R. Steach, A. G.
Lin, P. W. Washington, C. Sun, S. Lindtner, B. Lee, H. Wang, T. Shimogori,
York, M. H. Skadow, P. Yaghoubi, D. Dionne, A. Jarret, H. M. McGee, C. B. M.
J. L. R. Rubenstein, S. Blackshaw, Gene regulatory networks controlling
Porter, P. Licona-Limón, W. Bailis, R. Jackson, N. Gagliani, G. Gasteiger, R. M.
differentiation, survival, and diversification of hypothalamic Lhx6-expressing
Locksley, A. Regev, R. A. Flavell, Skin-resident innate lymphoid cells converge
GABAergic neurons. Commun Biology. 4, 95 (2021).
on a pathogenic effector state. Nature. 592, 128–132 (2021).
25. T. M. F. Tuohy, N. Wallingford, Y. Liu, F. H. Chan, T. Rizvi, R. Xing, B. ALTF 1015-2018 to KS

Bebo, M. S. Rao, L. S. Sherman, CD44 overexpression by oligodendrocytes: U.S. National Institutes of Health (NIH) grant U01
A novel mouse model of inflammation‐independent demyelination and
MH114812-02 to SL, ESL
dysmyelination. Glia. 47, 335–345 (2004).
Chan Zuckerberg Initiative and the Silicon Valley
26. L. A. Seeker, N. Bestard-Cuche, S. Jäkel, N.-L. Kazakou, S. M. K. Bøstrand, (NDCN 2018-191929) to EA, SL
A. M. Kilpatrick, D. V. Bruggen, M. Kabbe, F. B. Pohl, Z. Moslehi, N. C. Knut and Alice Wallenberg Foundation (2015.0041,
Henderson, C. A. Vallejos, G. L. Manno, G. Castelo-Branco, A. Williams, 2018.0172, 2018.0220 to SL; 2018.0232 to EA)
Biorxiv, in press, doi:10.1101/2022.03.22.485367.
Swedish Foundation for Strategic Research (SB16-
27. C. Escartin, E. Galea, A. Lakatos, J. P. O’Callaghan, G. C. Petzold, A. 0065) to SL, EA
Serrano-Pozo, C. Steinhäuser, A. Volterra, G. Carmignoto, A. Agarwal, N. Torsten Söderberg Foundation to SL
J. Allen, A. Araque, L. Barbeito, A. Barzilai, D. E. Bergles, G. Bonvento, A. European Union grants: ERC-ADG 884608, H2020
M. Butt, W.-T. Chen, M. Cohen-Salmon, C. Cunningham, B. Deneen, B. D.
Strooper, B. Díaz-Castro, C. Farina, M. Freeman, V. Gallo, J. E. Goldman, S. A.
874758 to EA
Goldman, M. Götz, A. Gutiérrez, P. G. Haydon, D. H. Heiland, E. M. Hol, M. Vetenskapsrådet 2020-01426, Karolinska Institutet
G. Holt, M. Iino, K. V. Kastanenka, H. Kettenmann, B. S. Khakh, S. Koizumi, C. (StratRegen SFO 2018) and Hjärnfonden FO2019-
J. Lee, S. A. Liddelow, B. A. MacVicar, P. Magistretti, A. Messing, A. Mishra, A. 0068to EA.
V. Molofsky, K. K. Murai, C. M. Norris, S. Okada, S. H. R. Oliet, J. F. Oliveira, Author contributions:
A. Panatier, V. Parpura, M. Pekna, M. Pekny, L. Pellerin, G. Perea, B. G.
Pérez-Nievas, F. W. Pfrieger, K. E. Poskanzer, F. J. Quintana, R. M. Ransohoff,
Study design: SL, EL, RDH, TB, SD, KS
M. Riquelme-Perez, S. Robel, C. R. Rose, J. D. Rothstein, N. Rouach, D. H. Tissue dissections and nuclei isolation: RDH, SD,
Rowitch, A. Semyanov, S. Sirko, H. Sontheimer, R. A. Swanson, J. Vitorica, I.-B. MC, TC, ND, JG, CDK, JN, HT, AMY
Wanner, L. B. Wood, J. Wu, B. Zheng, E. R. Zimmer, R. Zorec, M. V. Sofroniew, RNA sequencing: LH, KS
A. Verkhratsky, Reactive astrocyte nomenclature, definitions, and future RNA data preprocessing: PL
directions. Nat Neurosci. 24, 312–325 (2021). RNA data analysis: KS, SL
28. C. S. McGinnis, L. M. Murrow, Z. J. Gartner, DoubletFinder: Doublet RNA data annotation: KS, AMA
Detection in Single-Cell RNA Sequencing Data Using Artificial Nearest EEL experiments and analysis: AMA
Neighbors. Cell Syst. 8, 329-337.e4 (2019). Dopaminergic analysis: KWL, EA
29. Y. Wang, M. Wang, S. Yin, R. Jang, J. Wang, Z. Xue, T. Xu, NeuroPep: a RNAScope: RDH
comprehensive resource of neuropeptides. Database. 2015, bav038 (2015). Writing: KS, AMA, EA, KWL, RDH
Editing: KS, SL
30. N. L. Jorstad, J. H. T. Song, D. Exposito-Alonso, H. Suresh, N. Castro, F.
M. Krienen, A. M. Yanny, J. Close, E. Gelfand, K. J. Travaglini, S. Basu, M.
Beaudin, D. Bertagnolli, M. Crow, S.-L. Ding, J. Eggermont, A. Glandon, J. Competing interests: The authors declare that they
Goldy, T. Kroes, B. Long, D. McMillen, T. Pham, C. Rimorin, K. Siletti, S. have no competing interests.
Somasundaram, M. Tieu, A. Torkelson, K. Ward, G. Feng, W. D. Hopkins,
T. Höllt, C. D. Keene, S. Linnarsson, S. A. McCarroll, B. P. Lelieveldt, C. C.
Sherwood, K. Smith, C. A. Walsh, A. Dobin, J. Gillis, E. S. Lein, R. D. Hodge, T. Data and materials availability: The data were generated
E. Bakken, Biorxiv, in press, doi:10.1101/2022.09.19.508480. for the BRAIN Initiative Cell Census Network (BICCN,
31. L. Luebbert, L. Pachter, Biorxiv, in press, doi:10.1101/2022.05.17.492392. RRID:SCR_015820) and are available for download
from the Neuroscience Multi-omics Archive (NeMO,
RRID:SCR_016152) at: http://data.nemoarchive.
Acknowledgements: We are very grateful to members org/biccn/grant/u01_lein/linnarsson/transcriptome/
of the Linnarsson group for feedback and discussion. We sncell/10x_v3/human/raw/. The annotated expression
especially thank Anna Johnsson for lab management, Egle matrix, as well as code used for analysis and figure
Kvedaraite for immune-cell annotation, Elin Vinsland generation, is available on Github at linnarsson-lab/
for fibroblast and vasculature annotation, and Ivana
adult-human-brain. Auto-annotations are available at
Kapustová for the schematic drawing of the human brain.
linnarsson-lab/auto-annotation-ah.
We also acknowledge support from the National Genomics
Infrastructure in Stockholm funded by Science for Life
Laboratory, the Knut and Alice Wallenberg Foundation
Methods
and the Swedish Research Council, and SNIC/Uppsala
Multidisciplinary Center for Advanced Computational
Human postmortem tissue specimens
Science for assistance with massively parallel sequencing
and access to the UPPMAX computational infrastructure.
De-identified postmortem adult human brain tissue was
We thank the tissue procurement, tissue processing and
obtained after receiving permission from the deceased’s
facilities teams at the Allen Institute for Brain Science for
next-of-kin. Tissue collection was performed in accordance
assistance with the transport and processing of human
with the provisions of the United States Uniform
brain specimens and the technology team at the Allen
Anatomical Gift Act of 2006 described in the California
Institute for assistance with data management. The content
Health and Safety Code section 7150 (effective 1/1/2008)
of this article is solely the responsibility of the authors and
and other applicable state and federal laws and regulations.
does not necessarily represent the official views of the
The Western Institutional Review Board reviewed tissue
National Institutes of Health.
collection procedures and determined that they did not
Funding:
European Molecular Biology Organization fellowship constitute human subjects research requiring institutional
review board (IRB) review. The collection and processing
of human tissue was approved under Swedish law by the Data preprocessing
Swedish Ethical Review Authority (2019-03054).
Sequencing runs were demultiplexed with cellranger
Male and female donors 18–68 years of age with no mkfastq version 4.0.0 (10x Genomics) and filtered
known history of neuropsychiatric or neurological through the index-hopping-filter tool version 1.1.0 (10x
conditions were considered for inclusion in the study. Genomics). Unique molecular identifier (UMI) counts
Routine serological screening for infectious disease (HIV, were determined using STARSolo version 2.7.10a with the
Hepatitis B, and Hepatitis C) was conducted using donor following parameters:
blood samples and donors testing positive for infectious --soloFeatures Gene Velocyto
disease were excluded from the study. Specimens were --soloBarcodeReadLength 0
screened for RNA quality and samples with average --soloType CB_UMI_Simple
RNA integrity (RIN) values ≥7.0 were considered for --soloCellFilter EmptyDrops_CR %s 0.99 10 45000
inclusion in the study. Postmortem brain specimens were 90000 500 0.01 20000 0.01 10000
processed as previously described (dx.doi.org/10.17504/ --soloCBmatchWLtype
protocols.io.bf4ajqse). Briefly, coronal brain slabs were 1MM_multi_Nbase_pseudocounts
cut at 1 cm intervals, frozen in dry-ice cooled isopentane, --soloUMIfiltering MultiGeneUMI_CR
and transferred to vacuum-sealed bags for storage at --soloUMIdedup 1MM_CR
-80°C until the time of further use. To isolate regions of --clipAdapterType CellRanger4
interest, tissue slabs were briefly transferred to -20°C, and --outFilterScoreMin 30
the region of interest was removed and subdivided into
smaller blocks on a custom temperature controlled cold Barcode whitelists were downloaded from the 10x
table. Tissue blocks were stored at -80°C in vacuum-sealed Genomics website. Exonic, intronic, and ambiguous
bags until later use. counts were summed for clustering analysis.
Single-nucleus RNA sequencing The reference genome and transcript annotations were
based on the human GRCh38.p13 gencode V35 primary
Nucleus isolation for 10x Chromium RNA sequencing was sequence assembly. However, we filtered the reference.
conducted as described (dx.doi.org/10.17504/protocols. Because our pipeline only counted reads that uniquely
io.y6rfzd6). Gating on DAPI and NeuN fluorescence aligned to one gene, reads that aligned to more than one
intensity was described previously (2). NeuN+ and NeuN- gene were lost. Related genes therefore exhibited few or zero
nuclei were sorted into separate tubes and were pooled counts. To minimize this loss, we discarded overlapping
at a defined ratio after sorting. Sorted samples were genes or transcripts that overlapped or mapped to other
centrifuged, frozen in a solution of 1X PBS, 1% BSA, 10% genes or non-coding RNAs’ 3’ UTRs, leaving only one
DMSO, and 0.5% RNAsin Plus RNase inhibitor (Promega, of these transcripts in the genomic reference. We used
N2611), and stored at -80°C until the time of shipment on BLAST to align the last 400 nucleotides (3’ UTR) of all
dry ice from the Allen Institute to the Karolinska Institute protein-coding and non-coding transcripts to all other
for 10x chip loading. genes (maximum 4 mismatches, minimum alignment
length 300 nucleotides). We resolved all the matches by
Immediately before loading on the 10x Chromium the following procedure:
instrument, frozen nuclei were thawed in a 37˚C water • Fusion genes were filtered based on their names:
bath, spun down briefly, and pipetted several times to genes with names that contained both fusion genes
mix (dx.doi.org/10.17504/protocols.io.eq2lyrdpgx9k/v2). [“gene1-gene2”] were discarded.
Nuclei were then quantified and loaded according to the • Non-coding transcripts that matched a coding
10X Genomics protocol, targeting 5000 cells. We aimed transcript were discarded.
to sequence two replicates per sample. Nearly all samples • For overlapping transcripts where both transcripts
were processed with the 10x Genomics V3 kit; 12 samples were either coding or non-coding:
were sequenced with v3.1 (Table S1). Samples were 1. If one of the gene names matched the pattern
first sequenced to a shallow depth (approximately 1000 “XX######.#” its transcript was discarded.
reads/cell) on the Illumina NextSeq platform to validate 2. If one of the transcripts belonged to a gene
sample concentrations. Samples were then sequenced with one or more transcripts that were already
to approximately 100,000 reads/cell on the Illumina discarded during the procedure, it was discarded
NovaSeq platform. Sequencing saturation was examined as well.
for each sample using the preseq package (https://github. 3. We discarded transcripts of the gene with fewest
com/smithlabcode/preseq). Any samples that were not splice variants.
saturated to 60% were sequenced more deeply using 4. Otherwise, the transcript with a 3’ UTR that
preseq predictions. overlapped the other transcript was kept in the
genomic reference
5. For paralogs, we mapped all related genes (that
aligned to one another) and selected the paralog greater than the 99th percentile mean (0.056); or (2) if the
with most splice variants. All other highly similar Cytograph pipeline tagged the cluster with two or more of
paralogs were discarded. In special cases, we the following auto-annotations: NEUR, ASTRO, OLIGO,
manually chose the gene. OPC, MGL, ENDO, BERG, or VLMC. Auto-annotations
and further information about the auto-annotation
Altogether we filtered 387 fusion genes, 1140 overlapping procedure are available at github.com/linnarsson-lab/
transcripts, 414 non-coding transcripts, 1127 coding auto-annotation-ah. After removing these clusters, the
paralogs, and 350 non-coding paralogs. remaining cells in each supercluster were re-clustered.
This procedure was repeated until no clusters met these
Quality control criteria.
10x Chromium samples that captured more than Finally, clusters and subclusters at Levels 3 and 4,
15,000 cells were not analyzed further (Table respectively, underwent a merging procedure (“cytograph
S1). All remaining samples were pooled with merge”). A cluster that did not express at least one
their replicates and analyzed with “cytograph qc” statistically enriched gene was merged with the nearest
(commit 5f07e59768844e3c84fceb8fd7d1993059e13a7e), which cluster on the t-SNE. In short, FDR-corrected p-values
uses a modified version of DoubletFinder to calculate a were calculated by comparing Cytograph’s enrichment
doublet score for each cell (28). Cells with fewer than 800 scores with a null distribution obtained by permuting
UMIs, fewer than 40% unspliced molecules, or a doublet cluster labels across all cells in the supercluster.
score below 0.3 were removed from further analysis. These
thresholds were determined by examining distributions A t-distributed stochastic neighbor embedding (t-SNE)
across the dataset (Figure S1). and dendrogram were calculated for the final 461 clusters
(Figure 1). The dendrogram was calculated based on
Clustering Euclidean distances between cluster means for the 2,000
most variable genes. Cluster means were normalized
Cells were clustered using an updated version of the to median cluster means across the dataset and then
Cytograph package (github.com/linnarsson-lab/adult-human- standardized (scikit-learn StandardScaler).
brain, commit 22a83ebcf79a9fdcdff34beb0300aa1695d7039b) (1). The
command “cytograph process” was run with configuration Annotation
min_umis: 0, doublets_action: None, features: variance,
remove_low_quality: False, remove_doublets: False, Superclusters were manually annotated based on the
batch_keys: [“Donor”], clusterer: sknetwork. Default literature and on their regional composition. Clusters were
values were used for other parameters. In brief, principal not named with a single annotation but instead tagged
component analysis (PCA) was performed on the most with auto-annotations (github.com/linnarsson-lab/auto-
variable genes and used to construct a k-nearest neighbors annotation-ah) from four categories: neurotransmitters
graph, which provided input for Louvain clustering (auto-annotation-ha/Human_adult/Neurotransmission),
(scikit-network). Harmony with default parameters was neuropeptides (auto-annotation-ha/Neuropeptides),
used to integrate principal components across donors class (auto-annotation-ha/Human_adult/Class), and
before constructing the graph (6). subtype (auto-annotation-ha/Human_adult/Subtype).
Neuropeptide-related tags were manually compiled based
This clustering pipeline was used at four levels of analysis: primarily on the NeuroPep database (29). Subtype auto-
(1) All cells that passed QC were pooled into a single annotations included manually selected cell types and
dataset and analyzed. This top-level clustering was split states previously described in the literature. To identify
by the dendrogram into two groups that corresponded to cortical cell types, we also transferred labels from a
either neuronal or non-neuronal cells, with one exception: single-cell analysis of the middle temporal gyrus (MTG)
lymphocytes clustered with neurons. (2) Each of these (30). We performed PCA (50 components) on our full
two groups was separately analyzed. (3) Paris clustering dataset, trained a random forest classifier (scikit-learn,
was used on each graph to find “superclusters” within class_weight='balanced', max_depth=50) on the MTG
each group for further analysis (Figure 1) (7). The Paris labels, and then predicted labels for all cells. We labeled
dendrogram was arbitrarily cut so that the superclusters each cluster with the mode of its constituent cells if two
mirrored large islands on the t-SNE. Each supercluster conditions were met: more than 0.8 of predicted labels
was separately analyzed to produce “clusters.” (4) Each matched the mode, and the mean probability of these
of the 461 clusters was analyzed separately to produce predictions was greater than 0.8. Clusters that did not
“subclusters.” meet these criteria were labeled N/A.
At Level 3, additional cluster-level quality control was Gene ontology-enrichment analysis

performed. A cluster was removed from further analysis
if: (1) its mean percentage of mitochondrial UMIs was The gget package (https://github.com/pachterlab/gget) was
used to perform all gene-ontology enrichment analysis at least 3 separate sections from one human donor. Images
(function “enrichr” with database=“ontology”) (31). were assessed with FIJI distribution of ImageJ v1.52p and
Nikon NIS-Elements imaging software. Probes used were
Assessing supercluster complexity with PCA Hs-OTX2-C3 (#484581-C3), Hs-SOX14-C2 (#1055351-
C2), and Hs-CASR (#411931) from ACD Bio.
We used a permutation test to determine the number
of principal components necessary to describe each Extended analysis of dopaminergic neurons
supercluster, reasoning that this value would be
proportional to the supercluster’s complexity. We To identify additional dopaminergic subtypes, subclusters
therefore performed PCA twice on each supercluster 1870, 1871, 1873, and 1876 were reanalyzed separately as
(scikit-learn PCA, n_comps=50): once on the original described under “Clustering.”
expression matrix, and once on a permuted expression
matrix, where the expression values of each gene were To compare our data with a recently published census
randomly redistributed across cells. We compared the of midbrain dopaminergic neurons from Kamath et. al
percentage of variance that each principal component (4), we trained a random forest classifier on nuclei from
explained (attribute “explained_variance_ratio_”) in the their dataset annotated as dopaminergic neurons from
original and unpermuted matrix. We considered the first control samples with non-neurological disorders (Donor
component where explained_variance_ratio_ was similar IDs: 3298, 3322, 3345, 3346, 3482, 4956, 5610, 6173). We
between the two analyses (less than 10% higher in the trained the classifier with features that Cytograph selects
unpermuted analysis than in the permuted) to represent when run with the option to select cluster-enriched
the optimal number of components to describe the rather than variable genes (default parameters except
supercluster (Figure 3D, left). For each supercluster, we n_factors: 20 and features: enrichment). The scikit-
calculated an additional value using only the unpermuted learn RandomForestClassifier was regularized with the
analysis: the ratio between the percentages of variance parameters max_features and min_samples_leaf. Five-fold
explained by the first and second components (Figure 3D, cross-validation was performed on data from Kamath et al.
right). We reasoned that this value illustrates how quickly with scikit-learn StratifiedShuffledSplit for max_features
the explained_variance_ratio decays across principal 0.3, 0.4, 0.5 and min_samples_leaf between 3 to 50. We
components. The script is available as permuted_pca.py at selected max_features=0.3 and min_samples_leaf=4 to
github.com/linnarsson-lab/adult-human-brain. transfer labels to our data; oob_score was set true, and
other parameters were left as default.
RNAscope Multiplex fluorescent in situ hybridization
(mFISH) EEL-FISH
Fresh-frozen human postmortem brain tissues were
sectioned at 10 μm onto Superfrost Plus glass slides (Fisher EEL-FISH was performed as recently described (15). Donor
Scientific). Sections were dried for 10 minutes at -20°C H18.30.002 was used for the V1C sample; H20.30.002 for
and then vacuum sealed in plastic slide boxes and stored the AMY, MTG and A46 samples; H19.30.001 for the
at -80°C until use. The RNAscope multiplex fluorescent v2 pons sample; and H18.30.001 for the SN-RN and VTA
kit was used per the manufacturer’s instructions for fresh- samples. For all figure panels, genes were only plotted
frozen tissue sections (ACD Bio), except that fixation was for an individual cell if the cell contained more than two
performed for 60 minutes in 4% paraformaldehyde in 1X molecules of that gene. Expression below this threshold
PBS at 4°C and protease treatment was shortened to 15 was considered experimental noise.
minutes. Sections adjacent to those used for RNAscope
were fixed for 15 minutes in 4% paraformaldehyde in 1X Differential-expression testing
PBS, briefly washed in PBS, and stained with NeuroTrace
500/525 Green Fluorescent Nissl Stain and DAPI to aid Differential-expression testing was performed with the
in region localization. Nissl-stained sections were imaged diffxpy package, version 0.7.4+21.g12f1286. Datasets were
using a 10X objective on a Nikon TiE fluorescence subset before analysis; 25,000 cells were randomly selected
microscope equipped with NIS-Elements Advanced from each test group. Cells were normalized to the median
Research imaging software (v4.20, RRID:SCR_014329). total UMI count across the dataset and then used as input
RNAscope sections were imaged using a 40X oil immersion for a Mann-Whitney rank test (function “pairwise” with
lens on the same microscope. For mFISH experiments, test=“rank”).
positive cells were called by manually counting RNA
spots for each gene. Cells were called positive for a gene Topic modeling with Latent Dirichlet Allocation
if they contained ≥ 5 RNA spots for that gene. Lipofuscin
autofluorescence was distinguished from RNA spot signal We used tomotopy version 0.12.2 to perform topic
based on the larger size of lipofuscin granules and broad modeling with Latent Dirichlet Allocation (LDA). In brief
fluorescence spectrum of lipofuscin. Staining for the we trained a model with a specified number of topics “k”
probe combination was repeated with similar results on until perplexity stabilized (function “LDAModel” with
parameters alpha=50/k, eta=0.1). We trained with no

more than 200,000 cells and all protein-coding genes in the
genome that were expressed in at least 100 cells and fewer
than 60% of all cells. Some topics were donor-specific
or quality-specific, e.g, mitochondrial or ribosomal, and
some did not have an apparent biological meeting. We
therefore identified topics of interest manually. We also
identified representative genes for each topic by using
the topic probabilities reported for each gene (function
“get_topic_word_dist”). We filtered genes by specificity
(topic probabilities for each gene normalized by that
gene’s probability across all topics) and then sorted the
remaining genes by their unnormalized probabilities.
A script is available as optimize_lda.py at github.com/
linnarsson-lab/adult-human-brain.
Number of genes per cell Number of UMIs per cell
A
Number of samples
D
Cell-class proportions Hypothalamus Cerebral cortex
100000
150000
50000
10000
15000
A13
0
*
5000
100
HTHma
50
0
*
A13 HTHma-HTHtub A14
0 A14 HTHpo A19
A19 A1C
A1C HTHpo-HTHso
2500 A23 A23
HTHso
A25 A25
5000 A29-A30 HTHso-HTHtub
Macroglia
A32 HTHtub A29-A30
A35-A36
7500 A35r MN A32
A38 A35-A36
10000 A40
*
A35r
Excitatory neurons
A43
A44-A45 A38
12500 A46
A5-A7 Thalamus A40
Number of cells captured

ACC A43
15000 FI ANC
ITG CM A44-A45
17500 Idg A46
Ig CM-Pf
LEC A5-A7
*
ETH
M1C ACC
Immune cells
MEC LG
MTG LP AON
Pir
E
FI
Inhibitory neurons
Number of cells Pro LP-VPL
S1C MD ITG
0
1
2
STG Idg
TF MD-Re
0.0 TH-TL MG Ig
1e6
V1C
0.1 V2 Pul LEC
BL STH M1C
0.2 BM
BNST VA MEC
Vasculature
0.3 CEN MTG
CMN VLN
Choroid plexus
0.4 CaB Pir
* ** *
VPL
0.5 Cla Pro
CoA
0.6 GPe S1C
GPi STG
0.7 La
DoubletFinder score
0.8 NAC Midbrain TF
Pu TH-TL
0.9 SEP
SI IC V1C
Siletti et al. Transcriptomic diversity of cell types across the adult human brain
CA1 PAG V2
CA1-2R
CA1C-CA3C PAG-DR
Number of cells CA1R-CA2R-CA3R
*
PTR
CA1U-CA2U-CA3U
CA2U-CA3U RN
CA3R SC
10000
20000
CA4C-DGC Hippocampus
0
DGR-CA4 SN
DGR-CA4Rpy SN-RN CA1
0 DGU-CA4Upy
Sub CA1-2R
AON CA1C-CA3C
HTHma
1000 HTHma-HTHtub CA1R-CA2R-CA3R
HTHpo Cerebellum
HTHpo-HTHso CA1U-CA2U-CA3U
2000 HTHso CBL CA2U-CA3U
HTHso-HTHtub CBV
HTHtub CA3R
MN CbDN CA4C-DGC
Total UMIs
3000 ANC
CM DGR-CA4
CM-Pf DGR-CA4Rpy
4000 LG
LP DGU-CA4Upy
LP-VPL Pons Sub
MD
MD-Re DTg
C
MG PB
Pul
Number of cells STH Donor PN
VA PnAN
VLN
VPL PnEN Cerebral nuclei
0
ETH
IC PnRF Cla
250000
500000
750000
PAG CaB
0
PAG-DR
PTR GPe
0.0 RN GPi
SC Medulla
0.1 SN NAC
SN-RN IO Pu
25000
0.2 CBL
CBV MoAN SEP
0.3
*
CbDN MoRF-MoEN SI
DTg
0.4 IO MoSR BL
2500 cells
5000 cells
9000 cells
12500 cells
0.5 MoAN BM
MoRF-MoEN
MoSR
50000
0.6 CEN
PB
Number of UMIs per cell
PN CMN
0.7
*
PnAN Spinal cord CoA
Fraction unspliced UMIs

0.8 PnEN
PnRF La
*
SpC
0.9 SpC BNST

H19.30.002
H19.30.001
H18.30.002
H18.30.001
18
75000
Supplementary Figure 1 | Overview and quality of the dataset. (A) Anatomical locations dissected within each brain
region are shown along the x-axes. These locations are hereafter referred to as “dissections.” For each dissection, vertical
dots represent 10X Genomics samples that passed quality control. Dot size corresponds to the number of cells obtained
from each sample, as indicated by the legend. Dot color represents the sample’s donor of origin. Samples that were not
replicated across all three donors are starred. Some samples that appear not to have been replicated might have been
pooled differently across donors, such as red nucleus (RN), which was pooled with substantia nigra (SN) in two donors
but not the third. (B) For each dissection, box plots without fliers summarize the distributions of total unique molecular
integer (UMIs, top) and total genes (middle) per cell. Stacked bar plots summarize the proportions of cell classes sampled
from each dissection (bottom). Each bar is colored by class as indicated by the legend. (C) For each donor, box plots
without fliers summarize the distributions of total UMI counts per cell. (D) A histogram summarizes the distribution of
cells captured per 10X Genomics sample. (E) Histograms summarize the distributions of metrics used to perform quality
control on the complete dataset: DoubletFinder score per cell, total UMI count per cell, and percentage of unspliced
molecules in each cell. Orange bars in each plot indicate the threshold used to remove low-quality cells.
A B
Efferent nuclei of the pons (PNeN)
SOX14 OTX2 CASR OTX2+, CASR2+ cells
Max
Min
0
C D OTX2
SOX14
PnTg
CASR
500mm 10mm
E OTX2
PN
SOX14
CASR
1000mm
500mm 10mm
Supplementary Figure 2 | Midbrain-derived inhibitory neurons localize in the pons. (A) Cells from the efferent
nuclei of the pons (PnEN) are arranged by t-SNE and colored by their expression of marker genes for midbrain-derived
inhibitory neurons. (B) Cells on the t-SNE are colored purple if they express both OTX2 and CASR. (C) A Nissl-stained
overview of a section is adjacent to one processed with RNAscope. Pontine nucleus (PN) and the pontine tegmentum
(PnTg) that contains PnEN are delineated. Two boxes approximate the regions imaged in D and E. (D-E) On the left-hand
side, the cyan circles mark cells that are positive for midbrain-inhibitory markers OTX2, SOX14, and CASR. Black boxes
indicate the cells shown in the fluorescent images on the right-hand side.
A
LAMP5-LHX6
Misc. Upper IT Deep IT Deep CT/6b CGE interneurons MGE interneurons & Chandelier
Amygdala
222,100 cells
BL
BM
BNST
164
401
138
135
122
120
121
130
126
127
142
143
144
141
139
137
145
146
147
148
152
151
150
104
103
98
105
101
102
100
110
108
289
290
291
292
296
295
294
293
285
284
283
282
279
280
281
288
287
286
278
277
276
258
257
256
255
260
259
236
262
261
243
241
242
249
250
248
252
251
253
254
246
245
244
269
273
271
272
274
275
264
265
266
CEN
CMN
CA1-CA3 MSN Eccentric MSN Amygdala excitatory Splatter Thalamic CoA
La
187
163
427
430
221
219
218
220
217
216
213
214
215
211
212
208
207
426
227
226
225
230
231
232
233
234
228
229
223
224
419
407
406
405
156
157
158
161
162
159
160
153
154
155
174
400
393
392
363
390
382
420
421
415
422
424
423
425
416
377
378
428
429
410
238
237
235
409
403
459
LAMP5-LHX6
Misc. Upper IT Deep IT Deep NP Deep CT/6b CGE interneurons MGE interneurons & Chandelier
Basal forebrain
301,209 cells
CaB
Cla
GPe
GPi
MSN Eccentric MSN Splatter NAC
Pu
SEP
SI
Cerebellum
151,947 cells
Spl. Upper rho. Cerebellar inh.

CBL
CBV
CbDN
Misc. Upper IT Deep IT Deep CT/6b CGE interneurons MGE interneurons CA1
Hippocampus
347,947 cells
CA1-2R
CA1C-CA3C
CA1R-CA2R-CA3R
CA1U-CA2U-CA3U
CA2U-CA3U
LAMP5-LHX6 CA3R
& Chandelier Hippocampus CA1-CA3 Hippocampus CA4 Hippocampus DG MSN Amygdala excitatory Splatter
CA4C-DGC
DGR-CA4
DGR-CA4Rpy
DGU-CA4Upy
Sub
Amygdala excitatory
/6b LAMP5-LHX6 HTHma
Hypothalamus
134,471 cells
r IT IT CT & Chandelier
pe p p HTHma-HTHtub
Misc. Up Dee Dee CGE MGE MSN Eccentric MSN Mammillary body Splatter
HTHpo
HTHpo-HTHso
HTHso
HTHso-HTHtub
HTHtub
MN
IC
Midbrain
249,305 cells
Thalamic exc. PAG

CGE Upper rhombic lip
Misc. DG Splatter Midbrain-derived inh. Cerebellar inh. PAG-DR
PTR
RN
SC
SN
SN-RN
Lower rhombic lip

Medulla
85,487 cells
Splatter Lower rhombic lip IO

MoAN
MoRF-MoEN
MoSR
Midbrain-derived Upper rho. lip DTg

Pons
204,168 cells
Splatter inhibitory Lower rho. lip Cerebellar inh. PB

PN
PnAN
PnEN
PnRF
CHRNA1
DCDC2C TARID
GPR151 SMILR SHOX2
MMRN1 MATN3
CALCR 29% BNST
LHX8
Splatter Thalamic excitatory DG Amy.Exc.
Thalamus
343,320 cells
372
460
455
459
458
457
456
448
449
450
451
453
454
452
447
446
445
334
375
336
393
392
361
358
355
354
353
363
390
356
373
367
365
366
368
376
384
383
420
421
415
422
424
423
425
416
377
378
428
429
410
238
409
200
204
419
* *
PENK
LAMP5 GPR6
LHX6 Ecc.
CGE MGE Chand. MSN MSN Midbrain-derived inh. ANC LP Pul
CM LP-VPL STH
CM-Pf MD VA
ETH MD-Re VLN
LG MG VPL
261
296
294
293
288
287
286
270
273
274
427
219
217
215
211
212
208
207
210
209
225
233
234
223
433
439
440
437
438
435
434
436
444
443
442
441
Non-thalamic
B Eccentric MSNs C Eccentric-MSN score D Basal ganglia E CASZ1 DRD1 DRD2

Caudate body (CaB)
0.040
Eccentric MSNs
Max
0.035
0.030
0.025
0.020
0.015
0.010 Min
0
0.005
Supplementary Figure 3 | Superclusters and clusters are differentially distributed across brain regions. (A) Bar plots
represent neuronal cluster distributions for each region as in Fig. 2B. (B) Cells from the caudate body (CaB) in the
striatum are arranged by t-SNE; eccentric MSN cells are purple. (C) CaB cells are colored by a score that represents the
proportion of total UMIs that derive from the top 100 markers listed for mouse eccentric MSNs by Saunders et al. (D)
Eccentric MSNs on the t-SNE are purple if they derive from the basal ganglia. (E) Cells are colored by gene expression.
A Total UMI C E Percent of splatter cells F Top GO terms among 2000 most variable genes
Percent of splatter cells
50
5.5 50 Number of genes Adj. p-value
0
5.0 SLC17A7
SLC17A6 extracellular structure organization
4.5 SLC17A8
external encapsulating structure organization
SLC32A1
4.0 0 SLC6A1 chemical synaptic transmission
GAD1
anterograde trans-synaptic signaling
A1
SL 7A8
1
C1 6
GAD2
C 7
3.5
7A G D1
LC 2
2A
SL 17A
SL 7A
;S D
32
A
A
SLC6A9
C3
neuropeptide signaling pathway
C1
G
SL
3.0 TH
extracellular matrix organization
6
DDC
C1
B Fraction unspliced D
SLC6A3 regulation of cytosolic calcium ion concentration
SL
SLC18A2
adenylate cyclase-activating G protein-coupled receptor signaling pathway
Ce Percent of neurons in region SLC18A3
DBH adenylate cyclase-modulating G protein-coupled receptor signaling pathway
0.9 50 SLC6A2
that are splatter
negative regulation of inflammatory response to antigenic stimulus
PNMT
0.8 SLC6A4 collagen fibril organization
TPH2
0.7 regulation of inflammatory response to antigenic stimulus
CHAT
0 SLC5A7 phospholipase C-activating G protein-coupled receptor signaling pathway
0.6
ACHE
synaptic transmission, dopaminergic
th uc us
re Po in
al lla
Hy ebr camrtex
Th am i
M lam s
Ce bra s
Sp e um
rd
s
HDC
e
a u
id u
be n
l
po al n p
in du
co
0.5
l
o
l
SLC6A5 regulation of synaptic transmission, glutamatergic
Ce ipp al c
M
a
o
H br
re
68.0
45.33
22.67
0
0
0.01
r
G H I
Gene expression Subclusters Fraction shared subclusters
Permuted gene expression Cerebral cortex
1.0
0.04 Hippocampus
Explained variance
0.8
Cerebral nuclei
Hypothalamus
0.6
Thalamus
Midbrain
0.4
Pons
0.02 Cerebellum
*
0.2
Myelencephalon
Spinal cord
co n
Sp ep llu s
po al p x
in ha m
th nu us
a m i
M lam us
rd
le reb Poin
M C bra s
nc e n
Th ala cle
Hy br am te
al lo
id u
0.00 re oc or
Ceipp ral c
0 2 4 6 8 10 12 14 16 18
ye e
H eb
r
Ce
Component
J
Cerebral cortex Hippocampus Cerebral nuclei Hypothalamus Thalamus Midbrain Pons Cerebellum Medulla Spinal cord
K Total UMI L Superclusters M Subclusters N Distance to 25th neighbor

Astrocytes
5.00 1.0
Parabrachial nuclei
4.75 Lower rhombic Midbrain-derived 0.9

lip inhibitory
4.50 Upper 0.8
rhombic lip Microglia
4.25 0.7
4.00
OPCs
0.6
3.75 Splatter 0.5
Oligos
3.50
0.4
3.25 Cerebellar 0.3
3.00 inhibitory
0.2
O P Q S PnEN PTR
CALCA
SATB2
SLC17A6
SLC32A1
PB DTg
Total UMI
Parabrachial nuclei: splatter
5.00
4.75
4.50
4.25
4.00 R MoRF-MoEN PAG HTHpo-HTHso PAG-DR

Number of clusters
3.75 50
3.50
0
o
3.25
N
Hs
g
G
PB
DR
oE
DT
PA
PT
E
HT
Pn
G-
M
o-
PA
F-
Hp
oR
3.00
HT
M
Most common dissection
Supplementary Figure 4 | Splatter neurons are highly complex and undersampled. (A) Splatter cells are colored on
the t-SNE by log10 of their total UMI counts. (B) Cells on the t-SNE are colored by fraction of unspliced UMIs. (C)
The height of each bar indicates the percentage of splatter neurons that express the gene (or gene combination) shown
on the x-axis. (D) The height of each bar indicates the percentage of neurons from each brain region that are splatter
neurons. (E) The length of each bar indicates the percentage of splatter neurons that express the neurotransmitter shown
on the y-axis. (F) A bar plot shows the most enriched gene-ontology terms among the 2000 genes most variable across
splatter neurons. For each term, bar lengths represent the number of variable genes that overlapped with the term (left)
and p-value of the enrichment (right). Green line indicates 0.05. (G) An example plot illustrates how the values shown
in Fig. 3D were calculated for one supercluster. Bar height represents the fraction of total variance explained by each
principal component along the x-axis. Principal components were calculated using both the original gene-expression
matrix (blue) and a matrix in which each gene’s expression was permuted across cells (orange). The star indicates the
value that is plotted left in Fig. 3D: the component where the fraction of explained variance in the original dataset is less
than 10% greater than that of the permuted dataset. (H) Splatter cells on the t-SNE are colored by subcluster. (I) Each row
in the matrix represents the fractions of subclusters from each region that are shared with regions along the x-axis. To be
considered shared, subclusters contained at least five cells from both regions. (J) Splatter cells on each t-SNE are colored
if they derive from the indicated region. (K) Cells from the parabrachial nuclei are colored on the t-SNE by log10 of their
total UMI counts. (L) Cells are colored by supercluster, labeled on the panel. (M) Cells are colored by subcluster. (N) Cells
are colored by Euclidean distance to their 25th nearest neighbors. Distances were normalized by the maximum distance in
the dataset. (O) Splatter neurons from the parabrachial nuclei are colored on the t-SNE by log10 of their total UMI counts.
(P-Q) Cells are colored by expression levels of the gene they express most highly among those in the legend. Darker color
indicates higher expression; gray indicates no expression. (R) Bar height indicates the number of parabrachial splatter
subclusters for which the dissection along the x-axis was the most common dissection. (S) Cells are colored purple if they
derive from a subcluster predominated by the indicated region.
A B Max
C SLC6A4 D Total UMI E Donor
Cluster 397 SLC6A4 5.00 H18.30.001
H19.30.001
4.75 H19.30.002
4.50
4.25
4.00
Min
3.75
0
3.50
3.25
F Subclusters (reproduced
from Figure 3)
G SLC6A4+, SLC17A8+ H SLC6A4+, SLC32A1+
cells cells
Max
Clusters
Min
0
TPH2
DDC
HMX3
HOXB2
P2RY1
TECRL
GHRH
PIEZO2
SLC7A3
RHOJ
BTNL9
VIP
APCDD1
ANO2
CCDC141
PIK3R5
BDNF
POU3F1
MET
VAV3
DEFA6
TNS3
ABCA8
PTH2R
ELOVL2
MATN3
LIPG
ANKRD30A
KIT
SHISAL2B
KRT19
TMEM176B
TGFBI
AC068774.1
SATB2
EN2
CHST9
PAX5
ARHGAP36
TGFBR2
HOXB3
TAC1
NTN4
ANTXR2
CRHBP
CMTM8
WNT2
MYPN
RARB
TAC3
GCNT4
COL11A1
MC4R
LMX1A
PITX2
ROBO3
TTC6
LHFPL3
DCN
WIF1
IGFBP3
COL8A1
PKHD1L1
TRH
CWH43
NKX6-1
ADAMTS6
SYT10
PNOC
PAX6
EGFL6
RXFP2
MYO1B
PENK
J Cluster 395 K SLC6A3
L TH M Total UMI N Donor
H18.30.001
H19.30.001
5.00 H19.30.002
4.75
4.50
4.25
4.00
3.75
3.50
3.25
O P Q
Subcluster Cell-type prediction Accucary
F1 score
Subtype 0.78
SOX6_GFRA2
SOX6_AGTR1
Accucacy/ F1 Score
Region SOX6_DDT
SOX6_PART1
CALB1_RBP4
CALB1_CRYM_CCDC68
SOX10 CALB1_TRHR
CALB1_CALCR 0.75
MOG CALB1_PPP1R17
CALB1_GEM
LMX1A
NR4A2
PITX3 0.72
TH
DDC 5 25 45 0.3 0.4 0.5
SLC6A3 Min. samples leaf Max. features
SLC18A2
SLC18A1
R 1.0 S
SOX6
Prediction probability
ALDH1A1 1.0
CALB2
CALB1
Value
GAD2 0.6
GAD1 0.6
SLC32A1
PART1
GFRA2
AGTR1 0.2
LPL
CERKL
ion all re
cis Rec sco
0.0
CALCRL
Pre F1
EBF2
ADCYAP1
T U
PRLHR VIP 7.2
NPPC 5.0
CRYM 6.3 4.4 SLC18A2 and SLC32A1
CALCR 5.4 3.9 TH and GAD2
4.5 3.3
SLC18A2, SLC32A1, TH, and GAD2
NEUROD6
3.7 2.7
PPP1R17
2.8 2.1
SST 1.9 1.6
PAX5 1.0 1.0
SEMA3D PENK NPW
3.7 6.0
GEM
3.3 5.3
NPW 4.6
2.9
CARTPT 2.5 3.9
PENK 2.2 3.2
GCGR 1.8 2.4
NPPC 1.4 1.7
VIP 1.0 1.0
Supplementary Figure 5 | Serotonergic and dopaminergic neuronal heterogeneity within the splatter supercluster.
(A) Splatter cells on the t-SNE are colored if they belong to the serotonergic cluster 397. (B) Cells are colored by gene
expression. (C) Serotonergic cells are colored on the t-SNE by gene expression, legend in (B). (D) Cells are colored by log10
of their total UMI counts. (E) Cells are colored by donor. (F) Cells are colored by subcluster. (G) Cells are colored purple
if they express both serotonin-reuptake transporter SLC6A4 and glutamate transporter SLC17A8. (H) Cells are colored
purple if they express both serotonin-reuptake transporter SLC6A4 and GABA transporter SLC32A1. (I) Heatmap shows
mean gene-expression levels across clusters for serotonergic markers TPH2 and DDC, the rostral and caudal serotonergic
markers HMX3 and HOXB2, and the top five enriched genes in each cluster. Color bar on the left shows the corresponding
subcluster for each row as in (F). (J) Splatter cells on the t-SNE are colored if they belong to the dopaminergic cluster 395.
(K) Cells are colored by gene expression. (L) Dopaminergic cells are colored on the t-SNE by gene expression, legend in
(B). (M) Cells are colored by log10 of their total UMI counts. (N) Cells are colored by donor. (O) A heatmap represents
expression levels for select genes across dopaminergic cells. Top bar indicates the subcluster for each cell as in Fig. 3M;
second bar indicates subtype as in Fig. 3P. Third bar indicates the region of origin for each cell colored as in Fig. 3N.
(P) A random forest classifier was used to transfer labels from Kamath et. al. The classifier also reported a probability
for each prediction, reflecting its confidence in the label transfer. Cells are colored if their probability was greater than
60%; other cells are grey. The colors indicate predicted label according to the legend. (Q) The classifier’s performance was
tested at different values for two parameters: minimum number of samples permitted at a decision tree’s leaf (left) and
the maximum number of features used to fit a tree (right). Performance was evaluated by both accuracy (orange) and F1
score (blue). (R) Precision, recall, and F1 scores are reported for each predicted label, colored according to the legend in
(K). The classifier most poorly identified CALB1_RBP4 cells. (S) Cells on the t-SNE are colored if the classifier’s prediction
probability was above 60%; other cells are grey. Color indicates probability. (T) Cells are colored by gene expression. (U)
Cells are colored by the product of logged expression in multiple genes.
A B
TH GAD1 PVALB SST PENK
TH
SLC17A6 RN GAD2
CALB2 PVALB
SLC32A1
GAD1
GAD2
RN
400 um
TH PBP
PBP CALB2
1mm
TAC1 CARTPT ADCYAP1 NTS SPX
1mm
200 um
TH PBP TH PBP
GAD1 SST
GAD2
200 um 200 um
C PVALB
TH TPH2
1mm
PENK HOPX
D
SN-RN VTA Pons
ADCYAP1 ADCYAP1 Midbrain
CARTPT CARTPT
INH INH
Pons
NPY NPY
NTS NTS
~100 Cells
PENK PENK
~50 Cells
PVALB PVALB
~25Cells
SLC17A6
1 Cell
SLC17A6
SLC18A2 SLC18A2
y
SLC6A2 SLC6A2
SLC6A4 SLC6A4
SPX SPX
SST SST
TAC1 TAC1
TH TH
TPH2 TPH2
VIP VIP
VIP TPH2 TH TAC1 SST SPX SLC6A4 SLC6A2 SLC18A2SLC17A6 PVALB PENK NTS NPY INH CARTPT ADCYAP1
x
VIP
H2
TH
C1
SST
SLC X
SLC 4
SLC 2
SLC 2
6
B
K
S
Y
INH
PT
VIP
H2
TH
C1
SST
SLC X
SLC 4
SLC 2
SLC 2
6
B
K
S
Y
INH
PT
VIP
H2
TH
C1
SST
SLC X
SLC 4
SLC 2
SLC 2
6
B
K
S
Y
INH
PT
1
NT
NT
NT
6A
6A
18A
17A
AP
6A
6A
18A
17A
AP
6A
6A
18A
17A
AP
AL
AL
AL
NP
NP
NP
SP
SP
SP
PEN
PEN
PEN
TA
TA
TA
RT
RT
RT
TP
TP
TP
CY
CY
CY
PV
PV
PV
CA
CA
CA
AD
AD
AD
Supplementary Figure 6 | EEL-FISH profiles gene-expression distributions in the midbrain and hindbrain. (A) Main
panel shows TH, GAD1 or GAD2, and CALB2 expression in Tissue A that includes the substantia nigra, red nucleus (RN),
and parabrachial pigmented nucleus (PBP). Top right shows enlarged view of SLC17A6 and PVALB co-expression in the
RN. Other insets show enlarged views of cells in the PBP co-expressing TH with GAD1 or GAD2, CALB2, and SST. Each
dot represents a molecule. (B) Each dot represents a molecule of TH, GAD1 or GAD2 or SLC32A1, PVALB, SST, PENK,
TAC1, CARTPT, ADCYAP1, NTS, and SPX in tissue B. (C) Each dot represents a molecule of TH, TPH2, PVALB, PENK,
and HOPX in tissue section C. (D) For SN-RN, VTA, and pons, the size of each dot represents the number of cells that
co-express two genes along the x-axis and y-axis.
A Clusters B Total UMI C Donor D Mammilary nucleus E MDFIC F COL10A1

H18.30.001
H19.30.001
H19.30.002
0 Min Max 0 Min Max
G I
Topic 3 TMEM163 Most representative genes Topic 10 IFNG-AS1 Most representative genes
2.5
MT-RNR2 PLXDC2
2.5 MT-CO2 2.0 FCHSD2
MT-CO3 OPALIN
2.0 MT-CO1 1.5 DCC
Mean expreession
MT-ND4 ROR1
0 Min Max 0 Min Max 1.0
0 5
H 0 5
Topic 7 MT-ND4 3.0

CTNNA2
J Topic 1 ROR1
2.5 SGCZ
2.5 LAMA2
2.0 PLCL2
2.0
DPP10
1.5
IFNG-AS1 1.5
Mean expreession
1.0
0 5
0 5
Position along x-axis
M K Topic 14 SGCZ
2.5
RASGRF1
N
Clusters (Zeisel et al. 2018) Region 2.0
ACTN2
Mouse oligodendrocytes
SETBP1
ZSCAN31
Telencephalon
1.5
MFOL1
PLEKHA7
1.0
NFOL2 0 5
COP2
L Topic 6 PLEKHA7 5
NFOL1 MOL2 RBFOX1
COP1 4 AFF3
Non-telencephalon
MFOL2 3 FSTL5
PARD3
2
TMEM163
OPC 1
MOL3 0 5
MOL1 Position along x-axis
N
Topic 6 Rbfox1 Cd44 Nkx2-9 Vat1l VAT1L NKX2-8
Human
Mouse
O 0 Min Max 0 Min Max
32
33
OPCs
34
35
37
Oligodendrocytes COPs
38
39
40
44
45
46
47
48
49
50
A13
A14
A19
A1C
A23
A25
A29-A30
A32
A35-A36
A35r
A38
A40
A43
A44-A45
A46
A5-A7
ACC
AON
FI
ITG
Idg
Ig
LEC
M1C
MEC
MTG
Pir
Pro
S1C
STG
TF
TH-TL
V1C
V2
CA1
CA1-2R
CA1C-CA3C
CA1R-CA2R-CA3R
CA1U-CA2U-CA3U
CA2U-CA3U
CA3R
CA4C-DGC
DGR-CA4
DGR-CA4Rpy
DGU-CA4Upy
Sub
Cla
CaB
GPe
GPi
NAC
Pu
SEP
SI
BL
BM
CEN
CMN
CoA
La
BNST
HTHma
HTHma-HTHtub
HTHpo
HTHpo-HTHso
HTHso
HTHso-HTHtub
HTHtub
MN
ANC
CM
ETH
LG
LP
LP-VPL
MD
MD-Re
MG
Pul
STH
VA
VLN
VPL
IC
PAG
PAG-DR
PTR
RN
SC
SN
SN-RN
CBL
CbDN
DTg
PB
PN
PnAN
PnEN
PnRF
IO
MoAN
MoRF-MoEN
MoSR
SpC
CBV
CM-Pf
Cerebral cortex
Hippocampus
Cerebral nuclei
Hypothalamus
Region
Thalamus
Midbrain
Pons
Cerebellum
Medulla
Spinal cord
Supplementary Figure 7 | Oligodendrocyte-lineage cells differ between the telencephalon and the rest of the brain.
(A) Oligodendrocytes on the t-SNE are colored by cluster. (B) Cells are colored by log10 of their total UMI counts. (C)
Cells are colored by donor. (D) Cells are colored purple if they derive from the mammillary nucleus. (E-F) Cells are
colored by gene expression. (G-L) Cells are colored by their scores for each gene topic (left) and their expression of a
gene representative for that topic (middle). For each topic, expression values for five representative genes were binned
and averaged along the t-SNE’s x-axis (right). (M) A t-SNE was recalculated for mouse oligodendrocytes from Zeisel
et al. Cells on the t-SNE are colored by cluster. (N) Cells are colored blue if they derive from the telencephalon (top) or
outside the telencephalon (bottom). (N) Mouse oligodendrocytes were scored for human Topic 6. The topic-representative
gene Rbfox1 is not expressed in mouse oligodendrocytes. Correlations were calculated between all genes and the murine
Topic 6 scores in myelin-forming (MFOL) and mature (MOL) oligodendrocytes. Mouse and human cells are colored by
expression of select highly correlated genes Cd44, Nkx2-9 (NKX2-8 in humans), and Vat1l. Human CD44 expression is
plotted in Fig. 5. (O) Dot plot represents the distributions of oligodendrocyte-lineage cells across dissections. Within each
column, dot size is proportional to the percentage of cells in the indicated dissection that belong to each cluster.
A Gene ontology terms for top differentially expressed genes ( > 3-fold) Number of genes Adj. p-value
negative chemotaxis
negative regulation of axon extension involved in axon guidance
SEMA3D SEMA3A SEMA6D SLIT1 SEMA3E generation of neurons
regulation of axon extension involved in axon guidance
skeletal system development
negative regulation of axon extension
negative regulation of amyloid fibril formation
Notch signaling pathway
negative regulation of chemotaxis
nervous system development
IRX1 IRX2 IRX5 LHX2 SEMA3A CCK regulation of amyloid fibril formation
neural crest cell migration
semaphorin-plexin signaling pathway
neural crest cell development
regulation of low-density lipoprotein particle receptor catabolic process
12.0
8.0
4.0
0
0
0.01
0.03
0.04
NCAN TNFRSF11B PRELP HOXD3 BMPR1B HAPLN1
Max
NRARP HES1 MAML3 HOXD3 HES5

Min
0
B Mouse OPCs C D E F
(Zeisel et al. 2022) Clusters Mammillary nucleus Donor FMO3
Top2a
Irx5
Hes5
G Total UMI H FOXG1 I NELL1 J CNTN5
Supplementary Figure 8 | OPC heterogeneity within the oligodendrocyte lineage. (A) Differential expression analysis
identified 155 genes with greater than three-fold enrichment in either OPC Type 1 or Type 2. On the right-hand side,
bar plots show the top 15 gene-ontology terms found among the differentially expressed genes; bar lengths represents the
number of genes that overlapped with a term (left) and p-value of the enrichment (right). Green line indicates 0.05. On
the left-hand side, OPCs in each row are colored by their expression of genes that overlapped with the indicated gene-
ontology term. (B) A t-SNE was calculated on OPCs from Zeisel et al. 2018. Cells are colored by expression levels of the
gene they express most highly among those in the legend. Darker color indicates higher expression; gray indicates no
expression. Top2a denotes cycling cells, and Hes5 and Irx5 are enriched in human Type 1 and Type2 OPCs, respectively.
(C) Human OPCs are colored by cluster. (D) Cells are colored purple if they derive from the mammillary nucleus. (E)
Cells are colored by donor. (F) Cells are colored by gene expression. (G) Cells are colored by log10 of their total UMI
counts. (H-J) Cells are colored by gene expression.
A Top 15 GO terms: Most variable genes B Astrocyte clusters C

Total UMI
Number of genes Adj. p-value 4.75
modulation of chemical synaptic transmission 4.50
glutamate receptor signaling pathway 61 64 4.25
regulation of synapse assembly 58 4.00
axon guidance
axonogenesis
63 3.75
3.50
chemical synaptic transmission 3.25
axon extension involved in axon guidance 3.00
neuron projection extension involved in neuron projection guidance
regulation of trans-synaptic signaling
57 59
extracellular structure organization 52 53 D Donor H18.30.001
external encapsulating structure organization H19.30.001
regulation of synaptic transmission, glutamatergic 56 H19.30.002
anterograde trans-synaptic signaling

negative chemotaxis 54
negative regulation of vascular endothelial cell proliferation 62 60
55
55.0
36.67
18.33
0
0
0.04
0.08
0.11
E
52
53
54
55
61
58
57
56
64
63
62
60
59
A13
A14
A19
A1C
A23
A25
A29-A30
A32
A35-A36
A35r
A38
A40
A43
A44-A45
A46
A5-A7
ACC
AON
ITG
Idg
Ig
LEC
M1C
MEC
MTG
Pir
Pro
S1C
STG
TF
TH-TL
V1C
V2
CA1
CA1-2R
CA1C-CA3C
CA1R-CA2R-CA3R
CA1U-CA2U-CA3U
CA2U-CA3U
CA3R
CA4C-DGC
DGR-CA4
DGR-CA4Rpy
DGU-CA4Upy
Sub
Cla
CaB
GPe
GPi
NAC
Pu
SEP
SI
BL
BM
CEN
CMN
CoA
La
BNST
HTHma
HTHma-HTHtub
HTHpo
HTHpo-HTHso
HTHso
HTHso-HTHtub
HTHtub
MN
ANC
CM
ETH
LG
LP
LP-VPL
MD
MD-Re
MG
Pul
STH
VA
VLN
VPL
IC
PAG
PAG-DR
PTR
RN
SC
SN
SN-RN
CBL
CbDN
DTg
PB
PN
PnAN
PnEN
PnRF
IO
MoAN
MoRF-MoEN
MoSR
SpC
CBV
FI
CM-Pf
F Cerebral cortex
Hippocampus
Cerebral nuclei
Hypothalamus
Region
Thalamus
Midbrain
Pons
Cerebellum
Medulla
Spinal cord
G MEC GPe Pu CaB NAC SI
Entorhinal cortex
Cerebral nuclei
LEC Cla SEP BNST GPe GPi
0
ZNF98
HGF
PIP5K1B
PDE3B
DPYSL5
CCBE1
PLA2R1
WDR86
CACNA1A
TBX5
SLC8A1
KCNQ3
LAMA3
EPHA6
TMEM132C
CRYM
GABRA4
GFRA1
SHTN1
FYB2
IGFBP5
SFRP2
SORCS3
USH1C
SEMA5A
ADAMTS18
LMO2
FABP7
HS3ST3A1
TTR
ST6GALNAC3
SLC16A12
GPC6
HTR2C
ATP8A2
CNTNAP2
CNTN4
NRG1
MTUS2
NAA11
PIK3C2G
LRAT
COL24A1
AC072022.2
CTNNA3
TMEM163
NRXN3
CCDC85A
NR4A1
CPAMD8
GADD45G
ADM
PAX3
DUSP1
ITPRID1
SEMA3E
LGR6
PRLHR
GABBR2
PTPRT
PAK3
RPH3A
NETO2
MDGA2
PENK
Supplementary Figure 9 | Astrocytes exhibit transcriptomic diversity across brain regions. (A) Bar plots show the
top gene-ontology terms among the 2000 genes most variable across astrocytes. For each term, bar lengths represent the
number of variable genes that overlapped with the term (left) and p-value of the enrichment (right). Green line indicates
0.05. (B) Astrocytes on the t-SNE are colored by cluster. (C) Cells are colored by log10 of their total UMI counts. (D) Cells
are colored by donor. (E) Dot plot represents the distributions of astrocytes across dissections. Within each column, dot
size is proportional to the percentage of cells in the indicated dissection that belong to each cluster. (F) Cells are colored
if they derive from the region indicated by color according to the legend on the right. (G) Cells are colored black if they
derive from the indicated dissection. (H) Heatmap shows mean gene-expression levels across clusters for the top five
enriched genes in each cluster. Color bar on the left indicates cluster as in (B).
A White-matter score
1 Cerebral cortex
Hippocampus
Cerebral nuclei
Hypothalamus
Region
Thalamus
Midbrain
Pons
Cerebellum
Medulla
0 Spinal cord
B Topic Score by Region DAO

5
Topic 29
0
C 0.53
CRABP1
5
Topic 5
0
0.50
HGF
D 5
Topic 22
0
0.58
E CCL2
5
Topic 8
0
0.76
0 Min Max 0 Min Max
Supplementary Figure 10 | Latent Dirichlet Allocation captures astrocyte heterogeneity. (A) Violin plots summarize
distributions of white-matter scores within each brain region. Horizontal line indicates a distribution’s mean. (B-E)
Astrocytes are colored on the t-SNE by their scores for each gene topic (left). Scores were then split by region, and a
histogram shows the score distributions for each region (middle). Cells are colored by their expression of a gene
representative for that topic (right).
Supplementary Table 1 | Samples. All 10x Genomics samples that were sequenced are listed with metadata in the
Excel table.
Supplementary Table 2 | Annotations. All clusters and their annotations are listed in the Excel table as ordered in the
Figure 1A dendrogram. See methods for annotation details.
Supplementary Table 3 | Differential expression between Type 1 and Type 2 oligodendrocytes. Mann-Whitney rank-
test results for all genes are listed in the csv file.
Supplementary Table 4 | Differential expression between Type 1 and Type 2 OPCs. Mann-Whitney rank-test results
for all genes are listed in the csv file.

Transcriptomic Diversity of Cell Types Across The

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Transcriptomic Diversity of Cell Types Across The

Uploaded by

Copyright:

Available Formats

bioRxiv preprint doi: https://doi.org/10.1101/2022.10.12.511898; this version posted October 14, 2022.

The copyright holder for this preprint

Transcriptomic diversity of cell types

Figure 3 | Splatter neurons are highly complex

Estimated number of components

Explained variance ratio (Component 1/ Component 2)

(F) Splatter cells from the parabrachial nuclei are

SOX6 VIP CALB1

GABA B C LC Figure 4 | EEL-FISH reveals spatial distributions

GABA B5 GABA B6 GABA B7 GABA B8

Figure 5 | Oligodendrocyte-lineage cells differ

17 clusters right, respectively) found in each region. Colors

each group are jittered randomly along the x-axis for

y-axis in (G) was trimmed to 0.01 for visualization;

one telencephalic dissection contained 0.04% cycling

cells. (J) Bar heights represent the ratios of

Figure 6 | Astrocytes exhibit transcriptomic

Thalamus nucleus of the stria terminalis; SEP = septal nuclei; SI

Cerebellum mirrors GFAP expression and therefore likely reflects

Spinal cord white-matter association. (G) Violin plots

25. T. M. F. Tuohy, N. Wallingford, Y. Liu, F. H. Chan, T. Rizvi, R. Xing, B. ALTF 1015-2018 to KS

At Level 3, additional cluster-level quality control was Gene ontology-enrichment analysis

parameters alpha=50/k, eta=0.1). We trained with no

Number of cells captured

PnAN Spinal cord CoA

Fraction unspliced UMIs

0.9 SpC BNST

Spl. Upper rho. Cerebellar inh.

Thalamic exc. PAG

Lower rhombic lip

Splatter Lower rhombic lip IO

Midbrain-derived Upper rho. lip DTg

Splatter inhibitory Lower rho. lip Cerebellar inh. PB

B Eccentric MSNs C Eccentric-MSN score D Basal ganglia E CASZ1 DRD1 DRD2

Percent of splatter cells

K Total UMI L Superclusters M Subclusters N Distance to 25th neighbor

4.75 Lower rhombic Midbrain-derived 0.9

4.00 R MoRF-MoEN PAG HTHpo-HTHso PAG-DR

Most common dissection

TAC1 CARTPT ADCYAP1 NTS SPX

A Clusters B Total UMI C Donor D Mammilary nucleus E MDFIC F COL10A1

0 Min Max 0 Min Max

Topic 7 MT-ND4 3.0

O 0 Min Max 0 Min Max

NRARP HES1 MAML3 HOXD3 HES5

G Total UMI H FOXG1 I NELL1 J CNTN5

A Top 15 GO terms: Most variable genes B Astrocyte clusters C

anterograde trans-synaptic signaling

LEC Cla SEP BNST GPe GPi

B Topic Score by Region DAO

You might also like