Single-cell atlas of early human brain development highlights heterogeneity of human neuroepithelial cells and early radial glia

Abstract

The human cortex comprises diverse cell types that emerge from an initially uniform neuroepithelium that gives rise to radial glia, the neural stem cells of the cortex. To characterize the earliest stages of human brain development, we performed single-cell RNA-sequencing across regions of the developing human brain, including the telencephalon, diencephalon, midbrain, hindbrain and cerebellum. We identify nine progenitor populations physically proximal to the telencephalon, suggesting more heterogeneity than previously described, including a highly prevalent mesenchymal-like population that disappears once neurogenesis begins. Comparison of human and mouse progenitor populations at corresponding stages identifies two progenitor clusters that are enriched in the early stages of human cortical development. We also find that organoid systems display low fidelity to neuroepithelial and early radial glia cell types, but improve as neurogenesis progresses. Overall, we provide a comprehensive molecular and spatial atlas of early stages of human brain and cortical development. Eze et al. use single-cell sequencing and immunohistochemical validation to create an atlas of early human brain development. In the telencephalon, they discover a diversity of progenitor subtypes, including two that are enriched in humans.

Full text

ResouRce
https://doi.org/10.1038/s41593-020-00794-1
1Department of Neurology, University of California, San Francisco (UCSF), San Francisco, CA, USA. 2The Eli and Edythe Broad Center of Regeneration
Medicine and Stem Cell Research, University of California, San Francisco (UCSF), San Francisco, CA, USA. 3Genomics Institute, University of California,
Santa Cruz, Santa Cruz, CA, USA. 4Department of Anatomy, University of California, San Francisco (UCSF), San Francisco, CA, USA. 5These authors
contributed equally: Ugomma C. Eze, Aparna Bhaduri. ✉e-mail: Aparna.Bhaduri@ucsf.edu; Arnold.Kriegstein@ucsf.edu
The human brain consists of billions of cells across several
functionally interconnected structures that emerge from the
neuroectoderm. Many of these structures are substantially
expanded or distinct compared to other mammals, particularly the
cerebral cortex, the outermost layer of the human brain responsible
for perception and cognition. These differences emerge at devel-
opmental stages before birth, and thus exploring the cell types in
the developing human brain is essential to better characterize how
cell types across the brain are generated, how they may be affected
during the emergence of neurodevelopmental disorders and how
human neural stem cells can be directed to specific cell types for
modeling or treatment purposes.
The brain exponentially increases in size after the neural tube
closes1. Later in development, across brain regions, a series of sim-
ilar neurogenic and gliogenic processes give rise to the constituent
cell types. However, at the molecular level, the sequence of events
that leads to the emergence of these progenitor cells early in devel-
opment is less well understood. The diversity of brain structures
is known to emerge as a result of segmentation events that gen-
erate the prosencephalon, mesencephalon and rhombencephalon
that are then further specified into the anatomical structures (tel-
encephalon, diencephalon and so on) that were dissected in this
study2 (Supplementary Fig. 1a–c). It has been proposed that there
are core gene regulatory programs that enable the specification
of these regions and subsequent development of topographically
relevant cell types3. We sought to explore whether our data could
more comprehensively define regional signatures and also iden-
tify cell type-specific similarities and differences in these nascent
brain structures.
The human cerebral cortex is more than three times expanded
compared to our closest nonhuman primate relatives4. The cortex
emerges from an initially pseudostratified neuroepithelium that
gives rise to radial glia, the neural stem cells of the cortex1. Radial
glia generate neurons, initially through direct neurogenesis and
then indirectly through transit-amplifying IPCs5. A number of
subtypes of radial glia have been identified as the cortex matures,
and their primary role in neurogenesis declines late in the second
trimester, at which point they generate the glial populations of the
cortex6. Single-cell RNA-sequencing (scRNA-seq) has added sub-
stantially to our knowledge about cellular diversity and signaling
networks, particularly during stages of peak neurogenesis. However,
the first trimester of cortical development has not been described at
this level of molecular detail, and important questions remain about
the timing of neurogenesis, the presumed uniformity of the neuro-
epithelium and the signals that promote the transition to radial glia.
Results
Whole brain analysis. To identify cell types and trajectories that
lay the foundation for the development of the human brain, we
performed scRNA-seq using the droplet-based 10X Genomics
Chromium platform. We sequenced cells from ten individuals dur-
ing the first trimester of human development, spanning Carnegie
stages (CS) 12 to 22, corresponding to gestational weeks 6–10. We
also included cortical samples from one CS13 and one CS22 indi-
vidual that were analyzed in a previous study7. To identify the chief
cell populations across brain regions and to compare them to one
another, we sampled all available and identifiable structures, includ-
ing the telencephalon, diencephalon, midbrain, hindbrain, cerebel-
lum, ganglionic eminences, thalamus, hypothalamus and cortex
(Supplementary Figs. 1–3 and Supplementary Table 1). We vali-
dated that our sequencing did not contain substantial artifacts or
cell debri, and that it represented highly expressed transcripts from
Single-cell atlas of early human brain
development highlights heterogeneity of human
neuroepithelial cells and early radial glia
Ugomma C. Eze 1,2,5, Aparna Bhaduri 1,2,5 ✉ , Maximilian Haeussler 3, Tomasz J. Nowakowski 1,4
and Arnold R. Kriegstein 1,2 ✉
The human cortex comprises diverse cell types that emerge from an initially uniform neuroepithelium that gives rise to radial
glia, the neural stem cells of the cortex. To characterize the earliest stages of human brain development, we performed single-cell
RNA-sequencing across regions of the developing human brain, including the telencephalon, diencephalon, midbrain, hindbrain
and cerebellum. We identify nine progenitor populations physically proximal to the telencephalon, suggesting more heteroge-
neity than previously described, including a highly prevalent mesenchymal-like population that disappears once neurogenesis
begins. Comparison of human and mouse progenitor populations at corresponding stages identifies two progenitor clusters
that are enriched in the early stages of human cortical development. We also find that organoid systems display low fidelity to
neuroepithelial and early radial glia cell types, but improve as neurogenesis progresses. Overall, we provide a comprehensive
molecular and spatial atlas of early stages of human brain and cortical development.
NATURE NEUROSCIENCE | VOL 24 | APRIL 2021 | 584–594 | www.nature.com/natureneuroscience
584
Content courtesy of Springer Nature, terms of use apply. Rights reserved

ResouRce
NATuRe NeuROsCieNCe
bulk RNA-seq experiments8 (Supplementary Fig. 1). In total, we col-
lected 289,000 cells passing quality control.
Across brain regions, hierarchical analysis initially identified two
main cell classes; progenitors and neurons. To identify the develop-
mental region-specific gene signatures for each brain area, includ-
ing the hindbrain, midbrain, thalamus, ganglionic eminences and
cortex, we performed differential gene expression across areas at
each age (Supplementary Tables 2–4). Using samples at CS22, we
generated data-driven regional signatures and explored when the
most structure-specific genes emerged. We observed that charac-
teristic transcription factor expression (such as HOX genes (hind-
brain)9; PAX7 (midbrain)10; GBX2 (thalamus)11; NKX2-1 (medial
ganglionic eminence)12 and FOXG1 (cerebral cortex)13) segre-
gated these regions from one another as early as CS13. However,
the cell type-defining gene expression programs were largely con-
served across brain regions, resulting in subsets of progenitors
or neurons that transcriptomically appear similar across regions
(Supplementary Fig. 2b). This was reflected by the fact that coclus-
tering at the early stages could not segregate regional identities, but
that this became possible at later stages. At the earliest timepoint,
CS12, the differences between brain regions were minimal and did
not resemble the more advanced region-specific programs that were
identifiable later in the first trimester (Supplementary Figs. 2 and 3).
Single-cell sequencing of the telencephalon. To better understand
the early stages of cortical development, we focused on the 59,000
cells in our dataset that originated from the telencephalon (earliest
dissections) and the cortex specifically (when it was identifiable for
subdissection). Clustering this data revealed minimal batch effects
as most clusters contained contributions from multiple individuals,
and clustering segregated the samples based on early and late first
trimester stages (Fig. 1 and Supplementary Fig. 4). Each of the 63
identified clusters could be assigned to a cell type identity of neu-
roepithelial cells, radial glia, intermediate progenitor cells (IPCs),
neurons or mesenchymal-like cells. To differentiate between the
radial glia and neuroepithelial cells in our transcriptomic data, we
examined whether SOX2-positive progenitor clusters contained evi-
dence of neurogenic genes, considered to be a criteria of radial glia
identity14. SOX2-positive progenitors that did not have these char-
acteristics were labeled neuroepithelial cells. Additionally, we noted
‘other’ cell populations of endothelial cells, microglia and pericytes
(Supplementary Tables 5,6). Furthermore, we observed that the
mesenchymal cell population diminished over the course of the
first trimester. Comparison of these clusters to previously published
single-cell data, including a small number of first trimester cells,
showed a strong cluster-level correlation, indicating that our larger
dataset presented here recapitulated populations that were previ-
ously observed. Additionally, in our previous analysis, some clusters
from these early samples were marked ‘unknown’ in identity15; our
correlation analysis is sufficient to now assign cell type identities to
these clusters (Supplementary Fig. 5).
We observed that although neurogenesis ramps up by the end
of the second trimester, even the earliest samples contained a small
number of neurons (Fig. 1a). It seemed unlikely that they could be
migratory populations from other cortical regions since migratory
interneurons have not been identified until later developmental
timepoints16. Presumably these neurons were produced locally by
direct neurogenesis from radial glia that occurs early in nonhuman
models of cortical development as well as human cortical organ-
oids17, and would be consistent with the observation that the IPCs
that mediate indirect neurogenesis are nearly absent until later in
the first trimester (after CS16). However, without lineage tracing,
the inference that the earliest neuronal populations are a product
of direct neurogenesis remains a working hypothesis. Subclustering
of the neuronal populations resulted in subtype clusters, some of
which were strongly enriched in either younger samples (<CS16,
presumed to result from direct neurogenesis) or older samples
(>CS16) (Supplementary Fig. 6 and Supplementary Table 7). These
included expected differences such as clusters marked by NHLH1,
which was significantly higher in the older samples and has been
associated with newborn neurons derived from IPCs18. Other genes
significantly enriched in older samples included NEUROD6 and
BCL11B that are associated with neuronal and deep layer identity14,
as well as CALB2 that is expressed in migratory ventrally derived
interneurons and a subset of excitatory neurons19. We noted signifi-
cantly higher expression of MEF2C in the younger samples. This has
been identified as a regulator of early neuronal differentiation and
layer formation20, but is also a factor involved in synaptic matura-
tion at later stages21. We validated high expression of MEF2C at the
earliest timepoints using in situ hybridization, although some of the
expression was extra-cortical. By CS22 its expression had dimin-
ished, but MEF2C was highly expressed again by mid-gestation at
gestational week 14 (Supplementary Fig. 7). This expression pattern
is intriguing given the role of MEF2 transcription factors in regulat-
ing apoptosis22.
We also observed clusters marked by previously undescribed
genes, including the younger sample-specific LHX5-AS1 cluster.
We validated that LHX5-AS1 was strongly enriched at these early
timepoints with broad expression at CS13, but restriction to the
developing cortical plate by CS14 and CS16 (Supplementary Fig. 8).
LHX5-AS1 RNA had the same expression pattern as LHX5 protein
(Supplementary Fig. 9), indicating that it may play a repressive role
to the protein, which has been characterized in Cajal–Retzius cell
development23 but has largely remained unstudied in these early
cell populations. The remaining clusters consisted of expected
early-born neuronal populations including Cajal–Retzius cells
marked by RELN (Supplementary Fig. 10), and subplate cells marked
by TLE4 and NR4A2. The subplate clusters were unexpectedly het-
erogeneous (Supplementary Fig. 6c,d) and contained marker genes
not previously associated with subplate identity (Supplementary
Table 7), suggesting our analysis may provide additional character-
ization of early neuronal cell type gene expression patterns.
We also sought to describe the spatial organization of the cortical
populations across development. Thus, we performed immuno-
staining validation in five individuals from the first trimester com-
pared to an early second trimester sample at gestational week 14.
Fig. 1 | Cell types in the early human cortex. a, scRNA-seq of early cortical development. UMAP plot of 58,145 telencephalon or cortical dissections
colored by annotated cell type. Feature plots of markers of broad progenitors (SOX2), radial glia (NES), IPCs (PPP1R17) and neurons (BCL11B) are shown.
Stacked bar chart shows cell type composition at earliest (CS12–13), middle (CS14–16) and late (CS19–22) first trimester. b, Spatial immunostaining of
early cortical samples. Immunostaining for main cell type markers across first trimester stages. Because of limited sample availability, each sample was
immunostained once. Nuclei shown in blue (4,6-diamidino-2-phenylindole (DAPI); dotted line demarcates cortical span), newborn neurons marked by
DCX (green), progenitors by SOX2 (red), IPCs by TBR2 (yellow) and maturing neurons by CTIP2 (cyan). Scale bars, 50μM; in CS13, 25μM. c, RNA velocity
demonstrates streams of direct and indirect neurogenesis. UMAP colored by age shows the segregation of samples by early and late first trimester stages.
RNA-velocity trajectories are depicted by gray arrows in the middle UMAP plot with underlying color by cell types as annotated in a. Line thickness of the
arrows indicates the differences in gene signature between cell types, and red arrows show predicted direct and indirect neurogenesis trajectories. The
velocity plots on the right show the intensity of scored velocity for a progenitor gene, VIM, and a neuronal gene, MEF2C. Velocities highlight the distinction
between progenitors and neurons in the clustering and velocity analyses.
NATURE NEUROSCIENCE | VOL 24 | APRIL 2021 | 584–594 | www.nature.com/natureneuroscience 585
Content courtesy of Springer Nature, terms of use apply. Rights reserved

ResouRce NATuRe NeuROsCieNCe
We observed FOXG1 staining as early as CS16, but not earlier,
as has been described previously24 (Supplementary Fig. 11). FOXG1
staining ensured that we were identifying cell types in the develop-
ing forebrain. We further defined cortical regional and progenitor
identity by positive PAX6 expression and positive SOX2 expression.
PAX6 expression has previously been identified as a determinant for
neuroectoderm fate25 and SOX2 as a marker for stem cells. We were
also confident of cortical identity based on the anatomical presence
of the optic cups (Supplementary Fig. 12) just ventral, lateral and
caudal, as well as the nasal ridge, (Supplementary Fig. 12) located
just ventral to our regions of interest. Tilescan images of these mark-
ers are available for download and exploration in the image browser
a
IPC
Mesenchymal
Neuroepithelia
Radial glia
Neuron
Other
b
SOX2 NES
BCL11BPPP1R17
CS12
1050–5–10
–10
–5
0
5
10
UMAP 2
15
CS13
CS14
CS15
CS19
CS20
CS22
CS12–13
CS14–15
> CS16
0
0.5
1.0
Fraction of cells
Cell type composition during
early cortical development
IPC
Mesenchymal
Neuroepithelia
Radial glia
Neuron
Other
CS13 CS14
CS16 CS18
cVIM
Potential
indirect neurogenesis
Potential
direct neurogenesis
Potential
direct neurogenesis
MEF2C
DAPI DCX SOX2 TBR2 CTIP2
DAPI DCX SOX2 TBR2 CTIP2
DAPI DCX SOX2 TBR2 CTIP2
DAPI DCX SOX2 TBR2 CTIP2
UMAP 1
NATURE NEUROSCIENCE | VOL 24 | APRIL 2021 | 584–594 | www.nature.com/natureneuroscience
586
Content courtesy of Springer Nature, terms of use apply. Rights reserved

ResouRce
NATuRe NeuROsCieNCe
we created in conjunction with this study (https://cells-test.gi.ucsc.
edu/?ds=early-brain, images tab).
In concordance with our transcriptomic data, we observed
prevalent SOX2 staining as early as CS13 that persisted through
gestational week 14. Newborn neurons marked by doublecortin
(DCX) were identifiable as early as CS14, but markers of matur-
ing neuronal identity such as BCL11B (CTIP2) emerged after CS16.
As expected, IPCs identified by TBR2 staining did not emerge until
CS18 in the dorsal telencephalon26 (Fig. 1b and Supplementary Fig.
13), although we and others have observed TBR2 staining at earlier
timepoints (CS16) in the ventral telencephalon27.
RNA-velocity analysis of lineage. To characterize the lineage
relationship between chief cortical cell types, we performed
RNA-velocity analysis using the scVelo algorithm28. This algorithm
incorporates messenger RNA levels and inherent expression vari-
ability of individual cells to infer steady states that contribute to
potential lineage relationships. Across cortical cell types, the most
apparent lineage stream originated in progenitor cell populations
from older samples and followed a trajectory through IPCs to neu-
rons (Fig. 1c). This trajectory of neurogenesis is well described and
suggests that the stereotypical process of neuronal differentiation
emerges after CS19. At earlier timepoints, there were local examples
of radial glia giving rise to neurons, suggestive of potential direct
neurogenesis and also lineage relationships between the progenitors
themselves. In addition to velocity streams related to the excitatory
lineage, there was also a stream from a collection of putative blood–
brain barrier cells toward the middle mesenchymal mass. Given the
age of these samples, this stream is suggestive of the onset of vas-
culogenesis that occurs at this time29. This is supported by further
velocity analysis demonstrating an increase in velocity (indicated in
red) of presumptive endothelial (FN1) and pericyte (RGS5) mark-
ers, but not markers of microglia identity (AIF1) or the marker of
the mesenchymal population (LUM) (Supplementary Fig. 4d).
The transition from neuroepithelial cells to radial glia has tradi-
tionally been characterized by the reorganization of tight junctions
and the appearance of nestin RC2 (NES) immunoreactivity30. We
sought to visualize this process and its transition across our samples.
Using NES as a marker of radial glia, TJP1 (ZO-1) as an indica-
tor of neuroepithelial cells, and SOX2 as a label for all progenitor
populations, we saw small numbers of nestin-positive radial glia at
CS14, with substantial upregulation by CS22 that simultaneously
corresponded to a waning of ZO-1 (Fig. 2a). By gestational week
14, ZO-1 staining dissipated and was largely constrained to pre-
sumed vascular structures. During the first trimester, we found a
progressive, and incremental, shift away from neuroepithelial iden-
tity (Supplementary Fig. 14). To better understand the heterogene-
ity and spatiotemporal trajectories of neuroepithelial and radial glia
progenitor populations, we subclustered the progenitors. Because
the mesenchymal population also expressed strong progenitor cell
markers including VIM and SOX2, we removed all neurons, IPCs
and other cell populations (including microglia, pericytes and endo-
thelial cells) and subclustered the remaining cells. The resulting
analysis yielded nine subclusters all marked by strong VIM and
SOX2 expression (Fig. 2b). Each of the nine clusters contained some
cells from all first trimester age samples, suggesting that the veloc-
ity and maturation trajectories were not purely a function of age
(Supplementary Fig. 15b and Supplementary Table 8). To examine
if these trajectories correspond to expected cell identity transfor-
mations from neuroepithelial cells to radial glia, we explored the
expression of HES5 and FGF10, which have been described to
mediate the transition between these progenitor populations31,32.
As expected, HES5 was strongly enriched in radial glia compared
to other cell types, and FGF10 peaked in neuroepithelial popula-
tions (Supplementary Fig. 15). We explored whether any of these
progenitors expressed gene signatures that defined excitatory neu-
rons across cortical areas, and found that three-quarters of the pro-
genitors uniquely expressed an areal identity while the rest were
unspecified (Supplementary Fig. 15). We additionally performed
weighted gene coexpression network analysis (WGCNA) of the pro-
genitor clusters as an orthogonal metric to identify distinct popula-
tions15,33. We found strong correspondence between clusters derived
from WGCNA analysis and our nine progenitor subtypes, and also
found strong enrichment for radial glia-like networks in later stage
samples (Supplementary Fig. 15 and Supplementary Table 9).
To examine whether this progenitor distribution was distinct
at the earliest timepoints of our dataset, we subclustered the CS12
and CS13 samples individually. From this analysis we identified 29
populations that were characterized by SOX2 (progenitor), TOP2A
(dividing), LHX5-AS1 (previously uncharacterized cell population)
and LUM (mesenchymal cells). Comparison of these subpopula-
tions to the nine progenitor subtypes identified across all samples
demonstrated strong correspondence to these nine cell types, with
additional heterogeneity in the CS12 and CS13 mesenchymal
populations (Supplementary Fig. 5b and Supplementary Table 10).
Velocity analysis identified a strong gradient across the populations
from cluster 2 toward clusters 1 and 6. These data strongly indi-
cated a gradient of identity across the nine progenitor populations
(Supplementary Fig. 15d). We characterized the genes across all our
velocity analyses that most strongly influenced the velocity stream
(and thus would be hypothesized to be the chief regulators of cell
fate transition). The drivers of velocity were enriched for genes that
have been associated with neurodevelopmental or psychiatric dis-
orders, including autism, cortical malformations and schizophrenia
(Supplementary Table 11), indicating that very early events in corti-
cal development may result in susceptibility or vulnerability to these
disorders. More work at these early timepoints is required to fully
characterize the developmental implications.
Progenitor heterogeneity. To explore the timing and spatial local-
ization of each of the progenitor subpopulations, we performed
immunostaining of the most specific gene markers for each clus-
ter (Supplementary Figs. 7–10 and Supplementary Figs. 16–26).
Each of these immunostainings is available as a tilescan of the
entire brain section in our image browser (https://cells-test.gi.ucsc.
edu/?ds=early-brain, images tab). NTRK3 was a highly specific
Fig. 2 | Early progenitors can be divided into nine progenitor subtypes. a, Transition from primarily neuroepithelial to radial glia progenitor identity.
Immunostaining of early first trimester samples (including CS14, shown here) show strong staining for all progenitors (SOX2, red), as well as tight
junctions (ZO-1, cyan), but limited staining for radial glia (NES, green). By CS22, NES staining increases substantially, ZO-1 decreases and SOX2 expression
is maintained. Scale bars, 50μM. b, scRNA-seq identifies nine progenitor subpopulations. Left: a UMAP plot depicts subclustered progenitor cells; middle:
the velocity trajectory across progenitor subtypes. Right: feature plots show high expression of VIM and SOX2 marking all progenitor populations.
c, NTRK3 marks progenitor populations before becoming a neuronally enriched marker. Progenitor cluster 3 is specifically and uniquely labeled by NTRK3,
as shown in the violin plot. Immunostaining for NTRK3 (cyan) shows early labeling of SOX2 (red) positive progenitors at CS16 (white arrows), but
expression shifts to more closely coincide with newborn neurons marked by DCX (green) by CS18 (white arrows). Scale bars, 50μM. d, DLK1 marks a
subset of early progenitors. DLK1 (cyan) is the top marker for progenitor cluster 8 and is exclusively expressed in early first trimester, as shown in the violin
plot on the left. Immunostaining for DLK1 at CS16 shows colocalization with low SOX2 (red) expressing cells at the boundary of the cortical edge. This
staining disappears from the cortex entirely by CS18 when DCX (green) staining emerges. For all panels, each sample was immunostained once.
NATURE NEUROSCIENCE | VOL 24 | APRIL 2021 | 584–594 | www.nature.com/natureneuroscience 587
Content courtesy of Springer Nature, terms of use apply. Rights reserved

ResouRce NATuRe NeuROsCieNCe
marker of Progenitor Cluster 3, and intrigued us because it is com-
monly described as a TrkC receptor that enables survival of specific
neuronal populations34 and has broad neocortical expression at later
stages of mouse cortex development35. We observed that at early
stages (CS14, CS16), NTRK3 broadly colocalized with SOX2-positive
progenitor cell populations (Fig. 2c and Supplementary Fig. 16),
but after CS18, it shifted from progenitor to neuronal expression
as would be expected. DLK1 was a top marker gene for Progenitor
a
bVIM
SOX2
CS16
CS18
0
1
2
3
4
4
0
–4
–8 –4 0
UMAP_1
UMAP_2
4 8
123456789
NTRK3
Count distribution
Count distribution
c
CS14 CS22
0
1
2
3
4
5
Early Middle Late
DLK1
CS16
CS18
d
Progenitor cluster 1
Progenitor cluster 2
Progenitor cluster 3
Progenitor cluster 4
Progenitor cluster 5
Progenitor cluster 6
Progenitor cluster 7
Progenitor cluster 8
Progenitor cluster 9
NES SOX2 ZO-1 NES SOX2 ZO-1
DAPI DCX SOX2 DLK1
DAPI DCX SOX2 DLK1
DAPI DCX SOX2 NTRK3
DAPI DCX SOX2 NTRK3
NATURE NEUROSCIENCE | VOL 24 | APRIL 2021 | 584–594 | www.nature.com/natureneuroscience
588
Content courtesy of Springer Nature, terms of use apply. Rights reserved

ResouRce
NATuRe NeuROsCieNCe
CS16
a
CS22 CS22
ALX1
b
1
6
Expression level
IPC
Mesenchymal
Neuroepithelial
Neuronal
Other
Radial glia
CS12 13 14 15 19 20 22
1
4
Expression level
IPC
Mesenchymal
Neuroepithelial
Neuronal
Other
Radial glia
CS12 13 14 15 19 20 22
CS16
LUM
c
H28126 week 4 H28126 week 4
H1 week 7 13,324 week 10
DAPI SOX2 LUM DAPI PAX6 ALX1
DAPI SOX2 LUM
DAPI SOX2 LUM
DAPI SOX2 LUM
DAPI PAX6 ALX1
DAPI PAX6 ALX1
DAPI PAX6 ALX1
Fig. 3 | scRNA-seq identifies early mesenchymal cell population. a, LUM and ALX1 are mesenchymal cell type and early sample markers. In both
progenitor and cortex clusterings, LUM marks a separate population of cells, as shown by the feature plot on the left. LUM expression is highly specific to
the mesenchymal cell type and is enriched in early samples. ALX1 expression is highly correlated to LUM as shown in the right feature plot, and is similarly
enriched in early, mesenchymal populations. b, LUM is widely expressed early and diminishes later. Immunostaining for LUM (cyan) shows prevalent
expression in and between progenitors marked by SOX2 (red) at CS16, but this expression dissipates by CS22. However, expression of LUM does not begin
until week 7 in the H1 organoid. Scale bars are 50μM. c, ALX1 is sparsely expressed early and diminishes later. Immunostaining for ALX1 (cyan) shows sparse
expression in PAX6 (green) positive cells; it disappears from the cortex but is expressed in surrounding brain structures at CS22. ALX1 expression does not
begin until week 10 in the 13,234 cerebral organoids. For all panels, each sample was immunostained once. Scale bars, 50μM.
NATURE NEUROSCIENCE | VOL 24 | APRIL 2021 | 584–594 | www.nature.com/natureneuroscience 589
Content courtesy of Springer Nature, terms of use apply. Rights reserved

ResouRce NATuRe NeuROsCieNCe
Cluster 8 and also was highly enriched in early samples. Although
much of the DLK1 staining was in stromal cells peripheral to the
cortex, as has been previously reported36, we observed a small num-
ber of DLK1-positive cells that expressed low SOX2-levels at early
stages, such as CS16, after which DLK1 expression completely dis-
appeared (Fig. 2d and Supplementary Fig. 17). DLK1 is a nonca-
nonical ligand of Notch that inhibits its function and affects Notch
1 receptor distribution37, although it is not widely expressed in the
mammalian central nervous system during development38. DLK1
has also been described to play a role in enabling postnatal neu-
rogenesis in the mouse39, and may play a similar regulatory role
at the earliest neurogenic timepoints of human cortical develop-
ment. Both NTRK3 and DLK1 may play important roles in recep-
tor/ligand communication, as integration with other brain regions
increases (Supplementary Table 12). Across progenitor populations
we describe known and new gene expression patterns that define
the neuroepithelial and radial glia subpopulations and suggest that
there exists an expression gradient that marks the transition from
early to more mature progenitor populations.
Both in our initial clustering and in the more focused progeni-
tor analysis, a mesenchymal-like population labeled by the gene
LUM was distinctly segregated from other progenitor populations.
Lumican, the protein encoded by LUM, is an extracellular matrix
protein that is widely present in mesenchymal tissues throughout
the adult body40. It has also been used to promote folding in cor-
tical structures when added exogenously to media41. In our data,
LUM expression is highly enriched in the mesenchymal cell popula-
tion and diminishes substantially after early developmental time-
points (Fig. 3a). Another marker of this cluster was ALX1, which
also has one of the most correlated gene expression patterns to LUM
(Fig. 3b). ALX1 has been described as required in knockout mice
for the development of the forebrain mesenchyme42, but has not
been studied in humans. We observed prevalent LUM staining in
our samples at or before CS16, with ALX1 at the edges of the fore-
brain, including several colocalized PAX6-positive cells (Fig. 3c,d
and Supplementary Figs. 18 and 19). ALX1 staining substantially
diminishes later in development.
Canonical developmental signaling pathway activation. Because
canonical signaling pathways have been characterized for their role
in patterning the human brain and cortex, we sought to explore the
dynamics of their expression patterns in human cortical progeni-
tors. scRNA-seq data showed dynamic expression of the genes that
have been described as part of the FGF, Wnt, mTOR and Notch sig-
naling pathways (Fig. 4a) in cortical progenitors. The FGF signaling
family has been described to assign rostral identity in the develop-
ing neural tube. But FGF also plays a role in the transition to and
maintenance of the radial glia progenitor pool, in part promoted
by FGF10, by permitting the transition of neuroepithelial cells to
radial glia and by preventing the transition into intermediate pro-
genitors32. Furthermore, FGF signaling interacts with the Notch sig-
naling pathway during cortical development. Classic studies have
demonstrated that Notch signaling is generally required to preserve
stem cell pools. Constitutive Notch 1 activation increases radial
glial generation and leads to a decrease in the expression of pro-
neural genes, such as Neurogenin-2 (Ngn2)43. Consistent with this
observation, here we find that Notch 1 staining peaks and is largely
restricted to the ventricular zone at CS14 and CS16 (Fig. 4d).
mTOR activity in the developing forebrain has largely been
uncharacterized at early developmental timepoints, and descrip-
tions of its function have been restricted to its role in regulating
oRG cells15. However, upregulation of CDC42-dependent mTOR
signaling has been shown to be sufficient to generate neural pro-
genitors, mediated through increased HES5 and PAX6 expression44,
suggesting that mTOR signaling early on may be important for
driving the switch from neuroepithelial stem cell to radial glia.
We find that expression of phosphorylated S6 is highest in the earliest
samples and then diminishes later in the first trimester (Fig. 4c),
further supporting a role in the neuroepithelial to radial glia fate
transition. Furthermore, Wnt signaling has also been implicated in
medial-lateral patterning of the neural tube and in cell fate tran-
sitions. Constitutive Wnt activation leads to opposing actions of
increasing the progenitor pool by preventing neurogenesis but also
of increasing the neuronal pool by promoting IPC differentiation45.
This suggests that Wnt activity is variable and cell-dependent in
cortical development. In this study, we discover that active Wnt
signaling (phosphorylated β-catenin) peaks at CS16 and diminishes
subsequently (Fig. 4b). These data further clarify our understanding
of signaling in patterning cortical progenitors, with important impli-
cations for better modeling early stages of cortical development.
Conservation of progenitor populations across species. To evalu-
ate the similarities and differences between early forebrain develop-
ment in human and rodent, we performed scRNA-seq of the mouse
forebrain at embryonic days (E) 9 and 10, which correspond to
Theiler stages (TS) 14 and 16. Although the onset of neurogenesis
differs between human and mouse46, we used immunostaining to
verify that at these mouse stages, the forebrain was already express-
ing FOXG1 and undergoing neurogenesis, as marked by the presence
of DCX (Supplementary Figs. 11 and 13). To compare mouse and
human progenitor populations, we performed clustering and corre-
lation analysis between the mouse clusters and our nine human pro-
genitor populations. Using a correlative threshold, we observed that
seven of the nine human populations had at least one corresponding
cluster in the mouse single-cell data (Fig. 5a). However, there were
no high correlations for Progenitor Cluster 4 (marked most highly
by C1orf61) or Progenitor Cluster 7 (marked most highly by ID4).
To test the hypothesis that these populations were underrepresented
in the mouse data, we used immunostaining to explore the expres-
sion of the main progenitor population markers in the mouse.
Indeed, no fluorescent in situ signal for C1orf61 or immunostaining
signal for Id4 was identifiable in the mouse forebrain at TS14 or 16
(Fig. 5b and Supplementary Figs. 20–22). Although the TS14 and 16
samples were more advanced in terms of neurogenesis, they may be
at timepoints that are relatively immature in other ways compared
to our first trimester samples. To account for this difference, we also
explored the single-cell trajectory of Id4 from E9–E10 (our data) to
E11.5–E17.5 (ref. 47). In the mouse, Id4 peaks at E11.5 and is low
after E13, whereas human ID4 is sustained in its expression pattern
after CS13 (Fig. 5d). By contrast, both C1orf61 RNA and ID4 pro-
tein were detectable in macaque embryonic tissue and chimpanzee
induced pluripotent stem cell-derived organoids (Supplementary
Fig. 23). C1orf61 (CROC-4) has been characterized to be widely
expressed during late developmental stages and in the adult brain,
regulating c-FOS signaling48, but is otherwise uncharacterized. In
contrast, Id4 has been identified in the developing mouse cortex at
later stages (E15.5). ID4-deficient mice exhibit impaired brain size
and mistimed neurogenesis49, suggesting that earlier expression in
human progenitors compared to mouse could indicate a potential
mechanism by which early progenitors expand more rapidly in
humans than in mice.
Despite the general conservation of cell types between the early
development of the human and mouse cortex, the absence of spe-
cific human cell populations at early developmental stages in the
mouse suggests that alternatives to mouse are required to fully
model the earliest cell types involved in human cortical develop-
ment. Cortical organoids are an attractive model because they can
be generated from normal or patient-derived induced pluripotent
stem cells and are amenable to genetic modulation. We previously
generated an extensive catalog of single-cell sequencing data from
multiple organoids derived from four pluripotent stem cell lines
using three protocols analyzed from weeks 3 to 24 (ref. 7). Using
NATURE NEUROSCIENCE | VOL 24 | APRIL 2021 | 584–594 | www.nature.com/natureneuroscience
590
Content courtesy of Springer Nature, terms of use apply. Rights reserved

ResouRce
NATuRe NeuROsCieNCe
108,593 cells derived from weeks 3 and 5 of this dataset, we explored
the fidelity of organoid cell types to early human cortex. The pres-
ervation of excitatory neuronal identity mirrored that of later devel-
opmental stages (roughly 0.5, as we recently reported). However, the
remaining primary cell types had much lower fidelity in organoids
(average of 0.28) (Fig. 5c). These data indicate that early cell types in
organoids do not resemble their corresponding cortical progenitor
counterparts, although the fidelity of cell types does improve at later
stages of organoid culture. The two populations that were not well
represented in mice at TS14 and 16 were identifiable in organoid
cultures as evidenced by fluorescent in situ probing for C1orf61 and
immunostaining for ID4 (Supplementary Fig. 23).
0
–1
–2
Pathway signal
WNT
Notch
FGF
mTOR
Age (CS)
CS13
CS14
CS16
CS22
CS13
CS14
CS16
CS22
CS13
CS14
CS16
CS22
mTOR
12.5
2
1
15.0 17.5 20.0 22.5
Wnt
Notch
DAPI SOX2 pS6
ab
cd
DAPI SOX2 NICD/Notch1
DAPI SOX2 pβCatenin
DAPI SOX2 pβCatenin
DAPI SOX2 pβCatenin
DAPI SOX2 pβCatenin
DAPI SOX2 NICD/Notch1
DAPI SOX2 NICD/Notch1
DAPI SOX2 NICD/Notch1
DAPI SOX2 pS6
DAPI SOX2 pS6
DAPI SOX2 pS6
Fig. 4 | Signaling pathway oscillations in the first trimester human cortical progenitors. a, Signaling pathway oscillations in progenitor RNA expression
patterns of the FGF (green), Wnt (purple), mTOR (red) and Notch (blue) signaling pathways, as defined by KEGG pathway designations in progenitors
across ages sampled in this study. Error shading surrounding each bar indicates the loess regression 5–95% confidence interval. b, Wnt activity, as indicated
by phosphorylated β-catenin (cyan), is highest in the CS14 and CS16 samples and colocalizes with progenitors (SOX2, red), but dissipates by CS22.
c, mTOR activity, as indicated by phosphorylated S6 ribosomal protein (cyan), primarily localizes along the cortical plate (DCX, green) in the youngest
samples and diminishes by CS22. d, Notch activity, as indicated by cleaved Notch intracellular domain (NICD) of Notch 1, peaks in expression at CS14 and
localizes mainly to the ventricular zone. Nuclei are labeled by DAPI (blue). For all panels, each sample was immunostained once. Scale bars, 50μm.
NATURE NEUROSCIENCE | VOL 24 | APRIL 2021 | 584–594 | www.nature.com/natureneuroscience 591
Content courtesy of Springer Nature, terms of use apply. Rights reserved

ResouRce NATuRe NeuROsCieNCe
Discussion
Here, we present a comprehensive overview of scRNA-seq from the
first trimester of human development. As these samples are rarely
accessible, we seek to present both the single-cell sequencing as well
as the immunostaining validation provided as part of this study as
a community resource. Our analysis highlights the granular gene
expression programs that emerge across brain structures and across
cell types in the human neocortex as development progresses, and
provides a dataset by which other model organisms and in vitro
cortical organoids can be compared to their primary human coun-
terparts. The tissues used for the purpose of this study are fragile,
and dissections rely on morphological hallmarks without knowl-
edge of expression patterns before sample collection. As such, the
single-cell analysis becomes even more essential as a tool to explore
a
1.00
4.00
24.00
32.00
2.00
17.00
28.00
3.00
9.00
26.00
27.00
14.00
21.00
13.00
31.00
15.00
37.00
7.00
12.00
10.00
29.00
8.00
5.00
34.00
6.00
11.00
19.00
35.00
30.00
18.00
25.00
33.00
16.00
38.00
22.00
20.00
23.00
36.00
Progenitor 3, NTRK3
Progenitor 6, Dividing
Progenitor 9, CDH2
Progenitor 2, EEF1G
Progenitor 4, C1orf61
Progenitor 8, DLK1
Progenitor 7, ID4
Progenitor 1, CD44
Progenitor 5, ALX1/LUM
Mouse clusters
Human progenitor clusters
–1.00
E17.5E13.5E11.5E10E9
22201915141312
0
2
4
6
0
1
2
3
4
1.00
0
Pearson’s correlation
CS16 (human) CS16 (human)
TS16 (mouse)
bc
TS16 (mouse)
Human ID4 expression Mouse ID4 expression
CS Embryonic day
d
DAPI C1orf61 RNA
DAPI C1orf61 RNA DAPI SOX2 ID4
DAPI SOX2 ID4
Fig. 5 | Mouse models different aspects of early cell types. a, Comparison of mouse forebrain clusters to primary progenitor cell types. Heatmap showing
the comparison of mouse forebrain clusters from 16,053 cells to the primary progenitor populations identified in this dataset. Correlations are performed
using Pearson correlations between cluster marker sets and identify that progenitor clusters 4 and 7 do not have a counterpart in mouse data. b, C1orf61
is widely expressed in early human but not early mouse progenitors. Fluorescent in situ hybridization of human samples at CS16 shows broad expression
of C1orf61 (cyan) in progenitor cells labeled by SOX2 (red). Parallel in situ staining in TS16 mouse shows no C1orf61 expression. Scale bars, 50μM. c, ID4
is widely expressed in early human but not early mouse progenitors. Immunostaining of human samples at CS16 shows broad expression of ID4 (cyan) in
progenitor cells labeled by SOX2 (red). A parallel staining in TS16 mouse shows no ID4 expression. Scale bars, 50μM. d, Violin plots of ID4 RNA expression
across several CSs (human) and embryonic days (mouse). ID4 RNA expression in the human persists onward from CS13. However, ID4 RNA expression in
the mouse peaks at E11.5 (TS20) and dissipates. For all panels, each sample was immunostained once.
NATURE NEUROSCIENCE | VOL 24 | APRIL 2021 | 584–594 | www.nature.com/natureneuroscience
592
Content courtesy of Springer Nature, terms of use apply. Rights reserved

ResouRce
NATuRe NeuROsCieNCe
and validate accurate regional identities. The gene signatures that
we describe and validate may offer additional markers that can be
used to mark these brain regions, as well as the transition from neu-
roepithelia to radial glia cell identity.
Our analysis of several brain regions highlights many similarities
across progenitors that were only mildly distinguished by the expres-
sion of regional-specific transcription factors. These similarities in
gene programs across the developing human brain suggest that the
process by which initially uniform stem cells give rise to neuronal and
glia heterogeneity characteristic of the adult brain are parallel across
brain structures and that insights from one region, including the
degree of heterogeneity and the trajectories of differentiation, may
be able to cross-inform identification of regulatory gene programs
across the brain. We aspired to annotate specific gene programs that
distinguish progenitor groups within the cerebral cortex. Instead,
we observed a gradient of expression patterns, with clear neuroepi-
thelial and radial glia cell populations on the ends of the spectrum,
indicating that the transition from one population to another may be
gradual rather than distinct. However, because our data are limited
by the snapshot view of each timepoint sampled, and by imperfect
representation of all possible timepoints along this continuum, this
conclusion will need to be further explored through additional sam-
pling, validation and eventual mechanistic examination.
One unexpected population that was identified in our progeni-
tor dataset was a population that appeared mesenchymal in origin
(marked by ALX1 and LUM) by both single-cell sequencing and
immunostaining, that comprised a large swath of the telencepha-
lon at the earliest timepoints that we sampled. We hypothesize that
the ALX1-positive cells may resemble previously described neural
crest-derived cell populations that give rise to the meninges50. These
cells may be secreting LUM as a structural component to support
the physical formation of the telencephalon before the emergence
of the radial glial scaffold. In support of this hypothesis, we found
that the ALX1-positive cells were in the presumptive stroma sur-
rounding the cortex, while LUM was detected throughout the cor-
tex. Although cortical organoids do not express ALX1 or LUM at
early stages, their expression emerges after 7 weeks of development
(Fig. 3b,c), further prompting questions about their function and
role in cortical development.
Comparisons of primary data presented here to the single-cell
sequencing of previous cortical organoid populations raises inter-
esting differences between the systems. We were surprised to find
limited cell type correspondence at the earliest stages of cortical
organoid generation, and increased fidelity once the radial glia and
neuronal populations emerged. This may indicate that terminal
identity does not depend on a specific differentiation path. However,
there were also differences in the timing and cell type composition
between early human and mouse populations, further highlighting
that no model system is ideal for the study of all biological questions.
Moreover, populations present in primary human tissue that were
not present in early organoids, including the LUM and ALX1 clus-
ters, could be identified in mice (Supplementary Figs. 18 and 19),
indicating that depending on the questions being investigated,
either in vivo mouse or in vitro human models, may be most appro-
priate for the study of neuroepithelial and early radial glia popula-
tions. Our description of early human cortical cell populations may
also enable further refinement of in vitro culture systems to better
reflect human neuroepithelial and radial glia populations. Together,
the data we present here represent a characterization of main cell
types across the first trimester of human brain development and
highlight the subpopulations of progenitor cells that form the basis
for creating the human cortex.
Accession codes. The data analyzed in this study were produced
through the Brain Initiative Cell Census Network (BICCN)
(RRID:SCR_015820) and deposited in the Neuroscience Multi-omic
(NeMO) Archive (RRID:SCR_002001), https://assets.nemoarchive.
org/dat-0rsydy7.
Online content
Any methods, additional references, Nature Research report-
ing summaries, source data, extended data, supplementary infor-
mation, acknowledgements, peer review information; details of
author contributions and competing interests; and statements of
data and code availability are available at https://doi.org/10.1038/
s41593-020-00794-1.
Received: 21 September 2020; Accepted: 23 December 2020;
Published online: 15 March 2021
References
1. Breunig, J. J., Haydar, T. F. & Rakic, P. Neural stem cells: historical perspective
and future prospects. Neuron 70, 614–625 (2011).
2. Wilson, S. W. & Houart, C. Early steps in the development of the forebrain.
Dev. Cell 6, 167–181 (2004).
3. Beccari, L., Marco-Ferreres, R. & Bovolenta, P. e logic of gene regulatory
networks in early vertebrate forebrain patterning. Mech. Dev. 130,
95–111 (2013).
4. Lui, J. H., Hansen, D. V. & Kriegstein, A. R. Development and evolution of
the human neocortex. Cell 146, 18–36 (2011).
5. Gotz, M., Hartfuss, E. & Malatesta, P. Radial glial cells as neuronal
precursors: a new perspective on the correlation of morphology and lineage
restriction in the developing cerebral cortex of mice. Brain Res. Bull. 57,
777–788 (2002).
6. Miller, F. D. & Gauthier, A. S. Timing is everything: making neurons versus
glia in the developing cortex. Neuron 54, 357–369 (2007).
7. Bhaduri, A. et al. Cell stress in cortical organoids impairs molecular subtype
specication. Nature 578, 142–148 (2020).
8. Cardoso-Moreira, M. et al. Gene expression across mammalian organ
development. Nature 571, 505–509 (2019).
9. Tumpel, S., Wiedemann, L. M. & Krumlauf, R. Hox genes and segmentation
of the vertebrate hindbrain. Curr. Top. Dev. Biol. 88, 103–137 (2009).
10. Agoston, Z., Li, N., Haslinger, A., Wizenmann, A. & Schulte, D. Genetic and
physical interaction of Meis2, Pax3 and Pax7 during dorsal midbrain
development. BMC Dev. Biol. 12, 10 (2012).
11. Chen, L., Guo, Q. & Li, J. Y. Transcription factor Gbx2 acts
cell-nonautonomously to regulate the formation of lineage-restriction
boundaries of the thalamus. Development 136, 1317–1326 (2009).
12. Du, T., Xu, Q., Ocbina, P. J. & Anderson, S. A. NKX2.1 species cortical
interneuron fate by activating Lhx6. Development 135, 1559–1567 (2008).
13. Tao, W. & Lai, E. Telencephalon-restricted expression of BF-1, a new member
of the HNF-3/fork head gene family, in the developing rat brain. Neuron 8,
957–966 (1992).
14. Molyneaux, B. J., Arlotta, P., Hirata, T., Hibi, M. & Macklis, J. D. Fezl is
required for the birth and specication of corticospinal motor neurons.
Neuron 47, 817–831 (2005).
15. Nowakowski, T. J. et al. Spatiotemporal gene expression trajectories reveal
developmental hierarchies of the human cortex. Science 358, 1318–1323 (2017).
16. Alzu’bi, A., Homman-Ludiye, J., Bourne, J. A. & Clowry, G. J.
alamocortical aerents innervate the cortical subplate much earlier in
development in primate than in rodent. Cereb. Cortex 29, 1706–1718 (2019).
17. Cardenas, A. et al. Evolution of cortical neurogenesis in amniotes controlled
by robo signaling levels. Cell 174, 590–606 e521 (2018).
18. Li, C. M., Yan, R. T. & Wang, S. Z. Misexpression of a bHLH gene, cNSCL1,
results in abnormal brain development. Dev. Dyn. 215, 238–247 (1999).
19. Conde, F., Lund, J. S., Jacobowitz, D. M., Baimbridge, K. G. & Lewis, D. A.
Local circuit neurons immunoreactive for calretinin, calbindin D-28k or
parvalbumin in monkey prefrontal cortex: distribution and morphology.
J. Comp. Neurol. 341, 95–116 (1994).
20. Li, H. et al. Transcription factor MEF2C inuences neural stem/progenitor
cell dierentiation and maturation in vivo. Proc. Natl Acad. Sci. USA 105,
9397–9402 (2008).
21. Harrington, A. J. et al. MEF2C regulates cortical inhibitory and excitatory
synapses and behaviors relevant to neurodevelopmental disorders. eLife
https://doi.org/10.7554/eLife.20059 (2016).
22. Pottho, M. J. & Olson, E. N. MEF2: a central regulator of diverse
developmental programs. Development 134, 4131–4140 (2007).
23. Miquelajauregui, A. et al. LIM-homeobox gene Lhx5 is required for normal
development of Cajal–Retzius cells. J. Neurosci. 30, 10551–10562 (2010).
24. Manuel, M. N. et al. e transcription factor Foxg1 regulates telencephalic
progenitor proliferation cell autonomously, in part by controlling Pax6
expression levels. Neural Dev. 6, 9 (2011).
NATURE NEUROSCIENCE | VOL 24 | APRIL 2021 | 584–594 | www.nature.com/natureneuroscience 593
Content courtesy of Springer Nature, terms of use apply. Rights reserved

ResouRce NATuRe NeuROsCieNCe
25. Zhang, X. et al. Pax6 is a human neuroectoderm cell fate determinant.
Cell Stem Cell 7, 90–100 (2010).
26. Subramanian, L., Bershteyn, M., Paredes, M. F. & Kriegstein, A. R. Dynamic
behaviour of human neuroepithelial cells in the developing forebrain.
Nat. Commun. 8, 14167 (2017).
27. Qin, J. et al. Early forebrain neurons and scaold bers in human embryos.
Cereb. Cortex 30, 913–928 (2020).
28. Bergen, V., Lange, M., Peidli, S., Wolf, F. A. & eis, F. J. Generalizing RNA
velocity to transient cell states through dynamical modeling. Nat. Biotech. 38,
1408–1414 (2020).
29. Virgintino, D. et al. Astroglia-microvessel relationship in the developing
human telencephalon. Int. J. Dev. Biol. 42, 1165–1168 (1998).
30. Aaku-Saraste, E., Hellwig, A. & Huttner, W. B. Loss of occludin and
functional tight junctions, but not ZO-1, during neural tube closure—
remodeling of the neuroepithelium prior to neurogenesis. Dev. Biol. 180,
664–679 (1996).
31. Hatakeyama, J. et al. Hes genes regulate size, shape and histogenesis of the
nervous system by control of the timing of neural stem cell dierentiation.
Development 131, 5539–5550 (2004).
32. Sahara, S. & O’Leary, D. D. Fgf10 regulates transition period of cortical stem
cell dierentiation to radial glia controlling generation of neurons and basal
progenitors. Neuron 63, 48–62 (2009).
33. Langfelder, P. & Horvath, S. WGCNA: an R package for weighted correlation
network analysis. BMC Bioinform. 9, 559 (2008).
34. Qian, X., Riccio, A., Zhang, Y. & Ginty, D. D. Identication and
characterization of novel substrates of Trk receptors in developing neurons.
Neuron 21, 1017–1029 (1998).
35. Tessarollo, L. et al. trkC, a receptor for neurotrophin-3, is widely expressed in
the developing nervous system and in non-neuronal tissues. Development
118, 463–475 (1993).
36. Pellegrini, L. et al. Human CNS barrier-forming organoids with
cerebrospinal uid production. Science https://doi.org/10.1126/science.
aaz5626 (2020).
37. Bray, S. J., Takada, S., Harrison, E., Shen, S. C. & Ferguson-Smith, A. C. e
atypical mammalian ligand delta-like homologue 1 (Dlk1) can regulate Notch
signalling in Drosophila. BMC Dev. Biol. 8, 11 (2008).
38. Miller, A. J. & Cole, S. E. Multiple Dlk1 splice variants are expressed during
early mouse embryogenesis. Int. J. Dev. Biol. 58, 65–70 (2014).
39. Ferron, S. R. et al. Postnatal loss of Dlk1 imprinting in stem cells and niche
astrocytes regulates neurogenesis. Nature 475, 381–385 (2011).
40. Chakravarti, S. Functions of lumican and bromodulin: lessons from
knockout mice. Glycoconj. J. 19, 287–293 (2002).
41. Long, K. R. et al. Extracellular matrix components HAPLN1, Lumican, and
Collagen I cause hyaluronic acid-dependent folding of the developing human
neocortex. Neuron 99, 702–719 e706 (2018).
42. Zhao, Q., Behringer, R. R. & de Crombrugghe, B. Prenatal folic acid
treatment suppresses acrania and meroanencephaly in mice mutant for the
Cart1 homeobox gene. Nat. Genet. 13, 275–283 (1996).
43. Gaiano, N., Nye, J. S. & Fishell, G. Radial glial identity is promoted by
Notch1 signaling in the murine forebrain. Neuron 26, 395–404 (2000).
44. Endo, M., Antonyak, M. A. & Cerione, R. A. Cdc42-mTOR signaling pathway
controls Hes5 and Pax6 expression in retinoic acid-dependent neural
dierentiation. J. Biol. Chem. 284, 5107–5118 (2009).
45. Harrison-Uy, S. J. & Pleasure, S. J. Wnt signaling and forebrain development.
Cold Spring Harb. Perspect. Biol. 4, a008094 (2012).
46. Bystron, I., Rakic, P., Molnar, Z. & Blakemore, C. e rst neurons of the
human cerebral cortex. Nat. Neurosci. 9, 880–886 (2006).
47. Yuzwa, S. A. et al. Developmental emergence of adult neural stem
cells as revealed by single-cell transcriptional proling. Cell Rep. 21,
3970–3986 (2017).
48. Jerey, P. L. et al. CROC-4: a novel brain specic transcriptional activator of
c-fos expressed from proliferation through to maturation of multiple neuronal
cell types. Mol. Cell. Neurosci. 16, 185–196 (2000).
49. Bedford, L. et al. Id4 is required for the correct timing of neural
dierentiation. Dev. Biol. 280, 386–395 (2005).
50. Choe, Y., Zarbalis, K. S. & Pleasure, S. J. Neural crest-derived mesenchymal
cells require Wnt signaling for their development and drive invagination of
the telencephalic midline. PLoS ONE 9, e86025 (2014).
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in
published maps and institutional affiliations.
Open Access This article is licensed under a Creative Commons
Attribution 4.0 International License, which permits use, sharing,
adaptation, distribution and reproduction in any medium or format, as
long as you give appropriate credit to the original author(s) and the source, provide a link
to the Creative Commons license, and indicate if changes were made. The images or
other third party material in this article are included in the article’s Creative Commons
license, unless indicated otherwise in a credit line to the material. If material is not
included in the article’s Creative Commons license and your intended use is not
permitted by statutory regulation or exceeds the permitted use, you will need to obtain
permission directly from the copyright holder. To view a copy of this license, visit http://
creativecommons.org/licenses/by/4.0/.
© The Author(s) 2021
NATURE NEUROSCIENCE | VOL 24 | APRIL 2021 | 584–594 | www.nature.com/natureneuroscience
594
Content courtesy of Springer Nature, terms of use apply. Rights reserved

ResouRce
NATuRe NeuROsCieNCe
Methods
Additional methodological details can also be found in the Nature Research
Reporting Summary included in the Supplementary Material.
Sample processing. Acquisition of all primary human tissue samples was approved
by the UCSF Human Gamete, Embryo and Stem Cell Research Committee
(approval nos. 10-03379 and 10-05113). All experiments were performed in
accordance with protocol guidelines. Informed consent was obtained before sample
collection and use for this study. First trimester human samples were collected
from elective pregnancy terminations through the Human Developmental Biology
Resource, staged using crown-rump length and shipped overnight on ice in
Rosewell Park Memorial Institute media or in 4% paraformaldehyde. Samples
were donated and deidentified for sex. Although we can infer sex for sequenced
samples, we have not accounted for it because it is an unclear measurement during
the first trimester. Fixed samples were used for downstream immunostaining and
imaging. Mouse samples (CD-1 IGS Mouse) were killed at E9 and E10, and staged
using the somite number. Mice were housed in shared housing, five mice to a cage
with a 12 light/12 dark cycle and temperatures of roughly 18–23 °C (65–75 °F) with
40–60% humidity. Equal numbers of male and female embryonic mice were used.
All mouse experiments were approved by and conducted according to the UCSF
Institutional Animal Care and Use Committee (protocol AN078703-03A). Half of
the mouse samples were randomly assigned for fixation in 4% paraformaldehyde
and the remaining live samples were used for dissociation. Live samples were
subdissected into identifiable regions and dissociated using papain (Worthington,
LK003150) with DNase. Samples were minced and incubated in 1 ml activated
papain for 15 min at 37 °C, according to the manufacturer’s instructions. Samples
were then inverted for several times and incubated for an additional 15 min. The
dissociated cells were centrifuged at 300g for 5 min and the papain was removed.
Macaque tissue sections (E64) were a gift from A. Pollen (UCSF), originally a gift
from A. Tarantal (UC Davis).
scRNA-seq. Single-cell capture was performed following the 10X v.2 Chromium
manufacturer’s instructions. Each sample was its own batch. For each batch,
10,000 cells were targeted for capture and 12 cycles of amplification for each of the
complementary DNA and library amplifications were performed. Libraries were
sequenced according to the manufacturer’s instructions on the Illumnia NovaSeq
6000 S2 flow cell (RRID:SCR_016387).
scRNA analysis. scRNA-seq data were aligned to the GRCh38-0.1.2 reference
genome, and cells were identified using CellRanger v.2 (RRID:SCR_017344).
Quality control removed cells with fewer than 500 genes per cell and cells with
greater than 10% mitochondrial content. Clustering and batch effect were
performed as has been previously described in these established methods15. Batch
effects (as pertaining to day of capture) were removed by normalizing all cells in
a batch to the most counts, and then multiplying by the median counts in that
batch. These normalized counts were merged together across samples and log2
normalization was performed. Using default parameters of Seurat v.2, variable
genes were identified. Batch was regressed out in the space of variable genes during
scaling, again using default Seurat v.2 parameters. For three samples, there was
only one individual per timepoint, which may have resulted in confounding age
and batch, but because of the strong batch effects that result from 10X data, we
moved ahead with the batch correction. In the space of the scaled, batch corrected
variable genes, we performed principal component analysis. Significant principal
components were carried forward, as identified using the calculation detailed in
a previous publication. Using the RANN package, the top ten nearest neighbors
were identified for each cell in the space of significant principal components. We
used a custom script to calculate the Jaccard distance of these neighbors, and used
igraph to perform Louvain clustering for each analysis. Clustering with batch
effect correction was performed for each dissected area, and also without batch
correction by each individual, both across all brain regions and within cortex only.
Cluster markers were interpreted and assigned cluster identity by using known
literature cell type annotations, or by associating progenitor and neuronal genes
to other identifiable details. Progenitors were distinguished as radial glia if they
expressed neurogenic genes, else they were determined to be neuroepithelial
(Supplementary Table 6).
Subclustering within the neurons was performed by selecting and clustering
the cells from neuronal clusters and repeating the clustering procedure with
batch effect correction. Progenitor cells were subclustered by removing neuronal,
IPC and other (microglia, endothelial and pericyte) clusters from the data. After
subclustering, neuronal populations were identified and removed iteratively until
54 subclusters were identified and did not include neuronal populations. These 54
subclusters were recombined by correlating marker genes to one another, and then
performing a dendrogram cut to generate the nine clusters presented here.
Correlational analysis between mouse and human data was performed as we
previously described7. Briefly, cluster markers were generated for each dataset
individually, and gene scores integrating the specificity and fold enrichment for
each marker was calculated. A matrix for every marker gene and gene score was
generated across all clusters and used for correlations. This was performed in the
space of all mouse forebrain clusters as compared to primary human progenitor
clusters (Fig. 5), and in the space of all organoid clusters compared to the full set of
primary human clusters in our dataset (Supplementary Fig. 23).
RNA velocity. Velocity estimates were calculated using the veloctyo.py v0.17 and
scVelo (RRID:SCR_018168) algorithms. Reads that passed quality control after
clustering were used as input for the velocyto command line implementation. The
human expressed repeat annotation file was retrieved from the UCSC genome
browser (RRID:SCR_005780). The genome annotation file used was provided via
CellRanger. The output loom file was used as input to estimate velocity through
scVelo. For each individual analysis, cells were filtered based on the following
parameters: minimum total counts ≥200, minimum spliced counts ≥20 and
minimum unspliced counts ≥10. For the combined cortical analysis, the processed
loom files for each individual analysis were combined to generate a new unique
molecular identifier count matrix of 15,473 genes across 53,096 cells, for which the
velocity embedding was estimated using the stochastic model. For the combined
progenitor analysis, cells that were identified as progenitors were used to create
the loom file. The loom files for each of the individuals were combined for a
total count matrix of 14,207 genes across 30,562 cells for the velocity embedding
using the same criteria. Each embedding was visualized using uniform manifold
approximation and projection (UMAP) of dimension reduction.
Immunostaining. Primary human and mouse samples and organoids were
collected, fixed in 4% paraformaldehyde, washed with 1× PBS and immersed in
30% sucrose in 1× PBS until saturated. Samples were embedded in cryomolds
using 50% optimal temperature cutting (OCT) compound (Tissue-Tek catalog no.
4583) and 50% of 30% sucrose in 1× PBS and frozen at −80 °C. All samples were
sectioned at 16 μm onto SuperFrost Plus microscope slides. Citrate antigen retrieval
(Vector Laboratories, catalog no. H-3300) was performed for 20 min at roughly
95–100 °C. Slides were then washed with 1× PBS and blocked using 5% donkey
serum, 2% gelatin, 0.1% Triton in 1× PBS for 90 min at room temperature. Primary
antibody incubation occurred in blocking buffer overnight at 4 °C, and washed five
times using 0.3% Triton + 10 mM glycine. Secondary antibody incubation occurred
in blocking buffer for 2–3 h at room temperature. Primary antibodies used were
TRKC (1:200, R and D Systems catalog no. AF373, RRID:AB_355332); ALX1
(1:500, Santa Cruz Biotechnology catalog no. sc-130416, RRID:AB_2226324); ID4
(1:200, Santa Cruz Biotechnology catalog no. sc-365656, RRID:AB_10859382);
N-CADHERIN (1:300, Abcam catalog no. ab18203, RRID:AB_444317);
DLK1 (1:200, Abcam catalog no. ab119930, RRID:AB_10902607); DLK1
(1:100, Abcam catalog no. ab21682, RRID:AB_731965); CROC-4 (1:100,
Aviva Systems catalog no. ARP34802_P050, RRID:AB_2827813); LUM (1:50,
Thermo Fisher Scientific catalog no. MA5-29402, RRID:AB_2785270); ZO-1
(1:100, Thermo Fisher Scientific catalog no. 61-7300, RRID:AB_2533938);
SOX2 (1:100, R and D Systems catalog no. AF2018, RRID:AB_355110); SOX2
(1:250, Santa Cruz Biotechnology catalog no. sc-365823, RRID:AB_10842165);
CTIP2 (1:500, Abcam catalog no. ab18465, RRID:AB_2064130); KI67 (1:200,
Thermo Fisher Scientific catalog no. 14-5698, RRID: AB_10854564); HOPX
(1:250, Santa Cruz Biotechnology catalog no. sc-398703, RRID:AB_2687966);
HOPX (1:200, Proteintech catalog no. 11419-1-AP, RRID:AB_10693525);
TBR2 (1:250, Abcam catalog no. ab23345, RRID:AB_778267); TBR2 (1:250,
R and D Systems catalog no. AF6166, RRID:AB_10569705); NESTIN
(1:200, Millipore catalog no. MAB5326, RRID:AB_2251134), DCX
(1:500, Aves Laboratories catalog no. DCX, RRID:AB_2313540); NEUN
(1:250, Millipore catalog no. ABN91, RRID:AB_11205760); PAX6 (1:200,
BioLegend catalog no. 901301, RRID:AB_2565003); FOXG1 (1:1000, Abcam
catalog no. ab18259, RRID:AB_732415); SATB2 (1:250, Abcam catalog no.
Ab51502, RRID:AB_882455); LHX5 (1:100, R and D Systems catalog no.
AF6290, RRID:AB_10973257); REELIN (1:100, MBL catalog no. D223-3,
RRID:AB_843523); Phospho-β-CATENIN (1:100, Cell Signaling Technology
catalog no. 9561, RRID:AB_331729); Phospho-S6 (1:100, Cell Signaling
Technology catalog no. 2211S, RRID:AB_331679) and NICD/NOTCH1 (1:100,
Millipore catalog no. 07-1232, RRID:AB_1977387). All secondary antibodies
were AlexaFluor used at a dilution of 1:1,000. Secondary antibodies used
were donkey antimouse 488 (Thermo Fisher Scientific catalog no. A32766,
RRID:AB_2762823); donkey antirabbit 488 (Thermo Fisher Scientific catalog no.
A32790, RRID:AB_2762833); donkey antichicken 488 (Jackson ImmunoResearch
Laboratories catalog no. 703-545-155, RRID:AB_2340375); donkey antichicken
594 (Jackson ImmunoResearch Laboratories catalog no. 703-585-155,
RRID:AB_2340377); donkey antimouse 546 (Thermo Fisher Scientific catalog no.
A10036, RRID:AB_2534012); donkey antimouse 594 (Thermo Fisher Scientific
catalog no. A-21203, RRID:AB_141633); donkey antimouse 647 (Thermo
Fisher Scientific catalog no. A32787, RRID:AB_2762830); donkey antimouse
680 (Thermo Fisher Scientific catalog no. A32788, RRID:AB_2762831); donkey
antirabbit 546 (Thermo Fisher Scientific catalog no. A10040, RRID:AB_2534016);
donkey antirabbit 594 (Thermo Fisher Scientific catalog no. A-21207,
RRID:AB_141637); donkey antirabbit 647 (Thermo Fisher Scientific catalog
no. A32795, RRID:AB_2762835); donkey antirat 594 (Thermo Fisher Scientific
catalog no. A-21209, RRID:AB_2535795); donkey antirat 488 (Thermo Fisher
Scientific catalog no. A-21208, RRID:AB_2535794); donkey antigoat 546 (Thermo
Fisher Scientific catalog no. A-11056, RRID:AB_2534103); donkey antigoat 594
NATURE NEUROSCIENCE | www.nature.com/natureneuroscience
Content courtesy of Springer Nature, terms of use apply. Rights reserved

ResouRce NATuRe NeuROsCieNCe
(Thermo Fisher Scientific catalog no. A-11058, RRID:AB_2534105); donkey
antigoat 647 (Thermo Fisher Scientific catalog no. A32849, RRID:AB_2762840);
donkey antisheep 546 (Thermo Fisher Scientific catalog no. A-21098,
RRID:AB_2535752); donkey antisheep 594 (Thermo Fisher Scientific catalog no.
A-11016, RRID:AB_2534083); donkey antisheep 647 (Thermo Fisher Scientific
catalog no. A-21448, RRID:AB_2535865) and donkey antiguinea pig 647 (Jackson
ImmunoResearch Laboratories catalog no. 706-605-148, RRID:AB_2340476).
In situ hybridization. Primary fixed samples were treated using the protocol
for RNAScope Multiplex Fluorescence Assay v.2 (Advanced Cell Diagnostics
catalog no. 323100) for C1orf61 amplification, targeting nucleotides 60–897 of
NM_006365.3 (Advanced Cell Diagnostics Probe Design no. NPR-0003991);
MEF2C amplification, targeting nucleotides 1058–2575 of NM_002397.4
(Advanced Cell Diagnostics catalog no. 452881) and the protocol for BaseScope v.2
Assay (Advanced Cell Diagnostics catalog no. 323900) for chromogenic LHX5-AS1
amplification, targeting nucleotides 69–293 of NR_126425.1 (Advanced Cell
Diagnostics Probe Design no. NPR-0003991).
Imaging and image processing. Images were collected on the Leica SP8
(RRID:SCR_018169) inverted confocal microscope using a ×40 oil-immersion
objective. Because of the scarcity of first trimester primary tissue, only one sample
per panel was imaged. For each imaging panel, the parameters (including the gain,
offset, pinhole and laser power) for image acquisition was left constant for all
samples. Images were later processed using FIJI Image J (RRID:SCR_003070).
Statistical tests. No statistical methods were used to predetermine sample sizes.
No randomization was used in this study. Distributions of the data were not
tested. Data collection and analysis were not performed blind to the conditions
of the experiments. The Wilcoxon rank sum test was used to calculate cluster
markers within Seurat for a variety of analyses. A one-sided t-test was used in
Supplementary Figs. 1e and 15c. A loess regression was used to estimate smoothed
gene expression in Supplementary Fig. 11 and Fig. 4.
Reporting Summary. Further information on research design is available in the
Nature Research Reporting Summary linked to this article.
Data availability
The data that support the findings of this study are available from the
corresponding author upon request. Raw single-cell sequencing data are available
from the NeMO Repository at https://assets.nemoarchive.org/dat-0rsydy7.
Processed single-cell sequencing data and full tilescan images are available
for exploration and for download at our cell browser: https://cells-test.gi.ucsc.
edu/?ds=early-brain.
Code availability
No custom code was used in this study. Open source algorithms were used as
detailed in single-cell analysis methods. However, any details on how these
algorithms were used are available from the corresponding author upon request.
Acknowledgements
We thank S. Wang, W. Walantus, M.G. Andrews, G. Wilkins, L. Subramanian, A. Pollen,
M. Speir and members of the A.R.K. laboratory for providing resources, technical help
and helpful discussions. scRNA-seq data have been deposited at the NeMO archive
under dbGAP restricted access. All primary human tissue was obtained from the Human
Developmental Biology Resource, with special thanks to S. Lisgo and M. Crosier.
Author contributions
This study was supported by National Institutes of Health award no. U01MH114825 to
A.R.K., and nos. F32NS103266, K99NS111731 and the L’Oreal For Women in Science
Award through the American Association for the Advancement of Science to A.B.
A.B. and A.R.K. designed the study and conducted the analysis. Experiments were
performed by U.C.E., A.B. and T.J.N. Data analysis was performed by U.C.E. and A.B.
Data deposition into the NeMO Repository and configuration of the cell browser were
prepared by M.H. The study was supervised by A.B. and A.R.K. This manuscript was
prepared by A.B. and U.C.E. with input from all authors.
Competing interests
A.R.K. is a cofounder and board member of Neurona Therapeutics. The remaining
authors declare no competing interests.
Additional information
Supplementary information The online version contains supplementary material
available at https://doi.org/10.1038/s41593-020-00794-1.
Correspondence and requests for materials should be addressed to A.B. or A.R.K.
Peer review information Nature Neuroscience thanks Andre Goffinet and the other,
anonymous, reviewer(s) for their contribution to the peer review of this work.
Reprints and permissions information is available at www.nature.com/reprints.
NATURE NEUROSCIENCE | www.nature.com/natureneuroscience
Content courtesy of Springer Nature, terms of use apply. Rights reserved

Content courtesy of Springer Nature, terms of use apply. Rights reserved

2
nature research | reporting summary
October 2018
Field-specific reporting
Please select the one below that is the best fit for your research. If you are not sure, read the appropriate sections before making your selection.
Life sciences
Behavioural & social sciences
Ecological, evolutionary & environmental sciences
For a reference copy of the document with all sections, see nature.com/documents/nr-reporting-summary-flat.pdf
Life sciences study design
All studies must disclose on these points even when the disclosure is negative.
Sample size
Because of the scarcity of first trimester human samples, we used one sample per age for each panel of figures. For the organoid and mouse
samples, the data imaged represents a sample of 1. The single-cell data for the mouse samples represents a sample size of 3 individuals.
Sample sizes were chosen based upon the ability to get representative data with the number of single-cells for the comparison. As such, we
had 289,000 cells from the human and 16053 cells from mouse which is sufficient to identify cell types based upon numerous studies in the
field.
Data exclusions
In all the single-cell analyses, we excluded all cells that had fewer than 500 genes per cell and had greater than 10% mitochondrial content.
However, these data are available from the raw data. These exclusions were pre-established and were necessary to eliminate droplets that
might not contain actual cells or that contain dead cells from downstream analysis.
Replication
All imaged primary human data had one replicate due to the scarcity of the tissue. For the single-cell data of the human tissue, there were
two Carnegie Stage 14 , Carnegie Stage 15 and Carnegie Stage 22 replicates each. There was one replicate for the remaining ages due to
limited tissue availability. Each imaged mouse and organoid sample represents one replicate. All attempts for single-cell collection of data
were successful.
Randomization
In each immunostaining panel, the same tissue sample from one individual was sectioned and slices were evenly distributed across all imaging
panel conditions. No randomization was performed across other analysis, and this was not relevant because the data was processed with pre-
determined conditions and compared between methods (different individuals, immunostaining, etc).
Blinding
Blinding was not performed, and it was not meaningful because only one sample was collected at one point. During analysis, blinding was not
meaningful because all samples were treated equally by pre-determined analyses and thresholds. However, we used clearly defined imaging
and image processing criteria (described in our methods) for all our analyses and analyzed all included samples using the same rigorous
criteria in order to avoid bias.
Reporting for specific materials, systems and methods
We require information from authors about some types of materials, experimental systems and methods used in many studies. Here, indicate whether each material,
system or method listed is relevant to your study. If you are not sure if a list item applies to your research, read the appropriate section before selecting a response.
Materials & experimental systems
n/a
Involved in the study
Antibodies
Eukaryotic cell lines
Palaeontology
Animals and other organisms
Human research participants
Clinical data
Methods
n/a
Involved in the study
ChIP-seq
Flow cytometry
MRI-based neuroimaging
Antibodies
Antibodies used
All secondary antibodies were AlexaFluor used at a dilution 1:1000. Secondary antibodies: Donkey anti-Mouse 488 (Thermo
Fisher Scientific Cat# A32766, RRID:AB_2762823); Donkey anti-Rabbit 488 (Thermo Fisher Scientific Cat# A32790,
RRID:AB_2762833); Donkey anti-Chicken 488 (Jackson ImmunoResearch Labs Cat# 703-545-155, RRID:AB_2340375); Donkey
anti-chicken 594 (Jackson ImmunoResearch Labs Cat# 703-585-155, RRID:AB_2340377); Donkey anti-Mouse 546 (Thermo Fisher
Scientific Cat# A10036, RRID:AB_2534012); Donkey anti-Mouse 594 (Thermo Fisher Scientific Cat# A-21203, RRID:AB_141633);
Donkey anti-Mouse 647 (Thermo Fisher Scientific Cat# A32787, RRID:AB_2762830); Donkey anti-Mouse 680 (Thermo Fisher
Scientific Cat# A32788, RRID:AB_2762831); Donkey anti-Rabbit 546 (Thermo Fisher Scientific Cat# A10040, RRID:AB_2534016);
Donkey anti-Rabbit 594 (Thermo Fisher Scientific Cat# A-21207, RRID:AB_141637); Donkey anti-Rabbit 647 (Thermo Fisher
Scientific Cat# A32795, RRID:AB_2762835); Donkey anti-Rat 594 (Thermo Fisher Scientific Cat# A-21209, RRID:AB_2535795);
Donkey anti-Rat 488 (Thermo Fisher Scientific Cat# A-21208, RRID:AB_2535794); Donkey anti-Goat 546 (Thermo Fisher Scientific
Cat# A-11056, RRID:AB_2534103); Donkey anti-Goat 594 (Thermo Fisher Scientific Cat# A-11058, RRID:AB_2534105); Donkey
anti-Goat 647 (Thermo Fisher Scientific Cat# A32849, RRID:AB_2762840); Donkey anti-Sheep 546 (Thermo Fisher Scientific Cat#
Content courtesy of Springer Nature, terms of use apply. Rights reserved

3
nature research | reporting summary
October 2018
A-21098, RRID:AB_2535752); Donkey anti-Sheep 594 (Thermo Fisher Scientific Cat# A-11016, RRID:AB_2534083); Donkey anti-
Sheep 647 (Thermo Fisher Scientific Cat# A-21448, RRID:AB_2535865); Donkey anti-Guinea Pig 647 (Jackson ImmunoResearch
Labs Cat# 706-605-148, RRID:AB_2340476).
TRKC (1:200, R and D Systems Cat# AF373, RRID:AB_355332); ALX1 (1:500, Santa Cruz Biotechnology Cat# sc-130416,
RRID:AB_2226324); ID4 (1:200, Santa Cruz Biotechnology Cat# sc-365656, RRID:AB_10859382); N-CADHERIN (1:300, Abcam Cat#
ab18203, RRID:AB_444317); DLK1 (1:200, Abcam Cat# ab119930, RRID:AB_10902607); DLK1 (1:100, Abcam Cat# ab21682,
RRID:AB_731965); CROC-4 (1:100, Aviva Systems Cat# ARP34802_P050, RRID:AB_2827813); LUM (1:50, Thermo Fisher Scientific
Cat# MA5-29402, RRID:AB_2785270); ZO-1 (1:100, Thermo Fisher Scientific Cat# 61-7300, RRID:AB_2533938); SOX2 (1:100, R
and D Systems Cat# AF2018, RRID:AB_355110); SOX2 (1:250, Santa Cruz Biotechnology Cat# sc-365823, RRID:AB_10842165);
CTIP2 (1:500, Abcam Cat# ab18465, RRID:AB_2064130); KI67 (1:200, Thermo Fisher Scientific Cat# 14-5698, RRID:
AB_10854564); HOPX (1:250, Santa Cruz Biotechnology Cat# sc-398703, RRID:AB_2687966); HOPX (1:200, Proteintech Cat#
11419-1-AP, RRID:AB_10693525); TBR2 (1:250, Abcam Cat# ab23345, RRID:AB_778267); TBR2 (1:250, R and D Systems Cat#
AF6166, RRID:AB_10569705); NESTIN (1:200, Millipore Cat# MAB5326, RRID:AB_2251134), DCX (1:500, Aves Labs Cat# DCX,
RRID:AB_2313540); NEUN (1:250, Millipore Cat# ABN91, RRID:AB_11205760); PAX6 (1:200, BioLegend Cat# 901301,
RRID:AB_2565003); FOXG1 (1:1000, Abcam Cat# ab18259, RRID:AB_732415); SATB2 (1:250, Abcam Cat# Ab51502,
RRID:AB_882455); LHX5 (1:100, R and D Systems Cat# AF6290, RRID:AB_10973257); REELIN (1:100, MBL Cat# D223-3,
RRID:AB_843523); Phospho-B-CATENIN (1:100, Cell Signaling Technology Cat# 9561, RRID:AB_331729); Phospho-S6 (1:100, Cell
Signaling Technology Cat# 2211S, RRID:AB_331679); NICD/NOTCH1 (1:100, Millipore Cat# 07-1232, RRID:AB_1977387).
Validation
Validation:
TRKC R&D Systems, AF373: Western Blot validation at 164 kDa using the 12-230 kDA separation system under reducing
conditions https://www.rndsystems.com/products/human-trkc-antibody_af373
ALX1 Santa Cruz, sc-130416:raised agaisnt recombinant ALX1 of human origin. Western Blot analysis shows validation at the
~44kDa. https://www.scbt.com/p/alx1-antibody-96k
ID4 Santa Cruz, sc-365656: raised against amino acids 1-70 mapping the N-terminus of Id4 of human origin. Western blot
validaiton at ~20kDa. https://www.scbt.com/p/id4-antibody-b-5.
N-CADHERIN Abcam, ab18203: Synthetic peptide corresponding to Human N Cadherin aa 800-900 (internal sequence)
conjugated to keyhole limpet haemocyanin. Western blot validation at 125 kDA. https://www.abcam.com/n-cadherin-antibody-
ab18203.html
DLK1 Abcam, ab119930: Recombinant fragment corresponding to Human DLK-1 aa 174-349. Western blot validation at 41 kDa.
https://www.abcam.com/dlk-1-antibody-3a10-ab119930.html
DLK1 Abcam, ab21682: Synthetic peptide corresponding to Human DLK-1 aa 350 to the C-terminus (C terminal) conjugated to
keyhole limpet haemocyanin. Western Blot validation at 45 and 48 kDa. https://www.abcam.com/dlk-1-antibody-ab21682.html
CROC-4 Aviva Systems, ARP34802_P050: a synthetic peptide directed towards the N-terminal region of Human CROC4. Western
Blot validation at ~35 and 22 kDA. https://www.avivasysbio.com/c1orf61-antibody-n-terminal-region-arp34802-p050.html
LUM Thermo Fisher, MA5-29402: recombinant protein targeting the recombinant human lumican protein. https://
www.thermofisher.com/antibody/product/LUM-Antibody-clone-77-Recombinant-Monoclonal/MA5-29402
ZO-1 Thermo Fisher, 61-7300: A 69 kD fusion protein(1) corresponding to amino acids 463-1109 of human ZO-1 cDNA.(2) This
sequence lies N-terminal to the 80 amino acid region (the alpha-motif) present in the a+-isoform but absent in the a- isoform
due to alternative splicing. https://www.thermofisher.com/antibody/product/ZO-1-Antibody-Polyclonal/61-7300
SOX2 R&D Systems, AF2018: Polyclonal antibody raised against recombinant human SOX2. Western Blot validation at 36 kDa.
https://www.rndsystems.com/products/human-mouse-rat-sox2-antibody_af2018
SOX2 Santa Cruz, sc-365823: specific for an epitope mapping between amino acids 170-201 within an internal region of Sox-2 of
human origin. Western Blot validation at ~36 kDa. https://www.scbt.com/p/sox-2-antibody-e-4
CTIP2 Abcam, ab18465: Detects 2 bands representing Ctip2 at about 120kD (between aa 1-150 kDa). Western blot validation at
~128 and 129 kDa. https://www.abcam.com/ctip2-antibody-25b6-chip-grade-ab18465.html
KI67 Thermo Fisher, 14-5698: monoclonal antibody recognizes mouse and rat Ki-67, a 300 kDa nuclear protein. This Antibody
was verified by Cell treatment to ensure that the antibody binds to the antigen stated. https://www.thermofisher.com/antibody/
product/Ki-67-Antibody-clone-SolA15-Monoclonal/14-5698-82
HOPX Santa Cruz, sc-398703: raised against amino acids 1-73 representing full length Hop of human origin. Western Blot
validation at ~15 kDa. ttps://www.scbt.com/p/hop-antibody-e-1
HOPX Proteintech, 11419-1-AP. HopX fusion protein. Western Blot validation at ~ 10kDa. https://www.ptglab.com/Products/
HOPX-Antibody-11419-1-AP.htm
TBR2 Abcam, ab23345: Synthetic peptide corresponding to Mouse TBR2/ Eomes aa 650 to the C-terminus (C terminal)
conjugated to keyhole limpet haemocyanin. Western blot validation at 85 kDa. https://www.abcam.com/tbr2-eomes-antibody-
chip-grade-ab23345.html
TBR2 R&D Systems, AF6166: E. coli-derived recombinant human EOMES. Western blot validation at ~95 and 96 kDa. https://
www.rndsystems.com/products/human-eomes-antibody_af6166
Content courtesy of Springer Nature, terms of use apply. Rights reserved

4
nature research | reporting summary
October 2018
NESTIN Millipore Sigma, MAB5326: clone 10C2 for detection Nestin fusion protein. Western blot validation at 220 -240 kDa.
https://www.sigmaaldrich.com/catalog/product/mm/mab5326?lang=en&region=US
DCX Aves Labs, DCX: Two antipeptide antibodies were generated in chickens against sequences shared between the mouse
(AAT58219.1), rat (NP_445831.3) and human (CAA06617.1) gene products. https://www.aveslabs.com/products/doublecortin
NEUN, Millipore Sigma, ABN91: GST-tagged recombinant protein corresponding to the N-terminus of mouse NeuN. Western Blot
validationa t ~45kDa. https://www.sigmaaldrich.com/catalog/product/mm/abn91?lang=en&region=US
PAX6 Biolegend, 901301: antibody was generated against the peptide (QVPGSEPDMSQYWPRLQ) derived from the C-terminus of
the mouse Pax-6 protein. Western Blot validation at 46.6 and 48.2 kDa.https://www.biolegend.com/en-us/products/purified-
anti-pax-6-antibody-11511
FOXG1 Abcam, ab18259: Synthetic peptide corresponding to Human FOXG1 aa 400 to the C-terminus (C terminal) conjugated to
keyhole limpet haemocyanin. Western Blot validation at 50 kDa. https://www.abcam.com/foxg1-antibody-chip-grade-
ab18259.html
SATB2 Abcam, ab51502: Recombinant fragment corresponding to the C-terminal of Human SATB2. Western blot validation at
82kDa. https://www.abcam.com/satb2-antibody-satba4b10-c-terminal-ab51502.html
LHX5 R and D Systems, AF6290: Detects recombinant human LHX5 at 55-60 kDa in Western Blot. https://www.rndsystems.com/
products/human-mouse-rat-lhx5-antibody_af6290
REELIN MBL, D223-3: reacts with mouse Reelin. The CR-50 epitope is located between mouse Reelin amino acid 230 to 346 4.
https://www.mblintl.com/products/d223-3/
Phospho-B-Catenin Cell Signaling, 9561: detects endogenous levels of β-catenin only when phosphorylated at serines 33, 37 or
threonine 41. Polyclonal antibodies are produced by immunizing animals with a synthetic phosphopeptide corresponding to
residues surrounding Ser33, Ser37 and Thr41 of human B-catenin. https://www.cellsignal.com/products/primary-antibodies/
phospho-b-catenin-ser33-37-thr41-antibody/
Phospho-S6 Cell Signaling, 2211S: detects endogenous levels of ribosomal protein S6 only when phosphorylated at serine 235
and 236. Polyclonal antibodies are produced by immunizing animals with a synthetic phosphopeptide corresponding to residues
surrounding Ser235 and Ser236 of human ribosomal protein S6. https://www.cellsignal.com/products/primary-antibodies/
phospho-s6-ribosomal-protein-ser235-236-antibody/
NICD Millipore, 07-1232: Notch 1, cleaved N terminal. Only the cleaved intracellular (activated) form is detected. Synthetic
peptide from the N-terminal sequence of the cleaved N intracellular domain (NICD) human Notch 1. Western blot detection at
80 kDa. https://www.sigmaaldrich.com/catalog/product/mm/071232?lang=en&region=US
Eukaryotic cell lines
Policy information about cell lines
Cell line source(s)
H1 (WA01) embryonic stem cell line (source: WiCell)
1323-4 induced pluripotent stem cell line (source: Bruce Conklin, Gladstone Institute)
H28126 induced pluripotent stem cell line (source: Yoav Gilad, University of Chicago)
4955 induced pluripotent stem cell line (source: Yoav Gilad, University of Chicago)
Authentication
Each stem cell line was karyotyped and validated for pluripotency, prior to receipt. Every 10 passages, stem cells are tested
for karyotypic abnormalities and validated for pluripotency markers Sox2, Nanog, and Oct4.
Mycoplasma contamination
All cell lines tested negative for mycoplasma.
Commonly misidentified lines
(See ICLAC register)
No commonly misidentified lines were used.
Animals and other organisms
Policy information about studies involving animals; ARRIVE guidelines recommended for reporting animal research
Laboratory animals
CD-1® IGS Mouse, sacrificed at E9 and E10 with equal numbers of male and female embryonic mice used. Housing conditions are
described in the Methods.
Wild animals
No wild animals were used in this study.
Field-collected samples
No field-collected samples were used in this study.
Ethics oversight
All mouse experiments were approved by and conducted according to the UCSF Institutional Animal Care and Use Committee
(protocol AN078703-03A).
Note that full information on the approval of the study protocol must also be provided in the manuscript.
Content courtesy of Springer Nature, terms of use apply. Rights reserved

5
nature research | reporting summary
October 2018
Human research participants
Policy information about studies involving human research participants
Population characteristics
Because of the sensitivity of the samples, no population characteristics are known or recorded.
Recruitment
No recruitment criteria other than consent were required.
Ethics oversight
Acquisition of all primary human tissue samples was approved by the UCSF Human Gamete, Embryo and Stem Cell Research
Committee (GESCR, approval 10-03379 and 10-05113). All experiments were performed in accordance with protocol guidelines.
Informed consent was obtained before sample collection and use for this study.
Note that full information on the approval of the study protocol must also be provided in the manuscript.
Content courtesy of Springer Nature, terms of use apply. Rights reserved

1.
2.
3.
4.
5.
6.
Terms and Conditions
Springer Nature journal content, brought to you courtesy of Springer Nature Customer Service Center GmbH (“Springer Nature”).
Springer Nature supports a reasonable amount of sharing of research papers by authors, subscribers and authorised users (“Users”), for small-scale
personal, non-commercial use provided that all copyright, trade and service marks and other proprietary notices are maintained. By accessing,
sharing, receiving or otherwise using the Springer Nature journal content you agree to these terms of use (“Terms”). For these purposes, Springer
Nature considers academic use (by researchers and students) to be non-commercial.
These Terms are supplementary and will apply in addition to any applicable website terms and conditions, a relevant site licence or a personal
subscription. These Terms will prevail over any conflict or ambiguity with regards to the relevant terms, a site licence or a personal subscription (to
the extent of the conflict or ambiguity only). For Creative Commons-licensed articles, the terms of the Creative Commons license used will apply.
We collect and use personal data to provide access to the Springer Nature journal content. We may also use these personal data internally within
ResearchGate and Springer Nature and as agreed share it, in an anonymised way, for purposes of tracking, analysis and reporting. We will not
otherwise disclose your personal data outside the ResearchGate or the Springer Nature group of companies unless we have your permission as
detailed in the Privacy Policy.
While Users may use the Springer Nature journal content for small scale, personal non-commercial use, it is important to note that Users may not:
use such content for the purpose of providing other users with access on a regular or large scale basis or as a means to circumvent access control;
use such content where to do so would be considered a criminal or statutory offence in any jurisdiction, or gives rise to civil liability, or is
otherwise unlawful;
falsely or misleadingly imply or suggest endorsement, approval , sponsorship, or association unless explicitly agreed to by Springer Nature in
writing;
use bots or other automated methods to access the content or redirect messages
override any security feature or exclusionary protocol; or
share the content in order to create substitute for Springer Nature products or services or a systematic database of Springer Nature journal content.
In line with the restriction against commercial use, Springer Nature does not permit the creation of a product or service that creates revenue, royalties,
rent or income from our content or its inclusion as part of a paid for service or for other commercial gain. Springer Nature journal content cannot be
used for inter-library loans and librarians may not upload Springer Nature journal content on a large scale into their, or any other, institutional
repository.
These terms of use are reviewed regularly and may be amended at any time. Springer Nature is not obligated to publish any information or content on
this website and may remove it or features or functionality at our sole discretion, at any time with or without notice. Springer Nature may revoke this
licence to you at any time and remove access to any copies of the Springer Nature journal content which have been saved.
To the fullest extent permitted by law, Springer Nature makes no warranties, representations or guarantees to Users, either express or implied with
respect to the Springer nature journal content and all parties disclaim and waive any implied warranties or warranties imposed by law, including
merchantability or fitness for any particular purpose.
Please note that these rights do not automatically extend to content, data or other material published by Springer Nature that may be licensed from
third parties.
If you would like to use or distribute our Springer Nature journal content to a wider audience or on a regular basis or in any other manner not
expressly permitted by these Terms, please contact Springer Nature at
onlineservice@springernature.com