Available via license: CC BY-NC-ND 4.0
Content may be subject to copyright.
Stem Cell Reports
Article
Multi-layered regulation of neuroectoderm differentiation by retinoic acid in a
primitive streak-like context
Luigi Russo,
1,2
Hanna L. Sladitschek,
1,3
and Pierre A. Neveu
1,
*
1
Cell Biology and Biophysics Unit, European Molecular Biology Laboratory, 69117 Heidelberg, Germany
2
Joint PhD Degree from EMBL and Heidelberg University, Faculty of Biosciences, 69120 Heidelberg, Germany
3
Present address: Department of Molecular Medicine, University of Padua School of Medicine, 35126 Padua, Italy
*Correspondence: neveu@embl.de
https://doi.org/10.1016/j.stemcr.2021.12.014
SUMMARY
The formation of the primitive streak (PS) and the subsequent induction of neuroectoderm are hallmarks of gastrulation. Combining an
in vitro reconstitution of this process based on mouse embryonic stem cells (mESCs) with a collection of knockouts in reporter mESClines,
we identified retinoic acid (RA) as a critical mediator of early neural induction triggered by TGFbor Wnt signaling inhibition. Single-cell
RNA sequencing analysis captured the temporal unfolding of cell type diversification, up to the emergence of somite and neural fates. In
the absence of the RA-synthesizing enzyme Aldh1a2, a sensitive RA reporter revealed a hitherto unidentified residual RA signaling that
specified neural fate. Genetic evidence showed that the RA-degrading enzyme Cyp26a1 protected PS-like cells from neural induction,
even in the absence of TGFband Wnt antagonists. Overall, we characterized a multi-layered control of RA levels that regulates early neural
differentiation in an in vitro PS-like system.
INTRODUCTION
During gastrulation, cells of the epiblast are allocated to the
three germ layers (Tam and Behringer, 1997). Gastrulation
is initiated by the formation of the primitive streak (PS) and
subsequent induction of neuroectoderm. Seminal experi-
ments by Spemann and Mangold showed that the trans-
plantation of the dorsal blastopore in amphibians could
induce a secondary axis and neural tissue in the host
embryo (Spemann and Mangold, 1924). This region or
‘‘organizer’’ secretes a range of molecules mediating this in-
duction (De Robertis, 2006). Among them, antagonists of
the transforming growth factor b(TGFb) signaling
pathway, and in particular of bone morphogenetic
proteins, are considered pivotal for neuralization of the
ectoderm (Weinstein and Hemmati-Brivanlou, 1999). The
inhibition of the Wnt signaling pathway is another potent
inductive cue (Glinka et al., 1998). While most of the mo-
lecular mechanisms governing this process were deter-
mined in amphibians, they appear to be conserved in
mammals (Levine and Brivanlou, 2007). Indeed, the distal
tip of the mouse PS, the node, possesses organizer-like
properties (Tam and Behringer, 1997). The deletion of the
two TGFbinhibitors Chordin and Noggin (Bachiller et al.,
2000) or the knockout of the Wnt inhibitor Dkk1 (Mukho-
padhyay et al., 2001) lead to the absence of anterior neural
structures (forebrain) in mouse. Retinoic acid (RA) is
another signaling molecule with potent neuralizing activ-
ity (Rhinn and Dolle
´, 2012) that was found to be produced
by the Hensen’s node, the chick equivalent of the organizer
(Hogan et al., 1992). RA signaling was detected as well in
the mouse PS at E7.5 (Rossant et al., 1991). At this develop-
mental stage, ALDH1A2 is considered to be the only
enzyme synthesizing RA from retinal (Rhinn and Dolle
´,
2012). Both the presence of forebrain structures and an
absence of expression of an RA activity reporter in
Aldh1a2
/
embryos ruled out an involvement of RA in
early neural induction (Niederreither et al., 1999). This
contrasts with the widespread use of RA to induce neuronal
fates from pluripotent cells in vitro (Ying et al., 2003) and
the well-established role of RA in the formation of the pos-
terior neural axis. Here, the allocation of cell types to so-
mite and spinal cord fates from bipotent neuromesodermal
progenitors (NMPs) allows the extension of the body axis
(Henrique et al., 2015). It was demonstrated that RA has a
critical function in NMP differentiation to the neural line-
age (Diez del Corral et al., 2003). The RAR family of nuclear
receptors, which acts as transcription factors regulated by
RA, is the effector of the developmental functions of RA (Sa-
marut and Rochette-Egly, 2012). While in vivo work estab-
lished the importance of antagonizing TGFband Wnt
signaling pathways for neural induction and ruled out a
contribution of RA signaling in this process, the molecular
implementation of the neuroectoderm differentiation de-
cision is largely unexplored. In vitro systems based on
pluripotent stem cells enable to recapitulate crucial aspects
of early post-implantation mammalian development
(Shahbazi et al., 2019).
We adopted an mESC-based system in which we can
monitor the formation of neuroectoderm in the presence
of a PS-like population and profiled by single-cell RNA
sequencing the progression of the differentiating culture.
In this context, we determined that RA mediates early neu-
ral induction downstream of TGFband Wnt inhibition.
Stem Cell Reports jVol. 17 j231–244 jFebruary 8, 2022 jª2021 The Author(s). 231
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
The formation of neural progenitors, even in the absence of
the antagonists CHORDIN and NOGGIN or DKK1, was
enhanced by deleting the RA-degrading enzyme
CYP26A1. The development of a highly sensitive RA re-
porter enabled us to detect RA signaling in conditions
thought to lack RA synthesis ability. Finally, the knockout
of RA receptors highlighted their function as regulators of
loci critical for neural induction. Altogether, our results
add valuable insights into the multi-layered regulation of
RA signaling in the process of early neuroectoderm
formation.
RESULTS
Characterization of a system to investigate the
mechanisms of neuroectoderm formation
The generation of primitive streak-like cells in vitro
should enable the induction of a neuroectodermal fate
among differentiation-competent cells (Figure 1A). To
monitor the formation of PS-like cells and subsequent in-
duction of neural progenitors (NPs), we used a double
knock-in (2KI) reporter mESC line (Sladitschek and Ne-
veu, 2019)withSox1 locus targeted with GFP and T
(also known as Brachyury) locus targeted with H2B-
3xTagBFP. While Tis expressed in the PS (Wilkinson
et al., 1990), Sox1 marks exclusively NPs (Pevny et al.,
1998). We previously showed that the small molecule
IDE1, which phenocopies TGFbpathway agonists (Boro-
wiak et al., 2009), formed differentiation intermediates
resembling mouse post-implantation epiblast and PS
(Sladitschek and Neveu, 2019). Interestingly, putative
NP Sox1
GFP+
cells coexisted with T
TagB FP +
cells, the candi-
date PS-like cells (Figure 1B).
We monitored the composition of the differentiating cul-
ture by flow cytometry (Figure S1A). The increase in TagBFP
signal and Sox1
GFP+
cells detected by the third day matched
the increase in T and Sox1 mRNA levels (Figures S1A–S1D).
Increasing the number of cells seeded at the beginning of
the differentiation enhanced the fraction of Sox1
GFP+
cells
(Figures S1E and S1F), indicating that cell density played a
role in the formation of putative NPs.
To determine the identity of the different cell popula-
tions, we characterized the transcriptome of FACS-purified
cells expressing TagBFP or GFP (Figure 1C). T
TagBFP+
cells
expressed markers associated with post-implantation
epiblast and PS fates such as Fgf5,T,Mixl1, and Goosecoid
(Figures 1D and S1G). More importantly, the expression of
secreted antagonists associated with the in vivo organizer
was selectively higher in the T
TagBFP+
population. Among
these were the TGFbantagonists Chordin (Chrd) and
Noggin (Nog) and the Wnt antagonist Dkk1. The neuroec-
todermal identity of Sox1
GFP+
cells was confirmed by the
upregulation of NP markers, including Sox2 and Pax6 (Fig-
ures 1D and S1G).
scRNA-seq characterization of PS-like differentiation
To better characterize the cellular heterogeneity in our PS-
like differentiation, we conducted single-cell RNA
sequencing (scRNA-seq) between day 2 and day 5 and ob-
tained expression profiles for 46,700 cells (Figure 2A). Uni-
form manifold approximation and projection (UMAP)
analysis showed minimal overlap between consecutive
days (Figures 2B and S2A). Profiles of FACS-purified
T
TagBFP+
Sox1
GFP
and Sox1
GFP+
cells projected according
to the respective expression territories of Tand Sox1 (Fig-
ures 2C and S2B). Notably, cells with high Texpression
and Sox1-expressing cells formed distinct populations (Fig-
ure 2C). We could identify 35 cell subpopulations (Figures
2D and S2C) stratified by a number of markers (Figure S2D).
We observed a prevalence of epiblast and PS fates till day 3,
101
Sox1-GFP signal (a.u.)
T-TagBFP signal (a.u.)
102103
101
102
103
C
mRNA-Seq
Sox1-GFP+
mRNA-Seq
T-TagBFP+/Sox1-GFP-
T-H2B-3xTagBFP
Sox1-GFP
B
A
mESCs
neural progenitor
Primitive
Streak-like
T-TagBFP+
Sox1-GFP+
IDE1
T-H2B-3xTagBFP
Sox1-GFP
undifferentiated
cell
Eomes
Fgf5
Gsc
Lhx1
Mesp1
Mixl1
T
Lefty2
Chrd
Nog
Dkk1
Cer1
En1
En2
Hes5
Irx2
Lhx5
Olig3
Pax6
Pou3f2
Sox1
Sox2
Zic1
Nes
T-TagBFP+
Sox1-GFP- Sox1-GFP+
D04-4
log2(expression)
d4
d5
d6
d4
d5
d6
Figure 1. Characterization of neural induction by PS-like cells
(A) Experimental strategy to induce and monitor the formation of neuroectoderm by PS-like cells using a double knock-in mESC line
reporting on Tand Sox1 expression.
(B) T-TagBFP and Sox1-GFP reporter expression after 5 days of PS-like differentiation. Bar: 100 mm.
(C) Experimental strategy to characterize the populations present in the differentiating culture.
(D) Expression levels (as measured by mRNA-Seq) of PS and NP markers in FACS-purified populations after PS-like differentiation. See also
Figure S1.
232 Stem Cell Reports jVol. 17 j231–244 jFebruary 8, 2022
followed by the formation of NPs and PS derivatives later
on (Figure 2E).
A subpopulation of cells at day 2 displayed naive plu-
ripotency markers, while the rest initiated the expression
of primed epiblast markers (Figure 2F). Pluripotency fac-
tors had distinct behaviors: while Pou5f1 (also known
as Oct4) expression was retained till day 4 in the epiblast
and PS lineages, Nanog expression was transiently reacti-
vated in PS-like cells (Figure S2E). Sox2 was downregu-
lated in PS-like cells and their derivatives, whereas its
expression was maintained in NPs (Figure S2E). These
findings recapitulated the expression patterns of these
genes in E7.0 mouse embryos (Peng et al., 2019). PS
markers were expressed in different subpopulations (Fig-
ures 2GandS2D) corresponding to different regions of
the in vivo PS. Noteworthy, T expression encompassed
both bona fide PS and post-implantation epiblast cells,
with higher transcript levels in the former (Figures 2C
and S2D). Thus, the PS-like population marked by the
expression of the T
TagB FP
reporter at day 3 comprised a
mixture of these two fates.
PS derivatives were formed as differentiation pro-
ceeded. Indeed, presomitic mesoderm and distinct so-
mite fates were found at days 4 and 5, along with a
population resembling neuromesodermal progenitors
(NMPs) (Figures 2E, 2H, and S2D). Moreover, endoderm
and notochord fates were present by day 5 (Figure 2I).
Distinct neuroectodermal cell types gradually accumu-
lated, at the expense of epiblast and PS fates (Figures
2E, S2C, and S2F). Thus, despite the absence of defined
geometrical constraints, our in vitro system recapitulated
the fate diversification occurring in post-implantation
embryos and notably the temporal evolution of the PS
in vivo.
day 2
day 3
day 4
day 5
B
UMAP1
UMAP2
T
Sox1
C
mESCs
PGCLCs
Epiblast-like
Neural progenitors
Epithelial
cells
Primitive
streak-like
NMPs
PSM
Somite
Endoderm
Notochord
Neural tube
Neurons
D
Myotome
d0 d1 d2 d3 d4 d5
PS-like differentiation
A
scRNA-Seq
mESCs
Meox1
Tbx6
HFoxa2 Sox17 Shh
I
Mixl1
GGsc Lhx1
Zfp42 Fgf5
F
E
mESCs
PGLCs
Epithelial cells
Epiblast-like
Primitive
Streak-like
Endoderm
Notochord
NMPs
PSM
Somite
Myotome
Neural tube
Neural
Progenitors
Neurons mESCs
Epiblast-like Somite
day 2 day 3 day 4 day 5
Neural
Progenitors
Primitive
Streak-
like
NMPs
PSM
100050
10(expression levels)log
% of max
100050
10(expression levels) % of maxlog 100050
10(expression levels) % of maxlog
100050
10(expression levels) % of maxlog 100050
10(expression levels) % of maxlog
Figure 2. scRNA-seq characterization of neural induction by PS-like cells
(A) Experimental strategy to temporally monitor PS-like differentiation using scRNA-seq.
(B–D) UMAP (uniform manifold approximation and projection) of 46,700 cells colored by the collection day (B), by the scaled expression of
Tand Sox1 (C), or according to the identified populations (D) (NMPs: neuromesodermal progenitors, PSM: presomitic mesoderm, PGCLCs:
primordial germ cell-like cells).
(E) Alluvial plot showing the temporal evolution of the culture composition.
(F–I) UMAP colored by the scaled expression of the naive pluripotency marker Zfp42 and the post-implantation epiblast marker Fgf5 (F), of
PS (G), presomitic mesoderm and somite (H), or endoderm and notochord (I) markers. See also Figure S2.
Stem Cell Reports jVol. 17 j231–244 jFebruary 8, 2022 233
In vitro reconstitution of neural induction by Wnt and
TGFbantagonists
TGFbor Wnt signaling inhibition are critical for neuroecto-
derm induction from pluripotent epiblast in vivo (De Rob-
ertis, 2006). We sought to recapitulate this process in our
in vitro system by applying inhibitors once the T
TagBFP+
pop-
ulation was established at day 3 (Figure 3A). Adding a small
molecule antagonist of the TGFbpathway SB431542 or
blocking Wnt signaling using the tankyrase inhibitor
XAV939 led to an increase in Sox1
GFP+
cells (Figures 3B
and 3C). More Sox1
GFP+
cells could be detected as well
upon addition of recombinant NOGGIN and CHORDIN
or DKK1 (Figures S3A and S3B). Thus, the exogenous appli-
cation of inhibitors of either signaling pathway was able to
induce neural fate in our in vitro system. Bulk and scRNA-
seq data showed that Nog,Chrd, and Dkk1 were expressed
77.4% 5.46%
5.39%11.8%
64.4% 7.13%
11.9%16.5%
72.3% 4.64%
8.92%14.2%
6.32%
19.4%11.9%
36.1% 7.63%
23.2%33.1%
14.5% 2.13%
27.7%55.7%
56.2% 5.21%
9.21%29.4%
0.032% 0.077%
0.33%99.6%
Chrd-/-Nog-/-
mESCs
Dkk1-/-
mESCs
Wnt
TGFß
antagonist KO
PS-like diff.
101
Sox1-GFP signal (a.u.)
T-TagBFP signal (a.u.)
102103
101
102
103
EWT
101
Sox1-GFP signal (a.u.)
102103
Chrd-/-Nog-/-
101
Sox1-GFP signal (a.u.)
T-TagBFP signal (a.u.)
102103
101
102
103
BControl
101
Sox1-GFP signal (a.u.)
102103
+SB43
101
Sox1-GFP signal (a.u.)
102103
d0 d1 d2 d3 d4 d5
IDE1 (Control)
IDE1+SB43
(Tgfß inhib.)
IDE1+XAV
(Wnt inhib.)
IDE1
IDE1
APS-like state
T-H2B-3xTagBFP
Sox1-GFP
mESCs
101
Sox1-GFP signal (a.u.)
102103
+XAVNon-transgenic mESCs
101
Sox1-GFP signal (a.u.)
T-TagBFP signal (a.u.)
102103
101
102
103WT
101
Sox1-GFP signal (a.u.)
102103
Dkk1-/-
62.5%
D
***
***
C
Fraction of cells ± s.d.(%)
0
10
20
30
40
50
60
+SB43
+XAV
Control
BFP+ GFP+ GFP+
BFP+
**
*
F
Fraction of cells ± s.d.(%)
0
10
20
30
40
50
60
Chrd-/-Nog-/-
WT
BFP+ GFP+ GFP+
BFP+
**
**
70
80
*
** GH
Fraction of cells ± s.d.(%)
0
10
20
30
40
50
60
Dkk1-/-
WT
BFP+ GFP+ GFP+
BFP+
70
80
***
***
Figure 3. In vitro reconstitution of neural induction by Wnt and TGFbantagonists
(A) Experimental strategy to assess the impact of TGFbor Wnt pathway inhibition on fate induction. SB431542 (SB43) inhibits TGFb
receptors and XAV939 (XAV) is a tankyrase inhibitor.
(B) T-TagBFP and Sox1-GFP reporter expression as measured at day 5 by flow cytometry in non-transgenic mESCs and 2KI mESCs (Control),
or after inhibition of the TGFb(+SB43) or Wnt (+XAV) pathways. Dotted lines: gates fixed according to the non-transgenic mESCs negative
control.
(C) Quantification of (B) data (n = 3 independent experiments; *, p < .05; **, p < .01; ***, p < .001; two-sided unpaired t test; data
represented as mean ±SD).
(D) Experimental strategy to assess the impact of knockouts of antagonists of the TGFband Wnt signaling pathways.
(E) T-TagBFP and Sox1-GFP reporter expression as measured by flow cytometry after 5 days of PS-like differentiation of wild-type 2KI (WT)
or Chrd
/
Nog
/
mESCs. Dotted lines: gates fixed according to the non-transgenic mESCs negative control.
(F) Quantification of the data in (E) (n = 3 independent experiments; *, p < .05; **, p < .01; two-sided unpaired t test; data represented as
mean ±SD).
(G) T-TagBFP and Sox1-GFP reporter expression as measured by flow cytometry after 5 days of PS-like differentiation of wild-type 2KI (WT)
or Dkk1
/
mESCs. Dotted lines: gates fixed according to the non-transgenic mESCs negative control.
(H) Quantification of (G) data (n = 3 independent experiments; ***, p < .001; two-sided unpaired t test; data represented as mean ±SD).
See also Figure S3.
234 Stem Cell Reports jVol. 17 j231–244 jFebruary 8, 2022
in PS-like cells (Figures 1D, S1G, and S2D). To test whether
their endogenous expression was critical for the formation
of the Sox1
GFP+
cells, we generated Chrd
/
Nog
/
2KI and
Dkk1
/
2KI mESCs (Figures 3D, S3C, and S3D). The differ-
entiation of the knockout cells in both cases resulted in
reduced formation of Sox1
GFP+
NPs compared to their
wild-type counterparts (Figures 3E–3H). TGFband Wnt ag-
onists reduced the number of Sox1
GFP+
cells, with a nearly
complete repression of neuroectoderm generation upon
TGFbsignaling activation with ACTIVIN A or BMP4 (Fig-
ures S3E–S3H). These results argue that neuroectoderm for-
mation in our system depends on the balance between the
endogenous levels of agonists and inhibitors of the TGFb
and Wnt pathways.
Diverse neural progenitors emerge in the PS-like
differentiation
scRNA-seq data demonstrated the heterogeneity of the NP
population arising in the PS-like differentiation. Different
anteroposterior identities could be assigned to NP subtypes
according to the expression of markers characteristic of
anterior neural tissues (Hesx1), anterior hindbrain (Egr2
and Hoxa2), posterior hindbrain (Hoxd4), and spinal cord
(Hoxb9)(Figure 4A) (Gouti et al., 2014). We hypothesized
that the coexistence of multiple mechanisms of neural in-
duction in our system could explain the formation of NPs
with distinct developmental origin. To test this, we
compared the transcription profile of Sox1
GFP+
cells derived
from PS-like differentiation with the ones of NPs generated
by Wnt signaling inhibition, TGFbinhibition, or the neural
inducer RA (Ying et al., 2003)(Figure 4B). These NPs ex-
pressed distinct sets of transcription factors that spanned
the set upregulated in the heterogeneous PS-induced NP
population (Figure 4C). Wnt inhibition led to the upregu-
lation of anterior NP markers such as Lhx5,Otx2, and Six3
(Figure 4C), consistent with previous reports (Watanabe
et al., 2005). The differentiation triggered by TGFbinhibi-
tion increased the expression of markers of the posterior
neural axis (Figure 4C). However, these two different NP-in-
duction methods did not account for the full complexity of
the expression profile of NPs obtained in the PS-like differ-
entiation. Indeed, upregulation of markers such as Brn1,
Brn2,orIrx3 were only recapitulated by RA treatment (Fig-
ure 4C). This led us to infer that RA signaling might be in
part responsible for NP formation in our system.
RA signaling mediates neural induction in the PS-like
differentiation
The PS-like differentiation medium contains low concen-
trations of serum, which contains RA precursors that the
cells could convert in RA. To test the presence of RA
signaling, we stably inserted in the 2KI line a reporter
construct relying on the established DR5-based RA
response element (RARE) (Rossant et al., 1991) controlling
the expression of the fluorescent protein Scarlet. We de-
tected Scarlet
+
cells in increasing amount from day 3 of
the PS-like differentiation, demonstrating the activation
of RA signaling and particularly in T
TagBFP+
cells (Figures
4D, S4A, and S4B). This result paralleled the identification
of RA signaling in the mouse PS at E7.5 through a reporter
relying on the same RARE (Rossant et al., 1991).
To identify a possible relationship between RA and NP
formation, we perturbed RA signaling. Supplying the differ-
entiation medium with additional RA precursor, vitamin A
(also known as retinol), increased both RA signaling activa-
tion, as captured by the RA reporter, and the formation of
Sox1
GFP+
cells (Figures S4C–S4E). Inhibition of RA receptors
(RARs) with the small molecule AGN193109 (AGN) pre-
vented Scarlet expression and decreased the fraction of
Sox1
GFP+
cells (Figure S4F). Starting RAR inhibition at early
differentiation time points further reduced NP formation
(Figures S4G and S4H). This result suggested that blocking
RARs could prevent the formation of new neuroectodermal
cells but did not hamper the NPs already present in the
culture.
The impairment of neural induction upon RAR inhibi-
tion prompted us to test the existence of a crosstalk
between RA signaling and the mechanism of neural induc-
tion by TGFbor Wnt inhibition (Figure 4E). Blocking RARs
prevented the increase of the Sox1
GFP+
population nor-
mally associated with the inhibition of either of the two
pathways (Figures 4F and 4G). Moreover, we tested whether
the effects of RAR antagonism were limited to the differen-
tiation regime containing IDE1. As TGFbpathway activa-
tion is an established cue inducing the formation of the
PS in vivo and in hESCs (Gadue et al., 2006;Martyn et al.,
2018), we turned to the Nodal/TGFbagonist ACTIVIN A.
A pulse of ACTIVIN A between day 1 and day 2 generated
both T
TagBFP+
and Sox1
GFP+
cells (Figure 4H). The early
and short-term nature of the ACTIVIN A treatment was crit-
ical to avoid the repression of neural induction by Nodal/
TGFbsignaling, shown in Figure S3E. As for IDE1 differen-
tiation, RAR antagonism significantly reduced the fraction
of Sox1
GFP+
cells induced by TGFbor Wnt inhibitors in the
PS-like differentiation triggered by ACTIVIN A (Figures S4I–
S4K). These results indicated that RARs controlled a step
downstream of TGFbor Wnt inhibition in the cascade of
events leading to neuroectoderm formation.
Aldh1a2-independent RA signaling
We found that RA signaling mediated at least in part the
neuroectoderm induction by the antagonists of the TGFb
or Wnt pathways. Given that only the RA precursor
vitamin A was present in the differentiation medium, cells
had to synthesize RA themselves. The oxidation of retinal
in RA is performed by the retinaldehyde dehydrogenases
Stem Cell Reports jVol. 17 j231–244 jFebruary 8, 2022 235
ALDH1A1, ALDH1A2, or ALDH1A3 (Rhinn and Dolle
´,
2012). Aldh1a2 was upregulated in mesoderm cells and
T
TagBFP+
population compared with Sox1
GFP+
cells, whereas
Aldh1a1 and Aldh1a3 expression did not exceed back-
ground levels (Figures S5A–S5C). This reproduced the
expression pattern of these three genes in post-implanta-
tion mouse embryos, particularly Aldh1a2 expression in
the PS at E7.5 (Ribes et al., 2009).
The presence of forebrain structures in Aldh1a2
/
mouse embryos (Niederreither et al., 1999) seems to contra-
dict our finding that RA signaling mediates early neural in-
duction. We therefore probed whether RA signaling was
completely abolished in Aldh1a2
/
cells. We generated
Aldh1a2
/
mESCs (Figure S5D) bearing the RA activity re-
porter relying on the DR5-RARE used in the mouse model.
PS-like differentiation of Aldh1a2
/
mESCs led to the for-
mation of Sox1
GFP+
cells despite the absence of DR5-RARE-
Scarlet
+
cells (Figures 5A and S5E), in accordance with the
phenotype of Aldh1a2
/
mouse embryos (Niederreither
et al., 1999). We then provided extra precursor for RA
71.2% 14.0%
5.44%
9.45%
TGFß
inhibition
Wnt
inhibition
Retinoic
acid
PS
induced
Six3
Hoxc8
Otx2
Pax6
En2
Dmrt3
Dmrta2
Lmx1a
Pax8
Dach1
Elavl2
Prox1
Bach2
Hivep2
Npas3
Plagl1
Gas7
Lhx1
Msx3
Pou3f2
Irx5
Dbx1
Pou3f3
Btbd11
Irx1
Meis1
Nkx6-1
Olig2
Hes5
Irx3
Rarb
Irx2
Rfx4
Sox3
Sox2
Sox21
Sox5
Caskin1
Mafb
Insm1
Pax6
Lbh
Nr2f2
Pax7
Nr2f1
Prdm16
Tcf7l2
Tgfb1i1
Pax2
Sox9
Ank2
Lhx5
Foxb1
Gbx2
En2
Hes3
Nrarp
Hoxd8
Hoxb5
Hoxc5
Hoxa5
Hoxb6
Hoxa3
Hoxb3
Hoxb8
Hoxb7
Hoxd4
Hoxb4
Hoxd3
Hoxb9
Hoxc4
Hoxc6
Hoxa1
Hoxa2
Hoxb2
Aes
Trps1
Rara
Zic1
Nr3c1
Sall3
Hoxb1
Hoxa7
Hoxc8
Foxa2
Meox1
Eomes
Notch4
Msx2
Sp6
Ebf4
Twist1
Msx1
Prrx1
Egr2
Nkx2-9
Irf1
Mixl1
Nkx1-2
Evx1
Hoxd9
Hoxa9
Hoxc9
Sp5
Hes7
Tbx6
Cdx1
Etv4
Hoxc10
Hoxd10
Cdx2
Cdx4
Egr1
Junb
Maff
Egr3
Klf4
Fos
Fosb
Irf6
Nr4a1
Relb
Elf3
Zfp36
Rarg
Tcea3
Snai1
Lef1
T
Aebp1
Foxc1
Elf4
Klf5
Cited1
Lmo2
Pou3f1
Trp73
Ppp1r1b
Sox1
En1
Nr4a3
Zic2
Ets1
Tead4
Lmx1b
Prrx2
Sox10
Emx2
Ppp1r16b
Mef2c
Lhx2
Six3
Ar
Rxrg
Utf1
Pitx2
Foxh1
Pou5f1
Nfatc4
Hif3a
Stat6
Tal2
Otx1
Otx2
Foxo1
Ankrd1
Zic3
Hox genes
Cdx2
Brn1
Sox2
Brn2
Hoxc10
Lhx5
C
Irx3
04-4
log2(expression)
d4
d5
d6
d4
d5
d6
d5
d5
D
T-TagBFP
Sox1-GFP
DR5-RARE-Scarlet
B
TGFß inhib.
PS-induced
Wnt inhib.
Retinoic acid
Sox1-GFP+ neural progenitors isolation
mRNA-Seq
mESCs
T-H2B-3xTagBFP
Sox1-GFP
Egr2Hesx1
A
Hoxb9Hoxa2 Hoxd4
1
0.5
Fraction of population (a.u.)
0
100101103
Sox1-GFP si
g
nal (a.u.)
Control
+XAV
+XAV+AGN
102
1
0.5
Fraction of population (a.u.)
0
100101103
Sox1-GFP si
g
nal (a.u.)
Control
+SB43
+SB43+AGN
F
102
+AGN +AGN
G
Fraction GFP+ cells ± s.d.(%)
0
10
20
30
40
50
***
**
***
*** ***
d0 d1 d2 d3 d4 d5
IDE1 (Control)
IDE1+AGNIDE1
IDE1+XAV+AGNIDE1
IDE1+XAVIDE1
IDE1+SB43IDE1
IDE1+SB43+AGN
IDE1
E
Tgfß
inhib. Wnt
inhib.
RAR
inhib.
RAR
inhib.
101
Sox1-GFP signal (a.u.)
T-TagBFP signal (a.u.)
102103
101
102
103
H
log10(expression levels) 1000
% of max 50
Figure 4. Retinoic acid signaling underlies NP formation in the PS-like differentiation
(A) UMAP colored by the scaled expression of neuroectodermal markers, ordered by their expression along the anteroposterior axis in vivo.
(B) Scheme of the experimental strategy to characterize NPs induced by PS-like cells, TGFb, or Wnt pathway inhibition or retinoic acid (RA)
treatment.
(C) Expression levels of transcription factors and regulators differentially expressed in Sox1
GFP+
cells.
(D) Reporter expression after 5 days of differentiation of 2KI mESCs transgenic for a DR5-based RA signaling reporter. Bar: 100 mm.
(E) Experimental strategy to assess the crosstalk between RA signaling and TGFbor Wnt pathway inhibition on fate induction. AGN193109
(AGN) is a RAR antagonist.
(F) (left panel) Sox1-GFP expression levels after PS-like differentiation (black: control, orange: +AGN, purple: +SB43, pink: +SB43 + AGN).
(right panel) Sox1-GFP expression levels after PS-like differentiation (black: control, orange: +AGN, teal blue: +XAV, light green: +XAV +
AGN).
(G) Quantification of (F) data (n = 3 independent experiments; **, p < .01; ***, p < .001; one-way ANOVA followed by Tukey’s post-hoc
test; data represented as mean ±SD).
(H) T-TagBFP and Sox1-GFP reporter expression as measured at day 5 by flow cytometry. The differentiation was induced by a pulse of
ACTIVIN A from day 1 to day 2 and without IDE1. Dotted lines: gates fixed according to the non-transgenic mESCs negative control. See also
Figure S4.
236 Stem Cell Reports jVol. 17 j231–244 jFebruary 8, 2022
GFP+
13.2%
Scarlet+
6.13%
GFP+
19.9%
Scarlet+
7.81%
Scarlet+
12.7%
Scarlet+
0.018%
1
0.5
Fraction of population (a.u.)
0
100101103
Sox1-GFP signal (a.u.)
C
102
T-TagBFP
Sox1-GFP
cDR-RARE-Scarlet
Aldh1a2+/+
F
T-TagBFP
Sox1-GFP
DR5-RARE-Scarlet
Aldh1a2-/-
A
T-TagBFP
Sox1-GFP
cDR-RARE-Scarlet
Aldh1a2-/-
G
AGGTCAgaAGTTCAAGGTCA
DR2 DR0
DR8
min CMV
H2B-Scarlet
NLS-Scarlet-PEST2D
3x cDR-RARE
BGH pA
cDR-RARE:
E
T-H2B-3xTagBFP
Sox1-GFP
Aldh1a2-/- mESCs
RAR inhib
+AGN
+vit.A
3 day PS diff.
(IDE1)
control
assess pop. composition
2 day PS diff.
(IDE1)
B
1
0.5
Fraction of population (a.u.)
0
100101103
Sox1-GFP signal (a.u.)
102
WT
H
101
Sox1-GFP si
g
nal (a.u.)
cDR-RARE-Scarlet signal (a.u.)
102103
101
102
103
J2KI
101
Sox1-GFP si
g
nal (a.u.)
102103101
Sox1-GFP si
g
nal (a.u.)
102103101
Sox1-GFP si
g
nal (a.u.)
102103
Aldh1a2+/+ Aldh1a2-/- Aldh1a2-/-Rbp1-/-Stra6-/-
GFP+
30.6%
Rbp1-/-Stra6-/-
+vit.A
Rbp1-/-Stra6-/-
Aldh1a2-/-
Aldh1a2-/-+vit.A
Aldh1a2-/-+AGN
D
Fraction GFP+ cells ± s.d.(%)
0
10
20
40
50
60 Aldh1a2-/-
+vit.A
+AGN
30
70
**
**
Fraction GFP+ cells ± s.d.(%)
0
10
20
30
40
50
60
70
80
90
I
*
*** WT
Rbp1-/-Stra6-/-
+vit.A
Rbp1-/-Stra6-/-
DR5-RARE::Scarlet
GFP+
25.5%
Figure 5. Aldh1a2-independent RA signaling during PS-like differentiation
(A) Reporter expression after 5 days of PS-like differentiation of 2KI Aldh1a2
/
mESCs transgenic for an RA signaling reporter (DR5-RARE-
Scarlet). Bar: 100 mm.
(B) Scheme of the experimental principle to monitor the impact of perturbing RA signaling on the PS-like differentiation of Aldh1a2
/
mESCs.
(C and D) Sox1-GFP reporter expression in Aldh1a2
/
cells after PS-like differentiation. (C) (black: control, orange: AGN, red: vitamin A
and quantification). (D) (n = 3 independent experiments; **, p < .01; two-sided unpaired t test; data represented as mean ±SD).
(E) Scheme of an RA-responsive transcriptional reporter relying on three RA-responsive elements (RARE), each consisting of three RAR
binding sites (DR, direct repeats; cDR, composite direct repeat; min CMV, minimal CMV promoter; BGH pA, bovine growth hormone poly A).
(F and G) Reporter expression after 5 days of PS-like differentiation of 2KI wild-type (Aldh1a2
+/+
) (F) or 2KI Aldh1a2
/
(G) mESCs
transgenic for the cDR RA signaling reporter. Bar: 100 mm.
(legend continued on next page)
Stem Cell Reports jVol. 17 j231–244 jFebruary 8, 2022 237
synthesis or inhibited RARs during PS-like differentiation
(Figure 5B). The fraction of Sox1
GFP+
cells, in fact, should
be insensitive to these treatments in absence of RA produc-
tion. However, the RAR antagonist AGN led to a decrease of
the fraction of Sox1
GFP+
cells (Figures 5C and 5D). Upon
addition of vitamin A, a majority of cells expressed Sox1
GFP
and the DR5-RARE-Scarlet reporter could now be detected
in Aldh1a2
/
differentiating cultures (Figures 5C, 5D,
S5E, and S5F). Altogether, Aldh1a2 loss did not abrogate
RA signaling and this, in turn, could not be fully captured
by the DR5-RARE reporter.
We wondered whether Aldh1a2
/
cells could still
respond to RA at the concentration present in wild-type
cultures and, vice versa, whether the response of
Aldh1a2
+/+
cells to RA was affected by the presence of cells
impaired in RA synthesis. To address both questions, we set
up a co-culture experiment by mixing wild-type Aldh1a2
+/+
cells and mutant Aldh1a2
/
cells (Figure S5G). Under such
conditions, DR5-RARE-Scarlet
+
cells were found in the
Aldh1a2
/
fraction at a rate comparable to the one in
Aldh1a2
+/+
cells (Figures S5H and S5I). This proved that
RA signaling was paracrine in our in vitro system. Moreover,
the fraction of Aldh1a2
+/+
cells expressing the RA reporter
was reduced in the co-culture setting (Figure S5I) compared
to a pure wild-type culture (Figure S5H). This observation
implied that a cell’s response to RA did not depend on its
own RA production but rather on the overall RA level pre-
sent in the medium, that is the regulation of Scarlet expres-
sion was non-cell-autonomous.
A highly sensitive RA reporter captures Aldh1a2-
independent RA signaling
Our results stressed that the DR5-based RARE might cap-
ture only a subset of conditions in which RA signaling
was present. Thus, we turned to a composite RARE
(cDR-RARE) consisting of three RAR binding sites (Fig-
ure 5E) that was found to have much higher affinity for
RARs than the DR5-based RARE (Moutier et al., 2012). A
reporter construct relying on cDR-RARE driving Scarlet
expression could detect sub-nanomolar concentrations
of exogenously applied RA (Figure S5J). Many more cDR-
RARE-Scarlet
+
cells could be detected during PS-like differ-
entiation (Figure 5F) compared with the DR5-based re-
porter (Figures S5K and S5L). The complete abrogation
of cDR-RARE-Scarlet expression upon RAR antagonism
confirmed its reliance on RA signaling (Figure S5L).
Crucially, the PS-like differentiation of Aldh1a2
/
cells
bearing this reporter confirmed that RA signaling was
reduced but not absent (Figures 5G, S5L–S5N). As for the
DR5-based reporter, the cDR-RARE-Scarlet reporter was
expressed particularly in T
TagBFP+
cells (Figure S5M). Alto-
gether, a fraction of RA was produced in an Aldh1a2-inde-
pendent manner at sufficient levels to be detected by the
cDR-RARE reporter and to impact the formation of
Sox1
GFP+
cells.
Vitamin A availability regulates RA signaling levels
during PS-like differentiation
The increase of the fraction of Sox1
GFP+
cells with vitamin A
levels in the medium (Figure S4E) implied that cells were
sensitive to the external vitamin A concentration. Interest-
ingly, the transcript levels of the cellular retinol binding
protein Rbp1 and the Rbp-receptor Stra6, a major mediator
of the cellular uptake of vitamin A (Kawaguchi et al., 2007),
were upregulated during PS-like differentiation (Figures
S5O and S5P). Elevated expression levels of Rbp1 were
found in the mouse PS (Ruberte et al., 1991). RBP1 binds
to vitamin A and is thought to increase its intracellular con-
centration and to help RA synthesis (Napoli, 2016). We hy-
pothesized that vitamin A uptake through STRA6 and
intracellular storage by RBP1 could contribute to control
RA levels. Therefore, we generated Rbp1
/
Stra6
/
mESCs
(Figure S5Q) and submitted them to PS-like differentiation.
Rbp1
/
Stra6
/
cells generated fewer NPs compared to
their wild-type counterparts, but most cells were Sox1
GFP+
when increasing vitamin A concentration (Figures 5H
and 5I). Similarly, the fractions of both Sox1
GFP+
and
cDR-RARE-Scarlet
+
populations were decreased after PS-
like differentiation of Aldh1a2
/
Rbp1
/
Stra6
/
cells
compared with Aldh1a2
/
cells (Figures 5J and S5R). This
demonstrated that the control of intracellular vitamin A
levels via the Stra6-Rbp1 axis contributed to determine
RA signaling activation.
Cyp26a1 limits RA signaling during PS-like
differentiation
A possible additional mechanism to define the subset of
RA-responding cells is the control of RA degradation
mediated by cytochrome P450 CYP26 enzymes (Rhinn
and Dolle
´,2012). Cyp26a1 mRNA levels were highly
(H) Sox1-GFP reporter expression after PS-like differentiation of 2KI cells (gray) or Rbp1
/
Stra6
/
cells without (blue) or with (red)
additional vitamin A.
(I) Quantification of (H) data (n = 3 independent experiments; *, p < .05; ***, p < .001; two-sided unpaired t test; data represented as
mean ±SD).
(J) Sox1-GFP and cDR-RARE-Scarlet reporter expression after PS-like differentiation of 2KI mESCs, 2KI mESCs transgenic for the cDR RA
signaling reporter (Aldh1a2
+/+
), Aldh1a2
/
,orAldh1a2
/
Rbp1
/
Stra6
/
mESCs. Dotted lines: gates fixed according to the non-
transgenic mESCs negative control. See also Figure S5.
238 Stem Cell Reports jVol. 17 j231–244 jFebruary 8, 2022
upregulated during PS-like differentiation and particu-
larly in T
TagB FP +
cells compared with Sox1
GFP+
cells (Fig-
ures S6A and S6B), mirroring its expression pattern in
the PS in vivo at E7.0 (Fujii et al., 1997). To test whether
Cyp26a1 hampered neural induction by limiting RA
levels, we inactivated Cyp26a1 (Figure S6C). Compared
with wild-type cells, Cyp26a1
/
cultures produced
more Sox1
GFP+
cells and displayed many more cells
with active RA signaling (Figures 6A and 6B). Cyp26a1
loss further enhanced NP formation upon vitamin A
addition (Figures S6DandS6E).
We investigated how the loss of Cyp26a1 impacted neu-
ral fate acquisition in response to different RA concentra-
tions (Figure S6F). At low RA concentrations, the majority
of the Cyp26a1
/
cells were Sox1
GFP+
, in contrast to
wild-type cells (Figure S6G). However, Cyp26a1
/
and
wild-type cells displayed similar capacity to differentiate
to neuroectoderm at higher RA concentrations (Fig-
ure S6H). These findings demonstrated that Cyp26a1 plays
a key role in reducing RA levels and in the acquisition of
neural fate during PS-like differentiation, particularly in
response to low RA concentrations.
Impaired RA degradation increases neuroectoderm
formation in Chrd
/
Nog
/
and Dkk1
/
cells
We sought to test whether the impaired neuroectoderm
formation due to the absence of TGFbor Wnt inhibitors
could be counteracted by Cyp26a1 loss, which increases
the response to RA signaling. Therefore, we generated
mESCs lacking Cyp26a1 and either TGFbor Wnt
antagonists and subjected them to PS-like differentiation
(Figures 6C, S6I, and S6J). Chrd
/
Nog
/
Cyp26a1
/
and
Dkk1
/
Cyp26a1
/
cells formed more NPs compared
with Chrd
/
Nog
/
or Dkk1
/
cells (Figures 6D–6G).
RAR inhibition reversed this increase, proving that the ef-
fect was strictly dependent on RA signaling (Figures 6D–
6G). Altogether, Cyp26a1-mediated dampening of RA
signaling was a critical mechanism to reduce the exposure
of the PS-like population to the differentiating effects of the
RA they produce.
RA signaling status accounts for NP diversity
We went on to characterize the gene expression changes
associated with active (cDR-RARE-Scarlet
+
cells) or inactive
or low (AGN-treated cells and Aldh1a2
/
cells) RA
GFP+
31.6%
Scarlet+
15.6%
GFP+
12.5%
Scarlet+
3.31%
GFP+
0.28%
Scarlet+
0.13%
101
Sox1-GFP signal (a.u.)
DR5-RARE-Scarlet signal (a.u.)
102
100
103
101
102
103
104
A
101
Sox1-GFP signal (a.u.)
102103
1
0.5
Fraction of population (a.u.)
0
100101103
Sox1-GFP si
g
nal (a.u.)
102
F
1
0.5
Fraction of population (a.u.)
0
100101103
Sox1-GFP si
g
nal (a.u.)
102
D
101
Sox1-GFP signal (a.u.)
102103
Non-transgenic mESCs WT Cyp26a1-/-
Chrd-/-Nog-/-
Cyp26a1-/-Chrd-/-Nog-/-
Cyp26a1-/-Chrd-/-Nog-/-
+AGN
Dkk1-/-
Cyp26a1-/-Dkk1-/-
Cyp26a1-/-Dkk1-/-
+AGN
B
Fraction of cells ± s.d.(%)
0
10
20
30
40
Scarlet+
GFP+
**
*
E
Fraction GFP+ cells ± s.d.(%)
0
10
20
40
50
60
30
70
Chrd-/-Nog-/-
Cyp26a1-/-Chrd-/-Nog-/-
Cyp26a1-/-Chrd-/-Nog-/-
+AGN
*** ***
G
Fraction GFP+ cells ± s.d.(%)
0
10
20
40
50
60
30
70
*** ***
Dkk1-/-
Cyp26a1-/-Dkk1-/-
Cyp26a1-/-Dkk1-/-
+AGN
Chrd-/-Nog-/- Dkk1-/-
C
PS-like
diff.
assess pop. composition
+/- Cyp26a1 KO
mESCs
+/- Cyp26a1 KO
mESCs
WT
Cyp26a1-/-
Figure 6. Cyp26a1 limits RA levels and neuroectoderm differentiation during PS-like differentiation
(A) Sox1-GFP and DR5-RARE-Scarlet reporter expression after PS-like differentiation of non-transgenic wild-type (WT) or Cyp26a1
/
mESCs. Dotted lines: gates fixed according to the non-transgenic mESCs negative control.
(B) Quantification of (A) data (n = 3 independent experiments; **, p < .01; ***, p < .001; two-sided unpaired t test; data represented as
mean ±SD).
(C) Scheme to assess the interplay between Cyp26a1-mediated dampening of RA signaling and TGFbor Wnt signaling.
(D and E) Sox1-GFP reporter expression in Chrd
/
Nog
/
(pink) or Cyp26a1
/
Chrd
/
Nog
/
(purple, orange: AGN added after day 3)
cells after PS-like differentiation (D) and quantification (E) (n = 3 independent experiments; ***, p < .001; One-way ANOVA followed by
Tukey’s post hoc test; data represented as mean ±SD).
(F and G) Sox1-GFP reporter expression in Dkk1
/
(light green) or Cyp26a1
/
Dkk1
/
(dark green, orange: AGN added after day 3) cells
after PS-like differentiation (F) and quantification (G) (n = 3 independent experiments; ***, p < .001; one-way ANOVA followed by Tukey’s
post hoc test; data represented as mean ±SD). See also Figure S6.
Stem Cell Reports jVol. 17 j231–244 jFebruary 8, 2022 239
signaling both in T
TagBFP+
and Sox1
GFP+
subpopulations
(Figure 7A). The expression changes observed in the
T
TagBFP+
cells upon AGN treatment were well correlated
with the ones observed in Aldh1a2
/
cells (Figure 7B)
and were anti-correlated with the expression changes in
cDR-RARE-Scarlet
+
T
TagBFP+
cells (Figure 7C). Active RA
signaling either repressed (Figure S7A) or increased (Fig-
ure S7B) target gene expression. We next assessed genes
differentially expressed in the Sox1
GFP+
cells distinguished
by their RA signaling status and their expression in the NP
subpopulations defined by scRNA-seq (Figure 7D). The NP-
2, NP-3, NP-4, and NP-5 populations resembled the
B
-6 -3
log2(fold change)
T-TagBFP+ fraction
AGN vs. untreated
036
log2(fold change)
T-TagBFP+ fraction
Aldh1a2-/- vs. Aldh1a2+/+
-6
-3
0
3
6r=0.665
C
-6 -3
log2(fold change)
T-TagBFP+ fraction
AGN vs. untreated
036
log2(fold change)
T-TagBFP+ fraction
cDR-RARE-Scarlet+ vs. Scarlet-
-2
-1
0
1
2r=-0.359
Hox genes
J
WT
3Rar-/-
WT+AGN
02-2
log2(expression)
Sox2
Tbx6
T-TagBFP+ fraction
Zic1Irx2
Hoxb9
Hoxb8
E
genes
NP-1
NP-2
NP-3
NP-4
NP-5
NP-6
NP-7
NP-8
Sox1-GFP+/
cDR-RARE-Scarlet+
Sox1-GFP+/
cDR-RARE-Scarlet-
scRNA-Seq
cell types
D
03-3
log2(expression)
active RA signaling
NP-1
NP-2
NP-3 NP-4
NP-5
NP-6 NP-7
NP-8
low RA signaling
F
mESCs
RAR inhib
+AGN
3 day PS diff.
(IDE1)
control
FACS-purify & mRNA-Seq
2 day PS diff.
(IDE1)
A
Aldh1a2-/- mESCs
5 day PS diff.
(IDE1)
FACS-purify & mRNA-Seq
T-TagBFP
Sox1-GFP
cDR-RARE::Scarlet
mESCs
FACS-purify & mRNA-Seq
TagBFP+ GFP+ TagBFP+ GFP+ TagBFP+ GFP+ Scarlet- Scarlet+
TagBFP+GFP+TagBFP+GFP+
Normal
RA synthesis
& signaling
RAR block Impaired
RA synthesis
ActiveNot active
RA signaling
1
0.5
Fraction of population (a.u.)
0
10010110
Sox1-GFP signal (a.u.)
102
+vit.A
+AGN
T-TagBFP
Sox1-GFP
Rara-/-Rarb-/-Rarg-/- mESCs
3 day PS diff.
(IDE1)
RA signaling
modulation
2 day
PS diff.
(IDE1)
Wnt inhib.
+XAV
TGFß inhib.
+SB43
NP-inducing
si
g
nals
+vit. A
RAR inhib.
+AGN
assess pop. composition
G
Foxd4
1
0.5
Fraction of population (a.u.)
0
100101103
Sox1-GFP signal (a.u.)
102
H
+XAV
Rara-/-Rarb-/-Rarg-/-
+vit.A
+AGN
Rara-/-Rarb-/-Rarg-/-
+SB43
I
Fraction GFP+ cells ± s.d.(%)
0
10
20
30
40 Control
*
**
+AGN +XAV
+vit.A
+SB43
5 day PS diff.
(IDE1)
T-TagBFP
Sox1-GFP
DR5-RARE::Scarlet
T-TagBFP
Sox1-GFP
DR5-RARE::Scarlet
DR5-RARE::Scarlet
Cdx1
100050
10(expression levels) % of maxlog
n.s.
n.s.
Figure 7. Role of RA signaling and RARs in neural commitment and in establishing NP diversity
(A) Experimental strategy to assess the impact of RA signaling on gene expression during PS-like differentiation in T
TagBFP+
(TagBFP+) and
Sox1
GFP+
(GFP+) populations.
(B and C) Comparison of differential expression in T
TagBFP+
cells after AGN treatment or in Aldh1a2
/
cells (B) (Pearson’s r = 0.665, p = 10
–
181
), or in cDR-RARE-Scarlet
+
cells with active RA signaling (C) (Pearson’s r = 0.359, p = 2.10
–44
).
(D) Expression levels of genes differentially expressed between cDR-RARE-Scarlet
+
and Scarlet
–
subpopulations of the Sox1
GFP+
fraction
after PS-like differentiation and in the eight NP categories identified by scRNA-seq.
(E) UMAP colored by the scaled expression of markers identifying different NP territories.
(F) UMAP with the RA signaling status highlighted in the NP territories.
(G) Experimental strategy to monitor the impact of RA signaling modulation or NP-inducing cues in Rara
/
Rarb
/
Rarg
/
mESCs.
(H and I) Sox1-GFP reporter expression in Rara
/
Rarb
/
Rarg
/
cells after PS-like differentiation (H) (black: control, orange: AGN, red:
vitamin A, purple: SB43, teal blue: XAV) and quantification (I) (n = 3 independent experiments; *, p < .05; **, p < .01; n.s.: not significant;
two-sided unpaired t test; data represented as mean ±SD).
(J) Expression levels of transcription factors with differential expression in the T
TagBFP+
fraction after PS-like differentiation of
Rara
/
Rarb
/
Rarg
/
(3Rar
/
) cells or wild-type cells without (WT) or with (WT+AGN) treatment with the RAR antagonist AGN.
See also Figure S7.
240 Stem Cell Reports jVol. 17 j231–244 jFebruary 8, 2022
Sox1
GFP+
cells with active RA signaling, whereas the NP-6,
NP-7, and NP-8 classes shared a signature with cDR-
RARE-Scarlet negative cells (Figure 7D). Among the tran-
scription factors differentially expressed were NP markers
associated with distinct anteroposterior identity such as
Irx2,Zic1,Hoxb8, and Hoxb9 (Figure 7E). Genes character-
istic of anterior NP identity were upregulated in the NP-1
subpopulation (Figures S7C and S7D) and in AGN-treated
cells compared with cDR-RARE-Scarlet
-
(Figure S7E). Alto-
gether, the activation status of RA signaling accounted for
differences in the signatures of the NP subpopulations
identified by scRNA-seq. We broadly distinguished two
NP subgroups with high or low RA signaling, marked by
different expression levels of RA target genes such as
Rarb,Cdx1, and Neurog2 (Figures 7F and S7F).
RAR knockout cells exhibit increased propensity to
neuroectoderm differentiation
RARs are the transcriptional effectors of RA signaling (Cham-
bon, 1996).To obtain a condition where RA signalingcannot
be transduced, we derived mESCs lacking all three RARs
(RAR-null cells) (Figure S7G). Indeed, RA failed to induce
the expression of the DR5-RARE-based reporter in
Rara
/
Rarb
/
Rarg
/
mESCs (Figure S7H). Despite the
absence of RA signaling, RAR-null cells generated a Sox1
GFP+
population after PS-like differentiation. We tested whether
neural fate induction in these cells was responsive to alter-
ations of RA signaling or to TGFband Wnt inhibition (Fig-
ure 7G). The formation of Sox1
GFP+
cells was completely
insensitive to the addition of vitamin A or AGN (Figures 7H
and 7I), unlike the Aldh1a2
/
condition. A partial increase
of the fraction of NPs was observed after inhibiting TGFb
or Wnt signaling (Figures 7H and 7I). The results show a clear
functional distinction with regard to neuroectoderm forma-
tion between RAR inhibition and a complete RAR loss.
To understand the differences at the transcriptional level,
we compared the gene expression profile of the T
Tag BFP +
and
Sox1
GFP+
subpopulations of Rara
/
Rarb
/
Rarg
/
cells and
the ones of wild-type cells or of AGN-treated cells. Interest-
ingly, the expression of Tbx6, whose loss in vivo leads to
the formation of neural tissue at the expense of somites
(Chapman and Papaioannou, 1998), was reduced in T
Tag BFP +
RAR-null cells(Figure 7J). In addition, these cells upregulated
Sox2 (Figure 7J), whose misexpression in paraxial mesoderm
causes ectopicneural tube formation (Takemoto et al.,2011).
RAR-null Sox1
GFP+
cells downregulated genes of theHox and
Cdx families (Figure S7I). Furthermore, the expression of
some RA target genes, such as Rarb,Hoxb1,andNeurog2,
was lower in AGN-treated cells compared with RAR-null
ones (Figure S7I). Thus, the lack of RARs had distinct effects
compared with their pharmacological inhibition and could
not be assimilated to an absence of RA signaling. Interest-
ingly, in mediumdevoid of RA precursors, thedifferentiation
of RAR-null cells yielded many more Sox1
GFP+
cells
compared with wild-type cells, demonstrating the impor-
tance of RARs in the homeostasis of neuroectoderm forma-
tion (Figures S7JandS7K).
DISCUSSION
In this work, we combined a culture system reproducing
the maturation of primitive streak-like cells and the forma-
tion of both anterior and posterior neuroectodermal fates
with a large collection of reporter mESC lines harboring ge-
netic ablations of key signaling factors. In such a context,
we showed that RA signaling drove early neural induction
downstream of Wnt or TGFbinhibition and that multiple
components of the RA pathway contribute to neuroecto-
derm differentiation.
The inhibition of Wnt or TGFbpathways starts the for-
mation of neural lineage in the anterior epiblast of the
mouse conceptus. The reduced NP formation in
Chrd
/
Nog
/
or Dkk1
/
cells was reminiscent of the cor-
responding mutant mouse embryos lacking anterior neural
structures (Bachiller et al., 2000;Mukhopadhyay et al.,
2001). The PS-like differentiation generated a spectrum of
anterior and posterior neuroectoderm. These NP subpopu-
lations differed in their RA signaling status, with markers of
anterior fates being expressed in cells with low RA
signaling. This is in accordance with the proposed caudal-
izing effects of RA during development (Durston et al.,
1989). More importantly, we showed that the mechanisms
of neural induction at work in the PS-like culture were not
independent, because neural induction through inhibition
of Wnt or TGFbsignaling was hindered by blocking RA re-
ceptors. This suggested that the transduction of RA
signaling mediates a step downstream of Wnt and TGFbin-
hibition in the cascade of events leading to the acquisition
of neural fate.
Aldh1a2-mediated synthesis was crucial to generate high
levels of RA signaling during PS-like differentiation but did
not account for all RA production. While the presence of
RA signaling during the differentiation of Aldh1a2
/
cells
awaits confirmation in future in vivo studies, its significance
lies in the identification of alternative ways to respond to
RA. Indeed, the different sensitivity of distinct RAREs to
RA levels would enable cells to switch on different gene rep-
ertoires depending on RA concentration, thus generating
positional information.
In order to safeguard their developmental capabilities,
pluripotent cells such as the epiblast/PS-like population
should protect themselves from the neuralizing action of
RA and therefore need to carefully control the RA levels
they are exposed to. We found that the PS-like cells tuned
RA synthesis via the regulation of vitamin A availability
Stem Cell Reports jVol. 17 j231–244 jFebruary 8, 2022 241
through the Rbp1-Stra6 axis. Furthermore, we determined
that RA synthesis was not entirely dependent on Aldh1a2
and could not be attributed to another single aldehyde de-
hydrogenase by systematically knocking out all the ones
expressed in PS cells. Our co-culture experiment showed
that a cell’s response to RA was not intrinsically determined
by its own RA production. In this context, Cyp26a1-medi-
ated RA degradation is a crucial checkpoint, limiting the
differentiation toward the neural lineage of PS-like cells.
Altogether, cells exploited a three-tiered control of RA
levels regulating precursor availability, RA synthesis, and
degradation in order to induce neural differentiation only
in a subpopulation of cells.
The differentiation of Rara
/
Rarb
/
Rarg
/
mESCs
underlined an even more complex involvement of the RA
pathway in neuroectoderm differentiation. Unlike
Aldh1a2
/
cells, RAR-null cells were completely devoid
of RA signaling. NP formation in absence of RARs seems
in contradiction with the reduction of neuroectoderm in-
duction by blocking RA signaling. However, this can be ex-
plained by the binding of the receptors to their cognate
RAREs in the absence of RA (Chambon, 1996). Their phys-
ical absence in RAR-null cells would remove this control
mechanism and unmask binding sites, making them avail-
able to other nuclear receptors and transcription factors.
Thus, besides being effectors of RA signaling, RARs might
gate the expression at genomic loci important for neural
specification.
In conclusion, the flexibility of our in vitro system allowed
the manipulation of the external environment in a
controlled manner. This led us to recognizethe involvement
of RA signaling in early neural induction. Our results high-
light the potential of ESC-based systems to gain new insights
about lineage specification mechanisms. Notwithstanding
the strengths of this approach, it will be beneficial to exploit
the tools we developed and test our findings in in vivo models
or more complex tridimensional culture systems.
EXPERIMENTAL PROCEDURES
mESC maintenance
The parental mESC line was a Sox1-Brachyury double knock-in line
(Sladitschek and Neveu, 2019). mESCs were maintained in ‘‘LIF +
serum’’ as described previously (Sladitschek and Neveu, 2015b).
Generation of knockout mESC lines
RNA-guided Cas9 nucleases (Hsu et al., 2013) were used to inacti-
vate Aldh1a2,Chrd,Cyp26a1,Dkk1,Nog,Rara,Rarb,Rarg,Rbp1,
and Stra6. See supplemental experimental procedures for details.
Reporter constructs
Constructs were assembled following Sladitschek and Neveu
(2015a).
Transcriptional reporters relied on mScarlet (Bindels et al., 2017)
and different RAREs (Moutier et al., 2012;Rossant et al., 1991).
See supplemental experimental procedures for details.
Transgenic mESC lines
A list of all transgenic cell lines used in this study can be found in
supplemental experimental procedures.
Primitive streak-like differentiation
Differentiation toward a primitive streak-like fate was performed
using IDE1 (Sladitschek and Neveu, 2019) or a pulse of ACTIVIN
A. See supplemental experimental procedures for details.
Neural progenitor differentiation
mESCs were differentiated to NPs using RA or inhibition of TGFbor
Wnt signaling. See supplemental experimental procedures for
details.
Pharmacological treatments
Details of pharmacological treatments can be found in supple-
mental experimental procedures.
Imaging
Reporter fluorescence was assessed in live cells. Images were ac-
quired on an inverted SP8 confocal microscope (Leica) equipped
with a 403PLApo 1.1W objective and an incubation chamber at
37C and 5% CO
2
.
Flow cytometry
Cells were FACS-purified using an Aria Fusion sorter (BD BioSci-
ences). Samples were analyzed on an LSRFortessa flow cytometer
(BD BioSciences), and data was analyzed with FlowJo. See supple-
mental experimental procedures for details.
RNA-seq library construction
mRNA sequencing was conducted as previously described (Sladit-
schek et al., 2020). See supplemental experimental procedures for
details.
RNA-seq analysis
mRNA read counts were determined using Bowtie (Langmead
et al., 2009). edgeR (Robinson et al., 2010) was used for differential
gene expression analysis. See supplemental experimental proced-
ures for details.
Single-cell RNA sequencing
Samples for scRNA-seq were processed with a Chromium
Controller and reagents (103Genomics). See supplemental exper-
imental procedures for details.
scRNA-seq analysis
scRNA-seq data was pre-processed as described in supplemental
experimental procedures. 46,700 cells passed quality controls.
Expression levels were normalized using Seurat methods (Satija
et al., 2015). Dimensionality reduction was performed using
242 Stem Cell Reports jVol. 17 j231–244 jFebruary 8, 2022
UMAP (Becht et al., 2018). Clustering and marker determination
are described in supplemental experimental procedures.
Statistical analysis
Statistical tests were computed using R or the Python SciPy mod-
ule. Data is represented as mean ±SD. Two-sided unpaired Stu-
dent’s t test was used for pairwise comparison with a fixed control
condition. For multiple pairwise comparisons with different con-
trol and treatment conditions, one-way ANOVA analysis followed
by Tukey’s post hoc test was used. Values with p< 0.05 were consid-
ered significant.
Data and code availability
Sequencing results are deposited on ArrayExpress with accession
numbers ArrayExpress: E-MTAB-10242 and ArrayExpress: E-
MTAB-10243. In addition, we used the datasets ArrayExpress: E-
MTAB-2830, ArraxExpress: E-MTAB-3234 (Sladitschek and Neveu,
2015b), and ArrayExpress: E-MTAB-4904 (Sladitschek and Neveu,
2019).
SUPPLEMENTAL INFORMATION
Supplemental information can be found online at https://doi.org/
10.1016/j.stemcr.2021.12.014.
AUTHOR CONTRIBUTIONS
L.R. designed experiments, performed most experiments described
in the manuscript, and analyzed data. H.L.S. provided critical pre-
liminary data. P.A.N. conceived and supervised the study, per-
formed experiments, and analyzed data. L.R. and P.A.N. wrote
the paper, and H.L.S. commented on the manuscript.
CONFLICT OF INTERESTS
The authors declare no competing interests.
ACKNOWLEDGMENTS
We thank Lucia Cassella for advice on RNA-seq data analysis and
Laura Villacorta for help with scRNA-seq sample processing. This
work was technically supported by the EMBL Flow Cytometry
Core and Genomics Core facilities. The study was funded by
EMBL. L.R. was also supported by the EMBL International PhD Pro-
gram (EIPP).
Received: November 22, 2021
Revised: December 15, 2021
Accepted: December 16, 2021
Published: January 20, 2022
REFERENCES
Bachiller, D., Klingensmith, J., Kemp, C., Belo, J.A., Anderson,
R.M., May, S.R., McMahon, J.A., McMahon, A.P., Harland, R.M.,
Rossant, J., et al. (2000). The organizer factors Chordin and Noggin
are required for mouse forebrain development. Nature 403, 658–
661.
Becht, E., McInnes, L., Healy, J., Dutertre, C.-A., Kwok, I.W.H., Ng,
L.G., Ginhoux, F., and Newell, E.W. (2018). Dimensionality reduc-
tion for visualizing single-cell data using UMAP. Nat. Biotechnol.
37, 38–44.
Bindels, D.S., Haarbosch, L., van Weeren, L., Postma, M., Wiese,
K.E., Mastop, M., Aumonier, S., Gotthard, G., Royant, A., Hink,
M.A., et al. (2017). mScarlet: a bright monomeric red fluorescent
protein for cellular imaging. Nat. Methods 14, 53–56.
Borowiak, M., Maehr, R., Chen, S., Chen, A.E., Tang, W., Fox, J.L.,
Schreiber, S.L., and Melton, D.A. (2009). Small molecules effi-
ciently direct endodermal differentiation of mouse and human
embryonic stem cells. Cell Stem Cell 4, 348–358.
Chambon, P. (1996). A decade of molecular biology of retinoic acid
receptors. FASEB J. 10, 940–954.
Chapman, D., and Papaioannou, V. (1998). Three neural tubes in
mouse embryos with mutations in the t-box gene Tbx6. Nature
391, 695–697.
De Robertis, E.M. (2006). Spemann’s organizer and self-regulation
in amphibian embryos. Nat. Rev. Mol. Cell Biol. 7, 296–302.
Diez del Corral, R., Olivera-Martinez, I., Goriely, A., Gale, E., Ma-
den, M., and Storey, K. (2003). Opposing Fgf and retinoid pathways
control ventral neural pattern, neuronal differentiation, and seg-
mentation during body axis extension. Neuron 40, 65–79.
Durston, A., Timmermans, J., Hage, W., Hendriks, H., de Vries, N.,
Heideveld, M., and Nieuwkoop, P. (1989). Retinoic acid causes an
anteroposterior transformation in the developing central nervous
system. Nature 340, 140–144.
Fujii, H., Sato, T., Kaneko, S., Gotoh, O., Fujii-Kuriyama, Y., Osawa,
K., Kato, S., and Hamada, H. (1997). Metabolic inactivation of ret-
inoic acid by a novel p450 differentially expressed in developing
mouse embryos. EMBO J. 16, 4163–4173.
Gadue, P., Huber, T.L., Paddison, P.J., and Keller, G.M. (2006). Wnt
and Tgf-beta signaling are required for the induction of an in vitro
model of primitive streak formation using embryonic stem cells.
Proc. Natl. Acad. Sci. U S A 103, 16806–16811.
Glinka, A., Wu, W., Delius, H., Monaghan, A., Blumenstock, C.,
and Niehrs, C. (1998). Dickkopf-1 is a member of a new family of
secreted proteins and functions in head induction. Nature 391,
357–362.
Gouti, M., Tsakiridis, A., Wymeersch, F.J., Huang, Y., Kleinjung, J.,
Wilson, V., and Briscoe, J. (2014). In vitro generation of neurome-
sodermal progenitors reveals distinct roles for Wnt signalling in
the specification of spinal cord and paraxial mesoderm identity.
PLoS Biol. 12, e1001937.
Henrique, D., Abranches, E., Verrier, L., and Storey, K.G. (2015).
Neuromesodermal progenitors and the making of the spinal
cord. Development 142, 2864–2875.
Hogan, B.L., Thaller, C., and Eichele, G. (1992). Evidence that Hen-
sen’s node is a site of retinoic acid synthesis. Nature 359, 237–241.
Hsu, P.D., Scott, D.A., Weinstein, J.A., Ran, F.A., Konermann, S.,
Agarwala, V., Li, Y., Fine, E.J., Wu, X., Shalem, O., et al. (2013).
DNA targeting specificity of RNA-guided Cas9 nucleases. Nat. Bio-
technol. 31, 827–832.
Kawaguchi, R., Yu, J., Honda, J., Hu, J., Whitelegge, J., Ping, P.,
Wiita, P., Bok, D., and Sun, H. (2007). A membrane receptor for
Stem Cell Reports jVol. 17 j231–244 jFebruary 8, 2022 243
retinol binding protein mediates cellular uptake of vitamin A. Sci-
ence 315, 820–825.
Langmead, B., Trapnell, C., Pop, M., and Salzberg, S.L. (2009). Ul-
trafast and memory-efficient alignment of short DNA sequences
to the human genome. Genome Biol. 10, R25.
Levine, A.J., and Brivanlou, A.H. (2007). Proposal of a model of
mammalian neural induction. Dev. Biol. 308, 247–256.
Martyn, I., Kanno, T., Ruzo, A., Siggia, E., and Brivanlou, A. (2018).
Self-organization of a human organizer by combined Wnt and
Nodal signalling. Nature 558, 132–135.
Moutier, E., Ye, T., Choukrallah, M.-A., Urban, S., Osz, J., Chatag-
non, A., Delacroix, L., Langer, D., Rochel, N., Moras, D., et al.
(2012). Retinoic acid receptors recognize the mouse genome
through binding elements with diverse spacing and topology.
J. Biol. Chem. 287, 26328–26341.
Mukhopadhyay, M., Shtrom, S., Rodriguez-Esteban, C., Chen, L.,
Tsukui, T., Gomer, L., Dorward, D., Glinka, A., Grinberg, A., Huang,
S., et al. (2001). Dickkopf1 is required for embryonic head induc-
tion and limb morphogenesis in the mouse. Dev. Cell 1, 423–434.
Napoli, J.L. (2016). Functions of intracellular retinoid binding-pro-
teins. Subcell. Biochem. 81, 21–76.
Niederreither, K., Subbarayan, V., Dolle
´, P., and Chambon, P.
(1999). Embryonic retinoic acid synthesis is essential for early
mouse post-implantation development. Nat. Genet. 21, 444–448.
Peng, G., Suo, S., Cui, G., Yu, F., Wang, R., Chen, J., Chen, S., Liu, Z.,
Chen, G., Qian, Y., et al. (2019). Molecular architecture of lineage
allocation and tissue organization in early mouse embryo. Nature
572, 528–532.
Pevny, L., Sockanathan, S., Placzek, M., and Lovell-Badge, R.
(1998). A role for SOX1 in neural determination. Development
125, 1967–1978.
Rhinn, M., and Dolle
´, P. (2012). Retinoic acid signalling during
development. Development 139, 843–858.
Ribes, V., Le Roux, I., Rhinn, M., Schuhbaur, B., and Dolle
´,P.
(2009). Early mouse caudal development relies on crosstalk be-
tween retinoic acid, Shh and Fgf signalling pathways. Develop-
ment 136, 665–676.
Robinson, M.D., McCarthy, D.J., and Smyth, G.K. (2010). edgeR: a
Bioconductor package for differential expression analysis of digital
gene expression data. Bioinformatics 26, 139–140.
Rossant, J., Zirngibl, R., Cado, D., Shago, M., and Gigue
`re, V.
(1991). Expression of a retinoic acid response element-hsplacz
transgene defines specific domains of transcriptional activity dur-
ing mouse embryogenesis. Genes Dev. 5, 1333–1344.
Ruberte, E., Dolle, P., Chambon, P., and Morriss-Kay, G. (1991). Ret-
inoic acid receptors and cellular retinoid binding proteins. II. their
differential pattern of transcription during early morphogenesis in
mouse embryos. Development 111, 45–60.
Samarut, E., and Rochette-Egly, C. (2012). Nuclear retinoic acid re-
ceptors: conductors of the retinoic acid symphony during develop-
ment. Mol. Cell. Endocrinol. 348, 348–360.
Satija, R., Farrell, J.A., Gennert, D., Schier, A.F., and Regev, A.
(2015). Spatial reconstruction of single-cell gene expression data.
Nat. Biotechnol. 33, 495–502.
Shahbazi, M.N., Siggia, E.D., and Zernicka-Goetz, M. (2019). Self-
organization of stem cells into embryos: a window on early
mammalian development. Science 364, 948–951.
Sladitschek, H.L., Fiuza, U.-M., Pavlinic, D., Benes, V., Hufnagel, L.,
and Neveu, P.A. (2020). Morphoseq: full single-cell transcriptome
dynamics up to gastrulation in a chordate. Cell 181, 922–935.e21.
Sladitschek, H.L., and Neveu, P.A. (2015a). MXS-chaining: a highly
efficient cloning Platform for imaging and flow cytometry ap-
proaches in mammalian systems. PLoS One 10, e0124958.
Sladitschek, H.L., and Neveu, P.A. (2015b). The bimodally ex-
pressed microRNA miR-142 gates exit from pluripotency. Mol.
Syst. Biol. 11, 850.
Sladitschek, H.L., and Neveu, P.A. (2019). A gene regulatory
network controls the balance between mesendoderm and ecto-
derm at pluripotency exit. Mol. Syst. Biol. 15, e9043.
Spemann, H., and Mangold, H. (1924). U
¨ber die Induktion von
Embryonalanlagen durch Implantation artfremder Organisatoren.
Arch. Mikrosk Anat. Entwicklungsmech. 100, 599–638.
Takemoto, T., Uchikawa, M., Yoshida, M., Bell, D.M., Lovell-Badge,
R., Papaioannou, V.E., and Kondoh, H. (2011). Tbx6-dependent
Sox2 regulation determines neural or mesodermal fate in axial
stem cells. Nature 470, 394–398.
Tam, P.P., and Behringer, R.R. (1997). Mouse gastrulation: the for-
mation of a mammalian body plan. Mech. Dev. 68, 3–25.
Watanabe, K., Kamiya, D., Nishiyama, A., Katayama, T., Nozaki, S.,
Kawasaki, H., Watanabe, Y., Mizuseki, K., and Sasai, Y. (2005).
Directed differentiation of telencephalic precursors from embry-
onic stem cells. Nat. Neurosci. 8, 288–296.
Weinstein, D., and Hemmati-Brivanlou, A. (1999). Neural induc-
tion. Annu. Rev. Cell Dev. Biol. 15, 411–433.
Wilkinson, D., Bhatt, S., and Herrmann, B. (1990). Expression
pattern of the mouse T gene and its role in mesoderm formation.
Nature 343, 657–659.
Ying, Q.-L., Stavridis, M., Griffiths, D., Li, M., and Smith, A. (2003).
Conversion of embryonic stem cells into neuroectodermal precur-
sors in adherent monoculture. Nat. Biotechnol. 21, 183–186.
244 Stem Cell Reports jVol. 17 j231–244 jFebruary 8, 2022
