Cell Stem Cell

Resource

A Systems Biology Approach

for Defining the Molecular Framework
of the Hematopoietic Stem Cell Niche
Pierre Charbord,3,* Claire Pouget,4 Hans Binder,5 Florent Dumont,6 Grégoire Stik,1,2 Pacifique Levy,1,2 Fabrice Allain,1,2
Céline Marchal,1,2 Jenna Richter,4 Benjamin Uzan,7 Françoise Pflumio,7 Franck Letourneur,6 Henry Wirth,5
Elaine Dzierzak,8 David Traver,4 Thierry Jaffredo,1,2,9 and Charles Durand1,2,9,*
1Sorbonne Universités, UPMC Paris 06, IBPS, UMR 7622, Laboratoire de Biologie du Développement, 75005 Paris
2CNRS, INSERM U1156, IBPS, UMR 7622, Laboratoire de Biologie du Développement, 75005 Paris, France
3INSERM U972, University Paris 11, Hôpital Paul Brousse, 94807 Villejuif, France
4Department of Cell and Developmental Biology, University of California, San Diego, La Jolla, CA 92093-0380, USA
5Interdisciplinary Center for Bioinformatics, University of Leipzig, 04107 Leipzig, Germany
6Genomic Platform, Institut Cochin, INSERM U567, 75014 Paris, France
7UMR967 INSERM, LSHL/IRCM, CEA, University Paris 7, 92260 Fontenay-aux-Roses, France
8Department of Cell Biology, Erasmus Stem Cell Institute, Erasmus Medical Center, P.O. Box 2040, 3000 CA Rotterdam, the Netherlands
9Co-senior author

*Correspondence: pierre.charbord@inserm.fr (P.C.), charles.durand@upmc.fr (C.D.)

http://dx.doi.org/10.1016/j.stem.2014.06.005

SUMMARY Despite significant progress in identifying cells that comprise

the bone marrow (BM) niche and their molecular characterization
Despite progress in identifying the cellular composi- (Morrison and Scadden, 2014), a molecular comprehensive
tion of hematopoietic stem/progenitor cell (HSPC) understanding of the adult HSC niche has not yet been
niches, little is known about the molecular require- determined. This lack of information is largely due to the
ments of HSPC support. To address this issue, complexity of the BM microenvironment and the difficulty of ob-
we used a panel of six recognized HSPC-supportive taining and studying hematopoietic-supportive cells in living an-
imals. Additionally, it is currently controversial as to what cell
stromal lines and less-supportive counterparts orig-
types actually comprise the murine BM niche because endosteal
inating from embryonic and adult hematopoietic
cells, endothelial cells, perivascular cells, and mesenchymal
sites. Through comprehensive transcriptomic meta- stem cells (MSCs) have all been implicated as key players
analyses, we identified 481 mRNAs and 17 micro- (Calvi et al., 2003; Corselli et al., 2013; Ding et al., 2012; Green-
RNAs organized in a modular network implicated in baum et al., 2013; Kiel et al., 2005; Kunisaki et al., 2013; Méndez-
paracrine signaling. Further inclusion of 18 additional Ferrer et al., 2010; Zhang et al., 2003).
cell strains demonstrated that this mRNA subset was In addition to the BM niche, several additional supportive
predictive of HSPC support. Our gene set contains environments play key roles in HSC support during develop-
most known HSPC regulators as well as a number ment. The aorta gonad mesonephros (AGM) region is respon-
of unexpected ones, such as Pax9 and Ccdc80, as sible for HSC generation, whereas the fetal liver (FL) promotes
validated by functional studies in zebrafish embryos. HSC maturation and amplification (Dzierzak and Speck, 2008).
To better characterize these different HSC niches, we utilized
In sum, our approach has identified the core molecu-
clonal murine stromal cell lines generated from embryonic day
lar network required for HSPC support. These cues,
(E) 11 AGM, E14 FL, and adult BM (Chateauvieux et al., 2007;
along with a searchable web resource, will inform Moore et al., 1997; Oostendorp et al., 2002). These stromal lines
ongoing efforts to instruct HSPC ex vivo amplifica- immortalized with TSV40 exhibit mesenchymal differentiation
tion and formation from pluripotent precursors. potential (Chateauvieux et al., 2007; Durand et al., 2006) and
differentially support HSPCs in coculture experiments (Cha-
teauvieux et al., 2007; Hackney et al., 2002; Moore et al., 1997;
INTRODUCTION Oostendorp et al., 2002). Importantly, comparison of their gene
expression profiles has been instrumental in the identification
The molecular characterization of hematopoietic stem cell (HSC) of additional HSC regulators (Durand et al., 2007; Hackney
microenvironments (also termed niches) is a fundamental goal et al., 2002; Renström et al., 2009).
in the field of stem cell biology and regenerative medicine Here, we present in-depth analyses of the mRNA and micro-
(Wagers, 2012). Although signaling pathways and extracellular RNA (miR) transcriptomes expressed by these stromal lines.
matrix (ECM) components critical for HSC regulation have Using an ensemble of systems biology approaches, we have
been identified (Morrison and Spradling, 2008), the existence established shared molecular commonalities in HSPC niches
of a core genetic network responsible for HSC support remains from distinct temporal and spatial ontogenic locations. Our
elusive. data were incorporated into an interactive, searchable website

376 Cell Stem Cell 15, 376–391, September 4, 2014 ª2014 Elsevier Inc.
Cell Stem Cell
Molecular Analysis of the HSC Niche

Figure 1. General Outline of the Study and

Unsupervised Analysis of Transcripts
(A) Flow chart outlining the analyses of stromal cell
line transcriptomes.
(B) Heatmap and hierarchical clustering on the
basic set of six cell lines (1B6, 3B5, AFT, BFC, B9,
and B10, each in triplicate). Linkage average.
(C) PCA with the entire set of mRNAs as variables
and the basic set of six cell lines as observations.
PC1 versus PC2 score plot.

AFT024 (AFT, HSPC-supportive) and

BFC012 (BFC, nonsupportive) FL lines,
and BMC9 (B9, HSPC-supportive) and
BMC10 (B10, less-supportive) BM lines.

Identification of the Stromal Gene

Network Essential for HSPC
Support
To determine whether we could identify
unique genetic signatures present in
HSPC-supportive versus less-supportive
stromal cell lines, we performed hierarchi-
cal clustering (Figure 1B) and principal
component analysis (PCA; Figure 1C) of
the entire set of mRNAs. Our findings indi-
cated that there were no discernable dif-
ferences between HSPC-supportive and
less-supportive lines (Figure 1C). The first
two components accounted for 29% of
the variance of the results. This unsuper-
vised analysis uncovered the presence
of tissue-imprinted genes that clustered
with the AGM, FL, and BM lines. To
circumvent this signature, we subtracted,
for each site, the gene set significantly ex-
pressed in the less-supportive line from
(http://stemniche.snv.jussieu.fr/) in order to interrogate mRNA the set significantly expressed in the supportive one. For this
and miR networks in the HSPC niches during development and supervised analysis, gene expression levels were compared
adulthood. Altogether, information gleaned from these studies with two-way ANOVA with site and support as interacting
should help devise pharmaceutical drugs for the treatment of factors. Data were filtered with p values < 0.06 for transcripts
leukemias and determine methods for amplifying HSCs ex vivo regarded as statistically significant and fold changes of f R
and generating them from pluripotent stem cells, both of which 1.45 for upregulated genes and f % !1.45 for downregulated
are key issues in regenerative medicine approaches. genes in HSPC-supportive lines. Assuming that genes essential
for HSPC support would be conserved in at least two out of the
RESULTS three tissues analyzed, we defined a group of 481 transcripts
that we designated as set 1 (shaded area in Figure 2A, complete
A Systems Biology Approach gene list and annotations in Table S1 available online) that corre-
To determine the core of genes characteristic of the stromal sponds to 481 unique genes either up- or downregulated in at
HSPC-supportive capacity, we developed a systems biology least two HSPC-supportive cell lines. The lists of differentially
approach whereby high-throughput technology and bioinfor- expressed mRNAs between the HSPC-supportive and less-sup-
matics analyses were combined (Figure 1A). We analyzed portive stromal lines are accessible at http://stemniche.snv.
the mRNA and miR transcriptomes of stromal lines estab- jussieu.fr/. On the score plot (first two components) of PCA using
lished from mouse AGM, FL, and BM. For each site, we chose set 1 genes as variables and 18 samples as observations (Fig-
two lines with differing capacity to maintain HSPCs ex vivo as ure 2B), the segregation between the three stromal lines that pro-
revealed by repopulation assays and/or long-term cultures. vide a potent support for HSPCs (on the right) and the three that
The stromal cells were as follows: UG26.1B6 (1B6, HSPC- do not or less efficiently support HSPCs (on the left) indicate that
supportive) and UG26.3B5 (3B5, less-supportive) AGM lines, the first principal component (PC1) with the largest eigenvalue

Cell Stem Cell 15, 376–391, September 4, 2014 ª2014 Elsevier Inc. 377
 Cell Stem Cell
Molecular Analysis of the HSC Niche

Figure 2. Identification of the Gene Set

that Characterizes the HSPC-Supportive
Capacity of Stromal Cells
(A) Venn diagram of genes obtained by supervised
analysis. Set 1 is shaded.
(B) PCA with set 1 (481) genes as variables and
the basic set of six cell lines as observations
(18 samples). PC1 versus PC2 score plot.
(C) Study of known HSPC regulators belonging to
set 1. Blue bars represent correlations of gene
expression to the first principal component (PC1
loadings). Red bars represent gene rank metric
scores (R) given by GSEA. Green bars represent
gene scaled connectivities K given by WGCNA.
Value ranges: for PC1 and R [!1, 1], for K [0, 1].
(D) PCA with set 1 as variables and a set of 18
cell strains different from the basic set as obser-
vations (48 samples). PC1 versus PC2 score plot.
Green circles correspond to HSPC-supportive cell
strains, and red triangles correspond to non-
supportive strains.
(E) GSEA with set 1 as reference gene set and
the same 48 samples utilized for PCA as expres-
sion data set. Each sample belongs to either one of
the two phenotypes (supportive versus non-
supportive). The top shows distribution of down-
regulated genes in nonsupportive samples. The
bottom shows distribution of upregulated genes in
supportive stromal cells. NES, normalized enrich-
ment score.
See also Table S1.

lated. Poor PC1 correlation value for some

of the known regulators was due to signif-
icantly (p < 0.05) divergent expression in
two sites in comparison to the third (e.g.,
Cxcl12 upregulated in B9 and 1B6 but
downregulated in AFT).
To confirm that our subtractive strategy
has provided an adequate classification of
the known regulators, we utilized set 1
genes as reference gene set in gene-set
enrichment analysis (GSEA). The expres-
sion data set consisted in the same 18
samples used for PCA. Each sample was
corresponds to the factor support. The first two components labeled according to one of the two phenotypes and the genes
accounted for 57% of the variance. The difference in PC1 coef- were ranked on the basis of their level of expression. The values
ficients was maximal when comparing AFT to BFC, which was of the rank metric score (R) for the 30 most reported genes were
expected, given that these two lines were the most distinct in correlated and anticorrelated for positive and negative regula-
terms of hematopoietic support (Hackney et al., 2002). These tors, respectively (Figure 2C). This result further validates the
data indicate that we can effectively enrich for genes essential subtractive strategy with the use of a second line of analysis
for HSPC support by removing site-specific genes identified in (GSEA) totally distinct from the first (PCA).
our original unsupervised analyses. To find the structure of the gene network indicating how genes
Several genes present in set 1 have been previously identified from set 1 are interconnected, we used weighted gene correla-
as critical factors involved in HSPC support, validating our tion network analysis (WGCNA). WGCNA makes use of correla-
experimental approach. In Figure 2C, we indicate for the tion between genes to identify coordinately expressed genes
30 genes most reported in the literature (see Table S1) the value (Langfelder and Horvath, 2008). Moreover, this method indicates
of their correlation to the PC1 loadings. Remarkably, positive how genes are correlated to an external genetic trait, corre-
regulators of HSPCs (such as Col1a1, Ly6a, Ncam1, Pdgfrb, sponding in this work to the factor support quantified for each
S1pr1, Spp1, Ptn, Kitl, and Kirrel3) are positively correlated, line by PCA. Collectively, these analyses allowed network con-
whereas negative regulators (such as Dcn and F2r) are anticorre- struction corresponding to the set of 481 genes without relying

378 Cell Stem Cell 15, 376–391, September 4, 2014 ª2014 Elsevier Inc.
Cell Stem Cell
Molecular Analysis of the HSC Niche

upon literature mining. It is apparent from the values of the con- the variance, a value almost unchanged compared to that found
nectivity K in Figure 2C that some of the known genes highly in the PCA with our 18 initial samples (35%).
correlated to the ‘‘support’’ trait are highly connected to other Then, we utilized set 1 genes partitioned in two gene sets
genes of the network, defining hubs. Moreover, WGCNA allowed of reference (up- and downregulated) in GSEA. The expression
identification, within the network of 481 genes, of five subnet- data set consisted in the same 48 samples as used for PCA,
works (modules) significantly (p % 0.006) correlated to support, each sample being labeled according to one of the two pheno-
three positively (Figure 3A) and two negatively (data not shown). types (supportive versus nonsupportive). Given that the tran-
Although the three positively correlated networks differ in terms scriptomes of the samples had been carried out with different
of mean connectivity (density), they all contain highly connected Affymetrix platforms (Mouse Gene 1.0 ST, Mouse Exon 1.0 ST,
hubs (Figure 3B). For example, among Tgfbi, Snai2, Wnt10b, and and Mouse Expression 430 array), we designed our own
Ccdc80 that all belong to the same ‘‘blue’’ module, only Ccdc80 annotation file. As shown in Figure 2E, the enrichment score
and Snai2 are highly connected to the other genes of the module. plot was shifted to the left in supportive samples, correlating
In contrast, the interconnected Pax9 and Kitl genes are detected well with the HSPC-supportive nature of these cells. Conversely,
in the ‘‘turquoise’’ module in a marginal situation, indicating the enrichment score of nonsupportive cells was shifted to the
‘‘fuzzy’’ membership characteristic of nodes intermediate to right, correlating with their nonsupportive quality. The low values
several modules. Interestingly, Gata3 has a relatively high mem- of the false discovery rate (FDR) indicated that the difference
bership to the ‘‘black’’ module, supporting its recently observed in profiles was highly significant (FDR % 0.03). Altogether, these
role in HSPC regulation (Mirshekar-Syahkal et al., 2014). Alto- data strongly suggest that the list of genes in set 1 carries a
gether, these data identify a gene network characteristic of strong predictive value of stromal cells to support HSPCs.
the HSPC supportive capacity of stromal cells using a limited
number of stromal lines derived from distinct developmental Molecular Pathways Dedicated to HSPC Support
sites and differing in their ability to maintain HSPCs ex vivo. To identify the molecular pathways utilized by genes in the
network characteristic of the stromal supportive function, we per-
Predictive Value of the Gene Network formed Gene Ontology (GO) analysis with our basic set of six lines;
Then, we investigated whether the list of set 1 genes harbors i.e., 18 samples. Using the filters described above, we analyzed
a predictive advantage of supportive ability. To test this hypoth- the gene lists established for each HSPC-supportive line versus
esis, we studied 18 additional cell strains. To attain this number, its less-supportive counterpart and used database for annotation,
we generated additional transcriptomes from lines grown in visualization, and integrated discovery (DAVID) to determine the
our laboratory and also took advantage of the numerous GO categories with high enrichment score.
data sets stored in the Gene Expression Omnibus (GEO) of We found that 67 GO categories were differentially expressed
the NCBI. The additional strains included stromal cell lines, in at least one HSPC-supportive in comparison to its corre-
AM14.1C4 and AM30.3F4 from AGM (Oostendorp et al., 2002), sponding less-supportive counterpart. Using these GO cate-
EL08.1D2 from E11 embryonic liver (Oostendorp et al., 2002; gories as variables and pairwise comparisons between lines
Ledran et al., 2008), 2012 and AFT011 from FL (Wineman as observations, we found that PC1 still corresponded to factor
et al., 1996), and 14F1.1 (Zipori et al., 1985), MS-5 (Issaad support (Figure 4A, score plot on the left). The first two compo-
et al., 1993), OP-9 (Nakano et al., 1994), and OP9M2 (a nents accounted for 58.5% of the variance. In the loading plot
subclone of OP9) (Magnusson et al., 2013) from BM. We also shown on the right in Figure 4A, variables are represented as
included primary stromal cells recently identified as critical com- vectors (red arrows). This indicated that many variables whose
ponents of the BM HSC niche; i.e., nestin-GFP+ (Méndez-Ferrer extremity lies close to the unit circle were positively correlated
et al., 2010), SCF-GFP+ (Ding et al., 2012), Cxcl12-GFP+ and to the factor support (vectors in the left-to-right orientation).
Pdgfra+/Sca1+ cells (Greenbaum et al., 2013), and osteoblasts To find which of these variables were the most relevant, we
sorted from Col2.3-GFP+ transgenic mice (Eash et al., 2010). compared for each category the means of DAVID enrichment
Finally, we included embryonic stem cells (Zhao et al., 2009), car- scores in HSPC-supportive versus less-supportive cell lines.
diomyocytes, and cardiac fibroblasts (Ieda et al., 2010) as nega- This comparison revealed that 14 categories were significantly
tive control data sets. Gene expression levels were compared (p % 0.05) upregulated in the supportive lines (Table S2). The
with three-way ANOVA with site of origin, support capacity, cellular components ECM and membrane (M); the biological pro-
and experimental series as factors. All samples were appropri- cesses cell adhesion (A), cell migration (Mig), and vessel devel-
ately normalized by removing the factor series to provide a opment (V); and the molecular functions secreted (S), signal
unique 66 3 16,530 matrix. Given that all stromal strains studied (Sig), heparin binding (H), growth factor binding (GFb), cytokine
had been characterized in terms of HSPC support, the factor binding, peptidase, gpi anchor, Egf-like domain, and immuno-
support has been used as a discriminating genetic trait in subse- globulin domain were all upregulated in HSPC-supportive cell
quent analyses. As shown in Figure 2D, the score plot (first two lines. We found the same categories for set 1 when comparing
components) of PCA (using the 48 samples not included in our the genes upregulated to the downregulated ones (Figure 4A
initial study as observations and the set 1 genes as variables) and Table S2). Remarkably, the same categories were found
indicates that HSPC-supportive samples (green circles) segre- when analyzing the list of genes belonging to the WGCNA
gate clearly from the nonsupportive ones (red triangles). Interest- modules positively correlated to the factor support (Figure 3C).
ingly, supportive AGM and embryonic or fetal liver samples are However, categories were not all represented in each module
distinct from supportive BM samples, which indicates persistent because of their small number gene contents. Some of the cat-
imprinting of the site of origin. PC1 alone accounted for 37% of egories were predominant in a given module (e.g., GFb in the

Cell Stem Cell 15, 376–391, September 4, 2014 ª2014 Elsevier Inc. 379
 Cell Stem Cell
Molecular Analysis of the HSC Niche

(legend on next page)

380 Cell Stem Cell 15, 376–391, September 4, 2014 ª2014 Elsevier Inc.
Cell Stem Cell
Molecular Analysis of the HSC Niche

blue module and H in the black module), reflecting their differ- of a given site (Table S4). BM-derived supportive cells harbored
ence in the gene makeup. a mesenchymal phenotype, expressing genes specific to adipo-
Then, using Ingenuity Systems and Genomatix, we analyzed cytes (such as fatty acid or triglyceride metabolic process
literature-based molecular networks characteristic of each sup- and regulation of lipase activity) and to osteoblasts (response
portive line in comparison to its less-supportive counterpart. to vitamin). In contrast, AGM-derived supportive cells ex-
Results from this analysis are shown in Figure 4B for the AGM pressed genes involved in hemostasis, vascular smooth muscle
HSPC-supportive line 1B6 and Figures S1A and S1B for the cell contraction, and nucleotide phosphodiesterase activity.
HSPC-supportive FL (AFT) and BM (B9) lines. A network consist- FL-derived supportive cells expressed genes related to the cell
ing of genes implicated in cell communication is apparent, cycle, such as cyclin, DNA repair, and DNA replication. PCA
including molecules involved in ECM synthesis (yellow) and with 66 GO categories as variables and pairwise comparisons
degradation (purple) and transcripts coding for soluble and/or between cell strains as observations summarize these data (Fig-
transmembrane factors (gray). To validate the differential ure S2). Notably, the orthogonal axes had to be rotated in order
expression of candidate genes identified in our microarray to disclose factors easy to interpret, canonical pathways being
analyses, we performed quantitative RT-PCR (qRT-PCR) with highly correlated to factor 1, whereas BM specific pathways
TaqMan Low-Density Arrays (TLDAs) for 43 transcripts coding were correlated to factor 2.
for cytokines, morphogens, receptors, cell adhesion molecules, Thus, this study unravels three factors accounting for gene
and ECM proteins (Table S3). The regression between mRNA expression. The site-specific factor, apparent already at the
expression values obtained by microarray analyses and qRT- entire transcriptome level, corresponds to tissue-imprinted
PCR was significant (p < 0.0001), and the results for cytokines genes. It is clearly disclosed by hierarchical clustering with the
and morphogens are shown in Figure S1C. Importantly, we initial set of 18 samples (see Figure 1B). It is confirmed when
confirmed the higher expression level of numerous genes de- using the whole set of 57 stromal samples of AGM, FL, and
tected in the molecular network characteristic of HSPC support. BM origin, as shown in Figure S3. The two other factors are
An example is shown for the supportive AGM cell line 1B6 (Fig- discrete, requiring a subtractive strategy (core support factor)
ure 4C). Altogether, these data indicate that the gene network or a combination of strain comparisons (site/support-specific
characteristic of the HSPC supportive capacity of stromal cells factor). The three factors are intricate as shown by PCA with
includes well-conserved biological pathways implicated in cell- the samples not included in our initial study as observations
to-cell and cell-to-matrix communication. However, although and set 1 genes as variables where AGM and FL samples still
the core network consists in identical pathways, pairwise anal- segregate from the BM ones (see Figure 2C).
ysis indicates that the precise gene makeup of each pathway
is specific to each hematopoietic site. Analysis of Molecular Pathways after Contact
with HSPCs
Analysis of Site-Specific Gene Signatures To evaluate whether gene pathways upregulated in HSPC-sup-
To address the question of site-specific molecular pathways, portive lines defined a stromal cell state, we compared, for proof
we took advantage of the large panel of stromal cells studied of concept, the most distinctive FL lines and analyzed the tran-
and used GSEA and two different settings. First, the expression scriptomes of AFT and BFC stromal cells cultured in the pres-
profiles of the HSPC-supportive AGM, FL, and BM stromal ence or absence of BM HSPCs. Given that these stromal lines
cells were independently compared to those of nonsupportive do not exhibit contact inhibition because of the activity of the
samples. Second, the expression profiles of supportive cells thermosensitive SV40 T antigen, we set up 4-day cocultures
from one tissue were compared to supportive cells from the with BM c-Kit+Lin!Sca1+ (KLS) cells in order to assess whether
two other tissues. We retained the data sets with high normalized short time spans are sufficient for the development of hemato-
enrichment scores and low FDRs (%0.05), selected the genes poietic colonies. Data shown in Figure S4 confirm that AFT and
belonging to the leading edge of each retained data set, and BFC stromal cells differentially support HSPCs.
analyzed, using DAVID, the GO categories corresponding to AFT and BFC cells, exposed to KLS cells for 4 days or not,
the summing list of genes. Consistent with our previous observa- were sorted by flow cytometry on the basis of their lack of
tion, AGM, FL, and BM supportive stromal cells expressed genes expression of the CD45 antigen (Figure S4). AFT and BFC
involved in the well-conserved pathways described above. On cells cultured with KLS cells are referred to as AFTKLS and
top of these ‘‘canonical’’ pathways dedicated to HSPC support, BFCKLS, respectively. Total RNA was extracted from CD45-
we observed that AGM, FL, and BM stromal cells exhibited negative sorted stromal cells and used for transcriptome anal-
specific gene signatures corresponding to additional GO cate- ysis. AFTKLS cells retained the gene expressions corresponding
gories significantly (p % 0.05) upregulated in supportive lines to GO categories upregulated in supportive lines, whereas these

Figure 3. Structure of the Gene Network Identifying Coordinately Expressed Genes in Set 1
(A) Major network concepts for the three subnetworks (modules) positively correlated to the factor support. Module significance is the mean of the correlations to
the factor support of the genes belonging to the module. Module density approximates the mean connectivity of the genes belonging to the module. Bubble color
corresponds to the arbitrary color of the module given by WGCNA.
(B) Organization of the three modules. The node size is proportional to the gene connectivity. Genes emphasized in the text or in other figures are in bold.
(C) Major GO categories given by DAVID for genes belonging to the three modules positively correlated to the factor support and, for comparison, for the
upregulated genes in set 1. S, secreted; ECM, extracellular matrix; M, membrane; H, heparin binding; A, cell adhesion; GFb, growth factor binding.
See also Table S1.

Cell Stem Cell 15, 376–391, September 4, 2014 ª2014 Elsevier Inc. 381
 Cell Stem Cell
Molecular Analysis of the HSC Niche

Figure 4. Molecular Pathways Representative of HSPC Supportive Activity

(A) PCA (PC2 versus PC1) with 67 GO categories used as variables and eight pairwise comparisons of the six stromal lines (1B6 versus 3B5, AFT versus BFC, and
B9 versus B10) and of the gene set 1 (genes up- versus downregulated) used as observations. The 67 categories were selected as differentially expressed in at
least one supportive line in comparison to less-supportive counterpart. Left, score plot. Right, loading plot (variables as red arrows). The unit circle is shown in red.
Most of the 14 categories significantly (p % 0.05 using two-tailed Student’s t test) upregulated in the supportive lines are indicated in bold. V, vessel development;
Sig, signal; Mig, migration. Other abbreviations are indicated in the legend of Figure 3C.
(B) Literature-based gene network upregulated in 1B6 cell line in comparison to 3B5. Gene categories: ECM synthesis (yellow nodes), ECM degradation (purple),
soluble/transmembrane factors (gray). Hub genes are of larger size. Green circles indicate genes present in mRNA Set 1.
(C) qRT-PCR analysis of individual genes. Genes encircled in red in (B) were analyzed by qRT-PCR with TLDAs. Values for AGM lines are shown as red bars. For
comparison, values for FL and BM lines are shown as blue bars. +, HSPC-supportive line; !, less-supportive line. Mean of two independent experiments. Scale
bars represent SD.
See also Figures S1 and S2 and Tables S2–S4.

382 Cell Stem Cell 15, 376–391, September 4, 2014 ª2014 Elsevier Inc.
Cell Stem Cell
Molecular Analysis of the HSC Niche

GO categories did not appear in BFCKLS cells (Table S5). More- PC1 coincided with the location of cell lines in the score plot
over, a few additional GO categories were upregulated in AFTKLS, (e.g., miR-155 and the three 3B5 samples on Figure 6C), which
including growth factor activity, hexose metabolic process, gave a prediction on the relevant role of these miRs in the corre-
phospholipid binding, ion transmembrane transport, and sphin- sponding lines. Using qRT-PCR, we have confirmed the differen-
golipid metabolic process. Using the whole set of GO categories tial expression of miR-9* that was the only miR upregulated in the
as variables and pairwise comparisons as observations, PC1 still three supportive lines (Figure S5A).
corresponded to the factor support (Figure 5A, score plot on the Transcripts of chemokine Cxcl12 and cytokine receptor antag-
left). The loading plot (Figure 5A, right) revealed that the category onist Il13ra1, significantly (p % 0.01%) upregulated in the sup-
growth factor activity was correlated to PC1 along with the other portive lines 1B6 and B9 (see Table S1), are validated targets
GO categories characteristic of supportive pathways. On the of miR-155 that negatively correlated to the factor support
contrary, the other additional categories appearing in AFTKLS (PC1 loading = !0.42). Therefore, one explanation for the lack
were correlated to PC2. of supportive activity by the 3B5 and B10 lines may be the
The literature-based gene network in AFTKLS cells was essen- upregulation of miR-155 leading to downregulation of its targets
tially the same as for AFT (Figure 5B), but with the inclusion of in these lines. Another set of data concerned mRNAs upregu-
additional nodes, in particular those connected with Vegfa (entire lated in less-supportive lines. Several of these, significantly
list in Table S6), which is in agreement with the additional growth (p % 0.01%) upregulated in 3B5 or BFC, were common targets
factor activity category (Figure 5A). Using GSEA with set 1 as a of miRs that positively correlated to the factor support. There-
reference gene set, we found that the global transcriptomes of fore, the corresponding miRs may constitute essential nodes
AFTKLS and BFCKLS correlated with the phenotype supportive in gene networks activated in supportive lines. Some of their tar-
versus nonsupportive, respectively (Figure 5C). These data indi- gets (Krt19, Cdh1, Foxa2, and Isl1) are markers for, or transcrip-
cate that the network of upregulated genes in the FL-supportive tion factors affiliated with, endodermal differentiation.
line exists prior contact with KLS cells but is extended after con- We used GSEA to analyze motifs present in 30 untranslated
tact. Such gene pattern is not found in the nonsupportive line, regions of up- or downregulated genes, which include the
even after contact, which indicates that the function support binding sequence of the seed regions of miRs (Figure S5B).
cannot be induced de novo after contact with KLS cells. Finally, We hypothesized that miRs whose mRNA targets were found
the data further validate the set 1 gene list as strongly represen- increased in HSPC-supportive lines would be operative in less-
tative of the HSPC-supportive function. supportive lines. Several miRs predicted by GSEA as active
in one of the less-supportive lines were found significantly
MicroRNA Expression in HSPC-Supportive Cell Lines (p < 0.06) increased in the line (Figure S5B).
To determine whether HSPC-supportive cell lines also ex- Altogether, these results strongly suggest that miRs may be
pressed specific miR signatures, we examined the relative critical cell-autonomous regulators of the stromal function
expression of more than 600 characterized mouse miRs in the affecting the expression of positive regulators in less-supportive
18 samples corresponding to our initial set of stromal cell lines. lines or that of negative regulators in supportive ones.
Similar to what we found for mRNAs, we could not discriminate
by PCA HSPC-supportive from less-supportive lines with entire Global View of Gene Expression with Self-Organizing
gene sets (Figure 6A). The first two components accounted for Map Analysis
31.4% of the variance. Then, the supervised analysis was carried To confirm our results with an independent method as well as to
out with the same strategy adopted for mRNA transcriptomes obtain more detailed insights into the transcriptional expression
but with the additional step of ignoring genes with significantly of HSPC-supportive cell lines, we performed self-organizing
divergent expression in two cell lines in comparison to the third. map (SOM) analysis. In essence, SOM analysis portrays the
Seventeen miRs (five of which were represented by two probes) individual expression landscape of each sample in terms of
were up- or downregulated in the HSPC-supportive cell lines. mosaic images of metagenes, each representing a minicluster
This list, designated as miR set 1, corresponded to approxi- of coregulated genes (Wirth et al., 2011). The portraits generated
mately 3% of entire studied miR transcriptome (Figure 6B and for each of the replicated samples were very similar, reflecting
Table S7). The lists of differentially expressed miRs between high homogeneity of gene expression within each cell line
the HSPC-supportive and less-supportive stromal lines and their (data not shown). Next, we generated the mean portraits of
integration with the lists of differentially expressed mRNAs are each cell line as maps where red and blue spots indicate meta
accessible at http://stemniche.snv.jussieu.fr/. On the score genes that are over- or underexpressed in each line in compari-
plot (first two components) of PCA with miR set 1 genes as vari- son to expression in all samples (Figure 6D, top). Then, to deter-
ables and 18 samples as observations, the segregation between mine the meta gene pattern characteristic of HSPC support,
the three stromal lines that provide a potent support for HSPCs difference portraits were obtained by subtracting the mean
(on the right) and the three that do not or less efficiently support image of supportive lines from that of less-supportive ones (Fig-
HSPCs (on the left) indicated that PC1 corresponds to the factor ure 6D, middle). Comparison of portraits indicated differential
support (Figure 6C, left). The first two components accounted for expression of metagenes in several regions (red dotted lines
65% of the variance. Analysis of PC1 loading plots showed that encircle upregulated metagenes, and blue dashed lines encircle
most miRs were positively correlated to the factor support (e.g., downregulated metagenes). Interestingly, the difference por-
miR-143, miR-214*, and miR-9*), whereas a few were negatively traits for the FL cell lines showed the most distinctive spot pat-
correlated (e.g., miR-155; Figure 6C, right). Interestingly, the terns, confirming the results of PCA (see Figure 2B). Moreover,
location of some miRs in the 2D loading plot of PC2 versus there was strong similarity of the difference maps of FL cell lines

Cell Stem Cell 15, 376–391, September 4, 2014 ª2014 Elsevier Inc. 383
 Cell Stem Cell
Molecular Analysis of the HSC Niche

(legend on next page)

384 Cell Stem Cell 15, 376–391, September 4, 2014 ª2014 Elsevier Inc.
Cell Stem Cell
Molecular Analysis of the HSC Niche

exposed or not to KLS, confirming that the molecular pathways for the HSPC marker cmyb and the T cell marker rag1. Tgfbi,
are not substantially modified after contact with HSPCs. That snai2, pax9, and ccdc80, but not wnt10b, morphants exhibited
most of the support-related metagenes formed nonoverlapping strong hematopoietic defects in the 36 hr postfertilization dorsal
entities was made apparent when genes from set 1 were aorta at a time when HSPCs emerge (Bertrand et al., 2010; Kissa
superimposed to the SOM images (Figure 6D, bottom). The and Herbomel, 2010) (Figure 7A). Importantly, the fli1:eGFP+
subsets of differentially up- or downregulated genes of set 1 vasculature was intact, indicating that the reduction in HSPCs
mostly accumulate in regions where metagenes are differentially in tgfbi, snai2, pax9, and ccdc80 morphants was not resulting
expressed (encircled areas). These data indicate consistency from vascular abnormalities (Figure S6A). To analyze HSCs, we
between results from independent analyses, and illustrate that utilized cmyb:eGFP; kdrl:memCherry animals, whereby double-
set 1 gene set covers the full spectrum of the gene expression positive cells in the dorsal aorta are definitive HSCs (Bertrand
landscape. Remarkably, most target genes of the miR set 1 et al., 2010). Using confocal microscopy, we were able to
accumulated in the region of metagene strong differential down- precisely quantify the numbers of HSCs in the dorsal aorta of
regulation, indicating their repressive effect on gene expression. KD animals (Figure S6B). Figure 7B shows significant (p < 0.05)
Finally, we included in the SOM analysis gene sets correspond- reduction of cmyb+/kdrl+ cells in tgfbi-, snai2-, pax9-, and
ing to GO categories already identified in this study (Figure S5C). ccdc80-deficient embryos in comparison to controls.
These sets mainly accumulated in the regions where metagenes To examine later stages of definitive hematopoiesis, we exam-
were upregulated, matching well with support-related patterns. ined HSPCs in the caudal hematopoietic tissue (CHT, equivalent
In summary, comprehensive SOM analysis identifies and dis- of mammalian FL) and intrathymic T cell development. HSPC
entangles regulatory modes of gene expression due to HSPC numbers were severely impaired in tgfbi and snai2 morphants,
support with a simple-to-interpret visualization frame. as demonstrated by respective decreases in cmyb and rag1
expression (Figure 7C, top, rows 1–3). A reduction in cmyb and
In Vivo Validation of Developmental HSPC Regulators rag1 expression was also observed in pax9 and ccdc80 mor-
in the Zebrafish Embryo phants (Figure 7C, bottom, rows 1–3). In the developing kidney,
To functionally validate our computational analysis, we per- which is the future site of adult hematopoiesis, snai2, pax9, and
formed in vivo knockdown (KD) of gene function using antisense ccdc80 morphants showed a decrease in cmyb expression,
morpholinos (MOs) in zebrafish embryos. The zebrafish is whereas no cmyb+ cells were detected in tgfbi and wnt10b mor-
an excellent system for rapidly and robustly testing candidate phants (Figure 7C).
gene function in this manner (McKinney-Freeman et al., 2012; In addition, to examine whether our gene set is also predictive
Tijssen et al., 2011). We selected Tgfbi, Snai2, Wnt10b, Pax9, for negative HSPC regulators, we selected Grem1 encoding a
and Ccdc80 genes for KD experiments on the basis of their BMP antagonist, because its expression is negatively correlated
relative expression levels in our transcriptome analyses (Table to the factor support with PC1 loading of !0.74 (Figure 2C and
S1). Tgfbi plays an important role in cell-collagen interactions Table S1). Gain-of-function for grem1 reduced the expression
that affect mesenchymal differentiation; Tgfbi was significantly of cmyb in the dorsal aorta in comparison to control embryos
(p < 0.004) upregulated in the three HSPC-supportive cell lines (Figure S6C). This observation is in agreement with findings in
with PC1 loading of 0.88. The zinc-finger protein Snai2 is a tran- the mouse showing an inhibition of HSPC activity and a reduc-
scription factor involved in epithelial to mesenchyme transition; tion in the numbers of highly enriched HSCs when AGM explants
Snai2 was significantly (p % 0.03) upregulated in 1B6 and B9 are cultured in the presence of recombinant gremlin proteins
cell lines with PC1 loading of 0.74. Wnt10b plays a role in the ho- (Boisset et al., 2013; Durand et al., 2007).
meostasis of stem cells, including HSCs and MSCs; Wnt10b was These data confirm that our systems biology approach for
significantly (p < 0.004) upregulated in AFT and B9 lines with PC1 identifying critical regulators of HSPC specification and survival
loading of 0.87. Pax9 and Ccdc80 are significantly upregulated in can be functionally assessed with the zebrafish model sys-
AFT and B9 with PC1 loadings of 0.72 and 0.89, respectively. tem. Importantly, these results identify tgfbi, snai2, pax9, and
Pax9 encodes a paired box transcription factor involved in the ccdc80 as regulators of HSC specification and confirm the role
development of the thymus, parathyroid glands, teeth, and skel- of wnt10b for HSC maintenance.
etal elements of skull and larynx. Ccdc80 encodes an extracel-
lular protein implicated in cell adhesion and matrix assembly. DISCUSSION
Embryos were injected at the one-cell-stage with MOs
directed against zebrafish homologs of these genes and subse- The primary objective of the present work has been to provide a
quently analyzed by whole-mount in situ hybridization (WISH) list of core genes operative in several sites of hematopoiesis and

Figure 5. Modulation of the Transcriptome of FL Stromal Lines after Contact with HSPCs
(A) PCA with 82 GO categories used as variables and four pairwise comparisons (AFT > BFC, AFTKLS > BFCKLS, AFT < BFC, and AFTKLS < BFCKLS) used as
observations. The 82 categories were selected as differentially expressed in at least one of the AFT lines (±KLS) in comparison to BFC (±KLS). Left, score plot.
Right, loading plot (variables as red arrows). Some of the categories upregulated in AFT ± BFC are indicated in bold. GF, growth factor activity; Ion, ion
transmembrane transport; Phos, phospholipid binding; Sph, Sphingolipid metabolic process.
(B) Literature-based gene network upregulated in AFTKLS versus BFCKLS. White nodes, genes already upregulated in AFT versus BFC; black nodes, additional
genes upregulated in AFTKLS versus BFCKLS. Green circles indicate genes present in mRNA Set 1.
(C) GSEA with set 1 as reference data set and AFTKLS and BFCKLS as expression data set.
See also Figure S4 and Tables S5 and S6.

Cell Stem Cell 15, 376–391, September 4, 2014 ª2014 Elsevier Inc. 385
 Cell Stem Cell
Molecular Analysis of the HSC Niche

(legend on next page)

386 Cell Stem Cell 15, 376–391, September 4, 2014 ª2014 Elsevier Inc.
Cell Stem Cell
Molecular Analysis of the HSC Niche

indicate how these genes relate one to another so that they can correlated to the trait support. Tgfbi, snai2, pax9, and ccdc80, but
collectively implement the stromal function of HSPC support. To not wnt10b, morphants displayed a prominent phenotype in the
achieve this goal, we have used a number of analytical tools. dorsal aorta. The critical implication of tgfbi and ccdc80 known
The final output of these analyses was a gene set of 481 mRNAs to regulate cell adhesion and matrix assembly suggests an
(corresponding to 2%–3% of the global transcriptome). important role of these processes in HSPC specification. The
This discrete gene set included most of the genes previously presence of binding sites for the transcription factor Pax9 both
identified as positive or negative HSPC regulators. The resulting in the human and mouse Kitl promoter might account for the
network included three modules positively correlated to the trait high correlation between the expressions of these two genes.
support, with many known regulators presenting as hubs or This result suggests that Pax9 may affect hematopoiesis through
fuzzy genes. On the basis of the present literature, the gene direct regulation of cytokine gene expression. Snai2 is also an
set specified conserved biological processes and molecular interesting candidate because it is known to play a major role in
functions implicated in cell communication, ECM remodeling epithelial-to-mesenchyme transition, generation and migration
and vessel development. However, additional pathways are of neural crest cells, and mesenchymal stem cell differentiation.
likely at play, given that the former did not account for the Collectively, our data indicate that transcription factors, as well
upregulation in the supportive lines of many intracellular mole- as ECM and signaling molecules, might cooperate as niche fac-
cules (e.g., Arnt2 and Bach2, see Table S1). Such molecules tors to control definitive hematopoiesis in the AGM.
might also be implicated in the supportive capacity being ex- Our study was primarily designed to identify the molecular core
ported in microvesicles or secreted and secondarily internalized. of the HSPC support. However, specific signatures preferentially
Altogether, the stromal gene set 1 essential for HSPC support active in the AGM, FL, or BM are expected given that previous
can be envisioned as a network of connected genes, a set of studies have reported that mesenchymal cells from different
molecular pathways or a landscape of metagenes. anatomical sites express distin ct transcriptional program (Chang
Such results had to be statistically validated with samples et al., 2002). Moreover, such signatures may also directly reflect
different from the initial set. This has been performed by verifying the differences between AGM, FL, and BM hematopoietic micro-
that set 1 had a predictive value, allowing correct categorization environments, given that the AGM is the site where HSPCs
of stromal cells with already characterized supportive capacity emerge in the embryo, the FL the site where they are amplified,
and of cells originating from nonhematopoietic sites. Classifica- and the BM the tissue where HSPCs are maintained throughout
tions according to PCA and GSEA were completely concordant, life. Indeed, our analysis revealed that BM-supportive cells
allowing clear-cut discrimination of HSPC-supportive from non- exhibited a mesenchymal phenotype, whereas AGM-supportive
supportive cells. cells more specifically expressed genes implicated in blood
Biological validation of our in silico results has been two- vessel functions. Interestingly, FL-supportive cells expressed
pronged. First, we investigated how contact with HSPCs would genes related to the cell cycle, suggesting that not only HSCs
modulate the gene set. The pairwise comparison of the FL lines but also their niches may actively proliferate in the FL.
before and after exposure to KLS cells indicated that the genes The expression of the core and associated networks in the sup-
upregulated in the AFT-supportive line before contact were still portive lines results from underlying regulatory structures such as
upregulated after contact, although they were still not induced chromatin organization (de Wit et al., 2013) or networks of non-
in the nonsupportive BFC line. Moreover, in the supportive cell coding RNAs (Djebali et al., 2012). Therefore, we finalized this
line exposed to HSPCs, many genes were specifically induced, work by investigating miRs differentially expressed in the stromal
the most relevant being those implicated in angiogenesis and cell lines. The subtractive strategy yielded a short list of 17 miRs
cytokine activity. These results confirm previous studies indi- that segregated HSPC-supportive from less-supportive stromal
cating that MSCs and vascular cells play a critical role in the lines. Comparison of miR expression in a given HSPC-supportive
regulation of HSCs (Ding et al., 2012; Greenbaum et al., 2013; line and expression of its validated mRNA targets in its less-sup-
Kiel et al., 2005; Kunisaki et al., 2013; Méndez-Ferrer et al., portive counterpart, and the converse, suggested that miRs may
2010). Our second approach for biological validation has been exert their suppressive effect not only by decreasing the expres-
loss of function of some of the genes of the gene set. Due sion of positive hematopoietic regulators but also by inducing
to the conservation of major hematopoietic regulations adverse priming to nonmesodermal lineages.
throughout vertebrates, zebrafish is a useful model to study the In conclusion, our work not only validates previous findings on
role of molecules involved in hematopoiesis (McKinney-Freeman critical hematopoietic regulators but also provides insights into
et al., 2012; Tijssen et al., 2011). We selected five genes positively unexpected ones such as Tgfbi, Pax9, and Ccdc80. Among

Figure 6. MicroRNA Transcriptome and SOM Analyses

(A) PCA with entire expression data set as variables and basic set of six cell lines as observations. PC2 versus PC1 score plot.
(B) Venn diagram of genes obtained by supervised analysis. miR set 1 is shaded.
(C) PCA (PC2 versus PC1) with miR set 1 as variables and basic set of cell lines as observations. Left, score plot. Right, loading plot.
(D) SOM. Top, mean portraits of the different cell lines obtained by averaging the respective individual portraits. Middle, difference portraits obtained by sub-
tracting mean portraits of less-supportive cell lines from that of their HSPC-supportive counterparts. Red and blue dotted circles indicate regions containing
differentially up- and downregulated genes, respectively. Difference in scale for FL and FL + KLS (FLKLS) lines in comparison to AGM, and BM lines indicates
larger differences in the FL cell lines (areas encircled by dashed circles) than in AGM and BM cell lines (areas encircled by dotted circles). Bottom, projection of
mRNA set 1 genes and of target genes of miR Set 1 onto SOM images (each dot refers to at least one individual gene). Dotted circles indicate highly populated
regions.
See also Figure S5 and Table S7.

Cell Stem Cell 15, 376–391, September 4, 2014 ª2014 Elsevier Inc. 387
 Cell Stem Cell
Molecular Analysis of the HSC Niche

(legend on next page)

388 Cell Stem Cell 15, 376–391, September 4, 2014 ª2014 Elsevier Inc.
Cell Stem Cell
Molecular Analysis of the HSC Niche

the list of 481 genes, many have unknown functions as HSPC Zebrafish Studies
niche factors. To our knowledge, these data represent a Zebrafish were maintained according to University of California, San Diego,
International Animal Care and Use Committee guidelines. Embryos were
comprehensive list of genes involved in the support of HSPCs
collected, staged, fixed, and processed for in situ hybridization as previously
by stromal cell lines, therefore offering a unique resource for described (Thisse et al., 1993). Wild-type or transgenic zygotes were injected
research on the stromal regulation of hematopoiesis. From a with a morpholino solution and incubated at 28.5" C until they reached the
clinical standpoint, some of the genes unraveled in this work stage of interest. Tg(cmyb:eGFP) animals (North et al., 2007) were crossed
might prove invaluable for better definition of cell therapy proto- to Tg(kdrl:HsHRAS-mCherry)s896 animals (referred to as kdrl:memCherry
cols (ex vivo amplification of HSCs and generation of de novo for clarity) (Chi et al., 2008). Additional details of protocols and analyses are
given in the Supplemental Experimental Procedures.
HSCs from pluripotent stem cells) and as optimal targets for
pharmaceutical drugs for the treatment of hematologic diseases
and primarily leukemias. ACCESSION NUMBERS

The NCBI GEO accession number for the microarray data presented in this
EXPERIMENTAL PROCEDURES
paper is GSE44181.

Microarray Screening and Data Analysis

In this study, we used Affymetrix Mouse Gene 1.0 ST arrays and Agilent SUPPLEMENTAL INFORMATION
miRNA 8X15K arrays. Total RNA was extracted from confluent stromal
cultures or sorted stromal cells cultured with or without KLS cells with Trizol Supplemental Information contains Supplemental Experimental Procedures,
reagent (Invitrogen). RNA concentration and integrity were evaluated with the six figures, and eight tables and can be found with this article online at
Agilent Bioanalyzer 2100 (Genomic Facility Platform, Cochin Institute). To http://dx.doi.org/10.1016/j.stem.2014.06.005.
correct from probe set definition inaccuracy, we used the version 17 of the
custom ChIP definition file (Dai et al., 2005). This file eliminated probes
AUTHOR CONTRIBUTIONS
with multiple matching sequences. Unsupervised analyses were performed
on the global transcriptome, and data were represented with hierarchical
P.C., T.J., and C.D. designed research, performed experiments, analyzed
clustering, PCA, and SOMs. The global expression profile of the stromal lines
data, and wrote the paper. C.P. designed research, performed experiments,
was also analyzed with GSEA with the functional (curated) and motif data
and analyzed data. D.T. designed research, analyzed data and wrote the pa-
sets present in the Molecular Signature Database of the Broad Institute
per. G.S. and P.L. performed experiments and analyzed data. H.B., F.D.,
(www.broadinstitute.org/gsea) (Subramanian et al., 2005). In parallel, we
F.A., F.L., and H.W. analyzed data. C.M., J.R., and B.U. performed experi-
performed supervised analyses to identify genes specifically up- or down-
ments. F.P. and E.D. provided materials.
regulated in HSPC-supportive stromal cells. Then, the lists of differentially
expressed genes (from the supervised and the GSEA analyses) were
analyzed for GO with DAVID (http://david.abcc.ncifcrf.gov) and to uncover ACKNOWLEDGMENTS
literature-based molecular pathways with Ingenuity Systems (www.ingenuity.
com) and Genomatix (www.genomatix.de) databases and conversion with We are very grateful to Catherine Robin (Hubrecht Institute) for critical
Cytoscape software (http://www.cytoscape.org). The expression data were and constructive comments on this study. We thank David Stachura (Depart-
also analyzed with SOM as described previously (Wirth et al., 2011). To ment of Cell and Developmental Biology, UCSD) for critical reading of
find out the network structure of the gene set 1, we used WGCNA (Horvath, the manuscript. We thank Nicole Boggetto (Institut Jacques Monod,
2011). Adjacencies were given according to signed Pearson correlation. The ImagoSeine Bioimaging Core Facility) for cell sorting experiments. We also thank
soft threshold power b that resulted in approximate scale-free topology was Sophie Gournet (UMR CNRS 7622) for her excellent photographic and drawing
36. Modules were constructed with average linkage hierarchical clustering. assistance. These studies were supported by grants from the Fondation pour
Minimum module size was 30. For better readability, the connectivity la Recherche Médicale (DEQ20100318258) and Agence Nationale pour la
threshold was 0.1. Additional details are given in the Supplemental Experi- Recherche/California Institute for Regenerative Medicine (ANR/CIRM 0001-02).
mental Procedures.
Received: April 4, 2013
Hematopoietic Assays and Purification of Stromal Cells after Revised: April 4, 2014
Coculture Accepted: June 6, 2014
BM was obtained from adult C57BL/6 female mice (3–10 months of age). Mice Published: July 17, 2014
were bred at Janvier (Le Genest) and maintained in the animal facility of the
Laboratory of Developmental Biology (University Pierre and Marie Curie, REFERENCES
UMR7622, CNRS) according to institutional guidelines. Adult BM hematopoi-
etic cells enriched in stem cell activity were purified by fluorescence-activated Bertrand, J.Y., Cisson, J.L., Stachura, D.L., and Traver, D. (2010). Notch
cell sorting on the basis of the KLS cell-surface phenotype. Details are pro- signaling distinguishes 2 waves of definitive hematopoiesis in the zebrafish
vided in the Supplemental Experimental Procedures. embryo. Blood 115, 2777–2783.

Figure 7. Functional Validation of Developmental HSPC Regulators in Zebrafish Embryos

(A) WISH for cmyb expression in 36 hr postfertilization morphants embryos and their corresponding uninjected controls. Numbers of embryos with or without (out
of total) cmyb staining are indicated in the top right corner.
(B) Numbers of cmyb+/kdrl+ cells in ventral part of dorsal aorta in double cmyb:eGFP; kdrl:memCherry transgenic embryos. Data represent mean ± SD (n = 6–19
embryos). *p < 0.05 with two-tailed Student’s t test.
(C) WISH for cmyb and rag1 expression in uninjected embryos or embryos injected with morpholinos directed against tgfbi, snai2, and wnt10b (top) or directed
against pax9 and ccdc80 (bottom). Row 1, cmyb expression in CHT in 4 days postfertilization (dpf) embryos. Row 2, magnification of CHT region. Row 3, rag1
expression in 4 dpf embryos. Arrowheads indicate bilateral thymii. Row 4, cmyb expression in 4 dpf embryos. Dotted circles indicate pronephros. Numbers of
embryos with or without cmyb or rag1 staining are indicated in the top right corner.
See also Figure S6 and Table S8.

Cell Stem Cell 15, 376–391, September 4, 2014 ª2014 Elsevier Inc. 389
 Cell Stem Cell
Molecular Analysis of the HSC Niche

Boisset, J.C., Clapes, T., Van Der Linden, R., Dzierzak, E., and Robin, C. progenitor cells and reveal endothelial niches for stem cells. Cell 121,
(2013). Integrin aIIb (CD41) plays a role in the maintenance of hematopoietic 1109–1121.
stem cell activity in the mouse embryonic aorta. Biol. Open 2, 525–532. Kissa, K., and Herbomel, P. (2010). Blood stem cells emerge from aortic
Calvi, L.M., Adams, G.B., Weibrecht, K.W., Weber, J.M., Olson, D.P., Knight, endothelium by a novel type of cell transition. Nature 464, 112–115.
M.C., Martin, R.P., Schipani, E., Divieti, P., Bringhurst, F.R., et al. (2003). Kunisaki, Y., Bruns, I., Scheiermann, C., Ahmed, J., Pinho, S., Zhang, D.,
Osteoblastic cells regulate the haematopoietic stem cell niche. Nature 425, Mizoguchi, T., Wei, Q., Lucas, D., Ito, K., et al. (2013). Arteriolar niches maintain
841–846. haematopoietic stem cell quiescence. Nature 502, 637–643.
Chang, H.Y., Chi, J.T., Dudoit, S., Bondre, C., van de Rijn, M., Botstein, D., and
Langfelder, P., and Horvath, S. (2008). WGCNA: an R package for weighted
Brown, P.O. (2002). Diversity, topographic differentiation, and positional
correlation network analysis. BMC Bioinformatics 9, 559.
memory in human fibroblasts. Proc. Natl. Acad. Sci. USA 99, 12877–12882.
Ledran, M.H., Krassowska, A., Armstrong, L., Dimmick, I., Renström, J.,
Chateauvieux, S., Ichanté, J.L., Delorme, B., Frouin, V., Piétu, G., Langonné,
Lang, R., Yung, S., Santibanez-Coref, M., Dzierzak, E., Stojkovic, M.,
A., Gallay, N., Sensebé, L., Martin, M.T., Moore, K.A., and Charbord, P.
et al. (2008). Efficient hematopoietic differentiation of human embryonic
(2007). Molecular profile of mouse stromal mesenchymal stem cells. Physiol.
stem cells on stromal cells derived from hematopoietic niches. Cell Stem
Genomics 29, 128–138.
Cell 3, 85–98.
Chi, N.C., Shaw, R.M., De Val, S., Kang, G., Jan, L.Y., Black, B.L., and Stainier,
Magnusson, M., Sierra, M.I., Sasidharan, R., Prashad, S.L., Romero, M.,
D.Y. (2008). Foxn4 directly regulates tbx2b expression and atrioventricular ca-
Saarikoski, P., Van Handel, B., Huang, A., Li, X., and Mikkola, H.K. (2013).
nal formation. Genes Dev. 22, 734–739.
Expansion on stromal cells preserves the undifferentiated state of human
Corselli, M., Chin, C.J., Parekh, C., Sahaghian, A., Wang, W., Ge, S., hematopoietic stem cells despite compromised reconstitution ability. PLoS
Evseenko, D., Wang, X., Montelatici, E., Lazzari, L., et al. (2013). ONE 8, e53912.
Perivascular support of human hematopoietic stem/progenitor cells. Blood
121, 2891–2901. McKinney-Freeman, S., Cahan, P., Li, H., Lacadie, S.A., Huang, H.T., Curran,
M., Loewer, S., Naveiras, O., Kathrein, K.L., Konantz, M., et al. (2012). The
Dai, M., Wang, P., Boyd, A.D., Kostov, G., Athey, B., Jones, E.G., Bunney,
transcriptional landscape of hematopoietic stem cell ontogeny. Cell Stem
W.E., Myers, R.M., Speed, T.P., Akil, H., et al. (2005). Evolving gene/transcript
Cell 11, 701–714.
definitions significantly alter the interpretation of GeneChip data. Nucleic Acids
Res. 33, e175. Méndez-Ferrer, S., Michurina, T.V., Ferraro, F., Mazloom, A.R., Macarthur,
B.D., Lira, S.A., Scadden, D.T., Ma’ayan, A., Enikolopov, G.N., and Frenette,
de Wit, E., Bouwman, B.A., Zhu, Y., Klous, P., Splinter, E., Verstegen, M.J.,
P.S. (2010). Mesenchymal and haematopoietic stem cells form a unique
Krijger, P.H., Festuccia, N., Nora, E.P., Welling, M., et al. (2013). The pluripo-
bone marrow niche. Nature 466, 829–834.
tent genome in three dimensions is shaped around pluripotency factors.
Nature 501, 227–231. Mirshekar-Syahkal, B., Fitch, S.R., and Ottersbach, K. (2014). From green-
house to garden: The changing soil of the hematopoietic stem cell microenvi-
Ding, L., Saunders, T.L., Enikolopov, G., and Morrison, S.J. (2012). Endothelial
ronment during development. Stem Cells 32, 1691–700.
and perivascular cells maintain haematopoietic stem cells. Nature 481,
457–462. Moore, K.A., Ema, H., and Lemischka, I.R. (1997). In vitro maintenance of
highly purified, transplantable hematopoietic stem cells. Blood 89, 4337–4347.
Djebali, S., Davis, C.A., Merkel, A., Dobin, A., Lassmann, T., Mortazavi, A.,
Tanzer, A., Lagarde, J., Lin, W., Schlesinger, F., et al. (2012). Landscape of Morrison, S.J., and Scadden, D.T. (2014). The bone marrow niche for haema-
transcription in human cells. Nature 489, 101–108. topoietic stem cells. Nature 505, 327–334.
Durand, C., Robin, C., and Dzierzak, E. (2006). Mesenchymal lineage Morrison, S.J., and Spradling, A.C. (2008). Stem cells and niches: mechanisms
potentials of aorta-gonad-mesonephros stromal clones. Haematologica 91, that promote stem cell maintenance throughout life. Cell 132, 598–611.
1172–1179. Nakano, T., Kodama, H., and Honjo, T. (1994). Generation of lymphohema-
Durand, C., Robin, C., Bollerot, K., Baron, M.H., Ottersbach, K., and Dzierzak, topoietic cells from embryonic stem cells in culture. Science 265, 1098–
E. (2007). Embryonic stromal clones reveal developmental regulators of defin- 1101.
itive hematopoietic stem cells. Proc. Natl. Acad. Sci. USA 104, 20838–20843. North, T.E., Goessling, W., Walkley, C.R., Lengerke, C., Kopani, K.R., Lord,
Dzierzak, E., and Speck, N.A. (2008). Of lineage and legacy: the development A.M., Weber, G.J., Bowman, T.V., Jang, I.H., Grosser, T., et al. (2007).
of mammalian hematopoietic stem cells. Nat. Immunol. 9, 129–136. Prostaglandin E2 regulates vertebrate haematopoietic stem cell homeostasis.
Eash, K.J., Greenbaum, A.M., Gopalan, P.K., and Link, D.C. (2010). CXCR2 Nature 447, 1007–1011.
and CXCR4 antagonistically regulate neutrophil trafficking from murine bone Oostendorp, R.A., Harvey, K.N., Kusadasi, N., de Bruijn, M.F., Saris, C.,
marrow. J. Clin. Invest. 120, 2423–2431. Ploemacher, R.E., Medvinsky, A.L., and Dzierzak, E.A. (2002). Stromal cell
Greenbaum, A., Hsu, Y.M., Day, R.B., Schuettpelz, L.G., Christopher, M.J., lines from mouse aorta-gonads-mesonephros subregions are potent sup-
Borgerding, J.N., Nagasawa, T., and Link, D.C. (2013). CXCL12 in early porters of hematopoietic stem cell activity. Blood 99, 1183–1189.
mesenchymal progenitors is required for haematopoietic stem-cell mainte- Renström, J., Istvanffy, R., Gauthier, K., Shimono, A., Mages, J., Jardon-
nance. Nature 495, 227–230. Alvarez, A., Kröger, M., Schiemann, M., Busch, D.H., Esposito, I., et al.
Hackney, J.A., Charbord, P., Brunk, B.P., Stoeckert, C.J., Lemischka, I.R., and (2009). Secreted frizzled-related protein 1 extrinsically regulates cycling
Moore, K.A. (2002). A molecular profile of a hematopoietic stem cell niche. activity and maintenance of hematopoietic stem cells. Cell Stem Cell 5,
Proc. Natl. Acad. Sci. USA 99, 13061–13066. 157–167.
Horvath, S. (2011). Weighted Network Analysis. Applications in Genomics and Subramanian, A., Tamayo, P., Mootha, V.K., Mukherjee, S., Ebert, B.L.,
systems biology. (New York: Springer). Gillette, M.A., Paulovich, A., Pomeroy, S.L., Golub, T.R., Lander, E.S., and
Ieda, M., Fu, J.D., Delgado-Olguin, P., Vedantham, V., Hayashi, Y., Bruneau, Mesirov, J.P. (2005). Gene set enrichment analysis: a knowledge-based
B.G., and Srivastava, D. (2010). Direct reprogramming of fibroblasts into approach for interpreting genome-wide expression profiles. Proc. Natl.
functional cardiomyocytes by defined factors. Cell 142, 375–386. Acad. Sci. USA 102, 15545–15550.
Issaad, C., Croisille, L., Katz, A., Vainchenker, W., and Coulombel, L. (1993). A Thisse, C., Thisse, B., Schilling, T.F., and Postlethwait, J.H. (1993). Structure of
murine stromal cell line allows the proliferation of very primitive human the zebrafish snail1 gene and its expression in wild-type, spadetail and no tail
CD34++/CD38- progenitor cells in long-term cultures and semisolid assays. mutant embryos. Development 119, 1203–1215.
Blood 81, 2916–2924. Tijssen, M.R., Cvejic, A., Joshi, A., Hannah, R.L., Ferreira, R., Forrai, A.,
Kiel, M.J., Yilmaz, O.H., Iwashita, T., Yilmaz, O.H., Terhorst, C., and Morrison, Bellissimo, D.C., Oram, S.H., Smethurst, P.A., Wilson, N.K., et al. (2011).
S.J. (2005). SLAM family receptors distinguish hematopoietic stem and Genome-wide analysis of simultaneous GATA1/2, RUNX1, FLI1, and SCL

390 Cell Stem Cell 15, 376–391, September 4, 2014 ª2014 Elsevier Inc.
Cell Stem Cell
Molecular Analysis of the HSC Niche

binding in megakaryocytes identifies hematopoietic regulators. Dev. Cell 20, Zhang, J., Niu, C., Ye, L., Huang, H., He, X., Tong, W.G., Ross, J., Haug, J.,
597–609. Johnson, T., Feng, J.Q., et al. (2003). Identification of the haematopoietic
Wagers, A.J. (2012). The stem cell niche in regenerative medicine. Cell Stem stem cell niche and control of the niche size. Nature 425, 836–841.
Cell 10, 362–369. Zhao, X.Y., Li, W., Lv, Z., Liu, L., Tong, M., Hai, T., Hao, J., Guo, C.L., Ma, Q.W.,
Wineman, J., Moore, K., Lemischka, I., and Müller-Sieburg, C. (1996). Wang, L., et al. (2009). iPS cells produce viable mice through tetraploid
Functional heterogeneity of the hematopoietic microenvironment: rare stromal complementation. Nature 461, 86–90.
elements maintain long-term repopulating stem cells. Blood 87, 4082–4090. Zipori, D., Duksin, D., Tamir, M., Argaman, A., Toledo, J., and Malik, Z. (1985).
Wirth, H., Löffler, M., von Bergen, M., and Binder, H. (2011). Expression Cultured mouse marrow stromal cell lines. II. Distinct subtypes differing in
cartography of human tissues using self organizing maps. BMC morphology, collagen types, myelopoietic factors, and leukemic cell growth
Bioinformatics 12, 306. modulating activities. J. Cell. Physiol. 122, 81–90.

Cell Stem Cell 15, 376–391, September 4, 2014 ª2014 Elsevier Inc. 391