Gene function in early mouse embryonic stem cell differentiation.

Abstract

Little is known about the genes that drive embryonic stem cell differentiation. However, such knowledge is necessary if we are to exploit the therapeutic potential of stem cells. To uncover the genetic determinants of mouse embryonic stem cell (mESC) differentiation, we have generated and analyzed 11-point time-series of DNA microarray data for three biologically equivalent but genetically distinct mESC lines (R1, J1, and V6.5) undergoing undirected differentiation into embryoid bodies (EBs) over a period of two weeks.
We identified the initial 12 hour period as reflecting the early stages of mESC differentiation and studied probe sets showing consistent changes of gene expression in that period. Gene function analysis indicated significant up-regulation of genes related to regulation of transcription and mRNA splicing, and down-regulation of genes related to intracellular signaling. Phylogenetic analysis indicated that the genes showing the largest expression changes were more likely to have originated in metazoans. The probe sets with the most consistent gene changes in the three cell lines represented 24 down-regulated and 12 up-regulated genes, all with closely related human homologues. Whereas some of these genes are known to be involved in embryonic developmental processes (e.g. Klf4, Otx2, Smn1, Socs3, Tagln, Tdgf1), our analysis points to others (such as transcription factor Phf21a, extracellular matrix related Lama1 and Cyr61, or endoplasmic reticulum related Sc4mol and Scd2) that have not been previously related to mESC function. The majority of identified functions were related to transcriptional regulation, intracellular signaling, and cytoskeleton. Genes involved in other cellular functions important in ESC differentiation such as chromatin remodeling and transmembrane receptors were not observed in this set.
Our analysis profiles for the first time gene expression at a very early stage of mESC differentiation, and identifies a functional and phylogenetic signature for the genes involved. The data generated constitute a valuable resource for further studies. All DNA microarray data used in this study are available in the StemBase database of stem cell gene expression data 1 and in the NCBI's GEO database.

Full text

Page 1

BioMed Central
Page 1 of 21
(page number not for citation purposes)
BMC Genomics
Open Access
Research article
Gene function in early mouse embryonic stem cell differentiation
Kagnew Hailesellasse Sene, Christopher J Porter, Gareth Palidwor,
Carolina Perez-Iratxeta, Enrique M Muro, Pearl A Campbell,
Michael A Rudnicki and Miguel A Andrade-Navarro*
Address: Ontario Genomics Innovation Centre, Ottawa Health Research Institute, 501 Smyth, Ottawa, ON, K1H 8L6, Canada
Email: Kagnew Hailesellasse Sene - kagnewab@yahoo.com; Christopher J Porter - cporter@ohri.ca; Gareth Palidwor - gpalidwor@ohri.ca;
Carolina Perez-Iratxeta - cperez-iratxeta@ohri.ca; Enrique M Muro - emuro@ohri.ca; Pearl A Campbell - pcampbell@ohri.ca;
Michael A Rudnicki - mrudnicki@ohri.ca; Miguel A Andrade-Navarro* - mandrade@ohri.ca
* Corresponding author
Abstract
Background: Little is known about the genes that drive embryonic stem cell differentiation.
However, such knowledge is necessary if we are to exploit the therapeutic potential of stem cells.
To uncover the genetic determinants of mouse embryonic stem cell (mESC) differentiation, we
have generated and analyzed 11-point time-series of DNA microarray data for three biologically
equivalent but genetically distinct mESC lines (R1, J1, and V6.5) undergoing undirected
differentiation into embryoid bodies (EBs) over a period of two weeks.
Results: We identified the initial 12 hour period as reflecting the early stages of mESC
differentiation and studied probe sets showing consistent changes of gene expression in that
period. Gene function analysis indicated significant up-regulation of genes related to regulation of
transcription and mRNA splicing, and down-regulation of genes related to intracellular signaling.
Phylogenetic analysis indicated that the genes showing the largest expression changes were more
likely to have originated in metazoans. The probe sets with the most consistent gene changes in
the three cell lines represented 24 down-regulated and 12 up-regulated genes, all with closely
related human homologues. Whereas some of these genes are known to be involved in embryonic
developmental processes (e.g. Klf4, Otx2, Smn1, Socs3, Tagln, Tdgf1), our analysis points to others
(such as transcription factor Phf21a, extracellular matrix related Lama1 and Cyr61, or endoplasmic
reticulum related Sc4mol and Scd2) that have not been previously related to mESC function. The
majority of identified functions were related to transcriptional regulation, intracellular signaling, and
cytoskeleton. Genes involved in other cellular functions important in ESC differentiation such as
chromatin remodeling and transmembrane receptors were not observed in this set.
Conclusion: Our analysis profiles for the first time gene expression at a very early stage of mESC
differentiation, and identifies a functional and phylogenetic signature for the genes involved. The
data generated constitute a valuable resource for further studies. All DNA microarray data used in
this study are available in the StemBase database of stem cell gene expression data [1] and in the
NCBI's GEO database.
Published: 29 March 2007
BMC Genomics 2007, 8:85doi:10.1186/1471-2164-8-85
Received: 18 August 2006
Accepted: 29 March 2007
This article is available from: http://www.biomedcentral.com/1471-2164/8/85
© 2007 Hailesellasse Sene et al; licensee BioMed Central Ltd.
This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0),
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Page 2

BMC Genomics 2007, 8:85 http://www.biomedcentral.com/1471-2164/8/85
Page 2 of 21
(page number not for citation purposes)
1. Background
There is a growing interest in the identification of genes
responsible for stem cell phenotypes and behaviour, due
largely to the envisioned potential of the field of stem cell
therapeutics. However, relatively few genes have been
identified as associated with stem cell generation or main-
tenance [2]. Such genes are likely to be important in deriv-
ing methods to control and direct the differentiation of
stem cells for therapy, for example by expanding a stem
cell culture or by directing differentiation towards a tar-
geted cell type. The knowledge of the genes and processes
involved in stem cell control is particularly sparse for
embryonic stem cells (ESCs), though ESCs are the most
interesting for therapy as they have the potential to differ-
entiate into all cell types.
DNA microarrays allow the simultaneous profile of thou-
sands of mRNA transcripts, thereby providing a general
overview of the state of gene expression in a cell sample.
Here, we have used DNA microarrays to identify genes,
and thus cellular processes, that might be important to the
differentiation of mouse ESCs (mESCs). We have profiled
samples from three biologically equivalent but genetically
distinct mESC lines (J1, R1, and V6.5) undergoing in vitro
differentiation into embryoid bodies (EBs). Upon
removal of LIF (leukemia inhibitory factor), and in the
absence of murine embryonic feeder cells, mESCs can be
grown as unattached spheres termed EBs, which contain
mesodermal, ectodermal, and endodermal cells [3]. EB
formation from mESCs has been proposed as a model for
early embryonic development in terms of differentiation
capacity, morphological changes, and inductive signaling
events that drive these changes [4,5].
Measurements of gene expression were taken from biolog-
ical triplicates during differentiation at 11 time points
spanning a period of two weeks following LIF removal.
Measures were taken every six hours for the first 24 hours,
then at increasing intervals, with time points at: 0 h
(undifferentiated mESCs), 6 h, 12 h, 18 h, 24 h, 36 h, 48
h, 4 d, 7 d, 9 d, and 14 d. The cells' mRNA content was
assayed at each time point using the Affymetrix
MOE430A/B GeneChip set. Our hypothesis is that genes
important for the control of the differentiation process
should show large expression changes during the initial
period of differentiation.
We found that during the initial 12 hour period the gene
expression of transcription factors is still relatively
unchanged respect to the undifferentiated mESCs in all
three time courses. We conclude that this period repre-
sents cells not yet differentiated, and is, therefore, appro-
priate for the study of the early changes in gene expression
leading to differentiation.
To confirm that the gene expression data of the first 12
hours of the time series pinpoints to genes related to
mESC differentiation, we performed a heuristic method
for the analysis of the DNA microarray data that yielded
the genes with the largest changes in expression in that
period in two or three of the cell lines analyzed.
We classified the changes in gene expression in that period
with respect to gene function and phylogenetic profile of
the corresponding protein products. Our results showed a
significant abundance of proteins related to transcription
regulation, intracellular signaling, and the cytoskeleton;
and enrichment in protein sequences whose phylogenetic
distribution of homologs indicated that their evolution-
ary origin occurred along the metazoan lineage.
Finally, we applied a more standard method of DNA
microarray data analysis (Significance Analysis of Micro-
arrays; SAM), to select 36 genes with the most consistent
expression changes during the initial 12 hours of the time
series in two or three of the cell lines analyzed. We discuss
the functions and possible relations of those 36 genes
with mESC differentiation, which includes genes known
to be involved in embryonic differentiation processes but
also several never before linked to those processes.
Our study differs from previous genomic scale gene
expression surveys of human and mouse ESC differentia-
tion [6,7] in that it examines in parallel the differentiation
of three cell lines using a greater number of time points,
starting at a much earlier point after the initiation of dif-
ferentiation. We therefore expect that our findings expand
upon the information generated in those studies.
2. Results
2.1. Demarcation of an early period of differentiation
In order to identify early pre-differentiation changes in
mESC gene expression, we defined the period for each of
the three time series in which cell populations were still
predominantly undifferentiated ESCs. We reasoned that
the genes involved in triggering differentiation would be
among those whose expression changed the most during
that period. We performed two analyses to define this
early period.
Firstly, we performed two-dimensional hierarchical clus-
tering of the signal values of 1,605 probe sets correspond-
ing to 1,327 known or predicted transcription factors
against the eleven time points for each cell line (Figure 1).
The reasoning behind this was to identify for each cell line
the period during which changes in the genetic regulatory
network were yet relatively small. We identified the time
points that clustered closely with the first (0 h) time
points, when the culture consists primarily of ESCs, and
classified these as the early period. In the J1 and V6.5

Page 3

BMC Genomics 2007, 8:85http://www.biomedcentral.com/1471-2164/8/85
Page 3 of 21
(page number not for citation purposes)
lines, the expression patterns at 0, 6 and 12 hours clus-
tered together; in R1, the 0 h pattern formed a cluster with
the 6, 12, 18, and 24 hour patterns.
Secondly, in order to define the parts of the three time
series corresponding mainly to an ESC population, we
examined the signal values of probe sets corresponding to
known markers of ESCs. Out of ten marker genes exam-
ined, only three (Oct4/Pou5f1 [8], Cripto/Tdgf1 [9], and
Rex1 [10]) had high initial values of expression which
decreased gradually over time in all three cell lines. We
observed that the expression of these three marker genes
stayed between 50%-100% of their 0 h value at 6 h and 12
h (Figure 2), suggesting that the culture consists primarily
of ESC in that time range.
These two results suggest that the first three points of each
time series (representing the first 12 hours) can be used to
identify genes whose expression changes significantly in
the early stages of differentiation, before wide scale alter-
ations of gene expression. This observation is consistent
with, and conservative relative to, a previous study of
mESC showing that a major loss of pluripotency occurs
within the first 24 hours of differentiation into EBs [7].
Based on this result, we focused on probe sets showing
consistent changes of expression in the first 12 hours of
the time course in at least two of the three cell lines. These
genes are likely to include those involved in processes
controlling early differentiation.
2.2. Broad functional analysis
To confirm that the gene expression data of the first 12
hours of the time series can be used to detect genes related
to mESC differentiation, we studied the properties of
genes related to large or small expression changes in that
period using a heuristic approach.
We first pooled together and normalized the data from
the two GeneChips used (MOE430A and MOE430B) (see
Methods). Then, we pre-selected the probe sets that were
expressed in any of the 0 h, 6 h, or 12 h time points in all
three cell types. For this we used the detection call
(present (P), marginal (M), or absent (A)) generated by
the Affymetrix MAS5 software, which indicates whether a
transcript is detected, and by extension, whether a gene is
expressed. Accordingly, we selected probe sets with at least
one P call in any of the first three time points for all three
time series. A total of 16,752 of the 45,139 probe sets on
Clustered heat maps for differentiation time-course experiments in three mouse ES cell lines: R1, J1, and V6.5
Figure 1
Clustered heat maps for differentiation time-course experiments in three mouse ES cell lines: R1, J1, and V6.5.
Expression values for 1,605 probe sets, corresponding to 1,327 transcription factor genes (according to protein domain analy-
sis), were extracted from data from the MOE430A microarray for 11 time points from 0 hours to 14 days. The data for each
cell line were mean-centered by probe set and by array, and clustered using Cluster 3.0 (University of Tokyo). Clusters were
visualized using Java TreeView [72].
R1J1V6.5

Page 4

BMC Genomics 2007, 8:85http://www.biomedcentral.com/1471-2164/8/85
Page 4 of 21
(page number not for citation purposes)
Marker genes (Oct4/Pou5f1, Cripto/Tdgf1, and Rex1) expression pattern in the three time series
Figure 2
Marker genes (Oct4/Pou5f1, Cripto/Tdgf1, and Rex1) expression pattern in the three time series. The identifiers of
the probe sets used were 1417945_at (for Oct4/Pou5f1), 1450989_at (for Cripto/Tdgf1), and 1418362_at (for Rex1). Signal val-
ues were normalized to the value at 0 h.
0
0.2
0.4
0.6
0.8
1
1.2
0
0.2
0.4
0.6
0.8
1
1.2
0
0.2
0.4
0.6
0.8
1
1.2
14d9d7d 4d 48h36h24h18h 12h 6h0h
14d9d 7d 4d 48h36h24h18h12h6h 0h
14d
J1
9d7d 4d48h 36h24h18h12h 6h 0h
R1
V6.5
Oct4/Pou5f1
Cripto/Tdgf1 Rex1

Page 5

BMC Genomics 2007, 8:85http://www.biomedcentral.com/1471-2164/8/85
Page 5 of 21
(page number not for citation purposes)
12 h/0 h versus 6 h/0 h signal ratios for MOE430A/B probe sets
Figure 3
12 h/0 h versus 6 h/0 h signal ratios for MOE430A/B probe sets. Values are m6, the median of the three 6 h/0 h signal
ratios, and m12, the median of the three 12 h/0 h signal ratios, in the three time series (section 2.2). Marked data points corre-
spond to the 42 probe sets selected from the MOE430A data using SAM (section 2.4). Green squares: 15 probe sets down-
regulated at 6 h; green circles: 13 probe sets down-regulated at 12 h; red circles: 7 probe sets up-regulated at 12 h; red
squares: 7 probe sets up-regulated at 6 h. The data underlying this table are available [see Additional file 1].
-1
-0.5
0
0.5
1
-1-0.500.5
log(6h/0h)
log(12h/0h)

Page 6

BMC Genomics 2007, 8:85http://www.biomedcentral.com/1471-2164/8/85
Page 6 of 21
(page number not for citation purposes)
the MOE430A and MOE430B chips passed this filter [see
Additional file 1].
Next, we computed the ratios of the signal of these probe
sets at 6 h and 12 h relative to the initial time point (6 h/
0 h and 12 h/0 h) in the three time series. Defining m6
and m12, as the median of the 6 h/0 h ratios and of the 12
h/0 h ratios for a given probe set in the three time series,
respectively, we plotted m6 against m12 for each probe set
(Figure 3). The distribution of the values indicated a small
number of probe sets with particularly large fold changes.
In order to profile the functions of the genes whose
expression was modified during this time period, we used
the GOstat server [11], which searches for Gene Ontology
(GO) terms [12] significantly under- and over-represented
in a given set of probe sets or genes in relation to a refer-
ence set (see Methods for details). We first analyzed GO
terms for the genes which showed the largest expression
changes, regardless of the direction of the change. We
sorted the probe sets by their Euclidean distance from the
origin in the logs ratio graph represented in Figure 3,
(0,0), which corresponds to probe sets with invariant gene
expression at 0 h, 6 h, and 12 h. The top and bottom 10%
Phylogenetic distribution of proteins associated with probe sets
Figure 4
Phylogenetic distribution of proteins associated with probe sets. Percentage of proteins with homologues in a given
organism: Homo sapiens (human), Brachydanio rerio (fish), Xenopus laevis (frog), Drosophila melanogaster (fly), Caenorhabditis ele-
gans (worm), and Saccharomyces cerevisiae (yeast). For each organism, the leftmost column indicates homologues found for pro-
teins from the 1,675 probe sets with largest gene expression changes, the middle column those from the complete set of
16,752 probe sets, and the rightmost column those from the 1,675 probe sets with smallest gene expression changes. In each
column, the dark bottom part indicates the percentage of proteins aligned along their full length (less than 30 amino acids
unmatched at the N- and C-termini of both sequences), and the lighter upper part is the percentage of proteins with sequence
similarity (no length restriction). Proteins were considered similar with a BLAST E-value < 1e-6. 2
Human FrogFishFly WormYeast
100%
0%
80%
60%
40%
20%
large
small
any
full-
length
partial
gene
expression
change
homology

Page 7

BMC Genomics 2007, 8:85http://www.biomedcentral.com/1471-2164/8/85
Page 7 of 21
(page number not for citation purposes)
of probe sets (1,675) were selected as representing the
genes with the largest and smallest gene expression
changes in the first 12 hours of differentiation, respec-
tively. In the genes with the largest changes, the following
GO terms were over-represented: "organogenesis" (P-
value = 0.0004) and related "morphogenesis" (0.0008),
"intracellular signaling cascade" (0.17), and "cell differen-
tiation" (0.2). In contrast, for the genes with the smallest
changes in expression levels, over-represented GO terms
included "mitochondrion"
(0.008), and a series of terms related to housekeeping
functions, such as "ATP metabolism" (0.04) or "unfolded
protein binding" (0.04).
(0.0001), "cytoplasm"
To investigate the functions associated with a directional
change in expression, we performed GOstat analysis on
the 10% of probe sets with the largest (increasing expres-
sion) and smallest (decreasing expression) sums of m6
and m12 values. Over-represented GO terms in the genes
with increased expression indicated "DNA binding"
(0.013) and "regulation of transcription, DNA depend-
ent" (0.09). In the genes with decreased expression,
selected terms were "extracellular matrix (sensu Meta-
zoa)" (1.63e-05), "cell communication" (0.0004), "orga-
nogenesis" (0.007), "signal transduction" (0.014),
"protease inhibitor activity" (0.02), and "amino acid
metabolism" (0.07).
This analysis suggests that protein production and func-
tions related to communication between the cell and the
environment were down-regulated in the time period
under analysis, while the handling of genetic information
was up-regulated. These observations fit with what might
be expected; as differentiation is initiated, transcription
factors (which bind DNA and regulate transcription) are
expressed to initiate downstream changes in gene expres-
sion patterns. Meanwhile, genes for extracellular matrix
components are turned off as the cells change from grow-
ing attached to the plate, to growing in floating culture;
many of these same genes are related to cell communica-
tion.
2.3. Phylogenetic distribution
We hypothesized that genes important for stem cell func-
tion may have arisen along the metazoan lineage. There-
fore, genes with large expression changes in the period
studied should have a particular phylogenetic signature.
To verify this, we examined the phylogenetic distribution
of the homologues of the protein products assigned to
each of the 16,752 probe sets selected above, and con-
trasted it with the gene expression changes.
We used NetAffx [13] to identify mouse protein sequences
associated with each of the probe sets, and for each mouse
protein identified the most similar protein sequence in
the SPtrEMBL database [14] for six model eukaryotic
organisms: Homo sapiens, Danio rerio, Xenopus laevis, Dro-
sophila melanogaster, Caenorhabditis elegans, and Saccharo-
myces cerevisiae (see Figure 4). C. elegans is considered to
have stem cells [15] as are the other metazoan organisms
considered in this analysis.
We defined as homologues those proteins identified using
BLAST with a threshold E-value of 1e-6, based on the work
of Lopez-Bigas and Ouzonis [16]. Looking at data for all
16,752 probe sets (irrespective of their changes in gene
expression) we saw an increase in the number of homo-
logues between the worm and the yeast, which corre-
sponds to the global sequence similarities between these
genomes (Figure 4). Taking gene expression into account,
there are no large differences in the distribution of homo-
logues between the genes with largest and smallest gene
expression changes.
However, sequence similarity between two proteins often
covers only a small fraction of the total length of the
sequences being compared. This is not surprising if one
considers that the proteins analyzed have multiple
domains some of them present in many proteins. Partial
similarity does not imply functional equivalence between
the sequences compared. Accounting for full sequence
similarity is important, as exemplified by the human La
protein (Sjögren's syndrome antigen B) whose homologs
in fungi are not essential but which is critical for the sur-
vival of mESC cells [17]. The fact that the yeast sequence
(275 amino acids long) is much shorter than the human
and mouse proteins (408 and 415 amino acids long,
respectively), probably accounts for these striking func-
tional differences. A stricter definition of homology
requires the identical domain organization of the com-
pared proteins [18], which can be approximated to
sequence similarity extending over the full length of the
compared sequences. To do this, we required that fewer
than 30 amino acids were left unmatched at the C- or N-
termini of either of the two compared sequences.
With this additional constraint, differences in the distribu-
tion of homologues emerged (Figure 4). Relative to the
complete set described above, a smaller proportion of
proteins from genes with large expression changes was
found to have full-length homologues in non-mamma-
lian species (with the largest difference in the fly). This
indicates that the genes with large changes in gene expres-
sion are enriched in genes which appeared after the emer-
gence of metazoans (especially of arthropods), and before
the mammals' radiation.
For genes with the smallest expression changes, an
increased proportion of proteins have full-length homo-
logues in all species. This observation agrees with the

Page 8

BMC Genomics 2007, 8:85http://www.biomedcentral.com/1471-2164/8/85
Page 8 of 21
(page number not for citation purposes)
hypothesis that the smallest changes in expression would
be observed in housekeeping genes, which would be
expected to be conserved across a wide range of species.
2.4. Selection of a small set of probe sets and genes
To focus on a small set of genes for illustration, compari-
son to other analyses, and to suggest targets for experi-
mental work, we selected probe sets showing consistent
expression changes across replicates in the three cell lines
analyzed. We used RMA [19] to normalize the data for the
first three time points in each of the three time series (9
microarrays for each cell line, as triplicate arrays were run
for each time point). Then for each cell line, we used SAM
[20] to identify the top 100 probe sets with the most sig-
nificant changes in gene expression between 0 h and 6 h,
and between 0 h and 12 h, with separate analysis of the
MOE430A and MOE430B arrays.
SAM analysis of the MOE430B array indicated much
higher false discovery rates than were observed for the
MOE430A array (see Methods for details). As the
MOE430A array measures genes that are generally better
Signal values of 42 selected MOE430A probe sets in the three time series (R1, J1, and V6.5)
Figure 5
Signal values of 42 selected MOE430A probe sets in the three time series (R1, J1, and V6.5). Gene names and
probe set identifiers are indicated on the left. Relative to their signal value at time 0 h, top 15 probe sets displayed a significant
signal decrease at 6 h, next 13 displayed a significant signal decrease at 12 h, next 7 displayed a significant signal increase at 12
h, and bottom 7 displayed a significant signal increase at 6 h, in at least two of the three cell lines analyzed.
12h<0h
6h<0h
6h>0h
12h>0h

Page 9

BMC Genomics 2007, 8:85http://www.biomedcentral.com/1471-2164/8/85
Page 9 of 21
(page number not for citation purposes)
characterized than those on MOE430B, we proceeded to
report only the analysis of MOE430A.
We compared the top 100 lists for the three cell lines to
select probe sets present in at least two out of three lists
with fold changes in the same direction for each of the 6
h/0 h and 12 h/0 h analyses. This condition was fulfilled
by 15 probe sets corresponding to 11 genes down-regu-
lated at 6 h, for 13 probe sets (13 genes) down-regulated
at 12 h, for 7 probe sets (7 genes) up-regulated at 12 h,
and for 7 probe sets (5 genes) up-regulated at 6 h (see Fig-
ure 5 and Table 1). In total, we selected 42 probe sets cor-
responding to 36 genes for further analysis. When these
probe sets were highlighted on the expression fold change
graph (Figure 3), they were found towards the extremes of
the distribution, indicating a degree of agreement
between the two methods.
To allow statistical analysis of over-represented gene func-
tion in genes selected using SAM, we generated a longer
list of genes present in the top 1,000 of at least two of the
three time series for the 0 h/6 h and 0 h/12 h SAM com-
parisons, with fold changes in the same direction. This
new list was analyzed with GOstat. In the 0 h/6 h compar-
ison, a total of 163 up-regulated probe sets were selected.
Over-represented GO terms were "mRNA metabolism"
(P-value = 0.0012) and related terms (for example,
"nuclear mRNA splicing, via spliceosome", P-value =
0.03), "actin cytoskeleton" (0.02), "nucleic acid binding"
(0.03), and "regulation of transcription" (0.05). In the 0
h/6 h comparison, a total of 123 down-regulated probe
sets were selected. GO terms over-represented were "sterol
metabolism" (0.0004) and related, "immune cell activa-
tion" (0.006), and "organ development" (0.00856).
In the 0 h/12 h comparison, a smaller set of statistically
significant GO terms was observed. The 244 selected up-
regulated probe sets displayed no significant GO term.
The 113 down-regulated probe sets had over-represented
GO terms related to development, for example "organ
morphogenesis" (0.002), "growth factor activity" (0.008),
and "angiogenesis" (0.0004).
Functions related to cytoskeleton and transcriptional reg-
ulation were highlighted both here and in the broad func-
tional analysis (section 2.2) confirming the relative
equivalence of both analyses. The presence of mRNA
related functions in the up-regulated probe sets in this
analysis is however interesting as suggests that splicing has
an important role in the control of mESC differentiation.
2.5. Analysis of 36 selected proteins
The domain organization of the 36 selected gene products
are represented in Figure 6. Their functions, subcellular
localization, and up- or down-regulation are schematized
in Figure 7. The relation of some of these genes and pro-
teins to differentiation is known, but some are poorly
characterized.
The phylogenetic analysis of their sequences (following
the methodology described in section 2.3; details in Meth-
ods section) indicated that sequence similarity is found
for most of these proteins across the metazoan species
studied (for example, 30 of the 36 proteins, 83%, are sim-
ilar to fly proteins). However, full-length sequence simi-
larity was much more restricted (only five sequences,
14%, to fly proteins) [see additional file 2]. These proteins
contain many domains general to all metazoans, account-
ing for the high levels of partial sequence similarity (see
Figure 6), but their domain organization is more specific.
One such example is the Nr0b1 protein: towards the C-
terminus it contains a HOLI domain, which is also found
in fly and worm proteins, but its N-terminal sequence
(amino-acids 1–257), and thus its global domain compo-
sition, are unique to mammals. Another example is
Plekha2, which has two Pleckstrin Homology (PH)
domains (Figure 6); while these domains are present in
the eukaryote phylum down to the yeasts, the sequence
between the two PH domains (residues 116–198),
present in amphibians and fish (Xenopus laevis and Danio
rerio), is missing from fly, worm, and yeast sequences.
This analysis shows that the protein products of Myl9,
Tagln, Sc4mol, and Wdr1, have full-length similar
sequences in the yeast, which argues against their involve-
ment in stem cell related functions. Conversely, the pro-
tein products of Adm and Trh, do not appear to have any
significantly similar sequence (even partially similar) in
the worm and the fly, indicating they probable appeared
with the emergence of arthropods. Thus, their role in stem
cell (specific) related processes seems more plausible. All
36 proteins have very similar sequences in human, with at
least 28 having full-sequence-similarity.
The following sections describe these 36 proteins, their
functions, sub-cellular localization, and known or possi-
ble relation to mESC differentiation. Relevant protein syn-
onyms are indicated within brackets.
2.5.1. Intracellular signaling related proteins
Strikingly, five out of six protein products related to intra-
cellular signaling were associated with probe sets down-
regulated at 6 hr (Pim1, Pim3, SOCS3) or later at 12 hr
(Anxa3, Mras); only Plekha2 (TAPP2) was up-regulated
(Figure 6). The response of some of these genes to LIF is
known. SOCS-3, suppressor of cytokine signaling-3, a
negative regulator of the insulin receptor signaling path-
way [21], is known to be transcriptionally activated by LIF
[22]. This fits with the observed reduction in its expres-
sion upon LIF removal. The role of SOCS-3 in differentia-

Page 10

BMC Genomics 2007, 8:85http://www.biomedcentral.com/1471-2164/8/85
Page 10 of 21
(page number not for citation purposes)
Table 1: Probe sets selected by SAM analysis (section 2.4).
keyProbe set id Gen Gen descriptionFunction/process/component (GO)1
D61418133_atBcl3B-cell leukemia/lymphoma 3 DNA binding; transcription factor activity; regulation of transcription, DNA-
dependent; nucleus;
D61418025_at Bhlhb2basic helix-loop-helix domain containing, class B2 DNA binding; protein binding; nucleus; regulation of transcription, DNA-dependent;
transcriptional repressor activity; negative regulation of transcription, DNA-
dependent;
D61419076_a_at Brca2breast cancer 2nucleic acid binding; protein binding; nucleus; cytoplasm; DNA repair; response to
DNA damage stimulus; double-strand break repair via homologous recombination;
synaptonemal complex; single-stranded DNA binding; DNA replication; chromatin
remodeling; apoptosis; regulation of S phase of mitotic cell cycle; mitotic checkpoint;
transcriptional activator activity; regulation of transcription;
D6 1416039_x_atCyr61 cysteine rich protein 61regulation of cell growth; patterning of blood vessels; protein binding; insulin-like
growth factor binding; extracellular; extracellular space; chemotaxis; cell adhesion;
heparin binding; growth factor binding; sensory perception;
D61417394_at Klf4Kruppel-like factor 4 (gut) nucleic acid binding; DNA binding; nucleus; regulation of transcription, DNA-
dependent; zinc ion binding; transcription factor activity; transcriptional activator activity;
transcriptional repressor activity; negative regulation of transcription, DNA-dependent;
D61417395_at
D61435458_at Pim1proviral integration site 1protein kinase activity; protein serine/threonine kinase activity; protein-tyrosine
kinase activity; ATP binding; nucleus; protein amino acid phosphorylation; cell growth
and/or maintenance; kinase activity; transferase activity;
D6 1437100_x_atPim3proviral integration site 3protein kinase activity; protein serine/threonine kinase activity; ATP binding; protein
amino acid phosphorylation; kinase activity; transferase activity;
D6 1451069_at
D61423078_a_atSc4mol sterol-C4-methyl oxidase-likeC-4 methyl sterol oxidase activity; catalytic activity; endoplasmic reticulum; plasma
membrane; fatty acid metabolism; metabolism; steroid metabolism; integral to
membrane; sterol biosynthesis; oxidoreductase activity;
D61415823_atScd2stearoyl-Coenzyme A desaturase 2 fatty acid biosynthesis; superoxide metabolism; lipid biosynthesis; stearoyl-CoA 9-
desaturase activity; copper, zinc superoxide dismutase activity; iron ion binding;
oxidoreductase activity; metal ion binding; oxidoreductase activity, acting on paired
donors, with oxidation of a pair of donors resulting in the reduction of molecular
oxygen to two molecules of water; endoplasmic reticulum; membrane; integral to
membrane; endoplasmic reticulum
D6 1416576_atSocs3 suppressor of cytokine signaling 3regulation of cell growth; signal transduction; intracellular signaling cascade; regulation
of protein amino acid phosphorylation; protein binding; negative regulation of insulin receptor
signaling pathway;
D6 1455899_x_at
D61456212_x_at
D61450989_atTdgf1teratocarcinoma-derived growth factoractivation of MAPK; extracellular; extracellular space; epidermal growth factor
receptor signaling pathway; growth factor activity; positive regulation of cell
proliferation; determination of anterior/posterior axis, embryo; extrinsic to plasma
membrane;
D121460330_atAnxa3 annexin A3 phospholipase inhibitor activity; calcium ion binding; calcium-dependent phospholipid
binding; phospholipase A2 inhibitor activity;
D121449141_atFblim1filamin binding LIM protein 1
protein binding; cytoskeleton; cell adhesion; zinc ion binding; regulation of cell shape;
D12 1427238_atFbxo15F-box protein 15 protein binding; ubiquitin cycle; SCF ubiquitin ligase complex;
D121426858_atInhbb inhibin beta-Bhormone activity; extracellular; growth factor activity; cell growth and/or
maintenance; growth; ovarian follicle development; defense/immunity protein activity;
cytokine activity; response to external stimulus; protein homodimerization activity; host cell
surface receptor binding; positive regulation of follicle-stimulating hormone secretion; negative
regulation of follicle-stimulating hormone secretion; negative regulation of hepatocyte growth
factor biosynthesis;
D121451021_a_atKlf5Kruppel-like factor 5nucleic acid binding; DNA binding; nucleus; regulation of transcription, DNA-
dependent; zinc ion binding; angiogenesis; transcription factor activity; protein binding;
microvillus biogenesis; positive regulation of transcription;

Page 11

BMC Genomics 2007, 8:85http://www.biomedcentral.com/1471-2164/8/85
Page 11 of 21
(page number not for citation purposes)
D121418153_atLama1laminin, alpha 1 morphogenesis of an epithelial sheet; receptor binding; structural molecule activity;
extracellular matrix structural constituent; protein binding; extracellular matrix;
basement membrane; basal lamina; laminin-1; cell adhesion; cell surface receptor
linked signal transduction; histogenesis; regulation of cell adhesion; regulation of cell
migration; regulation of embryonic development; establishment of epithelial cell polarity;
D12 1424719_a_atMapt microtubule-associated protein tauprotein binding; cytoskeletal regulatory protein binding; cytoskeleton; microtubule
associated complex; microtubule-based process; microtubule stabilization; microtubule;
actin filament organization; microtubule-based movement;
D12 1449590_a_at Mras muscle and microspikes RAS GTP binding; small GTPase mediated signal transduction; small monomeric GTPase
activity; RAS small monomeric GTPase activity;
D121417760_atNr0b1 nuclear receptor subfamily 0, group B, member 1DNA binding; transcription factor activity; steroid hormone receptor activity;
receptor activity; ligand-dependent nuclear receptor activity; nucleus; cytoplasm;
regulation of transcription, DNA-dependent; negative regulation of transcription;
transcriptional repressor activity;
D121448612_at Sfnstratifinregulation of progression through cell cycle; regulation of cyclin dependent protein
kinase activity; protein domain specific binding; cytoplasm;
D12 1423505_atTagln transgelinmuscle development; NOT cytoskeleton organization and biogenesis;
D12 1418756_atTrh thyrotropin releasing hormonehormone activity; neuropeptide hormone activity; protein binding; extracellular;
extracellular space; thyrotropin-releasing hormone activity; hormone mediated
signaling;
D12 1450641_atVimvimentinstructural molecule activity; intermediate filament; type III intermediate filament;
intermediate filament-based process; oxygen transporter activity; oxygen transport;
U12 1427202_at 483344
2J19Rik
RIKEN cDNA 4833442J19 gene S-adenosylmethionine-dependent methyltransferase activity;nucleus;
U12 1416077_atAdmadrenomedullinhormone activity; neuropeptide hormone activity; extracellular; extracellular space;
neuropeptide signaling pathway;
U121449324_at Ero1l/
LOC43
4220
ERO1-like (S. cerevisiae)endoplasmic reticulum; electron transport; protein folding; transport; membrane;
oxidoreductase activity; integral to endoplasmic reticulum membrane;
U12 1421317_x_at Myb myeloblastosis oncogeneregulation of cell cycle; G1/S transition of mitotic cell cycle; DNA binding;
transcription factor activity; nucleus; regulation of transcription, DNA-dependent;
calcium ion transport; cell growth and/or maintenance; viral assembly;
U12 1425926_a_atOtx2orthodenticle homolog 2 (Drosophila)DNA binding; transcription factor activity; protein binding; nucleus; regulation of
transcription, DNA-dependent; regulation of transcription from Pol II promoter;
development; central nervous system development; organogenesis; anterior/posterior
pattern formation; dorsal/ventral pattern formation; transcription regulator activity;
forebrain development; midbrain development; eye photoreceptor cell fate
commitment; cell fate commitment;
U121418391_at Phf21a PHD finger protein 21Atranscription; regulation of transcription, DNA-dependent; DNA binding; protein
binding; zinc ion binding
U12 1417288_atPlekha2 pleckstrin homology domain-containing, family A
(phosphoinositide binding specific) member 2
phosphatidylinositol binding; nucleus; lipid binding; membrane;
U6 1448129_atArpc5 actin related protein 2/3 complex, subunit 5structural constituent of cytoskeleton; cytoplasm; Arp2/3 protein complex; cell
motility; actin filament organization; actin cytoskeleton; DNA binding; electron
transport; heme binding; lamellipodium; regulation of actin filament polymerization;
regulation of transcription;
U61434642_atDhrs8dehydrogenase/reductase (SDR family) member 8extracellular space; metabolism; oxidoreductase activity;
U6 1452670_atMyl9 myosin, light polypeptide 9, regulatorycalcium ion binding; muscle myosin; regulation of muscle contraction; muscle
development; structural constituent of muscle; myosin; motor activity;
U61426596_a_atSmn1 survival motor neuron 1nucleic acid binding; RNA binding; nucleus; cytoplasm; mRNA processing; spliceosome
assembly;
U61423054_atWdr1 WD repeat domain 1actin binding; cytoskeleton; perception of sound; actin cytoskeleton;
U61437591_a_at
U61450851_at
Column 1 indicates probe set behaviour: D6, down-regulated at 6 h; D12, down-regulated at 12 h; U12, up-regulated at 12 h; U6, up-regulated at 6
h. Associated information was taken from NetAffx unless otherwise indicated.
1GO-terms in italics were manually derived from database entries linked from NetAffx using the assistance of a data mining method [64]
Table 1: Probe sets selected by SAM analysis (section 2.4). (Continued)

Page 12

BMC Genomics 2007, 8:85http://www.biomedcentral.com/1471-2164/8/85
Page 12 of 21
(page number not for citation purposes)
tion remains unclear, as it has been observed to act both
as a promoter [23] and as a repressor of differentiation in
mESC [24].
Pim1, proviral integration site 1, and Pim3, proviral inte-
gration site 3, belong to a family of serine/threonine
kinases. PIM genes (Pim1, Pim2, Pim3) were initially dis-
covered through proviral insertional activation by murine
leukemia virus and have been shown to cooperate with c-
MYC in murine leukemia. Pim genes are related to multi-
ple growth processes including cytokine-mediated cell
growth and differentiation of bone marrow cells [25].
However, Pim knock-out mice are viable and fertile, which
makes it unlikely that these genes play an important role
in mESC differentiation. Induction of Pim expression is
likely mediated by JAK/STAT signaling [26] and that path-
way is induced by LIF, so the observed decrease in expres-
sion of both Pim1 and Pim3 is consistent.
Mras/Rras3, muscle and microspikes RAS, is a RAS small
GTPase. The human Mras is associated with the biogenesis
of the actin cytoskeleton in muscle cells and its product
interacts with the RAS-binding domain of a multitude of
proteins including RAF1 and RALGDS [27]. When acti-
vated by STAT3 [28], it mediates the activation of MAPK
in a cell dependant manner [29], thus linking the JAK/
STAT and RAS/MAPK pathways. This agrees with the
observed down-regulation of Mras as LIF induces JAK/
STAT, and STAT3 induces this gene.
Domain organization of the 36 selected proteins
Figure 6
Domain organization of the 36 selected proteins. Illustrations are adapted from Pfam [67] or SMART [66]. Names close
to or inside the domain pictures correspond to Pfam domain identifiers or, in the case of Tdgf1, Wdr1, Nr0b1, and MRas, to
SMART domain identifiers. Three-coloured boxes correspond to Pfam-b domains that are automatically predicted. All
sequences are at the same scale except Lama1.
Klf4
Zn fingers
other cytoplasmic
cytoskeleton
related
extra-cellular
Bcl3 Ankyrin repeats
Brca2
Brca2 repeats
Bhlhb2
Hairy orange
Sc4molScd2
Fbxo15
Fbox
Klf5Zn finger
Lama1
Laminin N
Laminin EGF domains
Laminin BLaminin BLaminin ILaminin IILaminin G1
Nr0b1
Sfn
4833442J19Rik
Ero1l
Wdr1
WD40 repeats
Smn1
Arpc5
Homeobox
TF_Otx
Tagln
Fblim1
Calponin
repeat
MaptTubulin binding domains
Myl9
EF hand domains
Vim
Filament_head
Tdgf1
Cyr61
TSP1
Cys knot
Inhbb
TGFb
Adm
Adrenomedullin
Dhrs8
Trh
Myb
Myb DNA binding domains
Phf21a
Otx2
nuclear
endoplasmic
reticulum
Socs3
Mras
SOCS box
Anxa3
Annexin repeats
intracellular
signaling
Pim3Pim1
Plekha2

Page 13

BMC Genomics 2007, 8:85http://www.biomedcentral.com/1471-2164/8/85
Page 13 of 21
(page number not for citation purposes)
Plekha2 (TAPP2) is an adaptor protein that can interact,
through its two PH domains, with the lipids of the plasma
membrane produced by the PI3 kinases, but its role is
unknown [30]; here we propose that it is associated with
early ESC differentiation.
2.5.2. Cytoskeleton related proteins
Seven protein products were related to the cytoskeleton:
three associated with probe sets up-regulated at 6 h
(Wdr1, Arpc5, and Myl9), and four associated with down-
regulated probe sets (Fblim1, Vim (vimentin), Tagln
(transgelin), and Mapt (microtubule-associated protein
tau)). The three down-regulated proteins are found in all
eukaryotes and therefore their importance to ESC devel-
opment is dubious: WDR1 protein (also AIP1, Actin inter-
acting protein 1) has been associated with cell
morphologic changes [31], ARPC5 is a component of the
Arp2/3 complex, an actin filament nucleator that activates
regulated actin assembly in response to extracellular sig-
nals in eukaryotic cells [32], and Myl9 (myosin, light
polypeptide 9, regulatory) is characterized by its similarity
to the human Myl9 or MLC2 [33], the myosin light chain
2.
The protein encoded by Fblim1 is known as Migfilin and
acts as mediator between the plasma membrane and the
actin cytoskeleton. Migfilin is also involved in cell mor-
phology [34]. The other three down-regulated have devel-
opmental roles, but they are markers of differentiated
cells: Vimentin of neuroepithelial cells, Tau of retinal gan-
glion cells, and Transgelin (SM22alpha) of smooth mus-
cle cells.
Function and sub-cellular localization of the 36 selected proteins
Figure 7
Function and sub-cellular localization of the 36 selected proteins. The colours of the labels indicate the behavior of
the corresponding probe sets (dark-green: down-regulated at 6 h; light green: down-regulated at 12 h; tan: up-regulated at 12
h; pink: up-regulated at 6 h). Pathways and other related proteins are indicated with white labels. Circular arrows indicate
transport between cytoplasm and nucleus. Solid and broken arrows indicate direct activation and activation through intermedi-
ate proteins, respectively. Information about the JAK-STAT and MAPK pathways was taken from the references listed in this
manuscript or from the KEGG database [73].
MAPK
signaling
pathway
Socs3
Adm
Klf4
Regulation of
transcription,
DNA-dependent
Nucleus
Bcl3
Bhlhb2
4833442J19Rik
Myb
Otx2
Phf21a (Bch80)
Fbxo15
Inhbb
Klf5
Mras
Nr0b1
Tdgf1
(Cripto-1)
Extracellular space
Extracellular
matrix
Endoplasmic
reticulum
Ero1l
Intracellular signaling
JAK
Lama1
Trh
Arpc5
Cytoskeleton
Wdr1
Myl9
Mapt
Vim
LIF
Sfn (Smfn/Rexo2)
Fblim1
STAT
Brca2
Pim1
Smn1
Anxa3
Pim3
Cyr61 (CCN1)
Tagln
Plekha2 (TAPP2)
Dhrs8 (PAN1B)
Sc4molScd2
Lipid biosynthesis

Page 14

BMC Genomics 2007, 8:85 http://www.biomedcentral.com/1471-2164/8/85
Page 14 of 21
(page number not for citation purposes)
2.5.3. Nuclear proteins
The importance of transcriptional regulation for the con-
trol of mESC differentiation is exhibited in the high per-
centage of nuclear proteins in our selected set. Four
proteins corresponded to up-regulated probe sets: Smn1,
Phf21a, Myb, and Otx2. Another six proteins corresponded
to down-regulated probe sets: Brca2, Bhlbh2, Bcl3, Klf4,
Klf5, and Nr0b1.
Eight of these ten proteins regulate transcription through
DNA interaction. For example, the transcription factor
Klf4 (down-regulated at 6 hr), Kruppel-like factor 4,
inhibits murine embryonic stem cell differentiation [23],
which agrees with the observed down-regulation. It has
been described as inducing and repressing a number of
signaling pathways that control macrophage activation
[35]. Otx2 encodes a homeobox domain containing pro-
tein, which is a regulator of neurogenesis in the develop-
ing brain in mouse embryo [36].
The two that are not transcription factors, SMN1 and
BRCA2, can both shuttle between nucleus and cytoplasm.
SMN1 is present in cytoplasm, nucleoplasm, and in the
Cajal bodies where it is needed for the biogenesis of spli-
ceosomal small nuclear ribonucleoproteins [37]; knock-
out of these gene leads to massive cell death in early mice
embryos [38]. BRCA2 is involved in DNA repair and can
be transported to the cytoplasm probably as a way to con-
trol its function [39].
2.5.4. Extracellular proteins
Six extracellular proteins were in our list of selected pro-
teins, all but one down-regulated. Three of them are
related to a receptor binding function: Trh (tyrothropin
releasing hormone), Lama1 (laminin alpha-1), and Tdgf1
(Cripto). The latter is an indirect activator of the Ras/
MAPK pathway [40] and well known as involved in the
determination of the anterior-posterior axis in the mouse
embryo [41]. Laminin alpha-1, is also involved in embry-
onic patterning and is one of the few essential extracellu-
lar matrix proteins in early embryogenesis [42].
The others are Inhbb (inhibin beta-B) that acts as activin-
B, a homodimer of two Inhbb, and two angiogenic fac-
tors, Adm (adrenomedullin) [43], and Cyr61 (CCN1), the
latter being present in the extracellular matrix [44].
2.5.5. Other intracellular proteins
There were eight other intracellular proteins in our list:
four cytoplasmic, Fbxo15, 4833442J19Rik, Sfn (Smfn/
Rexo2), and Dhrs8, and three located in the ER Ero1l,
Sc4mol, and Scd2. Two of the latter, Sc4mol (ERG25) and
Scd2, were up-regulated at 6 h and are related to lipid bio-
synthesis, whereas up-regulated Ero1l codes for an oxi-
doreductase involved in oxidative ER protein folding [45].
ERG25 is a sterol C-4 methyloxidase involved in the bio-
synthesis of sterol [46]. SCD2 is a stearoyl-CoA desaturase
involved in the biosynthesis of monounsaturated fatty
acids [47]. Lipids, besides their function in modulating
membrane function, can also act as second messengers in
developmental signaling. Changes in the expression of
genes coding for enzymes involved in their biosynthesis
could reflect alterations in the cellular membrane related
to the ESC differentiation process.
Cytoplasmic Sfn (Smfn/Rexo2) is a 3' to 5' exonuclease of
small (< 5nt) poly-nucleotides, hypothesized to be
involved in nucleotide recycling, and conserved in yeasts
and in Escherichia coli [48]. Fbxo15 (Fbx15), Fbox protein
15, is a protein of unknown function, known to be target
of Oct3/4 but not essential for development [49]; as it
contains an F-box domain this protein could recruit a tar-
get protein for ubiquitination. Dhrs8 (PAN1B) is an
enzyme with 17β-hydroxysteroid dehydrogenase activity
also involved in lipid biosynthesis [50]; 17β-dehydroge-
nases play roles in the activation and inactivation of both
estrogens and androgens.
In summary, our methodology primarily selected proteins
involved in intracellular signaling, cytoskeletal control,
and regulation of transcription. None of the 36 proteins
appear to have functions related to translation and
amino-acid biosynthesis, few are secreted, and the few
enzymes selected are involved in lipid biosynthesis.
3. Discussion
Mouse and human ESCs have been the subject of gene
expression profiling in previous studies using DNA micro-
arrays, Serial Analysis of Gene Expression (SAGE), and
Expressed Sequence Tag (EST) library generation (see e.g.
[7] and references therein). However, most of those stud-
ies compared undifferentiated ESCs to their fully differen-
tiated derivatives, and none followed gene expression
dynamics during the first 12 hours of the differentiation
process. Bhattacharya et al. [51] compared five lines of
hESCs to 8-day old EBs. Dvash et al. [52] compared H9
hESCs to 2 day, 10 day, and 30 day old EBs. Palmqvist et
al. [7] compared R1 mESC with ESC differentiated for 18
hours and 72 hours. Sekkai et al. [6] compared undiffer-
entiated Gs2 hESC with ESC differentiated for 16, 24, and
48 hours. The objective in most of these works was to
detect genes turned off or on temporarily during the early
differentiation period. Here, we have carried this idea fur-
ther by obtaining measurements of mESC differentiating
into EBs during a two-week period, including measure-
ments of gene expression as early as 6 hours and 12 hours
after differentiation was started.
Our analysis examined three genetically distinct mESC
lines (R1, J1, and V6.5) with all measurements in biolog-

Page 15

BMC Genomics 2007, 8:85http://www.biomedcentral.com/1471-2164/8/85
Page 15 of 21
(page number not for citation purposes)
ical triplicate, both to assess the robustness of the
observed changes and to avoid biases due to particular
mouse strains. We showed the existence and duration of
an early period in the time series of differentiation prior to
the onset of differentiation. This is important for the char-
acterization of gene expression in ESCs undergoing differ-
entiation before they differentiate into another cell type,
or a mixture of cell types. Clustering the data both by
expression levels of transcription factor genes (Figure 1)
and expression patterns of mESC markers (Figure 2) sug-
gested that the first three time points in the time series (i.e.
the start point, 6 hours, and 12 hours) corresponded to
undifferentiated stem cells in which major expression
changes had not yet begun. We took this as an indication
that the set of genes with expression changes during that
early period would include genes that trigger the differen-
tiation process.
Next we studied patterns of gene expression during this
early period compared to the initial values in the undiffer-
entiated state. We used the changes with respect to the sig-
nal at initial time point of each series (Figure 3). We
observed a distinct functional profile and phylogenetic
distribution for the genes with large expression changes
during the period under study, which were selected as pos-
sibly involved in the differentiation process.
Although this analysis provided a way to test several types
of gene expression profiles in a simple manner, we chose
a more sensitive method of gene expression analysis to
select a list of exemplar genes, SAM [20], which allows the
detection of consistent changes in gene expression using
the variation between replicates.
Using SAM analyses of gene expression to contrast the ini-
tial condition with either the 6 h or the 12 h time point,
we focused on 42 probe sets (representing 36 protein
products) with the most consistent changes of gene
expression in at least two of the three cell lines analyzed
(Figure 5). These probe sets had functional and phyloge-
netic properties consistent to those of the genes selected as
expressed and whose expression changed the most in the
broad analysis.
Among the genes selected, we note a majority of functions
related to the control of transcription (Klf4, Klf5, Bcl3,
Myb, Otx2, Bhlhb2, Phf21a, Nr0b1), intracellular signaling
(Socs3, Anxa3, Mras, Pim1, Pim3), and cytoskeletal related
(Fblim1, Arpc5, Tagln, Wdr1, Myl9, Vim, Mapt) (Figure 6).
Gene expression changes must in large part be modulated
through transcription factors, thus we would expect such
functions to be well represented in the lists of changing
genes. We note that this is one of the most represented
functions and that we found both up-regulated and
down-regulated transcription factors. This suggests that, at
least in the 0 h–12 h frame, the transcriptional control is
not limited to a small number of transcription factors.
Of the 36 protein products studied, six were associated
with the related JAK-STAT and MAPK signaling pathways
(Tdgf1, Socs3, Pim1, Pim3, Mras, Mapt), which are acti-
vated by LIF. These six genes were down-regulated. As
these pathways transduce a large variety of external sig-
nals, this down-regulation must have a major effect on the
cell state.
The abundance of cytoskeletal proteins is not surprising in
light of the number of intracellular signaling proteins
detected. It is well known that cytoskeletal proteins,
besides their structural function, serve to localize intracel-
lular signal pathways (two such examples in ESC are Pax-
illin [53] and beta 1 integrin [54]).
The broad reduction in the expression levels of genes
related to the extracellular matrix (two of which, Lama1
and Cyr61, are among the genes most significantly down-
regulated) reflects the changes in the cell interaction prop-
erties that happen upon differentiation of mESC. ES cells
are cultured as an attached monolayer, but their differen-
tiated derivatives grow as unattached spheres. Some stud-
ies are starting to shed light on how the interrelation
between mESC and their environment, mediated through
proteins of the extracellular matrix, controls their self-
renewal [55]. Our study suggests Lama1 and Cyr61 as can-
didate genes for the study of this process.
The absence of certain functions from the genes selected
in our broad and focused analyses also provides informa-
tion about the timing of cell processes involved in ESC
differentiation. Modifications in the mobility of chroma-
tin interacting proteins are known to occur during loss of
pluripotentiality of ESC [56]. However, we did not
observe expression changes in the genes for chromatin
interacting proteins other than BRCA2, which is impli-
cated in chromatin remodeling [57]. Neither did we
observe large modifications in the expression of genes
coding for surface receptors; such changes are occurring
later in human ESC differentiation into embryoid bodies
[52]. Apparently, the only proteins with enzymatic activ-
ity in our sort list of 36 were two endoplasmic reticulum
resident proteins, Sc4mol (ERG25) and Scd2, both up-reg-
ulated at 6 hours and related to lipid biosynthesis. This
may indicate a lipid involvement in signaling associated
with the differentiation process.
We hypothesized that the genes that control ES differenti-
ation are likely to have arisen in organisms which have
stem cells, and thus to be found only in metazoans. Other
genes which control specific mammalian aspects of ES dif-