Transcriptome Analysis of the Normal Human Mammary Cell Commitment and Differentiation Process

Abstract

Mature mammary epithelial cells are generated from undifferentiated precursors through a hierarchical process, but the molecular mechanisms involved, particularly in the human mammary gland, are poorly understood. To address this issue, we isolated highly purified subpopulations of primitive bipotent and committed luminal progenitor cells as well as mature luminal and myoepithelial cells from normal human mammary tissue and compared their transcriptomes obtained using three different methods. Elements unique to each subset of mammary cells were identified, and changes that accompany their differentiation in vivo were shown to be recapitulated in vitro. These include a stage-specific change in NOTCH pathway gene expression during the commitment of bipotent progenitors to the luminal lineage. Functional studies further showed NOTCH3 signaling to be critical for this differentiation event to occur in vitro. Taken together, these findings provide an initial foundation for future delineation of mechanisms that perturb primitive human mammary cell growth and differentiation.

Full text

Cell Stem Cell
Resource
Transcriptome Analysis of the Normal Human Mammary
Cell Commitment and Differentiation Process
Afshin Raouf,
1
Yun Zhao,
1
Karen To,
1
John Stingl,
1,7
Allen Delaney,
2
Mary Barbara,
4
Norman Iscove,
4
Steven Jones,
2
Steven McKinney,
3
Joanne Emerman,
5
Samuel Aparicio,
3
Marco Marra,
2,6
and Connie Eaves
1,6,
*
1
Terry Fox Laboratory
2
Genome Sciences Centre
3
Molecular Oncology and Breast Cancer Program
British Columbia Cancer Agency, Vancouver, BC V5Z 1L3, Canada
4
Division of Stem Cell and Developmental Biology, Ontario Cancer Institute, Toronto, ON M5G 2M9, Canada
5
Departments of Pathology and Laboratory Medicine, Anatomy and Cell Biology
6
Department of Medical Genetics
University of British Columbia, Vancouver, BC V6T 1Z4, Canada
7
Present address: Cancer Research UK Cambridge Research Institute, Li Ka Shing Centre, Robinson Way, Cambridge CB2 0RE, UK
*Correspondence: ceaves@bccrc.ca
DOI 10.1016/j.stem.2008.05.018
SUMMARY
Mature mammary epithelial cells are generated from
undifferentiated precursors through a hierarchical pro-
cess, but the molecular mechanisms involved, partic-
ularly in the human mammary gland, are poorly under-
stood. To address this issue, we isolated highly
purified subpopulations of primitive bipotent and
committed luminal progenitor cells as well as mature
luminal and myoepithelial cells from normal human
mammary tissue and compared their transcriptomes
obtained using three different methods. Elements
unique to each subset of mammary cells were identi-
fied, and changes that accompany their differentiation
in vivo were shown to be recapitulated in vitro. These
include a stage-specific change in NOTCH pathway
gene expression during the commitment of bipotent
progenitors to the luminal lineage. Functional studies
further showed NOTCH3 signaling to be critical for
this differentiation event to occur in vitro. Taken to-
gether, these findings provide an initial foundation for
future delineation of mechanisms that perturb primi-
tive human mammary cell growth and differentiation.
INTRODUCTION
The normal female mammary gland grows rapidly at puberty to
produce an elaborate bilayered tree-like structure composed
of an inner layer of luminal cells surrounded by an outer layer
of myoepithelial cells. Later cycles of expansion and involution
occur during each estrous cycle and even more dramatically
with each pregnancy (Howlin et al., 2006; Russo and Russo,
2004). This dynamic activity suggests the lifelong maintenance
within the normal mammary gland of a population of self-renew-
ing undifferentiated mammary stem cells. Support for this con-
cept was first provided by vector integration site analysis of
serially transplantable outgrowths generated in precleared
mammary fat pads of mice transplanted with retrovirally marked
tissue fragments (Kordon and Smith, 1998). More recently,
a quantitative in vivo assay for the cell of origin of these clonal
outgrowths has been identified and used to enable their partial
characterization and phenotypic distinction from most other
mammary epithelial cells in mice, including those capable of col-
ony formation in vitro (Asselin-Labat et al., 2006; Shackleton
et al., 2006; Stingl et al., 2006a).
Several lines of evidence suggest that the mammary gland of
normal adult women also contains a population of mammary
stem cells. These include in situ studies of X chromosome inac-
tivation patterns in normal human mammary tissue indicating
a common origin of cells in adjacent lobules and ducts (Tsai
et al., 1996) and in vitro experiments identifying a unique and
rare subset of human mammary progenitor cells that generate
mixed colonies containing both luminal and myoepithelial cells,
as well as other subsets of clonogenic cells that produce only
one or the other mature cell type (Stingl et al., 1998, 2001). The
properties exploited most effectively for the prospective isolation
of these different types of colony-forming cells (CFCs) include
their shared expression of an epithelial marker known as epithe-
lial cell adhesion molecule (EPCAM) (Stingl et al., 1998, 2001)
and a
6
-integrin (CD49f), a frequent marker of epithelial progeni-
tors (Stingl et al., 2001). The undifferentiated (bipotent) CFCs
have also been further distinguished by their expression of the
common acute lymphoblastic leukemia antigen (CALLA, also
called CD10), a marker of the more basally situated mature my-
oepithelial cells (Stingl et al., 1998), whereas the luminal-re-
stricted CFCs selectively express Mucin-1 (MUC1), a specific
marker of mature luminal cells in the mammary gland.
We now describe a procedure that utilizes additional markers
to subdivide normal human mammary epithelial cells into four
fractions, two of which are more highly and exclusively enriched
in bipotent CFCs and luminal-restricted CFCs than previously
achievable, the other two fractions representing the mature my-
oepithelial and luminal cells. Long serial analysis of gene expres-
sion (LongSAGE) and single-channel expression microarray (Af-
fymetrix) technologies were used to generate global transcript
profiles for each of these four subpopulations, and some of the
interesting differences were then further examined by quantita-
tive real-time PCR (Q-RT-PCR) measurements. The results
Cell Stem Cell 3, 109–118, July 2008 ª2008 Elsevier Inc. 109

identify many gene expression changes that accompany the dif-
ferentiation of primitive normal human mammary cells both in
vivo and in vitro. They also provide new evidence for a previously
unrecognized and nonredundant role of NOTCH3 signaling in the
commitment of bipotent mammary cells to the luminal lineage.
RESULTS
Isolation of Highly Purified Populations of Primitive and
Mature Subsets of Human Mammary Epithelial Cells
Human mammary CFCs constitute approximately 1% of freshly
dissociated, fibroblast-depleted suspensions of normal reduc-
tion mammoplasty samples, and this value can be increased
several-fold by culturing the cells in bulk for 3 days (Stingl
et al., 2001). Using such ‘‘precultured’’ cells and improved assay
conditions, we first showed that these allow CFC progenitor fre-
Figure 1. Detection and Purification of
Distinct, Functionally Defined Subsets of
Normal Human Mammary Epithelial Cells
(A) Linearity of colony formation in assays of un-
separated EPCAM
+
cells from 3-day cultured
mammoplasty cells (green, red, and blue lines
show data for three different samples, r R0.95
in each case). Values shown are the mean ±
SEM of three to six replicates in each experiment.
(B) Linearity of colony formation in assays of frac-
tions enriched in bipotent (dashed lines) or lumi-
nal-restricted (solid lines) CFCs isolated as shown
in (C) from the three samples shown in (A) (and
similarly color-coded, r R0.95 in each case).
(C) Strategy used to isolate subpopulations of nor-
mal human mammary EPCAM
+
viable (PI

) epithe-
lial cells selectively enriched in their content of bi-
potent CFCs, luminal-restricted CFCs, mature
myoepithelial cells, or mature luminal cells. A rep-
resentative FACS plot is shown. Typical mixed and
luminal colonies generated in assays of 50 cells
from the purified subsets indicated are shown
(magnified 323).
(D) Distribution of bipotent (circles) and luminal-
restricted (triangles) CFCs in the subsets indicated
(n = 16).
(E) To calculate the yield of bipotent and luminal-
restricted CFCs in each fraction, the values for
each progenitor type in the EPCAM
+
CD49f
+
frac-
tion were set = 100%.
The error bars represent the standard error of the
mean.
quencies to be measured independent of
the number of cells plated over a wide
range (Figures 1A and 1B). Additional ex-
periments showed that highly purified
and distinct populations of bipotent and
luminal-restricted CFCs were consis-
tently obtained by initial immunomag-
netic selection of the EPCAM
+
cells pres-
ent in 3-day cultures of dissociated
mammoplasty samples followed by sub-
dividing both the CD49f
+
and CD49f

compartments according to their com-
bined expression of either MUC1 and CD133, or CD10 and
THY1 (CD90) (Figure 1C). Using this protocol, we found that
the CD49f

fraction contained very few CFCs (2 ± 1 CFCs per
100 starting CD49f

cells, n = 4) in contrast to the CD49f
+
cells,
of which approximately 50% were CFCs and thus contained
>95% of the EPCAM
+
CFCs. Accordingly, the yield of CFCs
present in subsets of CD49f
+
cells analyzed according to their
expression of MUC1 and CD133, or CD10 and THY1, was
used to calculate progenitor yields from different starting popu-
lations. The MUC1

CD133

(CD10/THY1)
+
subset of the CD49f
+
fraction contained the majority (57%) of the bipotent CFCs
at a purity of 45% ± 3% with very few luminal-restricted CFCs
present (only 3% of this subset). The remaining (MUC1/
CD133)
+
CD10

THY1

CD49f
+
cells contained most (96%) of
the luminal-restricted CFCs, which were present in this fraction
at a purity of 32% ± 3% and were contaminated by the presence
Cell Stem Cell
Transcript Analysis of Human Mammary Cell Subsets
110 Cell Stem Cell 3, 109–118, July 2008 ª2008 Elsevier Inc.

of very few bipotent CFCs (only 5% of this subset) (Figures 1D
and 1E).
We next asked whether cells coexpressing cytokeratin 14 and
18 would be found in either of the CD49f
+
subsets and, if so,
whether their numbers would correlate with the number of CFC
present, since coexpression of these two cytokeratins has been
associated with a primitive state of mammary epithelial cells
(Smalley et al., 1999; Welm et al., 2002). Immunostaining of sorted
MUC1

CD133

(CD10/THY1)
+
and (Muc1/CD133)
+
CD10

THY1

cells from three different reduction mammoplasty samples re-
vealed that both of these fractions consistently contained detect-
able, albeit variable, frequencies of dually cytokeratin 14
+
and 18
+
cells. However,relatively high CFC frequencieswere seen in some
samples where dual cytokeratin 14/18
+
cells were rare and vice
versa (see Table S1 available online), thus indicating nonidentity
of this phenotype with any type of CFC.
Transcriptional Profiling of Different Purified Subsets
of Human Mammary Epithelial Cells
cDNA preparations were generated from each of the four sub-
sets of normal human mammary cells using PCR-amplification
methods previously shown to preserve transcript representation
in both LongSAGE libraries (Zhao et al., 2007) and array analyses
(Iscove et al., 2002). LongSAGE libraries were made from all four
fractions obtained from one mammoplasty sample (Table 1),
and each library was then sequenced to a depth of 200,000
tags (Gene Expression Omnibus [GEO] accession number
GSE11395). We found that each of these libraries contained
approximately 34,000 unique tags when a 99.9% sequence
quality cut-off was used. Twelve Affymetrix array hybridizations
were also performed, one for each of the four fractions obtained
from the same sample used for the LongSAGE libraries and the
other eight from the four fractions isolated from two additional
mammoplasty samples (Table 1 and GEO accession number
GSE11395). An unsupervised hierarchical clustering of the data
obtained by Pearson correlation analysis of the LongSAGE
library data demonstrated that the population of cells that was
most enriched in bipotent CFCs was most closely related to
the differentiated myoepithelial cells. Conversely, the population
most enriched in luminal-restricted CFCs was most closely re-
lated to the differentiated luminal cells (Figure 2A). Similar results
were obtained from the larger Affymetrix dataset (Figure 2B).
We then used DiscoverySpace software (Robertson et al.,
2007) to identify all tags in the LongSAGE libraries that mapped
to a RefSeq transcript at position 3 or lower and were signifi-
cantly differentially expressed (at a confidence level of 95%) in
pairwise comparisons of the four LongSAGE libraries (Tables
S2–S5). In terms of cell surface markers, both CD29 and PROCR
(endothelial protein C receptor/EPCR) transcripts were found to
be upregulated in the bipotent CFC-enriched fraction. Expres-
sion of the CD29 gene is also upregulated in mouse mammary
stem cells (Shackleton et al., 2006), and increased expression
of PROCR/EPCR is a feature of hematopoietic and hair-follicle
stem cells (Balazs et al., 2006; Blanpain et al., 2004).
From the Affymetrix data, we identified differentially expressed
transcripts using a 1.5-fold cut-off and a p value %0.05 (Tables
S6–S9). To determine whether and how these differences might
correlate with functionally distinct cell types, Volcano plots were
used to identify the most differentially expressed transcripts from
each of the six possible pairwise comparisons, and a K-mean
clustering algorithm was then performed using centered Pear-
son correlation analysis. From this analysis we obtained four
gene expression clusters (Figure 2C). One of these clusters
was exclusive for the luminal-restricted CFCs and included tran-
scripts for both CD24 and ALDH5A1. CD24 is of interest because
it was initially found to be expressed on certain breast cancer
stem cells (Al-Hajj et al., 2003) and subsequently was found to
be highly expressed on purified human luminal epithelial cells
(Jones et al., 2004), on luminal-restricted CFCs in the mouse (As-
selin-Labat et al., 2007; Stingl et al., 2006a), and at lower level on
mouse mammary stem cells (Asselin-Labat et al., 2007; Shackle-
ton et al., 2006; Stingl et al., 2006a). Consistent with these
Table 1. Transcriptome Profiling of Highly Purified Normal Human Mammary Epithelial Progenitor Populations
Transcriptome Profiling
Colony Phenotype LongSAGE Library Affymetrix Analysis
Sample Purified Fraction Luminal Mixed Myoepithelial
Amount of
RNA Used Total Tags
Total Tag
Types
Amount of
RNA Used
1 Bipotent CFC-enriched 0.5 51 0 10 ng 205,221 40,640 50 pg
Luminal-restricted CFC-enriched 28 6.5 0 10 ng 203,246 35,236 50 pg
Mature myoepithelial ND ND ND 10 ng 210,834 38,336 50 pg
Mature luminal ND ND ND 10 ng 201,037 32,632 50 pg
2 Bipotent CFC-enriched 3.3 46 1 50 pg
Luminal-restricted CFC-enriched 40.5 0.5 0 50 pg
Mature myoepithelial ND ND ND 50 pg
Mature luminal ND ND ND 50 pg
3 Bipotent CFC-enriched 5 37 2 50 pg
Luminal-restricted CFC-enriched 33 5 0 50 pg
Mature myoepithelial ND ND ND 50 pg
Mature luminal ND ND ND 50 pg
Bipotent and luminal-restricted progenitor-enriched fractions were isolated from three different reduction mammoplasty samples. The progenitor
frequency in each subfraction was calculated as described in the Experimental Procedures. ND, not determined.
Cell Stem Cell
Transcript Analysis of Human Mammary Cell Subsets
Cell Stem Cell 3, 109–118, July 2008 ª2008 Elsevier Inc. 111

findings, we determined from FACS analyses that virtually all of
the CD24
+
EPCAM
+
cells coexpressed MUC1 and/or CD133
and accounted for approximately 82%–96% of the (MUC1/
CD133)
+
subset (Figure S1). High ALDH5A1 expression in the lu-
minal CFC-enriched fraction is of interest in view of the recent re-
port that primitive normal and malignant human mammary cells
exhibit uniquely elevated ALDH activity (Ginestier et al., 2007;
Liu et al., 2008). In contrast to the luminal-restricted CFCs, a dis-
tinct cluster for the bipotent CFCs was not resolved. Instead,
a cluster that included both the bipotent CFCs and the differenti-
ated myoepithelial cells was identified. This cluster contained
Figure 2. Relatedness of Undifferentiated,
Lineage-Restricted, and Mature Normal
Human Mammary Epithelial Subpopula-
tions Revealed by Global Transcriptome
Comparisons
(A) Relatedness of the LongSAGE libraries deter-
mined by Pearson correlation analysis.
(B) Relatedness of the transcriptomes identified by
Affymetrix analyses of all four subsets of mam-
mary cells (three samples each) derived from an
unsupervised hierarchical clustering analysis.
(C) Cluster analysis performed by applying a
K-mean clustering algorithm of the most highly dif-
ferentially expressed transcripts identified in the
Affymetrix data comparisons of gene expression
in the four subsets of human mammary cells stud-
ied here.
(D) Distribution according to the Biological Pro-
cess (D) and Molecular Function (E) GO categories
at level 5 of transcripts found to be overexpressed
in the bipotent as compared to the luminal-re-
stricted CFC-enriched subpopulations (70 genes,
red bars) and in the luminal-restricted CFC-
enriched cells as compared to the bipotent CFC-
enriched subpopulations (108 genes, blue bars)
from both the LongSAGE and Affymetrix data
comparisons based on their assignment.
transcripts for THY1/CD90, which was
used to isolate these populations, and
also for both JAGGED1 (JAG1) and the
NOTCH4 receptor (Figure 2C).
A total of 178 genes were identified
as differentially expressed in both the
LongSAGE and Affymetrix comparisons
of the fractions enriched in bipotent ver-
sus luminal-restricted CFCs, 70 of which
were more highly expressed in the bipo-
tent CFC fraction and the other 108 of
which were more highly expressed in
the luminal CFC fraction (Table S10). Fur-
ther categorization of these differentially
expressed genes according to their
Gene Ontology Biological Process and
Molecular Function groups revealed that
the bipotent CFC molecular signature
contained a preponderance of tran-
scripts for proteins involved in cell migra-
tion, shape control, and morphogenesis,
and in calcium, integrin, and insulin signaling (Figures 2D and
2E). In contrast, the luminal-restricted CFC molecular signature
was found to contain a preponderance of transcripts for proteins
involved in transcription, RNA metabolism, phosphotransferase
activity, and retinoic acid synthesis (ALDH1A3) and signaling
(CRAPB2 and RBP7). Shared differentially expressed transcripts
for other comparisons are listed in Tables S11–S13, and their
numbers are shown diagrammatically in Figure S2. Together,
these data suggest at least 332 genes that undergo changes in
expression during normal human mammary epithelial cell differ-
entiation. Interestingly, CD44, a cell surface adhesion protein
Cell Stem Cell
Transcript Analysis of Human Mammary Cell Subsets
112 Cell Stem Cell 3, 109–118, July 2008 ª2008 Elsevier Inc.

whose expression has been reported to allow selective enrich-
ment of breast tumor-initiating cells (Al-Hajj et al., 2003), was
not found to be differentially expressed in either CFC-enriched
fraction by either transcriptome profiling method. This failure to
find CD44 expression to be a discriminating feature of any of
the subsets of EPCAM
+
cells studied here was further supported
by FACS analysis of antibody-stained preparations (Figure S1).
These showed 30%–50% of all EPCAM
+
cells are CD44
+
and
that among the CD49f
+
cells (which includes both the luminal
and bipotent CFCs subtypes, Figure 1), approximately 90% of
the cells coexpress CD44.
We then selected 32 genes for more extensive analysis of their
expression levels as determined by Q-RT-PCR. These genes in-
cluded 18 identified as differentially expressed in different mam-
mary subsets either in the LongSAGE libraries or the Affymetrix
data sets as well as 14 genes implicated in differentiation and/or
breastcancer. Transcriptlevels for each test genewere normalized
to GAPDH transcript levels in the same extract, which gave similar
results to those obtained when an alternative highly expressed
gene (RPL36) was used for the normalization (data not shown).
The results of the Q-RT-PCR assays demonstrated consistent dif-
ferentialexpression by 15 of the18 genes identified from the global
gene expression studies and 19 of all 32 surveyed (Table S14).
Included among these were the estrogen and progesterone re-
ceptor genes. These showed high progesterone receptor and low
estrogen receptor transcript levels in the bipotent CFC-enriched
fraction, which changed to low levels of progesterone receptor
transcripts and high levels of estrogen receptor transcripts in
Figure 3. Comparison of Q-RT-PCR Ex-
pression Data for LIF and NOTCH Pathway
Genes Expressed in Different Subsets of
Human Mammary Cells
(A–D) The level of expression of each transcript
type was assessed by normalizing its level to the
level of GAPDH transcripts in the same extract
and then to the levels of the same test transcript
in the bipotent CFC-enriched fraction (set = 1, dot-
ted lines). The error bars represent the standard
error of the mean.
the luminal-restricted CFC-enriched frac-
tion (Table S14). Q-RT-PCR also con-
firmed the LongSAGE data indicating LIF
expression to be upregulated (9-fold) in
the fraction enriched in luminal-restricted
CFCs as compared to the fraction en-
riched in bipotent CFCs and then similarly
reduced (9-fold) in the differentiated lu-
minal cells but remaining unchanged in
the mature myoepithelial cell fraction (Fig-
ure 3). Q-RT-PCR data further demon-
strated transcripts for both components
of the LIF receptor (LIFR and GP130) to
be present in all four cell fractions exam-
ined, albeit at constant levels.
Another regulatory gene that showed
a marked but progressive change in ex-
pression during luminal cell differentiation
(30-fold overall decrease from bipotent progenitors to mature lu-
minal cells) was the Iroquois-class homeobox gene IRX4 (Table
S14). IRX4 has been associated with the development of the
lungs and heart (Bruneau et al., 2001; van Tuyl et al., 2006). Inter-
estingly, transcripts for IRX2, another Iroquois-class homeobox
gene, were previously found to be preferentially expressed by lu-
minal and not basal epithelial cells in the mouse mammary gland
(Lewis et al., 1999).
All profiling methods also showed that transcripts for the
NOTCH ligand JAGGED2 were highest in the bipotent CFC-en-
riched fraction as compared to the luminal-restricted CFC-en-
riched fraction, with the opposite scenario for the NOTCH signal-
ing receptor, NOTCH3 (Table S10 and Figure 3). Expression of
NOTCH1 and NOTCH2 was also found to increase during the
process of luminal differentiation, although the changes in tran-
script levels for these genes were more modest. NOTCH4 ex-
pression showed the opposite pattern; transcripts levels were
relatively high in the bipotent CFC-enriched fraction and then de-
creased more than 50-fold during luminal differentiation and ap-
proximately 2-fold during myoepithelial cell differentiation. The
Q-RT-PCR data also showed higher (10-fold) expression of the
NOTCH receptor ligand genes DLL1,JAGGED1, and JAGGED 2
in the bipotent CFC-enriched fraction as compared to the lumi-
nal-restricted CFC-enriched fraction (Figure 3 and Table S14). In-
terestingly, transcripts for HES1,HES6, and HEY1, known target
genes of NOTCH, were found to be upregulated in parallel with
the upregulated expression of the NOTCH1,-2, and -3 receptors
in the luminal-restricted CFC-enriched fraction (Figure 3).
Cell Stem Cell
Transcript Analysis of Human Mammary Cell Subsets
Cell Stem Cell 3, 109–118, July 2008 ª2008 Elsevier Inc. 113

The CFC Assay Recapitulates Human Mammary
Cell Differentiation
As a first use of these gene expression data we asked the extent
to which the differentiation programs obtained in the in vitro CFC
assay replicate those executed largely in vivo. Accordingly,
we analyzed the levels of expression of human LIF,GAS6,
JAGGED2, and NOTCH3 in the predominantly myoepithelial
and luminal cells present after 8 days in cultures initiated with
EPCAM
+
cells enriched in either bipotent or luminal-restricted
CFCs, respectively, and compared these with the directly iso-
lated mature EPCAM
+
myoepithelial and luminal cells (that had
been in culture for a maximum of 3 days). The four genes se-
lected were chosen because they had shown large differences
between the CFC-enriched fractions and the two populations
of mature mammary cells (Table S14). To facilitate comparison
of lineage-specific differences in the expression of these four
genes in differently derived cells from multiple samples, we ex-
amined the ratio of the relative transcript levels in the two mature
cell types isolated from each source (Figure 4A). As can be seen
the relative levels of expression of the four selected genes in the
two sets of culture-derived ‘‘mature’’ cells show a pattern of
change that is similar to that characteristic of the in vivo genera-
tion of their mature counterparts.
NOTCH Signaling Regulates the Restriction of Bipotent
Mammary Cells to the Luminal Pathway
Given the gene profiling data suggesting that bipotent CFCs un-
dergo marked changes in NOTCH receptor expression when
they generate luminal-restricted CFCs, we hypothesized that
NOTCH expression may have an important functional role in
this lineage restriction process. To investigate this possibility,
we cultured purified bipotent and luminal CFCs in the presence
Figure 4. Changes in Gene Expression Seen
during the Differentiation of CFCs In Vitro
Are Similar to Those Evident in 3-Day Cul-
tured Cells and Evidence of a Role for Notch
in Normal Human Mammary Epithelial Cell
Commitment to the Luminal Lineage
(A) Transcript levels of six genes in 8-day colonies
produced by the luminal-restricted CFCs were
compared to the levels of the same transcripts
in 8-day colonies produced by the bipotent
CFCs (in vitro ratio). These ratios are compared
to the transcript levels in the mature luminal
[EPCAM
+
CD49f

(MUC1/CD133)
+
CD10

THY1

]or
myoepithelial [EPCAM
+
CD49f

MUC1

CD133

(CD10/THY1)
+
] cells isolated from 3-day cultured
mammoplasty samples (n = 3).
(B) Bipotent and luminal-restricted CFC-enriched
fractions were purified from six different 3-day
precultured mammoplasty samples and then as-
sayed for CFC with (10 mM) or without (+ DMSO)
DAPT. Colony counts were performed 8 days later.
(C) Representative FACS profiles of the EPCAM
+
cells present in the 8-day colonies generated in
(B). Staining with antibodies against EPCAM al-
lowed human mammary cells to be discriminated
from contaminating mouse fibroblast feeder cells.
Mean fluorescence intensities (MFI) were calcu-
lated using FlowJo software (Ashland, OR). The
arrow indicates the decreased expression of the
luminal cell markers (MUC1 and/or CD133) when
bipotent CFCs were cultured in the presence of
DAPT as compared to the DMSO control.
(D) A representative FACS profile of the cells pres-
ent in 8-day colonies generated from bipotent
CFCs infected with a lenti-dnMAML-GFP or a con-
trol lenti-GFPvirus (from one of two experiments) af-
ter staining the harvested cells as in (C). The mean
fluorescence (MFI) values show the decreased ex-
pression of MUC1 and/or CD133 in the GFP
+
cells.
(E) RepresentativeFACS profile of the cells present
in 8-day colonies generated from bipotent CFCs or
luminal-restricted CFCs infected with either Lenti-
shN3 or control Lenti-NS virus (from one of two ex-
periments) after staining the harvested cells as in
(C). The mean fluorescence (MFI) values show the
decreased expression of MUC1 and/or CD133 in
the GFP
+
cells.
The error bars represent the standard error of the
mean.
Cell Stem Cell
Transcript Analysis of Human Mammary Cell Subsets
114 Cell Stem Cell 3, 109–118, July 2008 ª2008 Elsevier Inc.

or absence of a g-secretase inhibitor, N-[N-(3, 5-difluorophen-
acetyl)-L-alanyl]-S-phenylglycine t-butyl ester (DAPT, 10 mM)
and then measured the effect on the numbers and types of col-
onies obtained. To ensure that any effects on initial cell attach-
ment were avoided, the DAPT (or vehicle control) was not added
until after the first 16 hr of culture and then every other day there-
after. We found no effect of this treatment on the size or total
number of colonies obtained in assays of either the bipotent or
luminal CFC-enriched fractions (Figure 4B and data not shown).
However, DAPT consistently and significantly (p = 0.0006,
Student’s t test) reduced the number of colonies that contained
a detectable luminal component in the assays of the bipotent
CFC-enriched fraction (Figure 4B). At the same time, the number
of apparently ‘‘pure’’ myoepithelial colonies was proportionately
increased (p = 0.005, Student’s t test). In contrast, DAPT had no
effect on the ability of the luminal-restricted CFCs to form colo-
nies of differentiated luminal cells (Figure 4B), in spite of an
equivalent extent of DAPT-mediated suppression of NOTCH
signaling as shown by a reduction in HEY1 transcripts in cultures
initiated with either purified bipotent or luminal-restricted CFCs
(2- and 6-fold, and 2- and 2-fold, respectively, in two experi-
ments). The differential effects of DAPT on bipotent and lumi-
nal-restricted CFCs were confirmed by flow cytometric analysis
of the immunophenotype of the cells present in 8-day colonies
derived from purified progenitors. These analyses showed a spe-
cific loss of (MUC1/AC133)
+
cells exclusively among the progeny
produced by the DAPT-treated bipotent CFCs and no change in
output of (CD10/THY1)
+
cells (Figure 4C).
As a second approach to investigating the requirement of bi-
potent cells for NOTCH signaling to undergo restriction to the
luminal pathway, we infected suspensions of purified bipotent
CFCs with a lentivirus expressing a dominant-negative (dn)
form of the human mastermind-like-1 gene (MAML) fused to
GFP (Lenti-dnMAML) or a Lenti-GFP control virus and then as-
sessed the phenotype of the GFP
+
cells harvested from bulk
CFC assays 8 days later (Figure 4D). MAML is a coactivator of
NOTCH signaling, and the dn cDNA used here was previously
shown to suppress NOTCH signaling (Maillard et al., 2004;
Weng et al., 2003). As shown in Figure 4D, forced expression
of the dnMAML also decreased the output of (MUC1/AC133)
+
cells from the bipotent progenitors.
We next asked whether this restriction step was directed by
specific activation of the NOTCH3 receptor using a similar strat-
egy in which purified bipotent CFCs were first transduced with
a GFP lentivirus also expressing a short hairpin (sh) NOTCH3
RNA or a nonsilencing control RNA. As shown in Figure 4E,
knockdown of the expression of NOTCH3 receptor in the bipotent
CFCs (90% and 85% in the two experiments undertaken; data not
shown) reduced their ability to generate (MUC1/AC133)
+
luminal
cells. Thus, activation of NOTCH3 appears to be critical for the
restriction of most bipotent progenitors to the luminal pathway,
and other NOTCH receptors cannot substitute for this activity.
DISCUSSION
The development of specific assays for cells at distinct stages
of differentiation in parallel with strategies for their selective
isolation at high purities constitute critical steps in delineating
the molecular mechanisms that regulate normal cell populations
and are likely targets for malignant transformation. In applying
this approach to the normal human mammary gland, we demon-
strate here the robustness of the 2D colony assay conditions
now available to distinguish undifferentiated (bipotent) and line-
age-restricted mammary progenitors of human origin. Second,
we describe a cell separation method that allows each of these
progenitor types to be isolated routinely at purities of more
than 30% and with less than 5% contamination with each other.
Attainment of such purities depends, however, on the use of
a starting population that has been cultured for an initial period
of 3 days to selectively enrich for CFCs while also allowing cells
damaged by the initial enzymatic dispersion process to be re-
moved. The CFC purification scheme we describe also takes ad-
vantage of the fact that the majority of bipotent progenitors share
a number of features with mature myoepithelial cells and lack
features of mature luminal cells, whereas the converse is true
for luminal-restricted progenitors (Russo and Russo, 2004; Stingl
et al., 1998; Tsai et al., 1996), in addition to the previous obser-
vation that primitive mammary cells with significant proliferative
potential express high levels of the a
6
-integrin (CD49f) in contrast
to the majority of mature luminal and myoepithelial mammary
cells (Stingl et al., 2001, 2006b).
Our first application of this technology enabled us to refute the
notion that dual expression of cytokeratins 14 and 18 is a feature
of undifferentiated or even primitive luminal-restricted progeni-
tors in the normal human mammary gland. A second application
was to generate molecular signatures for each of the four differ-
ent cell populations from their global gene expression profiles.
These allowed us to demonstrate that the differentiated cells
produced in the colony assays resemble their counterparts iso-
lated directly after only 3 days in vitro. However, the fact that
all of the gene expression data were from cells that had been
in culture for 3 days would predict that there would be some dif-
ferences with the corresponding cell types present in vivo. The
importance of culture conditions in replicating in vivo transfor-
mation events in human mammary cells has recently been docu-
mented in an elegant study by Ince et al. (2007).
The gene expression data presented here have also offered
some new insights into the molecular characteristics of normal
undifferentiated human mammary cells and their immediate lu-
minal-restricted progeny. For example, we found that bipotent
CFCs contain transcripts for the progesterone receptor but not
the estrogen receptor, with an opposite picture for the luminal-
restricted CFCs. Both of these patterns thus differ from the dou-
ble-negative profile described for normal mouse mammary stem
cells (Asselin-Labat et al., 2006; Shackleton et al., 2006) or any of
the four categories of human breast cancer defined by expres-
sion profiling methods (Hu et al., 2006). We also examined the
expression of transcripts for LIF and its receptor elements.
This cytokine has been previously implicated in controlling apo-
ptosis in the mammary gland (Kritikou et al., 2003) and is down-
regulated in some breast cancers (Hu et al., 2004). In addition,
Dontu et al. (2003) noted that LIF expression was elevated in cul-
tured human mammospheres, which represent mixed popula-
tions but contain elevated frequencies of primitive mammary
cells. The purity of the discrete populations obtained here now
demonstrates that the expression of LIF is markedly upregulated
specifically at the time of progenitor commitment to the luminal
differentiation pathway, at which time transcripts for the receptor
Cell Stem Cell
Transcript Analysis of Human Mammary Cell Subsets
Cell Stem Cell 3, 109–118, July 2008 ª2008 Elsevier Inc. 115

are also present. It is thereforeinviting to speculate that these cells
may be the most important physiological targets for paracrine or
autocrine effects of LIF in the normal mammary epithelium.
A similar stage-specific change in gene expression emerged
from a survey of transcripts for genes associated with the evo-
lutionary conserved NOTCH pathway. This pathway is widely
involved in cell-fate decisions (Fuchs and Raghavan, 2002;
Osborne and Minter, 2007) and has been shown to have a role
in mammary gland development and transformation (Callahan
and Egan, 2004; Farnie et al., 2007; Leong et al., 2007; Pece
et al., 2004; Shi and Harris, 2006). Upon binding of their ligands,
the NOTCH receptors (NOTCH1–4) undergo proteolytic cleav-
age, thereby releasing the NOTCH intracellular domain (NICD),
which then moves into the nucleus and activates genes such
as HES1 and HES6 (Callahan and Egan, 2004; Osborne and
Minter, 2007). Notch4 was first identified as an oncogene (Int3)
in mice by MMTV insertional activation (Raafat et al., 2004).
More recently, higher levels of NOTCH4 transcripts were found
in extracts of whole human mammospheres than in mature hu-
man mammary cells, and addition of soluble ligand to these cul-
tures increased the generation of CFCs in the mammospheres
(Dontu et al., 2004). We now show that NOTCH4 gene expres-
sion is highest in undifferentiated human clonogenic mammary
progenitors and is then markedly downregulated when these
cells become committed to the luminal lineage but before they
have lost their proliferative activity. We also found an opposite
pattern to hold for NOTCH3 and HES6, and to a lesser extent
for NOTCH1 and NOTCH2. This is interesting because the
NOTCH4 receptor possesses a shorter EGF-like repeat se-
quence in its extracellular domain than NOTCH3 and also lacks
cytokine response sequences in its intracellular domain (Allens-
pach et al., 2002; Bigas et al., 1998). Importantly, we also found
that two different methods of nonspecifically blocking NOTCH
signaling in bipotent progenitors (by exposing the cells to a
g-secretase inhibitor or by forced expression in them of a dn
form of MAML) selectively prevented them from generating
mature luminal progeny without affecting their ability to prolifer-
ate and generate mature myoepithelial cells. Nevertheless, the
same treatment had no effect on the production of differentiated
luminal cells from already committed luminal progenitors. Inter-
estingly, knockdown of NOTCH3 expression was sufficient to
produce this same result, suggesting that NOTCH1 and -2,
which appear to be coexpressed at similar levels in these cells,
do not share this unique commitment-requiring function of
NOTCH3 (Figure 5). The recent demonstration of an increased
production during alveogenesis of basal cells in Rbp-j or Pofut1
null mice (Buono et al., 2006) is consistent with this model, as is
the original observation that WAP-Int3 (=Notch4) transgenic
mice display a block in mammary cell differentiation.
The present analysis thusillustrates a number of key changes in
signaling molecules and points to a specificand previously unrec-
ognized role of NOTCH in the process of restriction of bipotent
progenitors to the luminal lineage. They also provide a compre-
hensive descriptionof the gene expression profiles of distinct sub-
sets of normal human mammary epithelial cells that establish an
important new framework for further interrogation of pathways in-
volved in the regulation of normal human mammary cell behavior.
EXPERIMENTAL PROCEDURES
Preparation and Isolation of Mammary Cell Subsets
Discarded reduction mammoplasty tissue was obtained with appropria te con-
sent; the cells were processed, frozen, and thawed; and single-cell suspen-
sions were prepared as described (Stingl et al., 2005). The cells were then cul-
tured for 3 days on top of irradiated 3T3 mouse fibroblasts (8 310
3
per cm
2
)in
EpiCult-B media supplemented with 5% fetal calf serum (FCS) (both from
StemCell Technologies Inc.), and the adherent cells were resuspended using
trypsin. EPCAM
+
cells (92.3% ± 4.7% EPCAM
+
, n = 7) were isolated immuno-
magnetically using the Human EPCAM positive selection kit (StemCell Tech-
nologies). Subsequently, cells were stained with anti-MUC1 (1:100 dilution,
StemCell Technologies), anti-CD133 (AC133, 1:100 dilution, Miltenyi Biotech),
anti-a6 integrin (CD49f) conjugated to fluorescein isothiocyanate (FITC) (1:40
dilution, Becton Dickenson PharMingen), anti-CD10 (CALLA, 1:10 dilution,
StemCell Technologies) conjugated to R-phycoerythrin (PE), and anti-CD90
(THY-1 = 5E10, 1:125 dilution, from Dr. P. Lansdorp, Terry Fox Laboratory,
Vancouver, BC, Canada). A goat anti-mouse antibody conjugated to allophy-
cocyanin (APC, 1:500 dilution, PharMingen) was used to detect cells express-
ing either MUC1 and/or CD133. IgG antibodies directly conjugated to FITC or
PE were used as isotype controls. To distinguish between live and dead cells,
propidium iodide (PI, Sigma) was added at 1 mg/ml to each sample (82% ± 4%
PI

cells in 12 experiments). Cells were sorted on a FACS Vantage SE using
gates that excluded 99.9% of events present in negatively stained control sam-
ples. Events with very high forward and side light scatter profiles were also ex-
cluded to improve sort efficiency. This was estimated to be 80%–98% as mea-
sured by sorting 10
5
MUC1
+
cells into a tube, and viable cell count was obtained
by a hemocytometer. To obtain RNA extracts for gene expression studies, al-
iquots of cells were sorted directly into Trizol reagent (Invitrogen, http://www.
invitrogen.com/) and DNase treated, and the quality of the RNA samples was
examined with the Agilent 2100 Bioanalyzer (Agilent Technologies).
Lentiviral Transduction of Isolated Progenitors
Human dnMAML-GFP(Maillardet al., 2004) was obtainedfrom Dr. A. Weng (Terry
Fox Laboratory, Vancouver, BC, Canada) and cloned into the KA391 lentiviral
vectorand virus-containingsupernatantsgenerated as previously described(Im-
ren et al., 2004). Bipotent and luminal-restricted CFC-enriched populationswere
infectedin 100 ml of EpiCult-Bgrowth media containing5% FCS, 5 310
5
lentiviral
particles(eitherLenti-dnMAML-GFP or Lenti-GFP,orLenti-shN3-GFP [Open Bio-
systems] or the Lenti-shNS-GFP [Open Biosystems]), and 5 mg/ml of protamine
sulfate for 4 hr at 37C. Subsequently, cells were washed three times in 2%
FCS in Hank’s solution and platedat clonal densities in CFC assays (see below).
CFC Assays
CFC assays were performed as described (Stingl et al., 1998, 2001) with the
modification that the plates were first precoated with type I collagen by adding
a solution of 70 mg/ml (Vitrogen 100, Collagen Biomaterials) for 1 hr at 37C
Figure 5. Model of the Proposed Role of NOTCH in Regulating the
Differentiation of Normal Human Mammary Progenitors
The initial commitment process is uniquely dependent on NOTCH3 signaling,
whereas subsequent execution of the luminal differentiation program pro-
ceeds independently.
Cell Stem Cell
Transcript Analysis of Human Mammary Cell Subsets
116 Cell Stem Cell 3, 109–118, July 2008 ª2008 Elsevier Inc.

followed by washing with PBS to remove unpolymerized collagen. After 8 days
the cultures were fixed with methanol acetone (1:1 ratio) and stained with
Wright-Giemsa (Sigma), and colonies containing R50 cells were scored and
typed using a microscope.
Construction and Analysis of LongSAGE Libraries
RNA from each isolated subpopulation was converted to cDNA and then am-
plified as described (Zhao et al., 2007). LongSAGE libraries were constructed
and sequenced using standard protocols (Saha et al., 2002). The relatedness
of the libraries was determined by calculating Pearson correlation values, and
the results were displayed as a tree diagram using Phylip software (http://
evolution.genetics.washington.edu/phylip.html). The complete data set has
been uploaded onto the GEO website (accession number GSE11395).
Microarray Analyses
RNA from each sample was exponentially amplified as described (Iscove et al.,
2002) except that the total number of PCR cycles was restricted to 42 and ex-
ecuted in a single reaction. From each sample, 50 pg aliquots were amplified in
each of ten replicate 20 ml reactions, each yielding 2–3 mg of amplified product .
The amplified cDNA preparations were then purified using the High Pure PCR
Product purification kit (Roche Applied Science), end-labeled with Biotin-N6
ddATP (StemCore Laboratories, OHRI, Ottawa, Ontario, Canada), and hybrid-
ized to Affymetrix human X3P GeneChip arrays according to the manufac-
turer’s protocol (Tietjen et al., 2003). The complete data set has been uploaded
onto the GEO website (accession number GSE11395).
Q-RT-PCR Analyses
cDNA was prepared from 1 ng of RNA using the SuperScript III reverse tran-
scriptase enzyme (Invitrogen) according to the manufacturer’s protocol and
subjected to real-time PCR (7500 Sequence Detection System, Applied Bio-
systems) using gene-specific primers for each transcript analyzed. To quantify
the relative expression of transcripts from the mature luminal and myoepithelial
cells generated in vitro, RNA was extracted from the pooled 8-day progeny of
cells from purified bipotent and luminal-restricted CFC-enriched fractions cul-
tured at 50 cells per plate and pooled for extraction. Relative levels of expres-
sion of each test transcript were calculated by normalizing to the level of
GAPDH transcripts in the same extract, and to compare results in particular
subsets isolated from different samples, values were normalized to those mea-
sured in the bipotent CFC-enriched fractions.
Statistical Analysis
Differentially expressed transcripts from different LongSAGE libraries were iden-
tified using a 95% statistical cut-off as determined by DiscoverySpace software
(Robertson et al., 2007). Transcripts represented by tags expressed only once
in any library (singletons) were excluded from further analysis. Differentially
expressed transcripts from the Affymetrix hybridizations were analyzed using
ArrayAssist software (http://Stratagene.com/software solutions) and R/Bio-
Conductor (Gentleman et al.,2004). The implemented GC-RMA algorithm, which
accountsfor probecomposition,was used to subtractthe backgroundand deter-
minethe correctedprobe intensityfor each probeset cluster.Transcriptsshowing
>1.5-folddifferences with p values%0.05 (after Bonferroni-Holmescorrectionfor
multiple comparisons) in a Student’s t test across replicates were deemed to be
significant. Student’s t test was used for pairwise comparisons to determinesta-
tistical significance. Volcano plots were generated using ArrayAssist software to
identifythe most differentially expressed transcripts. K-mean clustering was car-
riedout using Cluster3.01software(Eisen et al.,1998) using centered correlations.
ACCESSION NUMBERS
All raw data are available online under Gene Expression Omnibus (GEO)
accession number GSE11395.
SUPPLEMENTAL DATA
The Supplemental Data include two figures and 14 tables and can be found
with this article online at http://www.cellstemcell.com/cgi/content/full/3/1/
109/DC1/.
ACKNOWLEDGMENTS
The authors acknowledge excellent technical contributions from Darcy Wilkin-
son and Margaret Hale and the staff of the Flow Cytometry Facility of the Terry
Fox Laboratory, which is funded in part by the Michael Smith Foundation for
Health Research. Mammoplasty tissue was obtained with the assistance of
Drs. Jane Sproul, Peter Lennox, Nancy Van Laeken, and Richard Warren.
This project was funded by grants from Genome BC/Genome Canada, the
Stem Cell Network, The BC/Yukon Division of the Canadian Breast Cancer
Foundation (CBCF), and the Canadian Institute of Health Research (CIHR), in-
cluding the CIHR Centre for Molecular Pathology at the BC Cancer Agency.
A.R. was funded by CBCF and CIHR Fellowships. Y.Z. was funded by a Leuke-
mia Research Foundation of Canada Fellowship, and J.S. was funded by
a CBCF Fellowship and an Industrial Fellowship cofunded by the Natural Sci-
ences and Engineering Research Council of Canada and StemCell Technolo-
gies, Inc.
Received: November 5, 2007
Revised: April 9, 2008
Accepted: May 15, 2008
Published: July 2, 2008
REFERENCES
Al-Hajj, M., Wicha, M.S., Benito-Hernandez, A., Morrison, S.J., and Clarke,
M.F. (2003). Prospective identification of tumorigenic breast cancer cells.
Proc. Natl. Acad. Sci. USA 100, 3983–3988.
Allenspach, E.J., Maillard, I., Aster, J.C., and Pear, W.S. (2002). Notch signal-
ing in cancer. Cancer Biol. Ther. 1, 466–476.
Asselin-Labat, M.L., Shackleton, M., Stingl, J., Vaillant, F., Forrest, N.C.,
Eaves, C.J., Visvader, J.E., and Lindeman, G.J. (2006). Steroid hormone re-
ceptor status of mouse mammary stem cells. J. Natl. Cancer Inst. 98, 1011–
1014.
Asselin-Labat, M.L., Sutherland, K.D., Barker, H., Thomas, R., Shackleton, M.,
Forrest, N.C., Hartley, L., Robb, L., Grosveld, F.G., van der Wees, J., et al.
(2007). Gata-3 is an essential regulator of mammary-gland morphogenesis
and luminal-cell differentiation. Nat. Cell Biol. 9, 201–209.
Balazs, A.B., Fabian, A.J., Esmon, C.T., and Mulligan, R.C. (2006). Endothelial
protein C receptor (CD201) explicitly identifies hematopoietic stem cells in
murine bone marrow. Blood 107, 2317–2321.
Bigas, A., Martin, D.I., and Milner, L.A. (1998). Notch1 and Notch2 inhibit
myeloid differentiation in response to different cytokines. Mol. Cell. Biol. 18,
2324–2333.
Blanpain, C., Lowry, W.E., Geoghegan, A., Polak, L., and Fuchs, E. (2004).
Self-renewal, multipotency, and the existence of two cell populations within
an epithelial stem cell niche. Cell 118, 635–648.
Bruneau, B.G., Bao, Z.Z., Fatkin, D., Xavier-Neto, J., Georgakopoulos, D., Ma-
guire, C.T., Berul, C.I., Kass, D.A., Kuroski-de Bold, M.L., de Bold, A.J., et al.
(2001). Cardiomyopathy in Irx4-deficient mice is preceded by abnormal ven-
tricular gene expression. Mol. Cell. Biol. 21, 1730–1736.
Buono, K.D., Robinson, G.W., Martin, C., Shi, S., Stanley, P., Tanigaki, K.,
Honjo, T., and Hennighausen, L. (2006). The canonical Notch/RBP-J signaling
pathway controls the balance of cell lineages in mammary epithelium during
pregnancy. Dev. Biol. 293, 565–580.
Callahan, R., and Egan, S.E. (2004). Notch signaling in mammary developme nt
and oncogenesis. J. Mammary Gland Biol. Neoplasia 9, 145–163.
Dontu, G., Abdallah, W.M., Foley, J.M., Jackson, K.W., Clarke, M.F., Kawa-
mura, M.J., and Wicha, M.S. (2003). In vitro propagation and transcriptional
profiling of human mammary stem/progenitor cells. Genes Dev. 17, 1253–
1270.
Dontu, G., Jackson, K.W., McNicholas, E., Kawamura, M.J., Abdallah, W.M.,
and Wicha, M.S. (2004). Role of Notch signaling in cell-fate determination of
human mammary stem/progenitor cells. Breast Cancer Res. 6, R605–R615.
Cell Stem Cell
Transcript Analysis of Human Mammary Cell Subsets
Cell Stem Cell 3, 109–118, July 2008 ª2008 Elsevier Inc. 117

Eisen, M.B., Spellman, P.T., Brown, P.O., and Botstein, D. (1998). Cluster anal-
ysis and display of genome-wide expression patterns. Proc. Natl. Acad. Sci.
USA 95, 14863–14868.
Farnie, G., Clarke, R.B., Spence, K., Pinnock, N., Brennan, K., Anderson, N.G.,
and Bundred, N.J. (2007). Novel cell culture technique for primary ductal car-
cinoma in situ: role of Notch and epidermal growth factor receptor signaling
pathways. J. Natl. Cancer Inst. 99, 616–627.
Fuchs, E., and Raghavan, S. (2002). Getting under the skin of epidermal mor-
phogenesis. Nat. Rev. Genet. 3, 199–209.
Gentleman, R.C., Carey, V.J., Bates, D.M., Bolstad, B., Dettling, M., Dudoit, S.,
Ellis, B., Gautier, L., Ge, Y., Gentry, J., et al. (2004). Bioconductor: open soft-
ware development for computational biology and bioinformatics. Genome
Biol. 5, R80.
Ginestier, C., Hur, M.H., Charafe-Jauffret, E., Monville, F., Dutcher, J., Brown,
M., Jacquemier, J., Viens, P., Kleer, C.G., Liu, S., et al. (2007). ALDH1 is
a marker of normal and malignant human mammary stem cells and a predictor
of poor clinical outcome. Cell Stem Cell 1, 555–567.
Howlin, J., McBryan, J., and Martin, F. (2006). Pubertal mammary gland devel-
opment: insights from mouse models. J. Mammary Gland Biol. Neoplasia 11,
283–297.
Hu, Y., Sun, H., Drake, J., Kittrell, F., Abba, M.C., Deng, L., Gaddis, S., Sahin,
A., Baggerly, K., Medina, D., and Aldaz, C.M. (2004). From mice to humans:
identification of commonly deregulated genes in mammary cancer via com-
parative SAGE studies. Cancer Res. 64, 7748–7755.
Hu, Z., Fan, C., Oh, D.S., Marron, J.S., He, X., Qaqish, B.F., Livasy, C., Carey,
L.A., Reynolds, E., Dressler, L., et al. (2006). The molecular portraits of breast
tumors are conserved across microarray platforms. BMC Genomics 7, 96.
Imren, S., Fabry, M.E., Westerman, K.A., Pawliuk, R., Tang, P., Rosten, P.M.,
Nagel, R.L., Leboulch, P., Eaves, C.J., and Humphries, R.K. (2004). High-level
beta-globin expression and preferred intragenic integration after lentiviral
transduction of human cord blood stem cells. J. Clin. Invest. 114, 953–962.
Ince, T.A., Richardson, A.L., Bell, G.W., Saitoh, M., Godar, S., Karnoub, A.E.,
Iglehart, J.D., and Weinberg, R.A. (2007). Transformation of different human
breast epithelial cell types leads to distinct tumor phenotypes. Cancer Cell
12, 160–170.
Iscove, N.N., Barbara, M., Gu, M., Gibson, M., Modi, C., and Winegarden, N.
(2002). Representation is faithfully preserved in global cDNA amplified expo-
nentially from sub-picogram quantities of mRNA. Nat. Biotechnol. 20, 940–
943.
Jones, C., Mackay, A., Grigoriadis, A., Cossu, A., Reis-Filho, J.S., Fulford, L.,
Dexter, T., Davies, S., Bulmer, K., Ford, E., et al. (2004). Expression profiling of
purified normal human luminal and myoepithelial breast cells: identification of
novel prognostic markers for breast cancer. Cancer Res. 64, 3037–3045.
Kordon, E.C., and Smith, G.H. (1998). An entire functional mammary
gland may comprise the progeny from a single cell. Development 125, 1921–
1930.
Kritikou, E.A., Sharkey, A., Abell, K., Came, P.J., Anderson, E., Clarkson, R.W.,
and Watson, C.J. (2003). A dual, non-redundant, role for LIF as a regulator of
development and STAT3-mediated cell death in mammary gland. Develop-
ment 130, 3459–3468.
Leong, K.G., Niessen, K., Kulic, I., Raouf, A., Eaves, C., Pollet, I., and Karsan,
A. (2007). Jagged1-mediated Notch activation induces epithelial-to-mesen-
chymal transition through Slug-induced repression of E-cadherin. J. Exp.
Med. 204, 2935–2948.
Lewis, M.T., Ross, S., Strickland, P.A., Snyder, C.J., and Daniel, C.W. (1999).
Regulated expression patterns of IRX-2, an Iroquois-class homeobox gene, in
the human breast. Cell Tissue Res. 296, 549–554.
Liu, S., Ginestier, C., Charafe-Jauffret, E., Foco, H., Kleer, C.G., Merajver, S.D.,
Dontu, G., and Wicha, M.S. (2008). BRCA1 regulates human mammary stem/
progenitor cell fate. Proc. Natl. Acad. Sci. USA 105, 1680–1685.
Maillard, I., Weng, A.P., Carpenter, A.C., Rodriguez, C.G., Sai, H., Xu, L., All-
man, D., Aster, J.C., and Pear, W.S. (2004). Mastermind critically regulates
Notch-mediated lymphoid cell fate decisions. Blood 104, 1696–1702.
Osborne, B.A., and Minter, L.M. (2007). Notch signalling during peripheral
T-cell activation and differentiation. Nat. Rev. Immunol. 7, 64–75.
Pece, S., Serresi, M., Santolini, E., Capra, M., Hulleman, E., Galimberti, V., Zur-
rida, S., Maisonneuve, P., Viale, G., and Di Fiore, P.P. (2004). Loss of negative
regulation by Numb over Notch is relevant to human breast carcinogenesis. J.
Cell Biol. 167, 215–221.
Raafat, A., Bargo, S., Anver, M.R., and Callahan, R. (2004). Mammary develop-
ment and tumorigenesis in mice expressing a truncated human Notch4/Int3
intracellular domain (h-Int3sh). Oncogene 23, 9401–9407.
Robertson, N., Oveisi-Fordorei, M., Zuyderduyn, S.D., Varhol, R.J., Fjell, C.,
Marra, M., Jones, S., and Siddiqui, A. (2007). DiscoverySpace: an interactive
data analysis application. Genome Biol. 8, R6.
Russo, J., and Russo, I.H. (2004). Development of the human breast. Maturitas
49, 2–15.
Saha, S., Sparks, A.B., Rago, C., Akmaev, V., Wang, C.J., Vogelstein, B., Kin-
zler, K.W., and Velculescu, V.E. (2002). Using the transcriptome to annotate
the genome. Nat. Biotechnol. 20, 508–512.
Shackleton, M., Vaillant, F., Simpson, K.J., Stingl, J., Smyth, G.K., Asselin-
Labat, M.L., Wu, L., Lindeman, G.J., and Visvader, J.E. (2006). Generation of
a functional mammary gland from a single stem cell. Nature 439, 84–88.
Shi, W., and Harris, A.L. (2006). Notch signaling in breast cancer and tumor an-
giogenesis: cross-talk and therapeutic potentials. J. Mammary Gland Biol.
Neoplasia 11, 41–52.
Smalley, M.J., Titley, J., Paterson, H., Perusinghe, N., Clarke, C., and O’Hare,
M.J. (1999). Differentiation of separated mouse mammary luminal epithelial
and myoepithelial cells cultured on EHS matrix analyzed by indirect immuno-
fluorescence of cytoskeletal antigens. J. Histochem. Cytochem. 47, 1513–
1524.
Stingl, J., Eaves, C.J., Kuusk, U., and Emerman, J.T. (1998). Phenotypic and
functional characterization in vitro of a multipotent epithelial cell present in
the normal adult human breast. Differentiation 63, 201–213.
Stingl, J., Eaves, C.J., Zandieh, I., and Emerman, J.T. (2001). Characterization
of bipotent mammary epithelial progenitor cells in normal adult human breast
tissue. Breast Cancer Res. Treat. 67, 93–109.
Stingl, J., Raouf, A., Emerman, J.T., and Eaves, C.J. (2005). Epithelial progen-
itors in the normal human mammary gland. J. Mammary Gland Biol. Neoplasia
10, 49–59.
Stingl, J., Eirew, P., Ricketson, I., Shackleton, M., Vaillant, F., Choi, D., Li, H.I.,
and Eaves, C.J. (2006a). Purification and unique properties of mammary epi-
thelial stem cells. Nature 439, 993–997.
Stingl, J., Raouf, A., Eirew, P., and Eaves, C.J. (2006b). Deciphering the mam-
mary epithelial cell hierarchy. Cell Cycle 5, 1519–1522.
Tietjen, I., Rihel, J.M., Cao, Y., Koentges, G., Zakhary, L., and Dulac, C. (2003).
Single-cell transcriptional analysis of neuronal progenitors. Neuron 38, 161–
175.
Tsai, Y.C., Lu, Y., Nichols, P.W., Zlotnikov, G., Jones, P.A., and Smith, H.S.
(1996). Contiguous patches of normal human mammary epithelium derived
from a single stem cell: implications for breast carcinogenesis. Cancer Res.
56, 402–404.
van Tuyl, M., Liu, J., Groenman, F., Ridsdale, R., Han, R.N., Venkatesh, V., Tib-
boel, D., and Post, M. (2006). Iroquois genes influence proximo-distal morpho-
genesis during rat lung development. Am. J. Physiol. Lung Cell. Mol. Physiol.
290, L777–L789.
Welm, B.E., Tepera, S.B., Venezia, T., Graubert, T.A., Rosen, J.M., and Good-
ell, M.A. (2002). Sca-1(pos) cells in the mouse mammary gland represent an
enriched progenitor cell population. Dev. Biol. 245, 42–56.
Weng, A.P., Nam, Y., Wolfe, M.S., Pear, W.S., Griffin, J.D., Blacklow, S.C., and
Aster, J.C. (2003). Growth suppression of pre-T acute lymphoblastic leukemia
cells by inhibition of notch signaling. Mol. Cell. Biol. 23, 655–664.
Zhao, Y., Raouf, A., Kent, D., Khattra, J., Delaney, A., Schnerch, A., Asano, J.,
McDonald, H., Chan, C., Jones, S., et al. (2007). A modified polymerase
chain reaction-long serial analysis of gene expression protocol identifies novel
transcripts in human CD34+ bone marrow cells. Stem Cells 25, 1681–1689.
Cell Stem Cell
Transcript Analysis of Human Mammary Cell Subsets
118 Cell Stem Cell 3, 109–118, July 2008 ª2008 Elsevier Inc.