Kim et al.

Breast Cancer Research (2018) 20:16

https://doi.org/10.1186/s13058-018-0944-8

RESEARCH ARTICLE Open Access

The hypoxic tumor microenvironment in

vivo selects the cancer stem cell fate of
breast cancer cells
Hoon Kim, Qun Lin, Peter M. Glazer and Zhong Yun*

Abstract
Background: Tumor hypoxia is an independent prognostic factor associated with poor patient survival. Emerging
evidence suggests that hypoxia can potentially maintain or enhance the stem cell phenotype of both normal stem
cells and cancer cells. However, it remains to be determined whether cell fate is regulated in vivo by the hypoxic
tumor microenvironment (TME).
Methods: We established a hypoxia-sensing xenograft model to identify hypoxic tumor cell in vivo primarily using
human breast cancer cell lines MDA-MB-231 and MCF7. Hypoxic tumor cells were identified in situ by fluorescence
of green fluorescence protein. They were further isolated from xenografts, purified and sorted by flow cytometry for
detailed analysis of their stem cell characteristics.
Results: We have found that hypoxic tumor cells freshly isolated from xenografts contain increased subpopulations
of tumor cells with cancer stem cell (CSC)-like characteristics. The CSC characteristics of the hypoxic tumor cells are
further enhanced upon re-implantation in vivo, whereas secondary xenografts derived from the non-hypoxic tumor
cells remain similar to the primary xenografts. Interestingly, the phenotypes exhibited by the hypoxic tumor cells
are stable and remain distinctively different from those of the non-hypoxic tumor cells isolated from the same
tumor mass even when they are maintained under the same ambient culture conditions. Mechanistically, the PI3K/
AKT pathway is strongly potentiated in the hypoxic tumor cells and is required to maintain the CSC-like phenotype.
Importantly, the differential cell fates between hypoxic and non-hypoxic tumor cells are only found in tumor cells
isolated from the hypoxic TME in vivo and are not seen in tumor cells treated by hypoxia in vitro alone.
Conclusions: These previously unknown observations suggest that the hypoxic TME may promote malignant
progression and therapy resistance by coordinating induction, selection and/or preferential maintenance of the
CSC-like phenotype in tumor cells.
Keywords: AKT, Breast cancer cell, Cancer stem cell, Cell fate, Hypoxia, PI3K, Tumor microenvironment, Xenograft

Background perspective of cancer stem cells (CSCs). CSCs are defined

Tumor hypoxia, a hallmark of tumor microenvironment as a distinct subpopulation of tumor cells with unlimited
(TME), is an independent prognostic factor for advanced self-renewal and tumor-initiating potential [9, 10]. Current
disease progression and poor patient survival [1–3]. As an studies have shown that CSCs are likely to be the major
emerging paradigm, hypoxia has the potential to regulate cause of therapy resistance and tumor recurrence [9, 11].
cell fate and differentiation, especially the cellular proper- Similar to normal stem cells, maintenance and differenti-
ties associated with stem cells [4–8]. These observations ation of CSCs are also subject to regulation by both gen-
provide important insights into the role of TME in the etic and environmental or niche factors [10, 12–14].
regulation of malignant tumor progression from the Studies have shown that the hypoxic TME is strongly
associated with increased cancer cell “stemness”. In clinical
* Correspondence: zhong.yun@yale.edu
tumor specimens, tumor cells located in hypoxic regions
Department of Therapeutic Radiology, Yale University School of Medicine, P. appear poorly differentiated and express stem-cell-
O. Box 208040, New Haven, CT 06520-8040, USA

© The Author(s). 2018 Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0
International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and
reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to
the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver
(http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.
Kim et al. Breast Cancer Research (2018) 20:16 Page 2 of 15

associated genes [15, 16]. Poorly differentiated pancreatic sorted EGFP+ cells were then incubated under normal
cancer cells show strong nuclear accumulation of hypoxia- tissue culture conditions and EGFP− cells were sorted by
inducible factor 1α (HIF-1α) protein [17]. Higher HIF-α flow cytometry. Three rounds of positive and negative
protein levels are also found in the stemcell-like tumor cells selections were used to establish the hypoxia-sensing cell
in neuroblastomas [18, 19] or gliomas [20]. When main- lines. All cell lines are authenticated using the short tan-
tained under in vitro hypoxic conditions, tumor cells ex- dem repeat (STR) method.
hibit enhanced clonogenicity [21–23] and can activate an
embryonic stem cell-like transcription program in a HIF-- Xenografts
dependent manner [24]. Recent studies using three- MDA-MB-231/HRE-EGFP and MCF7/HRE-EGFP cells
dimensional culture systems have reported hypoxia- were mixed with growth factor-depleted Matrigel (1:1)
associated cell-type plasticity [25] and CSC heterogeneity and then injected either orthotopically into the mam-
[26]. These data suggest that tumor hypoxia is likely to have mary fat pads or ectopically under the skin in the upper
a direct and strong impact on the fate of cancer cells, back of female athymic mice (6–7 weeks of age) at a
especially CSCs. However, it remains to be clearly deter- concentration of 1–2 × 106 cells per injection site. For
mined whether the CSC fate is differentially regulated by tumor takeandgrowth curve analysis, 10 × 103 cells were
the hypoxic TME in vivo. injected per site. When the tumor size reached approxi-
In this study, we have developed a hypoxia-sensing mately 800 mm3, tumors were excised for isolation of
xenograft model using human breast cancer cell lines tumor cells.
and have made several new discoveries with regard to
cell fate determination by the hypoxic TME. First, we Tumor cell dissociation and purification
have obtained direct evidence that the CSC-like subpop- Excised xenograft tumors were minced and incubated at
ulation is significantly enriched in the enhanced green 37 °C for 2 h in a dissociation medium containing 10%
fluorescent protein (EGFP)+ hypoxic population. Second, fetal calf serum, 0.5 U/ml dispase (07913, STEMCELL
the CSC characteristics continue to be enhanced upon Technologies), 5 mg/ml Collagenase Type IV (CLS-4,
re-implantation of the EGFP+ hypoxic cells, suggesting Worthington Biochemical), and 100 U/ml penicillin
the CSC-like properties displayed by the hypoxic tumor streptomycin. Red blood cells were removed by incuba-
cells are stable and continue to be enriched. Consist- tion with a red blood cell lysis buffer containing 0.8%
ently, the phenotypes exhibited by the EGFP+ hypoxic NH4Cl for 10 min on ice. Host mouse cells were de-
tumor cells remain stably distinct from those of the pleted using a Mouse Cell Depletion Kit (130–104-694,
EGFP− non-hypoxic tumor cells isolated from the same Miltenyi Biotec.)
tumor mass even when they are maintained under the
same ambient culture condition. Third, the PI3K/AKT Analysis of cell surface marker by flow cytometry
pathway is strongly potentiated in the EGFP+ hypoxic All antibodies used for flow cytometry were purchased
tumor cells and is required to maintain the CSC-like from eBiosciences (ThermoScientific). For CD24 and
phenotype. Interestingly, the cell fate differences be- CD44 double stain, anti-CD24-PE-Cyanine 7 (1:20, #25–
tween hypoxic and non-hypoxic tumor cells are found 0247-42) and anti-CD44-eFlour 450 (1:50, #48–0441-82)
only in the tumor cells isolated from the hypoxic TME were used with mouse IgGκ-PE-Cyanine 7 (1:20, #25–
in vivo, but not in tumor cells treated by hypoxia in vitro 4714-42) and rat IgG2b-eFlour 450 as IgG controls
alone. These previously unrecognized observations (1:50, #48–4031). For CD44 and CD49f double stain,
suggest that tumor hypoxia may promote malignant pro- anti-CD44-eFlour 450 (1:50) and anti-CD49f-PE-Cya-
gression and therapy resistance by differential selection nine 7 (1:20, #25–0495-82) were used with rat IgG2b-
for and/or maintenance of the CSC-like tumor cells in eFlour 450 and mouse IgGκ-PE-Cyanine 7 as IgG
the hypoxic TME. controls. For human epithelial cell adhesion molecule
(EPCAM) stain, anti-CD326-PE-Cyanine 7 (1:20, #25–
Methods 9326-41) was used with mouse IgGκ-PE-Cyanine 7 as
Generation of the hypoxia-sensing cell lines IgG control. The BD LSR II flow cytometer was used for
MDA-MB-231 and MCF7 human breast cancer cells fluorescence-activated cell sorting (FACS). The instru-
(American Type Culture Collection (ATCC)) were trans- ments were calibrated daily. FACS data were analyzed
fected with 5HRE/GFP plasmid (a gift from Martin using the FlowJo™ software.
Brown and Thomas Foster, Addgene plasmid # 46926)
[27] and selected with 500 μg/ml of G418. For positive Flow-assisted cell sorting
selection, cells were incubated at 1% O2 in a hypoxia Human tumor cells were isolated from xenografts and
chamber (Invivo 400) for 24 h and EGFP+ cells were separated from the host mouse cells. After filtration
collected by flow cytometry. For negative selection, the through a 40-μm cell strainer, tumor cells were then
Kim et al. Breast Cancer Research (2018) 20:16 Page 3 of 15

sorted into the hypoxic (EGFP+, approximately the top Tumor cell invasion assay
30%) and non-hypoxic (EGFP−, approximately the bot- Invasion of tumor cells was determined using trans-well
tom 30%) fractions using the BD FACSAria™ II. chambers with polycarbonate membranes of 8-μm pore
size according to the manufacturer’s instructions. The
Side-population analysis dividing membranes were pre-coated with 50 μl 15%
Cells were suspended at 1 × 106 cells/ml in pre-warmed Matrigel solution in serum-free medium. Freshly sorted
Dulbecco’s modified Eagle’s medium (DMEM, Life Tech- tumor cells isolated from xenograft tumors were sus-
nologies) with 2% fetal calf serum (FCS) and 10 mM pended in a culture medium containing 0.5% FBS and
HEPES buffer. Hoechst 33342 (ThermoScientific) was loaded (10 × 103 cells per well) into the upper chamber
added at a final concentration of 5 μg/ml in the presence of the trans-well plates. The lower chambers were filled
or absence of verapamil hydrochloride (50 μM, Sigma- with a culture medium containing 10% fetal bovine
Aldrich). After incubation at 37 °C for 90 min, cells were serum (FBS). After incubation for 48 h, cells migrating
washed with ice-cold FACS buffer containing 3% FCS and to the underside of the membrane were stained with
10 mM HEPES, resuspended in the FACS buffer, and Diff-Quik™ Stain Set (B4132-1A, Siemens) and cells
filtered through a 40-μm cell strainer. Propidium iodide remaining on the upper surface of the membrane were
was added at a final concentration of 2 μg/ml before FACS removed with a cotton swab. The number of migrated
to gate viable cells. FACS analyses were done using the cells on each membrane was counted in 20 random
LSRII (BD Bioscience). The Hoechst 33342 dye was ex- fields at ×200 magnification under a light microscope.
cited at 350 nm and its fluorescence was dual-wavelength
analyzed (blue, 440 nm; red, 650–670 nm). Wound healing assay
Tumor cells were allowed to grow to 100% confluence
Detection of hypoxic regions in xenografts in 60-mm dishes. Wound tracks were created by manu-
Hypoxyprobe™-1 (pimonidazole HCl, Hypoxyprobe, Inc) ally scraping the cell monolayer with a P1000 pipet tip.
was injected intraperitoneally 2 h before xenograft After incubation for 12 and 24 h, the gap filling was
tumor isolation at a dose of 60 mg/kg bodyweight. examined under a light microscope. Wound healing is
Tumors were excised, fixed in formalin, and cryopreserved quantitatively shown as percent gap fill.
in optimum cutting temperature compound (OCT).
Tumor sections (7 μm thick) were incubated with anti- Western blot analysis
pimonidazole rabbit IgG1 antibody (PAB2627AP, Hypox- Antibodies to the following antigens were used for
yprobe, Inc.) and then a fluorescently labeled anti-rabbit western blots: phospho-AKT-S473 (#4060, 1:5000, Cell
IgG antibody. Nuclei were counter stained with Hoechst Signaling), total AKT (#4691, 1:5000, Cell Signaling),
33342. The fluorescence stains were examined under a phospho-ERK1/2-T202/T204 (#4377, 1:10,000, Cell
fluorescence microscope (Zeiss ZX10 Imager M2). Signaling), total ERK1/2 (#9102, 1:10,000, Cell Signal-
ing), phospho-PI3Kp85 (#4228, 1:1000, Cell Signaling),
Clonogenic assay phospho-mTOR-S2448 (#5536, 1:1000, Cell Signaling),
The clonogenic assay is based on our previously pub- HIF-1α (#14179, 1:1000, Cell Signaling), HIF-2α
lished protocols [22, 28]. Briefly, tumor cells were plated (#NB100–122, 1:1000, Novus Biologicals), β-actin
at 600 cells/well for MDA-MB-231 cells and at 1000 (#A5316, 1:10,000, Sigma Aldrich), and vinculin
cells/well for MCF7 cells in 6-well plates and incubated (ab18058, 1:10,000, Abcam). Protein bands were visual-
for 10 to 14 days. Colonies were stained with crystal vio- ized using ECL substrates (ThermoFisher Scientific,
let. Plating efficiency (PE) = numbers of colonies (≥50 #34080) and imaged using the Kodak X-OMAT 2000A.
cells/colony) divided by numbers of cells plated × 100%.
Real-time quantitative reverse transcription PCR (RT-qPCR)
Tumor sphere formation assay Total cellular RNA was isolated with the TRizol reagent
The tumor sphere formation assay has been described in (Invitrogen). First-strand cDNA was synthesized using
our previous publication [22]. Briefly, tumor cells were the High Capacity cDNA Reverse Transcription Kit
incubated as a single-cell suspension in a tumor sphere (Thermo Fisher Scientific). RT-qPCR was performed on
medium containing DMEM-F12, 3:1 (Invitrogen), 2% B27 StepOne Plus (Applied Biosystems) using iTaq Universal
supplement, 40 ng/ml basic fibroblast growth factor SYBR Green Supermix (Bio-Rad) under the following
(bFGF), and 20 ng/ml epidermal growth factor (EGF), and conditions: initiation at 95 °C × 30 s, 40 cycles at 95 °C
plated into tissue culture dishes pre-coated with polyhy- × 15 s, and 60 °C × 60 s. The housekeeping gene HPRT
droxyethylmethacrylate (polyHEMA, Sigma-Aldrich). After was used as a control for normalization. Specificity of
incubation for 4 to 6 days, the numbers of tumor spheres the primers was confirmed by a single peak on the dis-
were counted under a microscope. sociation curve. For CD44, TGCCGCTTTGCAGGTG
Kim et al. Breast Cancer Research (2018) 20:16 Page 4 of 15

TAT (forward) and GGCCTCCGTCCGAGAGA (re- female athymic nu/nu mice to examine the distribution of
verse); for CD24, AAACAACAACTGGAACTTCAAGT the EGFP+ tumor cells in the tumor microenvironment.
AACTC (forward) and GGTGGTGGCATTAGTTG The hypoxic regions in xenografts were independently
GATTT (reverse); for GLUT1, GATTGGCTCCTTCT identified using the bioreductive compound pimonidazole
CTGTGG (forward) and TCAAAGGACTTGCCCAG HCl (Hypoxyprobe-1). As shown in Fig. 1a, the EGFP+
TTT (reverse); for LOX1, GAACACAGGAACATCATC MDA-MB-231 cells were primarily localized in the
CTG (forward) and ACGCAGCACAGTCCTTGGTT (re- pimonidazole-positive regions in both orthotopic and sub-
verse); for HPRT, TATGGCGACCCGCAGCCCT (for- cutaneous xenografts. Similar results were found in
ward) and CATCTCGAGCAAGACGTTCAG (reverse). MCF7-derived xenografts (Additional file 1).
We next purified the human tumor cells from xeno-
Global gene expression analysis grafts (Additional file 3) and separated the EGFP+ and
Total cellular RNA was isolated with RNeasy® Mini Kit EGFP− cells by flow cytometry (Fig. 1c). Consistent with
(74,104, Qiagen) and treated with DNase I for 10 min. the literature [32], the basal-like MDA-MB-231 cell-
Global gene expression profile analysis was done using derived xenografts contained a larger hypoxic population
Affymetrix HTA microarrays at the Yale Center for than the xenografts derived from the luminal-like MCF7
Genomic Analysis. The hybridization patterns and signal cells (Fig. 1c). Using global gene expression analysis, we
intensities were analyzed and interpreted using the Affy- found that a panel of commonly observed hypoxia-
metrix GeneChip Expression Console. inducible genes [33–35], including BNIP3L, EGLN3,
LOXL4 and P4HA1, were significantly upregulated in the
Statistics EGFP+ MDA-MB-231 tumor cells, compared to the
Microarray data were analyzed using analysis of variance EGFP− cells (Fig. 1d). Collectively, these data demon-
(ANOVA) by Yale Center for Analytical Sciences (YCAS). strate that these HRE-EGFP-modified breast cancer cell
Two-group comparisons were analyzed using the two- lines can serve as a reliable hypoxia-sensing model to
tailed Student’s t test (GraphPad Prizm 7). A significant identify hypoxic tumor cells in xenografts.
difference was declared if the p value was < 0.05. The cell fate or the state of differentiation of tumor
cells is thought to be under the influence of the tumor
Results microenvironment [36–38]. In particular, the impact of
The cancer stem cell-like population of tumor cells is hypoxia on the regulation of CSC-associated phenotypes
enriched in the hypoxic TME in vivo and functions has been reported [7, 39–41]. However,
The cell fate regulation of tumor cells by the TME is crit- definitive in vivo evidence has remained elusive. We hy-
ical for understanding clonal heterogeneity and malignant pothesized that the hypoxic tumor microenvironment
progression but remains to be fully investigated. We have can alter tumor cell fate with preference for cancer
established two hypoxia-sensing tumor models by stably stem-cell-like phenotypes. Using FACS-based single-cell
expressing a hypoxia-responsive transcription enhancer analysis, we examined the stem cell fate of freshly iso-
element (five tandem repeats of HRE or 5XHRE)-driven lated tumor cells using common cancer stem cell
destabilized d2EGFP construct [27] in human breast can- markers [42], including CD44, CD24, and CD49f for
cer cell line MDA-MB-231 and MCF7, respectively. These breast CSCs. Tumor cells isolated from the basal-like
two cell lines represent the two most common types of MDA-MB-231 cell-derived xenografts are predominantly
breast cancer, i.e. the basal type (MDA-MB-231) and the CD44+ (Fig. 2a, b). The ex vivo EGFP+ or hypoxic popu-
luminal type (MCF7). The HRE-EGFP reporter gene is lation contains an average of 8% CD24+ cells, whereas
transcriptionally activated by hypoxia-inducible transcrip- the non-hypoxic or EGFP− population contains much
tion factor (HIF)-1 and/or HIF-2 [29]. Similar strategies higher numbers of CD24+ cells at approximately 13%
have been used in other tumor models [30, 31] to identify (Fig. 2a, b), suggesting that the hypoxic TME favors
hypoxic cells. We generated the hypoxia-reporter cell lines tumor cells with the CD24−/low CSC-like characteristics.
using three rounds of negative and positive selection to Similar results are obtained from both orthotopic and
ensure low background under normoxic conditions and subcutaneous xenografts. The differential expression of
robust induction of EGFP under hypoxic conditions the CD44 and CD24 surface markers was independently
(Additional file 1). We also confirmed that HIF accumula- confirmed by qRT-PCR using the freshly sorted cells
tion occurring during hypoxia treatment is readily revers- (Additional file 3). These results are consistent with the
ible in the ambient air in the sorted ex vivo tumor cells findings that the CSCs of basal-like breast cancer cells
from xenografts generated by our hypoxia-sensing cancer tend to display the CD44+/CD24− phenotype [43, 44].
cell lines (Additional file 2, panel A). Consistent with these observations, there was a slight in-
Next, we generated both orthotopic (mammary fat crease in the side population (SP) with low retention of
pads) and ectopic (subcutaneous sites) xenografts in Hoechst 33342 in the freshly sorted hypoxic tumor cells
Kim et al. Breast Cancer Research (2018) 20:16 Page 5 of 15

Fig. 1 The hypoxia-sensing human breast cancer xenograft model. a, b MDA-MB-231 cells stably expressing the HRE-EGPF reporter gene are implanted
either orthotopically in mammary fat pads (a) or ectopically in the hind back (b) of female athymic mice. Hypoxic regions are visualized by immunostaining
of the Hypoxyprobe (red). Expression of the hypoxia reporter gene is shown by fluorescence of enhanced green fluorescent protein (EGFP). Nuclei are
counterstained with Hoechst 33342. c The hypoxic populations from the MDA-MB-231/HRE-EGFP and MCF7/HRE-EGFP xenografts, respectively, are
analyzed by fluorescence-activated cell sorting. d Microarray analysis shows that expression of a panel of commonly observed hypoxia-induced genes is
significantly upregulated in the EGFP+ cells freshly isolated from the MDA-MB-231/HRE-EGFP xenografts, compared to the EGFP− cells from the same
xenografts (n = 3; analysis of variance, p < 0.05). SSC, side scatter

(Additional file 3) although MDA-MB-231 cells generally [45]. Clinically, CD24 overexpression is significantly
contain a very small SP. associated with unfavorable outcomes in patients with lu-
Among other CSC markers, we found the expression minal A breast cancers [46] or with breast tumors of
of the CSC marker CD133 and aldehyde dehydrogenase intermediate-grade differentiation [47]. Furthermore, the
(ALDH) activity were not significantly different in CD44+/CD49f+ population is also strongly increased
MDA-MB-231 cells isolated from different tumor micro- among the hypoxic (EGFP+) MCF7 tumor cells (Fig. 2f).
environments (data not shown), suggesting the hypoxic Our data suggest that the CSC-like population of the
TME exerts specific effects on the CD44/CD24 luminal-type MCF7 cells are likely to be characterized by
pathways. Interestingly, hypoxia in vitro does not signifi- the CD44+/CD24+ phenotype, which is strongly enhanced
cantly affect the CD24+ cell population (Fig. 2c). These in the hypoxic TME. Collectively, these data provide direct
findings underscore the unique ability of hypoxic TME ex vivo evidence demonstrating enrichment of the CSC-
in vivo to enhance and/or maintain the CSC phenotype, like breast tumor cells in the hypoxic TME.
which cannot be recapitulated simply by exposure to
hypoxia in vitro. The ex vivo hypoxic tumor cells possess aggressive CSC-
In contrast to the basal-like MDA-MB-231 cell- like properties
derived tumors, the hypoxic TME had a different impact Using robust in vitro functional assays, we examined the
on the cell fate of the luminal-like MCF7-derived xeno- TME-dependent stemness of the EGFP+ hypoxic and
grafts. The non-hypoxic (EGFP−) MCF7 cells exoko vivo EGFP− non-hypoxic tumor cells. Although the tumor
had very low levels of CD44 (Fig. 2d, e). However, there sphere formation assay has been widely used to assess
was a significant increase in the numbers of the CD44+ the self-renewal potential of CSC-like cells in some tu-
cells in the hypoxic (EGFP+) MCF7 tumor cell popula- mors or tumor cell lines, MDA-MB-231 tumor cells,
tion with the predominant increase of the CD44+/CD24+ however, do not form tumor spheres. We instead used
population (Fig. 2d, e). Interestingly, it has been shown the in vitro clonogenic assay to examine the ability of an
that the CD24+ MCF7 cells exhibit higher proliferation individual tumor cells to establish a colony. We have
and stronger invasion than their CD24− counterparts found, using the freshly isolated 4T1/HRE-EGFP mouse
Kim et al. Breast Cancer Research (2018) 20:16 Page 6 of 15

Fig. 2 Cancer stem cell (CSC)-like cells are enriched in the hypoxic populations freshly isolated from xenografts. Tumor cells are enzymatically
dissociated and isolated from either the MDA-MB-231/HRE-EGFP (a-c) or MCF7/HRE-EGFP (d-f) xenografts. Stem cell characteristics are evaluated by
fluorescence-activated cell sorting (FACS) for the expression of CSC-associated surface markers CD24, CD44 and CD49f. Representative FACS plots are
shown in a, c, d and f. Quantitative population analyses are shown in b (n = 4–5; *p < 0.05, ***p < 0.001, Student’s t test) and e (n = 4; ***p < 0.001,
Student’s t test). These results are confirmed by three or more independent experiments. EGFP, enhanced green fluorescent protein; SSC, side scatter

mammary tumor cells (Additional file 4), that the clono- more readily through the Matrigel barrier (Fig. 3b). Fur-
genic potential is closely correlated with the ability to thermore, the ex vivo EGFP+ hypoxic MDA-MB-231
grow as tumor spheres. As shown in Fig. 3a, the EGFP+ tumor cells were capable of healing a wounded monolayer
hypoxic tumor cells produced significantly more colonies in cell culture much more rapidly and efficiently than their
than the EGFP− tumor cells did when plated at clonal non-hypoxic counterparts (Fig. 3c). These data demon-
densities (<1 cell/mm2). The increased self-renewal ability strate that the ex vivo EGFP+ hypoxic tumor cells possess
and clonogenic potential of the ex vivo hypoxic MDA- enhanced self-renewal and clonogenic potential and highly
MB-231 tumor cells were observed in both orthotopic and invasive characteristics.
subcutaneous xenografts, suggesting a strong association We further examined the tumorigenic potential of the
with the hypoxic TME. ex vivo hypoxic and non-hypoxic MDA-MB-231 tumor
The ex vivo hypoxic MDA-MB-231 tumor cells also cells, respectively, by implanting the freshly sorted EGFP+
demonstrated strong invasion and migration. In contrast and EGFP− tumor cells in athymic mice. As shown in
to non-hypoxic tumor cells, we found that the freshly Fig. 3d, the ex vivo EGFP+ orthotopic tumor cells
sorted EGFP+ hypoxic MDA-MB-231 tumor cells invaded formed new xenografts at a tumor-take rate of 90%
Kim et al. Breast Cancer Research (2018) 20:16 Page 7 of 15

Fig. 3 The ex vivo hypoxic tumor cells possess properties functionally associated with self-renewal and tumorigenic potentials. Tumor cells are enzymatically
dissociated and isolated from the MDA-MB-231/HRE-EGFP xenografts. After sorting into the enhanced green fluorescent protein (EGFP)+ and EGFP−
populations, tumor cells were plated for in vitro assays (a-c) or directly re-implanted in athymic mice (d). Detailed experimental conditions are
described in “Methods”. a Clonogenic potential (n = 6, ***p < 0.001, Student’s t test). b Tumor cell invasion (n = 3, *p < 0.05, Student’s t test). c Wound
healing potentials (n = 5, *p < 0.001, Student’s t test). d Tumorigenic potentials in vivo are primarily reflected by percent tumor take (the ability of
implanted cells to produce a tumor)

versus 40% for the EGFP− tumor cells. Similarly, the model (Fig. 4a) in which we re-implanted freshly sorted
tumor-take rate was 60% for the freshly isolated EGFP+ hypoxic and EGFP− non-hypoxic MDA-MB-231
EGFP+ subcutaneous tumor cells versus 20% for the tumor cells isolated from the 1st xenograft tumor (1st
EGFP− tumor cells. The secondary xenografts origin- EGFP+ and 1st EGFP−), respectively, in tumor-naive female
ating from the ex vivo EGFP+ orthotopic tumor cells nude mice. In order to maintain relative consistency of
also grew at a higher rate than did xenografts formed the tumor microenvironment, breast tumor cells isolated
from the EGFP− tumor cells (Fig. 3d). These results from orthotopic xenografts were re-implanted into the
demonstrate that the hypoxic TME can select tumor mammary fat pads while tumor cells isolated from sub-
cells with biological properties associated with aggres- cutaneous sites were re-implanted ectopically.
sive and CSC-like phenotypes, including enhanced As shown in Fig. 4b, the difference in the CD24+ popu-
self-renewal, pronounced invasiveness and robust lation is further increased between the 2nd xenografts
tumor-initiating potential. derived from the 1st sorted EGFP+ tumor cells and those
derived from 1st sorted EGFP− cells. In particular, the
The CSC-like tumor cells are further enriched in the hyp- CD24+ population increased from approximately 13%
oxic TME upon sequential implantation in vivo (Fig. 2b) to approximately 25% (Fig. 4b) when the EGFP−
The results described above led to an interesting ques- MDA-MB-231 cells were re-implanted. Although both 2nd
tion as to whether the CSC characteristics continue to xenografts maintained high levels of CD44, the xenografts
evolve upon repeated exposure to hypoxia in vivo, an derived from the EGFP+ cells had slightly but significantly
important issue germane to the understanding of TME- higher levels of CD44 than the EGFP− cell-derived xeno-
driven malignant progression. Hence, we developed a grafts (right arrow in the FACS histogram, Fig. 4c). The
Kim et al. Breast Cancer Research (2018) 20:16 Page 8 of 15

Fig. 4 The cancer stem cell (CSC)-like population is further enriched in secondary xenografts derived from the enhanced green fluorescent protein
(EGFP)+ MDA-MB-231 cells. a Generation of the secondary xenografts. b, c Surface levels of CD24 and CD44 are analyzed by fluorescence-activated cell
sorting. b Average CD24+ populations from six individual tumors (***p < 0.001, Student’s t test). c Average CD44++ (right-pointing arrow) populations from
three individual tumors (*p < 0.05, Student’s t test). d Quantitative RT-PCR analysis of expression of CD24 and CD44 genes in the EGFP+ and EGFP− cells
freshly isolated from either the 2nd or 1st xenografts (n = 3; *p < 0.05, **p < 0.01, Student’s t test). e Clonogenic growth of sorted EGFP+ and EGFP− cells
freshly isolated from the 2nd xenografts in comparison to the unsorted tumor cells from the 1st xenografts (n = 3; *p < 0.05, ***p < 0.0001, Student’s t test)

differential changes in cell surface expression of CD44 and Nonetheless, the CD44+ CSC-like cells continue to be sig-
CD24 are consistent with their differential gene expression nificantly enriched in the hypoxic (EGFP+) population of
between the two types of 2nd xenografts and between the tumor cells ex vivo from 2nd xenografts derived from ei-
2nd and 1st xenografts (Fig. 4d and Additional file 5). Func- ther EGFP+ or EGFP− MCF7 cells (Fig. 5b). Consistently,
tionally, the 2nd xenograft tumor cells originating from the the side population (SP) were also enriched in the 2nd
1st sorted EGFP+ cells displayed significantly higher clono- xenografts derived from EGFP+ MCF7 cells compared to
genic potential than the 2nd xenograft tumor cells origin- those from EGFP− cells (Fig. 5c). Functionally, EGFP+
ating from the 1st sorted EGFP− cells and the unsorted MCF7 cells ex vivo from the 2nd xenografts exhibit
tumors cells isolated from the 1st xenografts (Fig. 4e). Col- significantly higher clonogenic potentials than their EGFP−
lectively, these data suggest that the EGFP− cells isolated counterparts from the same tumor (Fig. 5d). Together,
from the non-hypoxic TME are likely to be more prone to these findings strongly suggest that selective pressures in
differentiation whereas the EGFP+ hypoxic tumor cells the hypoxic TME favor the CSC-like cell fate of breast
continue to maintain their CSC-like cell fate and their ag- cancer cells independent of the tumor grade or type.
gressive properties.
As an independent model, we established 2nd xenografts The PI3K-AKT pathway is required for maintenance of the
using EGFP+ and EGFP− tumor cells ex vivo from the 1st CD44+/CD24− CSC phenotype
MCF7 orthotopic xenografts (Fig. 5). In contrast to the Upon examination of several stem-cell-related signaling
basal-like MDA-MB-231 cells, the luminal-like MCF7 pathways, we found that PI3K/AKT pathway, a critical
cells do not experience significant changes in the CD44+ pathway involved in pro-survival and pro-stem-cell
or CD44+/CD24+ phenotype upon re-implantation of the maintenance [48–52], was preferentially activated in the
sorted hypoxic (EGFP+) or non-hypoxic (EGFP−) cells. ex vivo hypoxia-selected breast cancer cells. We found
Kim et al. Breast Cancer Research (2018) 20:16 Page 9 of 15

Fig. 5 The cancer stem cell (CSC)-like characteristics of tumor cells isolated from the secondary enhanced green fluorescent protein (EGFP)+
MCF7/HRE-EGFP xenografts. a Generation of secondary MCF7/HRE-EGFP xenografts by re-implantation of sorted EGFP+ and EGFP− cells isolated
from the primary MCF7/HRE-EGFP xenografts. Tumor cells freshly isolated from the secondary xenografts were sorted into EGFP+ and EGFP−
populations for (b) fluorescence-activated cell sorting (FACS) analysis of the CD44+/CD24+ and CD44+ populations (n = 5 for EGFP+ cells, n = 4
for EGFP− cells; ***p < 0.001, Student’s t test). c Side population (SP) of the secondary xenograft-derived tumor cells. MCF7 tumor cells were
isolated from the secondary xenografts derived from EGFP+ and EGFP- tumor cells, respectively, and expanded in vitro for three passages. Cells
were stained with Hoechst 33342 for side population analysis by FACS. Verapamil (50 μM) was used to block nuclear export of Hoechst 33342.
These results were validated in two independent experiments. d Clonogenic potential of the freshly sorted EGFP+ and EGFP− populations from
the secondary xenografts (n = 6; ****p < 0.0001, ***p < 0.001, Student’s t test)

that the 2nd xenograft tumor cells originating from the distinct and stable phenotype compared to their neigh-
1st EGFP+ MDA-MB-231 cells showed robust AKT S473 boring non-hypoxic tumor cells.
phosphorylation in response to serum stimulation, com- To further understand the stable phenotypes of the
pared to the 2nd xenograft tumor cells originating from hypoxic TME-selected tumor cells, we established
the 1st EGFP− cells (Additional file 6, panel A). Similar short-term in vitro cultures of the 1st xenograft-derived
results were obtained from the 2nd xenograft tumor cells EGFP+ and EGFP− cells and interrogated the PI3K-AKT
derived from 1st EGFP+ and EGFP− MCF7 cells (Add- pathway that is strongly implicated in stem cell regula-
itional file 6, panel B). Because the preferential AKT acti- tion and malignant progression including that of breast
vation in the ex vivo hypoxic cells was stably maintained cancers [48–52]. Following serum starvation and then
even under the ambient non-hypoxic conditions, these stimulation in vitro, phosphorylation of AKT is much
data suggest that these hypoxic breast cancer cells iso- more strongly induced in the EGFP+ MDA-MB-231 cells
lated ex vivo from xenografts may have acquired a than in the EGFP− cells (Fig. 6a). The differences in
Kim et al. Breast Cancer Research (2018) 20:16 Page 10 of 15

Fig. 6 The PI3K/AKT pathway is required for maintenance of the CD44+/CD24− cancer stem cell (CSC) phenotype. The enhanced green
fluorescent protein (EGFP)+ and EGFP− cells sorted from the 1st MDA-MB-231 xenografts underwent serum starvation overnight. After serum
stimulation, phosphorylation of AKT (a), PI3Kp85 (b), mTOR (b), and ERK1/2 (c) was analyzed using Western blots. d Serum-stimulated AKT
phosphorylation in the parental MDA-MB-231/HRE-EGFP cells under normoxia and hypoxia (1% O2). e Increase in the CD24+ population induced
by the PI3K inhibitor LY294002 (20 μM). f Quantitative RT-PCR analysis of expression of CD24 and CD44 genes in response to LY294002. These
observations are confirmed by independent experiments

mTOR phosphorylation are also apparent with constitu- MDA-MB-231 cells derived from xenografts; exposure of
tively high levels of mTOR phosphorylation in the EGFP+ MDA-MB-231 cells to hypoxia in vitro, by itself, did not
MDA-MB-231 cells but much reduced mTOR phosphor- significantly affect AKT phosphorylation in response to
ylation in the EGFP− cells (Fig. 6b). Relatively smaller serum stimulation (Fig. 6d). Under the same in vitro
changes are found in phosphorylation of the PI3Kp85 conditions, the cellular microenvironment remained the
subunit (Fig. 6b). In contrast, the serum-induced phos- same for both the xenograft-derived EGFP+ and EGFP−
phorylation of ERK1/2 is somewhat reduced in the EGFP+ cells. Furthermore, the HIF-1α and HIF-2α proteins
MDA-MB-231 cells (Fig. 6c). Importantly, enhanced AKT become destabilized ex vivo in both EGFP+ and EGFP−
phosphorylation was observed only in the ex vivo EGFP+ tumor cells under the ambient tissue culture conditions
Kim et al. Breast Cancer Research (2018) 20:16 Page 11 of 15

(Additional file 2, pannel A), suggesting that the differen- In this study, we have established hypoxia-sensing
tial activation of the PI3K-AKT pathway that persists in xenograft models using the 5xHRE-driven d2EGFP re-
cell culture after isolation from the xenografts is inde- porter gene (HRE-EGFP) [27] in breast cancer cell lines.
pendent of HIF-1/2. The differential activation of the In the xenografts derived from these breast tumor cell
PI3K/AKT pathway exhibited by the ex vivo EGFP+ and lines, the EGFP+ tumor cells are primarily localized in
EGFP− tumor cells thus clearly illustrates the phenotypic the regions that are positively stained for pimonidazole,
differences between these two ex-vivo derived populations a widely used bioreductive marker of hypoxia. These
of breast cancer cells localized in two different compart- results indicate that our HRE-EGFP-based xenograft
ments of the TME, which is consistent with their models can reliably identify tumor cells that are hypoxic
corresponding CSC-like characteristic shown in Figs. 2, 3, in the TME. Because the expression of d2EGFP (half-life
4 and 5. These stable phenotypic differences suggest that ≅ 2 h) requires stabilized HIF-1α and/or HIF-2α, these
breast cancer cells undergo clonal evolution and/or selec- models are best suited for identifying hypoxic tumor
tion in the hypoxic TME that leads to acquisition of new cells around the time of observation or analysis. Similar
and sustained clonal properties. strategies have been used in other tumor models [30, 31]
The PI3K-AKT pathway is well-recognized as a key to identify hypoxic cells. Nonetheless, we cannot rule out
mechanism for regulating maintenance, proliferation and the possibility that some tumor cells might express EGFP
survival of both normal and cancer stem cells [53–55]. due to a condition of pseudohypoxia, i.e. stabilization of
We examined whether the PI3K/AKT pathway plays an HIF-α without oxygen deficiency.
important role in the maintenance of the CSC-like pheno- As shown by our combined in vitro and in vivo data,
type of the ex vivo EGFP+ hypoxic tumor cells using our xenograft model has the potential to identify hyp-
LY294002, a specific PI3K inhibitor, to suppress AKT oxic tumor cells at approximately ≤ 10 mmHg pO2 or ≤
activation (Additional file 2, pannel B). After incubation 1.3% O2, which is well within the hypoxic range in
with LY294002 for up to 5 days in vitro, both the human breast cancers clinically observed by direct pO2
xenograft-derived EGFP+ and EGFP− MDA-MB-231 cells measurement [64]. However, identification and isolation
showed progressive increases in the CD24+ population of viable hypoxic breast cancer tumor cells in the clinics
(Fig. 6e), indicating cell differentiation. Notably, the EGFP are extremely challenging, if possible at all. Nonetheless,
+
cells showed stronger upregulation of CD24. In contrast, a comprehensive approach using both genomics and
CD44 expression was not significantly affected in either proteomics could be employed to determine whether
cell population (data not shown). The increase in the breast cancer stem cells are enriched in the hypoxic re-
CD24+ population correlated with increased CD24 gene gions of clinical human breast cancers.
expression, whereas there were relatively small changes in Different breast cancer cell lines express different sets
CD44 gene expression (Fig. 6f). Although the EGFP+ of cancer stem cell markers, which may reflect their
tumor cells displayed robust mTOR phosphorylation, the respective status of development or differentiation.
mTOR inhibitor, rapamycin, did not strongly affect the Nonetheless, a number of studies have shown that CD44
CSC-associated CD24 and CD44 markers (data not expression is preferentially associated with the undiffer-
shown). Collectively, these data suggest that the enhanced entiated or progenitor-like phenotype whereas CD24 is
AKT activation is required for maintaining the CSC-like expressed in more differentiated breast cancer cells with
phenotype exhibited by the EGFP+ breast cancer cells iso- a luminal phenotype [65, 66]. MDA-MB-231 cells main-
lated ex vivo from xenografts. tain high levels of CD44. Increased CD24 expression
could signal differentiation. Our data indeed confirm that
the CD44+/CD24− MDA-MB-231 cells exhibit stronger
Discussion stem cell characteristics than their CD44+/CD24+ counter-
The hypoxia-dependent regulation of CSC-associated phe- parts do. Therefore, we consider the CD44+/CD24− MDA-
notypes and functions has been proposed and actively MB-231 cells as being more breast cancer stemcell-like. On
investigated [7, 39–41]. A number of in vitro studies have the other hand, MCF7 cells are mainly CD44−/low/CD24+.
shown that hypoxia or hypoxia-sensing pathways play a Emergence of CD44+/CD24+ MCF7 cells, also positive for
significant role in the maintenance of the CSC phenotype another stem cell marker CD49f, could result from in-
in breast cancer cells [56–62]. Hypoxia is also implicated creased stemcell-like populations in the hypoxic tumor
in increased CSC-like populations in breast cancer microenvironment in vivo. Importantly, the ex vivo hypoxic
xenografts treated by antiangiogenic agents [63]. However, tumor cells from both breast cancer xenograft models pos-
definitive in vivo evidence has remained elusive. Herein, sess biological functions closely associated with CSC char-
our data have provided direct evidence demonstrating that acteristics, including self-renewal and clonogenic potential,
the hypoxic TME favors enrichment and/or maintenance motility and invasion, and tumorigenicity. It is noteworthy
of the CSC-like tumor cells in vivo. that the enrichment of the CSC-like population occurs in
Kim et al. Breast Cancer Research (2018) 20:16 Page 12 of 15

the hypoxic TME in vivo only and exposure to hypoxia in Serum stimulation occurs when a previously hypoxic
vitro does not significantly affect cell fate. Consistent with area becomes re-oxygenated or when a hypoxic tumor
this point of view, a comprehensive bioinformatics study cell invades locally into a non-hypoxic area or enters the
[67] has shown that the hypoxia gene signatures obtained bloodstream. The preferential activation of the PI3K/
from various in vitro cell culture models have poor prog- AKT pathway may confer significant survival and/or
nostic values when applied to large clinical datasets of growth advantages on the previously hypoxic tumor cells
breast cancer patients. In contrast, the patient tumor- during their invasion and metastasis.
derived hypoxia gene signatures have statistically significant We have further found that pharmacological inhibition
prognostic values [67]. These findings strongly suggest that of the PI3K/AKT pathway leads to strong increases in
the phenotype and cell fate of tumor cells located in the the CD24+ population of MDA-MB-231 cells, suggesting
naturally hypoxic regions are likely to be determined col- that activity of the PI3K/AKT pathway is essential for
lectively by multiple factors in the hypoxic TME in combin- the maintenance of their CSC phenotype. In comparison
ation with O2 deficiency. to the ex vivo EGFP− cells, the CSC phenotype of the
Interestingly, the CSC characteristics of the EGFP+ EGFP+ hypoxic cells is much more sensitive to inhibition
hypoxic cells, but not the EGFP− non-hypoxic cells, are of PI3K/AKT. This observation is consistent with the well--
further increased upon re-implantation. These observa- recognized role of the PI3K/AKT pathway in the mainten-
tions suggest that tumor cells localized in the hypoxic ance of both normal and cancer stem cells [53–55].
TME might have acquired relatively stable phenotypes Nonetheless, it is highly possible that other as yet un-
especially those closely associated with the CSC charac- identified stem-cell-related pathways may also be in-
teristics and that they continue to evolve upon re- volved in cell fate determination by the hypoxic
implantation. Consistent with this new concept, we have tumor microenvironment.
found that the ex vivo EGFP+ hypoxic cells exhibit func-
tionally distinct cell signaling pathways, including the Conclusion
PI3K/AKT pathway, from the EGFP− non-hypoxic cells Cancer cell stemness, especially the capacity of self-
even when they are maintained under the same ambient renewal, is essential for enabling malignant progression
culture conditions. Although AKT activation in breast and clonal evolution. In this study, we have provided dir-
cancer cell lines has also been reported under in vitro ect evidence demonstrating that the hypoxic TME in
hypoxic conditions [63], we have found in this study that vivo favors the enrichment and/or selection of the CSC-
hypoxia in vitro alone is not sufficient to alter these sig- like characteristics. Importantly, the differential pheno-
naling pathways, suggesting the new cellular phenotype types of the tumor cells ex vivo from the hypoxic and
exhibited by the ex vivo EGFP+ hypoxic cells most likely non-hypoxic TME, respectively, are relatively stable even
results from complex regulations in the hypoxic TME. when they are subsequently maintained under normoxic
Nonetheless, these data strongly suggest that the hypoxic ambient culture conditions, which suggests active clonal
TME has the potential to cause tumor cells to evolve evolution and/or selection to yield a durable phenotype
and to acquire new and stable properties that are in the hypoxic TME in vivo.
distinct from those of tumor cells localized in the non- In light of the findings that hypoxia occurs in human
hypoxic TME within the same tumor mass. Consistent breast cancer [64] and is associated with poor treatment
with these findings, it has recently been shown that the outcomes [68], the novel observations presented in this
hypoxic TME can give rise to a subpopulation of tumor study strongly suggest that the hypoxic TME promotes
cells with a dormancy phenotype that is maintained even malignant progression and confers resistance to therapy,
after dissemination [30]. at least in part, by inducing or sustaining the cancer
As we have found, the ex vivo EGFP+ hypoxic cells stem cell phenotype.
show robust AKT activation in response to serum stimu-
lation even under non-hypoxic conditions compared to
the ex vivo EGFP− non-hypoxic cells that exhibit weak Additional files
to moderate response. This unique phenotype has po-
Additional file 1: Figure S1. The hypoxia-sensing human breast cancer
tentially significant biological implications because a xenograft model. A FACS analysis of EGFP+ populations in the selected
similar situation of serum starvation-stimulation could MDA-MB-231 cells stably expressing the HRE-EGPF reporter gene after
be encountered in vivo. Hypoxic areas in solid tumors exposure to hypoxia in vitro at 1% O2. B Co-localization of the EGFP+
tumor cells with immunofluorescent stains (red) of the Hypoxyprobe in
are often poorly perfused due to compromised blood MCF7/HRE-EGFP xenografts. (TIFF 9624 kb)
flow and vascular malfunctions. Therefore, concentra- Additional file 2: Figure S2. Examination of HIF-1α and HIF-2α in sorted
tions of serum-derived growth factors and other nutri- tumor cells from the MDA-MB-231/HRE-EGFP xenografts, and inhibition of
ents are expected to be much lower in the hypoxic areas AKT phosphorylation by LY294002. (A) Dynamic regulation of HIF-1α and
HIF-2α proteins in response to hypoxia and re-oxygenation is examined in
than those in the well-perfused non-hypoxic regions.
Kim et al. Breast Cancer Research (2018) 20:16 Page 13 of 15

Western blots. These results show that the O2-dependent regulation of Consent for publication
HIF-α stability in the sorted EGFP+ and EGFP− tumor cells is normal and Not applicable.
comparable to the parental MDA-MB-231 cells. (B) The PI3K-specific inhibitor
LY294002 (20 μM) blocks AKT phosphorylation in the sorted EGFP+ and Competing interests
EGFP− MDA-MB-231 cells. (TIFF 2045 kb) The authors declare that they have no competing interests.
+
Additional file 3: Figure S3. Characterization of the sorted EGFP and
EGFP− cells freshly isolated from the MDA-MB-231/HRE-EGFP xenografts.
(A) Purification of MDA-MB-231 cells from xenografts. The xenografts Publisher’s Note
contain approximately 75% human tumor cells, based on cell surface Springer Nature remains neutral with regard to jurisdictional claims in
expression of CD326 (human EpCAM). After depletion of mouse cells, published maps and institutional affiliations.
purity of tumor cells reaches 98%. (B) Expression of CSC-related markers,
CD24 and CD44, and hypoxia-induced genes, LOX1 and GLUT1, is analyzed Received: 8 August 2017 Accepted: 19 February 2018
by qRT-PCR. EGFP+ and EGFP− cells are freshly isolated from both orthotopic
and ectopic xenografts, respectively (n = 3–5; *p < 0.05, **p < 0.01, Student’s
t test). Gene expression is not affected by tumor sites. (C) Side population
(SP) of freshly isolated MDA-MB-231 cells from orthotopic xenografts. The References
unsorted tumor cells were stained with Hoechst 33342. The entire tumor 1. Dhani N, Fyles A, Hedley D, Milosevic M. The clinical significance of hypoxia
cell populations were then gated into the EGFP+ and EGFP− subpopulations, in human cancers. Semin Nucl Med. 2015;45(2):110–21.
respectively, for side population analysis by FACS. Verapamil (50 μM) was 2. Vaupel P, Mayer A. Hypoxia in cancer: significance and impact on clinical
used to block nuclear export of Hoechst 33342. These results were validated outcome. Cancer Metastasis Rev. 2007;26(2):225–39.
in three independent experiments. (TIFF 13956 kb) 3. Hockel M, Schlenger K, Aral B, Mitze M, Schaffer U, Vaupel P. Association
Additional file 4: Figure S4. Tumor sphere formation and clonogenic between tumor hypoxia and malignant progression in advanced cancer of
growth of sorted EGFP+ and EGFP− cells freshly isolated from mouse 4T1/HRE- the uterine cervix. Cancer Res. 1996;56(19):4509–15.
EGFP allogafts. The 4T1/HRE-EGFP cell line is established using the same 4. Lin Q, Yun Z. Impact of the hypoxic tumor microenvironment on the regulation
approach as that for MDA-MB-231 and MCF7 cell lines. Allografts are generated of cancer stem cell characteristics. Cancer Biol Ther. 2010;9(12):949–56.
by injection of 4T1/HRE-EGFP tumor cells either in the mammary fat pads 5. Simon MC, Keith B. The role of oxygen availability in embryonic development
(orthotopic) or in the hind back (subcutaneous) of female athymic mice. and stem cell function. Nat Rev Mol Cell Biol. 2008;9(4):285–96.
The EGFP+ and EGFP− tumor cells are sorted by FACS from enzymatically 6. Lin Q, Lee YJ, Yun Z. Differentiation arrest by hypoxia. J Biol Chem.
dissociated tumor mass. (A) The self-renewal potential is evaluated using 2006;281(41):30678–83.
the tumor sphere formation assay (n = 6; **p < 0.01, ***p < 0.001, Student’s t 7. Yun Z, Lin Q. Hypoxia and regulation of cancer cell stemness. Adv Exp Med
test). (B) Clonogenicity is examined by plating the sorted cells at a clonal Biol. 2014;772:41–53.
density (300 cells/well in 6-well plates, n = 6; ****p < 0.0001, Student’s t test). 8. Mohyeldin A, Garzon-Muvdi T, Quinones-Hinojosa A. Oxygen in stem cell
(TIFF 1025 kb) biology: a critical component of the stem cell niche. Cell Stem Cell.
2010;7(2):150–61.
Additional file 5: Figure S5. The CSC-like characteristics of tumor cells 9. Visvader JE, Lindeman GJ. Cancer stem cells in solid tumours: accumulating
isolated from the secondary MDA-MB-231/HRE-EGFP xenografts. (A, B) evidence and unresolved questions. Nat Rev Cancer. 2008;8(10):755–68.
The secondary MDA-MB-231 xenografts are generated by re-implanting 10. Kreso A, Dick JE. Evolution of the cancer stem cell model. Cell Stem Cell.
the sorted EGFP+ and EGFP− tumor cells, respectively. Gene expression 2014;14(3):275–91.
is analyzed by qRT-PCR (n = 3; *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 11. Liu C, Lin Q, Yun Z. Cellular and molecular mechanisms underlying oxygen-
0.0001, Student’s t test). (TIFF 1391 kb) dependent radiosensitivity. Radiat Res. 2015;183(5):487–96.
Additional file 6: Figure S6. Differential activation of AKT in sorted 12. Ku SY, Rosario S, Wang Y, Mu P, Seshadri M, Goodrich ZW, Goodrich MM,
EGFP+ and EGFP− cells isolated from xenografts. The EGFP+ and EGFP− Labbe DP, Gomez EC, Wang J, et al. Rb1 and Trp53 cooperate to suppress
cells isolated ex vivo from xenografts are maintained in vitro for ≤ 5 prostate cancer lineage plasticity, metastasis, and antiandrogen resistance.
passages. After overnight serum starvation, the tumor cells are stimulated Science. 2017;355(6320):78–83.
with serum (10% FBS in culture medium). AKT phosphorylation is 13. Mu P, Zhang Z, Benelli M, Karthaus WR, Hoover E, Chen CC, Wongvipat J, Ku SY,
examined by Western blotting of whole cell extracts of tumor cells Gao D, Cao Z, et al. SOX2 promotes lineage plasticity and antiandrogen resistance
from the 2nd MDA-MB-231 (A) and MCF7/HRE-EGFP (B) xenografts, in TP53- and RB1-deficient prostate cancer. Science. 2017;355(6320):84–8.
respectively. (TIFF 1903 kb) 14. Xin T, Greco V, Myung P. Hardwiring stem cell communication through
tissue structure. Cell. 2016;164(6):1212–25.
15. Das B, Tsuchida R, Malkin D, Koren G, Baruchel S, Yeger H. Hypoxia
Acknowledgements enhances tumor stemness by increasing the invasive and tumorigenic side
The authors thank Dr Guilin Wang of Yale Center for Genomic Analysis population fraction. Stem Cells. 2008;26(7):1818–30.
(YCGA) for microarray analysis and Dr Xiaoqing Yu of Yale Center for 16. Jogi A, Ora I, Nilsson H, Lindeheim A, Makino Y, Poellinger L, Axelson H,
Analytical Sciences (YCAS) for statistical analysis of microarray data. Pahlman S. Hypoxia alters gene expression in human neuroblastoma cells
toward an immature and neural crest-like phenotype. Proc Natl Acad Sci
Funding USA. 2002;99(10):7021–6.
This work is in part supported by a grant from CT Innovations (13-SCA-YALE-05 17. Couvelard A, O'Toole D, Turley H, Leek R, Sauvanet A, Degott C, Ruszniewski
to ZY) and an NIH grant (R01 CA178254 to ZY). PGM is supported by R01 P, Belghiti J, Harris AL, Gatter K, et al. Microvascular density and hypoxia-
ES005775 from the NIH. inducible factor pathway in pancreatic endocrine tumours: negative
correlation of microvascular density and VEGF expression with tumour
Availability of data and materials progression. Br J Cancer. 2005;92(1):94–101.
The data involved in this study are available upon reasonable request. 18. Pietras A, Gisselsson D, Ora I, Noguera R, Beckman S, Navarro S, Pahlman S.
High levels of HIF-2α highlight an immature neural crest-like neuroblastoma
Authors’ contributions cell cohort located in a perivascular niche. J Pathol. 2008;214(4):482–8.
HK, QL and ZY conceived and designed the study, and acquired data. All 19. Pietras A, Hansford LM, Johnsson AS, Bridges E, Sjolund J, Gisselsson D,
authors contributed to data analysis and manuscript preparation. All authors Rehn M, Beckman S, Noguera R, Navarro S, et al. HIF-2α maintains an
read and approved the final manuscript. undifferentiated state in neural crest-like human neuroblastoma tumor-
initiating cells. Proc Natl Acad Sci USA. 2009;106(39):16805–10.
Ethics approval and consent to participate 20. Li Z, Bao S, Wu Q, Wang H, Eyler C, Sathornsumetee S, Shi Q, Cao Y, Lathia
All animal procedures were reviewed and approved by the Institutional J, McLendon RE, et al. Hypoxia-inducible factors regulate tumorigenic
Animal Care & Use Committee (IACUC) of Yale University. capacity of glioma stem cells. Cancer Cell. 2009;15(6):501–13.
Kim et al. Breast Cancer Research (2018) 20:16 Page 14 of 15

21. Desplat V, Faucher JL, Mahon FX, Dello Sbarba P, Praloran V, Ivanovic Z. and ALDH1: expression distribution within intrinsic molecular subtype. J Clin
Hypoxia modifies proliferation and differentiation of CD34(+) CML cells. Pathol. 2011;64(11):937–46.
Stem Cells. 2002;20(4):347–54. 45. Kim HJ, Kim JB, Lee KM, Shin I, Han W, Ko E, Bae JY, Noh DY. Isolation of
22. Kim Y, Lin Q, Zelterman D, Yun Z. Hypoxia-regulated delta-like 1 CD24(high) and CD24(low/−) cells from MCF-7: CD24 expression is
homologue enhances cancer cell stemness and tumorigenicity. Cancer positively related with proliferation, adhesion and invasion in MCF-7. Cancer
Res. 2009;69(24):9271–80. Lett. 2007;258(1):98–108.
23. Schmaltz C, Hardenbergh PH, Wells A, Fisher DE. Regulation of proliferation- 46. Kwon MJ, Han J, Seo JH, Song K, Jeong HM, Choi JS, Kim YJ, Lee SH, Choi
survival decisions during tumor cell hypoxia. Mol Cell Biol. 1998;18(5):2845–54. YL, Shin YK. CD24 overexpression is associated with poor prognosis in
24. Mathieu J, Zhou W, Xing Y, Sperber H, Ferreccio A, Agoston Z, Kuppusamy luminal A and triple-negative breast cancer. PLoS One. 2015;10(10):
KT, Moon RT, Ruohola-Baker H. Hypoxia-inducible factors have distinct and e0139112.
stage-specific roles during reprogramming of human cells to pluripotency. 47. Mylona E, Giannopoulou I, Fasomytakis E, Nomikos A, Magkou C, Bakarakos
Cell Stem Cell. 2014;14(5):592–605. P, Nakopoulou L. The clinicopathologic and prognostic significance of CD44
25. Lehmann S, Te Boekhorst V, Odenthal J, Bianchi R, van Helvert S, Ikenberg K, +/CD24(−/low) and CD44-/CD24+ tumor cells in invasive breast carcinomas.
Ilina O, Stoma S, Xandry J, Jiang L, et al. Hypoxia Induces a HIF-1-dependent Hum Pathol. 2008;39(7):1096–102.
transition from collective-to-amoeboid dissemination in epithelial cancer 48. Masuda H, Zhang D, Bartholomeusz C, Doihara H, Hortobagyi GN, Ueno NT.
cells. Curr Biol. 2017;27(3):392–400. Role of epidermal growth factor receptor in breast cancer. Breast Cancer
26. Hubert CG, Rivera M, Spangler LC, Wu Q, Mack SC, Prager BC, Couce M, Res Treat. 2012;136(2):331–45.
McLendon RE, Sloan AE, Rich JN. A three-dimensional organoid culture 49. Moerkens M, Zhang Y, Wester L, van de Water B, Meerman JH. Epidermal
system derived from human glioblastomas recapitulates the hypoxic growth factor receptor signalling in human breast cancer cells operates
gradients and cancer stem cell heterogeneity of tumors found in vivo. parallel to estrogen receptor alpha signalling and results in tamoxifen
Cancer Res. 2016;76(8):2465–77. insensitive proliferation. BMC Cancer. 2014;14:283.
27. Vordermark D, Shibata T, Brown JM. Green fluorescent protein is a suitable 50. Clark AS, West K, Streicher S, Dennis PA. Constitutive and inducible Akt
reporter of tumor hypoxia despite an oxygen requirement for chromophore activity promotes resistance to chemotherapy, trastuzumab, or tamoxifen in
formation. Neoplasia. 2001;3(6):527–34. breast cancer cells. Mol Cancer Ther. 2002;1(9):707–17.
28. Li Q, Lin Q, Yun Z. Hypoxia-activated cytotoxicity of benznidazole against 51. Perez-Tenorio G, Stal O. Southeast Sweden Breast Cancer G. Activation of
clonogenic tumor cells. Cancer Biol Ther. 2016;17(12):1266–73. AKT/PKB in breast cancer predicts a worse outcome among endocrine
29. Shibata T, Giaccia AJ, Brown JM. Development of a hypoxia-responsive treated patients. Br J Cancer. 2002;86(4):540–5.
vector for tumor-specific gene therapy. Gene Ther. 2000;7(6):493–8. 52. Paplomata E, O'Regan R. The PI3K/AKT/mTOR pathway in breast cancer:
30. Fluegen G, Avivar-Valderas A, Wang Y, Padgen MR, Williams JK, Nobre AR, targets, trials and biomarkers. Ther Adv Med Oncol. 2014;6(4):154–66.
Calvo V, Cheung JF, Bravo-Cordero JJ, Entenberg D, et al. Phenotypic 53. D'Angelo RC, Wicha MS. Stem cells in normal development and cancer.
heterogeneity of disseminated tumour cells is preset by primary tumour Prog Mol Biol Transl Sci. 2010;95:113–58.
hypoxic microenvironments. Nat Cell Biol. 2017;19(2):120–32. 54. Voskas D, Ling LS, Woodgett JR. Signals controlling un-differentiated states
31. Le A, et al. Tumorigenicity of hypoxic respiring cancer cells revealed by a in embryonic stem and cancer cells: role of the phosphatidylinositol 3′
hypoxia-cell cycle dual reporter. Proc Natl Acad Sci USA. 2014;111(34):12486- kinase pathway. J Cell Physiol. 2014;229(10):1312–22.
91. 55. Yu JS, Cui W. Proliferation, survival and metabolism: the role of PI3K/AKT/
32. Pettersen EO, Ebbesen P, Gieling RG, Williams KJ, Dubois L, Lambin P, Ward mTOR signalling in pluripotency and cell fate determination. Development.
C, Meehan J, Kunkler IH, Langdon SP, et al. Targeting tumour hypoxia to 2016;143(17):3050–60.
prevent cancer metastasis. From biology, biosensing and technology to 56. Carcereri de Prati A, et al. Metastatic breast cancer cells enter into dormant
drug development: the METOXIA consortium. J Enzyme Inhib Med Chem. state and express cancer stem cells phenotype under chronic hypoxia. J Cell
2015;30(5):689–721. Biochem. 2017;118(10):3237-48.
33. Buffa FM, Harris AL, West CM, Miller CJ. Large meta-analysis of multiple cancers 57. Iriondo O, Rabano M, Domenici G, Carlevaris O, Lopez-Ruiz JA, Zabalza I,
reveals a common, compact and highly prognostic hypoxia metagene. Br J Berra E, Vivanco M. Distinct breast cancer stem/progenitor cell populations
Cancer. 2010;102(2):428–35. require either HIF1alpha or loss of PHD3 to expand under hypoxic
34. Eustace A, Mani N, Span PN, Irlam JJ, Taylor J, Betts GN, Denley H, Miller CJ, conditions. Oncotarget. 2015;6(31):31721–39.
Homer JJ, Rojas AM, et al. A 26-gene hypoxia signature predicts benefit 58. Shiraishi A, Tachi K, Essid N, Tsuboi I, Nagano M, Kato T, Yamashita T, Bando
from hypoxia-modifying therapy in laryngeal cancer but not bladder cancer. H, Hara H, Ohneda O. Hypoxia promotes the phenotypic change of
Clin Cancer Res. 2013;19(17):4879–88. aldehyde dehydrogenase activity of breast cancer stem cells. Cancer Sci.
35. Sorensen BS, Knudsen A, Wittrup CF, Nielsen S, Aggerholm-Pedersen N, 2017;108(3):362–72.
Busk M, Horsman M, Hoyer M, Bouchelouche PN, Overgaard J, et al. The 59. Xiang L, Gilkes DM, Hu H, Takano N, Luo W, Lu H, Bullen JW, Samanta D,
usability of a 15-gene hypoxia classifier as a universal hypoxia profile in Liang H, Semenza GL. Hypoxia-inducible factor 1 mediates TAZ expression
various cancer cell types. Radiother Oncol. 2015;116(3):346–51. and nuclear localization to induce the breast cancer stem cell phenotype.
36. Koren S, Bentires-Alj M. Breast tumor heterogeneity: source of fitness, hurdle Oncotarget. 2014;5(24):12509–27.
for therapy. Mol Cell. 2015;60(4):537–46. 60. Zhang C, Samanta D, Lu H, Bullen JW, Zhang H, Chen I, He X, Semenza GL.
37. Plaks V, Kong N, Werb Z. The cancer stem cell niche: how essential is the niche Hypoxia induces the breast cancer stem cell phenotype by HIF-dependent
in regulating stemness of tumor cells? Cell Stem Cell. 2015;16(3):225–38. and ALKBH5-mediated m(6)A-demethylation of NANOG mRNA. Proc Natl
38. Reya T, Morrison SJ, Clarke MF, Weissman IL. Stem cells, cancer, and cancer Acad Sci USA. 2016;113(14):E2047–56.
stem cells. Nature. 2001;414(6859):105–11. 61. Zhang C, Zhi WI, Lu H, Samanta D, Chen I, Gabrielson E, Semenza GL.
39. Hjelmeland AB, Lathia JD, Sathornsumetee S, Rich JN. Twisted tango: brain Hypoxia-inducible factors regulate pluripotency factor expression by
tumor neurovascular interactions. Nat Neurosci. 2011;14(11):1375–81. ZNF217- and ALKBH5-mediated modulation of RNA methylation in breast
40. Lee KE, Simon MC. From stem cells to cancer stem cells: HIF takes the stage. cancer cells. Oncotarget. 2016;7(40):64527–42.
Curr Opin Cell Biol. 2012;24(2):232–5. 62. Zhang H, Lu H, Xiang L, Bullen JW, Zhang C, Samanta D, Gilkes DM, He J,
41. Semenza GL. Regulation of the breast cancer stem cell phenotype by Semenza GL. HIF-1 regulates CD47 expression in breast cancer cells to
hypoxia-inducible factors. Clin Sci (Lond). 2015;129(12):1037–45. promote evasion of phagocytosis and maintenance of cancer stem cells.
42. Badve S, Nakshatri H. Breast-cancer stem cells-beyond semantics. Lancet Proc Natl Acad Sci USA. 2015;112(45):E6215–23.
Oncol. 2012;13(1):e43–8. 63. Conley SJ, Gheordunescu E, Kakarala P, Newman B, Korkaya H, Heath AN,
43. Kapucuoglu N, Bozkurt KK, Baspinar S, Kocer M, Eroglu HE, Akdeniz R, Akcil M. Clouthier SG, Wicha MS. Antiangiogenic agents increase breast cancer stem
The clinicopathological and prognostic significance of CD24, CD44, CD133, cells via the generation of tumor hypoxia. Proc Natl Acad Sci USA. 2012;
ALDH1 expressions in invasive ductal carcinoma of the breast: CD44/CD24 109(8):2784–9.
expression in breast cancer. Pathol Res Pract. 2015;211(10):740–7. 64. Vaupel P, Schlenger K, Knoop C, Hockel M. Oxygenation of human tumors:
44. Ricardo S, Vieira AF, Gerhard R, Leitao D, Pinto R, Cameselle-Teijeiro JF, evaluation of tissue oxygen distribution in breast cancers by computerized
Milanezi F, Schmitt F, Paredes J. Breast cancer stem cell markers CD44, CD24 O2 tension measurements. Cancer Res. 1991;51(12):3316–22.
Kim et al. Breast Cancer Research (2018) 20:16 Page 15 of 15

65. Fillmore C, Kuperwasser C. Human breast cancer stem cell markers CD44
and CD24: enriching for cells with functional properties in mice or in man?
Breast Cancer Res. 2007;9(3):303.
66. Shipitsin M, Campbell LL, Argani P, Weremowicz S, Bloushtain-Qimron N,
Yao J, Nikolskaya T, Serebryiskaya T, Beroukhim R, Hu M, et al. Molecular
definition of breast tumor heterogeneity. Cancer Cell. 2007;11(3):259–73.
67. Starmans MH, Chu KC, Haider S, Nguyen F, Seigneuric R, Magagnin MG,
Koritzinsky M, Kasprzyk A, Boutros PC, Wouters BG, et al. The prognostic
value of temporal in vitro and in vivo derived hypoxia gene-expression
signatures in breast cancer. Radiother Oncol. 2012;102(3):436–43.
68. Vaupel P, Briest S, Hockel M. Hypoxia in breast cancer: pathogenesis,
characterization and biological/therapeutic implications. Wien Med
Wochenschr. 2002;152(13–14):334–42.

Submit your next manuscript to BioMed Central

and we will help you at every step:
• We accept pre-submission inquiries
• Our selector tool helps you to find the most relevant journal
• We provide round the clock customer support
• Convenient online submission
• Thorough peer review
• Inclusion in PubMed and all major indexing services
• Maximum visibility for your research

Submit your manuscript at

www.biomedcentral.com/submit