HIF-1 regulates CD47 expression in breast cancer cellsto promote evasion of phagocytosis and maintenanceof cancer stem cellsHuimin Zhanga,b,c, Haiquan Lub,c, Lisha Xiangb,c, John W. Bullenb,c, Chuanzhao Zhangb,c, Debangshu Samantab,c,Daniele M. Gilkesd, Jianjun Hea, and Gregg L. Semenzab,c,d,e,f,g,h,1

aDepartment of Breast Surgery, The First Affiliated Hospital of Xi’an Jiaotong University Health Science Center, Xi’an, Shaanxi 710061, China; bInstitute forCell Engineering, The Johns Hopkins University School of Medicine, Baltimore, MD 21205; cMcKusick-Nathans Institute of Genetic Medicine, The JohnsHopkins University School of Medicine, Baltimore, MD 21205; dDepartment of Oncology, The Johns Hopkins University School of Medicine, Baltimore,MD 21205; eDepartment of Pediatrics, The Johns Hopkins University School of Medicine, Baltimore, MD 21205; fDepartment of Medicine, The Johns HopkinsUniversity School of Medicine, Baltimore, MD 21205; gDepartment of Radiation Oncology, The Johns Hopkins University School of Medicine, Baltimore,MD 21205; and hDepartment of Biological Chemistry, The Johns Hopkins University School of Medicine, Baltimore, MD 21205

Contributed by Gregg L. Semenza, October 9, 2015 (sent for review September 18, 2015; reviewed by Shinae Kizaka-Kondoh and Michail V. Sitkovsky)

Increased expression of CD47 has been reported to enable cancercells to evade phagocytosis by macrophages and to promote thecancer stem cell phenotype, but the molecular mechanisms regulat-ing CD47 expression have not been determined. Here we reportthat hypoxia-inducible factor 1 (HIF-1) directly activates transcriptionof the CD47 gene in hypoxic breast cancer cells. Knockdown of HIFactivity or CD47 expression increased the phagocytosis of breastcancer cells by bonemarrow-derivedmacrophages. CD47 expressionwas increased in mammosphere cultures, which are enriched forcancer stem cells, and CD47 deficiency led to cancer stem cell de-pletion. Analysis of datasets derived from thousands of patientswith breast cancer revealed that CD47 expression was correlatedwith HIF target gene expression and with patient mortality. Thus,CD47 expression contributes to the lethal breast cancer phenotypethat is mediated by HIF-1.

antitumor immunity | immune evasion | tumor-initiating cells |tumor microenvironment | “don’t eat me” signal

The pathogenesis of breast cancer reflects not only the con-sequence of somatic mutations that dysregulate oncogenes

and tumor suppressor genes, but also the effect of the changingtumor microenvironment, particularly the development of intra-tumoral hypoxia. The median pO2 within primary breast cancers is10 mmHg (1.4% O2) compared with 65 mmHg (9.3% O2) innormal breast tissue (1). Exposure of breast cancer cells to re-duced O2 availability induces the activity of hypoxia-induciblefactors (HIFs), which are heterodimeric transcriptional activators,consisting of an O2-regulated HIF-1α, HIF-2α, or HIF-3α subunitand a constitutively expressed HIF-1β subunit, that control theexpression of hundreds of target genes (2). Increased expression ofthe HIF-1α subunit, detected by immunohistochemistry in primarybreast cancer biopsies, is associated with a significantly increasedrisk of metastasis and mortality (2).Preclinical studies in mouse models have demonstrated that

loss of HIF-1α or HIF-2α expression impairs the metastasis ofbreast cancer cells to axillary lymph nodes (3), lungs (4, 5), andbone (6, 7). Specific HIF target genes have been identified thatare induced by hypoxia in breast cancer cells and promotecritical steps in the metastatic process, including stromal cellrecruitment (8, 9), cancer cell migration (10), invasion andintravasation (11–15), margination and extravasation (5), andpremetastatic niche formation (12, 16). Increased expression ofHIF target genes in primary breast cancers is associated withincreased patient mortality (17, 18).To give rise to a primary tumor, a tumor relapse, or a meta-

static tumor, a breast cancer cell must possess two importantcharacteristics: first, the cell must avoid destruction by the im-mune system and, second, the cell must possess stem-cell–likeproperties. Hypoxia induces the breast cancer stem cell (CSC)

phenotype (19, 20) through functional and physical interactionsof HIF-1 with the coactivator TAZ (20, 21) and by HIF-dependentexpression of pluripotency factors (22). Hypoxia also induces im-mune evasion by several HIF-dependent mechanisms (23, 24).A major mechanism by which cancer cells evade the innate im-mune system is by expression of CD47, which is a cell-surfaceprotein that interacts with signal regulatory protein α (SIRPα) onthe surface of macrophages to block phagocytosis (25, 26). Ex-pression of calreticulin (CRT) on the surface of cancer cells isthe primary trigger for phagocytosis by binding to low-densitylipoprotein-related protein (LRP) on the surface of macro-phages (27). The prophagocytic signal triggered by CRT–LRPligation is counterbalanced by the antiphagocytic signal trig-gered by CD47–SIRPα ligation (28). Analysis of circulatingtumor cells isolated from the blood of patients with breastcancer revealed that CD47 expression identified a subpopula-tion of cells with the capability to generate tumor xenografts inmice (29). Here, we report that CD47 expression is induced in aHIF-dependent manner when human breast cancer cells areexposed to hypoxia. Modest inhibition of CD47 expression wassufficient to increase the phagocytosis of breast cancer cells anddecrease the number of breast CSCs. Human breast cancerdatabase analysis revealed that high CD47 expression is corre-lated with increased HIF target gene expression and decreasedpatient survival.

Significance

Uncontrolled cell proliferation and abnormal blood vessel for-mation result in regions of breast cancers that are hypoxic(deprived of oxygen). Hypoxia-inducible factors (HIFs) stimu-late the expression of genes that enable cancer cells to invadeand metastasize, leading to patient mortality. In this paper, wereport that HIFs stimulate the production of CD47, a protein onthe cell surface that enables cancer cells to avoid destruction bymacrophages. CD47 is also important for maintaining cancerstem cells, which are a small population of cells that are re-quired for the formation of primary tumors and metastases.Reduction of HIF activity or CD47 levels in breast cancer cellsled to increased killing by macrophages and depletion of can-cer stem cells.

Author contributions: H.Z. and G.L.S. designed research; H.Z., H.L., L.X., J.W.B., C.Z., D.S.,and D.M.G. performed research; J.H. contributed new reagents/analytic tools; H.Z. andG.L.S. analyzed data; and H.Z. and G.L.S. wrote the paper.

Reviewers: S.K.-K., Tokyo Institute of Technology; and M.V.S., Northeastern University.

The authors declare no conflict of interest.1To whom correspondence should be addressed. Email: gsemenza@jhmi.edu.

www.pnas.org/cgi/doi/10.1073/pnas.1520032112 PNAS | Published online October 28, 2015 | E6215–E6223

MED

ICALSC

IENCE

SPN

ASPL

US

Dow

nloa

ded

by g

uest

on

Aug

ust 3

1, 2

021

ResultsHypoxia Induces Increased CD47 Expression in a HIF-Dependent Manner.Breast cancers are classified based on their expression of the estrogenreceptor (ER), progesterone receptor (PR), and human epidermalgrowth factor receptor 2 (HER2); they are also classified according toexpression of a 50-mRNA signature (PAM50) into luminal A, lu-minal B, HER2-enriched, basal-like, and normal-like subgroups (30).We analyzed the effect of hypoxia (1% O2 for 24 h) on CD47mRNA levels in six different human breast cell lines: MCF10A

is an immortalized but nontumorigenic mammary epithelial line;MCF7 is ER+PR+ and tumorigenic but nonmetastatic; HCC1954 isHER2+ and tumorigenic but nonmetastatic; and MDA-MB-231,MDA-MB-435, and SUM159 are ER−PR−HER2− (i.e., triplenegative), tumorigenic and metastatic. Reverse transcription (RT)and quantitative real-time PCR (qPCR) assays revealed thatexpression of CD47 mRNA was significantly induced byhypoxia in MCF7, HCC1954, MDA-MB-435, and SUM159cells but not in MCF10A or MDA-MB-231 cells (Fig. 1A).

A B C

D

HCC1954MCF7

SUM159 E

HCC1954MCF7 SUM159F G H

% o

f Max

20% O2 20% O2 1% O21% O2

NTC DKD ISO

NTC DKD ISO

NTC DKD ISO

NTC DKD ISO

20% O2 1% O2

ISO

MCF7 HCC1954 SUM159

Fig. 1. Hypoxia induces CD47 expression in a HIF-dependent manner. (A) Reverse transcription (RT) and quantitative real-time PCR (qPCR) were performed todetermine CD47 mRNA levels in human breast cell lines following exposure to 20% or 1% O2 for 24 h. For each sample, the expression of CD47 mRNAwas quantified relative to 18S rRNA and then normalized to the result obtained from MCF-10A cells at 20% O2 (mean ± SEM; n = 3). Significant increase at 1% O2

compared with 20%O2: *P < 0.05, **P < 0.01, ***P < 0.001 (Student’s t test). (B and C) CD47 mRNA expression was analyzed by RT-qPCR in MCF7 (B) and HCC1954(C) subclones expressing shRNA targeting HIF-1α (sh1α), HIF-2α (sh2α), or HIF-1α and HIF-2α (DKD), or expressing a nontargeting control shRNA (NTC), which wereexposed to 20% or 1% O2 for 24 h. Data were normalized to NTC at 20% O2 (mean ± SEM; n = 3). **P < 0.01, ***P < 0.001 vs. NTC at 20% O2;

##P < 0.01, ###P <0.001 vs. NTC at 1%O2 (two-way ANOVAwith Bonferroni posttest). (D) SUM159 cells were treatedwith vehicle (Con) or 2.5 μMacriflavine (ACF), exposed to 20% or1% O2 for 48 h, and expression of CD47 mRNA was assayed by RT-qPCR. ***P < 0.001 vs. Con at 20% O2;

##P < 0.01 vs. Con at 1% O2 (two-way ANOVA withBonferroni posttest). (E) Immunoblot assays were performed to analyze HIF-2α, HIF-1α, CD47, and Actin protein expression in whole cell lysates prepared fromMCF7(Left) and HCC1954 (Middle) subclones exposed to 20%O2 (N) or 1%O2 (H) for 24 h, and lysates from SUM159 cells (Right), whichwere treatedwith vehicle (Con) or2.5 μM acriflavine (ACF) and exposed to 20% or 1% O2 for 48 h. (F–H) MCF7 (F), HCC1954 (G), and SUM159 (H) cells were exposed to 20% or 1% O2 for 24 h andCD47 expression on the cell surface was determined by flow cytometry using anti-CD47 antibody or isotype control antibody (ISO) (Upper). The anti-CD47 medianfluorescence intensity (MFI) was determined and normalized to lane 1 in each bar graph (Lower;mean ± SEM; n = 3). *P < 0.05, **P < 0.01, ***P < 0.001 vs. NTC at20% O2;

##P < 0.01, ###P < 0.001 vs. NTC at 1% O2 (two-way ANOVA with Bonferroni posttest for MCF7 and HCC1954 subclones; Student’s t test for SUM159 cells).

E6216 | www.pnas.org/cgi/doi/10.1073/pnas.1520032112 Zhang et al.

Dow

nloa

ded

by g

uest

on

Aug

ust 3

1, 2

021

To determine whether HIF-1α or HIF-2α was required for CD47expression under hypoxic conditions, we analyzed MCF7 andHCC1954 subclones, which were stably transfected with an expressionvector encoding short hairpin RNA (shRNA) targeting HIF-1α(sh1α) or HIF-2α (sh2α) or both HIF-1α and HIF-2α (doubleknockdown, DKD) or a nontargeting control shRNA (NTC).Hypoxia-induced CD47 mRNA expression was lost in MCF7 sh1αand DKD subclones (Fig. 1B) and in HCC1954 sh1α, sh2α, andDKD subclones (Fig. 1C). Acriflavine inhibits the dimerization ofHIF-1α or HIF-2α with HIF-1β (31), leading to the degradation ofundimerized HIF-1α or HIF-2α subunits during prolonged (>24 h)incubation (18). Treatment of SUM159 cells with acriflavineblocked hypoxia-induced CD47 mRNA expression (Fig. 1D).Hypoxic induction of CD47 protein expression in whole celllysates was also abrogated in MCF7-DKD and HCC1954-DKDsubclones and in acriflavine-treated SUM159 cells (Fig. 1E).Because cell-surface expression of CD47 is required for its

antiphagocytic capacity, flow cytometry was performed. MCF7-DKD and HCC1954-DKD subclones showed decreased CD47cell-surface expression compared with the respective NTC sub-clones under both hypoxic and nonhypoxic (20% O2) conditions(Fig. 1 F and G). In HCC1954-NTC and SUM159 cells, hypoxiaincreased CD47 cell-surface expression (Fig. 1H). It was notpossible to perform flow cytometry on acriflavine-treated cellsdue to the inherent fluorescence of the drug. Taken together, thedata in Fig. 1 demonstrate that CD47 mRNA and protein ex-pression are induced by hypoxia in three breast cancer cell linesin a HIF-dependent manner, leading to increased cell-surfaceexpression of CD47 in HCC1954 and SUM159 cells.

CD47 Is a Direct HIF-1 Target Gene. To determine whether HIF-1directly regulates CD47 gene transcription, the human genomesequence was searched for matches to the consensus HIF bind-ing site 5′-(A/G)CGTG-3′ (32) that were located within DNaseI-hypersensitive domains at the CD47 locus. Chromatin immu-noprecipitation (ChIP) assays revealed that a DNA sequence,which encompassed 5′-GCGTG-3′ at −239 bp (site 1) and 5′-CACGC-3′ (5′-GCGTG-3′ on the antisense strand) at −200 bp(site 2), relative to the CD47 transcription start site (Fig. 2A,Top), was enriched by immunoprecipitation of chromatin fromhypoxic MCF7, HCC1954, and SUM159 cells with HIF-1α orHIF-1β antibodies (Fig. 2B), indicating that hypoxia induces di-rect binding of HIF-1 to the CD47 promoter.To determine whether DNA sequences encompassing either

site alone functioned as a hypoxia response element (HRE), a55-bp oligonucleotide spanning HIF binding site 1 or site 2 (Fig.2A, Bottom) was inserted into pGL2-promoter, which contains abasal SV40 promoter driving expression of firefly luciferase, togenerate reporters pGL2-HRE1 and pGL2-HRE2, respectively.As a negative control, we used pSV-Renilla, which encodesRenilla luciferase driven by the SV40 promoter alone. Hypoxiasignificantly increased the ratio of firefly:Renilla luciferase inHCC1954 and SUM159 cells that were cotransfected with eitherpGL2-HRE1 or pGL2-HRE2 and pSV-Renilla (Fig. 2C), dem-onstrating that each of the 55-bp oligonucleotides functions asan HRE. Taken together, the data presented in Fig. 2 indicatethat HIF-1 binds directly to the CD47 promoter to activate genetranscription.

B

C

A

HCC1954 SUM159 HCC1954 SUM159

pGL2-HRE1 pGL2-HRE2

HCC1954MCF7 SUM159

Fig. 2. CD47 is a direct HIF-1 target gene. (A) Two candidate HIF-1 binding sites in the 5′-flanking region of the human CD47 gene were identified: 5′-GCGTG-3′ (site 1,located 239 bp 5′ to the transcription start site) and 5′-CACGC-3′ (site 2, located 200 bp 5′ to the transcription start site), which is the complementary sequence of 5′-GCGTG-3′ on the antisense strand, are shown in red. (B) MCF-7 (Left), HCC1954 (Middle), and SUM159 (Right) cells were exposed to 20%or 1%O2 for 16 h and chromatinimmunoprecipitation (ChIP) assays were performed using IgG or antibodies against HIF-1α or HIF-1β. Primers flanking the entire sequence shown inAwere used for qPCRand results were normalized to IgG at 20%O2 (mean ± SEM; n = 3). *P < 0.05, **P < 0.01 vs. 20%O2 (Student’s t test). (C) Cells were cotransfected with pSV-Renilla andfirefly luciferase reporter pGL2-HRE1 or pGL2-HRE2, containing an oligonucleotide encompassing HIF binding site 1 or site 2, respectively, and exposed to 20% or 1%O2

for 24 h. Luciferase activity (firely:Renilla luciferase ratio) was determined and normalized to 20% O2 (mean ± SEM; n = 3). **P < 0.01 vs. 20% O2 (Student’s t test).

Zhang et al. PNAS | Published online October 28, 2015 | E6217

MED

ICALSC

IENCE

SPN

ASPL

US

Dow

nloa

ded

by g

uest

on

Aug

ust 3

1, 2

021

HIF Deficiency Increases the Phagocytosis of Breast Cancer Cells. Weobserved decreased expression of CD47 on the surface of HIF-deficient human breast cancer cells (Fig. 1 F and G), leading usto hypothesize that HIF inhibition may promote phagocytosis ofbreast cancer cells by macrophages. To test this hypothesis, weperformed in vitro phagocytosis assays on HCC1954 and MCF7subclones. The breast cancer cells were labeled with the fluo-rescent dye CFSE, exposed to 20% or 1% O2 for 24 h, coculturedwith bone marrow-derived macrophages for 2 h, incubated withallophycocyanin (APC)-labeled F4/80 antibody, and analyzed byflow cytometry to detect APC+CFSE+ cells, which representmacrophages that have phagocytosed breast cancer cells (UpperRight quadrant [Q2] of scatter plots in Fig. 3A). Phagocytosis wassignificantly increased in MCF7-DKD and HCC1954-DKD cellscompared with the respective NTC subclone, under both hypoxicand nonhypoxic conditions (Fig. 3 B and C). Hypoxia modestlyincreased phagocytosis of both NTC and DKD subclones, whichsuggests the existence of a prophagocytic signal that is induced byhypoxia in a HIF-independent manner.

CD47 Deficiency Increases the Phagocytosis of Breast Cancer Cells. Todetermine whether CD47 mediates protection of breast cancercells against phagocytosis, we transfected SUM159 cells with anexpression vector encoding one of five different shRNAs tar-geting CD47. Remarkably, we were not able to establish stablesubclones that expressed three of the shRNAs and expression ofthe other two shRNAs (shCD47-2 and shCD47-4) caused rela-tively modest (50–75%) inhibition of CD47 mRNA (Fig. 4A) andcell-surface protein (Fig. 4B) expression. However, phagocytosiswas significantly increased in the CD47-knockdown subclones(Fig. 4C) and CD47 levels were inversely correlated with the ex-tent of phagocytosis. Taken together, the data presented in Fig. 4demonstrate that CD47 expression protects breast cancer cellsagainst phagocytosis by bone marrow-derived macrophages.

CD47 Promotes the Breast CSC Phenotype. CD47 is preferentiallyexpressed on bladder, liver, and pancreatic CSCs compared withthe bulk cancer cells (non-CSCs) (33–36). To investigate whetherCD47 plays a role in breast CSCs, we first analyzed CD47mRNA levels in SUM159 cells, which were cultured as standardadherent monolayers or as nonadherent spheroids (mammo-spheres), which are highly enriched for CSCs (37). CD47 mRNAlevels were twofold higher in mammosphere cultures (Fig. 5A).Exposure of breast cancer cells to hypoxia for 3 d induces in-creased mammosphere formation in a HIF-1α–dependent man-ner (20). Knockdown of CD47 expression significantly decreasedmammosphere formation, with the greatest reduction observedin the shCD47-4 subclone (Fig. 5B), which had the greatest in-hibition of CD47 expression (Fig. 4 A and B). Analysis of alde-hyde dehydrogenase (ALDH) activity in breast cancer cells alsoidentifies a subpopulation that is enriched for CSCs (38) and thepercentage of ALDH+ cells increases in response to hypoxia (19,20). Hypoxia markedly increased the percentage of ALDH+ cellsin the NTC subclone, whereas knockdown of CD47 expressionsignificantly decreased the percentage of ALDH+ CSCs at both20% and 1% O2 (Fig. 5C). Taken together, the data presented inFig. 5 demonstrate that CD47 expression plays an important rolein promoting the breast CSC phenotype. We and others havepreviously reported that knockdown of HIF-1α blocks hypoxia-induced enrichment of breast CSCs, as determined by ALDHexpression or mammosphere formation assay (19, 20). Thus,CD47 loss-of-function phenocopies HIF-1α loss of function withrespect to CSC maintenance.

CD47 Expression Is Associated with HIF Target Gene Expression andPatient Mortality. To test whether the results obtained from theanalysis of breast cancer cell lines are relevant to patients with breastcancer, we analyzed gene expression data from 1,040 primary humanbreast cancer samples that are publicly available in The CancerGenome Atlas (TCGA) database (39). To determine whether HIFsregulate CD47 gene expression in human breast cancers, we com-pared CD47 mRNA levels with the levels of P4HA1, P4HA2,ANGPTL4, SLC2A1, CXCR3, VEGFA, PLOD1, PLOD2, L1CAM,and MET mRNA, which are all products of genes that are HIFregulated in breast cancer cells (18, 20). CD47 mRNA levels, likethose of other HIF target genes, were most highly expressed in breastcancers of the basal-like molecular subtype (Fig. 6A). Statisticalanalysis revealed that CD47 mRNA levels were significantly corre-lated with the levels of 8 of 10 mRNAs encoded by HIF target genes(Fig. 6B), which is comparable to correlations between members ofthis group (10, 15, 20). Expression of CD44, which is a marker ofbreast CSCs (40) and a HIF target gene (41), was also significantlycorrelated with CD47 expression in the 1,040 breast cancers (Fig. 6B).HIF target gene expression in human breast cancers is sig-

nificantly associated with patient mortality (18). To determine ifCD47 mRNA expression was also a prognostic factor in humanbreast cancer, we used two published datasets that contain bothgene expression and patient survival data (42, 43). Patients were

A

B C MCF7HCC1954

F4/8

0-A

PC

CFSE

Fig. 3. Increased phagocytosis of HIF-deficient breast cancer cells. (A) HCC1954NTC and DKD subclones were exposed to 20% or 1% O2 for 24 h, stained withCFSE, incubated with mouse bone marrow-derived macrophages for 2 h,stained with F4/80-APC antibody, and analyzed by flow cytometry. (B and C)Phagocytosis assays were performed with HCC1954 (B) and MCF7 (C) subclonesas described above and the percentage of CFSE+APC+ phagocytosed cancer cellswas normalized to NTC at 20% O2 (mean ± SEM; n = 3). **P < 0.01, ***P <0.001 vs. NTC at 20% O2;

##P < 0.01, ###P < 0.001 vs. NTC at 1% O2 (two-wayANOVA with Bonferroni posttest).

E6218 | www.pnas.org/cgi/doi/10.1073/pnas.1520032112 Zhang et al.

Dow

nloa

ded

by g

uest

on

Aug

ust 3

1, 2

021

stratified into two groups based on whether CD47 mRNA levelsin the primary tumor were greater or less than the median value.Increased CD47 mRNA expression levels were associated with asignificantly decreased probability of overall survival in the twoindependent datasets, which together comprised 1,954 patientswith breast cancer (Fig. 6C).

DiscussionGrowing evidence indicates that in various human cancers, CD47expression is required to avoid innate immune surveillance andelimination by phagocytosis. Blocking the interaction of CD47with its receptor, SIRPα, on macrophages enables phagocytosisand inhibits tumor growth (35, 36, 44–46). However, the mo-lecular mechanisms regulating the expression of CD47 by cancercells have not been delineated. In this study, we demonstrate thathypoxia, which is a critical microenvironmental stimulus in ad-vanced breast cancers, induced HIF-dependent expression ofCD47, leading to decreased phagocytosis of cancer cells bymacrophages and induction of the breast CSC phenotype, whichpromote cancer progression and patient mortality (Fig. 6D).Despite the recent interest in targeting CD47 for cancer therapy

(46), remarkably little is known about the transcriptional regulationof the CD47 gene. The results presented here represent to ourknowledge the first identification of a transcriptional regulator ofCD47 expression in breast cancer cells and further studies are re-quired to determine whether HIF-1 cooperates with other tran-scription factors that are induced by the tumor microenvironment,such as CREB, NF-κB, SMAD2, or STAT3. Increased NF-κBactivity was observed in hepatocellular carcinoma cells that de-veloped sorafenib resistance (36), which is of interest because theantiangiogenic effects of sorafenib induce intratumoral hypoxiathat causes HIF-1–dependent induction of breast CSCs (19). Im-munosuppressive functional interactions of HIF-1 and CREB havebeen proposed in T cells (47), but may also occur in cancer cells.Intratumoral hypoxia is a common finding in breast cancer

and HIFs activate the transcription of a large battery of genesencoding proteins that promote multiple steps in tumor progression,including tumor growth and vascularization, stromal cell recruitment,extracellular matrix remodeling, premetastatic niche formation, cellmotility, margination and extravasation of circulating tumor cells, andCSC specification/maintenance (2). Here, we find that HIFs also playan important role in immune evasion by activating the expression ofCD47, an antiphagocytic signal. In three different breast cancer celllines, which represent luminal/ER+ (MCF7), HER2-enriched(HCC1954), and basal-like/triple-negative (SUM159) subtypes ofbreast cancer, the expression of CD47 mRNA and protein was in-duced by hypoxia in a HIF-dependent manner. A recent studyreported that CD47 protein was detected in 5% of hormone re-ceptor positive, HER2− breast cancers (48). Our bioinformaticanalysis suggests that CD47 expression is likely to be higher amongtriple-negative breast cancers, most of which fall into the basal-likemolecular subtype that is characterized by increased expression ofthe HIF transcriptome (18, 39).Hypoxia did not induce increased CD47 expression in MCF10A

or MDA-MB-231 cells. This heterogeneous transcriptional re-sponse of breast cell lines to hypoxia is commonly observed. Forexample, analysis of mRNAs encoding lysyl oxidase (LOX) andLOX-like proteins (LOXL1–4) revealed that hypoxia induced

A

B

C

Anti-CD47-PE

% o

f Max

20% O2 1% O2

NTC sh-2 sh-4 ISO

NTC sh-2 sh-4 ISO

Fig. 4. CD47 deficiency increases the phagocytosis of breast cancer cells. (A) CD47mRNA levels were analyzed by RT-qPCR in SUM159 subclones expressing either oftwo different shRNAs targeting CD47 (shCD47) or a nontargeting control shRNA(NTC). Results were normalized to NTC (mean ± SEM; n = 3). ***P < 0.001 vs. NTC(one-way ANOVA with Bonferroni posttest). (B) CD47 protein expression on the cellsurfacewas determined by flow cytometry of SUM159 subclones exposed to 20%or1% O2 for 24 h (Upper). The anti-CD47 median fluorescence intensity (MFI) wasdetermined and normalized to NTC at 20% O2 (Lower; mean ± SEM; n = 3). **P <0.01, ***P < 0.001 vs. NTC at 20% O2;

###P < 0.001 vs. NTC at 1% O2 (two-way

ANOVA with Bonferroni posttest). (C) SUM159 subclones were exposed to20% or 1% O2 for 24 h, stained with CFSE, incubated with bone marrow-derived macrophages for 2 h, stained with F4/80-APC antibody, subjected toflow cytometry, and the percentage of CFSE+APC+ phagocytosed cancer cellswas determined and normalized to the NTC subclone at 20% O2 (mean ±SEM; n = 3). **P < 0.01 vs. NTC at 20% O2;

#P < 0.05, ##P < 0.01 vs. NTC at 1%O2 (two-way ANOVA with Bonferroni posttest).

Zhang et al. PNAS | Published online October 28, 2015 | E6219

MED

ICALSC

IENCE

SPN

ASPL

US

Dow

nloa

ded

by g

uest

on

Aug

ust 3

1, 2

021

increased expression of LOX and LOXL4 in MDA-MB-231,whereas LOXL2 expression was induced in MDA-MB-435cells (12). We also observed heterogeneity with respect to re-quirements for HIF-α subunits, as CD47 induction required onlyHIF-1α in MCF7 cells, but both HIF-1α and HIF-2α in HCC1954cells. Similarly, we previously reported that hypoxia-inducedLOXL2 expression required only HIF-1α in MDA-MB-435 cells,whereas expression of LOX and LOXL4 in MDA-MB-231 cells re-quired HIF-1α and HIF-2α (12). Heterogeneity at the level of geneexpression is recognized as a major obstacle to successful cancertherapy. However, in the cases described above, treatment with a HIFinhibitor, such as acriflavine, successfully blocked all hypoxia-inducedgene expression (Fig. 1D).Cells express varying levels of prophagocytic and antiphagocytic

signaling proteins, and it is the integration of both signals that de-termines whether the target cell will be phagocytosed. The relativelymodest changes in CD47 cell surface levels that resulted from ex-pression of shRNA targeting HIF-1α and HIF-2α or CD47 led tosignificant changes in phagocytosis, indicating a fine balance betweenpro- and antiphagocytic signaling, such that a hypoxic tumor micro-environment may significantly tip the balance toward immune eva-sion. Several prophagocytic signals have been identified, includingcell-surface expression of phosphatidylserine and CRT (49, 50). CRTis required for the phagocytosis of cancer cells when CD47–SIRPαinteraction is blocked (28). CRT expression was induced by ex-posure of cardiomyocytes to hypoxia and reoxygenation (51),

suggesting that the increased phagocytosis of NTC and DKDsubclones after hypoxic exposure may be due to HIF-inde-pendent CRT expression, but further studies are required totest this hypothesis. A recent report demonstrated that CD47blockade also promotes T-cell–mediated elimination of im-munogenic tumors (52) and HIF-1 is known to inhibit effectorT cells through the production of adenosine (23, 47), sug-gesting that HIF inhibitors may improve the response to CD47blockade both by inhibiting CD47 expression and by dis-inhibiting T-cell–mediated antitumor immunity.Two findings in our study suggested that hypoxia-induced ex-

pression of CD47 plays an important role that is independent ofits antiphagocytic function. First, although CD47 mRNA andprotein expression were induced by hypoxia in all three breastcancer cell lines analyzed, increased CD47 cell-surface expres-sion and decreased phagocytosis were only induced by hypoxia inSUM159 cells. These data suggest that hypoxia-induced CD47may fulfill another function that does not require cell-surfaceexpression of the protein. Second, attempts at generating CD47-deficient cells were only successful in establishing subclones withmodest knockdown of CD47, suggesting that expression of theprotein was required for clonal expansion. Using two establishedassays for breast CSCs, we demonstrated that CD47 expressionwas increased in breast CSCs (relative to bulk cancer cells) andwas required for breast CSC maintenance, as decreased CD47expression led to significantly reduced numbers of CSCs in a

A

C

B

Monolayer Mammosphere

Fig. 5. CD47 expression promotes the breast cancer stem cell phenotype. (A) CD47 mRNA expression relative to 18S rRNA was analyzed by RT-qPCR usingtotal RNA isolated from SUM159 cells that were grown in conventional adherent monolayer culture or as nonadherent mammospheres (described below).The results were normalized to monolayer cells (mean ± SEM; n = 3). ***P < 0.001 (Student’s t test). (B) SUM159 subclones were exposed to 20% or 1% O2

for 72 h in monolayer culture, harvested, seeded at a density of 5,000 cells per well on six-well ultra-low attachment culture plates, and cultured for 5 d inthe presence of 20% O2 (Upper). Mammospheres that were greater than 50 μm in diameter were counted by automated image analysis (Lower; mean ± SEM;n = 3). **P < 0.01 vs. NTC at 20% O2;

###P < 0.001 vs. NTC at 1%O2 (two-way ANOVA with Bonferroni posttest). (C) SUM159 subclones were exposed to 20% or1% O2 for 72 h in monolayer culture and the percentage of cells expressing aldehyde dehydrogenase (ALDH+) was determined by flow cytometry (mean ±SEM; n = 3). *P < 0.05, **P < 0.01 vs. NTC at 20% O2;

##P < 0.01, ###P < 0.001 vs. NTC at 1% O2 (two-way ANOVA with Bonferroni posttest).

E6220 | www.pnas.org/cgi/doi/10.1073/pnas.1520032112 Zhang et al.

Dow

nloa

ded

by g

uest

on

Aug

ust 3

1, 2

021

dose-dependent manner. Further studies are required to de-termine whether CD47 must be expressed at the cell surface tocontribute to the maintenance of breast CSCs.To generate a primary, recurrent, or metastatic tumor, a breast

cancer cell must avoid immune destruction and give rise to bothCSCs and differentiated breast cancer cells. A recent study foundthat among circulating tumor cells in the blood of women withmetastatic breast cancer, the cells that were capable of initiatingtumors when injected into mice expressed the cell surface proteinsCD47, CD44, and MET (29). Remarkably, as in the case of CD47,the expression of CD44 and MET is induced by hypoxia in a HIF-dependent manner (41, 53). Furthermore, our analysis of geneexpression data from over 1,000 human breast cancers revealedthat CD47 expression was significantly correlated with the ex-pression of CD44,MET, and other HIF target genes (Fig. 6B). We

also found that CD47 mRNA expression in primary human breastcancers was significantly associated with patient mortality in twolarge and independent datasets (Fig. 6C). Similar results wererecently reported regarding the immunohistochemical detection ofCD47 protein expression in breast cancer biopsies (48).We have recently demonstrated that exposure of breast cancer

cells to hypoxia or chemotherapy induces the breast CSC phe-notype through multiple HIF-dependent molecular pathways(18, 20, 54). Induction of CD47 expression provides anothermechanism by which hypoxia induces the CSC phenotype. Inaddition to interacting with SIRPα, CD47 engages in severalother functional interactions (55) and further studies are re-quired to determine the molecular mechanism by which the ex-pression of CD47, localized to the cell surface or perhaps anintracellular compartment, promotes the specification and/or

GENE CD47

P4HA1 *

P4HA2 n.s.

ANGPTL4 n.s.

SLC2A1 ***

CXCR3 ***

VEGFA *

PLOD1 ***

PLOD2 ***

L1CAM ***

MET ***

CD44 ***

A B

C Intratumoral Hypoxia

HIF-1

CD47

Evasion of phagocytosis

Specification and/or maintenance of CSCs

Cancer progression

Patient mortality

D

Fig. 6. CD47 expression in primary human breast cancers is associated with HIF target gene expression and decreased patient survival. (A) The relativeexpression levels of 10 mRNAs encoded by HIF target genes, as well as CD44 and CD47 mRNAs, are shown for 1,040 primary breast cancer tissuesamples from The Cancer Genome Atlas (TCGA), which were grouped according to the expression levels of 50 mRNAs (PAM50) that define the luminalA, luminal B, HER2+, basal-like, and normal-like breast cancer molecular subtypes (57). Color code for mRNA expression levels are as follows: red,greater than median; green, less than median. (B) CD47 mRNA expression in each primary breast cancer specimen was compared with the expressionof CD44 mRNA and HIF-regulated mRNAs using Pearson’s correlation test. *P < 0.05, ***P < 0.001; n.s., not significant. (C ) Kaplan–Meier curves wereconstructed to analyze the association of CD47 mRNA expression in the primary tumor with overall survival (OS) of patients with breast cancer usingtwo independent datasets (42, 43). Statistical analysis was performed using log-rank tests. (D) Intratumoral hypoxia induces HIF-1–dependent ex-pression of CD47 in breast cancer cells, leading to decreased phagocytosis of cancer cells by macrophages and induction of the cancer stem cell (CSC)phenotype, which promote cancer progression and patient mortality.

Zhang et al. PNAS | Published online October 28, 2015 | E6221

MED

ICALSC

IENCE

SPN

ASPL

US

Dow

nloa

ded

by g

uest

on

Aug

ust 3

1, 2

021

maintenance of CSCs. Genetic or pharmacologic inhibition ofHIF activity reduced the percentage of CSCs (18–20, 56), whichmay be due in part to the inhibition of CD47 expression. In-creased expression of HIF-1α mRNA or protein, or HIF targetgene mRNAs, in primary breast cancer biopsies is associatedwith increased patient mortality (2, 18). Our data suggest thataddition of HIF inhibitors to current treatment regimens mayimprove outcome in such high-risk patients in part by attackingCSCs and disinhibiting innate immunity.

Materials and MethodsCell Lines and Culture. Human breast cell lines MCF10A, MCF7, HCC1954,SUM159, MDA-MB-231, and MDA-MB-435 were cultured as previously de-scribed (18, 20, 21, 54). All cells were maintained at 37 °C in a 5% CO2 and95% air (20% O2) incubator. For hypoxia exposure, cells were placed in amodular incubator chamber (Billups–Rothenberg) that was flushed with agas mixture containing 1% O2, 5% CO2, and 94% N2.

shRNA, Lentiviruses, and Transduction. All lentiviral vectors contained a puro-mycin resistance gene. Vectors encoding shRNA targeting HIF-1α and HIF-2αwere described previously (5). Vectors encoding CD47 shRNAs were purchasedfrom Sigma-Aldrich; the clone ID and nucleotide sequences coding for shRNAare as follows: shCD47-2: NM_001777.2–1024s1c1; 5′-CCG GGC TTC CAA TCAGAA GAC TAT ACT CGA GTA TAG TCT TCT GAT TGG AAG CTT TTT-3′; shCD47-4:NM_001777.2–988s1c1; 5′-CCG GGC ACA ATT ACT TGG ACT AGT TCT CGAGAA CTA GTC CAA GTA ATT GTG CTT TTT-3′. Lentiviruses were packaged in293T cells by cotransfection with plasmid pCMV-dR8.91 and plasmid encodingvesicular stomatitis virus G protein using PolyJet (SignaGen). Supernatantcontaining viral particles was collected 48 h posttransfection, filtered (0.45-μmpore size), and transduced into MCF7, HCC1954, or SUM159 cells in the pres-ence of 8 μg/mL of Polybrene (Sigma-Aldrich). After 24 h, cells were main-tained in medium containing 0.5 μg/mL puromycin (Sigma-Aldrich).

RT-qPCR. Total RNA was extracted, cDNA was synthesized by RT, and qPCRanalysis was performed as described (18), using PCR primers with the fol-lowing sequences: CD47, 5′-AGA AGG TGA AAC GAT CAT CGA GC-3′ and5′-CTC ATC CAT ACC ACC GGA TCT-3′; and 18S rRNA, 5′-CGG CGA CGA CCCATT CGA AC-3′ and 5′-GAA TCG AAC CCT GAT TCC CCG TC-3′.

Immunoblot Assay. Whole-cell lysates were prepared in modified RIPA buffer;proteins were separated by SDS/PAGE, blotted onto nitrocellulose membranes,and probed with primary antibodies against HIF-1α, HIF-2α, and CD47 (NovusBiologicals). HRP-conjugated secondary antibodies (GE Healthcare) were used.Chemiluminescent signals were detected using ECL Plus (GE Healthcare). Blotswere stripped and reprobed with antiactin antibody (Santa Cruz).

ChIP Assay. Cells were exposed to 20% or 1% O2 for 16 h, then cross-linked inthe presence of 3.7% (vol/vol) formaldehyde for 10 min, quenched in0.125 M glycine for 5 min, and lysed with SDS lysis buffer. Chromatin wassheared by sonication, and lysates were precleared with salmon sperm DNA/protein A-agarose slurry (Millipore) and incubated with IgG or antibodiesagainst HIF-1α (Santa Cruz) or HIF-1β (Novus Biologicals) in the presence ofprotein A-conjugated agarose beads overnight. After sequential washes ofthe beads, DNA was eluted in 1% SDS with 0.1 M NaHCO3, and cross-linkswere reversed by addition of 0.2 M NaCl. DNA was purified by phenol-chloroform extraction and ethanol precipitation. Candidate binding siteswere analyzed by qPCR using flanking primers with the following sequences:5′-AGG AAC GGG TGC AAT GAG-3′ and 5-CTT CCA GGT CAC GTC CTG TC-3′.

Luciferase Reporter Assays.A total of 10,000 cells plated onto 48-well plateswere cotransfected with pGL2-HRE1 or pGL2-HRE2 and pSV-Renilla. Thesequences of the double-strand oligonucleotides used for pGL2-HREplasmid construction are as follows: HRE1, 5′-GAT CAA CGG GTG CAA

TGA GGT CCC CGG CGA GCG TGG GAA CAC AGG GTT CAG CCT CCT GC-3′and 5′-TCG AGC AGG AGG CTG AAC CCT GTG TTC CCA CGC TCG CCG GGGACC TCA TTG CAC CCG TT-3′; HRE2, 5′-GAT CGG TTC AGC CTC CTG CGGCGG GCG AGC ACG CGG ACC CCA GGG GCG GGC GGG TGC GA-3′ and 5′-TCG ATC GCA CCC GCC CGC CCC TGG GGT CCG CGT GCT CGC CCG CCGCAG GAG GCT GAA CC-3′. At 24 h after transfection, cells were exposed to20% or 1% O2 for 24 h and lysed. Luciferase activities were determined with amultiwell luminescence reader (PerkinElmer) using the Dual-Luciferase Re-porter Assay System (Promega).

Cell Surface Expression of CD47. Cells were harvested after 24-h exposure to20% or 1% O2 and a single cell suspension was prepared. Cells were in-cubated with Fc Block (BD Pharmingen) for 10 min, then stained with phy-coerythrin (PE)-conjugated CD47 antibody or isotype control (R&D Systems),and subjected to flow cytometry (FACScalibur, BD Biosciences). Dead cellswere gated out by side-scatter and forward-scatter analysis.

Isolation of Bone Marrow-Derived Macrophages. Bone marrow cells wereisolated from mouse long bones and cultured for 7 d in RPMI-1640 supple-mented with 10% (vol/vol) FBS, 1% penicillin/streptomycin, and 30 ng/mLCSF1 (R&D Systems) as previously described (9).

Phagocytosis Assay. Macrophages were plated (5 × 104 per well) in a 24-welltissue-culture plate in complete RPMI-1640 medium, which was supple-mented with CSF1, 24 h before the experiment. Breast cancer cells werestained with 2.5-μM carboxyfluorescein succinimidyl ester (CFSE) accordingto the manufacturer’s protocol (Invitrogen). Macrophages were incubated inserum-free medium for 2 h before adding 2 × 105 CFSE-labeled breast cancercells. After coculture for 2 h at 37 °C, cells were harvested, macrophageswere stained with APC-conjugated F4/80 (Novus Biologicals), and flowcytometry (FACScalibur, BD Biosciences) was performed. A total of 10,000cells in each sample were analyzed. Unstained control and single stainedcells were prepared for gating. Phagocytosis was calculated as the per-centage of F4/80+CFSE+ cells (Q2) among CSFE+ cells (Q2 + Q3): phagocytosis(%) = [Q2/(Q2 + Q3)] × 100%.

Mammosphere Assay. Cells were exposed to 20% or 1% O2 in monolayeradherent culture for 72 h. Single cells were plated in six-well ultra-low at-tachment culture plates (Corning) at a density of 5,000 cells per well incomplete mammocult medium (Stem Cell Technologies). Mammospheres(diameter ≥50 μm) were counted after 5 d. Mammosphere cultures were im-aged using an Olympus phase-contrast microscope and mammosphere diam-eters were determined using ImageJ software (National Institutes of Health).

ALDH Assay. Cells were exposed to 20% or 1% O2 for 72 h and harvested forAldefluor assay (Stem Cell Technologies). Cells were suspended in assaybuffer containing 1 μmol/L BODIPY-aminoacetaldehyde and incubated for45 min at 37 °C. As a negative control, for each sample an aliquot of cells wastreated with the ALDH inhibitor diethylaminobenzaldehyde (50 mM). Sam-ples were subjected to flow cytometry analysis (FACScalibur, BD Biosciences).

Statistical Analysis. Data are expressed as mean ± SEM and were analyzedusing Student’s t test for two groups or ANOVA with Bonferroni posttest formultiple groups. Kaplan–Meier curves were generated using two indepen-dent datasets containing gene expression and overall survival data on pa-tients with breast cancer (42, 43). The log rank test was performed todetermine whether observed differences between groups were statisticallysignificant. P values <0.05 were considered significant.

ACKNOWLEDGMENTS. We thank Karen Padgett of Novus Biologicals forproviding APC-F4/80, CD47, and HIF-1β antibodies. This work was supportedin part by American Cancer Society Grant 122437-RP-12-090-01-COUN. G.L.S.is an American Cancer Society Research Professor and the C. Michael ArmstrongProfessor at The Johns Hopkins University School of Medicine.

1. Vaupel P, Höckel M, Mayer A (2007) Detection and characterization of tumor hypoxiausing pO2 histography. Antioxid Redox Signal 9(8):1221–1235.

2. Semenza GL (2015) The hypoxic tumor microenvironment: A driving force for breastcancer metastasis. Biochim Biophys Acta, 10.1016/j.bbamcr.2015.05.036.

3. Schito L, et al. (2012) Hypoxia-inducible factor 1-dependent expression of platelet-derived growth factor B promotes lymphatic metastasis of hypoxic breast cancer cells.Proc Natl Acad Sci USA 109(40):E2707–E2716.

4. Liao D, Corle C, Seagroves TN, Johnson RS (2007) Hypoxia-inducible factor-1α is a keyregulator of metastasis in a transgenic model of cancer initiation and progression.Cancer Res 67(2):563–572.

5. ZhangH, et al. (2012) HIF-1-dependent expression of angiopoietin-like 4 and L1CAMmediatesvascular metastasis of hypoxic breast cancer cells to the lungs. Oncogene 31(14):1757–1770.

6. Hiraga T, Kizaka-Kondoh S, Hirota K, Hiraoka M, Yoneda T (2007) Hypoxia andhypoxia-inducible factor-1 expression enhance osteolytic bone metastases of breastcancer. Cancer Res 67(9):4157–4163.

7. Dunn LK, et al. (2009) Hypoxia and TGF-β drive breast cancer bone metastasesthrough parallel signaling pathways in tumor cells and the bone microenvironment.PLoS One 4(9):e6896.

8. Chaturvedi P, et al. (2013) Hypoxia-inducible factor-dependent breast cancer-mesenchy-mal stem cell bidirectional signaling promotes metastasis. J Clin Invest 123(1):189–205.

E6222 | www.pnas.org/cgi/doi/10.1073/pnas.1520032112 Zhang et al.

Dow

nloa

ded

by g

uest

on

Aug

ust 3

1, 2

021

9. Chaturvedi P, Gilkes DM, Takano N, Semenza GL (2014) Hypoxia-inducible factor-dependent signaling between triple-negative breast cancer cells and mesenchymalstem cells promotes macrophage recruitment. Proc Natl Acad Sci USA 111(20):E2120–E2129.

10. Gilkes DM, et al. (2014) Hypoxia-inducible factors mediate coordinated RhoA-ROCK1expression and signaling in breast cancer cells. Proc Natl Acad Sci USA 111(3):E384–E393.

11. Erler JT, et al. (2006) Lysyl oxidase is essential for hypoxia-induced metastasis. Nature440(7088):1222–1226.

12. Wong CC, et al. (2011) Hypoxia-inducible factor 1 is a master regulator of breastcancer metastatic niche formation. Proc Natl Acad Sci USA 108(39):16369–16374.

13. Gilkes DM, et al. (2013) Collagen prolyl hydroxylases are essential for breast cancermetastasis. Cancer Res 73(11):3285–3296.

14. Gilkes DM, et al. (2013) Procollagen lysyl hydroxylase 2 is essential for hypoxia-inducedbreast cancer metastasis. Mol Cancer Res 11(5):456–466.

15. Wang T, et al. (2014) Hypoxia-inducible factors and RAB22A mediate formation ofmicrovesicles that stimulate breast cancer invasion and metastasis. Proc Natl Acad SciUSA 111(31):E3234–E3242.

16. Erler JT, et al. (2009) Hypoxia-induced lysyl oxidase is a critical mediator of bonemarrow cell recruitment to form the premetastatic niche. Cancer Cell 15(1):35–44.

17. Favaro E, Lord S, Harris AL, Buffa FM (2011) Gene expression and hypoxia in breastcancer. Genome Med 3(8):55.

18. Samanta D, Gilkes DM, Chaturvedi P, Xiang L, Semenza GL (2014) Hypoxia-induciblefactors are required for chemotherapy resistance of breast cancer stem cells. Proc NatlAcad Sci USA 111(50):E5429–E5438.

19. Conley SJ, et al. (2012) Antiangiogenic agents increase breast cancer stem cells via thegeneration of tumor hypoxia. Proc Natl Acad Sci USA 109(8):2784–2789.

20. Xiang L, et al. (2014) Hypoxia-inducible factor 1 mediates TAZ expression and nuclearlocalization to induce the breast cancer stem cell phenotype. Oncotarget 5(24):12509–12527.

21. Xiang L, et al. (2015) HIF-1α and TAZ serve as reciprocal co-activators in human breastcancer cells. Oncotarget 6(14):11768–11778.

22. Mathieu J, et al. (2011) HIF induces human embryonic stem cell markers in cancer cells.Cancer Res 71(13):4640–4652.

23. Hatfield SM, et al. (2015) Immunological mechanisms of the antitumor effects ofsupplemental oxygenation. Sci Transl Med 7(277):277ra30.

24. Noman MZ, et al. (2015) Hypoxia: A key player in anti-tumor immune response. Am JPhysiol Cell Physiol, 10.1152/ajpcell.00207.2015.

25. Chao MP, Majeti R, Weissman IL (2012) Programmed cell removal: A new obstacle inthe road to developing cancer. Nat Rev Cancer 12(1):58–67.

26. KershawMH, Smyth MJ (2013) Immunology. Making macrophages eat cancer. Science341(6141):41–42.

27. Gardai SJ, et al. (2005) Cell-surface calreticulin initiates clearance of viable or apo-ptotic cells through trans-activation of LRP on the phagocyte. Cell 123(2):321–334.

28. Chao MP, et al. (2010) Calreticulin is the dominant pro-phagocytic signal on multiplehuman cancers and is counterbalanced by CD47. Sci Transl Med 2(63):63ra94.

29. Baccelli I, et al. (2013) Identification of a population of blood circulating tumor cellsfrom breast cancer patients that initiates metastasis in a xenograft assay. NatBiotechnol 31(6):539–544.

30. Weigelt B, Reis-Filho JS (2009) Histological and molecular types of breast cancer: Isthere a unifying taxonomy? Nat Rev Clin Oncol 6(12):718–730.

31. Lee K, et al. (2009) Acriflavine inhibits HIF-1 dimerization, tumor growth, and vas-cularization. Proc Natl Acad Sci USA 106(42):17910–17915.

32. Semenza GL, et al. (1996) Hypoxia response elements in the aldolase A, enolase 1, andlactate dehydrogenase A gene promoters contain essential binding sites for hypoxia-inducible factor 1. J Biol Chem 271(51):32529–32537.

33. Chan KS, et al. (2009) Identification, molecular characterization, clinical prognosis,and therapeutic targeting of human bladder tumor-initiating cells. Proc Natl Acad SciUSA 106(33):14016–14021.

34. Lee TK, et al. (2014) Blockade of CD47-mediated cathepsin S/protease-activated re-ceptor 2 signaling provides a therapeutic target for hepatocellular carcinoma.Hepatology 60(1):179–191.

35. Cioffi M, et al. (2015) Inhibition of CD47 effectively targets pancreatic cancer stemcells via dual mechanisms. Clin Cancer Res 21(10):2325–2337.

36. Lo J, et al. (2015) Nuclear factor κ B-mediated CD47 up-regulation promotes sorafenibresistance and its blockade synergizes the effect of sorafenib in hepatocellular car-cinoma in mice. Hepatology 62(2):534–545.

37. Dontu G, et al. (2003) In vitro propagation and transcriptional profiling of humanmammary stem/progenitor cells. Genes Dev 17(10):1253–1270.

38. Ginestier C, et al. (2007) ALDH1 is a marker of normal and malignant human mam-mary stem cells and a predictor of poor clinical outcome. Cell Stem Cell 1(5):555–567.

39. Cancer Genome Atlas Network (2012) Comprehensive molecular portraits of humanbreast tumours. Nature 490(7418):61–70.

40. Al-Hajj M, Wicha MS, Benito-Hernandez A, Morrison SJ, Clarke MF (2003) Prospectiveidentification of tumorigenic breast cancer cells. Proc Natl Acad Sci USA 100(7):3983–3988.

41. Krishnamachary B, et al. (2012) Hypoxia regulates CD44 and its variant isoformsthrough HIF-1α in triple negative breast cancer. PLoS One 7(8):e44078.

42. Ringnér M, Fredlund E, Häkkinen J, Borg Å, Staaf J (2011) GOBO: Gene expression-based outcome for breast cancer online. PLoS One 6(3):e17911.

43. Györffy B, et al. (2010) An online survival analysis tool to rapidly assess the effect of22,277 genes on breast cancer prognosis using microarray data of 1,809 patients.Breast Cancer Res Treat 123(3):725–731.

44. Jaiswal S, et al. (2009) CD47 is upregulated on circulating hematopoietic stem cellsand leukemia cells to avoid phagocytosis. Cell 138(2):271–285.

45. Majeti R, et al. (2009) CD47 is an adverse prognostic factor and therapeutic antibodytarget on human acute myeloid leukemia stem cells. Cell 138(2):286–299.

46. Willingham SB, et al. (2012) The CD47-signal regulatory protein alpha (SIRPa) in-teraction is a therapeutic target for human solid tumors. Proc Natl Acad Sci USA109(17):6662–6667.

47. Sitkovsky MV (2009) T regulatory cells: Hypoxia-adenosinergic suppression and re-direction of the immune response. Trends Immunol 30(3):102–108.

48. Baccelli I, et al. (2014) Co-expression of MET and CD47 is a novel prognosticator forsurvival of luminal breast cancer patients. Oncotarget 5(18):8147–8160.

49. Hoffmann PR, et al. (2001) Phosphatidylserine (PS) induces PS receptor-mediatedmacropinocytosis and promotes clearance of apoptotic cells. J Cell Biol 155(4):649–659.

50. Ogden CA, et al. (2001) C1q and mannose binding lectin engagement of cell surfacecalreticulin and CD91 initiates macropinocytosis and uptake of apoptotic cells. J ExpMed 194(6):781–795.

51. Wu X, Liu X, Zhu X, Tang C (2007) Hypoxic preconditioning induces delayed car-dioprotection through p38 MAPK-mediated calreticulin upregulation. Shock 27(5):572–577.

52. Liu X, et al. (2015) CD47 blockade triggers T cell-mediated destruction of immuno-genic tumors. Nat Med 21(10):1209–1215.

53. Pennacchietti S, et al. (2003) Hypoxia promotes invasive growth by transcriptionalactivation of the met protooncogene. Cancer Cell 3(4):347–361.

54. Lu H, et al. (2015) Chemotherapy triggers HIF-1-dependent glutathione synthesis andcopper chelation that induces the breast cancer stem cell phenotype. Proc Natl AcadSci USA 112(33):E4600–E4609.

55. Brown EJ, Frazier WA (2001) Integrin-associated protein (CD47) and its ligands. TrendsCell Biol 11(3):130–135.

56. Xiang L, et al. (2014) Ganetespib blocks HIF-1 activity and inhibits tumor growth,vascularization, stem cell maintenance, invasion, and metastasis in orthotopic mousemodels of triple-negative breast cancer. J Mol Med (Berl) 92(2):151–164.

57. Chia SK, et al. (2012) A 50-gene intrinsic subtype classifier for prognosis and pre-diction of benefit from adjuvant tamoxifen. Clin Cancer Res 18(16):4465–4472.

Zhang et al. PNAS | Published online October 28, 2015 | E6223

MED

ICALSC

IENCE

SPN

ASPL

US

Dow

nloa

ded

by g

uest

on

Aug

ust 3

1, 2

021