Hypoxia-inducible factors are required forchemotherapy resistance of breast cancer stem cellsDebangshu Samantaa,b, Daniele M. Gilkesa,b, Pallavi Chaturvedia,b, Lisha Xianga,b,c, and Gregg L. Semenzaa,b,d,e,f,g,h,1

aVascular Program, Institute for Cell Engineering, bMcKusick-Nathans Institute of Genetic Medicine, and Departments of dPediatrics, eMedicine, fOncology,gRadiation Oncology, and hBiological Chemistry, Johns Hopkins University School of Medicine, Baltimore, MD 21205; and cDepartment of Oncology andSouthwest Cancer Center, Southwest Hospital, Third Military Medical University, Chongqing 400038, China

Contributed by Gregg L. Semenza, November 10, 2014 (sent for review October 10, 2014; reviewed by Maria Alfonsina Desiderio and Youcef M. Rustum)

Triple negative breast cancers (TNBCs) are defined by the lack ofestrogen receptor (ER), progesterone receptor (PR), and humanepidermal growth factor receptor 2 expression, and are treatedwith cytotoxic chemotherapy such as paclitaxel or gemcitabine,with a durable response rate of less than 20%. TNBCs are enrichedfor the basal subtype gene expression profile and the presence ofbreast cancer stem cells, which are endowed with self-renewingand tumor-initiating properties and resistance to chemotherapy.Hypoxia-inducible factors (HIFs) and their target gene productsare highly active in TNBCs. Here, we demonstrate that HIF expres-sion and transcriptional activity are induced by treatment of MDA-MB-231, SUM-149, and SUM-159, which are human TNBC cell lines,as well as MCF-7, which is an ER+/PR+ breast cancer line, with pac-litaxel or gemcitabine. Chemotherapy-induced HIF activity enrichedthe breast cancer stem cell population through interleukin-6 andinterleukin-8 signaling and increased expression of multidrug resis-tance 1. Coadministration of HIF inhibitors overcame the resistanceof breast cancer stem cells to paclitaxel or gemcitabine, both in vitroand in vivo, leading to tumor eradication. Increased expression ofHIF-1α or HIF target genes in breast cancer biopsies was associatedwith decreased overall survival, particularly in patients with basalsubtype tumors and those treated with chemotherapy alone. Basedon these results, clinical trials are warranted to test whether treat-ment of patients with TNBC with a combination of cytotoxic che-motherapy and HIF inhibitors will improve patient survival.

digoxin | mammary fat pad implantation | tumor relapse

In the United States this year, more than 40,000 women will dieof breast cancer when the disease disseminates throughout the

body and fails to respond to chemotherapy. Breast cancer stemcells (BCSCs) represent a subpopulation that is central to both ofthese processes (1, 2). Although many breast cancer cells enterthe circulation, only BCSCs are capable of forming a secondarytumor (3). BCSCs are also resistant to chemotherapy and thepercentage of BCSCs following chemotherapy is increasedcompared with before therapy (4–6). A reduction in cancer cellburden of >95% may result in an apparent complete response totherapy, but may leave a population of residual BCSCs thatrepresent the source of subsequent disease recurrence that willeventually lead to death of the patient.Breast cancer therapy is based on the classification of tumors

into three groups: estrogen receptor/progesterone receptor[ER/PR]+ cancers (60–70%), which express the ER, PR, or both,and are treated with ER antagonists, such as tamoxifen, or aro-matase inhibitors, such as letrozole; human epidermal growthfactor receptor 2 (HER2+) cancers (15–20%), which overexpressHER2 and are treated with an anti-HER2 antibody, such astrastuzumab, or receptor tyrosine kinase inhibitor, such as lapa-tinib; and triple negative breast cancers (TNBCs), which do notexpress ER, PR, or HER2 and are treated with cytotoxic che-motherapy, such as paclitaxel or gemcitabine, with a durable re-sponse rate of less than 20% (7–9). TNBCs account for ∼20% ofall breast cancers (30% among African-American women) andare associated with an increased risk of metastasis and mortality.

TNBCs are characterized by an increased percentage of BCSCsand, at the molecular level, by a gene expression profile known asthe basal molecular subtype (10, 11).Microarray data from more than 500 human breast cancers (12)

revealed that one of the defining features of the basal molecularsubtype is the increased expression of a large battery of genes thatare regulated by hypoxia-inducible factors (HIFs), which aretranscription factors that function to maintain oxygen homeostasis(13). HIFs consist of a highly regulated HIF-1α or HIF-2α subunit,which heterodimerizes with a constitutively expressed HIF-1βsubunit. The homeostatic role of HIFs is to balance O2 supply anddemand, thereby preventing excessive production of reactive oxy-gen species (ROS) (14). ROS produced by cancer cells withintumors induce HIF activity (15). HIF-1 has been implicated inresistance to chemotherapy (16, 17) and in hypoxic colon cancercells, HIF-1 activated expression of multidrug resistance 1 (MDR1)(18), which mediates efflux of chemotherapy from cancer cells andrepresents a major mechanism underlying treatment failure (19).HIF-1α overexpression identified by immunohistochemistry of tu-mor biopsies is associated with increased mortality in multiplestudies involving thousands of patients with breast cancer (20).HIFs contribute to multiple steps in the metastasis of TNBCs (21).Increased expression of hypoxia-induced genes in breast cancer isalso associated with poor prognosis (22). Exposure of TNBC cellsto hypoxia has been shown to increase the percentage of BCSCs ina HIF-1α–dependent manner (23, 24). Treatment of cancer cellswith doxorubicin was shown to induce HIF-1α expression (25). Inthis study, we demonstrate that when TNBCs are treated withpaclitaxel or gemcitabine, induction of HIF activity is required forenrichment of BCSCs, both in vitro and in vivo.

ResultsPaclitaxel Induces HIF Expression and Transcriptional Activity in TNBCCells. MDA-MB-231 TNBC cells were exposed to 10 nM pacli-taxel for 0–4 d and whole cell lysates were subjected to immu-noblot assays, which revealed induction of HIF-1α protein

Significance

Breast cancer stem cells play essential roles in tumor growth,maintenance, and recurrence after chemotherapy. We reportthat treatment of human breast cancer cells with chemotherapyresults in an enrichment of breast cancer stem cells among thesurviving cells, which is dependent upon the activity of hypoxia-inducible factors (HIFs). Studies in mouse tumor models suggestthat combining chemotherapy with drugs that block HIF activitymay improve the survival of breast cancer patients.

Author contributions: D.S. and G.L.S. designed research; D.S., D.M.G., P.C., and L.X. per-formed research; D.S. and G.L.S. analyzed data; and D.S. and G.L.S. wrote the paper.

Reviewers: M.A.D., University of Milan School of Medicine; and Y.M.R., Roswell ParkCancer Institute.

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

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1421438111/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1421438111 PNAS | Published online December 1, 2014 | E5429–E5438

MED

ICALSC

IENCE

SPN

ASPL

US

expression on day 4 (Fig. 1A). Reverse transcription-quantitativereal-time PCR (RT-qPCR) analysis of total RNA isolated fromMDA-MB-231 cells revealed that paclitaxel significantly in-creased the expression of HIF-1α mRNA (Fig. 1B, Left). Expo-sure of MDA-MB-231, SUM-149, or SUM-159 TNBC cells topaclitaxel for 4 d at the concentration that inhibited growth by50% (IC50) induced HIF-1α protein expression (Fig. 1C). Wehave previously shown that digoxin and acriflavine are drugs thatinhibit HIF activity, primary tumor growth, and breast cancermetastasis (26–29). Induction of HIF-1α in MDA-MB-231 cellstreated with paclitaxel was blocked by coadministration of di-goxin (100 nM) or acriflavine (1 μM) (Fig. 1D). Although pre-vious studies indicated that exposure of cancer cells to acriflavinefor 24 h blocked dimerization of HIF-1α and HIF-1β withoutaffecting protein levels (27), exposure of TNBC cells to acrifla-vine for 4 d resulted in decreased HIF-1α protein levels, whichmay reflect decreased stability of the free (monomeric) HIF-1α

subunit. Exposure of MDA-MB-231 or SUM-159 cells to pacli-taxel for 4 d also induced expression of HIF-2α mRNA (Fig. 1B,Right) and protein (Fig. 1E), which was inhibited by co-administration of digoxin or acriflavine (Fig. 1E). Paclitaxelsignificantly increased the expression of CA9 and ENO1 mRNA,which are products of HIF target genes, whereas expression ofRPL13A mRNA, which is not HIF regulated, was unaffected(Fig. 1F). To determine the effect of paclitaxel treatment on HIFtranscriptional activity, MDA-MB-231 cells were cotransfectedwith p2.1, which is a HIF-dependent reporter plasmid thatcontains a 68-bp hypoxia response element from the humanENO1 gene upstream of a basal SV40 promoter and firefly lu-ciferase coding sequences, and pSV-RL, a control reporterplasmid that contains Renilla luciferase coding sequences down-stream of the SV40 promoter only. The ratio of firefly:Renillaluciferase activity serves as a measure of HIF transcriptional ac-tivity. In cotransfected MDA-MB-231 cells, paclitaxel treatmentsignificantly increased HIF transcriptional activity (Fig. 1G).Taken together, the data presented in Fig. 1 indicate that pacli-taxel induces HIF-α expression and HIF transcriptional activity inmultiple TNBC cell lines.

Paclitaxel Treatment Increases the Percentage of BCSCs. Severaldifferent assays identify subpopulations of breast cancer cells thatare enriched for BCSCs. The Aldefluor assay is based on the ac-tivity of aldehyde dehydrogenases (ALDH), which generate a fluo-rescent product in BCSCs that can be identified by flow cytometry(30, 31). Many breast cancer cell lines, including MDA-MB-231,SUM-149, SUM-159, and MCF-7 cells, have been shown to containan ALDH+ subpopulation that displays stem cell properties in vitroand in vivo (6, 23, 31). Approximately 1% of vehicle-treated MDA-MB-231 (Fig. 2A) or SUM-159 (Fig. 2B) TNBC cells were ALDH+.Treatment with digoxin had no significant effect, whereas treatmentwith 10 nM paclitaxel for 4 d increased the percentage of ALDH+

cells by 12-fold. Coadministration of digoxin significantly reducedthe paclitaxel-dependent increase in the percentage of ALDH+

cells (Fig. 2 A and B).No single marker, such as ALDH activity, has complete sensi-

tivity and specificity in the identification of BCSCs and a pheno-typic assay is therefore desirable. To confirm that changes in thepercentage of ALDH+ cells reflected changes in the percentage ofBCSCs, we performed mammosphere assays, which are based onthe ability of BCSCs to generate multicellular spheroids in sus-pension culture (32, 33). Exposure of SUM-159 (Fig. 2C) orSUM-149 (Fig. 2D) TNBC cells to paclitaxel for 4 d significantlyincreased the number of mammosphere-forming cells and thiseffect was completely abrogated by coadministration of digoxin.Primary mammosphere formation is a reflection of tumorigenic-ity, whereas secondary mammosphere formation provides in-creased evidence of self-renewal. Exposure of SUM-159 cells topaclitaxel again increased formation of primary mammospheres,which when dissociated and replated gave rise to increasednumbers of secondary mammospheres (Fig. 2E). Treatment withdigoxin alone significantly decreased primary and secondarymammosphere formation and abrogated paclitaxel-inducedmammosphere formation. Taken together, the data presented inFig. 2 indicate that paclitaxel increases the percentage of BCSCsin multiple TNBC cell lines in a HIF-dependent manner.

Paclitaxel-Induced Interleukin Expression Is Blocked by HIF Inhibitors.Interleukin 6 (IL-6) and IL-8 have been shown to regulate BCSCsurvival and self-renewal (6, 31, 34–38). Treatment of MDA-MB-231, SUM-159, or SUM-149 TNBC cells with paclitaxel in-duced IL-8 mRNA expression on day 4 (Fig. 3A). The inductionof IL-8 or IL-6 mRNA by exposure to paclitaxel for 4 d wasinhibited by coadministration of digoxin or acriflavine (Fig. 3B).Expression of RPL13A mRNA was unaffected by exposure topaclitaxel, alone or in combination with digoxin or acriflavine

Fig. 1. Paclitaxel induces HIF expression and signaling. (A) Immunoblotassays were performed to analyze HIF-1α and actin expression followingexposure of MDA-MB-231 (MDA-231) cells to 10 nM paclitaxel for 1–4 d orvehicle (0 d). (B) MDA-231 cells were treated with vehicle (V) or 10 nMpaclitaxel (P) for 4 d and aliquots of total RNA were assayed by RT-qPCRusing primers specific for HIF-1α or HIF-2α mRNA relative to 18S rRNA, andthe results were normalized to cells treated with V (mean ± SEM; n = 3). *P <0.001 by Student’s t test. (C) MDA-231 and SUM-159 cells were treated with0–10 nM paclitaxel, whereas SUM-149 cells were treated with 0–5 nM pac-litaxel for 4 d, and immunoblot assays of HIF-1α and actin were performed.(D and E) Cells were treated with vehicle (V) or 10 nM paclitaxel, eitheralone (P) or in combination with 100 nM digoxin (P/D) or 1 μM acriflavine(P/A), for 4 d, and the expression of actin, HIF-1α (D) and HIF-2α (E) was de-termined by immunoblot assays. (F) RT-qPCR was performed as describedabove to assay CA9, ENO1, or RPL13A mRNA (mean ± SEM; n = 3). *P < 0.001by Student’s t test. (G) MDA-MB-231 cells were transfected with HIF-depen-dent firefly luciferase reporter p2.1 and control Renilla luciferase reporterpSV-RL, then exposed to vehicle control (V) or 10 nM paclitaxel (P) for 4 d, andthe ratio of firefly:Renilla luciferase activity was determined. *P < 0.0001 byStudent’s t test.

E5430 | www.pnas.org/cgi/doi/10.1073/pnas.1421438111 Samanta et al.

(Fig. S1). Immunoblot assays confirmed that IL-6 protein levelsincreased in a dose-dependent manner in paclitaxel-treatedSUM-159 and SUM-149 cells (Fig. 3C), which mirrored theinduction of HIF-1α in the same cell lysates (Fig. 1C). Paclitaxel-induced expression of IL-6 protein was blocked by co-administration of digoxin or acriflavine (Fig. 3D). To determinewhether paclitaxel-induced BCSC enrichment was dependent onIL-6 or IL-8 signaling, we coadministered IgG or neutralizingantibodies against IL-6 or IL-8 during the exposure of MDA-MB-231 or SUM-159 cells to paclitaxel. Whereas IgG had noeffect, IL-6 or IL-8 neutralizing antibodies abrogated the in-creased percentage of mammosphere-forming cells (Fig. 3E) andALDH+ cells (Fig. 3F).

HIF-1α and HIF-2α Are Required for Paclitaxel-Induced BCSCEnrichment. To further analyze the role of HIFs in the responseto paclitaxel, we used MDA-MB-231 subclones that were stablytransfected with a lentiviral expression vector encoding a non-targeting control (NTC) short hairpin RNA (shRNA) orshRNAs targeting HIF-1α and HIF-2α (double knockdown,DKD), which have been extensively validated and used to in-vestigate the role of HIFs in breast cancer progression (29).Paclitaxel treatment markedly increased HIF-1α and HIF-2α(Fig. 4A) as well as IL-6 (Fig. 4B) protein levels and significantly

increased the percentage of ALDH+ cells (Fig. 4C) in the NTCsubclone, whereas neither HIF-1α, HIF-2α, nor IL-6 expressionwas induced and ALDH+ cells were markedly reduced in theDKD subclone, both before and after treatment with paclitaxel(Fig. 4 A–C). Analysis of subclones stably transfected witha lentiviral vector encoding shRNA targeting either HIF-1α orHIF-2α (shH1α and shH2α subclones, respectively) revealed thatexpression of both subunits was required for maximal enrich-ment of BCSCs in response to paclitaxel (Fig. 4D). Similarly,paclitaxel induced increased mammosphere formation in theNTC subclone but this response was markedly impaired in theshH1α, shH2α, and DKD subclones (Fig. 4E). Knockdown ofHIF-1α or HIF-2α impaired the induction of IL8 and IL6 gene

Fig. 2. Paclitaxel-induced BCSC enrichment is blocked by HIF inhibitors.(A and B) MDA-MB-231 and SUM-159 TNBC cells were treated with vehicle (V),100 nM digoxin (D), 10 nM paclitaxel (P), or digoxin and paclitaxel (P/D) for4 d, and the percentage of cells with aldehyde dehydrogenase activity(ALDH+) was determined (mean ± SEM; n = 3). *P < 0.001 compared withV, and #P < 0.01 compared with P, by Student’s t test. (C and D) SUM-159 (C)and SUM-149 (D) TNBC cells were treated as described above. After 4 d, thecells were transferred to ultra-low attachment plates, and 7 d later thenumber of mammospheres per field was counted (mean ± SEM; n = 3). *P <0.001, **P < 0.01 compared with V, and #P < 0.001 compared with P, byStudent’s t test. Representative photomicrographs of mammospheres areshown. (Scale bar, 2 mm.) (E) SUM-159 cells were treated as described above.Primary mammospheres were counted, collected, dissociated, transferred toultra-low attachment plates, and secondary mammospheres were counted7 d later (mean ± SEM; n = 3). *P < 0.001, **P < 0.01 compared with V, and#P < 0.001 compared with P, by Student’s t test.

Fig. 3. Paclitaxel induces expression of IL-6 and IL-8, which mediate BCSCenrichment. (A) TNBC cells were treated with 10 nM (MDA-MB-231 and SUM-159) or 5 nM (SUM-149) paclitaxel for 1–4 d or vehicle (0 d). RT-qPCR wasperformed to quantify IL-8 mRNA levels relative to 18S rRNA and normalizedto vehicle-treated cells (mean ± SEM; n = 3). *P < 0.001 compared with ve-hicle. (B) Cells were treated for 4 d with vehicle (V) or either 10 nM (MDA-MB-231 and SUM-159) or 5 nM (SUM-149) paclitaxel, either alone (P) or in com-bination with 100 nM digoxin (P/D) or 1 μM acriflavine (P/A). RT-qPCR wasperformed to assay IL-6 and IL-8 mRNA (mean ± SEM; n = 3). *P < 0.001compared with V, and #P < 0.001 compared with P, by Student’s t test. (C)Cells were exposed to the indicated concentration of paclitaxel for 4 d andimmunoblot assays were performed. (D) Cells were treated with vehicle (V) orpaclitaxel, either alone (P) or in combination with digoxin (P/D) or acriflavine(P/A) for 4 d and immunoblot assays were performed. (E and F) Cells weretreated with either vehicle (V) or paclitaxel, either alone or in the presence ofIgG or neutralizing antibody (500 ng/mL) against IL-6 or IL-8 for 4 d. At theend of treatment, cells were trypsinized and subjected to mammosphere (E)or Aldefluor assays (F) (mean ± SEM; n = 3). *P < 0.001 compared with 0 nMpaclitaxel, and #P < 0.001 compared with 10 nM paclitaxel, by Student’s t test.

Samanta et al. PNAS | Published online December 1, 2014 | E5431

MED

ICALSC

IENCE

SPN

ASPL

US

expression, whereas double knockdown of both subunits signifi-cantly decreased basal IL-6 and IL-8 mRNA expression andcompletely abrogated the response to paclitaxel (Fig. 4 F and G).Thus, inhibition of HIF activity by either pharmacologic (Figs. 2and 3) or genetic (Fig. 4) approaches blocks paclitaxel-inducedexpression of IL-6/IL-8 and enrichment of BCSCs.The histone demethylase JMJD1A, which is the product of

a HIF target gene, binds to the IL8 promoter and stimulates IL-8

mRNA expression (39). Paclitaxel treatment increased the ex-pression of JMJD1A mRNA, whereas coadministration of di-goxin or acriflavine blocked the effect of paclitaxel (Fig. 4H).Similarly, the histone demethylase JMJD3 is a HIF target geneproduct that regulates IL6 expression (40). Paclitaxel treatmentincreased JMJD3 mRNA expression and coadministration ofdigoxin or acriflavine blocked the effect of paclitaxel (Fig. 4I).These data suggest that HIFs may indirectly mediate paclitaxel-induced IL8 and IL6 gene expression by increasing the expres-sion of JMJD1A and JMJD3, respectively.

Paclitaxel-Induced SMAD2 and STAT3 Activity Is Insufficient to InduceBCSC Enrichment. A recent publication reported that TGF-β →SMAD2/4 → IL-8 signaling was necessary for paclitaxel-inducedBCSC enrichment (6). Paclitaxel induced SMAD2 phosphory-lation in both NTC and DKD subclones (Fig. S2A), despite themarked loss of ALDH+ cells in the DKD subclone (Fig. 4C),indicating that SMAD activation is independent of HIF activityand is not sufficient for BCSC enrichment in response to pac-litaxel. JAK2 → STAT3 signaling was also implicated as playinga critical role in the BCSC phenotype in TNBC (37). Paclitaxelinduced STAT3 phosphorylation in a HIF-independent manner(Fig. S2B). Thus, activation of SMAD2 and STAT3 in responseto paclitaxel is not sufficient to induce BCSC enrichment in theabsence of HIF activity.

TLR-4 and NF-κB Do Not Mediate Chemotherapy-Induced BCSCEnrichment. Previous studies suggested that paclitaxel induces ex-pression of TLR-4, which in turn increases NF-κB activity (41).However, exposure of MDA-MB-231 or SUM-159 cells to con-centrations of paclitaxel that enriched for BCSCs did not increaseTLR-4 mRNA expression (Fig. S3 A and B). Levels of phospho-IκBα were not affected by treatment with paclitaxel in the threeTNBC cell lines analyzed (Fig. S3C). Our results are consistentwith a recent report that the NF-κB pathway is constitutively ac-tivated in TNBC (42). Thus, induction of TLR-4 → NF-κB sig-naling does not play a role in paclitaxel-induced BCSC enrichmentin TNBC, although these results do not rule out constitutive in-volvement of NF-κB in promoting the BCSC phenotype.

Increased ROS Levels Mediate Induction of HIF-1α and Enrichment ofBCSCs. Short-term exposure of cancer cells to very high concen-trations of paclitaxel (100 nM for 8 h) has been shown to in-crease ROS levels (43, 44). We exposed TNBC cells to paclitaxelat IC50 for 4 d and stained the cells with MitoSOX Red, which israpidly and selectively targeted to mitochondria and producesfluorescence when oxidized by superoxide radicals, therebyserving as an indicator of ROS in live cells. Flow cytometryrevealed a marked increase in the percentage of MitoSOX+ cellsafter paclitaxel treatment of SUM-149 (Fig. 5A) or SUM-159(Fig. 5B) cells. The percentage of MitoSOX+ cells was signifi-cantly decreased when paclitaxel was coadministered with in-creasing concentrations of manganese (III) tetrakis (1-methyl-4-pyridyl) porphyrin (MnTMPyP), which is a cell-permeablesuperoxide scavenger (Fig. 5 A and B). MnTMPyP decreased thepaclitaxel-induced expression of HIF-1α in a dose-dependentmanner in both SUM-149 (Fig. 5C) and SUM-159 (Fig. 5D)cells. MnTMPyP also abrogated the paclitaxel-induced increasein the number of mammosphere-forming cells in both SUM-149(Fig. 5E) and SUM-159 (Fig. 5F) cells. Taken together, theresults presented in Fig. 5 suggest that increased ROS levelstrigger HIF-1 signaling leading to BCSC enrichment.

Paclitaxel-Induced MDR1 Expression Is Blocked by HIF Inhibitors.Treatment of MDA-MB-231 or SUM-159 TNBC cells with pac-litaxel increased expression of MDR1 mRNA (Fig. 6 A and B)and protein (Fig. 6C), which was abrogated by coadministrationof digoxin or acriflavine. Analysis of flow-sorted MDA-MB-231

Fig. 4. Paclitaxel-induced IL-6/IL-8 expression and BCSC enrichment are HIFdependent. (A–C) MDA-MB-231 subclones, which were stably transfectedwith a vector encoding a nontargeting control shRNA (NTC) or vectorsencoding shRNAs against HIF-1α and HIF-2α (DKD), were treated with eithervehicle (V) or 10 nM paclitaxel (P) for 4 d and immunoblot (A and B) orAldefluor assays (C) were performed (mean ± SEM; n = 3). *P < 0.001 com-pared with vehicle-treated NTC and #P < 0.001 compared with paclitaxel-treated NTC, by Student’s t test. (D) NTC and MDA-MB-231 subclones stablytransfected with vector encoding an shRNA against HIF-1α (shH1α) or HIF-2α(shH2α) were treated with 10 nM paclitaxel for 4 d and Aldefluor assays wereperformed (mean ± SEM; n = 3). *P < 0.001 compared with vehicle-treatedNTC, and #P < 0.001 compared with paclitaxel-treated NTC, by Student’s t test.(E) NTC and subclones stably transfected with vector encoding an shRNAagainst HIF-1α (shH1α), HIF-2α (shH2α), or both (DKD) were treated with10 nM paclitaxel for 4 d and mammosphere assays were performed (mean ±SEM; n = 3). *P < 0.001 compared with vehicle-treated NTC, and #P < 0.001compared with paclitaxel-treated NTC, by Student’s t test. (F and G) MDA-MB-231 subclones were treated with V or P for 4 d and analyzed by RT-qPCRusing primers specific for IL-8 (F) or IL-6 (G) mRNA (mean ± SEM; n = 3). *P <0.001, **P < 0.01 compared with vehicle-treated NTC, and #P < 0.001 com-pared with paclitaxel-treated NTC, by Student’s t test. (H and I) MDA-MB-231and SUM-149 cells were treated for 4 d with vehicle (V) or paclitaxel, eitheralone (P) or in combination with digoxin (P/D) or acriflavine (P/A). RT-qPCRwas performed to assay JMJD1A (H) and JMJD3 (I) mRNA relative to 18SrRNA and normalized to V (mean ± SEM; n = 3). *P < 0.001 compared with V,and #P < 0.001 compared with P, by Student’s t test.

E5432 | www.pnas.org/cgi/doi/10.1073/pnas.1421438111 Samanta et al.

(Fig. 6D) or SUM-159 (Fig. 6E) cells revealed a marked inductionof MDR1 mRNA expression in ALDH− cells and an even greaterincrease in ALDH+ cells. Deficiency of either HIF-1α or HIF-2αor both resulted in a dramatic reduction in MDR1 mRNA andprotein expression in the absence or presence of paclitaxel (Fig. 6F and G). Concurrent exposure of MDA-MB-231 cells to 10 nMpaclitaxel and 5 μM verapamil, which is a competitive inhibitor ofMDR1, significantly reduced the enrichment of BCSCs (Fig. 6H).Taken together, these data indicate that in addition to IL-6 andIL-8 signaling, induction of MDR1 expression and activity rep-resents another HIF-mediated molecular mechanism that con-tributes to paclitaxel-induced BCSC enrichment.

Digoxin Blocks Paclitaxel-Induced BCSC Enrichment in Vivo. To in-vestigate whether paclitaxel elicited similar effects in vivo, weimplanted MDA-MB-231 cells in the mammary fat pad (MFP) offemale severe combined immunodeficiency (Scid) mice. Whenthe orthotopic tumors reached a volume of 200 mm3 (designatedday 1), the mice were randomized to receive i.p. injections ofsaline, digoxin (2 mg/kg/d), paclitaxel (10 mg/kg on days 5 and10), or both digoxin and paclitaxel. Tumors were harvested onday 12. There was no significant change in body weight of themice (Fig. S4A). The combination of paclitaxel and digoxin had asignificantly greater inhibitory effect on tumor growth comparedwith either digoxin or paclitaxel alone (Fig. 7A). The freshlyharvested tumors were dissociated into single cell suspensions

and subjected to the Aldefluor assay, which revealed that com-pared with saline treatment, paclitaxel significantly increased thepercentage of ALDH+ cells (Fig. 7B). In contrast, digoxin sig-nificantly decreased the percentage of ALDH+ cells and abro-gated the paclitaxel-induced enrichment of ALDH+ cells (Fig.7B). In the same tumors, paclitaxel increased the expression ofIL-6, IL-8, and MDR1 mRNA and this effect was inhibited bydigoxin (Fig. 7C).To validate our results in another model, we injected SUM-

159 TNBC cells s.c. into female athymic (Nude) mice and startedtreatment when the tumor xenografts reached a volume of 500mm3 (designated day 1). The final tumor volume on day 12 wassignificantly decreased in mice treated with both paclitaxel anddigoxin compared with mice treated with saline (control) orpaclitaxel alone (Fig. 7D). Compared with saline, paclitaxeltreatment increased the fraction of ALDH+ cells in the residualtumor and digoxin treatment inhibited the effect of paclitaxel(Fig. 7E). Paclitaxel treatment also increased the number of

Fig. 5. Paclitaxel-induced ROS is required for HIF-1α expression and BCSCenrichment. (A and B) SUM-149 and SUM-159 cells were treated either withvehicle, paclitaxel, or paclitaxel and MnTMPyP. After 4 d, the percentage ofcells positive for MitoSox Red was determined by flow cytometry (mean ±SEM; n = 3). *P < 0.001 compared with vehicle-treated cells, and #P < 0.001compared with paclitaxel-treated cells, by Student’s t test. (C and D) Cellswere treated as described above and immunoblot assays were performed. (Eand F) Cells were treated with vehicle, paclitaxel, or paclitaxel and MnTMPyPfor 4 d and mammosphere assays were performed. *P < 0.001 comparedwith vehicle-treated cells, and #P < 0.001 compared with paclitaxel-treatedcells, by Student’s t test.

Fig. 6. Paclitaxel induces MDR1 expression in TNBC cells. (A and B) Cells weretreated with vehicle (V) or 10 nM paclitaxel, either alone (P) or in combinationwith 100 nM digoxin (P/D) or 1 μM acriflavine (P/A) for 4 d and RT-qPCR wasperformed to quantify MDR1mRNA relative to 18S rRNA and normalized to V(mean ± SEM; n = 3). *P < 0.001 compared with V, and #P < 0.001 comparedwith P, by Student’s t test. (C) Cells were treated as described above andimmunoblot assays were performed. (D and E) Cells were treated with 10 nMpaclitaxel for 4 d and sorted into ALDH− and ALDH+ populations. RT-qPCRwas performed to quantify MDR1 mRNA relative to 18S rRNA and normalizedto vehicle-treated ALDH− cells. *P < 0.001 compared with vehicle-treatedALDH− cells, and #P < 0.001 compared with paclitaxel-treated ALDH− cells, byStudent’s t test. (F) MDA-MB-231 subclones were treated with vehicle or10 nM paclitaxel for 4 d and analyzed by RT-qPCR using primers specific forMDR1 mRNA relative to 18S rRNA with results normalized to vehicle-treatedNTC (mean ± SEM; n = 3). *P < 0.001 compared with vehicle-treated NTC, and#P < 0.001 compared with paclitaxel-treated NTC, by Student’s t test. (G)MDA-MB-231 subclones were treated with vehicle (V) or 10 nM paclitaxel (P)for 4 d and immunoblot assays were performed. (H) MDA-MB-231 cellswere treated with vehicle (V), the MDR1 inhibitor verapamil (M; 50 μM),paclitaxel (P; 10 nM), or both paclitaxel and verapamil (P/M) for 4 d andAldefluor assays were performed (mean ± SEM; n = 3). *P < 0.001 com-pared with V, and #P < 0.001 compared with P, by Student’s t test.

Samanta et al. PNAS | Published online December 1, 2014 | E5433

MED

ICALSC

IENCE

SPN

ASPL

US

mammosphere-forming cells and digoxin significantly bluntedthe effect of paclitaxel (Fig. 7F). None of the treatments had anysignificant effect on body weight (Fig. S4B). Taken together,these results indicate that combination therapy of TNBC withdigoxin blocks paclitaxel-induced IL-6/IL-8/MDR1 expressionand BCSC enrichment in vivo.

Digoxin Blocks Gemcitabine-Induced IL-6 Expression and BCSCEnrichment. To analyze the effect of another chemotherapeuticagent, we treated TNBC cells with gemcitabine, which is a nu-cleoside analog that blocks DNA replication and acts by a dif-ferent mechanism of action than paclitaxel, which binds tomicrotubules and blocks their breakdown during cell division.

HIF-1α protein expression was induced in a dose-dependentmanner when MDA-MB-231 (Fig. 8A) or SUM-159 (Fig. 8B)cells were exposed to gemcitabine at IC50 for 4 d. Treatment ofMDA-MB-231 or SUM-159 cells with gemcitabine for 4 d increasedthe percentage of ALDH+ cells (Fig. 8C) and mammosphere-forming cells (Fig. 8D and Fig. S5A), respectively, and theseeffects were blocked by coadministration of digoxin. Gemcita-bine induced expression of IL-6 and MDR1 mRNA expression,

Fig. 7. Digoxin blocks paclitaxel-induced BCSC enrichment in TNBCorthografts and xenografts. (A–C) MDA-MB-231 cells were implanted intothe mammary fat pad of female Scid mice. When tumor volume reached200 mm3, mice were randomized to four groups, which were treated with:saline (control, C); digoxin (D; 2 mg/kg on days 1–12); paclitaxel (P; 10 mg/kgon days 5 and 10); or paclitaxel and digoxin (P/D). Tumor volumes weredetermined every 2–3 d (A). Tumors were harvested on day 12 for Aldefluor(B) and RT-qPCR (C) assays. Data are shown as mean ± SEM (n = 3). *P < 0.001compared with C, and #P < 0.001 compared with P, by Student’s t test. (D–F)SUM-159 cells were injected s.c. into female athymic Nude mice. When tu-mor volume reached 500 mm3, mice were randomized to four groups, whichwere treated as described above. Tumor volume was determined every 2–3 d(D). Tumors were harvested on day 12 for Aldefluor (E) or mammosphere (F)assays. Data are mean ± SEM (n = 3). *P < 0.01 compared with C, and #P <0.01 compared with P, by Student’s t test. Representative photomicrographsof mammospheres are shown. (Scale bar, 2 mm.)

Fig. 8. Gemcitabine induces BCSC enrichment in vitro and in vivo. (A and B)MDA-MB-231 and SUM-159 cells were treated with 0–20 nM gemcitabine(Gem) either alone or in combination with 100 nM digoxin (D) or 1 μM ac-riflavine (A) for 4 d and immunoblot assays were performed. (C and D) Cellswere treated with vehicle (V), digoxin (D), gemcitabine (G; 20 nM for MDA-MB-231 and 10 nM for SUM-159), or gemcitabine and digoxin (G/D) for 4 d.Cells were trypsinized and subjected to Aldefluor (C) and mammosphere (D)assays (mean ± SEM; n = 3). *P < 0.001 compared with V, and #P < 0.001compared with G, by Student’s t test. (E–G) MDA-231 cells were treated withvehicle (V) or 20 nM gemcitabine, either alone (G) or in combination with100 nM digoxin (G/D) or 1 μM acriflavine (G/A) for 4 d. RT-qPCR assays wereperformed for IL-6 (E), MDR1 (F), and IL-8 (G) mRNA relative to 18S rRNA,with expression normalized to vehicle-treated cells (mean ± SEM; n = 3). *P <0.001 compared with V, and #P < 0.001 compared with G, by Student’s t test.(H) MDA-MB-231 cells were exposed to vehicle or gemcitabine (Gem), eitheralone or in combination with IgG or neutralizing antibody against IL-6 or IL-8(500 ng/mL). After 4 d, the percentage of ALDH+ cells was determined (mean ±SEM; n = 3). *P < 0.001 compared with vehicle, and #P < 0.001 comparedwith gemcitabine, by Student’s t test. (I and J) MDA-MB-231 cells wereimplanted into the mammary fat pad of female Scid mice. When tumorvolume reached 200 mm3, the mice were randomized to four groups, whichwere treated with: saline control, digoxin (2 mg/kg on days 1–12), gemci-tabine (20 mg/kg on days 5 and 10), or gemcitabine and digoxin. Tumorvolumes were determined every 2–3 d (I). Tumors were harvested on day 12and the percentage of ALDH+ cells was determined (J; mean ± SEM; n = 3).*P < 0.001 compared with control and #P < 0.001 compared with gemcita-bine, by Student’s t test.

E5434 | www.pnas.org/cgi/doi/10.1073/pnas.1421438111 Samanta et al.

which was inhibited by combined treatment with digoxin or ac-riflavine, in MDA-MB-231 cells (Fig. 8 E and F) and SUM-159cells (Fig. S5 B and C). In contrast to paclitaxel, gemcitabine didnot induce IL-8 mRNA expression (Fig. 8G). However, treat-ment with neutralizing antibodies revealed that IL-8 signalingwas required for BCSC maintenance in the absence of chemo-therapy and that both IL-6 and IL-8 signaling were required forgemcitabine-induced BCSC enrichment (Fig. 8H).Next, we implanted MDA-MB-231 cells into the MFP of Scid

mice. When the tumor volume reached 200 mm3, mice wererandomized to receive i.p. injections of saline, digoxin (2 mg/kg/d),gemcitabine (20 mg/kg on days 5 and 10), or the combination ofdigoxin and gemcitabine. Combination therapy had a significantlygreater inhibitory effect on tumor growth than either drug alone(Fig. 8I) without any significant effect on body weight (Fig. S4C).Gemcitabine treatment increased the percentage of ALDH+ cellsin the residual tumor but digoxin abrogated the gemcitabine-induced increase in ALDH+ cells (Fig. 8J). Taken together, thedata presented in Fig. 8 (and Fig. S5) indicate that digoxin blocksHIF induction, IL-6 expression, and BCSC enrichment inducedby treatment of TNBC cells with gemcitabine.

Coadministration of Digoxin with Gemcitabine Prevents TumorRelapse. Patients with TNBC have the highest rate of relapsewithin 1–3 y despite adjuvant chemotherapy (10, 11). To examinethe effect of HIF inhibition on tumor relapse, MDA-MB-231cells were implanted into the MFP of Scid mice. When the tumorbecame palpable, the mice were randomized to receive i.p.injections of saline, gemcitabine (20 mg/kg on days 5, 10, 15, 20,and 25), or the combination of digoxin (2 mg/kg on days 1–25)and gemcitabine. Treatment was stopped when tumors were nolonger palpable in the mice receiving digoxin/gemcitabine com-bination therapy. Although treatment with gemcitabine mark-edly reduced tumor growth, after treatment was discontinued,rapid tumor growth resumed. In contrast, gemcitabine/digoxincombination therapy not only eliminated the primary tumor butalso prevented any immediate tumor relapse (Fig. 9). Takentogether, the data in Figs. 8 and 9 indicate that inhibiting theHIF pathway blocks chemotherapy-induced BCSC enrichment inthe primary tumor and thereby prevents rapid relapse of TNBC.

Effect of Cytotoxic Chemotherapy on HIF-1α Expression and BCSCEnrichment in Other Breast Cancer Subtypes. To investigatewhether chemotherapy-induced BCSC enrichment is limited tobasal subtype/TNBCs, we treated MCF-7 (luminal A subtype/ER+PR+) and HCC-1954 (HER2+) cells with paclitaxel at IC50.Exposure of MCF-7 cells to 10 nM paclitaxel for 4 d inducedHIF-1α and HIF-2α protein expression, whereas the inductionwas abrogated by coadministration of digoxin or acriflavine (Fig.S6A). Paclitaxel increased the percentage of ALDH+ cells,whereas digoxin decreased the percentage of ALDH+ cells, andcoadministration of digoxin significantly reduced the paclitaxel-dependent increase in the percentage of ALDH+ cells (Fig.S6B). Similarly, when HCC-1954 cells were treated with in-creasing doses of paclitaxel (Fig. S6C) or gemcitabine (Fig. S6E)for 4 d, there was a dose-dependent increase in HIF-1α proteinexpression, which was abrogated by coadministration of digoxinor acriflavine. However, there was no enrichment of ALDH+

cells after treatment with paclitaxel (Fig. S6D) or gemcitabine(Fig. S6F). The MCF-7 results demonstrate that chemotherapy-induced BCSC enrichment is not restricted to TNBC cells,whereas the HCC-1954 data indicate that induction of HIF-1α isnot sufficient for chemotherapy-induced BCSC enrichment.

HIF Signature Predicts Mortality of Patients Treated with Chemotherapy.To investigate the clinical relevance of HIF transcriptional activitywith regard to the survival of patients with breast cancer, we useda HIF-1 signature composed of 16 genes (PLOD1, VEGFA, LOX,

P4HA2, NDRG1, SLC2A1, ERO1L, ADM, LDHA, PGK1,ANGPTL4, SLC2A3, CA9, HIF1A, IL6, and IL8), which areexpressed in human breast cancer cell lines in a HIF-1–dependent manner. This HIF-1 signature was compared with thePAM50 signature, which consists of 50 genes, the expression ofwhich divides human breast cancer specimens into five groups,designated basal, HER2 enriched, luminal B, luminal A, andnormal (45, 46). Most TNBCs cluster in the basal subtype (12).Comparison of the HIF-1 and PAM50 signatures in 1,160 breastcancer specimens (47) revealed increased expression of HIF-1target genes in the basal and, to a lesser extent, the HER2-enriched subtypes (Fig. S7). Analysis of HIF-1α mRNA ex-pression in 3,458 human breast cancer specimens (48) revealedthat levels above the median were associated with significantlydecreased overall survival [P < 10−10; hazard ratio (HR) = 1.48;Fig. 10A], with an even larger survival difference when only breastcancers of the basal subtype were analyzed (HR = 1.56; Fig. 10B).Use of the HIF-1 signature resulted in an even more significantdifference in patient survival in the overall population (P < 10−15;HR = 1.63; Fig. 10C) and increased the hazard ratio in the sub-population with breast cancers of the basal subtype (HR = 1.90; Fig.10D). Finally, analysis of data from patients treated only with che-motherapy revealed that in this subpopulation, differences in ex-pression of the HIF-1 signature were associated with significantdifferences in survival (HR = 1.58; Fig. 10E), especially amongpatients with basal subtype tumors (HR = 3.45; Fig. 10F). Theseresults are consistent with our in vitro and in vivo studies demon-strating HIF-dependent BCSC enrichment after chemotherapy.

DiscussionEffective treatment for TNBC and other advanced breast can-cers requiring cytotoxic chemotherapy represents a major unmetclinical need. Recent studies have suggested that eradication ofBCSCs is required to achieve a complete remission and to pre-vent disease recurrence, but BCSCs are resistant to chemother-apy (4–6). Previous studies have demonstrated induction of HIF-1αexpression by doxorubicin (25) and increased sensitivity ofHIF-1α–deficient mouse embryo fibroblasts to killing by carbo-platin and etoposide (16). In the present study, we have dem-onstrated that treatment of three different TNBC cell lines withclinically relevant concentrations of paclitaxel or gemcitabineincreases ROS levels, which induce HIF-1α and HIF-2α mRNAand protein expression, leading to the HIF-dependent expression

Fig. 9. Combination therapy with gemcitabine and digoxin prevents tumorrelapse. MDA-MB-231 cells were implanted into the mammary fat pad offemale Scid mice. When a tumor was palpable (day 1), the mice were ran-domized to three groups, which were treated with i.p. injections of: saline,gemcitabine [20 mg/kg on days 5, 10, 15, 20, and 25 (arrows)], or gemcita-bine and digoxin (2 mg/kg on days 1–25). Tumor volumes were determinedevery 2–3 d and mean ± SEM (n = 3) are shown. The saline-treated mice wereeuthanized (E) on day 24 when tumor volume exceeded 1,000 mm3. Theexperiment was terminated on day 32, 7 d after treatment was discontinued.*P < 0.001 by Student’s t test.

Samanta et al. PNAS | Published online December 1, 2014 | E5435

MED

ICALSC

IENCE

SPN

ASPL

US

of MDR1, IL-8, and/or IL-6, and to BCSC enrichment, both invitro and in vivo (Fig. 11). Coordinate expression of IL-6 and IL-8has been shown to play an essential role in TNBC tumorigenesisand is associated with decreased survival in patients with breastcancer (37). Both IL-6 (34, 37) and IL-8 (6, 35) have been im-plicated in the maintenance of BCSCs and their resistance tochemotherapy, particularly in basal subtype/TNBC.It should be noted that cytotoxic chemotherapy induced HIF-1α

and HIF-2α expression under nonhypoxic conditions and that bothHIF-1α and HIF-2α were required for maximal chemotherapy-induced BCSC enrichment, whereas only HIF-1α was requiredfor BCSC enrichment in response to hypoxia (23), indicatingdifferences in the mechanisms leading to BCSC enrichment inresponse to chemotherapy vs. hypoxia that remain to be delin-eated. Although both paclitaxel and gemcitabine induced HIF-dependent IL-6 expression and BCSC enrichment, only paclitaxelinduced IL-8 expression. However, basal IL-8 signaling was stillrequired for gemcitabine-induced BCSC enrichment, which isconsistent with previous studies demonstrating that inhibitors ofCXCR1, which is the cognate receptor for IL-8, have profoundeffects on BCSCs (35).A previous study reported that paclitaxel-induced IL-8 expres-

sion and BCSC enrichment was dependent upon TGF-β pathwaysignaling to SMAD2/4 (6). Similarly, IL-6 expression was linkedto STAT3 activation and BCSCs (37). However, we found thatHIF deficiency in the MDA-MB-231-DKD subclone blockedpaclitaxel-induced IL-8 and IL-6 expression and BCSC enrich-ment, yet phosphorylation of SMAD2 and STAT3 was still in-duced in the DKD cells, indicating that induction of SMAD2 andSTAT3 activity was HIF independent and was not sufficient forBCSC enrichment. We also found that expression of HIF-1α, butnot IL-8, was induced in gemcitabine-treated MDA-MB-231cells, and that paclitaxel or gemcitabine treatment of HCC-1954cells induced HIF-1α expression but not BCSC enrichment, in-dicating that HIF-1α expression is not sufficient for IL-8 ex-pression or BCSC enrichment. Taken together, our data andprior studies suggest that HIFs, SMADs, and STAT3 all con-tribute to BCSC enrichment in response to cytotoxic chemo-therapy. However, from a translational point of view, the criticalresult is that inhibition of HIF activity is sufficient to block the

enrichment of BCSCs, regardless of the activity of other tran-scription factors.IL-6 and IL-8 signaling promotes BCSC enrichment by

inhibiting chemotherapy-induced apoptosis and by promotingthe BCSC phenotype. In contrast, MDR1 expression acts tomaintain cell viability by pumping cytotoxic chemotherapy agentsout of cancer cells. Our data indicate that treatment of TNBC

Fig. 10. Expression of HIF-1α and HIF-1 target gene mRNAs in primary breast cancers predicts clinical outcome. Kaplan–Meier analyses of overall survival(120 mo) were performed based on clinical and molecular data for 2,785 breast cancer patients, who were stratified by expression levels in the primary tumorof HIF-1α mRNA (A and B) or a 16-gene HIF-1 signature (C–F), which were greater (red) or less (black) than the median. The P value (log-rank test) and hazardratio (HR) for each comparison are shown. Patients of all breast cancer types (A and C) or only basal-type breast cancer (B and D) were analyzed. Within thesetwo groups, the subgroup of patients who received only chemotherapy (E and F) was also analyzed.

Fig. 11. Cytotoxic chemotherapy induces ROS-dependent expression of HIF-1αand HIF-2α, leading to HIF-mediated expression of IL-6, IL-8, and MDR1,which promote the survival of BCSCs, a response that is a critical determinantof treatment failure and patient mortality.

E5436 | www.pnas.org/cgi/doi/10.1073/pnas.1421438111 Samanta et al.

cells with either paclitaxel or gemcitabine increased MDR1expression in a HIF-dependent manner in ALDH− cells and toan even greater extent in ALDH+ cells, providing a mechanismfor increased resistance of BCSCs to chemotherapy relative tobulk cancer cells. IL-8 signaling also contributes to this differ-ential resistance because CXCR1 expression is restricted toBCSCs (35).Combination therapy with digoxin reversed the effects of

paclitaxel or gemcitabine at the molecular level (IL-6, IL-8,and MDR1 expression), resulting in improved tumor controland abrogation of chemotherapy-induced BCSC enrichmentin orthotopic and xenograft models of TNBC. We recentlyreported that, in contrast to cytotoxic chemotherapy, treat-ment of mice bearing orthotopic TNBC tumors with gane-tespib, which is a second-generation HSP90 inhibitor, inducedthe degradation of HIF-1α (but not HIF-2α), inhibited pri-mary tumor growth, and reduced the number of ALDH+ cellsin the residual tumor (49), providing further evidence thatHIF inhibition provides a mechanism to prevent BCSC en-richment. HIF inhibition by methylselenocysteine sensitizedhead and neck squamous cell carcinoma xenografts to thecytotoxic chemotherapy agent irinotecan (50), suggesting thatHIF-mediated cancer stem cell enrichment in response tochemotherapy may not be restricted to breast cancer treatedwith gemcitabine or paclitaxel and may be blocked by otherHIF inhibitors.The clinical relevance of these results is underscored by studies

demonstrating associations between: (i) ALDH expression andER−/basal subtype breast cancer (51); (ii) HIF-1α and ALDH+

cells in breast cancers, both before and after chemotherapy(51); (iii) ALDH+ breast cancer cells and resistance to se-quential paclitaxel and epirubicin chemotherapy (52); and (iv)increased ALDH+ cells after chemotherapy and decreaseddisease-free survival (51, 53). Taken together with these clin-ical studies, our results provide compelling evidence in supportof clinical trials combining HIF inhibitors and cytotoxic che-motherapy in women with TNBC. Several drugs have beenshown to inhibit HIF activity (49, 50, 54) and are candidatesfor such trials.

MethodsCell Culture. MDA-MB-231 and MCF-7 cells were maintained in high-glucose(4.5 mg/mL) Dulbecco’s modified Eagle medium (DMEM) with 10% (vol/vol)FBS and penicillin/streptomycin. SUM-159 and SUM-149 cells were main-tained in DMEM/F12 (50:50) with 10% (vol/vol) FBS, hydrocortisone, insulin,and 1% penicillin/streptomycin. HCC-1954 cells were maintained in RPMI-1640 media with 10% (vol/vol) FBS and penicillin/streptomycin. MDA-MB-231knockdown subclones were cultured in the presence of 0.5 μg/mL of puro-mycin. Cells were maintained at 37 °C in a 5% CO2, 95% air incubator (20%O2). Paclitaxel, digoxin, acriflavine, and gemcitabine were obtained fromSigma-Aldrich and dissolved in DMSO at 1,000× relative to final concentra-tion in tissue culture medium. MnTMPyP was obtained from Calbiochem anddissolved in deionized water.

RT-qPCR. Total RNA was extracted from cells and tissue using TRIzol (Invi-trogen) and treated with DNase I (Ambion). First-strand cDNA synthesis wasperformed using 1 μg of total RNA and the iScript cDNA Synthesis system(Bio-Rad), and qPCR was performed using human-specific primers and iQSYBR Green Supermix (Bio-Rad). For each primer pair, annealing tempera-ture was optimized by gradient PCR. The expression (E) of each target mRNArelative to 18S rRNAwas calculated based on the cycle threshold (Ct): E = 2–Δ(ΔCt),in which ΔCt = Cttarget – Ct18S and Δ(ΔCt) = ΔCttreatment – ΔCtcontrol. See Table S1for primer sequences.

Immunoblot Assays. Whole-cell lysates were prepared in RIPA lysis buffer.Blots were probed with antibodies against HIF-1α, HIF-2α, IL-6, IL-8, MDR1,phospho-SMAD2, phospho-STAT3, SMAD2, and STAT3 (Novus Biologicals).HRP-conjugated anti-rabbit (Roche) and anti-mouse (Santa Cruz) secondaryantibodies were used. Chemiluminescent signal was detected using ECL Plus

(GE Healthcare). Blots were stripped and reprobed with anti-actin antibody(Santa Cruz).

Luciferase Assay. 2 × 104 MDA-MB-231 cells were seeded onto 24-wellplates overnight and the cells were transfected with plasmid DNA usingPolyJet (SignaGen). Reporter plasmids pSV-RL (5 ng) and p2.1 (295 ng) werecotransfected. The media was changed after 6 h. Starting the next day, thecells were treated with either vehicle or 10 nM paclitaxel. The cells werelysed after 4 d and luciferase activities were determined with a multiwellluminescence reader (Perkin-Elmer Life Science) using a dual luciferase re-porter assay system (Promega).

Aldefluor Assay. After treatment of cultured cells for 4 d, the Aldefluor assay(StemCell Technologies) was performed to identify cells with ALDH activity.Cultured cells were trypsinized, whereas tumor tissue was minced, digestedwith 1 mg/mL of type 1 collagenase (Sigma) at 37 °C for 30 min, and filteredthrough a 70-μm cell strainer. The number of live cells was determined byTrypan blue assay and 1 × 106 live cells were suspended in assay buffercontaining the fluorogenic substrate BODIPY aminoacetaldehyde (1 μM) andincubated for 45 min at 37 °C. As a negative control, an aliquot of cells wastreated with the ALDH inhibitor diethylaminobenzaldehyde (50 mM). Sam-ples were then passed through a 35-μm strainer and analyzed by flowcytometry (FACSCalibur; BD Biosciences).

Mammosphere Assay. Single-cell suspensions were seeded in six-well ultra-lowattachment plates (Corning) at a density of 5,000 cells permilliliter in CompleteMammoCult Medium (StemCell Technologies). After 7 d, the cells were photo-graphed under an Olympus TH4-100 microscope with 4× apochromat objectivelens. Mammosphere number and volume were determined using ImageJ soft-ware. Mammospheres with area >500 pixels were counted in images of threefields per well in triplicate wells and the mean number of mammospheres perfield was determined. For secondary mammosphere formation, primary mam-mospheres were trypsinized, plated at a density of 5,000 cells per milliliter, in-cubated for 7 d, and analyzed as described above.

MitoSOX Staining. Intracellular ROS levels were determined by incubating thecells in 5 μM MitoSOX Red (Molecular Probes) at 37 °C for 45 min in PBS with5% FBS, followed by rinsing with PBS. Stained cells were filtered and sub-jected to flow cytometry (FACScan; BD Bioscience). All gain and amplifiersettings were held constant for all samples.

Animal Studies. Animal protocols were in accordance with the NationalInstitutes of Health Guide for the Care and Use of Laboratory Animals (55)and were approved by the Johns Hopkins University Animal Care and UseCommittee. Female 5- to 7-wk-old Scid and Nude mice (National CancerInstitute) were studied. Paclitaxel, gemcitabine, digoxin, and saline for in-jection were obtained from the research pharmacy of The Johns HopkinsHospital. Cells were harvested by trypsinization, rinsed with PBS, andresuspended at 2 × 107 cells per milliliter in a 1:1 solution of PBS/Matrigel.MDA-MB-231 cells were injected in the MFP of Scid mice and SUM-159 cellswere injected s.c. in Nude mice. Primary tumors were measured in threedimensions (a, b, and c), and volume (V) was calculated as V = abc × 0.52.

Statistical Analysis. Data are expressed as mean ± SEM. Differences betweentwo groups and multiple groups were analyzed by Student’s t test andANOVA, respectively. P values <0.05 were considered significant. For the HIF-1signature, the Breast Invasive Carcinoma Gene Expression Dataset of 1,162patients was analyzed (47). Kaplan–Meier curves were generated froma breast cancer dataset containing gene expression and overall survival dataon 2,785 patients (48). The log rank test was performed to determinewhether observed differences between groups were statistically significant.The analysis of HIF-1 signature expression according to breast cancer subtypewas performed using the GOBO database (56).

ACKNOWLEDGMENTS. We thank Vered Stearns of the Kimmel CancerCenter at Johns Hopkins University for helpful discussions and Karen Padgettof Novus Biologicals for providing IgG and antibodies against HIF-2α, IL-6,IL-8, MDR1, p-SMAD2, SMAD2, p-STAT3, STAT3, and p-IκBα. This work wassupported by Breast Cancer Research Program Impact Award W81XWH-12-1-0464 from the Department of Defense; Cigarette Restitution Fund ResearchGrant FH-B33-CRF from the State of Maryland Department of Health andMental Hygiene; the Cindy Rosencrans Fund for Triple Negative Breast Can-cer; and The WTFC. D.M.G. was supported by National Cancer Institute GrantK99-CA181352. G.L.S. is an American Cancer Society Research Professor andthe C. Michael Armstrong Professor at The Johns Hopkins University Schoolof Medicine.

Samanta et al. PNAS | Published online December 1, 2014 | E5437

MED

ICALSC

IENCE

SPN

ASPL

US

1. 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.

2. Heddleston JM, et al. (2010) Hypoxia inducible factors in cancer stem cells. Br J Cancer102(5):789–795.

3. Oskarsson T, Batlle E, Massagué J (2014) Metastatic stem cells: Sources, niches, andvital pathways. Cell Stem Cell 14(3):306–321.

4. Li X, et al. (2008) Intrinsic resistance of tumorigenic breast cancer cells to chemo-therapy. J Natl Cancer Inst 100(9):672–679.

5. Creighton CJ, et al. (2009) Residual breast cancers after conventional therapy displaymesenchymal as well as tumor-initiating features. Proc Natl Acad Sci USA 106(33):13820–13825.

6. Bhola NE, et al. (2013) TGF-β inhibition enhances chemotherapy action against triple-negative breast cancer. J Clin Invest 123(3):1348–1358.

7. Dent R, et al. (2007) Triple-negative breast cancer: Clinical features and patterns ofrecurrence. Clin Cancer Res 13(15 Pt 1):4429–4434.

8. 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.

9. Polyak K (2011) Heterogeneity in breast cancer. J Clin Invest 121(10):3786–3788.10. Foulkes WD, Smith IE, Reis-Filho JS (2010) Triple-negative breast cancer. N Engl J Med

363(20):1938–1948.11. Russnes HG, Navin N, Hicks J, Borresen-Dale AL (2011) Insight into the heterogeneity

of breast cancer through next-generation sequencing. J Clin Invest 121(10):3810–3818.

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

13. Prabhakar NR, Semenza GL (2012) Adaptive and maladaptive cardiorespiratory re-sponses to continuous and intermittent hypoxia mediated by hypoxia-inducible fac-tors 1 and 2. Physiol Rev 92(3):967–1003.

14. Semenza GL (2013) HIF-1 mediates metabolic responses to intratumoral hypoxia andoncogenic mutations. J Clin Invest 123(9):3664–3671.

15. Gao P, et al. (2007) HIF-dependent antitumorigenic effect of antioxidants in vivo.Cancer Cell 12(3):230–238.

16. Unruh A, et al. (2003) The hypoxia-inducible factor-1 α is a negative factor for tumortherapy. Oncogene 22(21):3213–3220.

17. Rohwer N, Cramer T (2011) Hypoxia-mediated drug resistance: Novel insights on thefunctional interaction of HIFs and cell death pathways. Drug Resist Updat 14(3):191–201.

18. Comerford KM, et al. (2002) Hypoxia-inducible factor-1-dependent regulation of themultidrug resistance (MDR1) gene. Cancer Res 62(12):3387–3394.

19. Pastan I, Gottesman MM (1991) Multidrug resistance. Annu Rev Med 42:277–286.20. Semenza GL (2014) Oxygen sensing, hypoxia-inducible factors, and disease patho-

physiology. Annu Rev Pathol 9:47–71.21. Gilkes DM, Semenza GL (2013) Role of hypoxia-inducible factors in breast cancer

metastasis. Future Oncol 9(11):1623–1636.22. Buffa FM, Harris AL, West CM, Miller CJ (2010) Large meta-analysis of multiple cancers

reveals a common, compact and highly prognostic hypoxia metagene. Br J Cancer102(2):428–435.

23. 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.

24. Schwab LP, et al. (2012) Hypoxia-inducible factor 1α promotes primary tumor growthand tumor-initiating cell activity in breast cancer. Breast Cancer Res 14(1):R6.

25. Cao Y, et al. (2013) Tumor cells upregulate normoxic HIF-1α in response to doxoru-bicin. Cancer Res 73(20):6230–6242.

26. Zhang H, et al. (2008) Digoxin and other cardiac glycosides inhibit HIF-1α synthesisand block tumor growth. Proc Natl Acad Sci USA 105(50):19579–19586.

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

28. Wong CC, et al. (2012) Inhibitors of hypoxia-inducible factor 1 block breast cancermetastatic niche formation and lung metastasis. J Mol Med (Berl) 90(7):803–815.

29. Zhang H, et al. (2012) HIF-1-dependent expression of angiopoietin-like 4 and L1CAMmediates vascular metastasis of hypoxic breast cancer cells to the lungs. Oncogene31(14):1757–1770.

30. 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.

31. Charafe-Jauffret E, et al. (2009) Breast cancer cell lines contain functional cancer stemcells with metastatic capacity and a distinct molecular signature. Cancer Res 69(4):1302–1313.

32. Dontu G, et al. (2003) In vitro propagation and transcriptional profiling of human

mammary stem/progenitor cells. Genes Dev 17(10):1253–1270.33. Ponti D, et al. (2005) Isolation and in vitro propagation of tumorigenic breast cancer

cells with stem/progenitor cell properties. Cancer Res 65(13):5506–5511.34. Sansone P, et al. (2007) IL-6 triggers malignant features in mammospheres from hu-

man ductal breast carcinoma and normal mammary gland. J Clin Invest 117(12):

3988–4002.35. Ginestier C, et al. (2010) CXCR1 blockade selectively targets human breast cancer stem

cells in vitro and in xenografts. J Clin Invest 120(2):485–497.36. Iliopoulos D, Hirsch HA, Wang G, Struhl K (2011) Inducible formation of breast cancer

stem cells and their dynamic equilibrium with non-stem cancer cells via IL6 secretion.

Proc Natl Acad Sci USA 108(4):1397–1402.37. Marotta LL, et al. (2011) The JAK2/STAT3 signaling pathway is required for growth of

CD44+CD24− stem cell-like breast cancer cells in human tumors. J Clin Invest 121(7):

2723–2735.38. Hartman ZC, et al. (2013) Growth of triple-negative breast cancer cells relies upon

coordinate autocrine expression of the proinflammatory cytokines IL-6 and IL-8.

Cancer Res 73(11):3470–3480.39. Mimura I, et al. (2012) Dynamic change of chromatin conformation in response to

hypoxia enhances the expression of GLUT3 (SLC2A3) by cooperative interaction of

hypoxia-inducible factor 1 and KDM3A. Mol Cell Biol 32(15):3018–3032.40. Przanowski P, et al. (2014) The signal transducers Stat1 and Stat3 and their novel

target Jmjd3 drive the expression of inflammatory genes in microglia. J Mol Med

(Berl) 92(3):239–254.41. Rajput S, Volk-Draper LD, Ran S (2013) TLR4 is a novel determinant of the response to

paclitaxel in breast cancer. Mol Cancer Ther 12(8):1676–1687.42. Kendellen MF, Bradford JW, Lawrence CL, Clark KS, Baldwin AS (2014) Canonical and

non-canonical NF-κB signaling promotes breast cancer tumor-initiating cells. Onco-

gene 33(10):1297–1305.43. Alexandre J, Hu Y, Lu W, Pelicano H, Huang P (2007) Novel action of paclitaxel against

cancer cells: Bystander effect mediated by reactive oxygen species. Cancer Res 67(8):

3512–3517.44. Ramanathan B, et al. (2005) Resistance to paclitaxel is proportional to cellular total

antioxidant capacity. Cancer Res 65(18):8455–8460.45. Parker JS, et al. (2009) Supervised risk predictor of breast cancer based on intrinsic

subtypes. J Clin Oncol 27(8):1160–1167.46. 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):4465–4472.47. Goldman M, et al. (2013) The UCSC Cancer Genomics Browser: Update 2013. Nucleic

Acids Res 41(Database issue):D949–D954.48. Györffy B, et al. (2010) An online survival analysis tool to rapidly assess the effect of

22,277 genes on breast cancer prognosis using microarray data of 1,809 patients.

Breast Cancer Res Treat 123(3):725–731.49. Xiang L, et al. (2014) Ganetespib blocks HIF-1 activity and inhibits tumor growth,

vascularization, stem cell maintenance, invasion, and metastasis in orthotopic mouse

models of triple-negative breast cancer. J Mol Med (Berl) 92(2):151–164.50. Chintala S, et al. (2010) Se-methylselenocysteine sensitizes hypoxic tumor cells to

irinotecan by targeting hypoxia-inducible factor 1α. Cancer Chemother Pharmacol

66(5):899–911.51. Tiezzi DG, et al. (2013) Expression of aldehyde dehydrogenase after neoadjuvant

chemotherapy is associated with expression of hypoxia-inducible factors 1 and 2 al-

pha and predicts prognosis in locally advanced breast cancer. Clinics (Sao Paulo) 68(5):

592–598.52. Tanei T, et al. (2009) Association of breast cancer stem cells identified by aldehyde

dehydrogenase 1 expression with resistance to sequential Paclitaxel and epirubicin-

based chemotherapy for breast cancers. Clin Cancer Res 15(12):4234–4241.53. Lee HE, et al. (2011) An increase in cancer stem cell population after primary systemic

therapy is a poor prognostic factor in breast cancer. Br J Cancer 104(11):1730–1738.54. Semenza GL (2012) Hypoxia-inducible factors: Mediators of cancer progression and

targets for cancer therapy. Trends Pharmacol Sci 33(4):207–214.55. Committee on Care and Use of Laboratory Animals (1996) Guide for the Care and Use

of Laboratory Animals (Natl Inst Health, Bethesda), DHHS Publ No (NIH) 85-23.56. 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.

E5438 | www.pnas.org/cgi/doi/10.1073/pnas.1421438111 Samanta et al.